From cbc5b8bdae1d97960116bf73c82115b7f52666dd Mon Sep 17 00:00:00 2001
From: sjplimp <sjplimp@f3b2605a-c512-4ea7-a41b-209d697bcdaa>
Date: Tue, 31 May 2016 15:30:30 +0000
Subject: [PATCH] git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@15076
f3b2605a-c512-4ea7-a41b-209d697bcdaa
---
src/Makefile | 2 +-
src/USER-CUDA/Install.sh | 202 --
src/USER-CUDA/README | 21 -
src/USER-CUDA/atom_vec_angle_cuda.cpp | 467 ---
src/USER-CUDA/atom_vec_angle_cuda.h | 69 -
src/USER-CUDA/atom_vec_atomic_cuda.cpp | 394 ---
src/USER-CUDA/atom_vec_atomic_cuda.h | 81 -
src/USER-CUDA/atom_vec_charge_cuda.cpp | 394 ---
src/USER-CUDA/atom_vec_charge_cuda.h | 69 -
src/USER-CUDA/atom_vec_full_cuda.cpp | 508 ---
src/USER-CUDA/atom_vec_full_cuda.h | 69 -
src/USER-CUDA/comm_cuda.cpp | 1375 --------
src/USER-CUDA/comm_cuda.h | 69 -
src/USER-CUDA/compute_pe_cuda.cpp | 61 -
src/USER-CUDA/compute_pe_cuda.h | 59 -
src/USER-CUDA/compute_pressure_cuda.cpp | 97 -
src/USER-CUDA/compute_pressure_cuda.h | 63 -
src/USER-CUDA/compute_temp_cuda.cpp | 215 --
src/USER-CUDA/compute_temp_cuda.h | 76 -
src/USER-CUDA/compute_temp_partial_cuda.cpp | 360 --
src/USER-CUDA/compute_temp_partial_cuda.h | 84 -
src/USER-CUDA/cuda.cpp | 1067 ------
src/USER-CUDA/cuda_data.h | 796 -----
src/USER-CUDA/cuda_modify_flags.h | 45 -
src/USER-CUDA/cuda_neigh_list.cpp | 184 --
src/USER-CUDA/cuda_neigh_list.h | 83 -
src/USER-CUDA/domain_cuda.cpp | 345 --
src/USER-CUDA/domain_cuda.h | 41 -
src/USER-CUDA/fft3d_cuda.cpp | 609 ----
src/USER-CUDA/fft3d_cuda.h | 148 -
src/USER-CUDA/fft3d_wrap_cuda.cpp | 111 -
src/USER-CUDA/fft3d_wrap_cuda.h | 68 -
src/USER-CUDA/fix_addforce_cuda.cpp | 193 --
src/USER-CUDA/fix_addforce_cuda.h | 64 -
src/USER-CUDA/fix_aveforce_cuda.cpp | 262 --
src/USER-CUDA/fix_aveforce_cuda.h | 68 -
src/USER-CUDA/fix_enforce2d_cuda.cpp | 171 -
src/USER-CUDA/fix_enforce2d_cuda.h | 55 -
src/USER-CUDA/fix_freeze_cuda.cpp | 137 -
src/USER-CUDA/fix_freeze_cuda.h | 57 -
src/USER-CUDA/fix_gravity_cuda.cpp | 180 -
src/USER-CUDA/fix_gravity_cuda.h | 60 -
src/USER-CUDA/fix_nh_cuda.cpp | 2072 ------------
src/USER-CUDA/fix_nh_cuda.h | 126 -
src/USER-CUDA/fix_npt_cuda.cpp | 75 -
src/USER-CUDA/fix_npt_cuda.h | 36 -
src/USER-CUDA/fix_nve_cuda.cpp | 157 -
src/USER-CUDA/fix_nve_cuda.h | 63 -
src/USER-CUDA/fix_nvt_cuda.cpp | 52 -
src/USER-CUDA/fix_nvt_cuda.h | 36 -
src/USER-CUDA/fix_set_force_cuda.cpp | 184 --
src/USER-CUDA/fix_set_force_cuda.h | 63 -
src/USER-CUDA/fix_shake_cuda.cpp | 2885 -----------------
src/USER-CUDA/fix_shake_cuda.h | 130 -
src/USER-CUDA/fix_temp_berendsen_cuda.cpp | 219 --
src/USER-CUDA/fix_temp_berendsen_cuda.h | 58 -
src/USER-CUDA/fix_temp_rescale_cuda.cpp | 224 --
src/USER-CUDA/fix_temp_rescale_cuda.h | 61 -
src/USER-CUDA/fix_temp_rescale_limit_cuda.cpp | 237 --
src/USER-CUDA/fix_temp_rescale_limit_cuda.h | 61 -
src/USER-CUDA/fix_viscous_cuda.cpp | 105 -
src/USER-CUDA/fix_viscous_cuda.h | 55 -
src/USER-CUDA/modify_cuda.cpp | 437 ---
src/USER-CUDA/modify_cuda.h | 83 -
src/USER-CUDA/neigh_full_cuda.cpp | 307 --
src/USER-CUDA/neighbor_cuda.cpp | 240 --
src/USER-CUDA/neighbor_cuda.h | 39 -
src/USER-CUDA/pair_born_coul_long_cuda.cpp | 183 --
src/USER-CUDA/pair_born_coul_long_cuda.h | 57 -
src/USER-CUDA/pair_buck_coul_cut_cuda.cpp | 170 -
src/USER-CUDA/pair_buck_coul_cut_cuda.h | 57 -
src/USER-CUDA/pair_buck_coul_long_cuda.cpp | 181 --
src/USER-CUDA/pair_buck_coul_long_cuda.h | 57 -
src/USER-CUDA/pair_buck_cuda.cpp | 166 -
src/USER-CUDA/pair_buck_cuda.h | 57 -
src/USER-CUDA/pair_eam_alloy_cuda.cpp | 326 --
src/USER-CUDA/pair_eam_alloy_cuda.h | 44 -
src/USER-CUDA/pair_eam_cuda.cpp | 265 --
src/USER-CUDA/pair_eam_cuda.h | 80 -
src/USER-CUDA/pair_eam_fs_cuda.cpp | 335 --
src/USER-CUDA/pair_eam_fs_cuda.h | 44 -
src/USER-CUDA/pair_gran_hooke_cuda.cpp | 250 --
src/USER-CUDA/pair_gran_hooke_cuda.h | 57 -
src/USER-CUDA/pair_lj96_cut_cuda.cpp | 179 -
src/USER-CUDA/pair_lj96_cut_cuda.h | 57 -
.../pair_lj_charmm_coul_charmm_cuda.cpp | 188 --
.../pair_lj_charmm_coul_charmm_cuda.h | 63 -
...ir_lj_charmm_coul_charmm_implicit_cuda.cpp | 183 --
...pair_lj_charmm_coul_charmm_implicit_cuda.h | 62 -
.../pair_lj_charmm_coul_long_cuda.cpp | 196 --
src/USER-CUDA/pair_lj_charmm_coul_long_cuda.h | 62 -
.../pair_lj_class2_coul_cut_cuda.cpp | 162 -
src/USER-CUDA/pair_lj_class2_coul_cut_cuda.h | 57 -
.../pair_lj_class2_coul_long_cuda.cpp | 175 -
src/USER-CUDA/pair_lj_class2_coul_long_cuda.h | 57 -
src/USER-CUDA/pair_lj_class2_cuda.cpp | 167 -
src/USER-CUDA/pair_lj_class2_cuda.h | 57 -
src/USER-CUDA/pair_lj_cut_coul_cut_cuda.cpp | 162 -
src/USER-CUDA/pair_lj_cut_coul_cut_cuda.h | 57 -
src/USER-CUDA/pair_lj_cut_coul_debye_cuda.cpp | 163 -
src/USER-CUDA/pair_lj_cut_coul_debye_cuda.h | 57 -
src/USER-CUDA/pair_lj_cut_coul_long_cuda.cpp | 216 --
src/USER-CUDA/pair_lj_cut_coul_long_cuda.h | 57 -
src/USER-CUDA/pair_lj_cut_cuda.cpp | 179 -
src/USER-CUDA/pair_lj_cut_cuda.h | 57 -
.../pair_lj_cut_experimental_cuda.cpp | 178 -
src/USER-CUDA/pair_lj_cut_experimental_cuda.h | 57 -
src/USER-CUDA/pair_lj_expand_cuda.cpp | 180 -
src/USER-CUDA/pair_lj_expand_cuda.h | 57 -
.../pair_lj_gromacs_coul_gromacs_cuda.cpp | 194 --
.../pair_lj_gromacs_coul_gromacs_cuda.h | 68 -
src/USER-CUDA/pair_lj_gromacs_cuda.cpp | 177 -
src/USER-CUDA/pair_lj_gromacs_cuda.h | 68 -
src/USER-CUDA/pair_lj_sdk_coul_long_cuda.cpp | 193 --
src/USER-CUDA/pair_lj_sdk_coul_long_cuda.h | 59 -
src/USER-CUDA/pair_lj_sdk_cuda.cpp | 184 --
src/USER-CUDA/pair_lj_sdk_cuda.h | 65 -
src/USER-CUDA/pair_lj_smooth_cuda.cpp | 177 -
src/USER-CUDA/pair_lj_smooth_cuda.h | 68 -
src/USER-CUDA/pair_morse_cuda.cpp | 177 -
src/USER-CUDA/pair_morse_cuda.h | 57 -
src/USER-CUDA/pair_sw_cuda.cpp | 207 --
src/USER-CUDA/pair_sw_cuda.h | 66 -
src/USER-CUDA/pair_tersoff_cuda.cpp | 204 --
src/USER-CUDA/pair_tersoff_cuda.h | 66 -
src/USER-CUDA/pair_tersoff_zbl_cuda.cpp | 220 --
src/USER-CUDA/pair_tersoff_zbl_cuda.h | 53 -
src/USER-CUDA/pppm_cuda.cpp | 1420 --------
src/USER-CUDA/pppm_cuda.h | 113 -
src/USER-CUDA/pppm_old.cpp | 2839 ----------------
src/USER-CUDA/pppm_old.h | 271 --
src/USER-CUDA/user_cuda.h | 159 -
src/USER-CUDA/verlet_cuda.cpp | 1230 -------
src/USER-CUDA/verlet_cuda.h | 63 -
134 files changed, 1 insertion(+), 31295 deletions(-)
delete mode 100755 src/USER-CUDA/Install.sh
delete mode 100644 src/USER-CUDA/README
delete mode 100644 src/USER-CUDA/atom_vec_angle_cuda.cpp
delete mode 100644 src/USER-CUDA/atom_vec_angle_cuda.h
delete mode 100644 src/USER-CUDA/atom_vec_atomic_cuda.cpp
delete mode 100644 src/USER-CUDA/atom_vec_atomic_cuda.h
delete mode 100644 src/USER-CUDA/atom_vec_charge_cuda.cpp
delete mode 100644 src/USER-CUDA/atom_vec_charge_cuda.h
delete mode 100644 src/USER-CUDA/atom_vec_full_cuda.cpp
delete mode 100644 src/USER-CUDA/atom_vec_full_cuda.h
delete mode 100644 src/USER-CUDA/comm_cuda.cpp
delete mode 100644 src/USER-CUDA/comm_cuda.h
delete mode 100644 src/USER-CUDA/compute_pe_cuda.cpp
delete mode 100644 src/USER-CUDA/compute_pe_cuda.h
delete mode 100644 src/USER-CUDA/compute_pressure_cuda.cpp
delete mode 100644 src/USER-CUDA/compute_pressure_cuda.h
delete mode 100644 src/USER-CUDA/compute_temp_cuda.cpp
delete mode 100644 src/USER-CUDA/compute_temp_cuda.h
delete mode 100644 src/USER-CUDA/compute_temp_partial_cuda.cpp
delete mode 100644 src/USER-CUDA/compute_temp_partial_cuda.h
delete mode 100644 src/USER-CUDA/cuda.cpp
delete mode 100644 src/USER-CUDA/cuda_data.h
delete mode 100644 src/USER-CUDA/cuda_modify_flags.h
delete mode 100644 src/USER-CUDA/cuda_neigh_list.cpp
delete mode 100644 src/USER-CUDA/cuda_neigh_list.h
delete mode 100644 src/USER-CUDA/domain_cuda.cpp
delete mode 100644 src/USER-CUDA/domain_cuda.h
delete mode 100644 src/USER-CUDA/fft3d_cuda.cpp
delete mode 100644 src/USER-CUDA/fft3d_cuda.h
delete mode 100644 src/USER-CUDA/fft3d_wrap_cuda.cpp
delete mode 100644 src/USER-CUDA/fft3d_wrap_cuda.h
delete mode 100644 src/USER-CUDA/fix_addforce_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_addforce_cuda.h
delete mode 100644 src/USER-CUDA/fix_aveforce_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_aveforce_cuda.h
delete mode 100644 src/USER-CUDA/fix_enforce2d_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_enforce2d_cuda.h
delete mode 100644 src/USER-CUDA/fix_freeze_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_freeze_cuda.h
delete mode 100644 src/USER-CUDA/fix_gravity_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_gravity_cuda.h
delete mode 100644 src/USER-CUDA/fix_nh_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_nh_cuda.h
delete mode 100644 src/USER-CUDA/fix_npt_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_npt_cuda.h
delete mode 100644 src/USER-CUDA/fix_nve_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_nve_cuda.h
delete mode 100644 src/USER-CUDA/fix_nvt_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_nvt_cuda.h
delete mode 100644 src/USER-CUDA/fix_set_force_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_set_force_cuda.h
delete mode 100644 src/USER-CUDA/fix_shake_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_shake_cuda.h
delete mode 100644 src/USER-CUDA/fix_temp_berendsen_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_temp_berendsen_cuda.h
delete mode 100644 src/USER-CUDA/fix_temp_rescale_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_temp_rescale_cuda.h
delete mode 100644 src/USER-CUDA/fix_temp_rescale_limit_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_temp_rescale_limit_cuda.h
delete mode 100644 src/USER-CUDA/fix_viscous_cuda.cpp
delete mode 100644 src/USER-CUDA/fix_viscous_cuda.h
delete mode 100644 src/USER-CUDA/modify_cuda.cpp
delete mode 100644 src/USER-CUDA/modify_cuda.h
delete mode 100644 src/USER-CUDA/neigh_full_cuda.cpp
delete mode 100644 src/USER-CUDA/neighbor_cuda.cpp
delete mode 100644 src/USER-CUDA/neighbor_cuda.h
delete mode 100644 src/USER-CUDA/pair_born_coul_long_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_born_coul_long_cuda.h
delete mode 100644 src/USER-CUDA/pair_buck_coul_cut_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_buck_coul_cut_cuda.h
delete mode 100644 src/USER-CUDA/pair_buck_coul_long_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_buck_coul_long_cuda.h
delete mode 100644 src/USER-CUDA/pair_buck_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_buck_cuda.h
delete mode 100644 src/USER-CUDA/pair_eam_alloy_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_eam_alloy_cuda.h
delete mode 100644 src/USER-CUDA/pair_eam_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_eam_cuda.h
delete mode 100644 src/USER-CUDA/pair_eam_fs_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_eam_fs_cuda.h
delete mode 100644 src/USER-CUDA/pair_gran_hooke_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_gran_hooke_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj96_cut_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj96_cut_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_charmm_coul_long_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_charmm_coul_long_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_class2_coul_cut_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_class2_coul_cut_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_class2_coul_long_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_class2_coul_long_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_class2_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_class2_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_cut_coul_cut_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_cut_coul_cut_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_cut_coul_debye_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_cut_coul_debye_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_cut_coul_long_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_cut_coul_long_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_cut_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_cut_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_cut_experimental_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_cut_experimental_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_expand_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_expand_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_gromacs_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_gromacs_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_sdk_coul_long_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_sdk_coul_long_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_sdk_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_sdk_cuda.h
delete mode 100644 src/USER-CUDA/pair_lj_smooth_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_lj_smooth_cuda.h
delete mode 100644 src/USER-CUDA/pair_morse_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_morse_cuda.h
delete mode 100644 src/USER-CUDA/pair_sw_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_sw_cuda.h
delete mode 100644 src/USER-CUDA/pair_tersoff_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_tersoff_cuda.h
delete mode 100644 src/USER-CUDA/pair_tersoff_zbl_cuda.cpp
delete mode 100644 src/USER-CUDA/pair_tersoff_zbl_cuda.h
delete mode 100644 src/USER-CUDA/pppm_cuda.cpp
delete mode 100644 src/USER-CUDA/pppm_cuda.h
delete mode 100755 src/USER-CUDA/pppm_old.cpp
delete mode 100644 src/USER-CUDA/pppm_old.h
delete mode 100644 src/USER-CUDA/user_cuda.h
delete mode 100644 src/USER-CUDA/verlet_cuda.cpp
delete mode 100644 src/USER-CUDA/verlet_cuda.h
diff --git a/src/Makefile b/src/Makefile
index 16d3e5326c..85b212d539 100755
--- a/src/Makefile
+++ b/src/Makefile
@@ -46,7 +46,7 @@ PACKAGE = asphere body class2 colloid compress coreshell dipole gpu \
kokkos kspace manybody mc meam misc molecule mpiio opt peri poems \
python qeq reax replica rigid shock snap srd voronoi
-PACKUSER = user-atc user-awpmd user-cg-cmm user-colvars user-cuda \
+PACKUSER = user-atc user-awpmd user-cg-cmm user-colvars \
user-diffraction user-dpd user-drude user-eff user-fep user-h5md \
user-intel user-lb user-manifold user-mgpt \
user-misc user-molfile user-omp user-phonon user-qmmm user-qtb \
diff --git a/src/USER-CUDA/Install.sh b/src/USER-CUDA/Install.sh
deleted file mode 100755
index 96345160a6..0000000000
--- a/src/USER-CUDA/Install.sh
+++ /dev/null
@@ -1,202 +0,0 @@
-# Install/unInstall package files in LAMMPS
-# mode = 0/1/2 for uninstall/install/update
-
-mode=$1
-
-# arg1 = file, arg2 = file it depends on
-
-action () {
- if (test $mode = 0) then
- rm -f ../$1
- elif (! cmp -s $1 ../$1) then
- if (test -z "$2" || test -e ../$2) then
- cp $1 ..
- if (test $mode = 2) then
- echo " updating src/$1"
- fi
- fi
- elif (test -n "$2") then
- if (test ! -e ../$2) then
- rm -f ../$1
- fi
- fi
-}
-
-# force rebuild of files with LMP_USER_CUDA switch
-
-touch ../accelerator_cuda.h
-
-# list of files with optional dependencies
-
-action atom_vec_angle_cuda.cpp atom_vec_angle.cpp
-action atom_vec_angle_cuda.h atom_vec_angle.cpp
-action atom_vec_atomic_cuda.cpp
-action atom_vec_atomic_cuda.h
-action atom_vec_charge_cuda.cpp
-action atom_vec_charge_cuda.h
-action atom_vec_full_cuda.cpp atom_vec_full.cpp
-action atom_vec_full_cuda.h atom_vec_full.cpp
-action comm_cuda.cpp
-action comm_cuda.h
-action compute_pe_cuda.cpp
-action compute_pe_cuda.h
-action compute_pressure_cuda.cpp
-action compute_pressure_cuda.h
-action compute_temp_cuda.cpp
-action compute_temp_cuda.h
-action compute_temp_partial_cuda.cpp
-action compute_temp_partial_cuda.h
-action cuda.cpp
-action cuda_data.h
-action cuda_modify_flags.h
-action cuda_neigh_list.cpp
-action cuda_neigh_list.h
-action domain_cuda.cpp
-action domain_cuda.h
-action fft3d_cuda.cpp pppm.cpp
-action fft3d_cuda.h pppm.cpp
-action fft3d_wrap_cuda.cpp pppm.cpp
-action fft3d_wrap_cuda.h pppm.cpp
-action fix_addforce_cuda.cpp
-action fix_addforce_cuda.h
-action fix_aveforce_cuda.cpp
-action fix_aveforce_cuda.h
-action fix_enforce2d_cuda.cpp
-action fix_enforce2d_cuda.h
-action fix_freeze_cuda.cpp fix_freeze.cpp
-action fix_freeze_cuda.h fix_freeze.cpp
-action fix_gravity_cuda.cpp
-action fix_gravity_cuda.h
-action fix_nh_cuda.cpp
-action fix_nh_cuda.h
-action fix_npt_cuda.cpp
-action fix_npt_cuda.h
-action fix_nve_cuda.cpp
-action fix_nve_cuda.h
-action fix_nvt_cuda.cpp
-action fix_nvt_cuda.h
-action fix_set_force_cuda.cpp
-action fix_set_force_cuda.h
-action fix_shake_cuda.cpp
-action fix_shake_cuda.h
-action fix_temp_berendsen_cuda.cpp
-action fix_temp_berendsen_cuda.h
-action fix_temp_rescale_cuda.cpp
-action fix_temp_rescale_cuda.h
-action fix_temp_rescale_limit_cuda.cpp
-action fix_temp_rescale_limit_cuda.h
-action fix_viscous_cuda.cpp
-action fix_viscous_cuda.h
-action modify_cuda.cpp
-action modify_cuda.h
-action neigh_full_cuda.cpp
-action neighbor_cuda.cpp
-action neighbor_cuda.h
-action pair_born_coul_long_cuda.cpp pair_born_coul_long.cpp
-action pair_born_coul_long_cuda.h pair_born_coul_long.cpp
-action pair_buck_coul_cut_cuda.cpp
-action pair_buck_coul_cut_cuda.h
-action pair_buck_coul_long_cuda.cpp pair_buck_coul_long.cpp
-action pair_buck_coul_long_cuda.h pair_buck_coul_long.cpp
-action pair_buck_cuda.cpp
-action pair_buck_cuda.h
-action pair_eam_alloy_cuda.cpp pair_eam_alloy.cpp
-action pair_eam_alloy_cuda.h pair_eam_alloy.cpp
-action pair_eam_cuda.cpp pair_eam.cpp
-action pair_eam_cuda.h pair_eam.cpp
-action pair_eam_fs_cuda.cpp pair_eam_fs.cpp
-action pair_eam_fs_cuda.h pair_eam_fs.cpp
-action pair_gran_hooke_cuda.cpp pair_gran_hooke.cpp
-action pair_gran_hooke_cuda.h pair_gran_hooke.cpp
-action pair_lj96_cut_cuda.cpp
-action pair_lj96_cut_cuda.h
-action pair_lj_charmm_coul_charmm_cuda.cpp pair_lj_charmm_coul_charmm.cpp
-action pair_lj_charmm_coul_charmm_cuda.h pair_lj_charmm_coul_charmm.cpp
-action pair_lj_charmm_coul_charmm_implicit_cuda.cpp pair_lj_charmm_coul_charmm_implicit.cpp
-action pair_lj_charmm_coul_charmm_implicit_cuda.h pair_lj_charmm_coul_charmm_implicit.cpp
-action pair_lj_charmm_coul_long_cuda.cpp pair_lj_charmm_coul_long.cpp
-action pair_lj_charmm_coul_long_cuda.h pair_lj_charmm_coul_long.cpp
-action pair_lj_class2_coul_cut_cuda.cpp pair_lj_class2_coul_cut.cpp
-action pair_lj_class2_coul_cut_cuda.h pair_lj_class2_coul_cut.cpp
-action pair_lj_class2_coul_long_cuda.cpp pair_lj_class2_coul_long.cpp
-action pair_lj_class2_coul_long_cuda.h pair_lj_class2_coul_long.cpp
-action pair_lj_class2_cuda.cpp pair_lj_class2.cpp
-action pair_lj_class2_cuda.h pair_lj_class2.cpp
-action pair_lj_cut_coul_cut_cuda.cpp
-action pair_lj_cut_coul_cut_cuda.h
-action pair_lj_cut_coul_debye_cuda.cpp
-action pair_lj_cut_coul_debye_cuda.h
-action pair_lj_cut_coul_long_cuda.cpp pair_lj_cut_coul_long.cpp
-action pair_lj_cut_coul_long_cuda.h pair_lj_cut_coul_long.cpp
-action pair_lj_cut_cuda.cpp
-action pair_lj_cut_cuda.h
-action pair_lj_cut_experimental_cuda.cpp
-action pair_lj_cut_experimental_cuda.h
-action pair_lj_expand_cuda.cpp
-action pair_lj_expand_cuda.h
-action pair_lj_gromacs_coul_gromacs_cuda.cpp
-action pair_lj_gromacs_coul_gromacs_cuda.h
-action pair_lj_gromacs_cuda.cpp
-action pair_lj_gromacs_cuda.h
-action pair_lj_sdk_coul_long_cuda.cpp pair_lj_sdk_coul_long.cpp
-action pair_lj_sdk_coul_long_cuda.h pair_lj_sdk_coul_long.cpp
-action pair_lj_sdk_cuda.cpp pair_lj_sdk.cpp
-action pair_lj_sdk_cuda.h pair_lj_sdk.cpp
-action pair_lj_smooth_cuda.cpp
-action pair_lj_smooth_cuda.h
-action pair_morse_cuda.cpp
-action pair_morse_cuda.h
-action pair_sw_cuda.cpp pair_sw.cpp
-action pair_sw_cuda.h pair_sw.cpp
-action pair_tersoff_cuda.cpp pair_tersoff.cpp
-action pair_tersoff_cuda.h pair_tersoff.cpp
-action pair_tersoff_zbl_cuda.cpp pair_tersoff_zbl.cpp
-action pair_tersoff_zbl_cuda.h pair_tersoff_zbl.cpp
-action pppm_cuda.cpp pppm.cpp
-action pppm_cuda.h pppm.cpp
-action pppm_old.cpp pppm.cpp
-action pppm_old.h pppm.cpp
-action user_cuda.h
-action verlet_cuda.cpp
-action verlet_cuda.h
-
-# edit 2 Makefile.package files to include/exclude package info
-
-if (test $1 = 1) then
-
- if (test -e ../Makefile.package) then
- sed -i -e 's/[^ \t]*cuda[^ \t]* //g' ../Makefile.package
- sed -i -e 's/[^ \t]*CUDA[^ \t]* //g' ../Makefile.package
- sed -i -e 's|^PKG_INC =[ \t]*|&-I..\/..\/lib\/cuda -DLMP_USER_CUDA |' ../Makefile.package
- sed -i -e 's|^PKG_PATH =[ \t]*|&-L..\/..\/lib\/cuda |' ../Makefile.package
- sed -i -e 's|^PKG_LIB =[ \t]*|&-llammpscuda |' ../Makefile.package
- sed -i -e 's|^PKG_SYSINC =[ \t]*|&$(user-cuda_SYSINC) |' ../Makefile.package
- sed -i -e 's|^PKG_SYSLIB =[ \t]*|&$(user-cuda_SYSLIB) |' ../Makefile.package
- sed -i -e 's|^PKG_SYSPATH =[ \t]*|&$(user-cuda_SYSPATH) |' ../Makefile.package
- fi
-
- if (test -e ../Makefile.package.settings) then
- sed -i -e '/^include.*cuda.*$/d' ../Makefile.package.settings
- # multiline form needed for BSD sed on Macs
- sed -i -e '4 i \
-include ..\/..\/lib\/cuda\/Makefile.lammps
-' ../Makefile.package.settings
-
- fi
-
-elif (test $1 = 0) then
- # need to delete a bunch of depenency files because they indirectly depend on user_cuda.h
- for f in input.d output.d pair.d fix_omp.d
- do \
- rm -f ../Obj_*/$f
- done
- if (test -e ../Makefile.package) then
- sed -i -e 's/[^ \t]*cuda[^ \t]* //g' ../Makefile.package
- sed -i -e 's/[^ \t]*CUDA[^ \t]* //g' ../Makefile.package
- fi
-
- if (test -e ../Makefile.package.settings) then
- sed -i -e '/^include.*cuda.*$/d' ../Makefile.package.settings
- fi
-
-fi
diff --git a/src/USER-CUDA/README b/src/USER-CUDA/README
deleted file mode 100644
index b9d2c07f8a..0000000000
--- a/src/USER-CUDA/README
+++ /dev/null
@@ -1,21 +0,0 @@
-This package provides acceleration of various LAMMPS pair styles, fix
-styles, compute styles, and long-range Coulombics via PPPM for NVIDIA
-GPUs.
-
-See this section of the manual to get started:
-
-doc/Section_accelerate.html, sub-section 5.4
-
-There are example scripts for using this package in
-examples/USER/cuda.
-
-This package uses an external library in lib/cuda which must be
-compiled before making LAMMPS. See the lib/cuda/README file and the
-LAMMPS manual for information on building LAMMPS with external
-libraries. The settings in the Makefile.lammps file in that directory
-must be correct for LAMMPS to build correctly with this package
-installed.
-
-The person who created this package is Christian Trott at the
-University of Technology Ilmenau, Germany (christian.trott at
-tu-ilmenau.de). Contact him directly if you have questions.
diff --git a/src/USER-CUDA/atom_vec_angle_cuda.cpp b/src/USER-CUDA/atom_vec_angle_cuda.cpp
deleted file mode 100644
index c393d58824..0000000000
--- a/src/USER-CUDA/atom_vec_angle_cuda.cpp
+++ /dev/null
@@ -1,467 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "atom_vec_angle_cuda.h"
-#include "comm_cuda_cu.h"
-#include "atom_vec_angle_cuda_cu.h"
-#include "atom.h"
-#include "domain.h"
-#include "modify.h"
-#include "fix.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "universe.h"
-#include "comm.h"
-
-using namespace LAMMPS_NS;
-
-#define BUFFACTOR 1.5
-#define BUFEXTRA 1000
-#define NCUDAEXCHANGE 12 //nextra x y z vx vy vz tag type mask image molecule
-
-#define BUF_CFLOAT double
-/* ---------------------------------------------------------------------- */
-
-AtomVecAngleCuda::AtomVecAngleCuda(LAMMPS *lmp) : AtomVecAngle(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- maxsend=0;
- cudable=true;
- cuda_init_done=false;
- max_nsend=0;
- cu_copylist=NULL;
- copylist=NULL;
- copylist2=NULL;
-}
-
-void AtomVecAngleCuda::grow_copylist(int new_max_nsend)
-{
- max_nsend=new_max_nsend;
- delete cu_copylist;
- delete [] copylist2;
- if(copylist) CudaWrapper_FreePinnedHostData((void*) copylist);
- copylist = (int*) CudaWrapper_AllocPinnedHostData(max_nsend*sizeof(int),false);
- copylist2 = new int[max_nsend];
- cu_copylist = new cCudaData<int, int, xx > (copylist, max_nsend);
-}
-
-void AtomVecAngleCuda::grow_send(int n,double** buf_send,int flag) //need to be able to grow the comm send_buffer since the array sahll be copied from the gpu in whole
-{
- int old_maxsend=*maxsend+BUFEXTRA;
- *maxsend = static_cast<int> (BUFFACTOR * n);
- if (flag)
- {
- if(cuda->pinned)
- {
- double* tmp = new double[old_maxsend];
- memcpy((void*) tmp,(void*) *buf_send,old_maxsend*sizeof(double));
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- memcpy(*buf_send,tmp,old_maxsend*sizeof(double));
- delete [] tmp;
- }
- else
- {
- *buf_send = (double *)
- memory->srealloc(*buf_send,(*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
- else {
- if(cuda->pinned)
- {
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- }
- else
- {
- memory->sfree(*buf_send);
- *buf_send = (double *) memory->smalloc((*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
-}
-
-void AtomVecAngleCuda::grow_both(int n)
-{
- if(cuda->finished_setup)
- {
- cuda->cu_special->upload();
- cuda->cu_nspecial->upload();
- cuda->downloadAll();
- }
- AtomVecAngle::grow(n);
- if(cuda->finished_setup)
- {
- cuda->checkResize();
- cuda->uploadAll();
- }
-}
-
-int AtomVecAngleCuda::pack_comm(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAngle::pack_comm(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackComm(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-
-int AtomVecAngleCuda::pack_comm_vel(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAngle::pack_comm_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackCommVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecAngleCuda::unpack_comm(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAngle::unpack_comm(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackComm(&cuda->shared_data,n,first,(void*)buf);
-}
-
-void AtomVecAngleCuda::unpack_comm_vel(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAngle::unpack_comm_vel(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackCommVel(&cuda->shared_data,n,first,(void*)buf);
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecAngleCuda::pack_reverse(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only forces are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAngle::pack_reverse(n,first,buf);
-
- int i,m,last;
- cuda->cu_f->download();
- m = 0;
- last = first + n;
- for (i = first; i < last; i++) {
- buf[m++] = f[i][0];
- buf[m++] = f[i][1];
- buf[m++] = f[i][2];
- }
- cuda->cu_f->upload();
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecAngleCuda::unpack_reverse(int n, int *list, double *buf)//usually this should not be called since comm->communicate handles the communication if only forces are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAngle::unpack_reverse(n,list,buf); return;}
-
- int i,j,m;
-
- m = 0;
- cuda->cu_f->download();
- for (i = 0; i < n; i++) {
- j = list[i];
- f[j][0] += buf[m++];
- f[j][1] += buf[m++];
- f[j][2] += buf[m++];
- }
- cuda->cu_f->upload();
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecAngleCuda::pack_border(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAngle::pack_border(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecAngleCuda_PackBorder(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- return m;
-}
-
-int AtomVecAngleCuda::pack_border_vel(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAngle::pack_border_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecAngleCuda_PackBorderVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
-
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecAngleCuda::unpack_border(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAngle::unpack_border(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax) //ensure there is enough space on device to unpack data
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecAngleCuda_UnpackBorder(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-}
-
-void AtomVecAngleCuda::unpack_border_vel(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAngle::unpack_border_vel(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax) //ensure there is enough space on device to unpack data
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecAngleCuda_UnpackBorderVel(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-}
-
-/* ----------------------------------------------------------------------
- pack data for atom I for sending to another proc
- xyz must be 1st 3 values, so comm::exchange() can test on them
-------------------------------------------------------------------------- */
-
-
-int AtomVecAngleCuda::pack_exchange(int dim, double *buf)
-{
- if(cuda->oncpu)
- return AtomVecAngle::pack_exchange(dim,buf);
-
- if(not cuda_init_done||domain->box_change)
- {
- Cuda_AtomVecAngleCuda_Init(&cuda->shared_data);
- cuda_init_done=true;
- }
- double** buf_pointer=(double**) buf;
- if(*maxsend<atom->nghost || *buf_pointer==NULL)
- {
- grow_send(atom->nghost>*maxsend?atom->nghost:*maxsend,buf_pointer,0);
- *maxsend=atom->nghost>*maxsend?atom->nghost:*maxsend;
- }
-
- if(max_nsend==0) grow_copylist(200);
-
- int nsend_atoms = Cuda_AtomVecAngleCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
-
- if(nsend_atoms>max_nsend) grow_copylist(nsend_atoms+100);
- if(nsend_atoms*NCUDAEXCHANGE>*maxsend)
- {
- grow_send((int) (nsend_atoms+100)*NCUDAEXCHANGE,buf_pointer,0);
- Cuda_AtomVecAngleCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
- }
-
- int nlocal=atom->nlocal-nsend_atoms;
-
- for(int i=0;i<nsend_atoms;i++) copylist2[i]=1;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i>=nlocal) copylist2[i-nlocal]=-1;
- }
-
- int actpos=0;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i<nlocal)
- {
- while(copylist2[actpos]==-1) actpos++;
- copylist[j-1]=nlocal+actpos;
- actpos++;
- }
- }
- cu_copylist->upload();
-
- cuda->shared_data.atom.nlocal=nlocal;
-
- int m = Cuda_AtomVecAngleCuda_PackExchange(&cuda->shared_data,nsend_atoms,*buf_pointer,cu_copylist->dev_data());
-
- my_times time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
-
- double* buf_p=*buf_pointer;
- for(int j=0;j<nsend_atoms;j++)
- {
- int i=static_cast <int> (buf_p[j+1]);
- int nextra=0;
- int k;
- buf_p[m++] = num_bond[i];
- for (k = 0; k < num_bond[i]; k++) {
- buf_p[m++] = bond_type[i][k];
- buf_p[m++] = bond_atom[i][k];
- }
- nextra+=2*num_bond[i]+1;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- buf_p[m++] = num_angle[i];
- for (k = 0; k < num_angle[i]; k++) {
- buf_p[m++] = angle_type[i][k];
- buf_p[m++] = angle_atom1[i][k];
- buf_p[m++] = angle_atom2[i][k];
- buf_p[m++] = angle_atom3[i][k];
- }
- nextra+=4*num_angle[i]+1;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- buf_p[m++] = nspecial[i][0];
- buf_p[m++] = nspecial[i][1];
- buf_p[m++] = nspecial[i][2];
- for (k = 0; k < nspecial[i][2]; k++) buf_p[m++] = special[i][k];
- nextra+=nspecial[i][2]+3;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- if (atom->nextra_grow)
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++)
- {
- int dm= modify->fix[atom->extra_grow[iextra]]->pack_exchange(i,&buf_p[m]);
- m+=dm;
- nextra+=dm;
- if(i<nlocal)modify->fix[atom->extra_grow[iextra]]->copy_arrays(copylist[j],i,1);
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
- }
-
- if(i<nlocal)AtomVecAngle::copy(copylist[j],i,1);
- (*buf_pointer)[j+1] = nextra;
- }
-
- my_gettime(CLOCK_REALTIME,&time2);
- cuda->shared_data.cuda_timings.comm_exchange_cpu_pack+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
-
- (*buf_pointer)[0] = nsend_atoms;
- atom->nlocal-=nsend_atoms;
- cuda->shared_data.atom.update_nlocal=2;
- //printf("End Pack Exchange\n");
- if(m==1) return 0;
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecAngleCuda::unpack_exchange(double *buf)
-{
-// printf("Begin UnPack Exchange\n");
- if(cuda->oncpu)
- return AtomVecAngle::unpack_exchange(buf);
-
- int dim=cuda->shared_data.exchange_dim;
- if(domain->box_change)
- Cuda_AtomVecAngleCuda_Init(&cuda->shared_data);
-
- int mfirst=0;
- for(int pi=0;pi<(comm->procgrid[dim]>2?2:1);pi++)
- {
- int nlocal = atom->nlocal;
- int nsend_atoms=static_cast<int> (buf[0]);
- if(nsend_atoms>max_nsend) grow_copylist(nsend_atoms+100);
-
- if (nlocal+nsend_atoms+atom->nghost>=atom->nmax) grow_both(nlocal+nsend_atoms*2+atom->nghost); //ensure there is enough space on device to unpack data
- int naccept = Cuda_AtomVecAngleCuda_UnpackExchange(&cuda->shared_data,nsend_atoms,buf,cu_copylist->dev_data());
- cu_copylist->download();
- int m = nsend_atoms*NCUDAEXCHANGE + 1;
- nlocal+=naccept;
-
- my_times time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
-
- for(int j=0;j<nsend_atoms;j++)
- {
- if(copylist[j]>-1)
- {
- int k;
- int i=copylist[j];
- num_bond[i] = static_cast<int> (buf[m++]);
- for (k = 0; k < num_bond[i]; k++) {
- bond_type[i][k] = static_cast<int> (buf[m++]);
- bond_atom[i][k] = static_cast<int> (buf[m++]);
- }
-
- num_angle[i] = static_cast<int> (buf[m++]);
- for (k = 0; k < num_angle[i]; k++) {
- angle_type[i][k] = static_cast<int> (buf[m++]);
- angle_atom1[i][k] = static_cast<int> (buf[m++]);
- angle_atom2[i][k] = static_cast<int> (buf[m++]);
- angle_atom3[i][k] = static_cast<int> (buf[m++]);
- }
-
- nspecial[i][0] = static_cast<int> (buf[m++]);
- nspecial[i][1] = static_cast<int> (buf[m++]);
- nspecial[i][2] = static_cast<int> (buf[m++]);
- for (k = 0; k < nspecial[i][2]; k++)
- special[i][k] = static_cast<int> (buf[m++]);
-
- if (atom->nextra_grow)
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++)
- m += modify->fix[atom->extra_grow[iextra]]->
- unpack_exchange(i,&buf[m]);
-
- }
- else
- m+=static_cast <int> (buf[j+1]);
- }
-
- my_gettime(CLOCK_REALTIME,&time2);
- cuda->shared_data.cuda_timings.comm_exchange_cpu_pack+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
-
- cuda->shared_data.atom.nlocal=nlocal;
- cuda->shared_data.atom.update_nlocal=2;
- atom->nlocal=nlocal;
- mfirst+=m;
- buf=&buf[m];
- }
- return mfirst;
-}
diff --git a/src/USER-CUDA/atom_vec_angle_cuda.h b/src/USER-CUDA/atom_vec_angle_cuda.h
deleted file mode 100644
index 13913da1c0..0000000000
--- a/src/USER-CUDA/atom_vec_angle_cuda.h
+++ /dev/null
@@ -1,69 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef ATOM_CLASS
-
-AtomStyle(angle/cuda,AtomVecAngleCuda)
-
-#else
-
-#ifndef LMP_ATOM_VEC_ANGLE_CUDA_H
-#define LMP_ATOM_VEC_ANGLE_CUDA_H
-
-#include "atom_vec_angle.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class AtomVecAngleCuda : public AtomVecAngle {
- public:
- AtomVecAngleCuda(class LAMMPS *);
- virtual ~AtomVecAngleCuda() {}
- void grow_copylist(int n);
- void grow_send(int n,double** buf_send,int flag);
- void grow_both(int n);
- int pack_comm(int, int *, double *, int, int *);
- int pack_comm_vel(int, int *, double *, int, int *);
- void unpack_comm(int, int, double *);
- void unpack_comm_vel(int, int, double *);
- int pack_reverse(int, int, double *);
- void unpack_reverse(int, int *, double *);
- int pack_border(int, int *, double *, int, int *);
- int pack_border_vel(int, int *, double *, int, int *);
- void unpack_border(int, int, double *);
- void unpack_border_vel(int, int, double *);
- int pack_exchange(int, double *);
- int unpack_exchange(double *);
- private:
- class Cuda *cuda;
- bool cuda_init_done;
- int* copylist;
- int* copylist2;
- cCudaData<int, int, xx >* cu_copylist;
- int max_nsend;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/atom_vec_atomic_cuda.cpp b/src/USER-CUDA/atom_vec_atomic_cuda.cpp
deleted file mode 100644
index c54f7d3127..0000000000
--- a/src/USER-CUDA/atom_vec_atomic_cuda.cpp
+++ /dev/null
@@ -1,394 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "atom_vec_atomic_cuda.h"
-#include "comm_cuda_cu.h"
-#include "atom_vec_atomic_cuda_cu.h"
-#include "atom.h"
-#include "domain.h"
-#include "modify.h"
-#include "fix.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "comm.h"
-
-using namespace LAMMPS_NS;
-
-#define BUFFACTOR 1.5
-#define BUFEXTRA 1000
-#define NCUDAEXCHANGE 11 //nextra x y z vx vy vz tag type mask image
-
-
-#define BUF_CFLOAT double
-/* ---------------------------------------------------------------------- */
-
-AtomVecAtomicCuda::AtomVecAtomicCuda(LAMMPS *lmp) : AtomVecAtomic(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- maxsend=0;
- cudable=true;
- cuda_init_done=false;
- max_nsend=0;
- cu_copylist=NULL;
- copylist=NULL;
- copylist2=NULL;
-}
-
-void AtomVecAtomicCuda::grow_copylist(int new_max_nsend)
-{
- max_nsend=new_max_nsend;
- delete cu_copylist;
- delete [] copylist2;
- if(copylist) CudaWrapper_FreePinnedHostData((void*) copylist);
- copylist = (int*) CudaWrapper_AllocPinnedHostData(max_nsend*sizeof(int),false);
- copylist2 = new int[max_nsend];
- cu_copylist = new cCudaData<int, int, xx > (copylist, max_nsend);
-}
-
-void AtomVecAtomicCuda::grow_send(int n,double** buf_send,int flag)
-{
- int old_maxsend=*maxsend+BUFEXTRA;
- *maxsend = static_cast<int> (BUFFACTOR * n);
- if (flag)
- {
- if(cuda->pinned)
- {
- double* tmp = new double[old_maxsend];
- memcpy((void*) tmp,(void*) *buf_send,old_maxsend*sizeof(double));
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- memcpy(*buf_send,tmp,old_maxsend*sizeof(double));
- delete [] tmp;
- }
- else
- {
- *buf_send = (double *)
- memory->srealloc(*buf_send,(*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
- else {
- if(cuda->pinned)
- {
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- }
- else
- {
- memory->sfree(*buf_send);
- *buf_send = (double *) memory->smalloc((*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
-}
-
-void AtomVecAtomicCuda::grow_both(int n)
-{
- if(cuda->finished_setup)
- cuda->downloadAll();
- AtomVecAtomic::grow(n);
- if(cuda->finished_setup)
- {
- cuda->checkResize();
- cuda->uploadAll();
- }
-}
-
-int AtomVecAtomicCuda::pack_comm(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAtomic::pack_comm(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackComm(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-
-int AtomVecAtomicCuda::pack_comm_vel(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAtomic::pack_comm_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackCommVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-/* ---------------------------------------------------------------------- */
-
-void AtomVecAtomicCuda::unpack_comm(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAtomic::unpack_comm(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackComm(&cuda->shared_data,n,first,(void*)buf);
-}
-
-void AtomVecAtomicCuda::unpack_comm_vel(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAtomic::unpack_comm_vel(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackCommVel(&cuda->shared_data,n,first,(void*)buf);
-}
-/* ---------------------------------------------------------------------- */
-
-int AtomVecAtomicCuda::pack_reverse(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAtomic::pack_reverse(n,first,buf);
-
- int i,m,last;
-
- m = 0;
- last = first + n;
- for (i = first; i < last; i++) {
- buf[m++] = f[i][0];
- buf[m++] = f[i][1];
- buf[m++] = f[i][2];
- }
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecAtomicCuda::unpack_reverse(int n, int *list, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAtomic::unpack_reverse(n,list,buf); return;}
-
- int i,j,m;
-
- m = 0;
- for (i = 0; i < n; i++) {
- j = list[i];
- f[j][0] += buf[m++];
- f[j][1] += buf[m++];
- f[j][2] += buf[m++];
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecAtomicCuda::pack_border(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAtomic::pack_border(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecAtomicCuda_PackBorder(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
-
- return m;
-}
-
-int AtomVecAtomicCuda::pack_border_vel(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecAtomic::pack_border_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecAtomicCuda_PackBorderVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
-
- return m;
-}
-/* ---------------------------------------------------------------------- */
-
-void AtomVecAtomicCuda::unpack_border(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAtomic::unpack_border(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax)
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecAtomicCuda_UnpackBorder(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-
-}
-
-void AtomVecAtomicCuda::unpack_border_vel(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecAtomic::unpack_border_vel(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax)
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecAtomicCuda_UnpackBorderVel(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-}
-/* ----------------------------------------------------------------------
- pack data for atom I for sending to another proc
- xyz must be 1st 3 values, so comm::exchange() can test on them
-------------------------------------------------------------------------- */
-
-
-int AtomVecAtomicCuda::pack_exchange(int dim, double *buf)
-{
- if(cuda->oncpu)
- return AtomVecAtomic::pack_exchange(dim,buf);
-
- if(not cuda_init_done||domain->box_change)
- {
- Cuda_AtomVecAtomicCuda_Init(&cuda->shared_data);
- cuda_init_done=true;
- }
- double** buf_pointer=(double**) buf;
- if(*maxsend<atom->nghost || *buf_pointer==NULL)
- {
- grow_send(atom->nghost>*maxsend?atom->nghost:*maxsend,buf_pointer,0);
- *maxsend=atom->nghost>*maxsend?atom->nghost:*maxsend;
- }
-
- if(max_nsend==0) grow_copylist(200);
-
- int nsend_atoms = Cuda_AtomVecAtomicCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
-
- if(nsend_atoms>max_nsend) {grow_copylist(nsend_atoms+100);}
- if(nsend_atoms*NCUDAEXCHANGE>*maxsend)
- {
- grow_send((int) (nsend_atoms+100)*NCUDAEXCHANGE,buf_pointer,0);
- Cuda_AtomVecAtomicCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
- }
-
- int nlocal=atom->nlocal-nsend_atoms;
-
- for(int i=0;i<nsend_atoms;i++) copylist2[i]=1;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i>=nlocal) copylist2[i-nlocal]=-1;
- }
-
- int actpos=0;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i<nlocal)
- {
- while(copylist2[actpos]==-1) actpos++;
- copylist[j-1]=nlocal+actpos;
- actpos++;
- }
- }
- cu_copylist->upload();
-
- cuda->shared_data.atom.nlocal=nlocal;
-
- int m = Cuda_AtomVecAtomicCuda_PackExchange(&cuda->shared_data,nsend_atoms,*buf_pointer,cu_copylist->dev_data());
- if (atom->nextra_grow)
- for(int j=0;j<nsend_atoms;j++)
- {
- int i=static_cast <int> ((*buf_pointer)[j+1]);
- int nextra=0;
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++) {
-
- int dm = modify->fix[atom->extra_grow[iextra]]->pack_exchange(i,&((*buf_pointer)[m]));
- m+=dm;
- nextra+=dm;
- if(i<nlocal)modify->fix[atom->extra_grow[iextra]]->copy_arrays(copylist[j],i,1);
- if(m>*maxsend) grow_send(m,buf_pointer,1);
- }
- (*buf_pointer)[j+1] = nextra;
-
- }
-
- (*buf_pointer)[0] = nsend_atoms;
- atom->nlocal-=nsend_atoms;
- cuda->shared_data.atom.update_nlocal=2;
-
- if(m==1) return 0;//m is at least 1 in cuda since buf[0] contains number of atoms
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecAtomicCuda::unpack_exchange(double *buf)
-{
- //printf("Unpack Begin\n");
- if(cuda->oncpu)
- return AtomVecAtomic::unpack_exchange(buf);
-
- int dim=cuda->shared_data.exchange_dim;
- if(domain->box_change)
- Cuda_AtomVecAtomicCuda_Init(&cuda->shared_data);
-
- int mfirst=0;
- for(int pi=0;pi<(comm->procgrid[dim]>2?2:1);pi++)
- {
- int nlocal = atom->nlocal;
-
- int nsend_atoms=static_cast<int> (buf[0]);
- if(nsend_atoms>max_nsend) grow_copylist(nsend_atoms+100);
-
- if (nlocal+nsend_atoms+atom->nghost>=atom->nmax) grow_both(nlocal+nsend_atoms*2+atom->nghost);
- int naccept = Cuda_AtomVecAtomicCuda_UnpackExchange(&cuda->shared_data,nsend_atoms,buf,cu_copylist->dev_data());
- cu_copylist->download();
- int m = nsend_atoms*NCUDAEXCHANGE + 1;
- nlocal+=naccept;
- if (atom->nextra_grow)
- for(int j=0;j<nsend_atoms;j++)
- {
- if(copylist[j]>-1)
- {
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++)
- m += modify->fix[atom->extra_grow[iextra]]->
- unpack_exchange(copylist[j],&buf[m]);
- }
- else
- {
- m+=static_cast <int> (buf[j+1]);
- }
- }
- cuda->shared_data.atom.nlocal=nlocal;
- if(atom->nlocal!=nlocal)
- cuda->shared_data.atom.update_nlocal=2;
- atom->nlocal=nlocal;
- mfirst+=m;
- buf=&buf[m];
- }
- return mfirst;
-}
diff --git a/src/USER-CUDA/atom_vec_atomic_cuda.h b/src/USER-CUDA/atom_vec_atomic_cuda.h
deleted file mode 100644
index dabbd9a215..0000000000
--- a/src/USER-CUDA/atom_vec_atomic_cuda.h
+++ /dev/null
@@ -1,81 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-#ifdef ATOM_CLASS
-
-AtomStyle(atomic/cuda,AtomVecAtomicCuda)
-
-#else
-
-#ifndef LMP_ATOM_VEC_ATOMIC_CUDA_H
-#define LMP_ATOM_VEC_ATOMIC_CUDA_H
-
-#include "atom_vec_atomic.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class AtomVecAtomicCuda : public AtomVecAtomic {
- public:
- AtomVecAtomicCuda(class LAMMPS *);
- virtual ~AtomVecAtomicCuda() {}
- void grow_copylist(int n);
- void grow_send(int n,double** buf_send,int flag);
- void grow_both(int n);
- int pack_comm(int, int *, double *, int, int *);
- int pack_comm_vel(int, int *, double *, int, int *);
- void unpack_comm(int, int, double *);
- void unpack_comm_vel(int, int, double *);
- int pack_reverse(int, int, double *);
- void unpack_reverse(int, int *, double *);
- int pack_border(int, int *, double *, int, int *);
- int pack_border_vel(int, int *, double *, int, int *);
- void unpack_border(int, int, double *);
- void unpack_border_vel(int, int, double *);
- int pack_exchange(int, double *);
- int unpack_exchange(double *);
- private:
- class Cuda *cuda;
- bool cuda_init_done;
- int* copylist;
- int* copylist2;
- cCudaData<int, int, xx >* cu_copylist;
- int max_nsend;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/atom_vec_charge_cuda.cpp b/src/USER-CUDA/atom_vec_charge_cuda.cpp
deleted file mode 100644
index 07140f3e84..0000000000
--- a/src/USER-CUDA/atom_vec_charge_cuda.cpp
+++ /dev/null
@@ -1,394 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "atom_vec_charge_cuda.h"
-#include "comm_cuda_cu.h"
-#include "atom_vec_charge_cuda_cu.h"
-#include "atom.h"
-#include "domain.h"
-#include "modify.h"
-#include "fix.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "comm.h"
-
-using namespace LAMMPS_NS;
-
-#define BUFFACTOR 1.5
-#define BUFEXTRA 1000
-#define NCUDAEXCHANGE 12 //nextra x y z vx vy vz tag type mask image q
-
-#define BUF_CFLOAT double
-/* ---------------------------------------------------------------------- */
-
-AtomVecChargeCuda::AtomVecChargeCuda(LAMMPS *lmp) : AtomVecCharge(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- maxsend=0;
- cudable=true;
- cuda_init_done=false;
- max_nsend=0;
- cu_copylist=NULL;
- copylist=NULL;
- copylist2=NULL;
-}
-
-void AtomVecChargeCuda::grow_copylist(int new_max_nsend)
-{
- max_nsend=new_max_nsend;
- delete cu_copylist;
- delete [] copylist2;
- if(copylist) CudaWrapper_FreePinnedHostData((void*) copylist);
- copylist = (int*) CudaWrapper_AllocPinnedHostData(max_nsend*sizeof(int),false);
- copylist2 = new int[max_nsend];
- cu_copylist = new cCudaData<int, int, xx > (copylist, max_nsend);
-}
-
-void AtomVecChargeCuda::grow_send(int n,double** buf_send,int flag) //need to be able to grow the comm send_buffer since the array sahll be copied from the gpu in whole
-{
- int old_maxsend=*maxsend+BUFEXTRA;
- *maxsend = static_cast<int> (BUFFACTOR * n);
- if (flag)
- {
- if(cuda->pinned)
- {
- double* tmp = new double[old_maxsend];
- memcpy((void*) tmp,(void*) *buf_send,old_maxsend*sizeof(double));
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- memcpy(*buf_send,tmp,old_maxsend*sizeof(double));
- delete [] tmp;
- }
- else
- {
- *buf_send = (double *)
- memory->srealloc(*buf_send,(*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
- else {
- if(cuda->pinned)
- {
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- }
- else
- {
- memory->sfree(*buf_send);
- *buf_send = (double *) memory->smalloc((*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
-}
-
-void AtomVecChargeCuda::grow_both(int n)
-{
- if(cuda->finished_setup)
- cuda->downloadAll();
- AtomVecCharge::grow(n);
- if(cuda->finished_setup)
- {
- cuda->checkResize();
- cuda->uploadAll();
- }
-}
-
-int AtomVecChargeCuda::pack_comm(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecCharge::pack_comm(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackComm(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-
-int AtomVecChargeCuda::pack_comm_vel(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecCharge::pack_comm_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackCommVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecChargeCuda::unpack_comm(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecCharge::unpack_comm(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackComm(&cuda->shared_data,n,first,(void*)buf);
-}
-
-void AtomVecChargeCuda::unpack_comm_vel(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecCharge::unpack_comm_vel(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackCommVel(&cuda->shared_data,n,first,(void*)buf);
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecChargeCuda::pack_reverse(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only forces are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecCharge::pack_reverse(n,first,buf);
-
- int i,m,last;
- cuda->cu_f->download();
- m = 0;
- last = first + n;
- for (i = first; i < last; i++) {
- buf[m++] = f[i][0];
- buf[m++] = f[i][1];
- buf[m++] = f[i][2];
- }
- cuda->cu_f->upload();
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecChargeCuda::unpack_reverse(int n, int *list, double *buf)//usually this should not be called since comm->communicate handles the communication if only forces are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecCharge::unpack_reverse(n,list,buf); return;}
-
- int i,j,m;
-
- m = 0;
- cuda->cu_f->download();
- for (i = 0; i < n; i++) {
- j = list[i];
- f[j][0] += buf[m++];
- f[j][1] += buf[m++];
- f[j][2] += buf[m++];
- }
- cuda->cu_f->upload();
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecChargeCuda::pack_border(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecCharge::pack_border(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecChargeCuda_PackBorder(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
-
- return m;
-}
-
-int AtomVecChargeCuda::pack_border_vel(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecCharge::pack_border_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecChargeCuda_PackBorderVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
-
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecChargeCuda::unpack_border(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecCharge::unpack_border(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax) //ensure there is enough space on device to unpack data
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecChargeCuda_UnpackBorder(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-}
-
-void AtomVecChargeCuda::unpack_border_vel(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecCharge::unpack_border_vel(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax) //ensure there is enough space on device to unpack data
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecChargeCuda_UnpackBorderVel(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-}
-
-/* ----------------------------------------------------------------------
- pack data for atom I for sending to another proc
- xyz must be 1st 3 values, so comm::exchange() can test on them
-------------------------------------------------------------------------- */
-
-
-int AtomVecChargeCuda::pack_exchange(int dim, double *buf)
-{
- if(cuda->oncpu)
- return AtomVecCharge::pack_exchange(dim,buf);
-
- if(not cuda_init_done||domain->box_change)
- {
- Cuda_AtomVecChargeCuda_Init(&cuda->shared_data);
- cuda_init_done=true;
- }
- double** buf_pointer=(double**) buf;
- if(*maxsend<atom->nghost || *buf_pointer==NULL)
- {
- grow_send(atom->nghost>*maxsend?atom->nghost:*maxsend,buf_pointer,0);
- *maxsend=atom->nghost>*maxsend?atom->nghost:*maxsend;
- }
-
- if(max_nsend==0) grow_copylist(200);
-
- int nsend_atoms = Cuda_AtomVecChargeCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
-
- if(nsend_atoms>max_nsend) grow_copylist(nsend_atoms+100);
- if(nsend_atoms*NCUDAEXCHANGE>*maxsend)
- {
- grow_send((int) (nsend_atoms+100)*NCUDAEXCHANGE,buf_pointer,0);
- Cuda_AtomVecChargeCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
- }
-
- int nlocal=atom->nlocal-nsend_atoms;
-
- for(int i=0;i<nsend_atoms;i++) copylist2[i]=1;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i>=nlocal) copylist2[i-nlocal]=-1;
- }
-
- int actpos=0;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i<nlocal)
- {
- while(copylist2[actpos]==-1) actpos++;
- copylist[j-1]=nlocal+actpos;
- actpos++;
- }
- }
- cu_copylist->upload();
-
- cuda->shared_data.atom.nlocal=nlocal;
-
- int m = Cuda_AtomVecChargeCuda_PackExchange(&cuda->shared_data,nsend_atoms,*buf_pointer,cu_copylist->dev_data());
-
- if (atom->nextra_grow)
- for(int j=0;j<nsend_atoms;j++)
- {
- int i=static_cast <int> ((*buf_pointer)[j+1]);
- int nextra=0;
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++) {
-
- int dm = modify->fix[atom->extra_grow[iextra]]->pack_exchange(i,&((*buf_pointer)[m]));
- m+=dm;
- nextra+=dm;
- if(i<nlocal)modify->fix[atom->extra_grow[iextra]]->copy_arrays(copylist[j],i,1);
- if(m>*maxsend) grow_send(m,buf_pointer,1);
- }
- (*buf_pointer)[j+1] = nextra;
- }
-
- (*buf_pointer)[0] = nsend_atoms;
- atom->nlocal-=nsend_atoms;
- cuda->shared_data.atom.update_nlocal=2;
-
- if(m==1) return 0;//m is at least 1 in cuda since buf[0] contains number of atoms
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecChargeCuda::unpack_exchange(double *buf)
-{
- if(cuda->oncpu)
- return AtomVecCharge::unpack_exchange(buf);
-
- int dim=cuda->shared_data.exchange_dim;
- if(domain->box_change)
- Cuda_AtomVecChargeCuda_Init(&cuda->shared_data);
-
- int mfirst=0;
- for(int pi=0;pi<(comm->procgrid[dim]>2?2:1);pi++)
- {
- int nlocal = atom->nlocal;
- int nsend_atoms=static_cast<int> (buf[0]);
- if(nsend_atoms>max_nsend) grow_copylist(nsend_atoms+100);
-
- if (nlocal+nsend_atoms+atom->nghost>=atom->nmax) grow_both(nlocal+nsend_atoms*2+atom->nghost);
- int naccept = Cuda_AtomVecChargeCuda_UnpackExchange(&cuda->shared_data,nsend_atoms,buf,cu_copylist->dev_data());
- cu_copylist->download();
- int m = nsend_atoms*NCUDAEXCHANGE + 1;
- nlocal+=naccept;
- if (atom->nextra_grow)
- for(int j=0;j<nsend_atoms;j++)
- {
- if(copylist[j]>-1)
- {
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++)
- m += modify->fix[atom->extra_grow[iextra]]->
- unpack_exchange(copylist[j],&buf[m]);
- }
- else
- m+=static_cast <int> (buf[j+1]);
- }
- cuda->shared_data.atom.nlocal=nlocal;
- cuda->shared_data.atom.update_nlocal=2;
- atom->nlocal=nlocal;
- mfirst+=m;
- buf=&buf[m];
- }
- return mfirst;
-}
diff --git a/src/USER-CUDA/atom_vec_charge_cuda.h b/src/USER-CUDA/atom_vec_charge_cuda.h
deleted file mode 100644
index 25d431c917..0000000000
--- a/src/USER-CUDA/atom_vec_charge_cuda.h
+++ /dev/null
@@ -1,69 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef ATOM_CLASS
-
-AtomStyle(charge/cuda,AtomVecChargeCuda)
-
-#else
-
-#ifndef LMP_ATOM_VEC_CHARGE_CUDA_H
-#define LMP_ATOM_VEC_CHARGE_CUDA_H
-
-#include "atom_vec_charge.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class AtomVecChargeCuda : public AtomVecCharge {
- public:
- AtomVecChargeCuda(class LAMMPS *);
- virtual ~AtomVecChargeCuda() {}
- void grow_copylist(int n);
- void grow_send(int n,double** buf_send,int flag);
- void grow_both(int n);
- int pack_comm(int, int *, double *, int, int *);
- int pack_comm_vel(int, int *, double *, int, int *);
- void unpack_comm(int, int, double *);
- void unpack_comm_vel(int, int, double *);
- int pack_reverse(int, int, double *);
- void unpack_reverse(int, int *, double *);
- int pack_border(int, int *, double *, int, int *);
- int pack_border_vel(int, int *, double *, int, int *);
- void unpack_border(int, int, double *);
- void unpack_border_vel(int, int, double *);
- int pack_exchange(int, double *);
- int unpack_exchange(double *);
- private:
- class Cuda *cuda;
- bool cuda_init_done;
- int* copylist;
- int* copylist2;
- cCudaData<int, int, xx >* cu_copylist;
- int max_nsend;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/atom_vec_full_cuda.cpp b/src/USER-CUDA/atom_vec_full_cuda.cpp
deleted file mode 100644
index dda12603db..0000000000
--- a/src/USER-CUDA/atom_vec_full_cuda.cpp
+++ /dev/null
@@ -1,508 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "atom_vec_full_cuda.h"
-#include "comm_cuda_cu.h"
-#include "atom_vec_full_cuda_cu.h"
-#include "atom.h"
-#include "domain.h"
-#include "modify.h"
-#include "fix.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "universe.h"
-#include "comm.h"
-
-using namespace LAMMPS_NS;
-
-#define BUFFACTOR 1.5
-#define BUFEXTRA 1000
-#define NCUDAEXCHANGE 13 //nextra x y z vx vy vz tag type mask image q molecule
-
-#define BUF_CFLOAT double
-/* ---------------------------------------------------------------------- */
-
-AtomVecFullCuda::AtomVecFullCuda(LAMMPS *lmp) :
- AtomVecFull(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- maxsend=0;
- cudable=true;
- cuda_init_done=false;
- max_nsend=0;
- cu_copylist=NULL;
- copylist=NULL;
- copylist2=NULL;
-}
-
-void AtomVecFullCuda::grow_copylist(int new_max_nsend)
-{
- max_nsend=new_max_nsend;
- delete cu_copylist;
- delete [] copylist2;
- if(copylist) CudaWrapper_FreePinnedHostData((void*) copylist);
- copylist = (int*) CudaWrapper_AllocPinnedHostData(max_nsend*sizeof(int),false);
- copylist2 = new int[max_nsend];
- cu_copylist = new cCudaData<int, int, xx > (copylist, max_nsend);
-}
-
-void AtomVecFullCuda::grow_send(int n,double** buf_send,int flag) //need to be able to grow the comm send_buffer since the array sahll be copied from the gpu in whole
-{
- int old_maxsend=*maxsend+BUFEXTRA;
- *maxsend = static_cast<int> (BUFFACTOR * n);
- if (flag)
- {
- if(cuda->pinned)
- {
- double* tmp = new double[old_maxsend];
- memcpy((void*) tmp,(void*) *buf_send,old_maxsend*sizeof(double));
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- memcpy(*buf_send,tmp,old_maxsend*sizeof(double));
- delete [] tmp;
- }
- else
- {
- *buf_send = (double *)
- memory->srealloc(*buf_send,(*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
- else {
- if(cuda->pinned)
- {
- if(*buf_send) CudaWrapper_FreePinnedHostData((void*) (*buf_send));
- *buf_send = (double*) CudaWrapper_AllocPinnedHostData((*maxsend+BUFEXTRA)*sizeof(double),false);
- }
- else
- {
- memory->sfree(*buf_send);
- *buf_send = (double *) memory->smalloc((*maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- }
-}
-
-void AtomVecFullCuda::grow_both(int n)
-{
- if(cuda->finished_setup)
- {
- cuda->cu_special->upload();
- cuda->cu_nspecial->upload();
- cuda->downloadAll();
- }
- AtomVecFull::grow(n);
- if(cuda->finished_setup)
- {
- cuda->checkResize();
- cuda->uploadAll();
- }
-}
-
-int AtomVecFullCuda::pack_comm(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecFull::pack_comm(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackComm(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-
-int AtomVecFullCuda::pack_comm_vel(int n, int* iswap, double *buf,
- int pbc_flag, int *pbc) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecFull::pack_comm_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_CommCuda_PackCommVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- if((sizeof(X_CFLOAT)!=sizeof(double)) && m)
- m=(m+1)*sizeof(X_CFLOAT)/sizeof(double);
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecFullCuda::unpack_comm(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecFull::unpack_comm(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackComm(&cuda->shared_data,n,first,(void*)buf);
-}
-
-void AtomVecFullCuda::unpack_comm_vel(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only positions are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecFull::unpack_comm_vel(n,first,buf); return;}
-
- Cuda_CommCuda_UnpackCommVel(&cuda->shared_data,n,first,(void*)buf);
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecFullCuda::pack_reverse(int n, int first, double *buf) //usually this should not be called since comm->communicate handles the communication if only forces are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecFull::pack_reverse(n,first,buf);
-
- int i,m,last;
- cuda->cu_f->download();
- m = 0;
- last = first + n;
- for (i = first; i < last; i++) {
- buf[m++] = f[i][0];
- buf[m++] = f[i][1];
- buf[m++] = f[i][2];
- }
- cuda->cu_f->upload();
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecFullCuda::unpack_reverse(int n, int *list, double *buf)//usually this should not be called since comm->communicate handles the communication if only forces are exchanged
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecFull::unpack_reverse(n,list,buf); return;}
-
- int i,j,m;
-
- m = 0;
- cuda->cu_f->download();
- for (i = 0; i < n; i++) {
- j = list[i];
- f[j][0] += buf[m++];
- f[j][1] += buf[m++];
- f[j][2] += buf[m++];
- }
- cuda->cu_f->upload();
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecFullCuda::pack_border(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecFull::pack_border(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecFullCuda_PackBorder(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
- return m;
-}
-
-int AtomVecFullCuda::pack_border_vel(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- return AtomVecFull::pack_border_vel(n,iswap,buf,pbc_flag,pbc);
-
- int m = Cuda_AtomVecFullCuda_PackBorderVel(&cuda->shared_data,n,*iswap,(void*) buf,pbc,pbc_flag);
-
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void AtomVecFullCuda::unpack_border(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecFull::unpack_border(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax) //ensure there is enough space on device to unpack data
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecFullCuda_UnpackBorder(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-}
-
-void AtomVecFullCuda::unpack_border_vel(int n, int first, double *buf)
-{
- if(not cuda->finished_setup || cuda->oncpu)
- {AtomVecFull::unpack_border_vel(n,first,buf); return;}
- while(atom->nghost+atom->nlocal+n>=cuda->shared_data.atom.nmax) //ensure there is enough space on device to unpack data
- {
- grow_both(0);
- }
- int flag=Cuda_AtomVecFullCuda_UnpackBorderVel(&cuda->shared_data,n,first,(void*)buf);
- if(flag) {printf(" # CUDA: Error: Failed to unpack Border atoms (This might be a bug).\n");}
-}
-
-/* ----------------------------------------------------------------------
- pack data for atom I for sending to another proc
- xyz must be 1st 3 values, so comm::exchange() can test on them
-------------------------------------------------------------------------- */
-
-
-int AtomVecFullCuda::pack_exchange(int dim, double *buf)
-{
- if(cuda->oncpu)
- return AtomVecFull::pack_exchange(dim,buf);
-
- if(not cuda_init_done||domain->box_change)
- {
- Cuda_AtomVecFullCuda_Init(&cuda->shared_data);
- cuda_init_done=true;
- }
- double** buf_pointer=(double**) buf;
- if(*maxsend<atom->nghost || *buf_pointer==NULL)
- {
- grow_send(atom->nghost>*maxsend?atom->nghost:*maxsend,buf_pointer,0);
- *maxsend=atom->nghost>*maxsend?atom->nghost:*maxsend;
- }
-
- if(max_nsend==0) grow_copylist(200);
-
- int nsend_atoms = Cuda_AtomVecFullCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
-
- if(nsend_atoms>max_nsend) grow_copylist(nsend_atoms+100);
- if(nsend_atoms*NCUDAEXCHANGE>*maxsend)
- {
- grow_send((int) (nsend_atoms+100)*NCUDAEXCHANGE,buf_pointer,0);
- Cuda_AtomVecFullCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
- }
-
- int nlocal=atom->nlocal-nsend_atoms;
-
- for(int i=0;i<nsend_atoms;i++) copylist2[i]=1;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i>=nlocal) copylist2[i-nlocal]=-1;
- }
-
- int actpos=0;
- for(int j=1;j<nsend_atoms+1;j++)
- {
- int i = static_cast <int> ((*buf_pointer)[j]);
- if(i<nlocal)
- {
- while(copylist2[actpos]==-1) actpos++;
- copylist[j-1]=nlocal+actpos;
- actpos++;
- }
- }
- cu_copylist->upload();
-
- cuda->shared_data.atom.nlocal=nlocal;
-
- int m = Cuda_AtomVecFullCuda_PackExchange(&cuda->shared_data,nsend_atoms,*buf_pointer,cu_copylist->dev_data());
-
- my_times time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
-
- double* buf_p=*buf_pointer;
- for(int j=0;j<nsend_atoms;j++)
- {
- int i=static_cast <int> (buf_p[j+1]);
- int nextra=0;
- int k;
- buf_p[m++] = num_bond[i];
- for (k = 0; k < num_bond[i]; k++) {
- buf_p[m++] = bond_type[i][k];
- buf_p[m++] = bond_atom[i][k];
- }
- nextra+=2*num_bond[i]+1;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- buf_p[m++] = num_angle[i];
- for (k = 0; k < num_angle[i]; k++) {
- buf_p[m++] = angle_type[i][k];
- buf_p[m++] = angle_atom1[i][k];
- buf_p[m++] = angle_atom2[i][k];
- buf_p[m++] = angle_atom3[i][k];
- }
- nextra+=4*num_angle[i]+1;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- buf_p[m++] = num_dihedral[i];
- for (k = 0; k < num_dihedral[i]; k++) {
- buf_p[m++] = dihedral_type[i][k];
- buf_p[m++] = dihedral_atom1[i][k];
- buf_p[m++] = dihedral_atom2[i][k];
- buf_p[m++] = dihedral_atom3[i][k];
- buf_p[m++] = dihedral_atom4[i][k];
- }
- nextra+=5*num_dihedral[i]+1;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- buf_p[m++] = num_improper[i];
- for (k = 0; k < num_improper[i]; k++) {
- buf_p[m++] = improper_type[i][k];
- buf_p[m++] = improper_atom1[i][k];
- buf_p[m++] = improper_atom2[i][k];
- buf_p[m++] = improper_atom3[i][k];
- buf_p[m++] = improper_atom4[i][k];
- }
- nextra+=5*num_improper[i]+1;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- buf_p[m++] = nspecial[i][0];
- buf_p[m++] = nspecial[i][1];
- buf_p[m++] = nspecial[i][2];
- for (k = 0; k < nspecial[i][2]; k++) buf_p[m++] = special[i][k];
- nextra+=nspecial[i][2]+3;
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
-
- if (atom->nextra_grow)
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++)
- {
- int dm= modify->fix[atom->extra_grow[iextra]]->pack_exchange(i,&buf_p[m]);
- m+=dm;
- nextra+=dm;
- if(i<nlocal)modify->fix[atom->extra_grow[iextra]]->copy_arrays(copylist[j],i,1);
- if(m>*maxsend) {grow_send(m,buf_pointer,1); buf_p=*buf_pointer;}
- }
-
- if(i<nlocal)AtomVecFull::copy(copylist[j],i,1);
- (*buf_pointer)[j+1] = nextra;
- }
-
- my_gettime(CLOCK_REALTIME,&time2);
- cuda->shared_data.cuda_timings.comm_exchange_cpu_pack+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
-
- (*buf_pointer)[0] = nsend_atoms;
- atom->nlocal-=nsend_atoms;
- cuda->shared_data.atom.update_nlocal=2;
- //printf("End Pack Exchange\n");
- if(m==1) return 0;
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int AtomVecFullCuda::unpack_exchange(double *buf)
-{
-// printf("Begin UnPack Exchange\n");
- if(cuda->oncpu)
- return AtomVecFull::unpack_exchange(buf);
-
- int dim=cuda->shared_data.exchange_dim;
- if(domain->box_change)
- Cuda_AtomVecFullCuda_Init(&cuda->shared_data);
-
- int mfirst=0;
- for(int pi=0;pi<(comm->procgrid[dim]>2?2:1);pi++)
- {
- int nlocal = atom->nlocal;
- int nsend_atoms=static_cast<int> (buf[0]);
- if(nsend_atoms>max_nsend) grow_copylist(nsend_atoms+100);
-
- if (nlocal+nsend_atoms+atom->nghost>=atom->nmax) grow_both(nlocal+nsend_atoms*2+atom->nghost); //ensure there is enough space on device to unpack data
- int naccept = Cuda_AtomVecFullCuda_UnpackExchange(&cuda->shared_data,nsend_atoms,buf,cu_copylist->dev_data());
- cu_copylist->download();
- int m = nsend_atoms*NCUDAEXCHANGE + 1;
- nlocal+=naccept;
-
- my_times time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
-
- for(int j=0;j<nsend_atoms;j++)
- {
- if(copylist[j]>-1)
- {
- int k;
- int i=copylist[j];
- num_bond[i] = static_cast<int> (buf[m++]);
- for (k = 0; k < num_bond[i]; k++) {
- bond_type[i][k] = static_cast<int> (buf[m++]);
- bond_atom[i][k] = static_cast<int> (buf[m++]);
- }
-
- num_angle[i] = static_cast<int> (buf[m++]);
- for (k = 0; k < num_angle[i]; k++) {
- angle_type[i][k] = static_cast<int> (buf[m++]);
- angle_atom1[i][k] = static_cast<int> (buf[m++]);
- angle_atom2[i][k] = static_cast<int> (buf[m++]);
- angle_atom3[i][k] = static_cast<int> (buf[m++]);
- }
-
- num_dihedral[i] = static_cast<int> (buf[m++]);
- for (k = 0; k < num_dihedral[i]; k++) {
- dihedral_type[i][k] = static_cast<int> (buf[m++]);
- dihedral_atom1[i][k] = static_cast<int> (buf[m++]);
- dihedral_atom2[i][k] = static_cast<int> (buf[m++]);
- dihedral_atom3[i][k] = static_cast<int> (buf[m++]);
- dihedral_atom4[i][k] = static_cast<int> (buf[m++]);
- }
-
- num_improper[i] = static_cast<int> (buf[m++]);
- for (k = 0; k < num_improper[i]; k++) {
- improper_type[i][k] = static_cast<int> (buf[m++]);
- improper_atom1[i][k] = static_cast<int> (buf[m++]);
- improper_atom2[i][k] = static_cast<int> (buf[m++]);
- improper_atom3[i][k] = static_cast<int> (buf[m++]);
- improper_atom4[i][k] = static_cast<int> (buf[m++]);
- }
-
- nspecial[i][0] = static_cast<int> (buf[m++]);
- nspecial[i][1] = static_cast<int> (buf[m++]);
- nspecial[i][2] = static_cast<int> (buf[m++]);
- for (k = 0; k < nspecial[i][2]; k++)
- special[i][k] = static_cast<int> (buf[m++]);
-
- if (atom->nextra_grow)
- for (int iextra = 0; iextra < atom->nextra_grow; iextra++)
- m += modify->fix[atom->extra_grow[iextra]]->
- unpack_exchange(i,&buf[m]);
-
- }
- else
- m+=static_cast <int> (buf[j+1]);
- }
-
- my_gettime(CLOCK_REALTIME,&time2);
- cuda->shared_data.cuda_timings.comm_exchange_cpu_pack+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
-
- cuda->shared_data.atom.nlocal=nlocal;
- cuda->shared_data.atom.update_nlocal=2;
- atom->nlocal=nlocal;
- mfirst+=m;
- buf=&buf[m];
- }
- return mfirst;
-}
diff --git a/src/USER-CUDA/atom_vec_full_cuda.h b/src/USER-CUDA/atom_vec_full_cuda.h
deleted file mode 100644
index 4ce1b24a51..0000000000
--- a/src/USER-CUDA/atom_vec_full_cuda.h
+++ /dev/null
@@ -1,69 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef ATOM_CLASS
-
-AtomStyle(full/cuda,AtomVecFullCuda)
-
-#else
-
-#ifndef LMP_ATOM_VEC_FULL_CUDA_H
-#define LMP_ATOM_VEC_FULL_CUDA_H
-
-#include "atom_vec_full.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class AtomVecFullCuda : public AtomVecFull {
- public:
- AtomVecFullCuda(class LAMMPS *);
- virtual ~AtomVecFullCuda() {}
- void grow_copylist(int n);
- void grow_send(int n,double** buf_send,int flag);
- void grow_both(int n);
- int pack_comm(int, int *, double *, int, int *);
- int pack_comm_vel(int, int *, double *, int, int *);
- void unpack_comm(int, int, double *);
- void unpack_comm_vel(int, int, double *);
- int pack_reverse(int, int, double *);
- void unpack_reverse(int, int *, double *);
- int pack_border(int, int *, double *, int, int *);
- int pack_border_vel(int, int *, double *, int, int *);
- void unpack_border(int, int, double *);
- void unpack_border_vel(int, int, double *);
- int pack_exchange(int, double *);
- int unpack_exchange(double *);
- private:
- class Cuda *cuda;
- bool cuda_init_done;
- int* copylist;
- int* copylist2;
- cCudaData<int, int, xx >* cu_copylist;
- int max_nsend;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/comm_cuda.cpp b/src/USER-CUDA/comm_cuda.cpp
deleted file mode 100644
index a03f873ce2..0000000000
--- a/src/USER-CUDA/comm_cuda.cpp
+++ /dev/null
@@ -1,1375 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author (triclinic) : Pieter in 't Veld (SNL)
-------------------------------------------------------------------------- */
-
-#ifdef LAMMPS_BIGBIG
-#error LAMMPS_BIGBIG not supported by this file
-#endif
-
-#include <mpi.h>
-#include <cmath>
-#include <cstring>
-#include <cstdio>
-#include <cstdlib>
-#include "comm_cuda.h"
-#include "atom.h"
-#include "atom_vec.h"
-#include "force.h"
-#include "pair.h"
-#include "domain.h"
-#include "neighbor.h"
-#include "modify.h"
-#include "fix.h"
-#include "group.h"
-#include "compute.h"
-#include "user_cuda.h"
-#include "error.h"
-#include "memory.h"
-#include "comm_cuda_cu.h"
-
-using namespace LAMMPS_NS;
-
-#define BUFFACTOR 1.5
-#define BUFMIN 1000
-#define BUFEXTRA 1000
-
-
-
-#define BIG 1.0e20
-
-enum{SINGLE,MULTI};
-
-/* ----------------------------------------------------------------------
- setup MPI and allocate buffer space
-------------------------------------------------------------------------- */
-
-CommCuda::CommCuda(LAMMPS *lmp) : CommBrick(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- cu_pbc=NULL;
- cu_slablo=NULL;
- cu_slabhi=NULL;
- cu_multilo=NULL;
- cu_multihi=NULL;
- cu_sendlist=NULL;
-
-
- memory->sfree(buf_send);
- memory->sfree(buf_recv);
- buf_send = NULL;
- buf_recv = NULL;
-
- CommBrick::free_swap();
- allocate_swap(maxswap);
-}
-
-/* ---------------------------------------------------------------------- */
-
-CommCuda::~CommCuda()
-{
- delete cu_sendlist;
- if(cuda->pinned)
- {
- CudaWrapper_FreePinnedHostData((void*)buf_send);
- CudaWrapper_FreePinnedHostData((void*)buf_recv);
- }
- else
- {
- memory->sfree(buf_send);
- memory->sfree(buf_recv);
- }
- buf_send=NULL;
- buf_recv=NULL;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void CommCuda::init()
-{
- if(not buf_send)
- grow_send(maxsend,0);
- if(not buf_recv)
- grow_recv(maxrecv);
- if(not cu_sendlist)
- {
- cu_sendlist=new cCudaData<int, int, xy> ((int*)sendlist,maxswap,BUFMIN);
- cuda->shared_data.comm.sendlist.dev_data=cu_sendlist->dev_data();
- cuda->shared_data.comm.maxswap=maxswap;
- cuda->shared_data.comm.maxlistlength=BUFMIN;
- cu_sendlist->upload();
- }
- delete cu_pbc;
- cu_pbc=new cCudaData<int, int, xy> ((int*)pbc,cuda->shared_data.comm.maxswap,6);
- cu_pbc->upload();
-
- delete cu_slablo;
- cu_slablo = new cCudaData<double, X_CFLOAT,x>(slablo,cuda->shared_data.comm.maxswap);
- cu_slablo->upload();
-
- delete cu_slabhi;
- cu_slabhi = new cCudaData<double, X_CFLOAT,x>(slabhi,cuda->shared_data.comm.maxswap);
- cu_slabhi->upload();
-
- cuda->shared_data.comm.pbc.dev_data=cu_pbc->dev_data();
- cuda->shared_data.comm.slablo.dev_data=cu_slablo->dev_data();
- cuda->shared_data.comm.slabhi.dev_data=cu_slabhi->dev_data();
-
- CommBrick::init();
-}
-
-/* ----------------------------------------------------------------------
- setup spatial-decomposition communication patterns
- function of neighbor cutoff(s) & cutghostuser & current box size
- single style sets slab boundaries (slablo,slabhi) based on max cutoff
- multi style sets type-dependent slab boundaries (multilo,multihi)
-------------------------------------------------------------------------- */
-
-void CommCuda::setup()
-{
- if(cuda->shared_data.pair.neighall) cutghostuser = MAX(2.0*neighbor->cutneighmax,cutghostuser);
- CommBrick::setup();
-
- //upload changed geometry to device
- if(style == SINGLE)
- {
- if(cu_slablo) cu_slablo->upload();
- if(cu_slabhi) cu_slabhi->upload();
- }
- else
- {
- if(cu_multilo) cu_multilo->upload();
- if(cu_multihi) cu_multihi->upload();
- }
-}
-
-/* ----------------------------------------------------------------------
- forward communication of atom coords every timestep
- other per-atom attributes may also be sent via pack/unpack routines
-------------------------------------------------------------------------- */
-
-void CommCuda::forward_comm(int mode)
-{
- if(mode==0) return forward_comm_cuda();
- if(mode==1) return forward_comm_pack_cuda();
- if(mode==2) return forward_comm_transfer_cuda();
- if(mode==3) return forward_comm_unpack_cuda();
-}
-
-
-void CommCuda::forward_comm_cuda()
-{
- my_times time1,time2,time3;
-
- int n;
- MPI_Request request;
- AtomVec *avec = atom->avec;
-
- cuda->shared_data.domain.xy=domain->xy;
- cuda->shared_data.domain.xz=domain->xz;
- cuda->shared_data.domain.yz=domain->yz;
- cuda->shared_data.domain.prd[0]=domain->prd[0];
- cuda->shared_data.domain.prd[1]=domain->prd[1];
- cuda->shared_data.domain.prd[2]=domain->prd[2];
- cuda->shared_data.domain.triclinic=domain->triclinic;
- if(not comm_x_only && not avec->cudable)
- {
- cuda->downloadAll();
- CommBrick::forward_comm();
- cuda->uploadAll();
- return;
- }
-
- // exchange data with another proc
- // if other proc is self, just copy
- // if comm_x_only set, exchange or copy directly to x, don't unpack
-
- for (int iswap = 0; iswap < nswap; iswap++) {
- if (sendproc[iswap] != me)
- {
- if (comm_x_only)
- {
-
- int size_forward_recv_now=0;
-
- if((sizeof(X_CFLOAT)!=sizeof(double)) && size_forward_recv[iswap]) //some complicated way to safe some transfer size if single precision is used
- size_forward_recv_now=(size_forward_recv[iswap]+1)*sizeof(X_CFLOAT)/sizeof(double);
- else
- size_forward_recv_now=size_forward_recv[iswap];
-my_gettime(CLOCK_REALTIME,&time1);
-
- MPI_Irecv(buf_recv,size_forward_recv_now,MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
- n = Cuda_CommCuda_PackComm(&cuda->shared_data,sendnum[iswap],iswap,(void*) buf_send,pbc[iswap],pbc_flag[iswap]);
-
-my_gettime(CLOCK_REALTIME,&time2);
-
- if((sizeof(X_CFLOAT)!=sizeof(double)) && n) //some complicated way to safe some transfer size if single precision is used
- n=(n+1)*sizeof(X_CFLOAT)/sizeof(double);
-
- //printf("RecvSize: %i SendSize: %i\n",size_forward_recv_now,n);
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
-
-my_gettime(CLOCK_REALTIME,&time3);
-cuda->shared_data.cuda_timings.comm_forward_mpi_upper+=
- time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
-cuda->shared_data.cuda_timings.comm_forward_mpi_lower+=
- time3.tv_sec-time2.tv_sec+1.0*(time3.tv_nsec-time2.tv_nsec)/1000000000;
-
- Cuda_CommCuda_UnpackComm(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],(void*)buf_recv,iswap); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
-
- }
- else if (ghost_velocity)
- {
- MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
-
- if(avec->cudable)
- n = avec->pack_comm_vel(sendnum[iswap],&iswap,
- buf_send,pbc_flag[iswap],pbc[iswap]);
- else
- n = avec->pack_comm_vel(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
-
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- avec->unpack_comm_vel(recvnum[iswap],firstrecv[iswap],buf_recv);
- }
- else
- {
- MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
-
- if(avec->cudable)
- n = avec->pack_comm(sendnum[iswap],&iswap,
- buf_send,pbc_flag[iswap],pbc[iswap]);
- else
- n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
-
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
- }
-
- }
- else //sendproc == me
- {
- cuda->self_comm=1;
- if (comm_x_only)
- {
- if (sendnum[iswap])
- {
- n = Cuda_CommCuda_PackComm_Self(&cuda->shared_data,sendnum[iswap],iswap,firstrecv[iswap],pbc[iswap],pbc_flag[iswap]);
- if(n<0) error->all(FLERR," # CUDA ERRROR on PackComm_Self");
- if((sizeof(X_CFLOAT)!=sizeof(double)) && n)
- n=(n+1)*sizeof(X_CFLOAT)/sizeof(double);
- }
- }
- else if (ghost_velocity)
- {
- n = avec->pack_comm_vel(sendnum[iswap],&iswap,
- (double*) firstrecv,pbc_flag[iswap],pbc[iswap]);
- //avec->unpack_comm_vel(recvnum[iswap],firstrecv[iswap],(double*) firstrecv);
- }
- else
- {
- n = avec->pack_comm(sendnum[iswap],&iswap,
- (double*) firstrecv,pbc_flag[iswap],pbc[iswap]);
- //avec->unpack_comm(recvnum[iswap],firstrecv[iswap],(double*) firstrecv);
- }
- cuda->self_comm=0;
- }
- }
-}
-
-void CommCuda::forward_comm_pack_cuda()
-{
- my_times time1,time2;
- int n; // initialize comm buffers & exchange memory
-
- MPI_Request request;
- AtomVec *avec = atom->avec;
-
- cuda->shared_data.domain.xy=domain->xy;
- cuda->shared_data.domain.xz=domain->xz;
- cuda->shared_data.domain.yz=domain->yz;
- cuda->shared_data.domain.prd[0]=domain->prd[0];
- cuda->shared_data.domain.prd[1]=domain->prd[1];
- cuda->shared_data.domain.prd[2]=domain->prd[2];
- cuda->shared_data.domain.triclinic=domain->triclinic;
- if(not comm_x_only && not avec->cudable) cuda->downloadAll(); //if not comm_x_only the communication routine of the atom_vec style class is used
-
- // exchange data with another proc
- // if other proc is self, just copy
- // if comm_x_only set, exchange or copy directly to x, don't unpack
-
- for (int iswap = 0; iswap < nswap; iswap++) {
- if (sendproc[iswap] != me)
- {
- if (comm_x_only)
- {
-
-
-my_gettime(CLOCK_REALTIME,&time1);
-
- // n = Cuda_CommCuda_PackComm(&cuda->shared_data,sendnum[iswap],iswap,(void*) cuda->shared_data.comm.buf_send[iswap],pbc[iswap],pbc_flag[iswap]);
- n = Cuda_CommCuda_PackComm(&cuda->shared_data,sendnum[iswap],iswap,(void*)buf_send,pbc[iswap],pbc_flag[iswap]);
-
-my_gettime(CLOCK_REALTIME,&time2);
-
- if((sizeof(X_CFLOAT)!=sizeof(double)) && n) //some complicated way to safe some transfer size if single precision is used
- n=(n+1)*sizeof(X_CFLOAT)/sizeof(double);
- cuda->shared_data.comm.send_size[iswap]=n;
- }
- else if (ghost_velocity)
- {
-my_gettime(CLOCK_REALTIME,&time1);
-
- // n = Cuda_CommCuda_PackComm_Vel(&cuda->shared_data,sendnum[iswap],iswap,(void*) &buf_send[iswap*maxsend],pbc[iswap],pbc_flag[iswap]);
-
-my_gettime(CLOCK_REALTIME,&time2);
-
- if((sizeof(X_CFLOAT)!=sizeof(double)) && n) //some complicated way to safe some transfer size if single precision is used
- n=(n+1)*sizeof(X_CFLOAT)/sizeof(double);
- cuda->shared_data.comm.send_size[iswap]=n;
- }
- else
- {
- MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
-
- if(avec->cudable)
- n = avec->pack_comm(sendnum[iswap],&iswap,
- cuda->shared_data.comm.buf_send[iswap],pbc_flag[iswap],pbc[iswap]);
- else
- n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
- cuda->shared_data.comm.buf_send[iswap],pbc_flag[iswap],pbc[iswap]);
-
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
- }
-
- }
- else //sendproc == me
- {
- if (comm_x_only)
- {
- if (sendnum[iswap])
- {
- n = Cuda_CommCuda_PackComm_Self(&cuda->shared_data,sendnum[iswap],iswap,firstrecv[iswap],pbc[iswap],pbc_flag[iswap]);
- if(n<0) error->all(FLERR," # CUDA ERRROR on PackComm_Self");
- if((sizeof(X_CFLOAT)!=sizeof(double)) && n)
- n=(n+1)*sizeof(X_CFLOAT)/sizeof(double);
- }
- }
- else if (ghost_velocity)
- {
- n = avec->pack_comm_vel(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
- avec->unpack_comm_vel(recvnum[iswap],firstrecv[iswap],buf_send);
- }
- else
- {
- n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
- avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_send);
- }
- }
- }
- if(not comm_x_only && not avec->cudable) cuda->uploadAll();
-}
-
-void CommCuda::forward_comm_transfer_cuda()
-{
- my_times time1,time2,time3;
- int n;
- MPI_Request request;
- AtomVec *avec = atom->avec;
- cuda->shared_data.domain.xy=domain->xy;
- cuda->shared_data.domain.xz=domain->xz;
- cuda->shared_data.domain.yz=domain->yz;
- cuda->shared_data.domain.prd[0]=domain->prd[0];
- cuda->shared_data.domain.prd[1]=domain->prd[1];
- cuda->shared_data.domain.prd[2]=domain->prd[2];
- cuda->shared_data.domain.triclinic=domain->triclinic;
- if(not comm_x_only && not avec->cudable) cuda->downloadAll(); //if not comm_x_only the communication routine of the atom_vec style class is used
-//printf("A\n");
- // exchange data with another proc
- // if other proc is self, just copy
- // if comm_x_only set, exchange or copy directly to x, don't unpack
-
- for (int iswap = 0; iswap < nswap; iswap++) {
- if (sendproc[iswap] != me)
- {
- if (comm_x_only)
- {
-
- int size_forward_recv_now=0;
-
- if((sizeof(X_CFLOAT)!=sizeof(double)) && size_forward_recv[iswap]) //some complicated way to safe some transfer size if single precision is used
- size_forward_recv_now=(size_forward_recv[iswap]+1)*sizeof(X_CFLOAT)/sizeof(double);
- else
- size_forward_recv_now=size_forward_recv[iswap];
-
- //printf("A: %i \n",size_forward_recv_now/1024*4);
- //MPI_Irecv(cuda->shared_data.comm.buf_recv[iswap],size_forward_recv_now,MPI_DOUBLE,
- // recvproc[iswap],0,world,&request);
- MPI_Irecv(buf_recv,size_forward_recv_now,MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
- //printf("%p %p %i\n",buf_send, cuda->shared_data.comm.buf_send_dev[iswap], cuda->shared_data.comm.send_size[iswap]*sizeof(double));
- //memcpy(buf_send,cuda->shared_data.comm.buf_send[iswap],cuda->shared_data.comm.send_size[iswap]*sizeof(double));
- // CudaWrapper_SyncStream(1);
- //printf("B: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
- CudaWrapper_DownloadCudaDataAsync((void*) buf_send, cuda->shared_data.comm.buf_send_dev[iswap], cuda->shared_data.comm.send_size[iswap]*sizeof(double),2);
- //MPI_Send(cuda->shared_data.comm.buf_send[iswap],cuda->shared_data.comm.send_size[iswap],MPI_DOUBLE,sendproc[iswap],0,world);
-my_gettime(CLOCK_REALTIME,&time1);
- CudaWrapper_SyncStream(2);
- //printf("C: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
-my_gettime(CLOCK_REALTIME,&time2);
-cuda->shared_data.cuda_timings.comm_forward_download+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
- MPI_Send(buf_send,cuda->shared_data.comm.send_size[iswap],MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- //printf("D: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
- CudaWrapper_UploadCudaDataAsync((void*) buf_recv,cuda->shared_data.comm.buf_recv_dev[iswap], size_forward_recv_now*sizeof(double),2);
-my_gettime(CLOCK_REALTIME,&time1);
- CudaWrapper_SyncStream(2);
- //printf("E: %i \n",cuda->shared_data.comm.send_size[iswap]/1024*4);
- //memcpy(cuda->shared_data.comm.buf_recv[iswap],buf_recv,size_forward_recv_now*sizeof(double));
- //printf("RecvSize: %i SendSize: %i\n",size_forward_recv_now*sizeof(double),cuda->shared_data.comm.send_size[iswap]*sizeof(double));
-my_gettime(CLOCK_REALTIME,&time3);
-cuda->shared_data.cuda_timings.comm_forward_upload+=
- time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
-cuda->shared_data.cuda_timings.comm_forward_mpi_lower+=
- time3.tv_sec-time2.tv_sec+1.0*(time3.tv_nsec-time2.tv_nsec)/1000000000;
-my_gettime(CLOCK_REALTIME,&time3);
-cuda->shared_data.cuda_timings.comm_forward_mpi_upper+=
- time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
- }
- else if (ghost_velocity)
- {
- /* int size_forward_recv_now=0;
-
- if((sizeof(X_CFLOAT)!=sizeof(double)) && size_forward_recv[iswap]) //some complicated way to safe some transfer size if single precision is used
- size_forward_recv_now=(size_forward_recv[iswap]+1)*sizeof(X_CFLOAT)/sizeof(double);
- else
- size_forward_recv_now=size_forward_recv[iswap];
-
-my_gettime(CLOCK_REALTIME,&time1);
-
- MPI_Irecv(cuda->shared_data.comm.buf_recv[iswap],size_forward_recv_now,MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
-
-my_gettime(CLOCK_REALTIME,&time2);
-
- MPI_Send(cuda->shared_data.comm.buf_send[iswap],cuda->shared_data.comm.send_size[iswap],MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
-
-my_gettime(CLOCK_REALTIME,&time3);
-cuda->shared_data.cuda_timings.comm_forward_mpi_upper+=
- time3.tv_sec-time1.tv_sec+1.0*(time3.tv_nsec-time1.tv_nsec)/1000000000;
-cuda->shared_data.cuda_timings.comm_forward_mpi_lower+=
- time3.tv_sec-time2.tv_sec+1.0*(time3.tv_nsec-time2.tv_nsec)/1000000000;*/
-
- }
- else
- {
- MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
-
- if(avec->cudable)
- n = avec->pack_comm(sendnum[iswap],&iswap,
- buf_send,pbc_flag[iswap],pbc[iswap]);
- else
- n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
-
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
- }
-
- }
- else //sendproc == me
- {
- if (comm_x_only)
- {
- if (sendnum[iswap])
- {
- }
- }
- else if (ghost_velocity)
- {
- }
- else
- {
- n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
- avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_send);
- }
- }
- }
- if(not comm_x_only && not avec->cudable) cuda->uploadAll();
-}
-
-void CommCuda::forward_comm_unpack_cuda()
-{
- int n;
- MPI_Request request;
- AtomVec *avec = atom->avec;
-
- cuda->shared_data.domain.xy=domain->xy;
- cuda->shared_data.domain.xz=domain->xz;
- cuda->shared_data.domain.yz=domain->yz;
- cuda->shared_data.domain.prd[0]=domain->prd[0];
- cuda->shared_data.domain.prd[1]=domain->prd[1];
- cuda->shared_data.domain.prd[2]=domain->prd[2];
- cuda->shared_data.domain.triclinic=domain->triclinic;
- if(not comm_x_only && not avec->cudable) cuda->downloadAll(); //if not comm_x_only the communication routine of the atom_vec style class is used
-
- // exchange data with another proc
- // if other proc is self, just copy
- // if comm_x_only set, exchange or copy directly to x, don't unpack
-
- for (int iswap = 0; iswap < nswap; iswap++) {
- if (sendproc[iswap] != me)
- {
- if (comm_x_only)
- {
-
- //Cuda_CommCuda_UnpackComm(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],cuda->shared_data.comm.buf_recv[iswap],iswap); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
- Cuda_CommCuda_UnpackComm(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],buf_recv,iswap); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
-
- }
- else if (ghost_velocity)
- {
- //Cuda_CommCuda_UnpackComm_Vel(&cuda->shared_data,recvnum[iswap],firstrecv[iswap],(void*)&buf_recv[iswap*maxrecv]); //Unpack for cpu exchange happens implicitely since buf==x[firstrecv]
- }
- else
- {
- MPI_Irecv(buf_recv,size_forward_recv[iswap],MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
-
- if(avec->cudable)
- n = avec->pack_comm(sendnum[iswap],&iswap,
- buf_send,pbc_flag[iswap],pbc[iswap]);
- else
- n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
-
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_recv);
- }
-
- }
- else //sendproc == me
- {
- if (comm_x_only)
- {
- if (sendnum[iswap])
- {
- }
- }
- else if (ghost_velocity)
- {
- }
- else
- {
- n = avec->pack_comm(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
- avec->unpack_comm(recvnum[iswap],firstrecv[iswap],buf_send);
- }
- }
- }
- if(not comm_x_only && not avec->cudable) cuda->uploadAll();
-}
-
-void CommCuda::forward_comm_pair(Pair *pair)
-{
- if(not cuda->shared_data.pair.cudable_force)
- {
- return CommBrick::forward_comm_pair(pair);
- }
-
- int iswap,n;
- double *buf;
- MPI_Request request;
-
- int nsize = pair->comm_forward;
-
- for (iswap = 0; iswap < nswap; iswap++) {
-
- // pack buffer
-
- n = pair->pack_forward_comm(sendnum[iswap],&iswap,
- buf_send,pbc_flag[iswap],pbc[iswap]);
- int nrecv = recvnum[iswap]*nsize;
- if(nrecv<0) nrecv=-(nrecv+1)/2;
- int nsend = n;
- if(nsend<0) nsend=-(nsend+1)/2;
-
- // exchange with another proc
- // if self, set recv buffer to send buffer
-
- if (sendproc[iswap] != me) {
- MPI_Irecv(buf_recv,nrecv,MPI_DOUBLE,recvproc[iswap],0,
- world,&request);
- MPI_Send(buf_send,nsend,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- buf = buf_recv;
- } else buf = buf_send;
-
- // unpack buffer
-
- pair->unpack_forward_comm(recvnum[iswap],firstrecv[iswap],buf);
- }
-}
-
-/* ----------------------------------------------------------------------
- reverse communication of forces on atoms every timestep
- other per-atom attributes may also be sent via pack/unpack routines
-------------------------------------------------------------------------- */
-
-void CommCuda::reverse_comm()
-{
- int n;
- MPI_Request request;
- AtomVec *avec = atom->avec;
- double *buf;
-
- if(not comm_f_only && not avec->cudable) cuda->downloadAll(); //not yet implemented in CUDA but only needed for non standard atom styles
-
- // exchange data with another proc
- // if other proc is self, just copy
- // if comm_f_only set, exchange or copy directly from f, don't pack
-
- for (int iswap = nswap-1; iswap >= 0; iswap--) {
- if (sendproc[iswap] != me) {
- if (comm_f_only) {
-
- int size_recv_now=size_reverse_recv[iswap];
- if((sizeof(F_CFLOAT)!=sizeof(double))&& size_reverse_recv[iswap])
- size_recv_now=(size_recv_now+1)*sizeof(F_CFLOAT)/sizeof(double);
- MPI_Irecv(buf_recv,size_recv_now,MPI_DOUBLE,
- sendproc[iswap],0,world,&request);
-
- buf=buf_send;
- if (size_reverse_send[iswap])
- {
- Cuda_CommCuda_PackReverse(&cuda->shared_data,size_reverse_send[iswap]/3,firstrecv[iswap],buf);
- }
- else buf=NULL;
- int size_reverse_send_now=size_reverse_send[iswap];
- if((sizeof(F_CFLOAT)!=sizeof(double))&& size_reverse_send[iswap])
- size_reverse_send_now=(size_reverse_send_now+1)*sizeof(F_CFLOAT)/sizeof(double);
- MPI_Send(buf,size_reverse_send_now,MPI_DOUBLE,
- recvproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- Cuda_CommCuda_UnpackReverse(&cuda->shared_data,sendnum[iswap],iswap,buf_recv);
-
- } else {
- MPI_Irecv(buf_recv,size_reverse_recv[iswap],MPI_DOUBLE,
- sendproc[iswap],0,world,&request);
- n = avec->pack_reverse(recvnum[iswap],firstrecv[iswap],buf_send);
- MPI_Send(buf_send,n,MPI_DOUBLE,recvproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
-
- avec->unpack_reverse(sendnum[iswap],sendlist[iswap],buf_recv);
- }
-
- } else {
- if (comm_f_only) {
- if (sendnum[iswap])
- Cuda_CommCuda_UnpackReverse_Self(&cuda->shared_data,sendnum[iswap],iswap,firstrecv[iswap]);
- } else {
- n = avec->pack_reverse(recvnum[iswap],firstrecv[iswap],buf_send);
- avec->unpack_reverse(sendnum[iswap],sendlist[iswap],buf_send);
- }
- }
- }
- if(not comm_f_only && not avec->cudable) cuda->uploadAll(); //not yet implemented in CUDA but only needed for non standard atom styles
-}
-
-/* ----------------------------------------------------------------------
- exchange: move atoms to correct processors
- atoms exchanged with all 6 stencil neighbors
- send out atoms that have left my box, receive ones entering my box
- atoms will be lost if not inside some proc's box
- can happen if atom moves outside of non-periodic bounary
- or if atom moves more than one proc away
- this routine called before every reneighboring
- for triclinic, atoms must be in lamda coords (0-1) before exchange is called
-------------------------------------------------------------------------- */
-
-void CommCuda::exchange()
-{
- AtomVec *avec = atom->avec;
-
- if(not cuda->oncpu && avec->cudable)
- return exchange_cuda();
-
- if(not cuda->oncpu) cuda->downloadAll();
-
- CommBrick::exchange();
-}
-
-
-void CommCuda::exchange_cuda()
-{
- int nsend,nrecv,nrecv1,nrecv2,nlocal;
- double *buf;
- MPI_Request request;
- AtomVec *avec = atom->avec;
- my_times time1,time2;
-
- // clear global->local map for owned and ghost atoms
- // b/c atoms migrate to new procs in exchange() and
- // new ghosts are created in borders()
- // map_set() is done at end of borders()
-
-
- if(map_style) cuda->cu_tag->download();
-
- if (map_style) atom->map_clear();
-
- // loop over dimensions
-
- for (int dim = 0; dim < 3; dim++) {
- // fill buffer with atoms leaving my box, using < and >=
- // when atom is deleted, fill it in with last atom
-
- cuda->shared_data.exchange_dim=dim;
-
- nlocal = atom->nlocal;
- avec->maxsend=&maxsend;
- nsend=avec->pack_exchange(dim,(double*) &buf_send);
- nlocal = atom->nlocal;
-
-
- atom->nlocal = nlocal;
-
- // send/recv atoms in both directions
- // if 1 proc in dimension, no send/recv, set recv buf to send buf
- // if 2 procs in dimension, single send/recv
- // if more than 2 procs in dimension, send/recv to both neighbors
-
- my_gettime(CLOCK_REALTIME,&time1);
-
- if (procgrid[dim] == 1) {
- nrecv = nsend;
- buf = buf_send;
-
- } else {
- MPI_Sendrecv(&nsend,1,MPI_INT,procneigh[dim][0],0,
- &nrecv1,1,MPI_INT,procneigh[dim][1],0,world,MPI_STATUS_IGNORE);
- nrecv = nrecv1;
- if (procgrid[dim] > 2) {
- MPI_Sendrecv(&nsend,1,MPI_INT,procneigh[dim][1],0,
- &nrecv2,1,MPI_INT,procneigh[dim][0],0,world,MPI_STATUS_IGNORE);
- nrecv += nrecv2;
- }
- if (nrecv+1 > maxrecv) grow_recv(nrecv+1);
-
- MPI_Irecv(buf_recv,nrecv1,MPI_DOUBLE,procneigh[dim][1],0,
- world,&request);
- MPI_Send(buf_send,nsend,MPI_DOUBLE,procneigh[dim][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
-
- if (procgrid[dim] > 2) {
- MPI_Irecv(&buf_recv[nrecv1],nrecv2,MPI_DOUBLE,procneigh[dim][0],0,
- world,&request);
- MPI_Send(buf_send,nsend,MPI_DOUBLE,procneigh[dim][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
-
- if((nrecv1==0)||(nrecv2==0)) buf_recv[nrecv]=0;
- }
-
- buf = buf_recv;
- }
- //printf("nsend: %i nrecv: %i\n",nsend,nrecv);
- // check incoming atoms to see if they are in my box
- // if so, add to my list
-my_gettime(CLOCK_REALTIME,&time2);
-cuda->shared_data.cuda_timings.comm_exchange_mpi+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
-
- if(nrecv)
- {
- avec->maxsend=&maxsend;
- avec->unpack_exchange(buf);
- }
- }
-
- if(atom->firstgroupname) cuda->downloadAll();
-
- if(atom->firstgroupname) atom->first_reorder();
-
- if(atom->firstgroupname) cuda->uploadAll();
-}
-
-/* ----------------------------------------------------------------------
- borders: list nearby atoms to send to neighboring procs at every timestep
- one list is created for every swap that will be made
- as list is made, actually do swaps
- this does equivalent of a communicate (so don't need to explicitly
- call communicate routine on reneighboring timestep)
- this routine is called before every reneighboring
- for triclinic, atoms must be in lamda coords (0-1) before borders is called
-------------------------------------------------------------------------- */
-
-
-void CommCuda::borders()
-{
- AtomVec *avec = atom->avec;
- if(not cuda->oncpu && avec->cudable)
- {
- if(cuda->shared_data.overlap_comm&&cuda->finished_setup)
- borders_cuda_overlap_forward_comm();
- else
- borders_cuda();
-
- return;
- }
-
- CommBrick::borders();
-
- cuda->setSystemParams();
- if(cuda->finished_setup) {cuda->checkResize(); cuda->uploadAll();}
- cuda->shared_data.atom.nghost=atom->nghost;
- cu_sendlist->upload();
-}
-
-void CommCuda::borders_cuda()
-{
- int n,iswap,dim,ineed,twoneed,smax,rmax;
- int nsend,nrecv,nfirst,nlast;
- double *buf;
- MPI_Request request;
- AtomVec *avec = atom->avec;
- my_times time1,time2;
-
- // clear old ghosts
-
- atom->nghost = 0;
-
- // do swaps over all 3 dimensions
-
- iswap = 0;
- smax = rmax = 0;
-
- cuda->shared_data.comm.nsend=0;
- for (dim = 0; dim < 3; dim++) {
- nlast = 0;
- twoneed = 2*maxneed[dim];
- for (ineed = 0; ineed < twoneed; ineed++) {
-
- // find atoms within slab boundaries lo/hi using <= and >=
- // check atoms between nfirst and nlast
- // for first swaps in a dim, check owned and ghost
- // for later swaps in a dim, only check newly arrived ghosts
- // store sent atom indices in list for use in future timesteps
-
- if (ineed % 2 == 0) {
- nfirst = nlast;
- nlast = atom->nlocal + atom->nghost;
- }
-
- nsend = 0;
-
- // find send atoms according to SINGLE vs MULTI
- // all atoms eligible versus atoms in bordergroup
- // only need to limit loop to bordergroup for first sends (ineed < 2)
- // on these sends, break loop in two: owned (in group) and ghost
- do
- {
- if(nsend>=maxsendlist[iswap]) grow_list(iswap,static_cast <int> (nsend*1.05));
- nsend=Cuda_CommCuda_BuildSendlist(&cuda->shared_data,bordergroup,ineed,style==SINGLE?1:0,atom->nfirst,nfirst,nlast,dim,iswap);
- }while(nsend>=maxsendlist[iswap]);
- // pack up list of border atoms
-
- if (nsend*size_border > maxsend)
- grow_send(nsend*size_border,0);
-
- if (ghost_velocity)
- n = avec->pack_border_vel(nsend,&iswap,buf_send,
- pbc_flag[iswap],pbc[iswap]);
- else
- n = avec->pack_border(nsend,&iswap,buf_send,
- pbc_flag[iswap],pbc[iswap]);
-
- // swap atoms with other proc
- // put incoming ghosts at end of my atom arrays
- // if swapping with self, simply copy, no messages
-
-my_gettime(CLOCK_REALTIME,&time1);
- if (sendproc[iswap] != me) {
- MPI_Sendrecv(&nsend,1,MPI_INT,sendproc[iswap],0,
- &nrecv,1,MPI_INT,recvproc[iswap],0,world,MPI_STATUS_IGNORE);
- if (nrecv*size_border > maxrecv)
- grow_recv(nrecv*size_border);
- MPI_Irecv(buf_recv,nrecv*size_border,MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- buf = buf_recv;
- } else {
- nrecv = nsend;
- buf = buf_send;
- }
-
-my_gettime(CLOCK_REALTIME,&time2);
-cuda->shared_data.cuda_timings.comm_border_mpi+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
-
- // unpack buffer
-
- if (ghost_velocity)
- avec->unpack_border_vel(nrecv,atom->nlocal+atom->nghost,buf);
- else
- avec->unpack_border(nrecv,atom->nlocal+atom->nghost,buf);
-
- // set all pointers & counters
-
- smax = MAX(smax,nsend);
- rmax = MAX(rmax,nrecv);
- sendnum[iswap] = nsend;
- recvnum[iswap] = nrecv;
- size_forward_recv[iswap] = nrecv*size_forward;
- size_reverse_send[iswap] = nrecv*size_reverse;
- size_reverse_recv[iswap] = nsend*size_reverse;
- firstrecv[iswap] = atom->nlocal + atom->nghost;
- atom->nghost += nrecv;
- iswap++;
- }
- }
-
- // insure send/recv buffers are long enough for all forward & reverse comm
-
- int max = MAX(maxforward*smax,maxreverse*rmax);
- if (max > maxsend) grow_send(max,0);
- max = MAX(maxforward*rmax,maxreverse*smax);
- if (max > maxrecv) grow_recv(max);
-
- // reset global->local map
- if(map_style)
- {
- cuda->cu_tag->download();
- atom->map_set();
- }
-
- cuda->setSystemParams();
- cuda->shared_data.atom.nghost+=n;
-}
-
-void CommCuda::borders_cuda_overlap_forward_comm()
-{
- int n,iswap,dim,ineed,twoneed,smax,rmax;
- int nsend,nrecv,nfirst,nlast;
- double *buf;
- MPI_Request request;
- AtomVec *avec = atom->avec;
- my_times time1,time2;
-
- // clear old ghosts
-
- atom->nghost = 0;
-
- // do swaps over all 3 dimensions
-
- iswap = 0;
- smax = rmax = 0;
-
- cuda->shared_data.comm.nsend=0;
- for (dim = 0; dim < 3; dim++) {
- nlast = 0;
- twoneed = 2*maxneed[dim];
- for (ineed = 0; ineed < twoneed; ineed++) {
-
- // find atoms within slab boundaries lo/hi using <= and >=
- // check atoms between nfirst and nlast
- // for first swaps in a dim, check owned and ghost
- // for later swaps in a dim, only check newly arrived ghosts
- // store sent atom indices in list for use in future timesteps
-
- if (ineed % 2 == 0) {
- nfirst = nlast;
- nlast = atom->nlocal + atom->nghost;
- }
-
- nsend = 0;
-
- // find send atoms according to SINGLE vs MULTI
- // all atoms eligible versus atoms in bordergroup
- // only need to limit loop to bordergroup for first sends (ineed < 2)
- // on these sends, break loop in two: owned (in group) and ghost
- do
- {
- if(nsend>=maxsendlist[iswap]) grow_list(iswap,static_cast <int> (nsend*1.05));
- nsend=Cuda_CommCuda_BuildSendlist(&cuda->shared_data,bordergroup,ineed,style==SINGLE?1:0,atom->nfirst,nfirst,nlast,dim,iswap);
- }while(nsend>=maxsendlist[iswap]);
- cuda->shared_data.comm.nsend_swap[iswap]=nsend;
- // pack up list of border atoms
-
- if (nsend*size_border > maxsend)
- grow_send(nsend*size_border,0);
-
- if (ghost_velocity)
- n = avec->pack_border_vel(nsend,&iswap,buf_send,
- pbc_flag[iswap],pbc[iswap]);
- else
- n = avec->pack_border(nsend,&iswap,buf_send,
- pbc_flag[iswap],pbc[iswap]);
-
- // swap atoms with other proc
- // put incoming ghosts at end of my atom arrays
- // if swapping with self, simply copy, no messages
-
-my_gettime(CLOCK_REALTIME,&time1);
- if (sendproc[iswap] != me) {
- MPI_Sendrecv(&nsend,1,MPI_INT,sendproc[iswap],0,
- &nrecv,1,MPI_INT,recvproc[iswap],0,world,MPI_STATUS_IGNORE);
- if (nrecv*size_border > maxrecv)
- grow_recv(nrecv*size_border);
- MPI_Irecv(buf_recv,nrecv*size_border,MPI_DOUBLE,
- recvproc[iswap],0,world,&request);
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- buf = buf_recv;
- } else {
- nrecv = nsend;
- buf = buf_send;
- }
-
-my_gettime(CLOCK_REALTIME,&time2);
-cuda->shared_data.cuda_timings.comm_border_mpi+=
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000;
-
- // unpack buffer
-
- if (ghost_velocity)
- avec->unpack_border_vel(nrecv,atom->nlocal+atom->nghost,buf);
- else
- avec->unpack_border(nrecv,atom->nlocal+atom->nghost,buf);
-
- // set all pointers & counters
-
- smax = MAX(smax,nsend);
- rmax = MAX(rmax,nrecv);
- sendnum[iswap] = nsend;
- recvnum[iswap] = nrecv;
- size_forward_recv[iswap] = nrecv*size_forward;
- size_reverse_send[iswap] = nrecv*size_reverse;
- size_reverse_recv[iswap] = nsend*size_reverse;
- firstrecv[iswap] = atom->nlocal + atom->nghost;
- atom->nghost += nrecv;
- iswap++;
- }
- }
-
- // insure send/recv buffers are long enough for all forward & reverse comm
-
- int max = MAX(maxforward*smax,maxreverse*rmax);
- if (max > maxsend) grow_send(max,0);
- max = MAX(maxforward*rmax,maxreverse*smax);
- if (max > maxrecv) grow_recv(max);
-
- // reset global->local map
- if(map_style)
- {
- cuda->cu_tag->download();
- atom->map_set();
- }
-
- cuda->setSystemParams();
- cuda->shared_data.atom.nghost+=n;
-}
-
-
-
-
-void CommCuda::forward_comm_fix(Fix *fix, int size)
-{
- int iswap,n;
- double *buf;
- MPI_Request request;
-
- int nsize = fix->comm_forward;
-
- for (iswap = 0; iswap < nswap; iswap++) {
- // pack buffer
- if(fix->cudable_comm&&cuda->finished_setup)
- {
- int swap=iswap;
- if(sendproc[iswap] == me) {swap=-iswap-1; buf=(double*)&(firstrecv[iswap]);}
- else buf=buf_send;
-
- n = fix->pack_forward_comm(sendnum[iswap],&swap,
- buf,pbc_flag[iswap],pbc[iswap]);
- if(sendproc[iswap] == me)
- {
- continue;
- }
- }
- else
- n = fix->pack_forward_comm(sendnum[iswap],sendlist[iswap],
- buf_send,pbc_flag[iswap],pbc[iswap]);
-
- // exchange with another proc
- // if self, set recv buffer to send buffer
-
- if (sendproc[iswap] != me) {
- MPI_Irecv(buf_recv,nsize*recvnum[iswap],MPI_DOUBLE,recvproc[iswap],0,
- world,&request);
- MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- buf = buf_recv;
- } else buf = buf_send;
-
- // unpack buffer
-
- fix->unpack_forward_comm(recvnum[iswap],firstrecv[iswap],buf);
- }
-}
-
-
-void CommCuda::grow_send(int n, int flag)
-{
- int oldmaxsend = (maxsend+BUFEXTRA)*sizeof(double);
- maxsend = static_cast<int> (BUFFACTOR * n);
- if (flag){
- if(cuda->pinned)
- {
- double* tmp = new double[oldmaxsend];
- memcpy((void*) tmp,(void*) buf_send,oldmaxsend*sizeof(double));
- if(buf_send) CudaWrapper_FreePinnedHostData((void*) (buf_send));
- buf_send = (double*) CudaWrapper_AllocPinnedHostData((maxsend+BUFEXTRA)*sizeof(double),false);
- memcpy(buf_send,tmp,oldmaxsend*sizeof(double));
- delete [] tmp;
- }
- else
- {
- buf_send = (double *)
- memory->srealloc(buf_send,(maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");printf("srealloc\n");
- }
- }
- else {
- if(cuda->pinned)
- {
- if(buf_send) CudaWrapper_FreePinnedHostData((void*) buf_send);
- buf_send = (double*) CudaWrapper_AllocPinnedHostData((maxsend+BUFEXTRA)*sizeof(double),false);
- }
- else
- {
- memory->sfree(buf_send);
- buf_send = (double *) memory->smalloc((maxsend+BUFEXTRA)*sizeof(double),
- "comm:buf_send");
- }
- for(int i=0;i<maxswap;i++)
- {
- if(cuda->shared_data.comm.buf_send_dev[i]) CudaWrapper_FreeCudaData(cuda->shared_data.comm.buf_send_dev[i],oldmaxsend);
- cuda->shared_data.comm.buf_send_dev[i]=CudaWrapper_AllocCudaData((maxsend+BUFEXTRA)*sizeof(double));
- }
- }
-}
-/* ----------------------------------------------------------------------
- free/malloc the size of the recv buffer as needed with BUFFACTOR
-------------------------------------------------------------------------- */
-
-
-void CommCuda::grow_recv(int n)
-{
- int oldmaxrecv = maxrecv*sizeof(double);
- maxrecv = static_cast<int> (BUFFACTOR * n);
- if(cuda->pinned)
- {
- if(buf_recv) CudaWrapper_FreePinnedHostData((void*)buf_recv);
- buf_recv = (double*) CudaWrapper_AllocPinnedHostData(maxrecv*sizeof(double), false,true);
- }
- else
- {
- memory->sfree(buf_recv);
- buf_recv = (double *) memory->smalloc(maxrecv*sizeof(double),
- "comm:buf_recv");
- }
- for(int i=0;i<maxswap;i++)
- {
- if(cuda->shared_data.comm.buf_recv_dev[i]) CudaWrapper_FreeCudaData(cuda->shared_data.comm.buf_recv_dev[i],oldmaxrecv);
- cuda->shared_data.comm.buf_recv_dev[i]=CudaWrapper_AllocCudaData((maxrecv)*sizeof(double));
- }
-}
-
-/* ----------------------------------------------------------------------
- realloc the size of the iswap sendlist as needed with BUFFACTOR
-------------------------------------------------------------------------- */
-
-void CommCuda::grow_list(int iswap, int n)
-{
-
- MYDBG(printf(" # CUDA CommCuda::grow_list\n");)
- if(cuda->finished_setup&&cu_sendlist) cu_sendlist->download();
- if(!cu_sendlist||n*BUFFACTOR>cu_sendlist->get_dim()[1]||n*BUFFACTOR>maxsendlist[iswap])
- {
- for(int i=0;i<maxswap;i++)
- {
- maxsendlist[i] = static_cast<int> (BUFFACTOR * n);
- sendlist[i] = (int *)
- memory->srealloc(sendlist[i],maxsendlist[i]*sizeof(int),
- "comm:sendlist[iswap]");
- }
- delete cu_sendlist;
- cu_sendlist=new cCudaData<int, int, xy> ((int*)sendlist,maxswap,maxsendlist[iswap]);
- cuda->shared_data.comm.sendlist.dev_data=cu_sendlist->dev_data();
- cuda->shared_data.comm.maxlistlength=maxsendlist[iswap];
- cu_sendlist->upload();
- }
- }
-
-/* ----------------------------------------------------------------------
- realloc the buffers needed for swaps
-------------------------------------------------------------------------- */
-
-void CommCuda::grow_swap(int n)
-{
- int oldmaxswap=maxswap;
- CommBrick::grow_swap(n);
- if(n>cu_sendlist->get_dim()[0])
- {
- MYDBG(printf(" # CUDA CommCuda::grow_swap\n");)
-
- delete cu_sendlist;
- cu_sendlist=new cCudaData<int, int, xy> ((int*)sendlist,n,BUFMIN);
- cuda->shared_data.comm.sendlist.dev_data=cu_sendlist->dev_data();
- cuda->shared_data.comm.maxlistlength=BUFMIN;
- cuda->shared_data.comm.maxswap=n;
- cuda->shared_data.comm.nsend_swap=new int[n];
- cuda->shared_data.comm.send_size=new int[n];
- cuda->shared_data.comm.recv_size=new int[n];
- }
- for(int i=0;i<oldmaxswap;i++)
- {
- if(cuda->shared_data.comm.buf_recv_dev[i]) CudaWrapper_FreeCudaData(cuda->shared_data.comm.buf_recv_dev[i],maxrecv*sizeof(double));
- if(cuda->shared_data.comm.buf_send_dev[i]) CudaWrapper_FreeCudaData(cuda->shared_data.comm.buf_send_dev[i],maxsend*sizeof(double));
- cuda->shared_data.comm.buf_recv_dev[i]=NULL;
- cuda->shared_data.comm.buf_send_dev[i]=NULL;
- }
- cuda->shared_data.comm.buf_send= new double*[n];
- cuda->shared_data.comm.buf_recv= new double*[n];
- cuda->shared_data.comm.buf_send_dev= new void*[n];
- cuda->shared_data.comm.buf_recv_dev= new void*[n];
- for(int i=0;i<n;i++)
- {
- cuda->shared_data.comm.buf_recv[i]=NULL;
- cuda->shared_data.comm.buf_send[i]=NULL;
- cuda->shared_data.comm.buf_recv_dev[i]=NULL;
- cuda->shared_data.comm.buf_send_dev[i]=NULL;
- }
- grow_send(maxsend,0);
- grow_recv(maxrecv);
-
- maxswap=n;
-}
-
-/* ----------------------------------------------------------------------
- allocation of swap info
-------------------------------------------------------------------------- */
-
-void CommCuda::allocate_swap(int n)
-{
- CommBrick::allocate_swap(n);
-
- delete cu_pbc;
- delete cu_slablo;
- delete cu_slabhi;
-
- cuda->shared_data.comm.maxswap=n;
- if(cu_sendlist)
- {
- cu_pbc=new cCudaData<int, int, xy> ((int*)pbc,n,6);
- cu_slablo = new cCudaData<double, X_CFLOAT,x>(slablo,n);
- cu_slabhi = new cCudaData<double, X_CFLOAT,x>(slabhi,n);
-
- cuda->shared_data.comm.pbc.dev_data=cu_pbc->dev_data();
- cuda->shared_data.comm.slablo.dev_data=cu_slablo->dev_data();
- cuda->shared_data.comm.slabhi.dev_data=cu_slabhi->dev_data();
- }
- cuda->shared_data.comm.nsend_swap=new int[n];
- cuda->shared_data.comm.send_size=new int[n];
- cuda->shared_data.comm.recv_size=new int[n];
- cuda->shared_data.comm.buf_send= new double*[n];
- cuda->shared_data.comm.buf_recv= new double*[n];
- cuda->shared_data.comm.buf_send_dev= new void*[n];
- cuda->shared_data.comm.buf_recv_dev= new void*[n];
- for(int i=0;i<n;i++) cuda->shared_data.comm.buf_send_dev[i]=NULL;
- for(int i=0;i<n;i++) cuda->shared_data.comm.buf_recv_dev[i]=NULL;
-}
-
-
-/* ----------------------------------------------------------------------
- allocation of multi-type swap info
-------------------------------------------------------------------------- */
-
-void CommCuda::allocate_multi(int n)
-{
- CommBrick::allocate_multi(n);
-
- delete cu_multilo;
- delete cu_multihi;
- cu_multilo = new cCudaData<double, X_CFLOAT,xy>(slablo,n,atom->ntypes+1);
- cu_multihi = new cCudaData<double, X_CFLOAT,xy>(slabhi,n,atom->ntypes+1);
-
- cuda->shared_data.comm.multilo.dev_data=cu_multilo->dev_data();
- cuda->shared_data.comm.multihi.dev_data=cu_multihi->dev_data();
-}
-
-/* ----------------------------------------------------------------------
- free memory for swaps
-------------------------------------------------------------------------- */
-
-void CommCuda::free_swap()
-{
-
- CommBrick::free_swap();
-
- delete cuda->shared_data.comm.nsend_swap; cuda->shared_data.comm.nsend_swap=NULL;
- delete cu_pbc; cu_pbc = NULL;
- delete cu_slablo; cu_slablo = NULL;
- delete cu_slabhi; cu_slabhi = NULL;
- for(int i=0;i<maxswap;i++)
- {
- if(cuda->shared_data.comm.buf_recv_dev[i]) CudaWrapper_FreeCudaData(cuda->shared_data.comm.buf_recv_dev[i],maxrecv*sizeof(double));
- if(cuda->shared_data.comm.buf_send_dev[i]) CudaWrapper_FreeCudaData(cuda->shared_data.comm.buf_send_dev[i],maxsend*sizeof(double));
- }
-
-
-}
-
-/* ----------------------------------------------------------------------
- free memory for multi-type swaps
-------------------------------------------------------------------------- */
-
-void CommCuda::free_multi()
-{
- CommBrick::free_multi();
- delete cu_multilo; cu_multilo = NULL;
- delete cu_multihi; cu_multihi = NULL;
-}
diff --git a/src/USER-CUDA/comm_cuda.h b/src/USER-CUDA/comm_cuda.h
deleted file mode 100644
index 5105018f32..0000000000
--- a/src/USER-CUDA/comm_cuda.h
+++ /dev/null
@@ -1,69 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifndef LMP_COMM_CUDA_H
-#define LMP_COMM_CUDA_H
-
-#include "pointers.h"
-
-#include "cuda_data.h"
-#include "comm_brick.h"
-
-namespace LAMMPS_NS {
-
-class CommCuda : public CommBrick {
-public:
- CommCuda(class LAMMPS *);
- ~CommCuda();
-
- virtual void init();
- virtual void setup(); // setup 3d communication pattern
- virtual void forward_comm(int mode=0); // forward communication of atom coords
- virtual void forward_comm_cuda();
- virtual void forward_comm_pack_cuda();
- virtual void forward_comm_transfer_cuda();
- virtual void forward_comm_unpack_cuda();
- virtual void forward_comm_pair(Pair *pair);
- virtual void reverse_comm(); // reverse communication of forces
- virtual void exchange(); // move atoms to new procs
- virtual void exchange_cuda(); // move atoms to new procs
- virtual void borders(); // setup list of atoms to communicate
- virtual void borders_cuda(); // setup list of atoms to communicate
- virtual void borders_cuda_overlap_forward_comm();
- virtual void forward_comm_fix(class Fix *, int size=0); // forward comm from a Fix
-
-
-
-
- protected:
- class Cuda *cuda;
- cCudaData<int, int, xy>* cu_pbc;
- cCudaData<double, X_CFLOAT, x>* cu_slablo;
- cCudaData<double, X_CFLOAT, x>* cu_slabhi;
- cCudaData<double, X_CFLOAT, xy>* cu_multilo;
- cCudaData<double, X_CFLOAT, xy>* cu_multihi;
-
- cCudaData<int, int, xy>* cu_sendlist;
- virtual void grow_send(int,int); // reallocate send buffer
- virtual void grow_recv(int); // free/allocate recv buffer
- virtual void grow_list(int, int); // reallocate one sendlist
- virtual void grow_swap(int); // grow swap and multi arrays
- virtual void allocate_swap(int); // allocate swap arrays
- virtual void allocate_multi(int); // allocate multi arrays
- virtual void free_swap(); // free swap arrays
- virtual void free_multi(); // free multi arrays
-};
-
-}
-
-#endif
diff --git a/src/USER-CUDA/compute_pe_cuda.cpp b/src/USER-CUDA/compute_pe_cuda.cpp
deleted file mode 100644
index b8661c9702..0000000000
--- a/src/USER-CUDA/compute_pe_cuda.cpp
+++ /dev/null
@@ -1,61 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <cstring>
-#include "compute_pe_cuda.h"
-#include "atom.h"
-#include "update.h"
-#include "force.h"
-#include "pair.h"
-#include "bond.h"
-#include "angle.h"
-#include "dihedral.h"
-#include "improper.h"
-#include "kspace.h"
-#include "modify.h"
-#include "domain.h"
-#include "error.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-ComputePECuda::ComputePECuda(LAMMPS *lmp, int narg, char **arg) :
- ComputePE(lmp, narg, arg)
-{
- cudable = 1;
-}
diff --git a/src/USER-CUDA/compute_pe_cuda.h b/src/USER-CUDA/compute_pe_cuda.h
deleted file mode 100644
index bc8b057762..0000000000
--- a/src/USER-CUDA/compute_pe_cuda.h
+++ /dev/null
@@ -1,59 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef COMPUTE_CLASS
-
-ComputeStyle(pe/cuda,ComputePECuda)
-
-#else
-
-#ifndef LMP_COMPUTE_PE_CUDA_H
-#define LMP_COMPUTE_PE_CUDA_H
-
-#include "compute_pe.h"
-
-namespace LAMMPS_NS {
-
-class ComputePECuda : public ComputePE {
- public:
- ComputePECuda(class LAMMPS *, int, char **);
- ~ComputePECuda() {}
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/compute_pressure_cuda.cpp b/src/USER-CUDA/compute_pressure_cuda.cpp
deleted file mode 100644
index c92e918ad0..0000000000
--- a/src/USER-CUDA/compute_pressure_cuda.cpp
+++ /dev/null
@@ -1,97 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <cstring>
-#include <cstdlib>
-#include "compute_pressure_cuda.h"
-#include "atom.h"
-#include "update.h"
-#include "domain.h"
-#include "modify.h"
-#include "fix.h"
-#include "force.h"
-#include "pair.h"
-#include "bond.h"
-#include "angle.h"
-#include "dihedral.h"
-#include "improper.h"
-#include "kspace.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-enum{DUMMY0,INVOKED_SCALAR,INVOKED_VECTOR,DUMMMY3,INVOKED_PERATOM};
-
-/* ---------------------------------------------------------------------- */
-
-ComputePressureCuda::ComputePressureCuda(LAMMPS *lmp, int narg, char **arg) :
- ComputePressure(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
- cudable = 1;
-
- // store temperature ID used by pressure computation
- // insure it is valid for temperature computation
-
- int n = strlen(arg[3]) + 1;
- char* id_temp = new char[n];
- strcpy(id_temp,arg[3]);
-
- int icompute = modify->find_compute(id_temp);
- delete [] id_temp;
- if (modify->compute[icompute]->cudable == 0)
- {
- error->warning(FLERR,"Compute pressure/cuda temperature ID is not cudable! Try a temp/cuda style.");
- cudable = 0;
- }
-
-}
-
-double ComputePressureCuda::compute_scalar()
-{
- if(not temperature->cudable && cuda->finished_setup) cuda->downloadAll();
- return ComputePressure::compute_scalar();
-}
-
-void ComputePressureCuda::compute_vector()
-{
- if(not temperature->cudable && cuda->finished_setup) cuda->downloadAll();
- ComputePressure::compute_vector();
-}
diff --git a/src/USER-CUDA/compute_pressure_cuda.h b/src/USER-CUDA/compute_pressure_cuda.h
deleted file mode 100644
index af48091708..0000000000
--- a/src/USER-CUDA/compute_pressure_cuda.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-#ifdef COMPUTE_CLASS
-
-ComputeStyle(pressure/cuda,ComputePressureCuda)
-
-#else
-
-#ifndef LMP_COMPUTE_PRESSURE_CUDA_H
-#define LMP_COMPUTE_PRESSURE_CUDA_H
-
-#include "compute_pressure.h"
-
-namespace LAMMPS_NS {
-
-class ComputePressureCuda : public ComputePressure {
- public:
- ComputePressureCuda(class LAMMPS *, int, char **);
- ~ComputePressureCuda() {}
- double compute_scalar();
- void compute_vector();
-
- private:
- class Cuda *cuda;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/compute_temp_cuda.cpp b/src/USER-CUDA/compute_temp_cuda.cpp
deleted file mode 100644
index 85afa07258..0000000000
--- a/src/USER-CUDA/compute_temp_cuda.cpp
+++ /dev/null
@@ -1,215 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "compute_temp_cuda.h"
-#include "compute_temp_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "force.h"
-#include "domain.h"
-#include "modify.h"
-#include "fix.h"
-#include "group.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-ComputeTempCuda::ComputeTempCuda(LAMMPS *lmp, int narg, char **arg) :
- Compute(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg != 3) error->all(FLERR,"Illegal compute temp/cuda command");
-
- scalar_flag = vector_flag = 1;
- size_vector = 6;
- extscalar = 0;
- extvector = 1;
- tempflag = 1;
-
- vector = new double[6];
- cu_t_vector = 0;
- cu_t_scalar = 0;
- cudable=true;
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-ComputeTempCuda::~ComputeTempCuda()
-{
- delete [] vector;
- delete cu_t_vector;
- delete cu_t_scalar;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void ComputeTempCuda::setup()
-{
- dynamic = 0;
- if (dynamic_user || group->dynamic[igroup]) dynamic = 1;
-
- fix_dof = 0;
- for (int i = 0; i < modify->nfix; i++)
- fix_dof += modify->fix[i]->dof(igroup);
- dof_compute();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void ComputeTempCuda::dof_compute()
-{
- double natoms = group->count(igroup);
- dof = domain->dimension * natoms;
- dof -= extra_dof + fix_dof;
- if (dof > 0.0) tfactor = force->mvv2e / (dof * force->boltz);
- else tfactor = 0.0;
-}
-
-/* ---------------------------------------------------------------------- */
-
-double ComputeTempCuda::compute_scalar()
-{
- if(cuda->begin_setup)
- {
- if(not cu_t_vector) cu_t_vector = new cCudaData<double, ENERGY_CFLOAT, x> (t_vector,6);
- if(not cu_t_scalar) cu_t_scalar = new cCudaData<double, ENERGY_CFLOAT, x> (&t_scalar,1);
- invoked_scalar = update->ntimestep;
- Cuda_ComputeTempCuda_Scalar(&cuda->shared_data,groupbit,(ENERGY_CFLOAT*) cu_t_scalar->dev_data());
- cu_t_scalar->download();
- }
- else
- {
- invoked_scalar = update->ntimestep;
-
- double **v = atom->v;
- double *mass = atom->mass;
- double *rmass = atom->rmass;
- int *type = atom->type;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- double t = 0.0;
-
- if (rmass) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- t += (v[i][0]*v[i][0] + v[i][1]*v[i][1] + v[i][2]*v[i][2]) * rmass[i];
- } else {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- t += (v[i][0]*v[i][0] + v[i][1]*v[i][1] + v[i][2]*v[i][2]) *
- mass[type[i]];
- }
- t_scalar=t;
- }
-
- MPI_Allreduce(&t_scalar,&scalar,1,MPI_DOUBLE,MPI_SUM,world);
- if (dynamic) dof_compute();
- scalar *= tfactor;
- if(scalar>1e15)
- {
- cuda->cu_v->download();
- cuda->cu_x->download();
- cuda->cu_type->download();
- double **v = atom->v;
- double **x = atom->x;
- printf("Out of v-range atoms: \n");
- for(int i=0;i<atom->nlocal;i++)
- if((v[i][0]*v[i][0] + v[i][1]*v[i][1] + v[i][2]*v[i][2])>1e5)
- printf("%i %i // %lf %lf %lf // %lf %lf %lf\n",atom->tag[i],atom->type[i],x[i][0], x[i][1], x[i][2],v[i][0], v[i][1], v[i][2]);
- error->all(FLERR,"Temperature out of range. Simulations will be abortet.\n");
- }
- return scalar;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void ComputeTempCuda::compute_vector()
-{
- int i;
- if(cuda->begin_setup)
- {
- if(not cu_t_vector) cu_t_vector = new cCudaData<double, ENERGY_CFLOAT, x> (t_vector,6);
- if(not cu_t_scalar) cu_t_scalar = new cCudaData<double, ENERGY_CFLOAT, x> (&t_scalar,1);
-
- invoked_vector = update->ntimestep;
-
- Cuda_ComputeTempCuda_Vector(&cuda->shared_data,groupbit,(ENERGY_CFLOAT*) cu_t_vector->dev_data());
- cu_t_vector->download();
- }
- else
- {
-
- invoked_vector = update->ntimestep;
-
- double **v = atom->v;
- double *mass = atom->mass;
- double *rmass = atom->rmass;
- int *type = atom->type;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- double massone,t[6];
- for (i = 0; i < 6; i++) t[i] = 0.0;
-
- for (i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- if (rmass) massone = rmass[i];
- else massone = mass[type[i]];
- t[0] += massone * v[i][0]*v[i][0];
- t[1] += massone * v[i][1]*v[i][1];
- t[2] += massone * v[i][2]*v[i][2];
- t[3] += massone * v[i][0]*v[i][1];
- t[4] += massone * v[i][0]*v[i][2];
- t[5] += massone * v[i][1]*v[i][2];
- }
-
- for (i = 0; i < 6; i++) t_vector[i]=t[i];
- }
- MPI_Allreduce(t_vector,vector,6,MPI_DOUBLE,MPI_SUM,world);
- for (i = 0; i < 6; i++) vector[i] *= force->mvv2e;
-}
diff --git a/src/USER-CUDA/compute_temp_cuda.h b/src/USER-CUDA/compute_temp_cuda.h
deleted file mode 100644
index 54b3338c08..0000000000
--- a/src/USER-CUDA/compute_temp_cuda.h
+++ /dev/null
@@ -1,76 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef COMPUTE_CLASS
-
-ComputeStyle(temp/cuda,ComputeTempCuda)
-
-#else
-
-#ifndef LMP_COMPUTE_TEMP_CUDA_H
-#define LMP_COMPUTE_TEMP_CUDA_H
-
-#include "compute.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class ComputeTempCuda : public Compute {
- public:
- ComputeTempCuda(class LAMMPS *, int, char **);
- ~ComputeTempCuda();
- void init() {}
- void setup();
- double compute_scalar();
- void compute_vector();
-
- private:
- class Cuda *cuda;
- int fix_dof;
- double tfactor;
-
- void dof_compute();
- double t_vector[6];
- double t_scalar;
- cCudaData<double , ENERGY_CFLOAT , x>* cu_t_scalar;
- cCudaData<double , ENERGY_CFLOAT , x>* cu_t_vector;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/compute_temp_partial_cuda.cpp b/src/USER-CUDA/compute_temp_partial_cuda.cpp
deleted file mode 100644
index b366c546f6..0000000000
--- a/src/USER-CUDA/compute_temp_partial_cuda.cpp
+++ /dev/null
@@ -1,360 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "compute_temp_partial_cuda.h"
-#include "compute_temp_partial_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "force.h"
-#include "domain.h"
-#include "modify.h"
-#include "fix.h"
-#include "group.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-ComputeTempPartialCuda::ComputeTempPartialCuda(LAMMPS *lmp, int narg, char **arg) :
- Compute(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg != 6) error->all(FLERR,"Illegal compute temp/partial command");
-
- scalar_flag = vector_flag = 1;
- size_vector = 6;
- extscalar = 0;
- extvector = 1;
- tempflag = 1;
- tempbias = 1;
-
- xflag = force->inumeric(FLERR,arg[3]);
- yflag = force->inumeric(FLERR,arg[4]);
- zflag = force->inumeric(FLERR,arg[5]);
- if (zflag && domain->dimension == 2)
- error->all(FLERR,"Compute temp/partial cannot use vz for 2d systemx");
-
- maxbias = 0;
- vbiasall = NULL;
-
- vector = new double[6];
- cu_t_vector = 0;
- cu_t_scalar = 0;
- cu_vbiasall=NULL;
- cudable=true;
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-ComputeTempPartialCuda::~ComputeTempPartialCuda()
-{
- memory->destroy(vbiasall);
- delete [] vector;
- delete cu_t_vector;
- delete cu_t_scalar;
- delete cu_vbiasall;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void ComputeTempPartialCuda::setup()
-{
- dynamic = 0;
- if (dynamic_user || group->dynamic[igroup]) dynamic = 1;
-
- fix_dof = 0;
- for (int i = 0; i < modify->nfix; i++)
- fix_dof += modify->fix[i]->dof(igroup);
- dof_compute();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void ComputeTempPartialCuda::dof_compute()
-{
- double natoms = group->count(igroup);
- int nper = xflag+yflag+zflag;
- dof = nper * natoms;
- dof -= (1.0*nper/domain->dimension)*fix_dof + extra_dof;
- if (dof > 0) tfactor = force->mvv2e / (dof * force->boltz);
- else tfactor = 0.0;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int ComputeTempPartialCuda::dof_remove(int i)
-{
- int nper = xflag+yflag+zflag;
- return (domain->dimension - nper);
-}
-
-/* ---------------------------------------------------------------------- */
-
-double ComputeTempPartialCuda::compute_scalar()
-{
- if(cuda->begin_setup)
- {
- if(not cu_t_vector) cu_t_vector = new cCudaData<double, ENERGY_CFLOAT, x> (t_vector,6);
- if(not cu_t_scalar) cu_t_scalar = new cCudaData<double, ENERGY_CFLOAT, x> (&t_scalar,1);
- invoked_scalar = update->ntimestep;
- Cuda_ComputeTempPartialCuda_Scalar(&cuda->shared_data,groupbit,(ENERGY_CFLOAT*) cu_t_scalar->dev_data(),xflag,yflag,zflag);
- cu_t_scalar->download();
- }
- else
- {
- invoked_scalar = update->ntimestep;
-
- double **v = atom->v;
- double *mass = atom->mass;
- double *rmass = atom->rmass;
- int *type = atom->type;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- double t = 0.0;
-
- if (rmass) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- t += (xflag*v[i][0]*v[i][0] + yflag*v[i][1]*v[i][1] + zflag*v[i][2]*v[i][2]) * rmass[i];
- } else {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- t += (xflag*v[i][0]*v[i][0] + yflag*v[i][1]*v[i][1] + zflag*v[i][2]*v[i][2]) *
- mass[type[i]];
- }
- t_scalar=t;
- }
-
- MPI_Allreduce(&t_scalar,&scalar,1,MPI_DOUBLE,MPI_SUM,world);
- if (dynamic) dof_compute();
- scalar *= tfactor;
- if(scalar>1e15)
- {
- cuda->cu_v->download();
- cuda->cu_x->download();
- cuda->cu_type->download();
- double **v = atom->v;
- double **x = atom->x;
- printf("Out of v-range atoms: \n");
- for(int i=0;i<atom->nlocal;i++)
- if((v[i][0]*v[i][0] + v[i][1]*v[i][1] + v[i][2]*v[i][2])>1e5)
- printf("%i %i // %lf %lf %lf // %lf %lf %lf\n",atom->tag[i],atom->type[i],x[i][0], x[i][1], x[i][2],v[i][0], v[i][1], v[i][2]);
- error->all(FLERR,"Temperature out of range. Simulations will be abortet.\n");
- }
- return scalar;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void ComputeTempPartialCuda::compute_vector()
-{
- int i;
- if(cuda->begin_setup)
- {
- if(not cu_t_vector) cu_t_vector = new cCudaData<double, ENERGY_CFLOAT, x> (t_vector,6);
- if(not cu_t_scalar) cu_t_scalar = new cCudaData<double, ENERGY_CFLOAT, x> (&t_scalar,1);
-
- invoked_vector = update->ntimestep;
-
- Cuda_ComputeTempPartialCuda_Vector(&cuda->shared_data,groupbit,(ENERGY_CFLOAT*) cu_t_vector->dev_data(),xflag,yflag,zflag);
- cu_t_vector->download();
- }
- else
- {
-
- invoked_vector = update->ntimestep;
-
- double **v = atom->v;
- double *mass = atom->mass;
- double *rmass = atom->rmass;
- int *type = atom->type;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- double massone,t[6];
- for (i = 0; i < 6; i++) t[i] = 0.0;
-
- for (i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- if (rmass) massone = rmass[i];
- else massone = mass[type[i]];
- t[0] += massone * xflag*v[i][0]*v[i][0];
- t[1] += massone * yflag*v[i][1]*v[i][1];
- t[2] += massone * zflag*v[i][2]*v[i][2];
- t[3] += massone * xflag*yflag*v[i][0]*v[i][1];
- t[4] += massone * xflag*zflag*v[i][0]*v[i][2];
- t[5] += massone * yflag*zflag*v[i][1]*v[i][2];
- }
-
- for (i = 0; i < 6; i++) t_vector[i]=t[i];
- }
- MPI_Allreduce(t_vector,vector,6,MPI_DOUBLE,MPI_SUM,world);
- for (i = 0; i < 6; i++) vector[i] *= force->mvv2e;
-}
-
-/* ----------------------------------------------------------------------
- remove velocity bias from atom I to leave thermal velocity
-------------------------------------------------------------------------- */
-
-void ComputeTempPartialCuda::remove_bias(int i, double *v)
-{
- if (!xflag) {
- vbias[0] = v[0];
- v[0] = 0.0;
- }
- if (!yflag) {
- vbias[1] = v[1];
- v[1] = 0.0;
- }
- if (!zflag) {
- vbias[2] = v[2];
- v[2] = 0.0;
- }
-}
-
-/* ----------------------------------------------------------------------
- remove velocity bias from all atoms to leave thermal velocity
-------------------------------------------------------------------------- */
-
-void ComputeTempPartialCuda::remove_bias_all()
-{
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- if (nlocal > maxbias) {
- memory->destroy(vbiasall);
- maxbias = atom->nmax;
- memory->create(vbiasall,maxbias,3,"temp/partial:vbiasall");
- delete cu_vbiasall;
- cu_vbiasall = new cCudaData<double, V_CFLOAT, yx> ((double*)vbiasall, atom->nmax, 3);
- }
- if(cuda->begin_setup)
- {
- Cuda_ComputeTempPartialCuda_RemoveBiasAll(&cuda->shared_data,groupbit,xflag,yflag,zflag,cu_vbiasall->dev_data());
- }
- else
- {
- if (!xflag) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- vbiasall[i][0] = v[i][0];
- v[i][0] = 0.0;
- }
- }
- if (!yflag) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- vbiasall[i][1] = v[i][1];
- v[i][1] = 0.0;
- }
- }
- if (!zflag) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- vbiasall[i][2] = v[i][2];
- v[i][2] = 0.0;
- }
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- add back in velocity bias to atom I removed by remove_bias()
- assume remove_bias() was previously called
-------------------------------------------------------------------------- */
-
-void ComputeTempPartialCuda::restore_bias(int i, double *v)
-{
- if (!xflag) v[0] += vbias[0];
- if (!yflag) v[1] += vbias[1];
- if (!zflag) v[2] += vbias[2];
-}
-
-/* ----------------------------------------------------------------------
- add back in velocity bias to all atoms removed by remove_bias_all()
- assume remove_bias_all() was previously called
-------------------------------------------------------------------------- */
-
-void ComputeTempPartialCuda::restore_bias_all()
-{
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
- if(cuda->begin_setup)
- {
- Cuda_ComputeTempPartialCuda_RestoreBiasAll(&cuda->shared_data,groupbit,xflag,yflag,zflag,cu_vbiasall->dev_data());
- }
- else
- {
-
- if (!xflag) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- v[i][0] += vbiasall[i][0];
- }
- if (!yflag) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- v[i][1] += vbiasall[i][1];
- }
- if (!zflag) {
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- v[i][2] += vbiasall[i][2];
- }
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-double ComputeTempPartialCuda::memory_usage()
-{
- double bytes = maxbias * sizeof(double);
- return bytes;
-}
diff --git a/src/USER-CUDA/compute_temp_partial_cuda.h b/src/USER-CUDA/compute_temp_partial_cuda.h
deleted file mode 100644
index 320bf17858..0000000000
--- a/src/USER-CUDA/compute_temp_partial_cuda.h
+++ /dev/null
@@ -1,84 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef COMPUTE_CLASS
-
-ComputeStyle(temp/partial/cuda,ComputeTempPartialCuda)
-
-#else
-
-#ifndef LMP_COMPUTE_TEMP_PARTIAL_CUDA_H
-#define LMP_COMPUTE_TEMP_PARTIAL_CUDA_H
-
-#include "compute.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class ComputeTempPartialCuda : public Compute {
- public:
- ComputeTempPartialCuda(class LAMMPS *, int, char **);
- ~ComputeTempPartialCuda();
- void init() {}
- void setup();
- double compute_scalar();
- void compute_vector();
-
- int dof_remove(int);
- void remove_bias(int, double *);
- void remove_bias_all();
- void restore_bias(int, double *);
- void restore_bias_all();
- double memory_usage();
-
- private:
- class Cuda *cuda;
- int xflag,yflag,zflag;
- int fix_dof;
- double tfactor;
-
- void dof_compute();
- double t_vector[6];
- double t_scalar;
- cCudaData<double , ENERGY_CFLOAT , x>* cu_t_scalar;
- cCudaData<double , ENERGY_CFLOAT , x>* cu_t_vector;
- cCudaData<double, V_CFLOAT, yx>* cu_vbiasall;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/cuda.cpp b/src/USER-CUDA/cuda.cpp
deleted file mode 100644
index 995289a792..0000000000
--- a/src/USER-CUDA/cuda.cpp
+++ /dev/null
@@ -1,1067 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "user_cuda.h"
-#include "atom.h"
-#include "domain.h"
-#include "force.h"
-#include "pair.h"
-#include "update.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "universe.h"
-#include "input.h"
-#include "atom_masks.h"
-#include "error.h"
-
-#include "cuda_neigh_list.h"
-//#include "pre_binning_cu.h"
-//#include "reverse_binning_cu.h"
-#include <ctime>
-#include <cmath>
-#include "cuda_pair_cu.h"
-#include "cuda_cu.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-Cuda::Cuda(LAMMPS* lmp) : Pointers(lmp)
-{
- cuda_exists = true;
- lmp->cuda = this;
-
- if (universe->me == 0) printf("# Using LAMMPS_CUDA \n");
-
- shared_data.me = universe->me;
-
- device_set = false;
- devicelist = NULL;
-
- Cuda_Cuda_GetCompileSettings(&shared_data);
-
- if (universe->me == 0) {
-
- if(shared_data.compile_settings.prec_glob != sizeof(CUDA_CFLOAT) / 4)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: Global Precision: cuda %i cpp %i\n\n",
- shared_data.compile_settings.prec_glob, (int) sizeof(CUDA_CFLOAT) / 4);
-
- if(shared_data.compile_settings.prec_x != sizeof(X_CFLOAT) / 4)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: X Precision: cuda %i cpp %i\n\n",
- shared_data.compile_settings.prec_x, (int) sizeof(X_CFLOAT) / 4);
-
- if(shared_data.compile_settings.prec_v != sizeof(V_CFLOAT) / 4)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: V Precision: cuda %i cpp %i\n\n",
- shared_data.compile_settings.prec_v, (int) sizeof(V_CFLOAT) / 4);
-
- if(shared_data.compile_settings.prec_f != sizeof(F_CFLOAT) / 4)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: F Precision: cuda %i cpp %i\n\n",
- shared_data.compile_settings.prec_f, (int) sizeof(F_CFLOAT) / 4);
-
- if(shared_data.compile_settings.prec_pppm != sizeof(PPPM_CFLOAT) / 4)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: PPPM Precision: cuda %i cpp %i\n\n",
- shared_data.compile_settings.prec_pppm, (int) sizeof(PPPM_CFLOAT) / 4);
-
- if(shared_data.compile_settings.prec_fft != sizeof(FFT_CFLOAT) / 4)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: FFT Precision: cuda %i cpp %i\n\n",
- shared_data.compile_settings.prec_fft, (int) sizeof(FFT_CFLOAT) / 4);
-
-#ifdef FFT_CUFFT
- if(shared_data.compile_settings.cufft != 1)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: cufft: cuda %i cpp %i\n\n",
- shared_data.compile_settings.cufft, 1);
-#else
- if(shared_data.compile_settings.cufft != 0)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: cufft: cuda %i cpp %i\n\n",
- shared_data.compile_settings.cufft, 0);
-#endif
-
- if(shared_data.compile_settings.arch != CUDA_ARCH)
- printf("\n\n # CUDA WARNING: Compile Settings of cuda and cpp code differ! \n"
- " # CUDA WARNING: arch: cuda %i cpp %i\n\n",
- shared_data.compile_settings.cufft, CUDA_ARCH);
- }
-
- cu_x = 0;
- cu_v = 0;
- cu_f = 0;
- cu_tag = 0;
- cu_type = 0;
- cu_mask = 0;
- cu_image = 0;
- cu_xhold = 0;
- cu_q = 0;
- cu_rmass = 0;
- cu_mass = 0;
- cu_virial = 0;
- cu_eatom = 0;
- cu_vatom = 0;
- cu_radius = 0;
- cu_density = 0;
- cu_omega = 0;
- cu_torque = 0;
-
- cu_special = 0;
- cu_nspecial = 0;
-
- cu_molecule = 0;
-
- cu_x_type = 0;
- x_type = 0;
- cu_v_radius = 0;
- v_radius = 0;
- cu_omega_rmass = 0;
- omega_rmass = 0;
-
- binned_id = 0;
- cu_binned_id = 0;
- binned_idnew = 0;
- cu_binned_idnew = 0;
-
- cu_map_array = 0;
-
- copy_buffer = 0;
- copy_buffersize = 0;
-
- neighbor_decide_by_integrator = 0;
- pinned = true;
-
- debugdata = 0;
-
- finished_setup = false;
- begin_setup = false;
- finished_run = false;
-
- setSharedDataZero();
-
- uploadtime = 0;
- downloadtime = 0;
- dotiming = false;
-
- dotestatom = false;
- testatom = 0;
- oncpu = true;
-
- self_comm = 0;
- MYDBG(printf("# CUDA: Cuda::Cuda Done...\n");)
- //cCudaData<double, float, yx >
-}
-
-/* ---------------------------------------------------------------------- */
-
-Cuda::~Cuda()
-{
- print_timings();
-
- if (universe->me == 0) printf("# CUDA: Free memory...\n");
-
- delete [] devicelist;
-
- delete cu_q;
- delete cu_x;
- delete cu_v;
- delete cu_f;
- delete cu_tag;
- delete cu_type;
- delete cu_mask;
- delete cu_image;
- delete cu_xhold;
- delete cu_mass;
- delete cu_rmass;
- delete cu_virial;
- delete cu_eng_vdwl;
- delete cu_eng_coul;
- delete cu_extent;
- delete cu_eatom;
- delete cu_vatom;
- delete cu_radius;
- delete cu_density;
- delete cu_omega;
- delete cu_torque;
- delete cu_molecule;
-
- delete cu_x_type;
- delete [] x_type;
- delete cu_v_radius;
- delete [] v_radius;
- delete cu_omega_rmass;
- delete [] omega_rmass;
-
- delete cu_debugdata;
- delete[] debugdata;
-
- delete cu_map_array;
-
- std::map<NeighList*, CudaNeighList*>::iterator p = neigh_lists.begin();
-
- while(p != neigh_lists.end()) {
- delete p->second;
- ++p;
- }
-}
-
-/* ----------------------------------------------------------------------
- package cuda command
- can be invoked multiple times: -c on, -pk, package command
- can only init GPUs once in activate(), so just store params here
-------------------------------------------------------------------------- */
-
-void Cuda::accelerator(int narg, char **arg)
-{
- // this error should not happen
-
- if (device_set) error->all(FLERR,"USER-CUDA device is already activated");
-
- // pppn = # of GPUs/node
-
- pppn = force->inumeric(FLERR,arg[0]);
- if (pppn <= 0) error->all(FLERR,"Illegal package cuda command");
-
- // optional args
-
- delete [] devicelist;
- devicelist = NULL;
- int newtonflag = 0;
-
- int iarg = 1;
- while (iarg < narg) {
- if (strcmp(arg[iarg],"newton") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal package cuda command");
- if (strcmp(arg[iarg+1],"off") == 0) newtonflag = 0;
- else if (strcmp(arg[iarg+1],"on") == 0) newtonflag = 1;
- else error->all(FLERR,"Illegal package cuda command");
- } else if (strcmp(arg[iarg],"gpuID") == 0) {
- if (iarg+pppn+1 > narg) error->all(FLERR,"Illegal package cuda command");
- devicelist = new int[pppn];
- for (int k = 0; k < pppn; k++)
- devicelist[k] = force->inumeric(FLERR,arg[iarg+k+1]);
- iarg += pppn + 1;
- } else if (strcmp(arg[iarg],"timing") == 0) {
- dotiming = true;
- iarg++;
- } else if (strcmp(arg[iarg],"test") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal package cuda command");
- testatom = force->numeric(FLERR,arg[iarg+1]);
- dotestatom = true;
- iarg += 2;
- } else if (strcmp(arg[iarg],"thread") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal package cuda command");
- if (strcmp(arg[iarg+1],"auto") == 0)
- shared_data.pair.override_block_per_atom = -1;
- else if (strcmp(arg[iarg+1],"tpa") == 0)
- shared_data.pair.override_block_per_atom = 0;
- else if (strcmp(arg[iarg+1],"bpa") == 0)
- shared_data.pair.override_block_per_atom = 1;
- else error->all(FLERR,"Illegal package cuda command");
- iarg += 2;
- }
-
- // undocumented options
-
- else if (strcmp(arg[iarg],"suffix") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal package cuda command");
- strcpy(lmp->suffix,arg[iarg+1]);
- iarg += 2;
- } else if (strcmp(arg[iarg],"overlap_comm") == 0) {
- shared_data.overlap_comm = 1;
- iarg++;
- } else if (strcmp(arg[iarg],"pinned") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal package cuda command");
- pinned = force->inumeric(FLERR,arg[iarg+1]) == 0 ? false : true;
- if ((pinned == false) && (universe->me == 0))
- printf(" #CUDA: Pinned memory is not used for communication\n");
- iarg += 2;
- } else error->all(FLERR,"Illegal package cuda command");
- }
-
- // set newton flags
-
- force->newton = force->newton_pair = force->newton_bond = newtonflag;
-}
-
-/* ----------------------------------------------------------------------
- activate the GPUs
- only done once with whatever settings used by the last package command
-------------------------------------------------------------------------- */
-
-void Cuda::activate()
-{
- if (device_set) return;
- device_set = true;
-
- if (universe->me == 0) printf("# CUDA: Activate GPU \n");
-
- CudaWrapper_Init(0, (char**)0, universe->me, pppn, devicelist);
- //if(shared_data.overlap_comm)
- CudaWrapper_AddStreams(3);
- cu_x = 0;
- cu_v = 0;
- cu_f = 0;
- cu_tag = 0;
- cu_type = 0;
- cu_mask = 0;
- cu_image = 0;
- cu_xhold = 0;
- cu_q = 0;
- cu_rmass = 0;
- cu_mass = 0;
- cu_virial = 0;
- cu_eatom = 0;
- cu_vatom = 0;
- cu_radius = 0;
- cu_density = 0;
- cu_omega = 0;
- cu_torque = 0;
-
- cu_special = 0;
- cu_nspecial = 0;
-
- cu_molecule = 0;
-
- cu_x_type = 0;
- cu_v_radius = 0;
- cu_omega_rmass = 0;
-
- cu_binned_id = 0;
- cu_binned_idnew = 0;
- allocate();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void Cuda::setSharedDataZero()
-{
- MYDBG(printf("# CUDA: Cuda::setSharedDataZero ...\n");)
- shared_data.atom.nlocal = 0;
- shared_data.atom.nghost = 0;
- shared_data.atom.nall = 0;
- shared_data.atom.nmax = 0;
- shared_data.atom.ntypes = 0;
- shared_data.atom.q_flag = 0;
- shared_data.atom.need_eatom = 0;
- shared_data.atom.need_vatom = 0;
- shared_data.atom.update_nmax = 1;
- shared_data.atom.update_nlocal = 1;
- shared_data.atom.update_neigh = 1;
-
- shared_data.pair.cudable_force = 0;
- shared_data.pair.collect_forces_later = 0;
- shared_data.pair.use_block_per_atom = 0;
- shared_data.pair.override_block_per_atom = -1;
- shared_data.pair.cut = 0;
- shared_data.pair.cutsq = 0;
- shared_data.pair.cut_inner = 0;
- shared_data.pair.cut_coul = 0;
- shared_data.pair.special_lj = 0;
- shared_data.pair.special_coul = 0;
-
- shared_data.pair.neighall = false;
-
- shared_data.pppm.cudable_force = 0;
-
- shared_data.buffersize = 0;
- shared_data.buffer_new = 1;
- shared_data.buffer = NULL;
-
- shared_data.comm.comm_phase = 0;
- shared_data.overlap_comm = 0;
-
- shared_data.comm.buffer = NULL;
- shared_data.comm.buffer_size = 0;
- shared_data.comm.overlap_split_ratio = 0;
- // setTimingsZero();
-}
-
-void Cuda::allocate()
-{
- MYDBG(printf("# CUDA: Cuda::allocate ...\n");)
-
- if(not cu_virial) {
- cu_virial = new cCudaData<double, ENERGY_CFLOAT, x > (NULL, & shared_data.pair.virial , 6);
- cu_eng_vdwl = new cCudaData<double, ENERGY_CFLOAT, x > (NULL, & shared_data.pair.eng_vdwl , 1);
- cu_eng_coul = new cCudaData<double, ENERGY_CFLOAT, x > (NULL, & shared_data.pair.eng_coul , 1);
- cu_extent = new cCudaData<double, double, x> (extent, 6);
- shared_data.flag = CudaWrapper_AllocCudaData(sizeof(int));
- int size = 2 * CUDA_MAX_DEBUG_SIZE;
- debugdata = new int[size];
- cu_debugdata = new cCudaData<int, int, x > (debugdata , size);
- shared_data.debugdata = cu_debugdata->dev_data();
- }
-
- checkResize();
- setSystemParams();
- MYDBG(printf("# CUDA: Cuda::allocate done...\n");)
-}
-
-void Cuda::setSystemParams()
-{
- MYDBG(printf("# CUDA: Cuda::setSystemParams ...\n");)
- shared_data.atom.nlocal = atom->nlocal;
- shared_data.atom.nghost = atom->nghost;
- shared_data.atom.nall = atom->nlocal + atom->nghost;
- shared_data.atom.ntypes = atom->ntypes;
- shared_data.atom.q_flag = atom->q_flag;
- shared_data.atom.rmass_flag = atom->rmass_flag;
- MYDBG(printf("# CUDA: Cuda::setSystemParams done ...\n");)
-}
-
-void Cuda::setDomainParams()
-{
- MYDBG(printf("# CUDA: Cuda::setDomainParams ...\n");)
- cuda_shared_domain* cu_domain = &shared_data.domain;
-
- cu_domain->triclinic = domain->triclinic;
-
- for(short i = 0; i < 3; ++i) {
- cu_domain->periodicity[i] = domain->periodicity[i];
- cu_domain->sublo[i] = domain->sublo[i];
- cu_domain->subhi[i] = domain->subhi[i];
- cu_domain->boxlo[i] = domain->boxlo[i];
- cu_domain->boxhi[i] = domain->boxhi[i];
- cu_domain->prd[i] = domain->prd[i];
- }
-
- if(domain->triclinic) {
- for(short i = 0; i < 3; ++i) {
- cu_domain->boxlo_lamda[i] = domain->boxlo_lamda[i];
- cu_domain->boxhi_lamda[i] = domain->boxhi_lamda[i];
- cu_domain->prd_lamda[i] = domain->prd_lamda[i];
- cu_domain->sublo[i] = domain->sublo_lamda[i];
- cu_domain->subhi[i] = domain->subhi_lamda[i];
- }
-
- cu_domain->xy = domain->xy;
- cu_domain->xz = domain->xz;
- cu_domain->yz = domain->yz;
- }
-
- for(int i = 0; i < 6; i++) {
- cu_domain->h[i] = domain->h[i];
- cu_domain->h_inv[i] = domain->h_inv[i];
- cu_domain->h_rate[i] = domain->h_rate[i];
- }
-
- cu_domain->update = 2;
- MYDBG(printf("# CUDA: Cuda::setDomainParams done ...\n");)
-}
-
-void Cuda::checkResize()
-{
- MYDBG(printf("# CUDA: Cuda::checkResize ...\n");)
- cuda_shared_atom* cu_atom = & shared_data.atom;
- cu_atom->q_flag = atom->q_flag;
- cu_atom->rmass_flag = atom->rmass ? 1 : 0;
- cu_atom->nall = atom->nlocal + atom->nghost;
- cu_atom->nlocal = atom->nlocal;
- cu_atom->nghost = atom->nghost;
-
- // do we have more atoms to upload than currently allocated memory on device? (also true if nothing yet allocated)
- if(atom->nmax > cu_atom->nmax || cu_tag == NULL) {
- delete cu_x;
- cu_x = new cCudaData<double, X_CFLOAT, yx> ((double*)atom->x , & cu_atom->x , atom->nmax, 3, 0, true); //cu_x->set_buffer(&(shared_data.buffer),&(shared_data.buffersize),true);
- delete cu_v;
- cu_v = new cCudaData<double, V_CFLOAT, yx> ((double*)atom->v, & cu_atom->v , atom->nmax, 3);
- delete cu_f;
- cu_f = new cCudaData<double, F_CFLOAT, yx> ((double*)atom->f, & cu_atom->f , atom->nmax, 3, 0, true);
- delete cu_tag;
- cu_tag = new cCudaData<int , int , x > (atom->tag , & cu_atom->tag , atom->nmax, 0, true);
- delete cu_type;
- cu_type = new cCudaData<int , int , x > (atom->type , & cu_atom->type , atom->nmax, 0, true);
- delete cu_mask;
- cu_mask = new cCudaData<int , int , x > (atom->mask , & cu_atom->mask , atom->nmax, 0, true);
- delete cu_image;
- cu_image = new cCudaData<int , int , x > (atom->image , & cu_atom->image , atom->nmax, 0, true);
-
- if(atom->rmass) {
- delete cu_rmass;
- cu_rmass = new cCudaData<double, V_CFLOAT, x > (atom->rmass , & cu_atom->rmass , atom->nmax);
- }
-
- if(cu_atom->q_flag) {
- delete cu_q;
- cu_q = new cCudaData<double, F_CFLOAT, x > ((double*)atom->q, & cu_atom->q , atom->nmax, 0 , true);
- }// cu_q->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
-
- if(atom->radius) {
- delete cu_radius;
- cu_radius = new cCudaData<double, X_CFLOAT, x > (atom->radius , & cu_atom->radius , atom->nmax);
- delete cu_v_radius;
- cu_v_radius = new cCudaData<V_CFLOAT, V_CFLOAT, x> (v_radius , & cu_atom->v_radius , atom->nmax * 4);
- delete cu_omega_rmass;
- cu_omega_rmass = new cCudaData<V_CFLOAT, V_CFLOAT, x> (omega_rmass , & cu_atom->omega_rmass , atom->nmax * 4);
- }
-
- if(atom->omega) {
- delete cu_omega;
- cu_omega = new cCudaData<double, V_CFLOAT, yx > (((double*) atom->omega) , & cu_atom->omega , atom->nmax, 3);
- }
-
- if(atom->torque) {
- delete cu_torque;
- cu_torque = new cCudaData<double, F_CFLOAT, yx > (((double*) atom->torque) , & cu_atom->torque , atom->nmax, 3);
- }
-
- if(atom->special) {
- delete cu_special;
- cu_special = new cCudaData<int, int, yx > (((int*) & (atom->special[0][0])) , & cu_atom->special , atom->nmax, atom->maxspecial, 0 , true);
- shared_data.atom.maxspecial = atom->maxspecial;
- }
-
- if(atom->nspecial) {
- delete cu_nspecial;
- cu_nspecial = new cCudaData<int, int, yx > (((int*) atom->nspecial) , & cu_atom->nspecial , atom->nmax, 3, 0, true);
- }
-
- if(atom->molecule) {
- delete cu_molecule;
- cu_molecule = new cCudaData<int, int, x > (((int*) atom->molecule) , & cu_atom->molecule , atom->nmax, 0 , true);
- }
-
- shared_data.atom.special_flag = neighbor->special_flag;
- shared_data.atom.molecular = atom->molecular;
-
- cu_atom->update_nmax = 2;
- cu_atom->nmax = atom->nmax;
-
- delete cu_x_type;
- cu_x_type = new cCudaData<X_CFLOAT, X_CFLOAT, x> (x_type , & cu_atom->x_type , atom->nmax * 4);
- }
-
- if(((cu_xhold == NULL) || (cu_xhold->get_dim()[0] < neighbor->maxhold)) && neighbor->xhold) {
- delete cu_xhold;
- cu_xhold = new cCudaData<double, X_CFLOAT, yx> ((double*)neighbor->xhold, & cu_atom->xhold , neighbor->maxhold, 3);
- shared_data.atom.maxhold = neighbor->maxhold;
- }
-
- if(atom->mass && !cu_mass) {
- cu_mass = new cCudaData<double, V_CFLOAT, x > (atom->mass , & cu_atom->mass , atom->ntypes + 1);
- }
-
- cu_atom->mass_host = atom->mass;
-
- if(atom->map_style == 1) {
- if(cu_map_array == NULL) {
- cu_map_array = new cCudaData<int, int, x > (atom->get_map_array() , & cu_atom->map_array , atom->get_map_size());
- } else if(cu_map_array->dev_size() / sizeof(int) < atom->get_map_size()) {
- delete cu_map_array;
- cu_map_array = new cCudaData<int, int, x > (atom->get_map_array() , & cu_atom->map_array , atom->get_map_size());
- }
- }
-
-
- // if any of the host pointers have changed (e.g. re-allocated somewhere else), set to correct pointer
- if(cu_x ->get_host_data() != atom->x) cu_x ->set_host_data((double*)(atom->x));
-
- if(cu_v ->get_host_data() != atom->v) cu_v ->set_host_data((double*)(atom->v));
-
- if(cu_f ->get_host_data() != atom->f) cu_f ->set_host_data((double*)(atom->f));
-
- if(cu_tag ->get_host_data() != atom->tag) cu_tag ->set_host_data(atom->tag);
-
- if(cu_type->get_host_data() != atom->type) cu_type->set_host_data(atom->type);
-
- if(cu_mask->get_host_data() != atom->mask) cu_mask->set_host_data(atom->mask);
-
- if(cu_image->get_host_data() != atom->image) cu_mask->set_host_data(atom->image);
-
- if(cu_xhold)
- if(cu_xhold->get_host_data() != neighbor->xhold) cu_xhold->set_host_data((double*)(neighbor->xhold));
-
- if(atom->rmass)
- if(cu_rmass->get_host_data() != atom->rmass) cu_rmass->set_host_data((double*)(atom->rmass));
-
- if(cu_atom->q_flag)
- if(cu_q->get_host_data() != atom->q) cu_q->set_host_data((double*)(atom->q));
-
- if(atom->radius)
- if(cu_radius->get_host_data() != atom->radius) cu_radius->set_host_data((double*)(atom->radius));
-
- if(atom->omega)
- if(cu_omega->get_host_data() != atom->omega) cu_omega->set_host_data((double*)(atom->omega));
-
- if(atom->torque)
- if(cu_torque->get_host_data() != atom->torque) cu_torque->set_host_data((double*)(atom->torque));
-
- if(atom->special)
- if(cu_special->get_host_data() != atom->special) {
- delete cu_special;
- cu_special = new cCudaData<int, int, yx > (((int*) atom->special) , & cu_atom->special , atom->nmax, atom->maxspecial);
- shared_data.atom.maxspecial = atom->maxspecial;
- }
-
- if(atom->nspecial)
- if(cu_nspecial->get_host_data() != atom->nspecial) cu_nspecial->set_host_data((int*)(atom->nspecial));
-
- if(atom->molecule)
- if(cu_molecule->get_host_data() != atom->molecule) cu_molecule->set_host_data((int*)(atom->molecule));
-
- if(force)
- if(cu_virial ->get_host_data() != force->pair->virial) cu_virial ->set_host_data(force->pair->virial);
-
- if(force)
- if(cu_eng_vdwl ->get_host_data() != &force->pair->eng_vdwl) cu_eng_vdwl ->set_host_data(&force->pair->eng_vdwl);
-
- if(force)
- if(cu_eng_coul ->get_host_data() != &force->pair->eng_coul) cu_eng_coul ->set_host_data(&force->pair->eng_coul);
-
- cu_atom->update_nlocal = 2;
- MYDBG(printf("# CUDA: Cuda::checkResize done...\n");)
-}
-
-void Cuda::evsetup_eatom_vatom(int eflag_atom, int vflag_atom)
-{
- if(eflag_atom) {
- if(not cu_eatom)
- cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > (force->pair->eatom, & (shared_data.atom.eatom) , atom->nmax); // cu_eatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
-
- if(cu_eatom->get_dim()[0] != atom->nmax) {
- //delete cu_eatom;
- //cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > (force->pair->eatom, & (shared_data.atom.eatom) , atom->nmax );// cu_eatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
- shared_data.atom.update_nmax = 2;
- }
-
- cu_eatom->set_host_data(force->pair->eatom);
- cu_eatom->memset_device(0);
- }
-
- if(vflag_atom) {
- if(not cu_vatom)
- cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)force->pair->vatom, & (shared_data.atom.vatom) , atom->nmax , 6);// cu_vatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
-
- if(cu_vatom->get_dim()[0] != atom->nmax) {
- //delete cu_vatom;
- //cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)force->pair->vatom, & (shared_data.atom.vatom) , atom->nmax ,6 );// cu_vatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
- shared_data.atom.update_nmax = 2;
- }
-
- cu_vatom->set_host_data((double*)force->pair->vatom);
- cu_vatom->memset_device(0);
- }
-}
-
-void Cuda::uploadAll()
-{
- MYDBG(printf("# CUDA: Cuda::uploadAll() ... start\n");)
- my_times starttime;
- my_times endtime;
-
- if(atom->nmax != shared_data.atom.nmax) checkResize();
-
- my_gettime(CLOCK_REALTIME, &starttime);
- cu_x ->upload();
- cu_v ->upload();
- cu_f ->upload();
- cu_tag ->upload();
- cu_type->upload();
- cu_mask->upload();
- cu_image->upload();
-
- if(shared_data.atom.q_flag) cu_q ->upload();
-
- if(atom->rmass) cu_rmass->upload();
-
- if(atom->radius) cu_radius->upload();
-
- if(atom->omega) cu_omega->upload();
-
- if(atom->torque) cu_torque->upload();
-
- if(atom->special) cu_special->upload();
-
- if(atom->nspecial) cu_nspecial->upload();
-
- if(atom->molecule) cu_molecule->upload();
-
- if(cu_eatom) cu_eatom->upload();
-
- if(cu_vatom) cu_vatom->upload();
-
- my_gettime(CLOCK_REALTIME, &endtime);
- uploadtime += (endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000);
- CUDA_IF_BINNING(Cuda_PreBinning(& shared_data);)
- CUDA_IF_BINNING(Cuda_Binning(& shared_data);)
-
- shared_data.atom.triggerneighsq = neighbor->triggersq;
- MYDBG(printf("# CUDA: Cuda::uploadAll() ... end\n");)
-}
-
-void Cuda::downloadAll()
-{
- MYDBG(printf("# CUDA: Cuda::downloadAll() ... start\n");)
- my_times starttime;
- my_times endtime;
-
- if(atom->nmax != shared_data.atom.nmax) checkResize();
-
- CUDA_IF_BINNING(Cuda_ReverseBinning(& shared_data);)
- my_gettime(CLOCK_REALTIME, &starttime);
- cu_x ->download();
- cu_v ->download();
- cu_f ->download();
- cu_type->download();
- cu_tag ->download();
- cu_mask->download();
- cu_image->download();
-
- //if(shared_data.atom.need_eatom) cu_eatom->download();
- //if(shared_data.atom.need_vatom) cu_vatom->download();
-
- if(shared_data.atom.q_flag) cu_q ->download();
-
- if(atom->rmass) cu_rmass->download();
-
- if(atom->radius) cu_radius->download();
-
- if(atom->omega) cu_omega->download();
-
- if(atom->torque) cu_torque->download();
-
- if(atom->special) cu_special->download();
-
- if(atom->nspecial) cu_nspecial->download();
-
- if(atom->molecule) cu_molecule->download();
-
- if(cu_eatom) cu_eatom->download();
-
- if(cu_vatom) cu_vatom->download();
-
- my_gettime(CLOCK_REALTIME, &endtime);
- downloadtime += (endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000);
- MYDBG(printf("# CUDA: Cuda::downloadAll() ... end\n");)
-}
-
-void Cuda::upload(int datamask)
-{
- MYDBG(printf("# CUDA: Cuda::upload() ... start\n");)
- my_times starttime;
- my_times endtime;
-
- if(atom->nmax != shared_data.atom.nmax) checkResize();
-
- my_gettime(CLOCK_REALTIME, &starttime);
- if(X_MASK & datamask) cu_x ->upload();
- if(V_MASK & datamask) cu_v ->upload();
- if(F_MASK & datamask) cu_f ->upload();
- if(TYPE_MASK & datamask) cu_type->upload();
- if(TAG_MASK & datamask) cu_tag ->upload();
- if(MASK_MASK & datamask) cu_mask->upload();
- if(IMAGE_MASK & datamask) cu_image->upload();
-
- //if(shared_data.atom.need_eatom) cu_eatom->upload();
- //if(shared_data.atom.need_vatom) cu_vatom->upload();
-
- if(shared_data.atom.q_flag)
- if(Q_MASK & datamask) cu_q ->upload();
-
- if(atom->rmass)
- if(RMASS_MASK & datamask) cu_rmass->upload();
-
- if(atom->radius)
- if(RADIUS_MASK & datamask) cu_radius->upload();
-
- if(atom->omega)
- if(OMEGA_MASK & datamask) cu_omega->upload();
-
- if(atom->torque)
- if(TORQUE_MASK & datamask) cu_torque->upload();
-
- if(atom->special)
- if(SPECIAL_MASK & datamask) cu_special->upload();
-
- if(atom->nspecial)
- if(SPECIAL_MASK & datamask) cu_nspecial->upload();
-
- if(atom->molecule)
- if(MOLECULE_MASK & datamask) cu_molecule->upload();
-
- if(cu_eatom) cu_eatom->upload();
-
- if(cu_vatom) cu_vatom->upload();
-
- my_gettime(CLOCK_REALTIME, &endtime);
- uploadtime += (endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000);
- MYDBG(printf("# CUDA: Cuda::upload() ... end\n");)
-}
-
-void Cuda::download(int datamask)
-{
- MYDBG(printf("# CUDA: Cuda::download() ... start\n");)
- my_times starttime;
- my_times endtime;
-
- if(atom->nmax != shared_data.atom.nmax) checkResize();
-
- CUDA_IF_BINNING(Cuda_ReverseBinning(& shared_data);)
- my_gettime(CLOCK_REALTIME, &starttime);
- if(X_MASK & datamask) cu_x ->download();
- if(V_MASK & datamask) cu_v ->download();
- if(F_MASK & datamask) cu_f ->download();
- if(TYPE_MASK & datamask) cu_type->download();
- if(TAG_MASK & datamask) cu_tag ->download();
- if(MASK_MASK & datamask) cu_mask->download();
- if(IMAGE_MASK & datamask) cu_image->download();
-
- //if(shared_data.atom.need_eatom) cu_eatom->download();
- //if(shared_data.atom.need_vatom) cu_vatom->download();
-
- if(shared_data.atom.q_flag)
- if(Q_MASK & datamask) cu_q ->download();
-
- if(atom->rmass)
- if(RMASS_MASK & datamask) cu_rmass->download();
-
- if(atom->radius)
- if(RADIUS_MASK & datamask) cu_radius->download();
-
- if(atom->omega)
- if(OMEGA_MASK & datamask) cu_omega->download();
-
- if(atom->torque)
- if(TORQUE_MASK & datamask) cu_torque->download();
-
- if(atom->special)
- if(SPECIAL_MASK & datamask) cu_special->download();
-
- if(atom->nspecial)
- if(SPECIAL_MASK & datamask) cu_nspecial->download();
-
- if(atom->molecule)
- if(MOLECULE_MASK & datamask) cu_molecule->download();
-
- if(cu_eatom) cu_eatom->download();
-
- if(cu_vatom) cu_vatom->download();
-
- my_gettime(CLOCK_REALTIME, &endtime);
- downloadtime += (endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000);
- MYDBG(printf("# CUDA: Cuda::download() ... end\n");)
-}
-
-void Cuda::downloadX()
-{
- Cuda_Pair_RevertXType(& this->shared_data);
- cu_x->download();
-}
-
-CudaNeighList* Cuda::registerNeighborList(class NeighList* neigh_list)
-{
- MYDBG(printf("# CUDA: Cuda::registerNeighborList() ... start a\n");)
- std::map<NeighList*, CudaNeighList*>::iterator p = neigh_lists.find(neigh_list);
-
- if(p != neigh_lists.end()) return p->second;
- else {
- CudaNeighList* neigh_list_cuda = new CudaNeighList(lmp, neigh_list);
- neigh_lists.insert(std::pair<NeighList*, CudaNeighList*>(neigh_list, neigh_list_cuda));
- return neigh_list_cuda;
- }
-
- MYDBG(printf("# CUDA: Cuda::registerNeighborList() ... end b\n");)
-}
-
-void Cuda::uploadAllNeighborLists()
-{
- MYDBG(printf("# CUDA: Cuda::uploadAllNeighborList() ... start\n");)
- std::map<NeighList*, CudaNeighList*>::iterator p = neigh_lists.begin();
-
- while(p != neigh_lists.end()) {
- p->second->nl_upload();
-
- if(not(p->second->neigh_list->cuda_list->build_cuda))
- for(int i = 0; i < atom->nlocal; i++)
- p->second->sneighlist.maxneighbors = MAX(p->second->neigh_list->numneigh[i], p->second->sneighlist.maxneighbors) ;
-
- ++p;
- }
-
- MYDBG(printf("# CUDA: Cuda::uploadAllNeighborList() ... done\n");)
-}
-
-void Cuda::downloadAllNeighborLists()
-{
- MYDBG(printf("# CUDA: Cuda::downloadAllNeighborList() ... start\n");)
- std::map<NeighList*, CudaNeighList*>::iterator p = neigh_lists.begin();
-
- while(p != neigh_lists.end()) {
- p->second->nl_download();
- ++p;
- }
-}
-
-void Cuda::update_xhold(int &maxhold, double* xhold)
-{
- if(this->shared_data.atom.maxhold < atom->nmax) {
- maxhold = atom->nmax;
- delete this->cu_xhold;
- this->cu_xhold = new cCudaData<double, X_CFLOAT, yx> ((double*)xhold, & this->shared_data.atom.xhold , maxhold, 3);
- }
-
- this->shared_data.atom.maxhold = maxhold;
- CudaWrapper_CopyData(this->cu_xhold->dev_data(), this->cu_x->dev_data(), 3 * atom->nmax * sizeof(X_CFLOAT));
-}
-
-void Cuda::setTimingsZero()
-{
- shared_data.cuda_timings.test1 = 0;
- shared_data.cuda_timings.test2 = 0;
-
- //communication
- shared_data.cuda_timings.comm_forward_total = 0;
- shared_data.cuda_timings.comm_forward_mpi_upper = 0;
- shared_data.cuda_timings.comm_forward_mpi_lower = 0;
- shared_data.cuda_timings.comm_forward_kernel_pack = 0;
- shared_data.cuda_timings.comm_forward_kernel_unpack = 0;
- shared_data.cuda_timings.comm_forward_upload = 0;
- shared_data.cuda_timings.comm_forward_download = 0;
-
- shared_data.cuda_timings.comm_exchange_total = 0;
- shared_data.cuda_timings.comm_exchange_mpi = 0;
- shared_data.cuda_timings.comm_exchange_kernel_pack = 0;
- shared_data.cuda_timings.comm_exchange_kernel_unpack = 0;
- shared_data.cuda_timings.comm_exchange_kernel_fill = 0;
- shared_data.cuda_timings.comm_exchange_cpu_pack = 0;
- shared_data.cuda_timings.comm_exchange_upload = 0;
- shared_data.cuda_timings.comm_exchange_download = 0;
-
- shared_data.cuda_timings.comm_border_total = 0;
- shared_data.cuda_timings.comm_border_mpi = 0;
- shared_data.cuda_timings.comm_border_kernel_pack = 0;
- shared_data.cuda_timings.comm_border_kernel_unpack = 0;
- shared_data.cuda_timings.comm_border_kernel_buildlist = 0;
- shared_data.cuda_timings.comm_border_kernel_self = 0;
- shared_data.cuda_timings.comm_border_upload = 0;
- shared_data.cuda_timings.comm_border_download = 0;
-
- //pair forces
- shared_data.cuda_timings.pair_xtype_conversion = 0;
- shared_data.cuda_timings.pair_kernel = 0;
- shared_data.cuda_timings.pair_virial = 0;
- shared_data.cuda_timings.pair_force_collection = 0;
-
- //neighbor
- shared_data.cuda_timings.neigh_bin = 0;
- shared_data.cuda_timings.neigh_build = 0;
- shared_data.cuda_timings.neigh_special = 0;
-
- //PPPM
- shared_data.cuda_timings.pppm_particle_map = 0;
- shared_data.cuda_timings.pppm_make_rho = 0;
- shared_data.cuda_timings.pppm_brick2fft = 0;
- shared_data.cuda_timings.pppm_poisson = 0;
- shared_data.cuda_timings.pppm_fillbrick = 0;
- shared_data.cuda_timings.pppm_fieldforce = 0;
- shared_data.cuda_timings.pppm_compute = 0;
-
- CudaWrapper_CheckUploadTime(true);
- CudaWrapper_CheckDownloadTime(true);
- CudaWrapper_CheckCPUBufUploadTime(true);
- CudaWrapper_CheckCPUBufDownloadTime(true);
-}
-
-void Cuda::print_timings()
-{
- if(universe->me != 0) return;
-
- if(not dotiming) return;
-
- printf("\n # CUDA: Special timings\n\n");
- printf("\n Transfer Times\n");
- printf(" PCIe Upload: \t %lf s\n", CudaWrapper_CheckUploadTime());
- printf(" PCIe Download:\t %lf s\n", CudaWrapper_CheckDownloadTime());
- printf(" CPU Tempbbuf Upload: \t %lf \n", CudaWrapper_CheckCPUBufUploadTime());
- printf(" CPU Tempbbuf Download: \t %lf \n", CudaWrapper_CheckCPUBufDownloadTime());
-
- printf("\n Communication \n");
-
- printf(" Forward Total \t %lf \n", shared_data.cuda_timings.comm_forward_total);
- printf(" Forward MPI Upper Bound \t %lf \n", shared_data.cuda_timings.comm_forward_mpi_upper);
- printf(" Forward MPI Lower Bound \t %lf \n", shared_data.cuda_timings.comm_forward_mpi_lower);
- printf(" Forward Kernel Pack \t %lf \n", shared_data.cuda_timings.comm_forward_kernel_pack);
- printf(" Forward Kernel Unpack \t %lf \n", shared_data.cuda_timings.comm_forward_kernel_unpack);
- printf(" Forward Kernel Self \t %lf \n", shared_data.cuda_timings.comm_forward_kernel_self);
- printf(" Forward Upload \t %lf \n", shared_data.cuda_timings.comm_forward_upload);
- printf(" Forward Download \t %lf \n", shared_data.cuda_timings.comm_forward_download);
- printf(" Forward Overlap Split Ratio\t %lf \n", shared_data.comm.overlap_split_ratio);
- printf("\n");
-
- printf(" Exchange Total \t %lf \n", shared_data.cuda_timings.comm_exchange_total);
- printf(" Exchange MPI \t %lf \n", shared_data.cuda_timings.comm_exchange_mpi);
- printf(" Exchange Kernel Pack \t %lf \n", shared_data.cuda_timings.comm_exchange_kernel_pack);
- printf(" Exchange Kernel Unpack \t %lf \n", shared_data.cuda_timings.comm_exchange_kernel_unpack);
- printf(" Exchange Kernel Fill \t %lf \n", shared_data.cuda_timings.comm_exchange_kernel_fill);
- printf(" Exchange CPU Pack \t %lf \n", shared_data.cuda_timings.comm_exchange_cpu_pack);
- printf(" Exchange Upload \t %lf \n", shared_data.cuda_timings.comm_exchange_upload);
- printf(" Exchange Download \t %lf \n", shared_data.cuda_timings.comm_exchange_download);
- printf("\n");
-
- printf(" Border Total \t %lf \n", shared_data.cuda_timings.comm_border_total);
- printf(" Border MPI \t %lf \n", shared_data.cuda_timings.comm_border_mpi);
- printf(" Border Kernel Pack \t %lf \n", shared_data.cuda_timings.comm_border_kernel_pack);
- printf(" Border Kernel Unpack \t %lf \n", shared_data.cuda_timings.comm_border_kernel_unpack);
- printf(" Border Kernel Self \t %lf \n", shared_data.cuda_timings.comm_border_kernel_self);
- printf(" Border Kernel BuildList \t %lf \n", shared_data.cuda_timings.comm_border_kernel_buildlist);
- printf(" Border Upload \t %lf \n", shared_data.cuda_timings.comm_border_upload);
- printf(" Border Download \t %lf \n", shared_data.cuda_timings.comm_border_download);
- printf("\n");
-
- //pair forces
- printf(" Pair XType Conversion \t %lf \n", shared_data.cuda_timings.pair_xtype_conversion);
- printf(" Pair Kernel \t %lf \n", shared_data.cuda_timings.pair_kernel);
- printf(" Pair Virial \t %lf \n", shared_data.cuda_timings.pair_virial);
- printf(" Pair Force Collection \t %lf \n", shared_data.cuda_timings.pair_force_collection);
- printf("\n");
-
- //neighbor
- printf(" Neighbor Binning \t %lf \n", shared_data.cuda_timings.neigh_bin);
- printf(" Neighbor Build \t %lf \n", shared_data.cuda_timings.neigh_build);
- printf(" Neighbor Special \t %lf \n", shared_data.cuda_timings.neigh_special);
- printf("\n");
-
- //pppm
- if(force->kspace) {
- printf(" PPPM Total \t %lf \n", shared_data.cuda_timings.pppm_compute);
- printf(" PPPM Particle Map \t %lf \n", shared_data.cuda_timings.pppm_particle_map);
- printf(" PPPM Make Rho \t %lf \n", shared_data.cuda_timings.pppm_make_rho);
- printf(" PPPM Brick2fft \t %lf \n", shared_data.cuda_timings.pppm_brick2fft);
- printf(" PPPM Poisson \t %lf \n", shared_data.cuda_timings.pppm_poisson);
- printf(" PPPM Fillbrick \t %lf \n", shared_data.cuda_timings.pppm_fillbrick);
- printf(" PPPM Fieldforce \t %lf \n", shared_data.cuda_timings.pppm_fieldforce);
- printf("\n");
- }
-
- printf(" Debug Test 1 \t %lf \n", shared_data.cuda_timings.test1);
- printf(" Debug Test 2 \t %lf \n", shared_data.cuda_timings.test2);
-
- printf("\n");
-}
diff --git a/src/USER-CUDA/cuda_data.h b/src/USER-CUDA/cuda_data.h
deleted file mode 100644
index bb778c12d3..0000000000
--- a/src/USER-CUDA/cuda_data.h
+++ /dev/null
@@ -1,796 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifndef _CUDA_DATA_H_
-#define _CUDA_DATA_H_
-
-
-enum copy_mode {x, xx, xy, yx, xyz, xzy}; // yxz, yzx, zxy, zyx not yet implemented since they were not needed yet
-//xx==x in atom_vec x is a member therefore copymode x produces compile errors
-#include "cuda_shared.h"
-#include "cuda_wrapper_cu.h"
-#include "cuda_data_cu.h"
-#include <ctime>
-
-#include <cstdio>
-#include <typeinfo>
-template <typename host_type, typename dev_type, copy_mode mode>
-class cCudaData
-{
- protected:
- void** buffer;
- int* buf_size;
- host_type* host_data;
- dev_array* dev_data_array;
- dev_type* temp_data;
- unsigned nbytes;
- bool owns_dev_array;
- bool current_data_on_device; //this is not yet working as intended and therefore deactivated
- bool current_data_on_host;
- bool is_continues;
- bool pinned;
-
- public:
- cCudaData(host_type* host_data, dev_array* dev_data_array, unsigned dim_x, unsigned dim_y=0, unsigned dim_z=0, bool is_pinned=false);
- cCudaData(host_type* host_data, unsigned dim_x, unsigned dim_y=0, unsigned dim_z=0, bool is_pinned=false);
- ~cCudaData();
- void* dev_data() {if(dev_data_array!=NULL) return dev_data_array->dev_data; else return NULL;};
- void set_dev_data(void* adev_data) {dev_data_array->dev_data=adev_data;};
- void set_dev_array(dev_array* adev_array) {dev_data_array=adev_array;};
- void set_host_data(host_type* host_data);
- void* get_host_data() { return host_data;};
- void set_buffer(void** buffer,int* buf_size,bool ais_continues);
- unsigned int* get_dim() {return dev_data_array->dim;};
- // if you want to upload data to the gpu, which will not change there, then set will_be_changed=false
- // if you want to upload data to the gpu and update it there, then set will_be_changed=true (default)
- void upload(bool will_be_changed=true);
- void uploadAsync(int stream, bool will_be_changed=true );
- // if you want to download data just to have a look at it, then set will_be_changed=false
- // if you are going to modify the downloaded data, then set will_be_changed=true (default)
- void download(bool will_be_changed=true);
- void downloadAsync(int stream);
- void memset_device(int value);
- void device_data_has_changed() {current_data_on_device=false;}
- void host_data_has_changed() {current_data_on_host=false;}
- int dev_size() {
- int size = dev_data_array->dim[0]*sizeof(dev_type);
- if(dev_data_array->dim[1]) size*=dev_data_array->dim[1];
- if(dev_data_array->dim[2]) size*=dev_data_array->dim[2];
- return size;}
-};
-
-
-template <typename host_type, typename dev_type, copy_mode mode>
-cCudaData<host_type, dev_type, mode>
-::cCudaData(host_type* host_data, dev_array* dev_data_array, unsigned dim_x, unsigned dim_y, unsigned dim_z, bool is_pinned)
-{
- pinned=is_pinned;
- owns_dev_array = false;
- current_data_on_device = false;
- current_data_on_host = false;
- is_continues = false;
- this->host_data = host_data;
- this->dev_data_array = dev_data_array;
- unsigned ndev;
- if((mode == x)||(mode==xx))
- {
- ndev = dim_x;
- dev_data_array->dim[0] = dim_x;
- dev_data_array->dim[1] = 0;
- dev_data_array->dim[2] = 0;
- }
- else if(mode == xy || mode == yx )
- {
- ndev = dim_x * dim_y;
- dev_data_array->dim[0] = dim_x;
- dev_data_array->dim[1] = dim_y;
- dev_data_array->dim[2] = 0;
- }
- else
- {
- ndev = dim_x * dim_y * dim_z;
- dev_data_array->dim[0] = dim_x;
- dev_data_array->dim[1] = dim_y;
- dev_data_array->dim[2] = dim_z;
- }
- nbytes = ndev * sizeof(dev_type);
- if(nbytes<=0)
- {
- host_data=NULL;
- temp_data=NULL;
- dev_data_array->dev_data=NULL;
- return;
- }
-
- dev_data_array->dev_data = CudaWrapper_AllocCudaData(nbytes);
- if(((mode!=x)&&(mode!=xx)) || typeid(host_type) != typeid(dev_type))
- {
- if(not pinned)
- temp_data = new dev_type[ndev];
- else
- {
- temp_data = (dev_type*) CudaWrapper_AllocPinnedHostData(ndev*sizeof(dev_type));
- }
- }
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-cCudaData<host_type, dev_type, mode>
-::cCudaData(host_type* host_data, unsigned dim_x, unsigned dim_y, unsigned dim_z, bool is_pinned)
-{
- pinned=is_pinned;
- this->dev_data_array = new dev_array;
- this->owns_dev_array = true;
- current_data_on_device = false;
- current_data_on_host = false;
- is_continues = false;
- this->host_data = host_data;
- unsigned ndev;
- if((mode == x)||(mode==xx))
- {
- ndev = dim_x;
- dev_data_array->dim[0] = dim_x;
- dev_data_array->dim[1] = 0;
- dev_data_array->dim[2] = 0;
- }
- else if(mode == xy || mode == yx )
- {
- ndev = dim_x * dim_y;
- dev_data_array->dim[0] = dim_x;
- dev_data_array->dim[1] = dim_y;
- dev_data_array->dim[2] = 0;
- }
- else
- {
- ndev = dim_x * dim_y * dim_z;
- dev_data_array->dim[0] = dim_x;
- dev_data_array->dim[1] = dim_y;
- dev_data_array->dim[2] = dim_z;
- }
- nbytes = ndev * sizeof(dev_type);
- if(nbytes<=0)
- {
- host_data=NULL;
- temp_data=NULL;
- dev_data_array->dev_data=NULL;
- return;
- }
-
- dev_data_array->dev_data = CudaWrapper_AllocCudaData(nbytes);
- if(((mode!=x)&&(mode!=xx)) || (typeid(host_type) != typeid(dev_type)))
- {
- if(not pinned)
- temp_data = new dev_type[ndev];
- else
- {
- temp_data = (dev_type*) CudaWrapper_AllocPinnedHostData(ndev*sizeof(dev_type));
- }
- }
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-cCudaData<host_type, dev_type, mode>
-::~cCudaData()
-{
- if(((mode!=x)&&(mode!=xx)) || typeid(host_type) != typeid(dev_type))
- {
- if(not pinned)
- delete [] temp_data;
- else
- {
- CudaWrapper_FreePinnedHostData((void*)temp_data);
- }
- }
- if((dev_data_array->dev_data)&&(nbytes>0))
- CudaWrapper_FreeCudaData(dev_data_array->dev_data,nbytes);
- if(owns_dev_array) delete dev_data_array;
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-void cCudaData<host_type, dev_type, mode>
-::set_host_data(host_type* host_data)
-{
- this->host_data = host_data;
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-void cCudaData<host_type, dev_type, mode>
-::upload(bool will_be_changed)
-{
- // if current data is already up, do not re-upload it
-// if(current_data_on_device) return;
- if(buffer&&is_continues)
- {
- printf("Actual Buffer: %p %i\n",*buffer,*buf_size);
- if(typeid(host_type)==typeid(double))
- {
- if(typeid(dev_type)==typeid(double))
- {
- CudaData_Upload_DoubleDouble((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- else if(typeid(dev_type)==typeid(float))
- {
- CudaData_Upload_DoubleFloat((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- }
- else if(typeid(host_type)==typeid(float))
- {
- if(typeid(dev_type)==typeid(double))
- {
- CudaData_Upload_FloatDouble((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- else if(typeid(dev_type)==typeid(float))
- {
- CudaData_Upload_FloatFloat((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- }
- else if(typeid(host_type)==typeid(int))
- {
- if(typeid(dev_type)==typeid(int))
- {
- CudaData_Upload_IntInt((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- }
- }
- switch(mode)
- {
- case x:
- {
- if(typeid(host_type) == typeid(dev_type))
- CudaWrapper_UploadCudaData(host_data, dev_data_array->dev_data, nbytes);
- else
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) temp_data[i] = static_cast<dev_type>(host_data[i]);
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- }
- break;
- }
-
- case xx:
- {
- if(typeid(host_type) == typeid(dev_type))
- CudaWrapper_UploadCudaData(host_data, dev_data_array->dev_data, nbytes);
- else
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) temp_data[i] = static_cast<dev_type>(host_data[i]);
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- }
- break;
- }
-
- case xy:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- dev_type* temp = &temp_data[i * dev_data_array->dim[1]];
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- temp[j] = static_cast<dev_type>((reinterpret_cast<host_type**>(host_data))[i][j]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- break;
- }
-
- case yx:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- dev_type* temp = &temp_data[j*dev_data_array->dim[0]];
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- temp[i] = static_cast<dev_type>(reinterpret_cast<host_type**>(host_data)[i][j]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- break;
- }
- case xyz:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- dev_type* temp = &temp_data[(i*dev_data_array->dim[1]+j)*dev_data_array->dim[2]];
- for(unsigned k=0; k<dev_data_array->dim[2]; ++k)
- {
- temp[k] = static_cast<dev_type>(reinterpret_cast<host_type***>(host_data)[i][j][k]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- break;
- }
-
- case xzy:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- for(unsigned k=0; k<dev_data_array->dim[2]; ++k)
- {
- dev_type* temp = &temp_data[(i*dev_data_array->dim[2]+k)*dev_data_array->dim[1]];
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- temp[j] = static_cast<dev_type>(reinterpret_cast<host_type***>(host_data)[i][j][k]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- break;
- }
- }
- // we have uploaded the data to the device, i.e.:
- current_data_on_device = true;
- // the data is going to change on the device, making the host data out-dated
- if(will_be_changed) current_data_on_host = false;
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-void cCudaData<host_type, dev_type, mode>
-::uploadAsync(int stream,bool will_be_changed)
-{
- // if current data is already up, do not re-upload it
-// if(current_data_on_device) return;
- if(buffer&&is_continues)
- {
- printf("Actual Buffer: %p %i\n",*buffer,*buf_size);
- if(typeid(host_type)==typeid(double))
- {
- if(typeid(dev_type)==typeid(double))
- {
- CudaData_Upload_DoubleDouble((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- else if(typeid(dev_type)==typeid(float))
- {
- CudaData_Upload_DoubleFloat((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- }
- else if(typeid(host_type)==typeid(float))
- {
- if(typeid(dev_type)==typeid(double))
- {
- CudaData_Upload_FloatDouble((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- else if(typeid(dev_type)==typeid(float))
- {
- CudaData_Upload_FloatFloat((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- }
- else if(typeid(host_type)==typeid(int))
- {
- if(typeid(dev_type)==typeid(int))
- {
- CudaData_Upload_IntInt((void*) host_data,dev_data_array->dev_data,
- dev_data_array->dim,mode,*buffer);
- current_data_on_device = true;
- if(will_be_changed) current_data_on_host = false;
- return;
- }
- }
- }
- switch(mode)
- {
- case x:
- {
- if(typeid(host_type) == typeid(dev_type))
- CudaWrapper_UploadCudaDataAsync(host_data, dev_data_array->dev_data, nbytes,stream);
- else
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) temp_data[i] = static_cast<dev_type>(host_data[i]);
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes,stream);
- }
- break;
- }
-
- case xx:
- {
- if(typeid(host_type) == typeid(dev_type))
- CudaWrapper_UploadCudaDataAsync(host_data, dev_data_array->dev_data, nbytes,stream);
- else
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) temp_data[i] = static_cast<dev_type>(host_data[i]);
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes,stream);
- }
- break;
- }
-
- case xy:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- dev_type* temp = &temp_data[i * dev_data_array->dim[1]];
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- temp[j] = static_cast<dev_type>((reinterpret_cast<host_type**>(host_data))[i][j]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes,stream);
- break;
- }
-
- case yx:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- dev_type* temp = &temp_data[j*dev_data_array->dim[0]];
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- temp[i] = static_cast<dev_type>(reinterpret_cast<host_type**>(host_data)[i][j]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes,stream);
- break;
- }
- case xyz:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- dev_type* temp = &temp_data[(i*dev_data_array->dim[1]+j)*dev_data_array->dim[2]];
- for(unsigned k=0; k<dev_data_array->dim[2]; ++k)
- {
- temp[k] = static_cast<dev_type>(reinterpret_cast<host_type***>(host_data)[i][j][k]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes,stream);
- break;
- }
-
- case xzy:
- {
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- for(unsigned k=0; k<dev_data_array->dim[2]; ++k)
- {
- dev_type* temp = &temp_data[(i*dev_data_array->dim[2]+k)*dev_data_array->dim[1]];
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- temp[j] = static_cast<dev_type>(reinterpret_cast<host_type***>(host_data)[i][j][k]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufUploadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- CudaWrapper_UploadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes,stream);
- break;
- }
- }
- // we have uploaded the data to the device, i.e.:
- current_data_on_device = true;
- // the data is going to change on the device, making the host data out-dated
- if(will_be_changed) current_data_on_host = false;
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-void cCudaData<host_type, dev_type, mode>
-::download(bool will_be_changed)
-{
- // if current data is already down, do not re-download it
-// if(current_data_on_host) return;
- switch(mode)
- {
- case x:
- {
- if(typeid(host_type) == typeid(dev_type))
- CudaWrapper_DownloadCudaData(host_data, dev_data_array->dev_data, nbytes);
- else
- {
- CudaWrapper_DownloadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) host_data[i] = static_cast<host_type>(temp_data[i]);
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufDownloadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- }
- break;
- }
-
- case xx:
- {
- if(typeid(host_type) == typeid(dev_type))
- CudaWrapper_DownloadCudaData(host_data, dev_data_array->dev_data, nbytes);
- else
- {
- CudaWrapper_DownloadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) host_data[i] = static_cast<host_type>(temp_data[i]);
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufDownloadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- }
- break;
- }
-
- case xy:
- {
- CudaWrapper_DownloadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- dev_type* temp = &temp_data[i * dev_data_array->dim[1]];
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- reinterpret_cast<host_type**>(host_data)[i][j] = static_cast<host_type>(temp[j]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufDownloadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- break;
- }
-
- case yx:
- {
- CudaWrapper_DownloadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- dev_type* temp = &temp_data[j*dev_data_array->dim[0]];
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- reinterpret_cast<host_type**>(host_data)[i][j] = static_cast<host_type>(temp[i]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufDownloadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- break;
- }
-
- case xyz:
- {
- CudaWrapper_DownloadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- dev_type* temp = &temp_data[(i * dev_data_array->dim[1]+j)*dev_data_array->dim[2]];
- for(unsigned k=0; k<dev_data_array->dim[2]; ++k)
- {
- reinterpret_cast<host_type***>(host_data)[i][j][k] = static_cast<host_type>(temp[k]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufDownloadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- break;
- }
-
- case xzy:
- {
- CudaWrapper_DownloadCudaData(temp_data, dev_data_array->dev_data, nbytes);
- timespec time1,time2;
- my_gettime(CLOCK_REALTIME,&time1);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- for(unsigned k=0; k<dev_data_array->dim[2]; ++k)
- {
- dev_type* temp = &temp_data[(i * dev_data_array->dim[2]+k)*dev_data_array->dim[1]];
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- reinterpret_cast<host_type***>(host_data)[i][j][k] = static_cast<host_type>(temp[j]);
- }
- }
- my_gettime(CLOCK_REALTIME,&time2);
- CudaWrapper_AddCPUBufDownloadTime(
- time2.tv_sec-time1.tv_sec+1.0*(time2.tv_nsec-time1.tv_nsec)/1000000000);
- break;
- }
- }
- // we have downloaded the data to the host, i.e.:
- current_data_on_host = true;
- // the data is going to change on the host, making the device data out-dated
- if(will_be_changed) current_data_on_device = false;
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-void cCudaData<host_type, dev_type, mode>
-::downloadAsync(int stream)
-{
- switch(mode)
- {
- case x:
- {
- if(typeid(host_type) == typeid(dev_type))
- {
- CudaWrapper_DownloadCudaDataAsync(host_data, dev_data_array->dev_data, nbytes, stream);
- CudaWrapper_SyncStream(stream);
- }
- else
- {
- CudaWrapper_DownloadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes, stream);
- CudaWrapper_SyncStream(stream);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) host_data[i] = static_cast<host_type>(temp_data[i]);
- }
- break;
- }
-
- case xx:
- {
- if(typeid(host_type) == typeid(dev_type))
- {
- CudaWrapper_DownloadCudaDataAsync(host_data, dev_data_array->dev_data, nbytes, stream);
- CudaWrapper_SyncStream(stream);
- }
- else
- {
- CudaWrapper_DownloadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes, stream);
- CudaWrapper_SyncStream(stream);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i) host_data[i] = static_cast<host_type>(temp_data[i]);
- }
- break;
- }
-
- case xy:
- {
- CudaWrapper_DownloadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes, stream);
- CudaWrapper_SyncStream(stream);
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- dev_type* temp = &temp_data[i * dev_data_array->dim[1]];
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- reinterpret_cast<host_type**>(host_data)[i][j] = static_cast<host_type>(temp[j]);
- }
- }
- break;
- }
-
- case yx:
- {
- CudaWrapper_DownloadCudaDataAsync(temp_data, dev_data_array->dev_data, nbytes, stream);
- CudaWrapper_SyncStream(stream);
- for(unsigned j=0; j<dev_data_array->dim[1]; ++j)
- {
- dev_type* temp = &temp_data[j*dev_data_array->dim[0]];
- for(unsigned i=0; i<dev_data_array->dim[0]; ++i)
- {
- reinterpret_cast<host_type**>(host_data)[i][j] = static_cast<host_type>(temp[i]);
- }
- }
- break;
- }
- }
-}
-
-
-template <typename host_type, typename dev_type, copy_mode mode>
-void cCudaData<host_type, dev_type, mode>
-::memset_device(int value)
-{
- CudaWrapper_Memset(dev_data_array->dev_data,value, nbytes);
-}
-
-template <typename host_type, typename dev_type, copy_mode mode>
-void cCudaData<host_type, dev_type, mode>
-::set_buffer(void** abuffer,int* abuf_size,bool ais_continues)
-{
- buffer = abuffer;
- buf_size = abuf_size;
- unsigned nbytes_buf=(nbytes/sizeof(dev_type))*sizeof(host_type);
- if(buffer!=NULL)
- if(not((typeid(host_type) == typeid(dev_type))&&(mode == x || mode == xx)))
- {
- printf("Allocate Buffer: %p %i\n",*buffer,*buf_size);
- if(((*buffer)!=NULL)&&(*buf_size<nbytes_buf))
- CudaWrapper_FreeCudaData(*buffer,*buf_size);
- if(*buf_size<nbytes_buf)
- {*buffer=CudaWrapper_AllocCudaData(nbytes_buf);*buf_size=nbytes_buf;}
- printf("Allocate Buffer2: %p %i\n",*buffer,*buf_size);
-
- }
- is_continues=ais_continues;
-}
-#endif // _CUDA_DATA_H_
diff --git a/src/USER-CUDA/cuda_modify_flags.h b/src/USER-CUDA/cuda_modify_flags.h
deleted file mode 100644
index e683456d27..0000000000
--- a/src/USER-CUDA/cuda_modify_flags.h
+++ /dev/null
@@ -1,45 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifndef CUDA_MODIFY_FLAGS_H
-#define CUDA_MODIFY_FLAGS_H
-
-#include "fix.h"
-
-namespace LAMMPS_NS {
-namespace FixConstCuda {
- static const int INITIAL_INTEGRATE_CUDA = FixConst::FIX_CONST_LAST << 0;
- static const int POST_INTEGRATE_CUDA = FixConst::FIX_CONST_LAST << 1;
- static const int PRE_EXCHANGE_CUDA = FixConst::FIX_CONST_LAST << 2;
- static const int PRE_NEIGHBOR_CUDA = FixConst::FIX_CONST_LAST << 3;
- static const int PRE_FORCE_CUDA = FixConst::FIX_CONST_LAST << 4;
- static const int POST_FORCE_CUDA = FixConst::FIX_CONST_LAST << 5;
- static const int FINAL_INTEGRATE_CUDA = FixConst::FIX_CONST_LAST << 6;
- static const int END_OF_STEP_CUDA = FixConst::FIX_CONST_LAST << 7;
- static const int THERMO_ENERGY_CUDA = FixConst::FIX_CONST_LAST << 8;
- static const int MIN_POST_FORCE_CUDA = FixConst::FIX_CONST_LAST << 9;
-}
-}
-// remember not to shift over 31 bits
-
-#endif // CUDA_MODIFY_FLAGS_H
diff --git a/src/USER-CUDA/cuda_neigh_list.cpp b/src/USER-CUDA/cuda_neigh_list.cpp
deleted file mode 100644
index 6e05cee274..0000000000
--- a/src/USER-CUDA/cuda_neigh_list.cpp
+++ /dev/null
@@ -1,184 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include "cuda_neigh_list.h"
-#include "neigh_list.h"
-#include <cstring>
-#include <vector>
-#include <map>
-#include <algorithm>
-#include "user_cuda.h"
-#include "atom.h"
-#include "error.h"
-
-using namespace LAMMPS_NS;
-
-CudaNeighList::CudaNeighList(LAMMPS *lmp, class NeighList* neigh_list) : Pointers(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- MYDBG(printf("# CUDA: CudaNeighList::cudaNeighList() ... start\n");)
- this->neigh_list = neigh_list;
- neigh_list->cuda_list=this;
- sneighlist.maxlocal = neigh_list->get_maxlocal();
- sneighlist.maxneighbors = 32;
- sneighlist.maxcut = 0.0;
- sneighlist.cutneighsq = NULL;
- cu_neighbors = NULL;
- cu_neighbors_border = NULL;
- cu_neighbors_inner = NULL;
- cu_numneigh_border = NULL;
- cu_numneigh_inner = NULL;
- cu_numneigh = NULL;
- cu_ilist = NULL;
- cu_ilist_border = NULL;
- cu_inum_border = NULL;
- inum_border = 0;
- neighbors = NULL;
- neighbors_inner = NULL;
- neighbors_border = NULL;
- numneigh_border = NULL;
- numneigh_inner = NULL;
- ilist_border = NULL;
-
- build_cuda = false;
- sneighlist.binned_id=NULL;
- sneighlist.bin_dim=new int[3];
- sneighlist.bin_dim[0]=0;
- sneighlist.bin_dim[1]=0;
- sneighlist.bin_dim[2]=0;
-
- cu_ex_type = NULL;
- cu_ex1_bit = NULL;
- cu_ex2_bit = NULL;
- cu_ex_mol_bit = NULL;
- sneighlist.nex_type=0;
- sneighlist.nex_group=0;
- sneighlist.nex_mol=0;
-
- sneighlist.bin_nmax=0;
- sneighlist.bin_extraspace=0.05;
- MYDBG(printf("# CUDA: CudaNeighList::cudaNeighList() ... end\n");)
-
-}
-
-CudaNeighList::~CudaNeighList()
-{
- dev_free();
-}
-
-void CudaNeighList::dev_alloc()
-{
- MYDBG( printf("# CUDA: CudaNeighList::dev_alloc() ... start\n"); )
- cu_ilist = new cCudaData<int , int , x> (neigh_list->ilist , & sneighlist.ilist , sneighlist.maxlocal );
- cu_numneigh = new cCudaData<int , int , x> (neigh_list->numneigh, & sneighlist.numneigh , sneighlist.maxlocal );
- neighbors = new int[atom->nmax*sneighlist.maxneighbors];
- cu_neighbors= new cCudaData<int, int, x> (neighbors , & sneighlist.neighbors, atom->nmax*sneighlist.maxneighbors );
-
- if(cuda->shared_data.overlap_comm)
- {
- ilist_border = new int[sneighlist.maxlocal];
- numneigh_border = new int[sneighlist.maxlocal];
- numneigh_inner = new int[sneighlist.maxlocal];
- cu_inum_border = new cCudaData<int , int , x> (&inum_border , & sneighlist.inum_border , 1 );
- cu_ilist_border = new cCudaData<int , int , x> (ilist_border , & sneighlist.ilist_border , sneighlist.maxlocal );
- cu_numneigh_border = new cCudaData<int , int , x> (numneigh_border , & sneighlist.numneigh_border , sneighlist.maxlocal );
- cu_numneigh_inner = new cCudaData<int , int , x> (numneigh_inner , & sneighlist.numneigh_inner , sneighlist.maxlocal );
- neighbors_border = new int[sneighlist.maxlocal*sneighlist.maxneighbors];
- cu_neighbors_border= new cCudaData<int, int, x> (neighbors_border , & sneighlist.neighbors_border, sneighlist.maxlocal*sneighlist.maxneighbors );
- neighbors_inner = new int[sneighlist.maxlocal*sneighlist.maxneighbors];
- cu_neighbors_inner = new cCudaData<int, int, x> (neighbors_inner , & sneighlist.neighbors_inner , sneighlist.maxlocal*sneighlist.maxneighbors );
- }
- cuda->shared_data.atom.update_neigh=2;
- MYDBG( printf("# CUDA: CudaNeighList::dev_alloc() ... end\n"); )
-}
-
-void CudaNeighList::dev_free()
-{
- MYDBG( printf("# CUDA: CudaNeighList::dev_free() ... start\n"); )
- delete cu_numneigh;
- delete cu_ilist;
- delete [] neighbors;
- delete cu_neighbors;
-
- if(cuda->shared_data.overlap_comm)
- {
- delete [] ilist_border;
- delete [] numneigh_border;
- delete [] numneigh_inner;
- delete [] neighbors_border;
- delete [] neighbors_inner;
- delete cu_inum_border;
- delete cu_neighbors_border;
- delete cu_neighbors_inner;
- delete cu_numneigh_border;
- delete cu_numneigh_inner;
- delete cu_ilist_border;
- }
- MYDBG( printf("# CUDA: CudaNeighList::dev_free() ... end\n"); )
-}
-
-void CudaNeighList::grow_device()
-{
- MYDBG(printf("# CUDA: CudaNeighList::grow_device() ... start\n");)
- // if host has allocated more memory for atom arrays than device has, then allocate more memory on device
- int new_maxlocal = neigh_list->get_maxlocal();
- if(sneighlist.maxlocal < new_maxlocal)
- {
- sneighlist.maxlocal = new_maxlocal;
- dev_free();
- dev_alloc();
- }
-
- if(!cu_ilist || !cu_numneigh) dev_alloc();
-
- // check, if hosts data has been allocated somewhere else
- if(cu_ilist ->get_host_data() != neigh_list->ilist) cu_ilist ->set_host_data(neigh_list->ilist);
- if(cu_numneigh->get_host_data() != neigh_list->numneigh) cu_numneigh->set_host_data(neigh_list->numneigh);
-
- MYDBG(printf("# CUDA: CudaNeighList::grow_device() ... end\n");)
-}
-
-
-void CudaNeighList::nl_upload(bool will_be_changed)
-{
- //return;
- MYDBG(printf("# CUDA: CudaNeighList::nl_upload() ... start\n");)
- if(cu_ilist)
- cu_ilist->upload();
- if(cu_numneigh)
- cu_numneigh->upload();
- MYDBG(printf("# CUDA: CudaNeighList::nl_upload() ... end\n");)
-}
-
-void CudaNeighList::nl_download(bool will_be_changed)
-{
- MYDBG(printf("# CUDA: CudaNeighList::nl_download() ... start\n");)
- if(cu_ilist)
- cu_ilist->download();
- if(cu_numneigh)
- cu_numneigh->download();
- MYDBG(printf("# CUDA: CudaNeighList::nl_download() ... end\n");)
-}
diff --git a/src/USER-CUDA/cuda_neigh_list.h b/src/USER-CUDA/cuda_neigh_list.h
deleted file mode 100644
index f733cdfd61..0000000000
--- a/src/USER-CUDA/cuda_neigh_list.h
+++ /dev/null
@@ -1,83 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifndef LMP_NEIGH_LIST_CUDA_H
-#define LMP_NEIGH_LIST_CUDA_H
-
-#include "pointers.h"
-#include "cuda_data.h"
-#include "neigh_list.h"
-
-namespace LAMMPS_NS
-{
-
-class CudaNeighList : protected Pointers
-{
- public:
- cCudaData<int , int , x>* cu_ilist;
- cCudaData<int , int , x>* cu_numneigh;
- cCudaData<int , int , x>* cu_inum_border;
- cCudaData<int , int , x>* cu_ilist_border;
- cCudaData<int , int , x>* cu_numneigh_border;
- cCudaData<int , int , x>* cu_numneigh_inner;
- cCudaData<int , int , x>* cu_neighbors;
- cCudaData<int , int , x>* cu_neighbors_border;
- cCudaData<int , int , x>* cu_neighbors_inner;
- cCudaData<int , int , x>* cu_ex_type;
- cCudaData<int , int , x>* cu_ex1_bit;
- cCudaData<int , int , x>* cu_ex2_bit;
- cCudaData<int , int , x>* cu_ex_mol_bit;
-
-
- cuda_shared_neighlist sneighlist;
-
- int* neighbors;
- int* neighbors_inner;
- int* neighbors_border;
- int inum_border;
- int* ilist_border;
- int* numneigh_border;
- int* numneigh_inner;
- int nex_type;
- int nex_group;
- int nex_mol;
-
- bool build_cuda;
-
- CudaNeighList(class LAMMPS *, class NeighList* neigh_list);
- ~CudaNeighList();
- void grow_device(); // will grow pages memory on device, keeping old pages. will grow lists memory on device, deleting old lists
- void nl_upload(bool will_be_changed=true);
- void nl_download(bool will_be_changed=true);
- NeighList* neigh_list;
-
- void dev_alloc();
- void dev_free();
-
- private:
- class Cuda *cuda;
-};
-
-}
-
-#endif
diff --git a/src/USER-CUDA/domain_cuda.cpp b/src/USER-CUDA/domain_cuda.cpp
deleted file mode 100644
index 997a42a681..0000000000
--- a/src/USER-CUDA/domain_cuda.cpp
+++ /dev/null
@@ -1,345 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author (triclinic) : Pieter in 't Veld (SNL)
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <stdlib.h>
-#include <string.h>
-#include <stdio.h>
-#include <math.h>
-#include "domain_cuda.h"
-#include "style_region.h"
-#include "atom.h"
-#include "force.h"
-#include "update.h"
-#include "modify.h"
-#include "fix.h"
-#include "fix_deform.h"
-#include "region.h"
-#include "lattice.h"
-#include "comm.h"
-#include "memory.h"
-#include "error.h"
-
-#include "user_cuda.h"
-#include "domain_cu.h"
-
-using namespace LAMMPS_NS;
-
-#define BIG 1.0e20
-#define SMALL 1.0e-4
-#define DELTA 1
-
-enum {NO_REMAP, X_REMAP, V_REMAP}; // same as fix_deform.cpp
-
-/* ----------------------------------------------------------------------
- default is periodic
-------------------------------------------------------------------------- */
-
-DomainCuda::DomainCuda(LAMMPS* lmp) : Domain(lmp)
-{
- cuda = lmp->cuda;
-
- if(cuda == NULL)
- error->all(FLERR, "You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-}
-
-/* ---------------------------------------------------------------------- */
-
-void DomainCuda::init()
-{
- Domain::init();
-
- if(not cuda->finished_run) {
- cuda->setDomainParams();
- Cuda_Domain_Init(&cuda->shared_data);
- }
-}
-
-/* ----------------------------------------------------------------------
- set global box params
- assumes boxlo/hi and triclinic tilts are already set
-------------------------------------------------------------------------- */
-
-void DomainCuda::set_global_box()
-{
- // one-time activation of CUDA
- // do it here, b/c is now too late for further package commands
- // activation must occur before any USER-CUDA class communicates with GPUs
-
- cuda->activate();
-
- Domain::set_global_box();
-
- if(not cuda->finished_run) {
- cuda->setDomainParams();
- }
-}
-
-/* ----------------------------------------------------------------------
- set lamda box params, only need be done one time
- assumes global box is defined and proc assignment has been made by comm
- for uppermost proc, insure subhi = 1.0 (in case round-off occurs)
-------------------------------------------------------------------------- */
-
-void DomainCuda::set_lamda_box()
-{
- Domain::set_lamda_box();
-
- if(not cuda->finished_run) {
- cuda->setDomainParams();
- }
-}
-
-/* ----------------------------------------------------------------------
- set local subbox params
- assumes global box is defined and proc assignment has been made
- for uppermost proc, insure subhi = boxhi (in case round-off occurs)
-------------------------------------------------------------------------- */
-
-void DomainCuda::set_local_box()
-{
- Domain::set_local_box();
-
- if(not cuda->finished_run) {
- // cuda->setDomainParams();
- //Cuda_Domain_Init(&cuda->shared_data);
- }
-}
-
-/* ----------------------------------------------------------------------
- reset global & local boxes due to global box boundary changes
- if shrink-wrapped, determine atom extent and reset boxlo/hi
- if shrink-wrapped and triclinic, perform shrink-wrap in box coords
-------------------------------------------------------------------------- */
-
-void DomainCuda::reset_box()
-{
- if(nonperiodic == 2) {
-
- // convert back to box coords for shrink-wrap operation
-
- if(triclinic) lamda2x(atom->nlocal);
-
- // compute extent of atoms on this proc
-
- double extent[3][2], all[3][2];
-
- extent[2][0] = extent[1][0] = extent[0][0] = BIG;
- extent[2][1] = extent[1][1] = extent[0][1] = -BIG;
-
- double** x = atom->x;
- int nlocal = atom->nlocal;
-
- if(cuda->finished_setup && (!cuda->oncpu)) {
- extent[0][0] = cuda->extent[0];
- extent[0][1] = cuda->extent[1];
- extent[1][0] = cuda->extent[2];
- extent[1][1] = cuda->extent[3];
- extent[2][0] = cuda->extent[4];
- extent[2][1] = cuda->extent[5];
- } else
- for(int i = 0; i < nlocal; i++) {
- extent[0][0] = MIN(extent[0][0], x[i][0]);
- extent[0][1] = MAX(extent[0][1], x[i][0]);
- extent[1][0] = MIN(extent[1][0], x[i][1]);
- extent[1][1] = MAX(extent[1][1], x[i][1]);
- extent[2][0] = MIN(extent[2][0], x[i][2]);
- extent[2][1] = MAX(extent[2][1], x[i][2]);
- }
-
- // compute extent across all procs
- // flip sign of MIN to do it in one Allreduce MAX
-
- extent[0][0] = -extent[0][0];
- extent[1][0] = -extent[1][0];
- extent[2][0] = -extent[2][0];
-
- MPI_Allreduce(extent, all, 6, MPI_DOUBLE, MPI_MAX, world);
-
- // in shrink-wrapped dims, set box by atom extent
- // if minimum set, enforce min box size settings
-
- if(triclinic == 0) {
- if(xperiodic == 0) {
- if(boundary[0][0] == 2) boxlo[0] = -all[0][0] - small[0];
- else if(boundary[0][0] == 3)
- boxlo[0] = MIN(-all[0][0] - small[0], minxlo);
-
- if(boundary[0][1] == 2) boxhi[0] = all[0][1] + small[0];
- else if(boundary[0][1] == 3) boxhi[0] = MAX(all[0][1] + small[0], minxhi);
-
- if(boxlo[0] > boxhi[0]) error->all(FLERR, "Illegal simulation box");
- }
-
- if(yperiodic == 0) {
- if(boundary[1][0] == 2) boxlo[1] = -all[1][0] - small[1];
- else if(boundary[1][0] == 3)
- boxlo[1] = MIN(-all[1][0] - small[1], minylo);
-
- if(boundary[1][1] == 2) boxhi[1] = all[1][1] + small[1];
- else if(boundary[1][1] == 3) boxhi[1] = MAX(all[1][1] + small[1], minyhi);
-
- if(boxlo[1] > boxhi[1]) error->all(FLERR, "Illegal simulation box");
- }
-
- if(zperiodic == 0) {
- if(boundary[2][0] == 2) boxlo[2] = -all[2][0] - small[2];
- else if(boundary[2][0] == 3)
- boxlo[2] = MIN(-all[2][0] - small[2], minzlo);
-
- if(boundary[2][1] == 2) boxhi[2] = all[2][1] + small[2];
- else if(boundary[2][1] == 3) boxhi[2] = MAX(all[2][1] + small[2], minzhi);
-
- if(boxlo[2] > boxhi[2]) error->all(FLERR, "Illegal simulation box");
- }
-
- } else {
- double lo[3], hi[3];
-
- if(xperiodic == 0) {
- lo[0] = -all[0][0];
- lo[1] = 0.0;
- lo[2] = 0.0;
- Domain::lamda2x(lo, lo);
- hi[0] = all[0][1];
- hi[1] = 0.0;
- hi[2] = 0.0;
- Domain::lamda2x(hi, hi);
-
- if(boundary[0][0] == 2) boxlo[0] = lo[0] - small[0];
- else if(boundary[0][0] == 3) boxlo[0] = MIN(lo[0] - small[0], minxlo);
-
- if(boundary[0][1] == 2) boxhi[0] = hi[0] + small[0];
- else if(boundary[0][1] == 3) boxhi[0] = MAX(hi[0] + small[0], minxhi);
-
- if(boxlo[0] > boxhi[0]) error->all(FLERR, "Illegal simulation box");
- }
-
- if(yperiodic == 0) {
- lo[0] = 0.0;
- lo[1] = -all[1][0];
- lo[2] = 0.0;
- Domain::lamda2x(lo, lo);
- hi[0] = 0.0;
- hi[1] = all[1][1];
- hi[2] = 0.0;
- Domain::lamda2x(hi, hi);
-
- if(boundary[1][0] == 2) boxlo[1] = lo[1] - small[1];
- else if(boundary[1][0] == 3) boxlo[1] = MIN(lo[1] - small[1], minylo);
-
- if(boundary[1][1] == 2) boxhi[1] = hi[1] + small[1];
- else if(boundary[1][1] == 3) boxhi[1] = MAX(hi[1] + small[1], minyhi);
-
- if(boxlo[1] > boxhi[1]) error->all(FLERR, "Illegal simulation box");
-
- //xy *= (boxhi[1]-boxlo[1]) / yprd;
- }
-
- if(zperiodic == 0) {
- lo[0] = 0.0;
- lo[1] = 0.0;
- lo[2] = -all[2][0];
- Domain::lamda2x(lo, lo);
- hi[0] = 0.0;
- hi[1] = 0.0;
- hi[2] = all[2][1];
- Domain::lamda2x(hi, hi);
-
- if(boundary[2][0] == 2) boxlo[2] = lo[2] - small[2];
- else if(boundary[2][0] == 3) boxlo[2] = MIN(lo[2] - small[2], minzlo);
-
- if(boundary[2][1] == 2) boxhi[2] = hi[2] + small[2];
- else if(boundary[2][1] == 3) boxhi[2] = MAX(hi[2] + small[2], minzhi);
-
- if(boxlo[2] > boxhi[2]) error->all(FLERR, "Illegal simulation box");
-
- //xz *= (boxhi[2]-boxlo[2]) / xprd;
- //yz *= (boxhi[2]-boxlo[2]) / yprd;
- }
- }
- }
-
- set_global_box();
- set_local_box();
-
- if(not cuda->finished_run) {
- cuda->setDomainParams();
- Cuda_Domain_Init(&cuda->shared_data);
- }
-
- // if shrink-wrapped, convert to lamda coords for new box
- // must re-invoke pbc() b/c x2lamda result can be outside 0,1 due to roundoff
-
- if(nonperiodic == 2 && triclinic) {
- x2lamda(atom->nlocal);
- pbc();
- }
-}
-
-/* ----------------------------------------------------------------------
- enforce PBC and modify box image flags for each atom
- called every reneighboring and by other commands that change atoms
- resulting coord must satisfy lo <= coord < hi
- MAX is important since coord - prd < lo can happen when coord = hi
- if fix deform, remap velocity of fix group atoms by box edge velocities
- for triclinic, atoms must be in lamda coords (0-1) before pbc is called
- image = 10 bits for each dimension
- increment/decrement in wrap-around fashion
-------------------------------------------------------------------------- */
-
-void DomainCuda::pbc()
-{
- if(cuda->finished_setup && (!cuda->oncpu)) {
- cuda->setDomainParams();
- Cuda_Domain_PBC(&cuda->shared_data, deform_vremap, deform_groupbit, cuda->extent);
- return;
- }
-
- Domain::pbc();
-}
-
-
-/* ----------------------------------------------------------------------
- convert triclinic 0-1 lamda coords to box coords for all N atoms
- x = H lamda + x0;
-------------------------------------------------------------------------- */
-
-void DomainCuda::lamda2x(int n)
-{
- if(cuda->finished_setup && (!cuda->oncpu)) {
- Cuda_Domain_lamda2x(&cuda->shared_data, n);
- return;
- }
-
- Domain::lamda2x(n);
-}
-
-/* ----------------------------------------------------------------------
- convert box coords to triclinic 0-1 lamda coords for all N atoms
- lamda = H^-1 (x - x0)
-------------------------------------------------------------------------- */
-
-void DomainCuda::x2lamda(int n)
-{
- if(cuda->finished_setup && (!cuda->oncpu)) {
- Cuda_Domain_x2lamda(&cuda->shared_data, n);
- return;
- }
-
- Domain::x2lamda(n);
-}
diff --git a/src/USER-CUDA/domain_cuda.h b/src/USER-CUDA/domain_cuda.h
deleted file mode 100644
index ede402ffba..0000000000
--- a/src/USER-CUDA/domain_cuda.h
+++ /dev/null
@@ -1,41 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifndef LMP_DOMAIN_CUDA_H
-#define LMP_DOMAIN_CUDA_H
-
-#include "pointers.h"
-#include "domain.h"
-
-namespace LAMMPS_NS {
-
-class DomainCuda : public Domain {
- public:
- DomainCuda(class LAMMPS *);
- void init();
- void set_global_box();
- void set_lamda_box();
- void set_local_box();
- void reset_box();
- void pbc();
-
- virtual void lamda2x(int);
- virtual void x2lamda(int);
-
- protected:
- class Cuda *cuda;
-};
-
-}
-
-#endif
diff --git a/src/USER-CUDA/fft3d_cuda.cpp b/src/USER-CUDA/fft3d_cuda.cpp
deleted file mode 100644
index 42ec4b16ee..0000000000
--- a/src/USER-CUDA/fft3d_cuda.cpp
+++ /dev/null
@@ -1,609 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing authors: Jim Shepherd (GA Tech) added SGI SCSL support
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <cstdio>
-#include <cstdlib>
-#include <cmath>
-#include "fft3d_cuda.h"
-#include "fft3d_cuda_cu.h"
-#include "remap.h"
-#include <ctime>
-#include "cuda_wrapper_cu.h"
-
-#ifdef FFT_CUFFT
-#endif
-
-#define MIN(A,B) ((A) < (B)) ? (A) : (B)
-#define MAX(A,B) ((A) > (B)) ? (A) : (B)
-
-/* ----------------------------------------------------------------------
- Data layout for 3d FFTs:
-
- data set of Nfast x Nmid x Nslow elements is owned by P procs
- on input, each proc owns a subsection of the elements
- on output, each proc will own a (possibly different) subsection
- my subsection must not overlap with any other proc's subsection,
- i.e. the union of all proc's input (or output) subsections must
- exactly tile the global Nfast x Nmid x Nslow data set
- when called from C, all subsection indices are
- C-style from 0 to N-1 where N = Nfast or Nmid or Nslow
- when called from F77, all subsection indices are
- F77-style from 1 to N where N = Nfast or Nmid or Nslow
- a proc can own 0 elements on input or output
- by specifying hi index < lo index
- on both input and output, data is stored contiguously on a processor
- with a fast-varying, mid-varying, and slow-varying index
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Perform 3d FFT
-
- Arguments:
- in starting address of input data on this proc
- out starting address of where output data for this proc
- will be placed (can be same as in)
- flag 1 for forward FFT, -1 for inverse FFT
- plan plan returned by previous call to fft_3d_create_plan
-------------------------------------------------------------------------- */
-
-void fft_3d_cuda(FFT_DATA *in, FFT_DATA *out, int flag, struct fft_plan_3d *plan)
-{
-#ifdef FFT_CUFFT
- plan->iterate++;
- my_times starttime,starttime2;
- my_times endtime,endtime2;
-
- int i,total,length,offset,num;
- double norm;
- FFT_DATA *data,*copy;
- // system specific constants
-
-
- // pre-remap to prepare for 1st FFTs if needed
- // copy = loc for remap result
- int nprocs=plan->nprocs;
-if(nprocs>1)
-{
- if(plan->init)
- my_gettime(CLOCK_REALTIME,&starttime);
- if (plan->pre_plan) {
- if (plan->pre_target == 0) copy = out;
- else copy = plan->copy;
- if(plan->init) remap_3d((double *) in, (double *) out, (double *) plan->scratch,plan->pre_plan);
- data = out;
- }
- else
- data = in;
-}
- cufftResult retvalc;
- if(plan->init)
- {
- if(nprocs>1)
- {
- if(sizeof(FFT_CFLOAT)==sizeof(double))cudaMemcpy((void*) (plan->cudata2), (void*) data, plan->cudatasize/2,cudaMemcpyHostToDevice);
- if(sizeof(FFT_CFLOAT)==sizeof(float)) cudaMemcpy((void*) (plan->cudata2), (void*) data, plan->cudatasize,cudaMemcpyHostToDevice);
- initfftdata((double*)plan->cudata2,(FFT_CFLOAT*)plan->cudata,plan->nfast,plan->nmid,plan->nslow);
- }
- }
- if (flag == -1)
- {
- retvalc=cufft(plan->plan_3d, plan->cudata, plan->cudata2,CUFFT_FORWARD);
- }
- else
- {
- retvalc=cufft(plan->plan_3d, plan->cudata, plan->cudata2,CUFFT_INVERSE);
- }
- if(retvalc!=CUFFT_SUCCESS) {printf("ErrorCUFFT: %i\n",retvalc);exit(EXIT_FAILURE);}
-
- FFTsyncthreads();
-#endif
-}
-/* ----------------------------------------------------------------------
- Create plan for performing a 3d FFT
-
- Arguments:
- comm MPI communicator for the P procs which own the data
- nfast,nmid,nslow size of global 3d matrix
- in_ilo,in_ihi input bounds of data I own in fast index
- in_jlo,in_jhi input bounds of data I own in mid index
- in_klo,in_khi input bounds of data I own in slow index
- out_ilo,out_ihi output bounds of data I own in fast index
- out_jlo,out_jhi output bounds of data I own in mid index
- out_klo,out_khi output bounds of data I own in slow index
- scaled 0 = no scaling of result, 1 = scaling
- permute permutation in storage order of indices on output
- 0 = no permutation
- 1 = permute once = mid->fast, slow->mid, fast->slow
- 2 = permute twice = slow->fast, fast->mid, mid->slow
- nbuf returns size of internal storage buffers used by FFT
-------------------------------------------------------------------------- */
-
-struct fft_plan_3d *fft_3d_create_plan_cuda(
- MPI_Comm comm, int nfast, int nmid, int nslow,
- int in_ilo, int in_ihi, int in_jlo, int in_jhi,
- int in_klo, int in_khi,
- int out_ilo, int out_ihi, int out_jlo, int out_jhi,
- int out_klo, int out_khi,
- int scaled, int permute, int *nbuf,bool ainit)
-{
-#ifdef FFT_CUFFT
- struct fft_plan_3d *plan;
- int me,nprocs;
- int i,num,flag,remapflag,fftflag;
- int first_ilo,first_ihi,first_jlo,first_jhi,first_klo,first_khi;
- int second_ilo,second_ihi,second_jlo,second_jhi,second_klo,second_khi;
- int third_ilo,third_ihi,third_jlo,third_jhi,third_klo,third_khi;
- int out_size,first_size,second_size,third_size,copy_size,scratch_size;
- int np1,np2,ip1,ip2;
- int list[50];
-
- // system specific variables
-
- // query MPI info
-
- MPI_Comm_rank(comm,&me);
- MPI_Comm_size(comm,&nprocs);
-
-#ifndef FFT_CUFFT
- error->all(FLERR,"ERROR: Trying to use cuda fft without FFT_CUFFT set. Recompile with make option 'cufft=1'.");
-#endif
- // compute division of procs in 2 dimensions not on-processor
- bifactor_cuda(nprocs,&np1,&np2);
- ip1 = me % np1;
- ip2 = me/np1;
-
- // in case of CUDA FFT every proc does the full FFT in order to avoid data transfers (the problem is other wise heavily bandwidth limited)
-
- int ip1out = ip1;
- int ip2out = ip2;
- int np1out = np1;
- int np2out = np2;
-
- ip1 = 0;
- ip2 = 0;
- np1 = 1;
- np2 = 1;
-
- // allocate memory for plan data struct
-
- plan = (struct fft_plan_3d *) malloc(sizeof(struct fft_plan_3d));
- if (plan == NULL) return NULL;
- plan->init=ainit;
-
- // remap from initial distribution to layout needed for 1st set of 1d FFTs
- // not needed if all procs own entire fast axis initially
- // first indices = distribution after 1st set of FFTs
-
- if (in_ilo == 0 && in_ihi == nfast-1)
- flag = 0;
- else
- flag = 1;
-
- if(nprocs>1)flag=1;
-
- MPI_Allreduce(&flag,&remapflag,1,MPI_INT,MPI_MAX,comm);
-
- if (remapflag == 0) {
- first_ilo = in_ilo;
- first_ihi = in_ihi;
- first_jlo = in_jlo;
- first_jhi = in_jhi;
- first_klo = in_klo;
- first_khi = in_khi;
- plan->pre_plan = NULL;
- }
- else {
- first_ilo = 0;
- first_ihi = nfast - 1;
- first_jlo = ip1*nmid/np1;
- first_jhi = (ip1+1)*nmid/np1 - 1;
- first_klo = ip2*nslow/np2;
- first_khi = (ip2+1)*nslow/np2 - 1;
- int members=2;
- if(plan->init) members=1;
- plan->pre_plan =
- remap_3d_create_plan(comm,in_ilo,in_ihi,in_jlo,in_jhi,in_klo,in_khi,
- first_ilo,first_ihi,first_jlo,first_jhi,
- first_klo,first_khi,
- members,0,0,2,0);
- if (plan->pre_plan == NULL) return NULL;
- }
-
- // 1d FFTs along fast axis
-
- plan->length1 = nfast;
- plan->total1 = nfast * nmid * nslow;
-
- // remap from 1st to 2nd FFT
- // choose which axis is split over np1 vs np2 to minimize communication
- // second indices = distribution after 2nd set of FFTs
-
- second_ilo = ip1*nfast/np1;
- second_ihi = (ip1+1)*nfast/np1 - 1;
- second_jlo = 0;
- second_jhi = nmid - 1;
- second_klo = ip2*nslow/np2;
- second_khi = (ip2+1)*nslow/np2 - 1;
- plan->mid1_plan =
- remap_3d_create_plan(comm,
- first_ilo,first_ihi,first_jlo,first_jhi,
- first_klo,first_khi,
- second_ilo,second_ihi,second_jlo,second_jhi,
- second_klo,second_khi,
- 2,1,0,2,0);
- if (plan->mid1_plan == NULL) return NULL;
-
- // 1d FFTs along mid axis
-
- plan->length2 = nmid;
- plan->total2 = nfast * nmid * nslow;
-
- // remap from 2nd to 3rd FFT
- // if final distribution is permute=2 with all procs owning entire slow axis
- // then this remapping goes directly to final distribution
- // third indices = distribution after 3rd set of FFTs
-
- flag=1;
-
- MPI_Allreduce(&flag,&remapflag,1,MPI_INT,MPI_MAX,comm);
-
- if (remapflag == 0) {
- third_ilo = out_ilo;
- third_ihi = out_ihi;
- third_jlo = out_jlo;
- third_jhi = out_jhi;
- third_klo = out_klo;
- third_khi = out_khi;
- }
- else {
- third_ilo = ip1*nfast/np1;
- third_ihi = (ip1+1)*nfast/np1 - 1;
- third_jlo = ip2*nmid/np2;
- third_jhi = (ip2+1)*nmid/np2 - 1;
- third_klo = 0;
- third_khi = nslow - 1;
- }
-
- plan->mid2_plan =
- remap_3d_create_plan(comm,
- second_jlo,second_jhi,second_klo,second_khi,
- second_ilo,second_ihi,
- third_jlo,third_jhi,third_klo,third_khi,
- third_ilo,third_ihi,
- 2,1,0,2,0);
- if (plan->mid2_plan == NULL) return NULL;
-
- // 1d FFTs along slow axis
-
- plan->length3 = nslow;
- plan->total3 = nfast * nmid * nslow;
-
- // remap from 3rd FFT to final distribution
- // not needed if permute = 2 and third indices = out indices on all procs
-
- flag=1;
-
- MPI_Allreduce(&flag,&remapflag,1,MPI_INT,MPI_MAX,comm);
-
- if (remapflag == 0)
- plan->post_plan = NULL;
- else {
- plan->post_plan =
- remap_3d_create_plan(comm,
- third_klo,third_khi,third_ilo,third_ihi,
- third_jlo,third_jhi,
- out_klo,out_khi,out_ilo,out_ihi,
- out_jlo,out_jhi,
- 2,(permute+1)%3,0,2,0);
- if (plan->post_plan == NULL) return NULL;
- }
-
- // configure plan memory pointers and allocate work space
- // out_size = amount of memory given to FFT by user
- // first/second/third_size = amount of memory needed after pre,mid1,mid2 remaps
- // copy_size = amount needed internally for extra copy of data
- // scratch_size = amount needed internally for remap scratch space
- // for each remap:
- // out space used for result if big enough, else require copy buffer
- // accumulate largest required remap scratch space
-
- out_size = (out_ihi-out_ilo+1) * (out_jhi-out_jlo+1) * (out_khi-out_klo+1);
- first_size = (first_ihi-first_ilo+1) * (first_jhi-first_jlo+1) *
- (first_khi-first_klo+1);
- second_size = (second_ihi-second_ilo+1) * (second_jhi-second_jlo+1) *
- (second_khi-second_klo+1);
- third_size = (third_ihi-third_ilo+1) * (third_jhi-third_jlo+1) *
- (third_khi-third_klo+1);
-
- plan->ihi_out=out_ihi;
- plan->ilo_out=out_ilo;
- plan->jhi_out=out_jhi;
- plan->jlo_out=out_jlo;
- plan->khi_out=out_khi;
- plan->klo_out=out_klo;
-
- copy_size = 0;
- scratch_size = 0;
-
- if (plan->pre_plan) {
- if (first_size <= out_size)
- plan->pre_target = 0;
- else {
- plan->pre_target = 1;
- copy_size = MAX(copy_size,first_size);
- }
- scratch_size = MAX(scratch_size,first_size);
- }
-
- if (plan->mid1_plan) {
- if (second_size <= out_size)
- plan->mid1_target = 0;
- else {
- plan->mid1_target = 1;
- copy_size = MAX(copy_size,second_size);
- }
- scratch_size = MAX(scratch_size,second_size);
- }
-
- if (plan->mid2_plan) {
- if (third_size <= out_size)
- plan->mid2_target = 0;
- else {
- plan->mid2_target = 1;
- copy_size = MAX(copy_size,third_size);
- }
- scratch_size = MAX(scratch_size,third_size);
- }
-
- if (plan->post_plan)
- scratch_size = MAX(scratch_size,out_size);
-
- *nbuf = copy_size + scratch_size;
-
- if (copy_size) {
- plan->copy = (FFT_DATA *) malloc(copy_size*sizeof(FFT_DATA));
- if (plan->copy == NULL) return NULL;
- }
- else plan->copy = NULL;
-
- if (scratch_size) {
- plan->scratch = (FFT_DATA *) malloc(scratch_size*sizeof(FFT_DATA));
- if (plan->scratch == NULL) return NULL;
- }
- else plan->scratch = NULL;
-
- // system specific pre-computation of 1d FFT coeffs
- // and scaling normalization
-
- cufftResult retvalc;
- int nfft = (in_ihi-in_ilo+1) * (in_jhi-in_jlo+1) *
- (in_khi-in_klo+1);
- int nfft_brick = (out_ihi-out_ilo+1) * (out_jhi-out_jlo+1) *
- (out_khi-out_klo+1);
-
- int nfft_both = MAX(nfft,nfft_brick);
- nfft_both=nfast*nmid*nslow;
-
- plan->cudatasize=nfft_both*sizeof(FFT_DATA);
-
- //retvalc=cufftPlan1d(&(plan->plan_fast), nfast, CUFFT_PLAN,plan->total1/nfast);
- //if(retvalc!=CUFFT_SUCCESS) printf("ErrorCUFFT1: %i\n",retvalc);
- plan->nfast=nfast;
-
- //retvalc=cufftPlan1d(&(plan->plan_mid), nmid, CUFFT_PLAN,plan->total2/nmid);
- //if(retvalc!=CUFFT_SUCCESS) printf("ErrorCUFFT2: %i\n",retvalc);
- plan->nmid=nmid;
-
- //retvalc=cufftPlan1d(&(plan->plan_slow), nslow, CUFFT_PLAN,plan->total3/nslow);
- //if(retvalc!=CUFFT_SUCCESS) printf("ErrorCUFFT3: %i\n",retvalc);
- plan->nslow=nslow;
-
- retvalc=cufftPlan3d(&(plan->plan_3d), nslow,nmid,nfast, CUFFT_PLAN);
- if(retvalc!=CUFFT_SUCCESS) printf("ErrorCUFFT3: %i\n",retvalc);
-
- plan->nprocs=nprocs;
- plan->me=me;
- if (scaled == 0)
- plan->scaled = 0;
- else {
- plan->scaled = 1;
- plan->norm = 1.0/(nfast*nmid*nslow);
- plan->normnum = (out_ihi-out_ilo+1) * (out_jhi-out_jlo+1) *
- (out_khi-out_klo+1);
- }
-
- plan->coretime=0;
- plan->iterate=0;
- plan->ffttime=0;
- return plan;
- #endif
-}
-
-/* ----------------------------------------------------------------------
- Destroy a 3d fft plan
-------------------------------------------------------------------------- */
-
-void fft_3d_destroy_plan_cuda(struct fft_plan_3d *plan)
-{
-#ifdef FFT_CUFFT
- if (plan->pre_plan) remap_3d_destroy_plan(plan->pre_plan);
- if (plan->mid1_plan) remap_3d_destroy_plan(plan->mid1_plan);
- if (plan->mid2_plan) remap_3d_destroy_plan(plan->mid2_plan);
- if (plan->post_plan) remap_3d_destroy_plan(plan->post_plan);
-
- if (plan->copy) free(plan->copy);
- if (plan->scratch) free(plan->scratch);
-
-
- //cufftDestroy(plan->plan_fast);
- //cufftDestroy(plan->plan_mid);
- //cufftDestroy(plan->plan_slow);
- cufftDestroy(plan->plan_3d);
- free(plan);
-#endif
-}
-
-/* ----------------------------------------------------------------------
- recursively divide n into small factors, return them in list
-------------------------------------------------------------------------- */
-
-void factor_cuda(int n, int *num, int *list)
-{
- if (n == 1) {
- return;
- }
- else if (n % 2 == 0) {
- *list = 2;
- (*num)++;
- factor_cuda(n/2,num,list+1);
- }
- else if (n % 3 == 0) {
- *list = 3;
- (*num)++;
- factor_cuda(n/3,num,list+1);
- }
- else if (n % 5 == 0) {
- *list = 5;
- (*num)++;
- factor_cuda(n/5,num,list+1);
- }
- else if (n % 7 == 0) {
- *list = 7;
- (*num)++;
- factor_cuda(n/7,num,list+1);
- }
- else if (n % 11 == 0) {
- *list = 11;
- (*num)++;
- factor_cuda(n/11,num,list+1);
- }
- else if (n % 13 == 0) {
- *list = 13;
- (*num)++;
- factor_cuda(n/13,num,list+1);
- }
- else {
- *list = n;
- (*num)++;
- return;
- }
-}
-
-/* ----------------------------------------------------------------------
- divide n into 2 factors of as equal size as possible
-------------------------------------------------------------------------- */
-
-void bifactor_cuda(int n, int *factor1, int *factor2)
-{
- int n1,n2,facmax;
-
- facmax = static_cast<int> (sqrt((double) n));
-
- for (n1 = facmax; n1 > 0; n1--) {
- n2 = n/n1;
- if (n1*n2 == n) {
- *factor1 = n1;
- *factor2 = n2;
- return;
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- perform just the 1d FFTs needed by a 3d FFT, no data movement
- used for timing purposes
-
- Arguments:
- in starting address of input data on this proc, all set to 0.0
- nsize size of in
- flag 1 for forward FFT, -1 for inverse FFT
- plan plan returned by previous call to fft_3d_create_plan
-------------------------------------------------------------------------- */
-
-void fft_1d_only_cuda(FFT_DATA *data, int nsize, int flag, struct fft_plan_3d *plan)
-{
-#ifdef FFT_CUFFT
- int i,total,length,offset,num;
- double norm;
-
- // system specific constants
-
-
-
- // total = size of data needed in each dim
- // length = length of 1d FFT in each dim
- // total/length = # of 1d FFTs in each dim
- // if total > nsize, limit # of 1d FFTs to available size of data
-
- int total1 = plan->total1;
- int length1 = plan->length1;
- int total2 = plan->total2;
- int length2 = plan->length2;
- int total3 = plan->total3;
- int length3 = plan->length3;
-
- if (total1 > nsize) total1 = (nsize/length1) * length1;
- if (total2 > nsize) total2 = (nsize/length2) * length2;
- if (total3 > nsize) total3 = (nsize/length3) * length3;
-
- // perform 1d FFTs in each of 3 dimensions
- // data is just an array of 0.0
-
-
- cudaMemcpy((void**) &(plan->cudata), (void*) data, plan->cudatasize,cudaMemcpyHostToDevice);
- if (flag == -1) {
- cufft(plan->plan_3d, plan->cudata, plan->cudata,CUFFT_FORWARD);
- /*cufft(plan->plan_fast, plan->cudata, plan->cudata,CUFFT_FORWARD);
- cufft(plan->plan_mid, plan->cudata, plan->cudata,CUFFT_FORWARD);
- cufft(plan->plan_slow, plan->cudata, plan->cudata,CUFFT_FORWARD);*/
- } else {
- cufft(plan->plan_3d, plan->cudata, plan->cudata,CUFFT_FORWARD);
- /*cufft(plan->plan_fast, plan->cudata, plan->cudata,CUFFT_INVERSE);
- cufft(plan->plan_mid,plan->cudata, plan->cudata,CUFFT_INVERSE);
- cufft(plan->plan_slow, plan->cudata, plan->cudata,CUFFT_INVERSE);*/
- }
- cudaMemcpy((void*) data, (void**) &(plan->cudata), plan->cudatasize,cudaMemcpyDeviceToHost);
-
- // scaling if required
- // limit num to size of data
-
-#endif
-}
diff --git a/src/USER-CUDA/fft3d_cuda.h b/src/USER-CUDA/fft3d_cuda.h
deleted file mode 100644
index 059ac977f7..0000000000
--- a/src/USER-CUDA/fft3d_cuda.h
+++ /dev/null
@@ -1,148 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-// User-settable FFT precision
-
-// FFT_PRECISION = 1 is single-precision complex (4-byte real, 4-byte imag)
-// FFT_PRECISION = 2 is double-precision complex (8-byte real, 8-byte imag)
-#include "cuda_precision.h"
-//#define FFT_PRECISION 2
-
-// -------------------------------------------------------------------------
-
-// Data types for single-precision complex
-
-#if FFT_PRECISION_CU == 1
-
-#ifdef FFT_CUFFT
-#include "cuda_runtime.h"
-#include "cufft.h"
-typedef struct {
- float re;
- float im;
-} FFT_DATA;
-typedef cufftComplex cufftData;
-typedef cufftReal cufftDataInit;
-#define cufft cufftExecC2C
-#define cufftinit cufftExecR2C
-#define CUFFT_PLAN CUFFT_C2C
-#define CUFFT_PLAN_INIT CUFFT_R2C
-#else
-typedef struct {
- float re;
- float im;
-} FFT_DATA;
-#endif
-
-#endif
-
-// -------------------------------------------------------------------------
-
-// Data types for double-precision complex
-
-#if FFT_PRECISION_CU == 2
-
-
-#ifdef FFT_CUFFT
-#include "cuda_runtime.h"
-#include "cufft.h"
-typedef cufftDoubleComplex cufftData;
-typedef cufftDoubleReal cufftDataInit;
-typedef struct {
- double re;
- double im;
-} FFT_DATA;
-#define cufft cufftExecZ2Z
-#define cufftinit cufftExecD2Z
-#define CUFFT_PLAN CUFFT_Z2Z
-#define CUFFT_PLAN_INIT CUFFT_D2Z
-#endif
-
-#endif
-
-// -------------------------------------------------------------------------
-
-// details of how to do a 3d FFT
-
-struct fft_plan_3d {
- struct remap_plan_3d *pre_plan; // remap from input -> 1st FFTs
- struct remap_plan_3d *mid1_plan; // remap from 1st -> 2nd FFTs
- struct remap_plan_3d *mid2_plan; // remap from 2nd -> 3rd FFTs
- struct remap_plan_3d *post_plan; // remap from 3rd FFTs -> output
- FFT_DATA *copy; // memory for remap results (if needed)
- FFT_DATA *scratch; // scratch space for remaps
- int total1,total2,total3; // # of 1st,2nd,3rd FFTs (times length)
- int length1,length2,length3; // length of 1st,2nd,3rd FFTs
- int pre_target; // where to put remap results
- int mid1_target,mid2_target;
- int scaled; // whether to scale FFT results
- int normnum; // # of values to rescale
- double norm; // normalization factor for rescaling
-
- double coretime;
- double ffttime;
- int iterate;
- // system specific 1d FFT info
-
-#ifdef FFT_CUFFT
- //CUdeviceptr cudata;
- cufftData* cudata;
- cufftData* cudata2;
- unsigned int cudatasize;
- cufftHandle plan_fast;
- cufftHandle plan_mid;
- cufftHandle plan_slow;
- cufftHandle plan_3d;
- int nfast;
- int nmid;
- int nslow;
- int ihi_out,ilo_out,jhi_out,jlo_out,khi_out,klo_out;
- int me,nprocs;
-#endif
- int init;
-};
-
-// function prototypes
-
-void fft_3d_destroy_plan_cuda(struct fft_plan_3d *);
-void factor_cuda(int, int *, int *);
-void bifactor_cuda(int, int *, int *);
-void fft_1d_only_cuda(FFT_DATA *, int, int, struct fft_plan_3d *);
-void fft_3d_cudaA(FFT_DATA *, FFT_DATA *, int, struct fft_plan_3d *);
-void fft_3d_cuda(FFT_DATA *, FFT_DATA *, int, struct fft_plan_3d *);
-struct fft_plan_3d *fft_3d_create_plan_cuda(MPI_Comm, int, int, int,
- int, int, int, int, int, int, int, int, int, int, int, int,
- int, int, int *,bool init);
diff --git a/src/USER-CUDA/fft3d_wrap_cuda.cpp b/src/USER-CUDA/fft3d_wrap_cuda.cpp
deleted file mode 100644
index f02c38d831..0000000000
--- a/src/USER-CUDA/fft3d_wrap_cuda.cpp
+++ /dev/null
@@ -1,111 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include "fft3d_wrap_cuda.h"
-#include "error.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-FFT3dCuda::FFT3dCuda(LAMMPS *lmp, MPI_Comm comm, int nfast, int nmid, int nslow,
- int in_ilo, int in_ihi, int in_jlo, int in_jhi,
- int in_klo, int in_khi,
- int out_ilo, int out_ihi, int out_jlo, int out_jhi,
- int out_klo, int out_khi,
- int scaled, int permute, int *nbuf,bool init) : Pointers(lmp)
-{
-#ifdef FFT_CUFFT
- plan = fft_3d_create_plan_cuda(comm,nfast,nmid,nslow,
- in_ilo,in_ihi,in_jlo,in_jhi,in_klo,in_khi,
- out_ilo,out_ihi,out_jlo,out_jhi,out_klo,out_khi,
- scaled,permute,nbuf,init);
-#endif
-#ifndef FFT_CUFFT
- plan = fft_3d_create_plan(comm,nfast,nmid,nslow,
- in_ilo,in_ihi,in_jlo,in_jhi,in_klo,in_khi,
- out_ilo,out_ihi,out_jlo,out_jhi,out_klo,out_khi,
- scaled,permute,nbuf,0);
-#endif
- if (plan == NULL) error->one(FLERR,"Could not create 3d FFT plan");
-}
-
-/* ---------------------------------------------------------------------- */
-
-FFT3dCuda::~FFT3dCuda()
-{
-#ifdef FFT_CUFFT
- fft_3d_destroy_plan_cuda(plan);
-#endif
-#ifndef FFT_CUFFT
- fft_3d_destroy_plan(plan);
-#endif
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FFT3dCuda::compute(double *in, double *out, int flag)
-{
-#ifdef FFT_CUFFT
- fft_3d_cuda((FFT_DATA *) in,(FFT_DATA *) out,flag,plan);
-#endif
-#ifndef FFT_CUFFT
- fft_3d((FFT_DATA *) in,(FFT_DATA *) out,flag,plan);
-#endif
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FFT3dCuda::timing1d(double *in, int nsize, int flag)
-{
-#ifdef FFT_CUFFT
- fft_1d_only_cuda((FFT_DATA *) in,nsize,flag,plan);
-#endif
-#ifndef FFT_CUFFT
- fft_1d_only((FFT_DATA *) in,nsize,flag,plan);
-#endif
-}
-
-#ifdef FFT_CUFFT
-void FFT3dCuda::set_cudata(void* cudata,void* cudata2)
-{
-
- plan->cudata=(cufftData*) cudata;
- plan->cudata2=(cufftData*) cudata2;
-
-}
-#endif
diff --git a/src/USER-CUDA/fft3d_wrap_cuda.h b/src/USER-CUDA/fft3d_wrap_cuda.h
deleted file mode 100644
index cc6baa9ebe..0000000000
--- a/src/USER-CUDA/fft3d_wrap_cuda.h
+++ /dev/null
@@ -1,68 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifndef FFT3D_WRAP_CUDA_H_
-#define FFT3D_WRAP_CUDA_H_
-
-#include "pointers.h"
-
-#ifdef FFT_CUFFT
- #include "fft3d_cuda.h"
-#endif
-#ifndef FFT_CUFFT
- #include "fft3d.h"
-#endif
-
-namespace LAMMPS_NS {
-
-class FFT3dCuda : protected Pointers {
- public:
- FFT3dCuda(class LAMMPS *, MPI_Comm,int,int,int,int,int,int,int,int,int,
- int,int,int,int,int,int,int,int,int *,bool);
- ~FFT3dCuda();
- void compute(double *, double *, int);
- void timing1d(double *, int, int);
-
-#ifdef FFT_CUFFT
- void set_cudata(void* cudata,void* cudata2);
-#endif
- private:
- struct fft_plan_3d *plan;
-};
-
-}
-
-#endif /*FFT3D_WRAP_CUDA_H_*/
diff --git a/src/USER-CUDA/fix_addforce_cuda.cpp b/src/USER-CUDA/fix_addforce_cuda.cpp
deleted file mode 100644
index 5462668f4f..0000000000
--- a/src/USER-CUDA/fix_addforce_cuda.cpp
+++ /dev/null
@@ -1,193 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-
-#include <cstring>
-#include <cstdlib>
-#include "fix_addforce_cuda.h"
-#include "fix_addforce_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "respa.h"
-#include "error.h"
-#include "force.h"
-#include "domain.h"
-#include "user_cuda.h"
-#include "memory.h"
-#include "cuda_modify_flags.h"
-
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixAddForceCuda::FixAddForceCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg < 6) error->all(FLERR,"Illegal fix addforce/cuda command");
-
- scalar_flag = 1;
- vector_flag = 1;
- size_vector = 3;
- global_freq = 1;
- extscalar = 1;
- extvector = 1;
-
- xvalue = force->numeric(FLERR,arg[3]);
- yvalue = force->numeric(FLERR,arg[4]);
- zvalue = force->numeric(FLERR,arg[5]);
-
- // optional args
-
- iregion = -1;
-
- int iarg = 6;
- while (iarg < narg) {
- if (strcmp(arg[iarg],"region") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix addforce/cuda command");
- iregion = domain->find_region(arg[iarg+1]);
- if (iregion == -1) error->all(FLERR,"Fix addforce/cuda region ID does not exist");
- iarg += 2;
- } else error->all(FLERR,"Illegal fix addforce/cuda command");
- }
-
- if(iregion!=-1) error->all(FLERR,"Error: fix addforce/cuda does not currently support 'region' option");
-
- force_flag = 0;
- foriginal[0] = foriginal[1] = foriginal[2] = foriginal[3] = 0.0;
- cu_foriginal = NULL;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixAddForceCuda::setmask()
-{
- int mask = 0;
- mask |= POST_FORCE_CUDA;
- mask |= THERMO_ENERGY_CUDA;
- mask |= POST_FORCE_RESPA;
- mask |= MIN_POST_FORCE_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAddForceCuda::init()
-{
- if(not cu_foriginal)
- cu_foriginal = new cCudaData<double, F_CFLOAT, x> (foriginal,4);
- if (strstr(update->integrate_style,"respa"))
- nlevels_respa = ((Respa *) update->integrate)->nlevels;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAddForceCuda::setup(int vflag)
-{
- MYDBG( printf("# CUDA: FixAddForceCuda::setup\n"); )
-
- if (strstr(update->integrate_style,"verlet"))
- {
- Cuda_FixAddForceCuda_Init(&cuda->shared_data);
- cuda->cu_f->upload();
- post_force(vflag);
- cuda->cu_f->download();
-
- }
- else {
- ((Respa *) update->integrate)->copy_flevel_f(nlevels_respa-1);
- cuda->cu_f->download();
- post_force_respa(vflag,nlevels_respa-1,0);
- cuda->cu_f->upload();
- ((Respa *) update->integrate)->copy_f_flevel(nlevels_respa-1);
- }
- MYDBG( printf("# CUDA: FixAddForceCuda::setup done\n"); )
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAddForceCuda::min_setup(int vflag)
-{
- post_force(vflag);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAddForceCuda::post_force(int vflag)
-{
- MYDBG( printf("# CUDA: FixAddForceCuda::postforce start\n"); )
- force_flag = 0;
- cu_foriginal->memset_device(0);
- Cuda_FixAddForceCuda_PostForce(&cuda->shared_data, groupbit, xvalue, yvalue,zvalue,(F_CFLOAT*) cu_foriginal->dev_data());
- cu_foriginal->download();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAddForceCuda::post_force_respa(int vflag, int ilevel, int iloop)
-{
- if (ilevel == nlevels_respa-1) post_force(vflag);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAddForceCuda::min_post_force(int vflag)
-{
- post_force(vflag);
-}
-
-/* ----------------------------------------------------------------------
- potential energy of added force
-------------------------------------------------------------------------- */
-
-double FixAddForceCuda::compute_scalar()
-{
- // only sum across procs one time
-
- if (force_flag == 0) {
- MPI_Allreduce(foriginal,foriginal_all,4,MPI_DOUBLE,MPI_SUM,world);
- force_flag = 1;
- }
- return foriginal_all[0];
-}
-
-/* ----------------------------------------------------------------------
- return components of total force on fix group before force was changed
-------------------------------------------------------------------------- */
-
-double FixAddForceCuda::compute_vector(int n)
-{
- // only sum across procs one time
-
- if (force_flag == 0) {
- MPI_Allreduce(foriginal,foriginal_all,4,MPI_DOUBLE,MPI_SUM,world);
- force_flag = 1;
- }
- return foriginal_all[n+1];
-}
diff --git a/src/USER-CUDA/fix_addforce_cuda.h b/src/USER-CUDA/fix_addforce_cuda.h
deleted file mode 100644
index 043cae6d21..0000000000
--- a/src/USER-CUDA/fix_addforce_cuda.h
+++ /dev/null
@@ -1,64 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(addforce/cuda,FixAddForceCuda)
-
-#else
-
-#ifndef LMP_FIX_ADD_FORCE_CUDA_H
-#define LMP_FIX_ADD_FORCE_CUDA_H
-
-#include "fix.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class FixAddForceCuda : public Fix {
- public:
- FixAddForceCuda(class LAMMPS *, int, char **);
- int setmask();
- void init();
- void setup(int);
- void min_setup(int);
- void post_force(int);
- void post_force_respa(int, int, int);
- void min_post_force(int);
- double compute_scalar();
- double compute_vector(int);
-
- private:
- class Cuda *cuda;
- int iregion;
- double xvalue,yvalue,zvalue;
- double foriginal[4],foriginal_all[4];
- cCudaData<double , F_CFLOAT , x>* cu_foriginal;
- int force_flag;
- int nlevels_respa;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_aveforce_cuda.cpp b/src/USER-CUDA/fix_aveforce_cuda.cpp
deleted file mode 100644
index 9b4ceaa67c..0000000000
--- a/src/USER-CUDA/fix_aveforce_cuda.cpp
+++ /dev/null
@@ -1,262 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-
-#include <mpi.h>
-#include <cstring>
-#include <cstdlib>
-#include "fix_aveforce_cuda.h"
-#include "fix_aveforce_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "respa.h"
-#include "domain.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-#include "variable.h"
-#include "input.h"
-#include "modify.h"
-#include "atom_masks.h"
-#include "error.h"
-#include "force.h"
-
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-enum{NONE,CONSTANT,EQUAL};
-
-/* ---------------------------------------------------------------------- */
-
-FixAveForceCuda::FixAveForceCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg != 6) error->all(FLERR,"Illegal fix aveforce command");
-
- vector_flag = 1;
- size_vector = 3;
- global_freq = 1;
- extvector = 1;
-
- xstr = ystr = zstr = NULL;
- xvalue = yvalue = zvalue = 0;
-
- if (strstr(arg[3],"v_") == arg[3]) {
- int n = strlen(&arg[3][2]) + 1;
- xstr = new char[n];
- strcpy(xstr,&arg[3][2]);
- } else if (strcmp(arg[3],"NULL") == 0) {
- xstyle = NONE;
- } else {
- xvalue = force->numeric(FLERR,arg[3]);
- xstyle = CONSTANT;
- }
- if (strstr(arg[4],"v_") == arg[4]) {
- int n = strlen(&arg[4][2]) + 1;
- ystr = new char[n];
- strcpy(ystr,&arg[4][2]);
- } else if (strcmp(arg[4],"NULL") == 0) {
- ystyle = NONE;
- } else {
- yvalue = force->numeric(FLERR,arg[4]);
- ystyle = CONSTANT;
- }
- if (strstr(arg[5],"v_") == arg[5]) {
- int n = strlen(&arg[5][2]) + 1;
- zstr = new char[n];
- strcpy(zstr,&arg[5][2]);
- } else if (strcmp(arg[5],"NULL") == 0) {
- zstyle = NONE;
- } else {
- zvalue = force->numeric(FLERR,arg[5]);
- zstyle = CONSTANT;
- }
-
- // optional args
-
- iregion = -1;
-
- int iarg = 6;
- while (iarg < narg) {
- if (strcmp(arg[iarg],"region") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix aveforce command");
- iregion = domain->find_region(arg[iarg+1]);
- if (iregion == -1) error->all(FLERR,"Fix aveforce region ID does not exist");
- iarg += 2;
- } else error->all(FLERR,"Illegal fix aveforce command");
-
- }
-
- if(iregion!=-1) error->all(FLERR,"Error: fix aveforce/cuda does not currently support 'region' option");
-
- foriginal_all[0] = foriginal_all[1] = foriginal_all[2] = foriginal_all[3] = 0.0;
- foriginal[0] = foriginal[1] = foriginal[2] = foriginal[3] = 0.0;
- cu_foriginal = NULL;
-
-}
-
-FixAveForceCuda::~FixAveForceCuda()
-{
- delete [] xstr;
- delete [] ystr;
- delete [] zstr;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixAveForceCuda::setmask()
-{
- int mask = 0;
- mask |= POST_FORCE_CUDA;
- mask |= POST_FORCE_RESPA;
- mask |= MIN_POST_FORCE_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAveForceCuda::init()
-{
- if(not cu_foriginal)
- cu_foriginal = new cCudaData<double, F_CFLOAT, x> (foriginal,4);
-
- if (xstr) {
- xvar = input->variable->find(xstr);
- if (xvar < 0)
- error->all(FLERR,"Variable name for fix aveforce does not exist");
- if (input->variable->equalstyle(xvar)) xstyle = EQUAL;
- else error->all(FLERR,"Variable for fix aveforce is invalid style");
- }
- if (ystr) {
- yvar = input->variable->find(ystr);
- if (yvar < 0)
- error->all(FLERR,"Variable name for fix aveforce does not exist");
- if (input->variable->equalstyle(yvar)) ystyle = EQUAL;
- else error->all(FLERR,"Variable for fix aveforce is invalid style");
- }
- if (zstr) {
- zvar = input->variable->find(zstr);
- if (zvar < 0)
- error->all(FLERR,"Variable name for fix aveforce does not exist");
- if (input->variable->equalstyle(zvar)) zstyle = EQUAL;
- else error->all(FLERR,"Variable for fix aveforce is invalid style");
- }
-
- if (xstyle == EQUAL || ystyle == EQUAL || zstyle == EQUAL) varflag = EQUAL;
- else varflag = CONSTANT;
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAveForceCuda::setup(int vflag)
-{
- if (strstr(update->integrate_style,"verlet"))
- {
- Cuda_FixAveForceCuda_Init(&cuda->shared_data);
- cuda->cu_f->upload();
- post_force(vflag);
- cuda->cu_f->download();
-
- }
- else
- {
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAveForceCuda::min_setup(int vflag)
-{
- post_force(vflag);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAveForceCuda::post_force(int vflag)
-{
- // sum forces on participating atoms
-
- cu_foriginal->memset_device(0);
- Cuda_FixAveForceCuda_PostForce_FOrg(&cuda->shared_data, groupbit,(F_CFLOAT*) cu_foriginal->dev_data());
- cu_foriginal->download();
-
- // average the force on participating atoms
- // add in requested amount
-
- MPI_Allreduce(foriginal,foriginal_all,4,MPI_DOUBLE,MPI_SUM,world);
- int ncount = static_cast<int> (foriginal_all[3]);
- if (ncount == 0) return;
-
- if (varflag == EQUAL) {
- unsigned int datamask = EMPTY_MASK;
- if (xstyle == EQUAL) datamask &= input->variable->data_mask(xstr);
- if (ystyle == EQUAL) datamask &= input->variable->data_mask(ystr);
- if (zstyle == EQUAL) datamask &= input->variable->data_mask(zstr);
-
- cuda->download(datamask);
- modify->clearstep_compute();
- if (xstyle == EQUAL) xvalue = input->variable->compute_equal(xvar);
- if (ystyle == EQUAL) yvalue = input->variable->compute_equal(yvar);
- if (zstyle == EQUAL) zvalue = input->variable->compute_equal(zvar);
- modify->addstep_compute(update->ntimestep + 1);
- }
-
- double fave[3];
- fave[0] = foriginal_all[0]/ncount + xvalue;
- fave[1] = foriginal_all[1]/ncount + yvalue;
- fave[2] = foriginal_all[2]/ncount + zvalue;
-
- // set force of all participating atoms to same value
- // only for active dimensions
-
- Cuda_FixAveForceCuda_PostForce_Set(&cuda->shared_data, groupbit,!(xstyle==NONE),!(ystyle==NONE),!(zstyle==NONE),fave[0],fave[1],fave[2]);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAveForceCuda::post_force_respa(int vflag, int ilevel, int iloop)
-{
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixAveForceCuda::min_post_force(int vflag)
-{
- post_force(vflag);
-}
-
-/* ----------------------------------------------------------------------
- return components of total force on fix group before force was changed
-------------------------------------------------------------------------- */
-
-double FixAveForceCuda::compute_vector(int n)
-{
- return foriginal_all[n];
-}
diff --git a/src/USER-CUDA/fix_aveforce_cuda.h b/src/USER-CUDA/fix_aveforce_cuda.h
deleted file mode 100644
index c22e702ee2..0000000000
--- a/src/USER-CUDA/fix_aveforce_cuda.h
+++ /dev/null
@@ -1,68 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(aveforce/cuda,FixAveForceCuda)
-
-#else
-
-
-#ifndef LMP_FIX_AVE_FORCE_CUDA_H
-#define LMP_FIX_AVE_FORCE_CUDA_H
-
-#include "fix.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class FixAveForceCuda : public Fix {
- public:
- FixAveForceCuda(class LAMMPS *, int, char **);
- ~FixAveForceCuda();
- int setmask();
- void init();
- void setup(int);
- void min_setup(int);
- void post_force(int);
- void post_force_respa(int, int, int);
- void min_post_force(int);
- double compute_vector(int);
-
- private:
- class Cuda *cuda;
- char *xstr,*ystr,*zstr;
- int iregion;
- double xvalue,yvalue,zvalue;
- double foriginal_all[4];
- double foriginal[4];
- cCudaData<double , F_CFLOAT , x>* cu_foriginal;
- int varflag;
- int xvar,yvar,zvar,xstyle,ystyle,zstyle;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_enforce2d_cuda.cpp b/src/USER-CUDA/fix_enforce2d_cuda.cpp
deleted file mode 100644
index d10edf1cce..0000000000
--- a/src/USER-CUDA/fix_enforce2d_cuda.cpp
+++ /dev/null
@@ -1,171 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstring>
-#include "fix_enforce2d_cuda.h"
-#include "fix_enforce2d_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "domain.h"
-#include "respa.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixEnforce2DCuda::FixEnforce2DCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg != 3) error->all(FLERR,"Illegal fix enforce2d command");
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixEnforce2DCuda::setmask()
-{
- int mask = 0;
- mask |= POST_FORCE_CUDA;
- mask |= POST_FORCE_RESPA;
- mask |= MIN_POST_FORCE_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixEnforce2DCuda::init()
-{
- if (domain->dimension == 3)
- error->all(FLERR,"Cannot use fix enforce2d/cuda with 3d simulation");
- if (atom->omega_flag)
- error->warning(FLERR,"Enforce2d/cuda does not support omega_flag on gpu yet. Will be handled on cpu.");
-
- if (atom->angmom_flag)
- error->warning(FLERR,"Enforce2d/cuda does not support angmom_flag (angular momentum) on gpu yet. Will be handled on cpu.");
-
- if (atom->torque_flag)
- error->warning(FLERR,"Enforce2d/cuda does not support torque_flag on gpu yet. Will be handled on cpu.");
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixEnforce2DCuda::setup(int vflag)
-{
- if (strstr(update->integrate_style,"verlet"))
- {
- Cuda_FixEnforce2dCuda_Init(&cuda->shared_data);
- cuda->cu_f->upload();
- cuda->cu_v->upload();
- post_force(vflag);
- cuda->cu_f->download();
- cuda->cu_v->download();
- }
- else {
- int nlevels_respa = ((Respa *) update->integrate)->nlevels;
- for (int ilevel = 0; ilevel < nlevels_respa; ilevel++) {
- ((Respa *) update->integrate)->copy_flevel_f(ilevel);
- post_force_respa(vflag,ilevel,0);
- ((Respa *) update->integrate)->copy_f_flevel(ilevel);
- }
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixEnforce2DCuda::min_setup(int vflag)
-{
- post_force(vflag);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixEnforce2DCuda::post_force(int vflag)
-{
- Cuda_FixEnforce2dCuda_PostForce(&cuda->shared_data, groupbit);
-
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
- if (igroup == atom->firstgroup) nlocal = atom->nfirst;
-
- if (atom->omega_flag) {
- double **omega = atom->omega;
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- omega[i][0] = 0.0;
- omega[i][1] = 0.0;
- }
- }
-
- if (atom->angmom_flag) {
- double **angmom = atom->angmom;
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- angmom[i][0] = 0.0;
- angmom[i][1] = 0.0;
- }
- }
-
- if (atom->torque_flag) {
- double **torque = atom->torque;
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- torque[i][0] = 0.0;
- torque[i][1] = 0.0;
- }
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixEnforce2DCuda::post_force_respa(int vflag, int ilevel, int iloop)
-{
- post_force(vflag);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixEnforce2DCuda::min_post_force(int vflag)
-{
- post_force(vflag);
-}
diff --git a/src/USER-CUDA/fix_enforce2d_cuda.h b/src/USER-CUDA/fix_enforce2d_cuda.h
deleted file mode 100644
index 63bf289d9e..0000000000
--- a/src/USER-CUDA/fix_enforce2d_cuda.h
+++ /dev/null
@@ -1,55 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(enforce2d/cuda,FixEnforce2DCuda)
-
-#else
-
-#ifndef LMP_FIX_ENFORCE2D_CUDA_H
-#define LMP_FIX_ENFORCE2D_CUDA_H
-
-#include "fix.h"
-
-namespace LAMMPS_NS {
-
-class FixEnforce2DCuda : public Fix {
- public:
- FixEnforce2DCuda(class LAMMPS *, int, char **);
- int setmask();
- void init();
- void setup(int);
- void min_setup(int);
- void post_force(int);
- void post_force_respa(int, int, int);
- void min_post_force(int);
-
- private:
- class Cuda *cuda;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_freeze_cuda.cpp b/src/USER-CUDA/fix_freeze_cuda.cpp
deleted file mode 100644
index c4a04af564..0000000000
--- a/src/USER-CUDA/fix_freeze_cuda.cpp
+++ /dev/null
@@ -1,137 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-#include <cstring>
-#include <cstdlib>
-#include "fix_freeze_cuda.h"
-#include "fix_freeze_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "respa.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "memory.h"
-#include "modify.h"
-#include "cuda_modify_flags.h"
-
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixFreezeCuda::FixFreezeCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg != 3) error->all(FLERR,"Illegal fix freeze command");
-
- if (!atom->torque_flag)
- error->all(FLERR,"Fix freeze requires atom attribute torque");
-
- vector_flag = 1;
- size_vector = 3;
- global_freq = 1;
- extvector = 1;
-
-
-
- force_flag = 0;
- foriginal[0] = foriginal[1] = foriginal[2] = 0.0;
- cu_foriginal=NULL;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixFreezeCuda::setmask()
-{
- int mask = 0;
- mask |= POST_FORCE_CUDA;
- mask |= THERMO_ENERGY_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixFreezeCuda::init()
-{
- if(not cu_foriginal)
- cu_foriginal = new cCudaData<double, F_CFLOAT, x> (foriginal,3);
- int count = 0;
- for (int i = 0; i < modify->nfix; i++)
- if (strcmp(modify->fix[i]->style,"freeze") == 0) count++;
- if (count > 1) error->all(FLERR,"More than one fix freeze");
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixFreezeCuda::setup(int vflag)
-{
- MYDBG( printf("# CUDA: FixFreezeCuda::setup\n"); )
-
- if (strstr(update->integrate_style,"verlet"))
- {
- Cuda_FixFreezeCuda_Init(&cuda->shared_data);
- cuda->cu_f->upload();
- post_force(vflag);
- cuda->cu_f->download();
-
- }
-
- MYDBG( printf("# CUDA: FixFreezeCuda::setup done\n"); )
-}
-
-/* ---------------------------------------------------------------------- */
-
-/* ---------------------------------------------------------------------- */
-
-void FixFreezeCuda::post_force(int vflag)
-{
- MYDBG( printf("# CUDA: FixFreezeCuda::postforce start\n"); )
- force_flag = 0;
- cu_foriginal->memset_device(0);
- Cuda_FixFreezeCuda_PostForce(&cuda->shared_data, groupbit, (F_CFLOAT*) cu_foriginal->dev_data());
- cu_foriginal->download();
-}
-
-/* ---------------------------------------------------------------------- */
-
-
-
-/* ----------------------------------------------------------------------
- return components of total force on fix group before force was changed
-------------------------------------------------------------------------- */
-
-double FixFreezeCuda::compute_vector(int n)
-{
- // only sum across procs one time
-
- if (force_flag == 0) {
- MPI_Allreduce(foriginal,foriginal_all,3,MPI_DOUBLE,MPI_SUM,world);
- force_flag = 1;
- }
- return foriginal_all[n+1];
-}
diff --git a/src/USER-CUDA/fix_freeze_cuda.h b/src/USER-CUDA/fix_freeze_cuda.h
deleted file mode 100644
index 9f6a1a99f0..0000000000
--- a/src/USER-CUDA/fix_freeze_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(freeze/cuda,FixFreezeCuda)
-
-#else
-
-#ifndef LMP_FIX_FREEZE_CUDA_H
-#define LMP_FIX_FREEZE_CUDA_H
-
-#include "fix.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class FixFreezeCuda : public Fix {
- public:
- FixFreezeCuda(class LAMMPS *, int, char **);
- int setmask();
- void init();
- void setup(int);
- void post_force(int);
- double compute_vector(int);
-
- private:
- class Cuda *cuda;
- double foriginal[3],foriginal_all[3];
- cCudaData<double , F_CFLOAT , x>* cu_foriginal;
- int force_flag;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_gravity_cuda.cpp b/src/USER-CUDA/fix_gravity_cuda.cpp
deleted file mode 100644
index 34107ed593..0000000000
--- a/src/USER-CUDA/fix_gravity_cuda.cpp
+++ /dev/null
@@ -1,180 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-#include <cmath>
-#include <cstdio>
-#include <cstring>
-#include <cstdlib>
-#include "fix_gravity_cuda.h"
-#include "fix_gravity_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "domain.h"
-#include "respa.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-#include "math_const.h"
-#include "error.h"
-#include "force.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-using namespace MathConst;
-
-enum{CHUTE,SPHERICAL,GRADIENT,VECTOR};
-
-/* ---------------------------------------------------------------------- */
-
-FixGravityCuda::FixGravityCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg < 5) error->all(FLERR,"Illegal fix gravity command");
-
- magnitude = force->numeric(FLERR,arg[3]);
-
- if (strcmp(arg[4],"chute") == 0) {
- if (narg != 6) error->all(FLERR,"Illegal fix gravity command");
- style = CHUTE;
- phi = 0.0;
- theta = 180.0 - force->numeric(FLERR,arg[5]);
- } else if (strcmp(arg[4],"spherical") == 0) {
- if (narg != 7) error->all(FLERR,"Illegal fix gravity command");
- style = SPHERICAL;
- phi = force->numeric(FLERR,arg[5]);
- theta = force->numeric(FLERR,arg[6]);
- } else if (strcmp(arg[4],"gradient") == 0) {
- if (narg != 9) error->all(FLERR,"Illegal fix gravity command");
- style = GRADIENT;
- phi = force->numeric(FLERR,arg[5]);
- theta = force->numeric(FLERR,arg[6]);
- phigrad = force->numeric(FLERR,arg[7]);
- thetagrad = force->numeric(FLERR,arg[8]);
- } else if (strcmp(arg[4],"vector") == 0) {
- if (narg != 8) error->all(FLERR,"Illegal fix gravity command");
- style = VECTOR;
- xdir = force->numeric(FLERR,arg[5]);
- ydir = force->numeric(FLERR,arg[6]);
- zdir = force->numeric(FLERR,arg[7]);
- } else error->all(FLERR,"Illegal fix gravity command");
-
- degree2rad = MY_PI/180.0;
-
- if (style == CHUTE || style == SPHERICAL || style == GRADIENT) {
- if (domain->dimension == 3) {
- xgrav = sin(degree2rad * theta) * cos(degree2rad * phi);
- ygrav = sin(degree2rad * theta) * sin(degree2rad * phi);
- zgrav = cos(degree2rad * theta);
- } else {
- xgrav = sin(degree2rad * theta);
- ygrav = cos(degree2rad * theta);
- zgrav = 0.0;
- }
- } else if (style == VECTOR) {
- if (domain->dimension == 3) {
- double length = sqrt(xdir*xdir + ydir*ydir + zdir*zdir);
- xgrav = xdir/length;
- ygrav = ydir/length;
- zgrav = zdir/length;
- } else {
- double length = sqrt(xdir*xdir + ydir*ydir);
- xgrav = xdir/length;
- ygrav = ydir/length;
- zgrav = 0.0;
- }
- }
-
- time_origin = update->ntimestep;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixGravityCuda::setmask()
-{
- int mask = 0;
- mask |= POST_FORCE_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixGravityCuda::init()
-{
- dt = update->dt;
-
- xacc = magnitude*xgrav;
- yacc = magnitude*ygrav;
- zacc = magnitude*zgrav;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixGravityCuda::setup(int vflag)
-{
- MYDBG( printf("# CUDA: FixGravityCuda::setup\n"); )
-
- if (strstr(update->integrate_style,"verlet"))
- {
- Cuda_FixGravityCuda_Init(&cuda->shared_data);
- cuda->cu_f->upload();
- post_force(vflag);
- cuda->cu_f->download();
-
- }
- else {
- }
- MYDBG( printf("# CUDA: FixGravityCuda::setup done\n"); )
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixGravityCuda::post_force(int vflag)
-{
- // update direction of gravity vector if gradient style
-
- if (style == GRADIENT) {
- if (domain->dimension == 3) {
- double phi_current = degree2rad *
- (phi + (update->ntimestep - time_origin)*dt*phigrad*360.0);
- double theta_current = degree2rad *
- (theta + (update->ntimestep - time_origin)*dt*thetagrad*360.0);
- xgrav = sin(theta_current) * cos(phi_current);
- ygrav = sin(theta_current) * sin(phi_current);
- zgrav = cos(theta_current);
- } else {
- double theta_current = degree2rad *
- (theta + (update->ntimestep - time_origin)*dt*thetagrad*360.0);
- xgrav = sin(theta_current);
- ygrav = cos(theta_current);
- }
- xacc = magnitude*xgrav;
- yacc = magnitude*ygrav;
- zacc = magnitude*zgrav;
- }
-
- MYDBG( printf("# CUDA: FixGravityCuda::postforce start\n"); )
- Cuda_FixGravityCuda_PostForce(&cuda->shared_data, groupbit, xacc,yacc,zacc);
-}
diff --git a/src/USER-CUDA/fix_gravity_cuda.h b/src/USER-CUDA/fix_gravity_cuda.h
deleted file mode 100644
index 98d2586660..0000000000
--- a/src/USER-CUDA/fix_gravity_cuda.h
+++ /dev/null
@@ -1,60 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(gravity/cuda,FixGravityCuda)
-
-#else
-
-#ifndef LMP_FIX_GRAVITY_CUDA_H
-#define LMP_FIX_GRAVITY_CUDA_H
-
-#include "fix.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class FixGravityCuda : public Fix {
- public:
- FixGravityCuda(class LAMMPS *, int, char **);
- int setmask();
- void init();
- void setup(int);
- void post_force(int);
-
- private:
- class Cuda *cuda;
- int style;
- double magnitude,dt;
- double phi,theta,phigrad,thetagrad;
- double xdir,ydir,zdir;
- double xgrav,ygrav,zgrav,xacc,yacc,zacc;
- double degree2rad;
- int time_origin;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_nh_cuda.cpp b/src/USER-CUDA/fix_nh_cuda.cpp
deleted file mode 100644
index 1a5092a68f..0000000000
--- a/src/USER-CUDA/fix_nh_cuda.cpp
+++ /dev/null
@@ -1,2072 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing authors: Mark Stevens (SNL), Aidan Thompson (SNL)
-------------------------------------------------------------------------- */
-
-#include <cstring>
-#include <cstdlib>
-#include <cmath>
-#include "fix_nh_cuda.h"
-#include "atom.h"
-#include "force.h"
-#include "comm.h"
-#include "modify.h"
-#include "fix_deform.h"
-#include "compute.h"
-#include "kspace.h"
-#include "update.h"
-#include "respa.h"
-#include "domain.h"
-#include "memory.h"
-#include "error.h"
-#include "math_extra.h"
-#include "user_cuda.h"
-#include "fix_nh_cuda_cu.h"
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-enum{NOBIAS,BIAS};
-enum{NONE,XYZ,XY,YZ,XZ};
-enum{ISO,ANISO,TRICLINIC};
-
-/* ----------------------------------------------------------------------
- NVT,NPH,NPT integrators for improved Nose-Hoover equations of motion
- ---------------------------------------------------------------------- */
-
-FixNHCuda::FixNHCuda(LAMMPS *lmp, int narg, char **arg) : Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg < 4) error->all(FLERR,"Illegal fix nvt/npt/nph command");
-
- restart_global = 1;
- time_integrate = 1;
- scalar_flag = 1;
- vector_flag = 1;
- global_freq = 1;
- extscalar = 1;
- extvector = 0;
-
- triggerneighsq = -1;
- // default values
-
- pcouple = NONE;
- drag = 0.0;
- allremap = 1;
- mtchain = mpchain = 3;
- nc_tchain = nc_pchain = 1;
- mtk_flag = 1;
- deviatoric_flag = 0;
- nreset_h0 = 0;
-
- // Used by FixNVTSllod to preserve non-default value
-
- mtchain_default_flag = 1;
-
- tstat_flag = 0;
- double t_period = 0.0;
-
- double p_period[6];
- for (int i = 0; i < 6; i++) {
- p_start[i] = p_stop[i] = p_period[i] = 0.0;
- p_flag[i] = 0;
- }
-
- // process keywords
-
- dimension = domain->dimension;
-
- int iarg = 3;
-
- while (iarg < narg) {
- if (strcmp(arg[iarg],"temp") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- tstat_flag = 1;
- t_start = force->numeric(FLERR,arg[iarg+1]);
- t_stop = force->numeric(FLERR,arg[iarg+2]);
- t_period = force->numeric(FLERR,arg[iarg+3]);
- if (t_start < 0.0 || t_stop <= 0.0)
- error->all(FLERR,"Target T for fix nvt/npt/nph cannot be 0.0");
- iarg += 4;
-
- } else if (strcmp(arg[iarg],"iso") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- pcouple = XYZ;
- p_start[0] = p_start[1] = p_start[2] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[0] = p_stop[1] = p_stop[2] = force->numeric(FLERR,arg[iarg+2]);
- p_period[0] = p_period[1] = p_period[2] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[0] = p_flag[1] = p_flag[2] = 1;
- if (dimension == 2) {
- p_start[2] = p_stop[2] = p_period[2] = 0.0;
- p_flag[2] = 0;
- }
- iarg += 4;
- } else if (strcmp(arg[iarg],"aniso") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- pcouple = NONE;
- p_start[0] = p_start[1] = p_start[2] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[0] = p_stop[1] = p_stop[2] = force->numeric(FLERR,arg[iarg+2]);
- p_period[0] = p_period[1] = p_period[2] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[0] = p_flag[1] = p_flag[2] = 1;
- if (dimension == 2) {
- p_start[2] = p_stop[2] = p_period[2] = 0.0;
- p_flag[2] = 0;
- }
- iarg += 4;
- } else if (strcmp(arg[iarg],"tri") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- pcouple = NONE;
- p_start[0] = p_start[1] = p_start[2] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[0] = p_stop[1] = p_stop[2] = force->numeric(FLERR,arg[iarg+2]);
- p_period[0] = p_period[1] = p_period[2] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[0] = p_flag[1] = p_flag[2] = 1;
- p_start[3] = p_start[4] = p_start[5] = 0.0;
- p_stop[3] = p_stop[4] = p_stop[5] = 0.0;
- p_period[3] = p_period[4] = p_period[5] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[3] = p_flag[4] = p_flag[5] = 1;
- if (dimension == 2) {
- p_start[2] = p_stop[2] = p_period[2] = 0.0;
- p_flag[2] = 0;
- p_start[3] = p_stop[3] = p_period[3] = 0.0;
- p_flag[3] = 0;
- p_start[4] = p_stop[4] = p_period[4] = 0.0;
- p_flag[4] = 0;
- }
- iarg += 4;
-
- } else if (strcmp(arg[iarg],"x") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- p_start[0] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[0] = force->numeric(FLERR,arg[iarg+2]);
- p_period[0] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[0] = 1;
- deviatoric_flag = 1;
- iarg += 4;
- } else if (strcmp(arg[iarg],"y") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- p_start[1] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[1] = force->numeric(FLERR,arg[iarg+2]);
- p_period[1] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[1] = 1;
- deviatoric_flag = 1;
- iarg += 4;
- } else if (strcmp(arg[iarg],"z") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- p_start[2] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[2] = force->numeric(FLERR,arg[iarg+2]);
- p_period[2] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[2] = 1;
- deviatoric_flag = 1;
- iarg += 4;
- if (dimension == 2)
- error->all(FLERR,"Invalid fix nvt/npt/nph command for a 2d simulation");
-
- } else if (strcmp(arg[iarg],"yz") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- p_start[3] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[3] = force->numeric(FLERR,arg[iarg+2]);
- p_period[3] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[3] = 1;
- deviatoric_flag = 1;
- iarg += 4;
- if (dimension == 2)
- error->all(FLERR,"Invalid fix nvt/npt/nph command for a 2d simulation");
- } else if (strcmp(arg[iarg],"xz") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- p_start[4] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[4] = force->numeric(FLERR,arg[iarg+2]);
- p_period[4] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[4] = 1;
- deviatoric_flag = 1;
- iarg += 4;
- if (dimension == 2)
- error->all(FLERR,"Invalid fix nvt/npt/nph command for a 2d simulation");
- } else if (strcmp(arg[iarg],"xy") == 0) {
- if (iarg+4 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- p_start[5] = force->numeric(FLERR,arg[iarg+1]);
- p_stop[5] = force->numeric(FLERR,arg[iarg+2]);
- p_period[5] = force->numeric(FLERR,arg[iarg+3]);
- p_flag[5] = 1;
- deviatoric_flag = 1;
- iarg += 4;
-
- } else if (strcmp(arg[iarg],"couple") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- if (strcmp(arg[iarg+1],"xyz") == 0) pcouple = XYZ;
- else if (strcmp(arg[iarg+1],"xy") == 0) pcouple = XY;
- else if (strcmp(arg[iarg+1],"yz") == 0) pcouple = YZ;
- else if (strcmp(arg[iarg+1],"xz") == 0) pcouple = XZ;
- else if (strcmp(arg[iarg+1],"none") == 0) pcouple = NONE;
- else error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
-
- } else if (strcmp(arg[iarg],"drag") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- drag = force->numeric(FLERR,arg[iarg+1]);
- if (drag < 0.0) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else if (strcmp(arg[iarg],"dilate") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- if (strcmp(arg[iarg+1],"all") == 0) allremap = 1;
- else if (strcmp(arg[iarg+1],"partial") == 0) allremap = 0;
- else error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else if (strcmp(arg[iarg],"tchain") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- mtchain = force->inumeric(FLERR,arg[iarg+1]);
- if (mtchain < 1) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else if (strcmp(arg[iarg],"pchain") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- mpchain = force->inumeric(FLERR,arg[iarg+1]);
- if (mpchain < 0) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else if (strcmp(arg[iarg],"mtk") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- if (strcmp(arg[iarg+1],"yes") == 0) mtk_flag = 1;
- else if (strcmp(arg[iarg+1],"no") == 0) mtk_flag = 0;
- else error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else if (strcmp(arg[iarg],"tloop") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- nc_tchain = force->inumeric(FLERR,arg[iarg+1]);
- if (nc_tchain < 0) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else if (strcmp(arg[iarg],"ploop") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- nc_pchain = force->inumeric(FLERR,arg[iarg+1]);
- if (nc_pchain < 0) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else if (strcmp(arg[iarg],"nreset") == 0) {
- if (iarg+2 > narg) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- nreset_h0 = force->inumeric(FLERR,arg[iarg+1]);
- if (nreset_h0 < 0) error->all(FLERR,"Illegal fix nvt/npt/nph command");
- iarg += 2;
- } else error->all(FLERR,"Illegal fix nvt/npt/nph command");
- }
-
- // error checks
-
- if (dimension == 2 && (p_flag[2] || p_flag[3] || p_flag[4]))
- error->all(FLERR,"Invalid fix nvt/npt/nph command for a 2d simulation");
- if (dimension == 2 && (pcouple == YZ || pcouple == XZ))
- error->all(FLERR,"Invalid fix nvt/npt/nph command for a 2d simulation");
-
- if (pcouple == XYZ && (p_flag[0] == 0 || p_flag[1] == 0))
- error->all(FLERR,"Invalid fix nvt/npt/nph command pressure settings");
- if (pcouple == XYZ && dimension == 3 && p_flag[2] == 0)
- error->all(FLERR,"Invalid fix nvt/npt/nph command pressure settings");
- if (pcouple == XY && (p_flag[0] == 0 || p_flag[1] == 0))
- error->all(FLERR,"Invalid fix nvt/npt/nph command pressure settings");
- if (pcouple == YZ && (p_flag[1] == 0 || p_flag[2] == 0))
- error->all(FLERR,"Invalid fix nvt/npt/nph command pressure settings");
- if (pcouple == XZ && (p_flag[0] == 0 || p_flag[2] == 0))
- error->all(FLERR,"Invalid fix nvt/npt/nph command pressure settings");
-
- if (p_flag[0] && domain->xperiodic == 0)
- error->all(FLERR,"Cannot use fix nvt/npt/nph on a non-periodic dimension");
- if (p_flag[1] && domain->yperiodic == 0)
- error->all(FLERR,"Cannot use fix nvt/npt/nph on a non-periodic dimension");
- if (p_flag[2] && domain->zperiodic == 0)
- error->all(FLERR,"Cannot use fix nvt/npt/nph on a non-periodic dimension");
- if (p_flag[3] && domain->zperiodic == 0)
- error->all(FLERR,"Cannot use fix nvt/npt/nph on a 2nd non-periodic dimension");
- if (p_flag[4] && domain->zperiodic == 0)
- error->all(FLERR,"Cannot use fix nvt/npt/nph on a 2nd non-periodic dimension");
- if (p_flag[5] && domain->yperiodic == 0)
- error->all(FLERR,"Cannot use fix nvt/npt/nph on a 2nd non-periodic dimension");
-
- if (!domain->triclinic && (p_flag[3] || p_flag[4] || p_flag[5]))
- error->all(FLERR,"Can not specify Pxy/Pxz/Pyz in "
- "fix nvt/npt/nph with non-triclinic box");
-
- if (pcouple == XYZ && dimension == 3 &&
- (p_start[0] != p_start[1] || p_start[0] != p_start[2] ||
- p_stop[0] != p_stop[1] || p_stop[0] != p_stop[2] ||
- p_period[0] != p_period[1] || p_period[0] != p_period[2]))
- error->all(FLERR,"Invalid fix nvt/npt/nph pressure settings");
- if (pcouple == XYZ && dimension == 2 &&
- (p_start[0] != p_start[1] || p_stop[0] != p_stop[1] ||
- p_period[0] != p_period[1]))
- error->all(FLERR,"Invalid fix nvt/npt/nph pressure settings");
- if (pcouple == XY &&
- (p_start[0] != p_start[1] || p_stop[0] != p_stop[1] ||
- p_period[0] != p_period[1]))
- error->all(FLERR,"Invalid fix nvt/npt/nph pressure settings");
- if (pcouple == YZ &&
- (p_start[1] != p_start[2] || p_stop[1] != p_stop[2] ||
- p_period[1] != p_period[2]))
- error->all(FLERR,"Invalid fix nvt/npt/nph pressure settings");
- if (pcouple == XZ &&
- (p_start[0] != p_start[2] || p_stop[0] != p_stop[2] ||
- p_period[0] != p_period[2]))
- error->all(FLERR,"Invalid fix nvt/npt/nph pressure settings");
-
- if ((tstat_flag && t_period <= 0.0) ||
- (p_flag[0] && p_period[0] <= 0.0) ||
- (p_flag[1] && p_period[1] <= 0.0) ||
- (p_flag[2] && p_period[2] <= 0.0) ||
- (p_flag[3] && p_period[3] <= 0.0) ||
- (p_flag[4] && p_period[4] <= 0.0) ||
- (p_flag[5] && p_period[5] <= 0.0))
- error->all(FLERR,"Fix nvt/npt/nph damping parameters must be > 0.0");
-
- // set pstat_flag and box change variables
-
- pstat_flag = 0;
- for (int i = 0; i < 6; i++)
- if (p_flag[i]) pstat_flag = 1;
-
- if (pstat_flag) {
- if (p_flag[0] || p_flag[1] || p_flag[2]) box_change_size = 1;
- if (p_flag[3] || p_flag[4] || p_flag[5]) box_change_shape = 1;
- no_change_box = 1;
- if (allremap == 0) restart_pbc = 1;
- }
-
- // pstyle = TRICLINIC if any off-diagonal term is controlled -> 6 dof
- // else pstyle = ISO if XYZ coupling or XY coupling in 2d -> 1 dof
- // else pstyle = ANISO -> 3 dof
-
- if (p_flag[3] || p_flag[4] || p_flag[5]) pstyle = TRICLINIC;
- else if (pcouple == XYZ || (dimension == 2 && pcouple == XY)) pstyle = ISO;
- else pstyle = ANISO;
-
- // convert input periods to frequencies
-
- t_freq = 0.0;
- p_freq[0] = p_freq[1] = p_freq[2] = p_freq[3] = p_freq[4] = p_freq[5] = 0.0;
-
- if (tstat_flag) t_freq = 1.0 / t_period;
- if (p_flag[0]) p_freq[0] = 1.0 / p_period[0];
- if (p_flag[1]) p_freq[1] = 1.0 / p_period[1];
- if (p_flag[2]) p_freq[2] = 1.0 / p_period[2];
- if (p_flag[3]) p_freq[3] = 1.0 / p_period[3];
- if (p_flag[4]) p_freq[4] = 1.0 / p_period[4];
- if (p_flag[5]) p_freq[5] = 1.0 / p_period[5];
-
- // Nose/Hoover temp and pressure init
-
- size_vector = 0;
-
- if (tstat_flag) {
- int ich;
- eta = new double[mtchain];
-
- // add one extra dummy thermostat, set to zero
-
- eta_dot = new double[mtchain+1];
- eta_dot[mtchain] = 0.0;
- eta_dotdot = new double[mtchain];
- for (ich = 0; ich < mtchain; ich++) {
- eta[ich] = eta_dot[ich] = eta_dotdot[ich] = 0.0;
- }
- eta_mass = new double[mtchain];
- size_vector += 2*2*mtchain;
- }
-
- if (pstat_flag) {
- omega[0] = omega[1] = omega[2] = 0.0;
- omega_dot[0] = omega_dot[1] = omega_dot[2] = 0.0;
- omega_mass[0] = omega_mass[1] = omega_mass[2] = 0.0;
- omega[3] = omega[4] = omega[5] = 0.0;
- omega_dot[3] = omega_dot[4] = omega_dot[5] = 0.0;
- omega_mass[3] = omega_mass[4] = omega_mass[5] = 0.0;
- if (pstyle == ISO) size_vector += 2*2*1;
- else if (pstyle == ANISO) size_vector += 2*2*3;
- else if (pstyle == TRICLINIC) size_vector += 2*2*6;
-
- if (mpchain) {
- int ich;
- etap = new double[mpchain];
-
- // add one extra dummy thermostat, set to zero
-
- etap_dot = new double[mpchain+1];
- etap_dot[mpchain] = 0.0;
- etap_dotdot = new double[mpchain];
- for (ich = 0; ich < mpchain; ich++) {
- etap[ich] = etap_dot[ich] =
- etap_dotdot[ich] = 0.0;
- }
- etap_mass = new double[mpchain];
- size_vector += 2*2*mpchain;
- }
-
- if (deviatoric_flag) size_vector += 1;
- }
-
- nrigid = 0;
- rfix = NULL;
-
- // initialize vol0,t0 to zero to signal uninitialized
- // values then assigned in init(), if necessary
-
- vol0 = t0 = 0.0;
-}
-
-/* ---------------------------------------------------------------------- */
-
-FixNHCuda::~FixNHCuda()
-{
- delete [] rfix;
-
- // delete temperature and pressure if fix created them
-
- if (tflag) modify->delete_compute(id_temp);
- delete [] id_temp;
-
- if (tstat_flag) {
- delete [] eta;
- delete [] eta_dot;
- delete [] eta_dotdot;
- delete [] eta_mass;
- }
-
- if (pstat_flag) {
- if (pflag) modify->delete_compute(id_press);
- delete [] id_press;
- if (mpchain) {
- delete [] etap;
- delete [] etap_dot;
- delete [] etap_dotdot;
- delete [] etap_mass;
- }
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixNHCuda::setmask()
-{
- int mask = 0;
- mask |= INITIAL_INTEGRATE_CUDA;
- mask |= FINAL_INTEGRATE_CUDA;
- mask |= THERMO_ENERGY_CUDA;
- //mask |= INITIAL_INTEGRATE_RESPA;
- //mask |= FINAL_INTEGRATE_RESPA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNHCuda::init()
-{
- // insure no conflict with fix deform
-
- if (pstat_flag)
- for (int i = 0; i < modify->nfix; i++)
- if (strcmp(modify->fix[i]->style,"deform") == 0) {
- int *dimflag = ((FixDeform *) modify->fix[i])->dimflag;
- if ((p_flag[0] && dimflag[0]) || (p_flag[1] && dimflag[1]) ||
- (p_flag[2] && dimflag[2]) || (p_flag[3] && dimflag[3]) ||
- (p_flag[4] && dimflag[4]) || (p_flag[5] && dimflag[5]))
- error->all(FLERR,"Cannot use fix npt and fix deform on "
- "same component of stress tensor");
- }
-
- // set temperature and pressure ptrs
-
- int icompute = modify->find_compute(id_temp);
- if (icompute < 0)
- error->all(FLERR,"Temperature ID for fix nvt/nph/npt does not exist");
- temperature = modify->compute[icompute];
-
- if (temperature->tempbias) which = BIAS;
- else which = NOBIAS;
-
- if (pstat_flag) {
- icompute = modify->find_compute(id_press);
- if (icompute < 0) error->all(FLERR,"Pressure ID for fix npt/nph does not exist");
- pressure = modify->compute[icompute];
- }
-
- // set timesteps and frequencies
-
- dtv = update->dt;
- dtf = 0.5 * update->dt * force->ftm2v;
- dthalf = 0.5 * update->dt;
- dt4 = 0.25 * update->dt;
- dt8 = 0.125 * update->dt;
- dto = dthalf;
-
- p_freq_max = 0.0;
- if (pstat_flag) {
- p_freq_max = MAX(p_freq[0],p_freq[1]);
- p_freq_max = MAX(p_freq_max,p_freq[2]);
- if (pstyle == TRICLINIC) {
- p_freq_max = MAX(p_freq_max,p_freq[3]);
- p_freq_max = MAX(p_freq_max,p_freq[4]);
- p_freq_max = MAX(p_freq_max,p_freq[5]);
- }
- pdrag_factor = 1.0 - (update->dt * p_freq_max * drag / nc_pchain);
- }
-
- if (tstat_flag)
- tdrag_factor = 1.0 - (update->dt * t_freq * drag / nc_tchain);
-
- // tally the number of dimensions that are barostatted
- // also compute the initial volume and reference cell
- // set initial volume and reference cell, if not already done
-
- if (pstat_flag) {
- pdim = p_flag[0] + p_flag[1] + p_flag[2];
- if (vol0 == 0.0) {
- if (dimension == 3) vol0 = domain->xprd * domain->yprd * domain->zprd;
- else vol0 = domain->xprd * domain->yprd;
- h0_inv[0] = domain->h_inv[0];
- h0_inv[1] = domain->h_inv[1];
- h0_inv[2] = domain->h_inv[2];
- h0_inv[3] = domain->h_inv[3];
- h0_inv[4] = domain->h_inv[4];
- h0_inv[5] = domain->h_inv[5];
- }
- }
-
- boltz = force->boltz;
- nktv2p = force->nktv2p;
-
- if (force->kspace) kspace_flag = 1;
- else kspace_flag = 0;
-
- if (strcmp(update->integrate_style,"respa") == 0) {
- nlevels_respa = ((Respa *) update->integrate)->nlevels;
- step_respa = ((Respa *) update->integrate)->step;
- dto = 0.5*step_respa[0];
- }
-
- // detect if any rigid fixes exist so rigid bodies move when box is remapped
- // rfix[] = indices to each fix rigid
-
- delete [] rfix;
- nrigid = 0;
- rfix = NULL;
-
- for (int i = 0; i < modify->nfix; i++)
- if (modify->fix[i]->rigid_flag) nrigid++;
- if (nrigid) {
- rfix = new int[nrigid];
- nrigid = 0;
- for (int i = 0; i < modify->nfix; i++)
- if (modify->fix[i]->rigid_flag) rfix[nrigid++] = i;
- }
- triggerneighsq= cuda->shared_data.atom.triggerneighsq;
- cuda->neighbor_decide_by_integrator=1;
- Cuda_FixNHCuda_Init(&cuda->shared_data,dtv,dtf);
-
-}
-
-/* ----------------------------------------------------------------------
- compute T,P before integrator starts
-------------------------------------------------------------------------- */
-
-void FixNHCuda::setup(int vflag)
-{
- // initialize some quantities that were not available earlier
-
- //if (mtk_flag) mtk_factor = 1.0 + 1.0/atom->natoms;
- //else mtk_factor = 1.0;
- tdof = temperature->dof;
-
- // t_target is used by compute_scalar(), even for NPH
-
- if (tstat_flag) t_target = t_start;
- else if (pstat_flag) {
-
- // t0 = initial value for piston mass and energy conservation
- // cannot be done in init() b/c temperature cannot be called there
- // is b/c Modify::init() inits computes after fixes due to dof dependence
- // guesstimate a unit-dependent t0 if actual T = 0.0
- // if it was read in from a restart file, leave it be
-
- if (t0 == 0.0) {
- t0 = temperature->compute_scalar();
- if (t0 == 0.0) {
- if (strcmp(update->unit_style,"lj") == 0) t0 = 1.0;
- else t0 = 300.0;
- }
- }
- t_target = t0;
- }
-
- if (pstat_flag) compute_press_target();
-
- t_current = temperature->compute_scalar();
- if (pstat_flag) {
- if (pstyle == ISO) double tmp = pressure->compute_scalar();
- else pressure->compute_vector();
- couple();
- pressure->addstep(update->ntimestep+1);
- }
-
- // initial forces on thermostat variables
-
- if (tstat_flag) {
- eta_mass[0] = tdof * boltz * t_target / (t_freq*t_freq);
- for (int ich = 1; ich < mtchain; ich++)
- eta_mass[ich] = boltz * t_target / (t_freq*t_freq);
- for (int ich = 1; ich < mtchain; ich++) {
- eta_dotdot[ich] = (eta_mass[ich-1]*eta_dot[ich-1]*eta_dot[ich-1] -
- boltz*t_target) / eta_mass[ich];
- }
- }
-
- if (pstat_flag) {
- double kt = boltz * t_target;
- double nkt = atom->natoms * kt;
-
- for (int i = 0; i < 3; i++)
- if (p_flag[i])
- omega_mass[i] = nkt/(p_freq[i]*p_freq[i]);
-
- if (pstyle == TRICLINIC) {
- for (int i = 3; i < 6; i++)
- if (p_flag[i]) omega_mass[i] = nkt/(p_freq[i]*p_freq[i]);
- }
-
- // initial forces on barostat thermostat variables
-
- if (mpchain) {
- etap_mass[0] = boltz * t_target / (p_freq_max*p_freq_max);
- for (int ich = 1; ich < mpchain; ich++)
- etap_mass[ich] = boltz * t_target / (p_freq_max*p_freq_max);
- for (int ich = 1; ich < mpchain; ich++)
- etap_dotdot[ich] =
- (etap_mass[ich-1]*etap_dot[ich-1]*etap_dot[ich-1] -
- boltz*t_target) / etap_mass[ich];
- }
-
- // compute appropriately coupled elements of mvv_current
-
- //if (mtk_flag) couple_ke();
- }
-}
-
-/* ----------------------------------------------------------------------
- 1st half of Verlet update
-------------------------------------------------------------------------- */
-
-void FixNHCuda::initial_integrate(int vflag)
-{
- if(!temperature->cudable) cuda->downloadAll();
-
- if(triggerneighsq!=cuda->shared_data.atom.triggerneighsq)
- {
- triggerneighsq= cuda->shared_data.atom.triggerneighsq;
- Cuda_FixNHCuda_Init(&cuda->shared_data,dtv,dtf);
- }
-
- // update eta_press_dot
-
- if (pstat_flag && mpchain) nhc_press_integrate();
-
- // update eta_dot
-
- if (tstat_flag) {
- double delta = update->ntimestep - update->beginstep;
- delta /= update->endstep - update->beginstep;
- t_target = t_start + delta * (t_stop-t_start);
- eta_mass[0] = tdof * boltz * t_target / (t_freq*t_freq);
- for (int ich = 1; ich < mtchain; ich++)
- eta_mass[ich] = boltz * t_target / (t_freq*t_freq);
- nhc_temp_integrate();
- }
-
- // need to recompute pressure to account for change in KE
- // t_current is up-to-date, but compute_temperature is not
- // compute appropriately coupled elements of mvv_current
-
- if (pstat_flag) {
- if (pstyle == ISO) {
- temperature->compute_scalar();
- double tmp = pressure->compute_scalar();
- } else {
- temperature->compute_vector();
- pressure->compute_vector();
- }
- couple();
- pressure->addstep(update->ntimestep+1);
- //if (mtk_flag) couple_ke();
- }
-
- if(which==NOBIAS)
- {
- if (pstat_flag) {
- compute_press_target();
- nh_omega_dot();
- factor[0] = exp(-dt4*(omega_dot[0]+mtk_term2));
- factor[1] = exp(-dt4*(omega_dot[1]+mtk_term2));
- factor[2] = exp(-dt4*(omega_dot[2]+mtk_term2));
- Cuda_FixNHCuda_nh_v_press_and_nve_v_NoBias(&cuda->shared_data, groupbit, factor,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal,(pstyle == TRICLINIC)?1:0);
- }
- else
- Cuda_FixNHCuda_nve_v(&cuda->shared_data,groupbit,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal);
- }
- else if(which==BIAS)
- {
- if(pstat_flag)
- {
- compute_press_target();
- nh_omega_dot();
- factor[0] = exp(-dt4*(omega_dot[0]+mtk_term2));
- factor[1] = exp(-dt4*(omega_dot[1]+mtk_term2));
- factor[2] = exp(-dt4*(omega_dot[2]+mtk_term2));
- if(!temperature->cudable)
- {
- nh_v_press();
- cuda->cu_v->upload();
- }
- else
- {
- int groupbit_org=temperature->groupbit;
- temperature->groupbit=groupbit;
- temperature->remove_bias_all();
- Cuda_FixNHCuda_nh_v_press(&cuda->shared_data, groupbit, factor,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal,(pstyle == TRICLINIC)?1:0);
- temperature->restore_bias_all();
- temperature->groupbit=groupbit_org;
- }
- }
- Cuda_FixNHCuda_nve_v(&cuda->shared_data,groupbit,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal);
- }
-
- // remap simulation box by 1/2 step
-
- if (pstat_flag) remap();
-
- Cuda_FixNHCuda_nve_x(&cuda->shared_data,groupbit,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal);
-
- // remap simulation box by 1/2 step
- // redo KSpace coeffs since volume has changed
-
- if (pstat_flag) {
- remap();
- if (kspace_flag) force->kspace->setup();
- }
-}
-
-/* ----------------------------------------------------------------------
- 2nd half of Verlet update
-------------------------------------------------------------------------- */
-
-void FixNHCuda::final_integrate()
-{
- if(!temperature->cudable) cuda->downloadAll();
-
- if(which==NOBIAS)
- {
- if(pstat_flag)
- {
- factor[0] = exp(-dt4*(omega_dot[0]+mtk_term2));
- factor[1] = exp(-dt4*(omega_dot[1]+mtk_term2));
- factor[2] = exp(-dt4*(omega_dot[2]+mtk_term2));
-
- Cuda_FixNHCuda_nve_v_and_nh_v_press_NoBias(&cuda->shared_data, groupbit, factor,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal,(pstyle == TRICLINIC)?1:0);
- }
- else
- Cuda_FixNHCuda_nve_v(&cuda->shared_data,groupbit,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal);
- }
- else if(which==BIAS)
- {
- Cuda_FixNHCuda_nve_v(&cuda->shared_data,groupbit,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal);
-
- if(pstat_flag)
- {
- factor[0] = exp(-dt4*(omega_dot[0]+mtk_term2));
- factor[1] = exp(-dt4*(omega_dot[1]+mtk_term2));
- factor[2] = exp(-dt4*(omega_dot[2]+mtk_term2));
- if(!temperature->cudable)
- {
- cuda->cu_v->download();
- nh_v_press();
- cuda->cu_v->upload();
- }
- else
- {
- int groupbit_org=temperature->groupbit;
- temperature->groupbit=groupbit;
- temperature->remove_bias_all();
- Cuda_FixNHCuda_nh_v_press(&cuda->shared_data, groupbit, factor,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal,(pstyle == TRICLINIC)?1:0);
- temperature->restore_bias_all();
- temperature->groupbit=groupbit_org;
- }
- }
- }
- // compute new T,P
- // compute appropriately coupled elements of mvv_current
-
- if(!temperature->cudable) cuda->cu_v->download();
- t_current = temperature->compute_scalar();
- if (pstat_flag) {
- if (pstyle == ISO) double tmp = pressure->compute_scalar();
- else pressure->compute_vector();
- couple();
- pressure->addstep(update->ntimestep+1);
- }
-
- if (pstat_flag) nh_omega_dot();
-
- // update eta_dot
- // update eta_press_dot
-
- if (tstat_flag) nhc_temp_integrate();
- if (pstat_flag && mpchain) nhc_press_integrate();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNHCuda::initial_integrate_respa(int vflag, int ilevel, int iloop)
-{
- // set timesteps by level
-
- dtv = step_respa[ilevel];
- dtf = 0.5 * step_respa[ilevel] * force->ftm2v;
- dthalf = 0.5 * step_respa[ilevel];
-
- // outermost level - update eta_dot and omega_dot, apply to v, remap box
- // all other levels - NVE update of v
- // x,v updates only performed for atoms in group
-
- if (ilevel == nlevels_respa-1) {
-
- // update eta_press_dot
-
- if (pstat_flag && mpchain) nhc_press_integrate();
-
- // update eta_dot
-
- if (tstat_flag) {
- double delta = update->ntimestep - update->beginstep;
- delta /= update->endstep - update->beginstep;
- t_target = t_start + delta * (t_stop-t_start);
- eta_mass[0] = tdof * boltz * t_target / (t_freq*t_freq);
- for (int ich = 1; ich < mtchain; ich++)
- eta_mass[ich] = boltz * t_target / (t_freq*t_freq);
- nhc_temp_integrate();
- }
-
- // recompute pressure to account for change in KE
- // t_current is up-to-date, but compute_temperature is not
- // compute appropriately coupled elements of mvv_current
-
- if (pstat_flag) {
- if (pstyle == ISO) {
- temperature->compute_scalar();
- double tmp = pressure->compute_scalar();
- } else {
- temperature->compute_vector();
- pressure->compute_vector();
- }
- couple();
- pressure->addstep(update->ntimestep+1);
- if (mtk_flag) couple_ke();
- }
-
- if (pstat_flag) {
- compute_press_target();
- nh_omega_dot();
- nh_v_press();
- }
-
- nve_v();
-
- } else nve_v();
-
- // innermost level - also update x only for atoms in group
- // if barostat, perform 1/2 step remap before and after
-
- if (ilevel == 0) {
- if (pstat_flag) remap();
- nve_x();
- if (pstat_flag) remap();
- }
-
- // if barostat, redo KSpace coeffs at outermost level,
- // since volume has changed
-
- if (ilevel == nlevels_respa-1 && kspace_flag && pstat_flag)
- force->kspace->setup();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNHCuda::final_integrate_respa(int ilevel, int iloop)
-{
- // set timesteps by level
-
- dtf = 0.5 * step_respa[ilevel] * force->ftm2v;
- dthalf = 0.5 * step_respa[ilevel];
-
- // outermost level - update eta_dot and omega_dot, apply via final_integrate
- // all other levels - NVE update of v
-
- if (ilevel == nlevels_respa-1) final_integrate();
- else nve_v();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNHCuda::couple()
-{
- double *tensor = pressure->vector;
-
- if (pstyle == ISO)
- p_current[0] = p_current[1] = p_current[2] = pressure->scalar;
- else if (pcouple == XYZ) {
- double ave = 1.0/3.0 * (tensor[0] + tensor[1] + tensor[2]);
- p_current[0] = p_current[1] = p_current[2] = ave;
- } else if (pcouple == XY) {
- double ave = 0.5 * (tensor[0] + tensor[1]);
- p_current[0] = p_current[1] = ave;
- p_current[2] = tensor[2];
- } else if (pcouple == YZ) {
- double ave = 0.5 * (tensor[1] + tensor[2]);
- p_current[1] = p_current[2] = ave;
- p_current[0] = tensor[0];
- } else if (pcouple == XZ) {
- double ave = 0.5 * (tensor[0] + tensor[2]);
- p_current[0] = p_current[2] = ave;
- p_current[1] = tensor[1];
- } else {
- p_current[0] = tensor[0];
- p_current[1] = tensor[1];
- p_current[2] = tensor[2];
- }
-
- // switch order from xy-xz-yz to Voigt
-
- if (pstyle == TRICLINIC) {
- p_current[3] = tensor[5];
- p_current[4] = tensor[4];
- p_current[5] = tensor[3];
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNHCuda::couple_ke()
-{
- double *tensor = temperature->vector;
- if (pstyle == ISO)
- mvv_current[0] = mvv_current[1] = mvv_current[2] =
- tdof * boltz * t_current/dimension;
- else if (pcouple == XYZ) {
- double ave = 1.0/3.0 * (tensor[0] + tensor[1] + tensor[2]);
- mvv_current[0] = mvv_current[1] = mvv_current[2] = ave;
- } else if (pcouple == XY) {
- double ave = 0.5 * (tensor[0] + tensor[1]);
- mvv_current[0] = mvv_current[1] = ave;
- mvv_current[2] = tensor[2];
- } else if (pcouple == YZ) {
- double ave = 0.5 * (tensor[1] + tensor[2]);
- mvv_current[1] = mvv_current[2] = ave;
- mvv_current[0] = tensor[0];
- } else if (pcouple == XZ) {
- double ave = 0.5 * (tensor[0] + tensor[2]);
- mvv_current[0] = mvv_current[2] = ave;
- mvv_current[1] = tensor[1];
- } else {
- mvv_current[0] = tensor[0];
- mvv_current[1] = tensor[1];
- mvv_current[2] = tensor[2];
- }
-}
-
-/* ----------------------------------------------------------------------
- change box size
- remap all atoms or fix group atoms depending on allremap flag
- if rigid bodies exist, scale rigid body centers-of-mass
-------------------------------------------------------------------------- */
-
-void FixNHCuda::remap()
-{
- int i;
- double oldlo,oldhi,ctr;
-
- double **x = atom->x;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
- double *h = domain->h;
-
- // omega is not used, except for book-keeping
-
- for (int i = 0; i < 6; i++) omega[i] += dto*omega_dot[i];
-
- // convert pertinent atoms and rigid bodies to lamda coords
- if (allremap) domain->x2lamda(nlocal);
- else {
- for (i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- domain->x2lamda(x[i],x[i]);
- }
-
- if (nrigid)
- for (i = 0; i < nrigid; i++)
- modify->fix[rfix[i]]->deform(0);
-
- // reset global and local box to new size/shape
-
- // This operation corresponds to applying the
- // translate and scale operations
- // corresponding to the solution of the following ODE:
- //
- // h_dot = omega_dot * h
- //
- // where h_dot, omega_dot and h are all upper-triangular
- // 3x3 tensors. In Voigt notation, the elements of the
- // RHS product tensor are:
- // h_dot = [0*0, 1*1, 2*2, 1*3+3*2, 0*4+5*3+4*2, 0*5+5*1]
- //
- // Ordering of operations preserves time symmetry.
-
- double dto2 = dto/2.0;
- double dto4 = dto/4.0;
- double dto8 = dto/8.0;
-
- if (pstyle == TRICLINIC) {
-
- h[4] *= exp(dto8*omega_dot[0]);
- h[4] += dto4*(omega_dot[5]*h[3]+omega_dot[4]*h[2]);
- h[4] *= exp(dto8*omega_dot[0]);
-
- h[3] *= exp(dto4*omega_dot[1]);
- h[3] += dto2*(omega_dot[3]*h[2]);
- h[3] *= exp(dto4*omega_dot[1]);
-
- h[5] *= exp(dto4*omega_dot[0]);
- h[5] += dto2*(omega_dot[5]*h[1]);
- h[5] *= exp(dto4*omega_dot[0]);
-
- h[4] *= exp(dto8*omega_dot[0]);
- h[4] += dto4*(omega_dot[5]*h[3]+omega_dot[4]*h[2]);
- h[4] *= exp(dto8*omega_dot[0]);
-
- }
-
- for (i = 0; i < 3; i++) {
- if (p_flag[i]) {
- oldlo = domain->boxlo[i];
- oldhi = domain->boxhi[i];
- ctr = 0.5 * (oldlo + oldhi);
- domain->boxlo[i] = (oldlo-ctr)*exp(dto*omega_dot[i]) + ctr;
- domain->boxhi[i] = (oldhi-ctr)*exp(dto*omega_dot[i]) + ctr;
- }
- }
-
- if (pstyle == TRICLINIC) {
-
- h[4] *= exp(dto8*omega_dot[0]);
- h[4] += dto4*(omega_dot[5]*h[3]+omega_dot[4]*h[2]);
- h[4] *= exp(dto8*omega_dot[0]);
-
- h[3] *= exp(dto4*omega_dot[1]);
- h[3] += dto2*(omega_dot[3]*h[2]);
- h[3] *= exp(dto4*omega_dot[1]);
-
- h[5] *= exp(dto4*omega_dot[0]);
- h[5] += dto2*(omega_dot[5]*h[1]);
- h[5] *= exp(dto4*omega_dot[0]);
-
- h[4] *= exp(dto8*omega_dot[0]);
- h[4] += dto4*(omega_dot[5]*h[3]+omega_dot[4]*h[2]);
- h[4] *= exp(dto8*omega_dot[0]);
-
- domain->yz = h[3];
- domain->xz = h[4];
- domain->xy = h[5];
-
- if (domain->yz < -0.5*domain->yprd || domain->yz > 0.5*domain->yprd ||
- domain->xz < -0.5*domain->xprd || domain->xz > 0.5*domain->xprd ||
- domain->xy < -0.5*domain->xprd || domain->xy > 0.5*domain->xprd)
- error->all(FLERR,"Fix npt/nph has tilted box too far - "
- "box flips are not yet implemented");
- }
-
- domain->set_global_box();
- domain->set_local_box();
-
- // convert pertinent atoms and rigid bodies back to box coords
-
- if (allremap) domain->lamda2x(nlocal);
- else {
- for (i = 0; i < nlocal; i++)
- if (mask[i] & groupbit)
- domain->lamda2x(x[i],x[i]);
- }
-
- if (nrigid)
- for (i = 0; i < nrigid; i++)
- modify->fix[rfix[i]]->deform(1);
-}
-
-/* ----------------------------------------------------------------------
- pack entire state of Fix into one write
-------------------------------------------------------------------------- */
-
-void FixNHCuda::write_restart(FILE *fp)
-{
- int nsize = 2;
- if (tstat_flag) nsize += 1 + 2*mtchain;
- if (pstat_flag) {
- nsize += 16 + 2*mpchain;
- if (deviatoric_flag) nsize += 6;
- }
-
- double* list = (double *) memory->smalloc(nsize*sizeof(double),"nh:list");
-
- int n = 0;
-
- list[n++] = tstat_flag;
- if (tstat_flag) {
- list[n++] = mtchain;
- for (int ich = 0; ich < mtchain; ich++)
- list[n++] = eta[ich];
- for (int ich = 0; ich < mtchain; ich++)
- list[n++] = eta_dot[ich];
- }
-
- list[n++] = pstat_flag;
- if (pstat_flag) {
- list[n++] = omega[0];
- list[n++] = omega[1];
- list[n++] = omega[2];
- list[n++] = omega[3];
- list[n++] = omega[4];
- list[n++] = omega[5];
- list[n++] = omega_dot[0];
- list[n++] = omega_dot[1];
- list[n++] = omega_dot[2];
- list[n++] = omega_dot[3];
- list[n++] = omega_dot[4];
- list[n++] = omega_dot[5];
- list[n++] = vol0;
- list[n++] = t0;
- list[n++] = mpchain;
- if (mpchain) {
- for (int ich = 0; ich < mpchain; ich++)
- list[n++] = etap[ich];
- for (int ich = 0; ich < mpchain; ich++)
- list[n++] = etap_dot[ich];
- }
-
- list[n++] = deviatoric_flag;
- if (deviatoric_flag) {
- list[n++] = h0_inv[0];
- list[n++] = h0_inv[1];
- list[n++] = h0_inv[2];
- list[n++] = h0_inv[3];
- list[n++] = h0_inv[4];
- list[n++] = h0_inv[5];
- }
- }
-
- if (comm->me == 0) {
- int size = nsize * sizeof(double);
- fwrite(&size,sizeof(int),1,fp);
- fwrite(list,sizeof(double),nsize,fp);
- }
-
- memory->sfree(list);
-}
-
-/* ----------------------------------------------------------------------
- use state info from restart file to restart the Fix
-------------------------------------------------------------------------- */
-
-void FixNHCuda::restart(char *buf)
-{
- int n = 0;
- double *list = (double *) buf;
- int flag = static_cast<int> (list[n++]);
- if (flag) {
- int m = static_cast<int> (list[n++]);
- if (tstat_flag && m == mtchain) {
- for (int ich = 0; ich < mtchain; ich++)
- eta[ich] = list[n++];
- for (int ich = 0; ich < mtchain; ich++)
- eta_dot[ich] = list[n++];
- } else n += 2*m;
- }
- flag = static_cast<int> (list[n++]);
- if (flag) {
- omega[0] = list[n++];
- omega[1] = list[n++];
- omega[2] = list[n++];
- omega[3] = list[n++];
- omega[4] = list[n++];
- omega[5] = list[n++];
- omega_dot[0] = list[n++];
- omega_dot[1] = list[n++];
- omega_dot[2] = list[n++];
- omega_dot[3] = list[n++];
- omega_dot[4] = list[n++];
- omega_dot[5] = list[n++];
- vol0 = list[n++];
- t0 = list[n++];
- int m = static_cast<int> (list[n++]);
- if (pstat_flag && m == mpchain) {
- for (int ich = 0; ich < mpchain; ich++)
- etap[ich] = list[n++];
- for (int ich = 0; ich < mpchain; ich++)
- etap_dot[ich] = list[n++];
- } else n+=2*m;
- flag = static_cast<int> (list[n++]);
- if (flag) {
- h0_inv[0] = list[n++];
- h0_inv[1] = list[n++];
- h0_inv[2] = list[n++];
- h0_inv[3] = list[n++];
- h0_inv[4] = list[n++];
- h0_inv[5] = list[n++];
- }
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixNHCuda::modify_param(int narg, char **arg)
-{
- if (strcmp(arg[0],"temp") == 0) {
- if (narg < 2) error->all(FLERR,"Illegal fix_modify command");
- if (tflag) {
- modify->delete_compute(id_temp);
- tflag = 0;
- }
- delete [] id_temp;
- int n = strlen(arg[1]) + 1;
- id_temp = new char[n];
- strcpy(id_temp,arg[1]);
-
- int icompute = modify->find_compute(arg[1]);
- if (icompute < 0) error->all(FLERR,"Could not find fix_modify temperature ID");
- temperature = modify->compute[icompute];
-
- if (temperature->tempflag == 0)
- error->all(FLERR,"Fix_modify temperature ID does not compute temperature");
- if (temperature->igroup != 0 && comm->me == 0)
- error->warning(FLERR,"Temperature for fix modify is not for group all");
-
- // reset id_temp of pressure to new temperature ID
-
- if (pstat_flag) {
- icompute = modify->find_compute(id_press);
- if (icompute < 0)
- error->all(FLERR,"Pressure ID for fix modify does not exist");
- modify->compute[icompute]->reset_extra_compute_fix(id_temp);
- }
-
- return 2;
-
- } else if (strcmp(arg[0],"press") == 0) {
- if (narg < 2) error->all(FLERR,"Illegal fix_modify command");
- if (!pstat_flag) error->all(FLERR,"Illegal fix_modify command");
- if (pflag) {
- modify->delete_compute(id_press);
- pflag = 0;
- }
- delete [] id_press;
- int n = strlen(arg[1]) + 1;
- id_press = new char[n];
- strcpy(id_press,arg[1]);
-
- int icompute = modify->find_compute(arg[1]);
- if (icompute < 0) error->all(FLERR,"Could not find fix_modify pressure ID");
- pressure = modify->compute[icompute];
-
- if (pressure->pressflag == 0)
- error->all(FLERR,"Fix_modify pressure ID does not compute pressure");
- return 2;
- }
-
- return 0;
-}
-
-/* ---------------------------------------------------------------------- */
-
-double FixNHCuda::compute_scalar()
-{
- int i;
- double volume;
- double energy;
- double kt = boltz * t_target;
- double lkt = tdof * kt;
- double lkt_press = kt;
- int ich;
- if (dimension == 3) volume = domain->xprd * domain->yprd * domain->zprd;
- else volume = domain->xprd * domain->yprd;
-
- energy = 0.0;
-
- // thermostat chain energy is equivalent to Eq. (2) in
- // Martyna, Tuckerman, Tobias, Klein, Mol Phys, 87, 1117
- // Sum(0.5*p_eta_k^2/Q_k,k=1,M) + L*k*T*eta_1 + Sum(k*T*eta_k,k=2,M),
- // where L = tdof
- // M = mtchain
- // p_eta_k = Q_k*eta_dot[k-1]
- // Q_1 = L*k*T/t_freq^2
- // Q_k = k*T/t_freq^2, k > 1
-
- if (tstat_flag) {
- energy += lkt * eta[0] + 0.5*eta_mass[0]*eta_dot[0]*eta_dot[0];
- for (ich = 1; ich < mtchain; ich++)
- energy += kt * eta[ich] + 0.5*eta_mass[ich]*eta_dot[ich]*eta_dot[ich];
- }
-
- // barostat energy is equivalent to Eq. (8) in
- // Martyna, Tuckerman, Tobias, Klein, Mol Phys, 87, 1117
- // Sum(0.5*p_omega^2/W + P*V),
- // where N = natoms
- // p_omega = W*omega_dot
- // W = N*k*T/p_freq^2
- // sum is over barostatted dimensions
-
- if (pstat_flag) {
- for (i = 0; i < 3; i++)
- if (p_flag[i])
- energy += 0.5*omega_dot[i]*omega_dot[i]*omega_mass[i] +
- p_hydro*(volume-vol0) / (pdim*nktv2p);
-
- if (pstyle == TRICLINIC) {
- for (i = 3; i < 6; i++)
- if (p_flag[i])
- energy += 0.5*omega_dot[i]*omega_dot[i]*omega_mass[i];
- }
-
- // extra contributions from thermostat chain for barostat
-
- if (mpchain) {
- energy += lkt_press * etap[0] + 0.5*etap_mass[0]*etap_dot[0]*etap_dot[0];
- for (ich = 1; ich < mpchain; ich++)
- energy += kt * etap[ich] +
- 0.5*etap_mass[ich]*etap_dot[ich]*etap_dot[ich];
- }
-
- // extra contribution from strain energy
-
- if (deviatoric_flag) energy += compute_strain_energy();
- }
-
- return energy;
-}
-
-/* ----------------------------------------------------------------------
- return a single element of the following vectors, in this order:
- eta[tchain], eta_dot[tchain], omega[ndof], omega_dot[ndof]
- etap[pchain], etap_dot[pchain], PE_eta[tchain], KE_eta_dot[tchain]
- PE_omega[ndof], KE_omega_dot[ndof], PE_etap[pchain], KE_etap_dot[pchain]
- PE_strain[1]
- if no thermostat exists, related quantities are omitted from the list
- if no barostat exists, related quantities are omitted from the list
- ndof = 1,3,6 degrees of freedom for pstyle = ISO,ANISO,TRI
-------------------------------------------------------------------------- */
-
-double FixNHCuda::compute_vector(int n)
-{
- int ilen;
-
- if (tstat_flag) {
- ilen = mtchain;
- if (n < ilen) return eta[n];
- n -= ilen;
- ilen = mtchain;
- if (n < ilen) return eta_dot[n];
- n -= ilen;
- }
-
- if (pstat_flag) {
- if (pstyle == ISO) {
- ilen = 1;
- if (n < ilen) return omega[n];
- n -= ilen;
- } else if (pstyle == ANISO) {
- ilen = 3;
- if (n < ilen) return omega[n];
- n -= ilen;
- } else {
- ilen = 6;
- if (n < ilen) return omega[n];
- n -= ilen;
- }
-
- if (pstyle == ISO) {
- ilen = 1;
- if (n < ilen) return omega_dot[n];
- n -= ilen;
- } else if (pstyle == ANISO) {
- ilen = 3;
- if (n < ilen) return omega_dot[n];
- n -= ilen;
- } else {
- ilen = 6;
- if (n < ilen) return omega_dot[n];
- n -= ilen;
- }
-
- if (mpchain) {
- ilen = mpchain;
- if (n < ilen) return etap[n];
- n -= ilen;
- ilen = mpchain;
- if (n < ilen) return etap_dot[n];
- n -= ilen;
- }
- }
-
- double volume;
- double kt = boltz * t_target;
- double lkt = tdof * kt;
- double lkt_press = kt;
- int ich;
- if (dimension == 3) volume = domain->xprd * domain->yprd * domain->zprd;
- else volume = domain->xprd * domain->yprd;
-
- if (tstat_flag) {
- ilen = mtchain;
- if (n < ilen) {
- ich = n;
- if (ich == 0)
- return lkt * eta[0];
- else
- return kt * eta[ich];
- }
- n -= ilen;
- ilen = mtchain;
- if (n < ilen) {
- ich = n;
- if (ich == 0)
- return 0.5*eta_mass[0]*eta_dot[0]*eta_dot[0];
- else
- return 0.5*eta_mass[ich]*eta_dot[ich]*eta_dot[ich];
- }
- n -= ilen;
- }
-
- if (pstat_flag) {
- if (pstyle == ISO) {
- ilen = 1;
- if (n < ilen)
- return p_hydro*(volume-vol0) / nktv2p;
- n -= ilen;
- } else if (pstyle == ANISO) {
- ilen = 3;
- if (n < ilen)
- if (p_flag[n])
- return p_hydro*(volume-vol0) / (pdim*nktv2p);
- else
- return 0.0;
- n -= ilen;
- } else {
- ilen = 6;
- if (n < ilen)
- if (n > 2) return 0.0;
- else if (p_flag[n])
- return p_hydro*(volume-vol0) / (pdim*nktv2p);
- else
- return 0.0;
- n -= ilen;
- }
-
- if (pstyle == ISO) {
- ilen = 1;
- if (n < ilen)
- return pdim*0.5*omega_dot[n]*omega_dot[n]*omega_mass[n];
- n -= ilen;
- } else if (pstyle == ANISO) {
- ilen = 3;
- if (n < ilen)
- if (p_flag[n])
- return 0.5*omega_dot[n]*omega_dot[n]*omega_mass[n];
- else return 0.0;
- n -= ilen;
- } else {
- ilen = 6;
- if (n < ilen)
- if (p_flag[n])
- return 0.5*omega_dot[n]*omega_dot[n]*omega_mass[n];
- else return 0.0;
- n -= ilen;
- }
-
- if (mpchain) {
- ilen = mpchain;
- if (n < ilen) {
- ich = n;
- if (ich == 0) return lkt_press * etap[0];
- else return kt * etap[ich];
- }
- n -= ilen;
- ilen = mpchain;
- if (n < ilen) {
- ich = n;
- if (ich == 0)
- return 0.5*etap_mass[0]*etap_dot[0]*etap_dot[0];
- else
- return 0.5*etap_mass[ich]*etap_dot[ich]*etap_dot[ich];
- }
- n -= ilen;
- }
-
- if (deviatoric_flag) {
- ilen = 1;
- if (n < ilen)
- return compute_strain_energy();
- n -= ilen;
- }
- }
-
- return 0.0;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNHCuda::reset_dt()
-{
- dtv = update->dt;
- dtf = 0.5 * update->dt * force->ftm2v;
- dthalf = 0.5 * update->dt;
- dt4 = 0.25 * update->dt;
- dt8 = 0.125 * update->dt;
- dto = dthalf;
-
- // If using respa, then remap is performed in innermost level
-
- if (strcmp(update->integrate_style,"respa") == 0)
- dto = 0.5*step_respa[0];
-
- p_freq_max = 0.0;
- if (pstat_flag) {
- p_freq_max = MAX(p_freq[0],p_freq[1]);
- p_freq_max = MAX(p_freq_max,p_freq[2]);
- if (pstyle == TRICLINIC) {
- p_freq_max = MAX(p_freq_max,p_freq[3]);
- p_freq_max = MAX(p_freq_max,p_freq[4]);
- p_freq_max = MAX(p_freq_max,p_freq[5]);
- }
- pdrag_factor = 1.0 - (update->dt * p_freq_max * drag / nc_pchain);
- }
-
- if (tstat_flag)
- tdrag_factor = 1.0 - (update->dt * t_freq * drag / nc_tchain);
-}
-
-/* ----------------------------------------------------------------------
- perform half-step update of chain thermostat variables
-------------------------------------------------------------------------- */
-
-void FixNHCuda::nhc_temp_integrate()
-{
- int ich;
- double expfac;
-
- double lkt = tdof * boltz * t_target;
- double kecurrent = tdof * boltz * t_current;
- eta_dotdot[0] = (kecurrent - lkt)/eta_mass[0];
-
- double ncfac = 1.0/nc_tchain;
- for (int iloop = 0; iloop < nc_tchain; iloop++) {
-
- for (ich = mtchain-1; ich > 0; ich--) {
- expfac = exp(-ncfac*dt8*eta_dot[ich+1]);
- eta_dot[ich] *= expfac;
- eta_dot[ich] += eta_dotdot[ich] * ncfac*dt4;
- eta_dot[ich] *= tdrag_factor;
- eta_dot[ich] *= expfac;
- }
-
- expfac = exp(-ncfac*dt8*eta_dot[1]);
- eta_dot[0] *= expfac;
- eta_dot[0] += eta_dotdot[0] * ncfac*dt4;
- eta_dot[0] *= tdrag_factor;
- eta_dot[0] *= expfac;
-
- factor_eta = exp(-ncfac*dthalf*eta_dot[0]);
- if(which==NOBIAS)
- Cuda_FixNHCuda_nh_v_temp(&cuda->shared_data,groupbit,factor_eta,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal);
- else if(which==BIAS)
- {
- if(!temperature->cudable)
- {
- cuda->downloadAll();
- nh_v_temp();
- cuda->cu_v->upload();
- }
- else
- {
- int groupbit_org=temperature->groupbit;
- temperature->groupbit=groupbit;
- temperature->remove_bias_all();
- Cuda_FixNHCuda_nh_v_temp(&cuda->shared_data,groupbit,factor_eta,(igroup == atom->firstgroup)?atom->nfirst:atom->nlocal);
- temperature->restore_bias_all();
- temperature->groupbit=groupbit_org;
- }
-
- }
- // rescale temperature due to velocity scaling
- // should not be necessary to explicitly recompute the temperature
-
- t_current *= factor_eta*factor_eta;
- kecurrent = tdof * boltz * t_current;
- eta_dotdot[0] = (kecurrent - lkt)/eta_mass[0];
-
- for (ich = 0; ich < mtchain; ich++)
- eta[ich] += ncfac*dthalf*eta_dot[ich];
-
- eta_dot[0] *= expfac;
- eta_dot[0] += eta_dotdot[0] * ncfac*dt4;
- eta_dot[0] *= expfac;
-
- for (ich = 1; ich < mtchain; ich++) {
- expfac = exp(-ncfac*dt8*eta_dot[ich+1]);
- eta_dot[ich] *= expfac;
- eta_dotdot[ich] = (eta_mass[ich-1]*eta_dot[ich-1]*eta_dot[ich-1]
- - boltz * t_target)/eta_mass[ich];
- eta_dot[ich] += eta_dotdot[ich] * ncfac*dt4;
- eta_dot[ich] *= expfac;
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- perform half-step update of chain thermostat variables for barostat
- scale barostat velocities
-------------------------------------------------------------------------- */
-
-void FixNHCuda::nhc_press_integrate()
-{
- int ich,i;
- double expfac,factor_etap,kecurrent;
- double kt = boltz * t_target;
- double lkt_press = kt;
-
- kecurrent = 0.0;
- for (i = 0; i < 3; i++)
- if (p_flag[i]) kecurrent += omega_mass[i]*omega_dot[i]*omega_dot[i];
-
- if (pstyle == TRICLINIC) {
- for (i = 3; i < 6; i++)
- if (p_flag[i]) kecurrent += omega_mass[i]*omega_dot[i]*omega_dot[i];
- }
-
- etap_dotdot[0] = (kecurrent - lkt_press)/etap_mass[0];
-
- double ncfac = 1.0/nc_pchain;
- for (int iloop = 0; iloop < nc_pchain; iloop++) {
-
- for (ich = mpchain-1; ich > 0; ich--) {
- expfac = exp(-ncfac*dt8*etap_dot[ich+1]);
- etap_dot[ich] *= expfac;
- etap_dot[ich] += etap_dotdot[ich] * ncfac*dt4;
- etap_dot[ich] *= pdrag_factor;
- etap_dot[ich] *= expfac;
- }
-
- expfac = exp(-ncfac*dt8*etap_dot[1]);
- etap_dot[0] *= expfac;
- etap_dot[0] += etap_dotdot[0] * ncfac*dt4;
- etap_dot[0] *= pdrag_factor;
- etap_dot[0] *= expfac;
-
- for (ich = 0; ich < mpchain; ich++)
- etap[ich] += ncfac*dthalf*etap_dot[ich];
-
- factor_etap = exp(-ncfac*dthalf*etap_dot[0]);
- for (i = 0; i < 3; i++)
- if (p_flag[i]) omega_dot[i] *= factor_etap;
-
- if (pstyle == TRICLINIC) {
- for (i = 3; i < 6; i++)
- if (p_flag[i]) omega_dot[i] *= factor_etap;
- }
-
- kecurrent = 0.0;
- for (i = 0; i < 3; i++)
- if (p_flag[i]) kecurrent += omega_mass[i]*omega_dot[i]*omega_dot[i];
-
- if (pstyle == TRICLINIC) {
- for (i = 3; i < 6; i++)
- if (p_flag[i]) kecurrent += omega_mass[i]*omega_dot[i]*omega_dot[i];
- }
-
- etap_dotdot[0] = (kecurrent - lkt_press)/etap_mass[0];
-
- etap_dot[0] *= expfac;
- etap_dot[0] += etap_dotdot[0] * ncfac*dt4;
- etap_dot[0] *= expfac;
-
- for (ich = 1; ich < mpchain; ich++) {
- expfac = exp(-ncfac*dt8*etap_dot[ich+1]);
- etap_dot[ich] *= expfac;
- etap_dotdot[ich] =
- (etap_mass[ich-1]*etap_dot[ich-1]*etap_dot[ich-1] - boltz*t_target) /
- etap_mass[ich];
- etap_dot[ich] += etap_dotdot[ich] * ncfac*dt4;
- etap_dot[ich] *= expfac;
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- perform half-step barostat scaling of velocities
------------------------------------------------------------------------*/
-
-void FixNHCuda::nh_v_press()
-{
- double factor[3];
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
- if (igroup == atom->firstgroup) nlocal = atom->nfirst;
-
- factor[0] = exp(-dt4*(omega_dot[0]+mtk_term2));
- factor[1] = exp(-dt4*(omega_dot[1]+mtk_term2));
- factor[2] = exp(-dt4*(omega_dot[2]+mtk_term2));
-
- if (which == NOBIAS) {
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- v[i][0] *= factor[0];
- v[i][1] *= factor[1];
- v[i][2] *= factor[2];
- if (pstyle == TRICLINIC) {
- v[i][0] += -dthalf*(v[i][1]*omega_dot[5] + v[i][2]*omega_dot[4]);
- v[i][1] += -dthalf*v[i][2]*omega_dot[3];
- }
- v[i][0] *= factor[0];
- v[i][1] *= factor[1];
- v[i][2] *= factor[2];
- }
- }
- } else if (which == BIAS) {
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- temperature->remove_bias(i,v[i]);
- v[i][0] *= factor[0];
- v[i][1] *= factor[1];
- v[i][2] *= factor[2];
- if (pstyle == TRICLINIC) {
- v[i][0] += -dthalf*(v[i][1]*omega_dot[5] + v[i][2]*omega_dot[4]);
- v[i][1] += -dthalf*v[i][2]*omega_dot[3];
- }
- v[i][0] *= factor[0];
- v[i][1] *= factor[1];
- v[i][2] *= factor[2];
- temperature->restore_bias(i,v[i]);
- }
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- perform half-step update of velocities
------------------------------------------------------------------------*/
-
-void FixNHCuda::nve_v()
-{
- double dtfm;
- double **v = atom->v;
- double **f = atom->f;
- double *rmass = atom->rmass;
- double *mass = atom->mass;
- int *type = atom->type;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
- if (igroup == atom->firstgroup) nlocal = atom->nfirst;
-
- if (rmass) {
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- dtfm = dtf / rmass[i];
- v[i][0] += dtfm*f[i][0];
- v[i][1] += dtfm*f[i][1];
- v[i][2] += dtfm*f[i][2];
- }
- }
- } else {
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- dtfm = dtf / mass[type[i]];
- v[i][0] += dtfm*f[i][0];
- v[i][1] += dtfm*f[i][1];
- v[i][2] += dtfm*f[i][2];
- }
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- perform full-step update of positions
------------------------------------------------------------------------*/
-
-void FixNHCuda::nve_x()
-{
- double **x = atom->x;
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
- if (igroup == atom->firstgroup) nlocal = atom->nfirst;
-
- // x update by full step only for atoms in group
-
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- x[i][0] += dtv * v[i][0];
- x[i][1] += dtv * v[i][1];
- x[i][2] += dtv * v[i][2];
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- perform half-step thermostat scaling of velocities
------------------------------------------------------------------------*/
-
-void FixNHCuda::nh_v_temp()
-{
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
- if (igroup == atom->firstgroup) nlocal = atom->nfirst;
-
- if (which == NOBIAS) {
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- v[i][0] *= factor_eta;
- v[i][1] *= factor_eta;
- v[i][2] *= factor_eta;
- }
- }
- } else if (which == BIAS) {
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- temperature->remove_bias(i,v[i]);
- v[i][0] *= factor_eta;
- v[i][1] *= factor_eta;
- v[i][2] *= factor_eta;
- temperature->restore_bias(i,v[i]);
- }
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- compute sigma tensor
- needed whenever p_target or h0_inv changes
------------------------------------------------------------------------*/
-
-void FixNHCuda::compute_sigma()
-{
- // if nreset_h0 > 0, reset vol0 and h0_inv
- // every nreset_h0 timesteps
-
- if (nreset_h0 > 0) {
- int delta = update->ntimestep - update->beginstep;
- if (delta % nreset_h0 == 0) {
- if (dimension == 3) vol0 = domain->xprd * domain->yprd * domain->zprd;
- else vol0 = domain->xprd * domain->yprd;
- h0_inv[0] = domain->h_inv[0];
- h0_inv[1] = domain->h_inv[1];
- h0_inv[2] = domain->h_inv[2];
- h0_inv[3] = domain->h_inv[3];
- h0_inv[4] = domain->h_inv[4];
- h0_inv[5] = domain->h_inv[5];
- }
- }
-
- // generate upper-triangular half of
- // sigma = vol0*h0inv*(p_target-p_hydro)*h0inv^t
- // units of sigma are are PV/L^2 e.g. atm.A
- //
- // [ 0 5 4 ] [ 0 5 4 ] [ 0 5 4 ] [ 0 - - ]
- // [ 5 1 3 ] = [ - 1 3 ] [ 5 1 3 ] [ 5 1 - ]
- // [ 4 3 2 ] [ - - 2 ] [ 4 3 2 ] [ 4 3 2 ]
-
- sigma[0] =
- vol0*(h0_inv[0]*((p_target[0]-p_hydro)*h0_inv[0] +
- p_target[5]*h0_inv[5]+p_target[4]*h0_inv[4]) +
- h0_inv[5]*(p_target[5]*h0_inv[0] +
- (p_target[1]-p_hydro)*h0_inv[5]+p_target[3]*h0_inv[4]) +
- h0_inv[4]*(p_target[4]*h0_inv[0]+p_target[3]*h0_inv[5] +
- (p_target[2]-p_hydro)*h0_inv[4]));
- sigma[1] =
- vol0*(h0_inv[1]*((p_target[1]-p_hydro)*h0_inv[1] +
- p_target[3]*h0_inv[3]) +
- h0_inv[3]*(p_target[3]*h0_inv[1] +
- (p_target[2]-p_hydro)*h0_inv[3]));
- sigma[2] =
- vol0*(h0_inv[2]*((p_target[2]-p_hydro)*h0_inv[2]));
- sigma[3] =
- vol0*(h0_inv[1]*(p_target[3]*h0_inv[2]) +
- h0_inv[3]*((p_target[2]-p_hydro)*h0_inv[2]));
- sigma[4] =
- vol0*(h0_inv[0]*(p_target[4]*h0_inv[2]) +
- h0_inv[5]*(p_target[3]*h0_inv[2]) +
- h0_inv[4]*((p_target[2]-p_hydro)*h0_inv[2]));
- sigma[5] =
- vol0*(h0_inv[0]*(p_target[5]*h0_inv[1]+p_target[4]*h0_inv[3]) +
- h0_inv[5]*((p_target[1]-p_hydro)*h0_inv[1]+p_target[3]*h0_inv[3]) +
- h0_inv[4]*(p_target[3]*h0_inv[1]+(p_target[2]-p_hydro)*h0_inv[3]));
-}
-
-/* ----------------------------------------------------------------------
- compute strain energy
------------------------------------------------------------------------*/
-
-double FixNHCuda::compute_strain_energy()
-{
- // compute strain energy = 0.5*Tr(sigma*h*h^t) in energy units
-
- double* h = domain->h;
- double d0,d1,d2;
-
- d0 =
- sigma[0]*(h[0]*h[0]+h[5]*h[5]+h[4]*h[4]) +
- sigma[5]*( h[1]*h[5]+h[3]*h[4]) +
- sigma[4]*( h[2]*h[4]);
- d1 =
- sigma[5]*( h[5]*h[1]+h[4]*h[3]) +
- sigma[1]*( h[1]*h[1]+h[3]*h[3]) +
- sigma[3]*( h[2]*h[3]);
- d2 =
- sigma[4]*( h[4]*h[2]) +
- sigma[3]*( h[3]*h[2]) +
- sigma[2]*( h[2]*h[2]);
-
- double energy = 0.5*(d0+d1+d2)/nktv2p;
- return energy;
-}
-
-/* ----------------------------------------------------------------------
- compute deviatoric barostat force = h*sigma*h^t
------------------------------------------------------------------------*/
-
-void FixNHCuda::compute_deviatoric()
-{
- // generate upper-triangular part of h*sigma*h^t
- // units of fdev are are PV, e.g. atm*A^3
- // [ 0 5 4 ] [ 0 5 4 ] [ 0 5 4 ] [ 0 - - ]
- // [ 5 1 3 ] = [ - 1 3 ] [ 5 1 3 ] [ 5 1 - ]
- // [ 4 3 2 ] [ - - 2 ] [ 4 3 2 ] [ 4 3 2 ]
-
- double* h = domain->h;
-
- fdev[0] =
- h[0]*(sigma[0]*h[0]+sigma[5]*h[5]+sigma[4]*h[4]) +
- h[5]*(sigma[5]*h[0]+sigma[1]*h[5]+sigma[3]*h[4]) +
- h[4]*(sigma[4]*h[0]+sigma[3]*h[5]+sigma[2]*h[4]);
- fdev[1] =
- h[1]*( sigma[1]*h[1]+sigma[3]*h[3]) +
- h[3]*( sigma[3]*h[1]+sigma[2]*h[3]);
- fdev[2] =
- h[2]*( sigma[2]*h[2]);
- fdev[3] =
- h[1]*( sigma[3]*h[2]) +
- h[3]*( sigma[2]*h[2]);
- fdev[4] =
- h[0]*( sigma[4]*h[2]) +
- h[5]*( sigma[3]*h[2]) +
- h[4]*( sigma[2]*h[2]);
- fdev[5] =
- h[0]*( sigma[5]*h[1]+sigma[4]*h[3]) +
- h[5]*( sigma[1]*h[1]+sigma[3]*h[3]) +
- h[4]*( sigma[3]*h[1]+sigma[2]*h[3]);
-}
-
-/* ----------------------------------------------------------------------
- compute hydrostatic target pressure
------------------------------------------------------------------------*/
-
-void FixNHCuda::compute_press_target()
-{
- double delta = update->ntimestep - update->beginstep;
- if (update->endstep > update->beginstep)
- delta /= update->endstep - update->beginstep;
- else delta = 0.0;
-
- p_hydro = 0.0;
- for (int i = 0; i < 3; i++)
- if (p_flag[i]) {
- p_target[i] = p_start[i] + delta * (p_stop[i]-p_start[i]);
- p_hydro += p_target[i];
- }
- p_hydro /= pdim;
-
- if (pstyle == TRICLINIC)
- for (int i = 3; i < 6; i++)
- p_target[i] = p_start[i] + delta * (p_stop[i]-p_start[i]);
-
- // if deviatoric, recompute sigma each time p_target changes
-
- if (deviatoric_flag) compute_sigma();
-}
-
-/* ----------------------------------------------------------------------
- update omega_dot, omega, dilation
------------------------------------------------------------------------*/
-
-void FixNHCuda::nh_omega_dot()
-{
- double f_omega,volume;
-
- if (dimension == 3) volume = domain->xprd*domain->yprd*domain->zprd;
- else volume = domain->xprd*domain->yprd;
-
- if (deviatoric_flag) compute_deviatoric();
-
- mtk_term1 = 0.0;
- if (mtk_flag)
- if (pstyle == ISO) {
- mtk_term1 = tdof * boltz * t_current;
- mtk_term1 /= pdim * atom->natoms;
- } else {
- double *mvv_current = temperature->vector;
- for (int i = 0; i < 3; i++)
- if (p_flag[i])
- mtk_term1 += mvv_current[i];
- mtk_term1 /= pdim * atom->natoms;
- }
-
- for (int i = 0; i < 3; i++)
- if (p_flag[i]) {
- f_omega = (p_current[i]-p_hydro)*volume /
- (omega_mass[i] * nktv2p) + mtk_term1 / omega_mass[i];
- if (deviatoric_flag) f_omega -= fdev[i]/(omega_mass[i] * nktv2p);
- omega_dot[i] += f_omega*dthalf;
- omega_dot[i] *= pdrag_factor;
- }
-
- mtk_term2 = 0.0;
- if (mtk_flag) {
- for (int i = 0; i < 3; i++)
- if (p_flag[i])
- mtk_term2 += omega_dot[i];
- mtk_term2 /= pdim * atom->natoms;
- }
-
- if (pstyle == TRICLINIC) {
- for (int i = 3; i < 6; i++) {
- if (p_flag[i]) {
- f_omega = p_current[i]*volume/(omega_mass[i] * nktv2p);
- if (deviatoric_flag)
- f_omega -= fdev[i]/(omega_mass[i] * nktv2p);
- omega_dot[i] += f_omega*dthalf;
- omega_dot[i] *= pdrag_factor;
- }
- }
- }
-}
diff --git a/src/USER-CUDA/fix_nh_cuda.h b/src/USER-CUDA/fix_nh_cuda.h
deleted file mode 100644
index 3cb97873c0..0000000000
--- a/src/USER-CUDA/fix_nh_cuda.h
+++ /dev/null
@@ -1,126 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifndef LMP_FIX_NH_CUDA_H
-#define LMP_FIX_NH_CUDA_H
-
-#include "fix.h"
-#include "cuda_precision.h"
-
-namespace LAMMPS_NS {
-
-class FixNHCuda : public Fix {
- public:
- FixNHCuda(class LAMMPS *, int, char **);
- virtual ~FixNHCuda();
- int setmask();
- virtual void init();
- void setup(int);
- virtual void initial_integrate(int);
- virtual void final_integrate();
- void initial_integrate_respa(int, int, int);
- void final_integrate_respa(int, int);
- double compute_scalar();
- double compute_vector(int);
- void write_restart(FILE *);
- void restart(char *);
- int modify_param(int, char **);
- void reset_dt();
-
- protected:
- class Cuda *cuda;
- int dimension,which;
- double dtv,dtf,dthalf,dt4,dt8,dto;
- double boltz,nktv2p,tdof;
- double vol0,t0;
-
- double t_start,t_stop;
- double t_current,t_target;
- double t_freq;
-
- int tstat_flag; // 1 if control T
- int pstat_flag; // 1 if control P
-
- int pstyle,pcouple,allremap;
- int p_flag[6]; // 1 if control P on this dim, 0 if not
- double p_start[6],p_stop[6];
- double p_freq[6],p_target[6];
- double omega[6],omega_dot[6];
- double omega_mass[6];
- double p_current[6],dilation[6];
- double drag,tdrag_factor; // drag factor on particle thermostat
- double pdrag_factor; // drag factor on barostat
- double factor[6]; // velocity scaling due to barostat
- int kspace_flag; // 1 if KSpace invoked, 0 if not
- int nrigid; // number of rigid fixes
- int *rfix; // indices of rigid fixes
-
- int nlevels_respa;
- double *step_respa;
-
- char *id_temp,*id_press;
- class Compute *temperature,*pressure;
- int tflag,pflag;
-
- double *eta,*eta_dot; // chain thermostat for particles
- double *eta_dotdot;
- double *eta_mass;
- int mtchain; // length of chain
-
- double *etap; // chain thermostat for barostat
- double *etap_dot;
- double *etap_dotdot;
- double *etap_mass;
- int mpchain; // length of chain
-
- int mtk_flag; // 0 if using Hoover barostat
- double mtk_term1,mtk_term2;
- int mtchain_default_flag;
- int pdim; // number of barostatted dims
- double mvv_current[3]; // diagonal of KE tensor
- double mtk_factor; // MTK factor
- double p_freq_max; // maximum barostat frequency
-
- double p_hydro; // hydrostatic target pressure
-
- int nc_tchain,nc_pchain;
- double factor_eta;
- double sigma[6]; // scaled target stress
- double fdev[6]; // deviatoric force on barostat
- int deviatoric_flag; // 0 if target stress tensor is hydrostatic
- double h0_inv[6]; // h_inv of reference (zero strain) box
- int nreset_h0; // interval for resetting h0
-
- void couple();
- void couple_ke();
- void remap();
- void nhc_temp_integrate();
- void nhc_press_integrate();
-
- virtual void nve_x(); // may be overwritten by child classes
- virtual void nve_v();
- virtual void nh_v_press();
- virtual void nh_v_temp();
-
- void compute_sigma();
- void compute_deviatoric();
- double compute_strain_energy();
- void compute_press_target();
- void nh_omega_dot();
-
- X_CFLOAT triggerneighsq;
-};
-
-}
-
-#endif
diff --git a/src/USER-CUDA/fix_npt_cuda.cpp b/src/USER-CUDA/fix_npt_cuda.cpp
deleted file mode 100644
index ea2dd1fe04..0000000000
--- a/src/USER-CUDA/fix_npt_cuda.cpp
+++ /dev/null
@@ -1,75 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstring>
-#include "fix_npt_cuda.h"
-#include "modify.h"
-#include "error.h"
-
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixNPTCuda::FixNPTCuda(LAMMPS *lmp, int narg, char **arg) :
- FixNHCuda(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (!tstat_flag)
- error->all(FLERR,"Temperature control must be used with fix npt");
- if (!pstat_flag)
- error->all(FLERR,"Pressure control must be used with fix npt");
-
- // create a new compute temp style
- // id = fix-ID + temp
- // compute group = all since pressure is always global (group all)
- // and thus its KE/temperature contribution should use group all
-
- int n = strlen(id) + 6;
- id_temp = new char[n];
- strcpy(id_temp,id);
- strcat(id_temp,"_temp");
-
- char **newarg = new char*[3];
- newarg[0] = id_temp;
- newarg[1] = (char *) "all";
- newarg[2] = (char *) "temp/cuda";
-
- modify->add_compute(3,newarg);
- delete [] newarg;
- tflag = 1;
-
- // create a new compute pressure style
- // id = fix-ID + press, compute group = all
- // pass id_temp as 4th arg to pressure constructor
-
- n = strlen(id) + 7;
- id_press = new char[n];
- strcpy(id_press,id);
- strcat(id_press,"_press");
-
- newarg = new char*[4];
- newarg[0] = id_press;
- newarg[1] = (char *) "all";
- newarg[2] = (char *) "pressure/cuda";
- newarg[3] = id_temp;
- modify->add_compute(4,newarg);
- delete [] newarg;
- pflag = 1;
-}
diff --git a/src/USER-CUDA/fix_npt_cuda.h b/src/USER-CUDA/fix_npt_cuda.h
deleted file mode 100644
index e10efb6a9a..0000000000
--- a/src/USER-CUDA/fix_npt_cuda.h
+++ /dev/null
@@ -1,36 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(npt/cuda,FixNPTCuda)
-
-#else
-
-#ifndef LMP_FIX_NPTCuda_H
-#define LMP_FIX_NPTCuda_H
-
-#include "fix_nh_cuda.h"
-
-namespace LAMMPS_NS {
-
-class FixNPTCuda : public FixNHCuda {
- public:
- FixNPTCuda(class LAMMPS *, int, char **);
- ~FixNPTCuda() {}
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_nve_cuda.cpp b/src/USER-CUDA/fix_nve_cuda.cpp
deleted file mode 100644
index fbe85b176f..0000000000
--- a/src/USER-CUDA/fix_nve_cuda.cpp
+++ /dev/null
@@ -1,157 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstdio>
-#include <cstring>
-#include "fix_nve_cuda.h"
-#include "fix_nve_cuda_cu.h"
-#include "atom.h"
-#include "force.h"
-#include "update.h"
-#include "respa.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixNVECuda::FixNVECuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
-
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (strcmp(style,"nve/sphere") != 0 && narg < 3)
- error->all(FLERR,"Illegal fix nve command");
-
- time_integrate = 1;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixNVECuda::setmask()
-{
- int mask = 0;
- mask |= INITIAL_INTEGRATE_CUDA;
- mask |= FINAL_INTEGRATE_CUDA;
- // mask |= INITIAL_INTEGRATE_RESPA_CUDA;
- // mask |= FINAL_INTEGRATE_RESPA_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNVECuda::init()
-{
- dtv = update->dt;
- dtf = 0.5 * update->dt * force->ftm2v;
-
- if (strstr(update->integrate_style,"respa"))
- step_respa = ((Respa *) update->integrate)->step;
-
- triggerneighsq= cuda->shared_data.atom.triggerneighsq;
- cuda->neighbor_decide_by_integrator=1;
- Cuda_FixNVECuda_Init(&cuda->shared_data,dtv,dtf);
-
-}
-
-/* ----------------------------------------------------------------------
- allow for both per-type and per-atom mass
-------------------------------------------------------------------------- */
-
-void FixNVECuda::initial_integrate(int vflag)
-{
- if(triggerneighsq!=cuda->shared_data.atom.triggerneighsq)
- {
- triggerneighsq= cuda->shared_data.atom.triggerneighsq;
- Cuda_FixNVECuda_Init(&cuda->shared_data,dtv,dtf);
- }
- int nlocal = atom->nlocal;
- if(igroup == atom->firstgroup) nlocal = atom->nfirst;
-
- Cuda_FixNVECuda_InitialIntegrate(& cuda->shared_data, groupbit,nlocal);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNVECuda::final_integrate()
-{
- int nlocal = atom->nlocal;
- if(igroup == atom->firstgroup) nlocal = atom->nfirst;
-
- Cuda_FixNVECuda_FinalIntegrate(& cuda->shared_data, groupbit,nlocal);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNVECuda::initial_integrate_respa(int vflag, int ilevel, int flag)
-{
- //this point should not be reached yet since RESPA is not supported
- if (flag) return; // only used by NPT,NPH
-
- dtv = step_respa[ilevel];
- dtf = 0.5 * step_respa[ilevel] * force->ftm2v;
-
- // innermost level - NVE update of v and x
- // all other levels - NVE update of v
-
- if(ilevel == 0) initial_integrate(vflag);
- else final_integrate();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNVECuda::final_integrate_respa(int ilevel, int iloop)
-{
- //this point should not be reached yet since RESPA is not supported
- dtf = 0.5 * step_respa[ilevel] * force->ftm2v;
- final_integrate();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixNVECuda::reset_dt()
-{
- dtv = update->dt;
- dtf = 0.5 * update->dt * force->ftm2v;
- Cuda_FixNVECuda_Init(&cuda->shared_data,dtv,dtf);
-}
diff --git a/src/USER-CUDA/fix_nve_cuda.h b/src/USER-CUDA/fix_nve_cuda.h
deleted file mode 100644
index 090d327db5..0000000000
--- a/src/USER-CUDA/fix_nve_cuda.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(nve/cuda,FixNVECuda)
-
-#else
-
-#ifndef LMP_FIX_NVE_CUDA_H
-#define LMP_FIX_NVE_CUDA_H
-
-#include "fix.h"
-#include "cuda_precision.h"
-
-namespace LAMMPS_NS {
-
-class FixNVECuda : public Fix
-{
- public:
- FixNVECuda(class LAMMPS *, int, char **);
- int setmask();
- virtual void init();
- virtual void initial_integrate(int);
- virtual void final_integrate();
- void initial_integrate_respa(int, int, int);
- void final_integrate_respa(int, int);
- void reset_dt();
-
- X_CFLOAT triggerneighsq;
-
- protected:
- class Cuda *cuda;
- double dtv, dtf;
- double *step_respa;
- int mass_require;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_nvt_cuda.cpp b/src/USER-CUDA/fix_nvt_cuda.cpp
deleted file mode 100644
index e1380b0005..0000000000
--- a/src/USER-CUDA/fix_nvt_cuda.cpp
+++ /dev/null
@@ -1,52 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstring>
-#include "fix_nvt_cuda.h"
-#include "group.h"
-#include "modify.h"
-#include "error.h"
-
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixNVTCuda::FixNVTCuda(LAMMPS *lmp, int narg, char **arg) :
- FixNHCuda(lmp, narg, arg)
-{
- if (!tstat_flag)
- error->all(FLERR,"Temperature control must be used with fix nvt");
- if (pstat_flag)
- error->all(FLERR,"Pressure control can not be used with fix nvt");
-
- // create a new compute temp style
- // id = fix-ID + temp
-
- int n = strlen(id) + 6;
- id_temp = new char[n];
- strcpy(id_temp,id);
- strcat(id_temp,"_temp");
-
- char **newarg = new char*[3];
- newarg[0] = id_temp;
- newarg[1] = group->names[igroup];
- newarg[2] = (char *) "temp/cuda";
-
- modify->add_compute(3,newarg);
- delete [] newarg;
- tflag = 1;
-}
diff --git a/src/USER-CUDA/fix_nvt_cuda.h b/src/USER-CUDA/fix_nvt_cuda.h
deleted file mode 100644
index 65f38e05d6..0000000000
--- a/src/USER-CUDA/fix_nvt_cuda.h
+++ /dev/null
@@ -1,36 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(nvt/cuda,FixNVTCuda)
-
-#else
-
-#ifndef LMP_FIX_NVTCuda_H
-#define LMP_FIX_NVTCuda_H
-
-#include "fix_nh_cuda.h"
-
-namespace LAMMPS_NS {
-
-class FixNVTCuda : public FixNHCuda {
- public:
- FixNVTCuda(class LAMMPS *, int, char **);
- ~FixNVTCuda() {}
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_set_force_cuda.cpp b/src/USER-CUDA/fix_set_force_cuda.cpp
deleted file mode 100644
index b7000a5548..0000000000
--- a/src/USER-CUDA/fix_set_force_cuda.cpp
+++ /dev/null
@@ -1,184 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-#include <cstring>
-#include <cstdlib>
-#include "fix_set_force_cuda.h"
-#include "fix_set_force_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "respa.h"
-#include "error.h"
-#include "force.h"
-#include "user_cuda.h"
-#include "memory.h"
-#include "cuda_modify_flags.h"
-
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixSetForceCuda::FixSetForceCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg != 6) error->all(FLERR,"Illegal fix setforce/cuda command");
-
- vector_flag = 1;
- size_vector = 3;
- global_freq = 1;
- extvector = 1;
-
- flagx = flagy = flagz = 1;
- if (strcmp(arg[3],"NULL") == 0) flagx = 0;
- else xvalue = force->numeric(FLERR,arg[3]);
- if (strcmp(arg[4],"NULL") == 0) flagy = 0;
- else yvalue = force->numeric(FLERR,arg[4]);
- if (strcmp(arg[5],"NULL") == 0) flagz = 0;
- else zvalue = force->numeric(FLERR,arg[5]);
-
- force_flag = 0;
- foriginal[0] = foriginal[1] = foriginal[2] = 0.0;
- cu_foriginal=NULL;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixSetForceCuda::setmask()
-{
- int mask = 0;
- mask |= POST_FORCE_CUDA;
- mask |= THERMO_ENERGY_CUDA;
- mask |= POST_FORCE_RESPA;
- mask |= MIN_POST_FORCE_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixSetForceCuda::init()
-{
- if(not cu_foriginal)
- cu_foriginal = new cCudaData<double, F_CFLOAT, x> (foriginal,3);
- if (strstr(update->integrate_style,"respa"))
- nlevels_respa = ((Respa *) update->integrate)->nlevels;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixSetForceCuda::setup(int vflag)
-{
- MYDBG( printf("# CUDA: FixSetForceCuda::setup\n"); )
-
- if (strstr(update->integrate_style,"verlet"))
- {
- Cuda_FixSetForceCuda_Init(&cuda->shared_data);
- cuda->cu_f->upload();
- post_force(vflag);
- cuda->cu_f->download();
-
- }
- else {
- ((Respa *) update->integrate)->copy_flevel_f(nlevels_respa-1);
- cuda->cu_f->download();
- post_force_respa(vflag,nlevels_respa-1,0);
- cuda->cu_f->upload();
- ((Respa *) update->integrate)->copy_f_flevel(nlevels_respa-1);
- }
- MYDBG( printf("# CUDA: FixSetForceCuda::setup done\n"); )
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixSetForceCuda::min_setup(int vflag)
-{
- post_force(vflag);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixSetForceCuda::post_force(int vflag)
-{
- MYDBG( printf("# CUDA: FixSetForceCuda::postforce start\n"); )
- force_flag = 0;
- cu_foriginal->memset_device(0);
- Cuda_FixSetForceCuda_PostForce(&cuda->shared_data, groupbit, xvalue, yvalue,zvalue,(F_CFLOAT*) cu_foriginal->dev_data(),flagx,flagy,flagz);
- cu_foriginal->download();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixSetForceCuda::post_force_respa(int vflag, int ilevel, int iloop)
-{
- if (ilevel == nlevels_respa-1) post_force(vflag);
- else {
- cuda->cu_f->download();
- cuda->cu_mask->download();
-
- double **f = atom->f;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- foriginal[0] = foriginal[1] = foriginal[2] = 0.0;
- force_flag = 0;
-
- for (int i = 0; i < nlocal; i++)
- if (mask[i] & groupbit) {
- foriginal[0] += f[i][0];
- foriginal[1] += f[i][1];
- foriginal[2] += f[i][2];
- if (flagx) f[i][0] = 0.0;
- if (flagy) f[i][1] = 0.0;
- if (flagz) f[i][2] = 0.0;
- }
- cuda->cu_f->upload();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixSetForceCuda::min_post_force(int vflag)
-{
- post_force(vflag);
-}
-
-
-/* ----------------------------------------------------------------------
- return components of total force on fix group before force was changed
-------------------------------------------------------------------------- */
-
-double FixSetForceCuda::compute_vector(int n)
-{
- // only sum across procs one time
-
- if (force_flag == 0) {
- MPI_Allreduce(foriginal,foriginal_all,3,MPI_DOUBLE,MPI_SUM,world);
- force_flag = 1;
- }
- return foriginal_all[n+1];
-}
diff --git a/src/USER-CUDA/fix_set_force_cuda.h b/src/USER-CUDA/fix_set_force_cuda.h
deleted file mode 100644
index a195aec0ec..0000000000
--- a/src/USER-CUDA/fix_set_force_cuda.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(setforce/cuda,FixSetForceCuda)
-
-#else
-
-#ifndef LMP_FIX_SET_FORCE_CUDA_H
-#define LMP_FIX_SET_FORCE_CUDA_H
-
-#include "fix.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class FixSetForceCuda : public Fix {
- public:
- FixSetForceCuda(class LAMMPS *, int, char **);
- int setmask();
- void init();
- void setup(int);
- void min_setup(int);
- void post_force(int);
- void post_force_respa(int, int, int);
- void min_post_force(int);
- double compute_vector(int);
-
- private:
- class Cuda *cuda;
- int flagx,flagy,flagz;
- double xvalue,yvalue,zvalue;
- double foriginal[3],foriginal_all[3];
- cCudaData<double , F_CFLOAT , x>* cu_foriginal;
- int force_flag;
- int nlevels_respa;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_shake_cuda.cpp b/src/USER-CUDA/fix_shake_cuda.cpp
deleted file mode 100644
index 92274d1d46..0000000000
--- a/src/USER-CUDA/fix_shake_cuda.cpp
+++ /dev/null
@@ -1,2885 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <cmath>
-#include <cstdlib>
-#include <cstring>
-#include <cstdio>
-#include <ctime>
-#include "fix_shake_cuda.h"
-#include "fix_shake_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "respa.h"
-#include "modify.h"
-#include "domain.h"
-#include "force.h"
-#include "bond.h"
-#include "angle.h"
-#include "comm.h"
-#include "group.h"
-#include "fix_respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-#include "math_const.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-using namespace MathConst;
-
-#define BIG 1.0e20
-#define MASSDELTA 0.1
-
-/* ---------------------------------------------------------------------- */
-
-FixShakeCuda::FixShakeCuda(LAMMPS* lmp, int narg, char** arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
-
- if(cuda == NULL)
- error->all(FLERR, "You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if(atom->map_style != 1)
- error->all(FLERR, "Fix shake/cuda needs atom map style array. In particular it does not currently work with hash-tables.");
-
- MPI_Comm_rank(world, &me);
- MPI_Comm_size(world, &nprocs);
- neighbor_step = true;
-
- virial_flag = 1;
- create_attribute = 1;
- dof_flag = 1;
-
- // error check
-
- if(atom->molecular == 0)
- error->all(FLERR, "Cannot use fix shake with non-molecular system");
-
- // perform initial allocation of atom-based arrays
- // register with Atom class
-
- shake_flag = NULL;
- shake_atom = shake_type = NULL;
- xshake = NULL;
- cu_shake_flag = NULL;
- cu_shake_atom = NULL;
- cu_shake_type = NULL;
- cu_xshake = NULL;
- cu_list = NULL;
- cu_bond_distance = NULL;
- cu_angle_distance = NULL;
- cu_virial = new cCudaData<double , ENERGY_CFLOAT , xx >(virial, 6);
- grow_arrays(atom->nmax);
- atom->add_callback(0);
-
- // set comm size needed by this fix
-
- comm_forward = 3;
-
- // parse SHAKE args
-
- if(narg < 8) error->all(FLERR, "Illegal fix shake command");
-
- tolerance = force->numeric(FLERR,arg[3]);
- max_iter = force->inumeric(FLERR,arg[4]);
- output_every = force->inumeric(FLERR,arg[5]);
-
- // parse SHAKE args for bond and angle types
- // will be used by find_clusters
- // store args for "b" "a" "t" as flags in (1:n) list for fast access
- // store args for "m" in list of length nmass for looping over
- // for "m" verify that atom masses have been set
-
- bond_flag = new int[atom->nbondtypes + 1];
-
- for(int i = 1; i <= atom->nbondtypes; i++) bond_flag[i] = 0;
-
- angle_flag = new int[atom->nangletypes + 1];
-
- for(int i = 1; i <= atom->nangletypes; i++) angle_flag[i] = 0;
-
- type_flag = new int[atom->ntypes + 1];
-
- for(int i = 1; i <= atom->ntypes; i++) type_flag[i] = 0;
-
- mass_list = new double[atom->ntypes];
- nmass = 0;
-
- char mode = '\0';
- int next = 6;
-
- while(next < narg) {
-
- if(strcmp(arg[next], "b") == 0) mode = 'b';
- else if(strcmp(arg[next], "a") == 0) mode = 'a';
- else if(strcmp(arg[next], "t") == 0) mode = 't';
- else if(strcmp(arg[next], "m") == 0) {
- mode = 'm';
- atom->check_mass();
-
- } else if(mode == 'b') {
- int i = force->inumeric(FLERR,arg[next]);
-
- if(i < 1 || i > atom->nbondtypes)
- error->all(FLERR, "Invalid bond type index for fix shake");
-
- bond_flag[i] = 1;
-
- } else if(mode == 'a') {
- int i = force->inumeric(FLERR,arg[next]);
-
- if(i < 1 || i > atom->nangletypes)
- error->all(FLERR, "Invalid angle type index for fix shake");
-
- angle_flag[i] = 1;
-
- } else if(mode == 't') {
- int i = force->inumeric(FLERR,arg[next]);
-
- if(i < 1 || i > atom->ntypes)
- error->all(FLERR, "Invalid atom type index for fix shake");
-
- type_flag[i] = 1;
-
- } else if(mode == 'm') {
- double massone = force->numeric(FLERR,arg[next]);
-
- if(massone == 0.0) error->all(FLERR, "Invalid atom mass for fix shake");
-
- if(nmass == atom->ntypes) error->all(FLERR, "Too many masses for fix shake");
-
- mass_list[nmass++] = massone;
-
- } else error->all(FLERR, "Illegal fix shake command");
-
- next++;
- }
-
- // allocate bond and angle distance arrays, indexed from 1 to n
-
- bond_distance = new double[atom->nbondtypes + 1];
- angle_distance = new double[atom->nangletypes + 1];
-
- cu_bond_distance = new cCudaData<double, X_CFLOAT, xx> (bond_distance, atom->nbondtypes + 1);
- cu_angle_distance = new cCudaData<double, X_CFLOAT, xx> (angle_distance, atom->nangletypes + 1);
-
- // allocate statistics arrays
-
- if(output_every) {
- int nb = atom->nbondtypes + 1;
- b_count = new int[nb];
- b_count_all = new int[nb];
- b_ave = new double[nb];
- b_ave_all = new double[nb];
- b_max = new double[nb];
- b_max_all = new double[nb];
- b_min = new double[nb];
- b_min_all = new double[nb];
-
- int na = atom->nangletypes + 1;
- a_count = new int[na];
- a_count_all = new int[na];
- a_ave = new double[na];
- a_ave_all = new double[na];
- a_max = new double[na];
- a_max_all = new double[na];
- a_min = new double[na];
- a_min_all = new double[na];
- }
-
- cudable_comm = true;
- // identify all SHAKE clusters
-
- find_clusters();
-
- // initialize list of SHAKE clusters to constrain
-
- maxlist = 0;
- list = NULL;
- Cuda_FixShakeCuda_Init(&cuda->shared_data, dtv, dtfsq,
- cu_shake_flag->dev_data(), cu_shake_atom->dev_data(), cu_shake_type->dev_data(), cu_xshake->dev_data(),
- cu_bond_distance->dev_data(), cu_angle_distance->dev_data(), cu_virial->dev_data(),
- max_iter, tolerance);
-
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-FixShakeCuda::~FixShakeCuda()
-{
- // unregister callbacks to this fix from Atom class
-
- atom->delete_callback(id, 0);
-
- // set bond_type and angle_type back to positive for SHAKE clusters
- // must set for all SHAKE bonds and angles stored by each atom
-
- int** bond_type = atom->bond_type;
- int** angle_type = atom->angle_type;
- int nlocal = atom->nlocal;
-
- int n;
-
- for(int i = 0; i < nlocal; i++) {
- if(shake_flag[i] == 0) continue;
- else if(shake_flag[i] == 1) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][2]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = anglefind(i, shake_atom[i][1], shake_atom[i][2]);
-
- if(n >= 0) angle_type[i][n] = -angle_type[i][n];
- } else if(shake_flag[i] == 2) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
- } else if(shake_flag[i] == 3) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][2]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
- } else if(shake_flag[i] == 4) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][2]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][3]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
- }
- }
-
- // delete locally stored arrays
-
- memory->destroy(shake_flag);
- memory->destroy(shake_atom);
- memory->destroy(shake_type);
- memory->destroy(xshake);
-
- delete [] bond_flag;
- delete [] angle_flag;
- delete [] type_flag;
- delete [] mass_list;
-
- delete [] bond_distance;
- delete [] angle_distance;
-
- if(output_every) {
- delete [] b_count;
- delete [] b_count_all;
- delete [] b_ave;
- delete [] b_ave_all;
- delete [] b_max;
- delete [] b_max_all;
- delete [] b_min;
- delete [] b_min_all;
-
- delete [] a_count;
- delete [] a_count_all;
- delete [] a_ave;
- delete [] a_ave_all;
- delete [] a_max;
- delete [] a_max_all;
- delete [] a_min;
- delete [] a_min_all;
- }
-
- memory->destroy(list);
-
- delete cu_shake_flag;
- delete cu_shake_atom;
- delete cu_shake_type;
- delete cu_xshake;
- delete cu_list;
- delete cu_bond_distance;
- delete cu_angle_distance;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixShakeCuda::setmask()
-{
- int mask = 0;
- mask |= PRE_NEIGHBOR_CUDA;
- mask |= POST_FORCE_CUDA;
- mask |= POST_FORCE_RESPA;
- return mask;
-}
-
-/* ----------------------------------------------------------------------
- set bond and angle distances
- this init must happen after force->bond and force->angle inits
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::init()
-{
- int i, m, flag, flag_all, type1, type2, bond1_type, bond2_type;
- double rsq, angle;
-
- // error if more than one shake fix
-
- int count = 0;
-
- for(i = 0; i < modify->nfix; i++)
- if(strcmp(modify->fix[i]->style, "shake") == 0) count++;
-
- if(count > 1) error->all(FLERR, "More than one fix shake");
-
- // cannot use with minimization since SHAKE turns off bonds
- // that should contribute to potential energy
-
- if(update->whichflag == 2)
- error->all(FLERR, "Fix shake cannot be used with minimization");
-
- // error if npt,nph fix comes before shake fix
-
- for(i = 0; i < modify->nfix; i++) {
- if(strcmp(modify->fix[i]->style, "npt") == 0) break;
-
- if(strcmp(modify->fix[i]->style, "nph") == 0) break;
- }
-
- if(i < modify->nfix) {
- for(int j = i; j < modify->nfix; j++)
- if(strcmp(modify->fix[j]->style, "shake") == 0)
- error->all(FLERR, "Shake fix must come before NPT/NPH fix");
- }
-
- // if rRESPA, find associated fix that must exist
- // could have changed locations in fix list since created
- // set ptrs to rRESPA variables
-
- if(strstr(update->integrate_style, "respa")) {
- for(i = 0; i < modify->nfix; i++)
- if(strcmp(modify->fix[i]->style, "RESPA") == 0) ifix_respa = i;
-
- nlevels_respa = ((Respa*) update->integrate)->nlevels;
- loop_respa = ((Respa*) update->integrate)->loop;
- step_respa = ((Respa*) update->integrate)->step;
- }
-
- // set equilibrium bond distances
-
- if(force->bond == NULL)
- error->all(FLERR, "Bond potential must be defined for SHAKE");
-
- for(i = 1; i <= atom->nbondtypes; i++)
- bond_distance[i] = force->bond->equilibrium_distance(i);
-
- // set equilibrium angle distances
-
- int nlocal = atom->nlocal;
-
- for(i = 1; i <= atom->nangletypes; i++) {
- if(angle_flag[i] == 0) continue;
-
- if(force->angle == NULL)
- error->all(FLERR, "Angle potential must be defined for SHAKE");
-
- // scan all atoms for a SHAKE angle cluster
- // extract bond types for the 2 bonds in the cluster
- // bond types must be same in all clusters of this angle type,
- // else set error flag
-
- flag = 0;
- bond1_type = bond2_type = 0;
-
- for(m = 0; m < nlocal; m++) {
- if(shake_flag[m] != 1) continue;
-
- if(shake_type[m][2] != i) continue;
-
- type1 = MIN(shake_type[m][0], shake_type[m][1]);
- type2 = MAX(shake_type[m][0], shake_type[m][1]);
-
- if(bond1_type > 0) {
- if(type1 != bond1_type || type2 != bond2_type) {
- flag = 1;
- break;
- }
- }
-
- bond1_type = type1;
- bond2_type = type2;
- }
-
- // error check for any bond types that are not the same
-
- MPI_Allreduce(&flag, &flag_all, 1, MPI_INT, MPI_MAX, world);
-
- if(flag_all) error->all(FLERR, "Shake angles have different bond types");
-
- // insure all procs have bond types
-
- MPI_Allreduce(&bond1_type, &flag_all, 1, MPI_INT, MPI_MAX, world);
- bond1_type = flag_all;
- MPI_Allreduce(&bond2_type, &flag_all, 1, MPI_INT, MPI_MAX, world);
- bond2_type = flag_all;
-
- // if bond types are 0, no SHAKE angles of this type exist
- // just skip this angle
-
- if(bond1_type == 0) {
- angle_distance[i] = 0.0;
- continue;
- }
-
- // compute the angle distance as a function of 2 bond distances
-
- angle = force->angle->equilibrium_angle(i);
- rsq = 2.0 * bond_distance[bond1_type] * bond_distance[bond2_type] *
- (1.0 - cos(angle));
- angle_distance[i] = sqrt(rsq);
- }
-}
-
-/* ----------------------------------------------------------------------
- SHAKE as pre-integrator constraint
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::setup(int vflag)
-{
- pre_neighbor();
-
- if(output_every) stats();
-
- // setup SHAKE output
-
- int ntimestep = update->ntimestep;
- next_output = ntimestep + output_every;
-
- if(output_every == 0) next_output = update->laststep + 1;
-
- if(output_every && ntimestep % output_every != 0)
- next_output = (ntimestep / output_every) * output_every + output_every;
-
- // half timestep constraint on pre-step, full timestep thereafter
-
- if(strstr(update->integrate_style, "verlet")) {
- dtv = update->dt;
- dtfsq = 0.5 * update->dt * update->dt * force->ftm2v;
- post_force(vflag);
- dtfsq = update->dt * update->dt * force->ftm2v;
- } else {
- dtv = step_respa[0];
- dtf_innerhalf = 0.5 * step_respa[0] * force->ftm2v;
- dtf_inner = dtf_innerhalf;
- ((Respa*) update->integrate)->copy_flevel_f(nlevels_respa - 1);
- post_force_respa(vflag, nlevels_respa - 1, 0);
- ((Respa*) update->integrate)->copy_f_flevel(nlevels_respa - 1);
- dtf_inner = step_respa[0] * force->ftm2v;
- }
-
- Cuda_FixShakeCuda_Init(&cuda->shared_data, dtv, dtfsq,
- cu_shake_flag->dev_data(), cu_shake_atom->dev_data(), cu_shake_type->dev_data(), cu_xshake->dev_data(),
- cu_bond_distance->dev_data(), cu_angle_distance->dev_data(), cu_virial->dev_data(),
- max_iter, tolerance);
-}
-
-/* ----------------------------------------------------------------------
- build list of SHAKE clusters to constrain
- if one or more atoms in cluster are on this proc,
- this proc lists the cluster exactly once
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::pre_neighbor()
-{
- int atom1, atom2, atom3, atom4;
-
- // local copies of atom quantities
- // used by SHAKE until next re-neighboring
-
- x = atom->x;
- v = atom->v;
- f = atom->f;
- mass = atom->mass;
- rmass = atom->rmass;
- type = atom->type;
- nlocal = atom->nlocal;
-
- // extend size of SHAKE list if necessary
-
- if(nlocal > maxlist) {
- maxlist = nlocal;
- memory->destroy(list);
- memory->create(list, maxlist, "shake:list");
- delete cu_list;
- cu_list = new cCudaData<int , int , xx >(list, maxlist);
- }
-
- // build list of SHAKE clusters I compute
-
- nlist = 0;
- int count2 = 0, count3 = 0, count4 = 0, count3a = 0;
-
- for(int i = 0; i < nlocal; i++)
- if(shake_flag[i]) {
- if(shake_flag[i] == 2) count2++;
-
- if(shake_flag[i] == 3) count3++;
-
- if(shake_flag[i] == 4) count4++;
-
- if(shake_flag[i] == 1) count3a++;
-
- if(shake_flag[i] == 2) {
- atom1 = atom->map(shake_atom[i][0]);
- atom2 = atom->map(shake_atom[i][1]);
-
- if(atom1 == -1 || atom2 == -1) {
- char str[128];
- sprintf(str,
- "Shake atoms %d %d missing on proc %d at step " BIGINT_FORMAT,
- shake_atom[i][0], shake_atom[i][1], me, update->ntimestep);
- error->one(FLERR, str);
- }
-
- if(i <= atom1 && i <= atom2) list[nlist++] = i;
- } else if(shake_flag[i] % 2 == 1) {
- atom1 = atom->map(shake_atom[i][0]);
- atom2 = atom->map(shake_atom[i][1]);
- atom3 = atom->map(shake_atom[i][2]);
-
- if(atom1 == -1 || atom2 == -1 || atom3 == -1) {
- char str[128];
- sprintf(str,
- "Shake atoms %d %d %d missing on proc %d at step "
- BIGINT_FORMAT,
- shake_atom[i][0], shake_atom[i][1], shake_atom[i][2],
- me, update->ntimestep);
- error->one(FLERR, str);
- }
-
- if(i <= atom1 && i <= atom2 && i <= atom3) list[nlist++] = i;
- } else {
- atom1 = atom->map(shake_atom[i][0]);
- atom2 = atom->map(shake_atom[i][1]);
- atom3 = atom->map(shake_atom[i][2]);
- atom4 = atom->map(shake_atom[i][3]);
-
- if(atom1 == -1 || atom2 == -1 || atom3 == -1 || atom4 == -1) {
- char str[128];
- sprintf(str,
- "Shake atoms %d %d %d %d missing on proc %d at step "
- BIGINT_FORMAT,
- shake_atom[i][0], shake_atom[i][1],
- shake_atom[i][2], shake_atom[i][3],
- me, update->ntimestep);
- error->one(FLERR, str);
- }
-
- if(i <= atom1 && i <= atom2 && i <= atom3 && i <= atom4)
- list[nlist++] = i;
- }
- }
-
- count2 /= 2;
- count3 /= 3;
- count4 /= 4;
- count3a /= 3;
- count3 += count2;
- count4 += count3;
- count3a += count4;
-
- for(int k = 0, l = count2; k < count2; k++) {
- if(shake_flag[list[k]] != 2) {
- while(shake_flag[list[l]] != 2 && l < nlist - 1) l++;
-
- if(shake_flag[list[l]] != 2) {
- printf("FixShakeCuda: Error in List SortA %i %i\n", k, l);
- return;
- }
-
- int tmp = list[k];
- list[k] = list[l];
- list[l] = tmp;
- }
- }
-
- for(int k = count2, l = count3; k < count3; k++) {
- if(shake_flag[list[k]] != 3) {
- while(shake_flag[list[l]] != 3 && l < nlist - 1) l++;
-
- if(shake_flag[list[l]] != 3) {
- printf("FixShakeCuda: Error in List SortB %i %i\n", k, l);
- return;
- }
-
- int tmp = list[k];
- list[k] = list[l];
- list[l] = tmp;
- }
- }
-
- for(int k = count3, l = count4; k < count4; k++) {
- if(shake_flag[list[k]] != 4) {
- while(shake_flag[list[l]] != 4 && l < nlist - 1) l++;
-
- if(shake_flag[list[l]] != 4) {
- printf("FixShakeCuda: Error in List SortC %i %i\n", k, l);
- return;
- }
-
- int tmp = list[k];
- list[k] = list[l];
- list[l] = tmp;
- }
- }
-
- cu_list->upload();
- cu_bond_distance->upload();
- cu_angle_distance->upload();
- cu_shake_flag->upload();
- cu_shake_atom->upload();
- cu_shake_type->upload();
-
- neighbor_step = true;
-}
-
-/* ----------------------------------------------------------------------
- compute the force adjustment for SHAKE constraint
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::post_force(int vflag)
-{
- my_times starttime;
- my_times endtime;
-
-
- if(cuda->finished_setup && neighbor_step) {
- Cuda_FixShakeCuda_Init(&cuda->shared_data, dtv, dtfsq,
- cu_shake_flag->dev_data(), cu_shake_atom->dev_data(), cu_shake_type->dev_data(), cu_xshake->dev_data(),
- cu_bond_distance->dev_data(), cu_angle_distance->dev_data(), cu_virial->dev_data(),
- max_iter, tolerance);
-
- }
-
- if(not cuda->finished_setup)
- cuda->downloadAll();
-
- if(update->ntimestep == next_output) {
- if(cuda->finished_setup)
- cuda->cu_x->download();
-
- stats();
- }
-
- // xshake = unconstrained move with current v,f
-
- unconstrained_update();
-
- // communicate results if necessary
-
- //if(cuda->finished_setup) cu_xshake->download();
-
- if(nprocs > 1) {
- //if(cuda->finished_setup)
- //cu_xshake->download();
- comm->forward_comm_fix(this);
- //if(cuda->finished_setup)
- //cu_xshake->upload();
- }
-
- // virial setup
-
- if(vflag) v_setup(vflag);
- else evflag = 0;
-
- // loop over clusters
-
- my_gettime(CLOCK_REALTIME, &starttime);
-
- if(cuda->finished_setup) {
- cu_virial->upload();
-
- if(vflag_atom) cuda->cu_vatom->upload();
-
- Cuda_FixShakeCuda_Shake(&cuda->shared_data, vflag, vflag_atom, (int*)cu_list->dev_data(), nlist);
- cu_virial->download();
-
- if(vflag_atom) cuda->cu_vatom->download();
-
- } else
- for(int i = 0; i < nlist; i++) {
- int m = list[i];
-
- if(shake_flag[m] == 2) shake2(m);
- else if(shake_flag[m] == 3) shake3(m);
- else if(shake_flag[m] == 4) shake4(m);
- else shake3angle(m);
- }
-
- if((not cuda->finished_setup)) cuda->cu_f->upload();
-
- my_gettime(CLOCK_REALTIME, &endtime);
-
- if(cuda->finished_setup)
- time_postforce += (endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000);
- else
- time_postforce = 0.0;
-
- //printf("Postforce time: %lf\n",time_postforce);
-}
-
-/* ----------------------------------------------------------------------
- count # of degrees-of-freedom removed by SHAKE for atoms in igroup
-------------------------------------------------------------------------- */
-
-int FixShakeCuda::dof(int igroup)
-{
- int groupbit = group->bitmask[igroup];
-
- int* mask = atom->mask;
- int* tag = atom->tag;
- int nlocal = atom->nlocal;
-
- // count dof in a cluster if and only if
- // the central atom is in group and atom i is the central atom
-
- int n = 0;
-
- for(int i = 0; i < nlocal; i++) {
- if(!(mask[i] & groupbit)) continue;
-
- if(shake_flag[i] == 0) continue;
-
- if(shake_atom[i][0] != tag[i]) continue;
-
- if(shake_flag[i] == 1) n += 3;
- else if(shake_flag[i] == 2) n += 1;
- else if(shake_flag[i] == 3) n += 2;
- else if(shake_flag[i] == 4) n += 3;
- }
-
- int nall;
- MPI_Allreduce(&n, &nall, 1, MPI_INT, MPI_SUM, world);
- return nall;
-}
-
-/* ----------------------------------------------------------------------
- identify whether each atom is in a SHAKE cluster
- only include atoms in fix group and those bonds/angles specified in input
- test whether all clusters are valid
- set shake_flag, shake_atom, shake_type values
- set bond,angle types negative so will be ignored in neighbor lists
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::find_clusters()
-{
- int i, j, m, n;
- int flag, flag_all, messtag, loop, nbuf, nbufmax, size;
- double massone;
- int* buf, *bufcopy;
- MPI_Request request;
- MPI_Status status;
-
- if(me == 0 && screen) fprintf(screen, "Finding SHAKE clusters ...\n");
-
- // local copies of atom ptrs
-
- int* tag = atom->tag;
- int* type = atom->type;
- int* mask = atom->mask;
- double* mass = atom->mass;
- double* rmass = atom->rmass;
- int** bond_type = atom->bond_type;
- int** angle_type = atom->angle_type;
- int** nspecial = atom->nspecial;
- int** special = atom->special;
- int nlocal = atom->nlocal;
-
- // setup ring of procs
-
- int next = me + 1;
- int prev = me - 1;
-
- if(next == nprocs) next = 0;
-
- if(prev < 0) prev = nprocs - 1;
-
- // -----------------------------------------------------
- // allocate arrays for self (1d) and bond partners (2d)
- // max = max # of bond partners for owned atoms = 2nd dim of partner arrays
- // npartner[i] = # of bonds attached to atom i
- // nshake[i] = # of SHAKE bonds attached to atom i
- // partner_tag[i][] = global IDs of each partner
- // partner_mask[i][] = mask of each partner
- // partner_type[i][] = type of each partner
- // partner_massflag[i][] = 1 if partner meets mass criterion, 0 if not
- // partner_bondtype[i][] = type of bond attached to each partner
- // partner_shake[i][] = 1 if SHAKE bonded to partner, 0 if not
- // partner_nshake[i][] = nshake value for each partner
- // -----------------------------------------------------
-
- int max = 0;
-
- for(i = 0; i < nlocal; i++) max = MAX(max, nspecial[i][0]);
-
- int* npartner, *nshake;
- memory->create(npartner, nlocal, "shake:npartner");
- memory->create(nshake, nlocal, "shake:nshake");
-
- int** partner_tag, **partner_mask, **partner_type, **partner_massflag;
- int** partner_bondtype, **partner_shake, **partner_nshake;
- memory->create(partner_tag, nlocal, max, "shake:partner_tag");
- memory->create(partner_mask, nlocal, max, "shake:partner_mask");
- memory->create(partner_type, nlocal, max, "shake:partner_type");
- memory->create(partner_massflag, nlocal, max, "shake:partner_massflag");
- memory->create(partner_bondtype, nlocal, max, "shake:partner_bondtype");
- memory->create(partner_shake, nlocal, max, "shake:partner_shake");
- memory->create(partner_nshake, nlocal, max, "shake:partner_nshake");
-
- // -----------------------------------------------------
- // set npartner and partner_tag from special arrays
- // -----------------------------------------------------
-
- for(i = 0; i < nlocal; i++) {
- npartner[i] = nspecial[i][0];
-
- for(j = 0; j < npartner[i]; j++) partner_tag[i][j] = special[i][j];
- }
-
- // -----------------------------------------------------
- // set partner_mask, partner_type, partner_massflag, partner_bondtype
- // for bonded partners
- // requires communication for off-proc partners
- // -----------------------------------------------------
-
- // fill in mask, type, massflag, bondtype if own bond partner
- // info to store in buf for each off-proc bond = nper = 6
- // 2 atoms IDs in bond, space for mask, type, massflag, bondtype
- // nbufmax = largest buffer needed to hold info from any proc
-
- int nper = 6;
-
- nbuf = 0;
-
- for(i = 0; i < nlocal; i++) {
- for(j = 0; j < npartner[i]; j++) {
- partner_mask[i][j] = 0;
- partner_type[i][j] = 0;
- partner_massflag[i][j] = 0;
- partner_bondtype[i][j] = 0;
-
- m = atom->map(partner_tag[i][j]);
-
- if(m >= 0 && m < nlocal) {
- partner_mask[i][j] = mask[m];
- partner_type[i][j] = type[m];
-
- if(nmass) {
- if(rmass) massone = rmass[m];
- else massone = mass[type[m]];
-
- partner_massflag[i][j] = masscheck(massone);
- }
-
- n = bondfind(i, tag[i], partner_tag[i][j]);
-
- if(n >= 0) partner_bondtype[i][j] = bond_type[i][n];
- else {
- n = bondfind(m, tag[i], partner_tag[i][j]);
-
- if(n >= 0) partner_bondtype[i][j] = bond_type[m][n];
- }
- } else nbuf += nper;
- }
- }
-
- MPI_Allreduce(&nbuf, &nbufmax, 1, MPI_INT, MPI_MAX, world);
-
- buf = new int[nbufmax];
- bufcopy = new int[nbufmax];
-
- // fill buffer with info
-
- size = 0;
-
- for(i = 0; i < nlocal; i++) {
- for(j = 0; j < npartner[i]; j++) {
- m = atom->map(partner_tag[i][j]);
-
- if(m < 0 || m >= nlocal) {
- buf[size] = tag[i];
- buf[size + 1] = partner_tag[i][j];
- buf[size + 2] = 0;
- buf[size + 3] = 0;
- buf[size + 4] = 0;
- n = bondfind(i, tag[i], partner_tag[i][j]);
-
- if(n >= 0) buf[size + 5] = bond_type[i][n];
- else buf[size + 5] = 0;
-
- size += nper;
- }
- }
- }
-
- // cycle buffer around ring of procs back to self
- // when receive buffer, scan bond partner IDs for atoms I own
- // if I own partner:
- // fill in mask and type and massflag
- // search for bond with 1st atom and fill in bondtype
-
- messtag = 1;
-
- for(loop = 0; loop < nprocs; loop++) {
- i = 0;
-
- while(i < size) {
- m = atom->map(buf[i + 1]);
-
- if(m >= 0 && m < nlocal) {
- buf[i + 2] = mask[m];
- buf[i + 3] = type[m];
-
- if(nmass) {
- if(rmass) massone = rmass[m];
- else massone = mass[type[m]];
-
- buf[i + 4] = masscheck(massone);
- }
-
- if(buf[i + 5] == 0) {
- n = bondfind(m, buf[i], buf[i + 1]);
-
- if(n >= 0) buf[i + 5] = bond_type[m][n];
- }
- }
-
- i += nper;
- }
-
- if(me != next) {
- MPI_Irecv(bufcopy, nbufmax, MPI_INT, prev, messtag, world, &request);
- MPI_Send(buf, size, MPI_INT, next, messtag, world);
- MPI_Wait(&request, &status);
- MPI_Get_count(&status, MPI_INT, &size);
-
- for(j = 0; j < size; j++) buf[j] = bufcopy[j];
- }
- }
-
- // store partner info returned to me
-
- m = 0;
-
- while(m < size) {
- i = atom->map(buf[m]);
-
- for(j = 0; j < npartner[i]; j++)
- if(buf[m + 1] == partner_tag[i][j]) break;
-
- partner_mask[i][j] = buf[m + 2];
- partner_type[i][j] = buf[m + 3];
- partner_massflag[i][j] = buf[m + 4];
- partner_bondtype[i][j] = buf[m + 5];
- m += nper;
- }
-
- delete [] buf;
- delete [] bufcopy;
-
- // error check for unfilled partner info
- // if partner_type not set, is an error
- // partner_bondtype may not be set if special list is not consistent
- // with bondatom (e.g. due to delete_bonds command)
- // this is OK if one or both atoms are not in fix group, since
- // bond won't be SHAKEn anyway
- // else it's an error
-
- flag = 0;
-
- for(i = 0; i < nlocal; i++)
- for(j = 0; j < npartner[i]; j++) {
- if(partner_type[i][j] == 0) flag = 1;
-
- if(!(mask[i] & groupbit)) continue;
-
- if(!(partner_mask[i][j] & groupbit)) continue;
-
- if(partner_bondtype[i][j] == 0) flag = 1;
- }
-
- MPI_Allreduce(&flag, &flag_all, 1, MPI_INT, MPI_SUM, world);
-
- if(flag_all) error->all(FLERR, "Did not find fix shake partner info");
-
- // -----------------------------------------------------
- // identify SHAKEable bonds
- // set nshake[i] = # of SHAKE bonds attached to atom i
- // set partner_shake[i][] = 1 if SHAKE bonded to partner, 0 if not
- // both atoms must be in group, bondtype must be > 0
- // check if bondtype is in input bond_flag
- // check if type of either atom is in input type_flag
- // check if mass of either atom is in input mass_list
- // -----------------------------------------------------
-
- int np;
-
- for(i = 0; i < nlocal; i++) {
- nshake[i] = 0;
- np = npartner[i];
-
- for(j = 0; j < np; j++) {
- partner_shake[i][j] = 0;
-
- if(!(mask[i] & groupbit)) continue;
-
- if(!(partner_mask[i][j] & groupbit)) continue;
-
- if(partner_bondtype[i][j] <= 0) continue;
-
- if(bond_flag[partner_bondtype[i][j]]) {
- partner_shake[i][j] = 1;
- nshake[i]++;
- continue;
- }
-
- if(type_flag[type[i]] || type_flag[partner_type[i][j]]) {
- partner_shake[i][j] = 1;
- nshake[i]++;
- continue;
- }
-
- if(nmass) {
- if(partner_massflag[i][j]) {
- partner_shake[i][j] = 1;
- nshake[i]++;
- continue;
- } else {
- if(rmass) massone = rmass[i];
- else massone = mass[type[i]];
-
- if(masscheck(massone)) {
- partner_shake[i][j] = 1;
- nshake[i]++;
- continue;
- }
- }
- }
- }
- }
-
- // -----------------------------------------------------
- // set partner_nshake for bonded partners
- // requires communication for off-proc partners
- // -----------------------------------------------------
-
- // fill in partner_nshake if own bond partner
- // info to store in buf for each off-proc bond =
- // 2 atoms IDs in bond, space for nshake value
- // nbufmax = largest buffer needed to hold info from any proc
-
- nbuf = 0;
-
- for(i = 0; i < nlocal; i++) {
- for(j = 0; j < npartner[i]; j++) {
- m = atom->map(partner_tag[i][j]);
-
- if(m >= 0 && m < nlocal) partner_nshake[i][j] = nshake[m];
- else nbuf += 3;
- }
- }
-
- MPI_Allreduce(&nbuf, &nbufmax, 1, MPI_INT, MPI_MAX, world);
-
- buf = new int[nbufmax];
- bufcopy = new int[nbufmax];
-
- // fill buffer with info
-
- size = 0;
-
- for(i = 0; i < nlocal; i++) {
- for(j = 0; j < npartner[i]; j++) {
- m = atom->map(partner_tag[i][j]);
-
- if(m < 0 || m >= nlocal) {
- buf[size] = tag[i];
- buf[size + 1] = partner_tag[i][j];
- size += 3;
- }
- }
- }
-
- // cycle buffer around ring of procs back to self
- // when receive buffer, scan bond partner IDs for atoms I own
- // if I own partner, fill in nshake value
-
- messtag = 2;
-
- for(loop = 0; loop < nprocs; loop++) {
- i = 0;
-
- while(i < size) {
- m = atom->map(buf[i + 1]);
-
- if(m >= 0 && m < nlocal) buf[i + 2] = nshake[m];
-
- i += 3;
- }
-
- if(me != next) {
- MPI_Irecv(bufcopy, nbufmax, MPI_INT, prev, messtag, world, &request);
- MPI_Send(buf, size, MPI_INT, next, messtag, world);
- MPI_Wait(&request, &status);
- MPI_Get_count(&status, MPI_INT, &size);
-
- for(j = 0; j < size; j++) buf[j] = bufcopy[j];
- }
- }
-
- // store partner info returned to me
-
- m = 0;
-
- while(m < size) {
- i = atom->map(buf[m]);
-
- for(j = 0; j < npartner[i]; j++)
- if(buf[m + 1] == partner_tag[i][j]) break;
-
- partner_nshake[i][j] = buf[m + 2];
- m += 3;
- }
-
- delete [] buf;
- delete [] bufcopy;
-
- // -----------------------------------------------------
- // error checks
- // no atom with nshake > 3
- // no connected atoms which both have nshake > 1
- // -----------------------------------------------------
-
- flag = 0;
-
- for(i = 0; i < nlocal; i++) if(nshake[i] > 3) flag = 1;
-
- MPI_Allreduce(&flag, &flag_all, 1, MPI_INT, MPI_SUM, world);
-
- if(flag_all) error->all(FLERR, "Shake cluster of more than 4 atoms");
-
- flag = 0;
-
- for(i = 0; i < nlocal; i++) {
- if(nshake[i] <= 1) continue;
-
- for(j = 0; j < npartner[i]; j++)
- if(partner_shake[i][j] && partner_nshake[i][j] > 1) flag = 1;
- }
-
- MPI_Allreduce(&flag, &flag_all, 1, MPI_INT, MPI_SUM, world);
-
- if(flag_all) error->all(FLERR, "Shake clusters are connected");
-
- // -----------------------------------------------------
- // set SHAKE arrays that are stored with atoms & add angle constraints
- // zero shake arrays for all owned atoms
- // if I am central atom set shake_flag & shake_atom & shake_type
- // for 2-atom clusters, I am central atom if my atom ID < partner ID
- // for 3-atom clusters, test for angle constraint
- // angle will be stored by this atom if it exists
- // if angle type matches angle_flag, then it is angle-constrained
- // shake_flag[] = 0 if atom not in SHAKE cluster
- // 2,3,4 = size of bond-only cluster
- // 1 = 3-atom angle cluster
- // shake_atom[][] = global IDs of 2,3,4 atoms in cluster
- // central atom is 1st
- // for 2-atom cluster, lowest ID is 1st
- // shake_type[][] = bondtype of each bond in cluster
- // for 3-atom angle cluster, 3rd value is angletype
- // -----------------------------------------------------
-
- for(i = 0; i < nlocal; i++) {
- shake_flag[i] = 0;
- shake_atom[i][0] = 0;
- shake_atom[i][1] = 0;
- shake_atom[i][2] = 0;
- shake_atom[i][3] = 0;
- shake_type[i][0] = 0;
- shake_type[i][1] = 0;
- shake_type[i][2] = 0;
-
- if(nshake[i] == 1) {
- for(j = 0; j < npartner[i]; j++)
- if(partner_shake[i][j]) break;
-
- if(partner_nshake[i][j] == 1 && tag[i] < partner_tag[i][j]) {
- shake_flag[i] = 2;
- shake_atom[i][0] = tag[i];
- shake_atom[i][1] = partner_tag[i][j];
- shake_type[i][0] = partner_bondtype[i][j];
- }
- }
-
- if(nshake[i] > 1) {
- shake_flag[i] = 1;
- shake_atom[i][0] = tag[i];
-
- for(j = 0; j < npartner[i]; j++)
- if(partner_shake[i][j]) {
- m = shake_flag[i];
- shake_atom[i][m] = partner_tag[i][j];
- shake_type[i][m - 1] = partner_bondtype[i][j];
- shake_flag[i]++;
- }
- }
-
- if(nshake[i] == 2) {
- n = anglefind(i, shake_atom[i][1], shake_atom[i][2]);
-
- if(n < 0) continue;
-
- if(angle_type[i][n] < 0) continue;
-
- if(angle_flag[angle_type[i][n]]) {
- shake_flag[i] = 1;
- shake_type[i][2] = angle_type[i][n];
- }
- }
- }
-
- // -----------------------------------------------------
- // set shake_flag,shake_atom,shake_type for non-central atoms
- // requires communication for off-proc atoms
- // -----------------------------------------------------
-
- // fill in shake arrays for each bond partner I own
- // info to store in buf for each off-proc bond =
- // all values from shake_flag, shake_atom, shake_type
- // nbufmax = largest buffer needed to hold info from any proc
-
- nbuf = 0;
-
- for(i = 0; i < nlocal; i++) {
- if(shake_flag[i] == 0) continue;
-
- for(j = 0; j < npartner[i]; j++) {
- if(partner_shake[i][j] == 0) continue;
-
- m = atom->map(partner_tag[i][j]);
-
- if(m >= 0 && m < nlocal) {
- shake_flag[m] = shake_flag[i];
- shake_atom[m][0] = shake_atom[i][0];
- shake_atom[m][1] = shake_atom[i][1];
- shake_atom[m][2] = shake_atom[i][2];
- shake_atom[m][3] = shake_atom[i][3];
- shake_type[m][0] = shake_type[i][0];
- shake_type[m][1] = shake_type[i][1];
- shake_type[m][2] = shake_type[i][2];
- } else nbuf += 9;
- }
- }
-
- MPI_Allreduce(&nbuf, &nbufmax, 1, MPI_INT, MPI_MAX, world);
-
- buf = new int[nbufmax];
- bufcopy = new int[nbufmax];
-
- // fill buffer with info
-
- size = 0;
-
- for(i = 0; i < nlocal; i++) {
- if(shake_flag[i] == 0) continue;
-
- for(j = 0; j < npartner[i]; j++) {
- if(partner_shake[i][j] == 0) continue;
-
- m = atom->map(partner_tag[i][j]);
-
- if(m < 0 || m >= nlocal) {
- buf[size] = partner_tag[i][j];
- buf[size + 1] = shake_flag[i];
- buf[size + 2] = shake_atom[i][0];
- buf[size + 3] = shake_atom[i][1];
- buf[size + 4] = shake_atom[i][2];
- buf[size + 5] = shake_atom[i][3];
- buf[size + 6] = shake_type[i][0];
- buf[size + 7] = shake_type[i][1];
- buf[size + 8] = shake_type[i][2];
- size += 9;
- }
- }
- }
-
- // cycle buffer around ring of procs back to self
- // when receive buffer, scan for ID that I own
- // if I own ID, fill in shake array values
-
- messtag = 3;
-
- for(loop = 0; loop < nprocs; loop++) {
- i = 0;
-
- while(i < size) {
- m = atom->map(buf[i]);
-
- if(m >= 0 && m < nlocal) {
- shake_flag[m] = buf[i + 1];
- shake_atom[m][0] = buf[i + 2];
- shake_atom[m][1] = buf[i + 3];
- shake_atom[m][2] = buf[i + 4];
- shake_atom[m][3] = buf[i + 5];
- shake_type[m][0] = buf[i + 6];
- shake_type[m][1] = buf[i + 7];
- shake_type[m][2] = buf[i + 8];
- }
-
- i += 9;
- }
-
- if(me != next) {
- MPI_Irecv(bufcopy, nbufmax, MPI_INT, prev, messtag, world, &request);
- MPI_Send(buf, size, MPI_INT, next, messtag, world);
- MPI_Wait(&request, &status);
- MPI_Get_count(&status, MPI_INT, &size);
-
- for(j = 0; j < size; j++) buf[j] = bufcopy[j];
- }
- }
-
- delete [] buf;
- delete [] bufcopy;
-
- // -----------------------------------------------------
- // free local memory
- // -----------------------------------------------------
-
- memory->destroy(npartner);
- memory->destroy(nshake);
- memory->destroy(partner_tag);
- memory->destroy(partner_mask);
- memory->destroy(partner_type);
- memory->destroy(partner_massflag);
- memory->destroy(partner_bondtype);
- memory->destroy(partner_shake);
- memory->destroy(partner_nshake);
-
- // -----------------------------------------------------
- // set bond_type and angle_type negative for SHAKE clusters
- // must set for all SHAKE bonds and angles stored by each atom
- // -----------------------------------------------------
-
- for(i = 0; i < nlocal; i++) {
- if(shake_flag[i] == 0) continue;
- else if(shake_flag[i] == 1) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][2]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = anglefind(i, shake_atom[i][1], shake_atom[i][2]);
-
- if(n >= 0) angle_type[i][n] = -angle_type[i][n];
- } else if(shake_flag[i] == 2) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
- } else if(shake_flag[i] == 3) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][2]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
- } else if(shake_flag[i] == 4) {
- n = bondfind(i, shake_atom[i][0], shake_atom[i][1]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][2]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
-
- n = bondfind(i, shake_atom[i][0], shake_atom[i][3]);
-
- if(n >= 0) bond_type[i][n] = -bond_type[i][n];
- }
- }
-
- // -----------------------------------------------------
- // print info on SHAKE clusters
- // -----------------------------------------------------
-
- int count1, count2, count3, count4;
- count1 = count2 = count3 = count4 = 0;
-
- for(i = 0; i < nlocal; i++) {
- if(shake_flag[i] == 1) count1++;
- else if(shake_flag[i] == 2) count2++;
- else if(shake_flag[i] == 3) count3++;
- else if(shake_flag[i] == 4) count4++;
- }
-
- for(int i = 0; i < nlocal; i++) {
- }
-
-
- int tmp;
- tmp = count1;
- MPI_Allreduce(&tmp, &count1, 1, MPI_INT, MPI_SUM, world);
- tmp = count2;
- MPI_Allreduce(&tmp, &count2, 1, MPI_INT, MPI_SUM, world);
- tmp = count3;
- MPI_Allreduce(&tmp, &count3, 1, MPI_INT, MPI_SUM, world);
- tmp = count4;
- MPI_Allreduce(&tmp, &count4, 1, MPI_INT, MPI_SUM, world);
-
- if(me == 0) {
- if(screen) {
- fprintf(screen, " %d = # of size 2 clusters\n", count2 / 2);
- fprintf(screen, " %d = # of size 3 clusters\n", count3 / 3);
- fprintf(screen, " %d = # of size 4 clusters\n", count4 / 4);
- fprintf(screen, " %d = # of frozen angles\n", count1 / 3);
- }
-
- if(logfile) {
- fprintf(logfile, " %d = # of size 2 clusters\n", count2 / 2);
- fprintf(logfile, " %d = # of size 3 clusters\n", count3 / 3);
- fprintf(logfile, " %d = # of size 4 clusters\n", count4 / 4);
- fprintf(logfile, " %d = # of frozen angles\n", count1 / 3);
- }
- }
-
- cu_shake_flag->upload();
- cu_shake_atom->upload();
- cu_shake_type->upload();
- Cuda_FixShakeCuda_Init(&cuda->shared_data, dtv, dtfsq,
- cu_shake_flag->dev_data(), cu_shake_atom->dev_data(), cu_shake_type->dev_data(), cu_xshake->dev_data(),
- cu_bond_distance->dev_data(), cu_angle_distance->dev_data(), cu_virial->dev_data(),
- max_iter, tolerance);
-
-}
-
-void FixShakeCuda::swap_clusters(int i, int j)
-{
- int tmp;
- tmp = shake_flag[i];
- shake_flag[i] = shake_flag[j];
- shake_flag[j] = tmp;
- tmp = shake_atom[i][0];
- shake_atom[i][0] = shake_atom[j][0];
- shake_atom[j][0] = tmp;
- tmp = shake_atom[i][1];
- shake_atom[i][1] = shake_atom[j][1];
- shake_atom[j][1] = tmp;
- tmp = shake_atom[i][2];
- shake_atom[i][2] = shake_atom[j][2];
- shake_atom[j][2] = tmp;
- tmp = shake_atom[i][3];
- shake_atom[i][3] = shake_atom[j][3];
- shake_atom[j][3] = tmp;
- tmp = shake_type[i][0];
- shake_type[i][0] = shake_type[j][0];
- shake_type[j][0] = tmp;
- tmp = shake_type[i][1];
- shake_type[i][1] = shake_type[j][1];
- shake_type[j][1] = tmp;
- tmp = shake_type[i][2];
- shake_type[i][2] = shake_type[j][2];
- shake_type[j][2] = tmp;
-}
-
-/* ----------------------------------------------------------------------
- check if massone is within MASSDELTA of any mass in mass_list
- return 1 if yes, 0 if not
-------------------------------------------------------------------------- */
-
-int FixShakeCuda::masscheck(double massone)
-{
- for(int i = 0; i < nmass; i++)
- if(fabs(mass_list[i] - massone) <= MASSDELTA) return 1;
-
- return 0;
-}
-
-/* ----------------------------------------------------------------------
- update the unconstrained position of each atom
- only for SHAKE clusters, else set to 0.0
- assumes NVE update, seems to be accurate enough for NVT,NPT,NPH as well
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::unconstrained_update()
-{
- if(cuda->finished_setup) {
- Cuda_FixShakeCuda_UnconstrainedUpdate(&cuda->shared_data);
- return;
- }
-
- double dtfmsq;
-
- if(rmass) {
- for(int i = 0; i < nlocal; i++) {
- if(shake_flag[i]) {
- dtfmsq = dtfsq / rmass[i];
- xshake[i][0] = x[i][0] + dtv * v[i][0] + dtfmsq * f[i][0];
- xshake[i][1] = x[i][1] + dtv * v[i][1] + dtfmsq * f[i][1];
- xshake[i][2] = x[i][2] + dtv * v[i][2] + dtfmsq * f[i][2];
- } else xshake[i][2] = xshake[i][1] = xshake[i][0] = 0.0;
- }
- } else {
- for(int i = 0; i < nlocal; i++) {
- if(shake_flag[i]) {
- dtfmsq = dtfsq / mass[type[i]];
- xshake[i][0] = x[i][0] + dtv * v[i][0] + dtfmsq * f[i][0];
- xshake[i][1] = x[i][1] + dtv * v[i][1] + dtfmsq * f[i][1];
- xshake[i][2] = x[i][2] + dtv * v[i][2] + dtfmsq * f[i][2];
- } else xshake[i][2] = xshake[i][1] = xshake[i][0] = 0.0;
- }
- }
-
- cu_xshake->upload();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixShakeCuda::shake2(int m)
-{
- int nlist, list[2];
- double v[6];
- double invmass0, invmass1;
-
- // local atom IDs and constraint distances
-
- int i0 = atom->map(shake_atom[m][0]);
- int i1 = atom->map(shake_atom[m][1]);
- double bond1 = bond_distance[shake_type[m][0]];
-
- // r01 = distance vec between atoms, with PBC
-
- double r01[3];
- r01[0] = x[i0][0] - x[i1][0];
- r01[1] = x[i0][1] - x[i1][1];
- r01[2] = x[i0][2] - x[i1][2];
- domain->minimum_image(r01);
-
- // s01 = distance vec after unconstrained update, with PBC
-
- double s01[3];
- s01[0] = xshake[i0][0] - xshake[i1][0];
- s01[1] = xshake[i0][1] - xshake[i1][1];
- s01[2] = xshake[i0][2] - xshake[i1][2];
- domain->minimum_image(s01);
-
- // scalar distances between atoms
-
- double r01sq = r01[0] * r01[0] + r01[1] * r01[1] + r01[2] * r01[2];
- double s01sq = s01[0] * s01[0] + s01[1] * s01[1] + s01[2] * s01[2];
-
- // a,b,c = coeffs in quadratic equation for lamda
-
- if(rmass) {
- invmass0 = 1.0 / rmass[i0];
- invmass1 = 1.0 / rmass[i1];
- } else {
- invmass0 = 1.0 / mass[type[i0]];
- invmass1 = 1.0 / mass[type[i1]];
- }
-
- double a = (invmass0 + invmass1) * (invmass0 + invmass1) * r01sq;
- double b = 2.0 * (invmass0 + invmass1) *
- (s01[0] * r01[0] + s01[1] * r01[1] + s01[2] * r01[2]);
- double c = s01sq - bond1 * bond1;
-
- // error check
-
- double determ = b * b - 4.0 * a * c;
-
- if(determ < 0.0) {
- error->warning(FLERR, "Shake determinant < 0.0");
- determ = 0.0;
- }
-
- // exact quadratic solution for lamda
-
- double lamda, lamda1, lamda2;
- lamda1 = (-b + sqrt(determ)) / (2.0 * a);
- lamda2 = (-b - sqrt(determ)) / (2.0 * a);
-
- if(fabs(lamda1) <= fabs(lamda2)) lamda = lamda1;
- else lamda = lamda2;
-
- // update forces if atom is owned by this processor
- lamda /= dtfsq;
-
- if(i0 < nlocal) {
- f[i0][0] += lamda * r01[0];
- f[i0][1] += lamda * r01[1];
- f[i0][2] += lamda * r01[2];
- }
-
- if(i1 < nlocal) {
- f[i1][0] -= lamda * r01[0];
- f[i1][1] -= lamda * r01[1];
- f[i1][2] -= lamda * r01[2];
- }
-
- if(evflag) {
- nlist = 0;
-
- if(i0 < nlocal) list[nlist++] = i0;
-
- if(i1 < nlocal) list[nlist++] = i1;
-
- v[0] = lamda * r01[0] * r01[0];
- v[1] = lamda * r01[1] * r01[1];
- v[2] = lamda * r01[2] * r01[2];
- v[3] = lamda * r01[0] * r01[1];
- v[4] = lamda * r01[0] * r01[2];
- v[5] = lamda * r01[1] * r01[2];
-
- v_tally(nlist, list, 2.0, v);
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixShakeCuda::shake3(int m)
-{
- int nlist, list[3];
- double v[6];
- double invmass0, invmass1, invmass2;
-
- // local atom IDs and constraint distances
-
- int i0 = atom->map(shake_atom[m][0]);
- int i1 = atom->map(shake_atom[m][1]);
- int i2 = atom->map(shake_atom[m][2]);
- double bond1 = bond_distance[shake_type[m][0]];
- double bond2 = bond_distance[shake_type[m][1]];
-
- // r01,r02 = distance vec between atoms, with PBC
-
- double r01[3];
- r01[0] = x[i0][0] - x[i1][0];
- r01[1] = x[i0][1] - x[i1][1];
- r01[2] = x[i0][2] - x[i1][2];
- domain->minimum_image(r01);
-
- double r02[3];
- r02[0] = x[i0][0] - x[i2][0];
- r02[1] = x[i0][1] - x[i2][1];
- r02[2] = x[i0][2] - x[i2][2];
- domain->minimum_image(r02);
-
- // s01,s02 = distance vec after unconstrained update, with PBC
-
- double s01[3];
- s01[0] = xshake[i0][0] - xshake[i1][0];
- s01[1] = xshake[i0][1] - xshake[i1][1];
- s01[2] = xshake[i0][2] - xshake[i1][2];
- domain->minimum_image(s01);
-
- double s02[3];
- s02[0] = xshake[i0][0] - xshake[i2][0];
- s02[1] = xshake[i0][1] - xshake[i2][1];
- s02[2] = xshake[i0][2] - xshake[i2][2];
- domain->minimum_image(s02);
-
- // scalar distances between atoms
-
- double r01sq = r01[0] * r01[0] + r01[1] * r01[1] + r01[2] * r01[2];
- double r02sq = r02[0] * r02[0] + r02[1] * r02[1] + r02[2] * r02[2];
- double s01sq = s01[0] * s01[0] + s01[1] * s01[1] + s01[2] * s01[2];
- double s02sq = s02[0] * s02[0] + s02[1] * s02[1] + s02[2] * s02[2];
-
- // matrix coeffs and rhs for lamda equations
-
- if(rmass) {
- invmass0 = 1.0 / rmass[i0];
- invmass1 = 1.0 / rmass[i1];
- invmass2 = 1.0 / rmass[i2];
- } else {
- invmass0 = 1.0 / mass[type[i0]];
- invmass1 = 1.0 / mass[type[i1]];
- invmass2 = 1.0 / mass[type[i2]];
- }
-
- double a11 = 2.0 * (invmass0 + invmass1) *
- (s01[0] * r01[0] + s01[1] * r01[1] + s01[2] * r01[2]);
- double a12 = 2.0 * invmass0 *
- (s01[0] * r02[0] + s01[1] * r02[1] + s01[2] * r02[2]);
- double a21 = 2.0 * invmass0 *
- (s02[0] * r01[0] + s02[1] * r01[1] + s02[2] * r01[2]);
- double a22 = 2.0 * (invmass0 + invmass2) *
- (s02[0] * r02[0] + s02[1] * r02[1] + s02[2] * r02[2]);
-
- // inverse of matrix
-
- double determ = a11 * a22 - a12 * a21;
-
- if(determ == 0.0) error->one(FLERR, "Shake determinant = 0.0");
-
- double determinv = 1.0 / determ;
-
- double a11inv = a22 * determinv;
- double a12inv = -a12 * determinv;
- double a21inv = -a21 * determinv;
- double a22inv = a11 * determinv;
-
- // quadratic correction coeffs
-
- double r0102 = (r01[0] * r02[0] + r01[1] * r02[1] + r01[2] * r02[2]);
-
- double quad1_0101 = (invmass0 + invmass1) * (invmass0 + invmass1) * r01sq;
- double quad1_0202 = invmass0 * invmass0 * r02sq;
- double quad1_0102 = 2.0 * (invmass0 + invmass1) * invmass0 * r0102;
-
- double quad2_0202 = (invmass0 + invmass2) * (invmass0 + invmass2) * r02sq;
- double quad2_0101 = invmass0 * invmass0 * r01sq;
- double quad2_0102 = 2.0 * (invmass0 + invmass2) * invmass0 * r0102;
-
- // iterate until converged
-
- double lamda01 = 0.0;
- double lamda02 = 0.0;
- int niter = 0;
- int done = 0;
-
- double quad1, quad2, b1, b2, lamda01_new, lamda02_new;
-
- while(!done && niter < max_iter) {
- quad1 = quad1_0101 * lamda01 * lamda01 + quad1_0202 * lamda02 * lamda02 +
- quad1_0102 * lamda01 * lamda02;
- quad2 = quad2_0101 * lamda01 * lamda01 + quad2_0202 * lamda02 * lamda02 +
- quad2_0102 * lamda01 * lamda02;
-
- b1 = bond1 * bond1 - s01sq - quad1;
- b2 = bond2 * bond2 - s02sq - quad2;
-
- lamda01_new = a11inv * b1 + a12inv * b2;
- lamda02_new = a21inv * b1 + a22inv * b2;
-
- done = 1;
-
- if(fabs(lamda01_new - lamda01) > tolerance) done = 0;
-
- if(fabs(lamda02_new - lamda02) > tolerance) done = 0;
-
- lamda01 = lamda01_new;
- lamda02 = lamda02_new;
- niter++;
- }
-
- // update forces if atom is owned by this processor
-
- lamda01 = lamda01 / dtfsq;
- lamda02 = lamda02 / dtfsq;
-
- if(i0 < nlocal) {
- f[i0][0] += lamda01 * r01[0] + lamda02 * r02[0];
- f[i0][1] += lamda01 * r01[1] + lamda02 * r02[1];
- f[i0][2] += lamda01 * r01[2] + lamda02 * r02[2];
- }
-
- if(i1 < nlocal) {
- f[i1][0] -= lamda01 * r01[0];
- f[i1][1] -= lamda01 * r01[1];
- f[i1][2] -= lamda01 * r01[2];
- }
-
- if(i2 < nlocal) {
- f[i2][0] -= lamda02 * r02[0];
- f[i2][1] -= lamda02 * r02[1];
- f[i2][2] -= lamda02 * r02[2];
- }
-
- if(evflag) {
- nlist = 0;
-
- if(i0 < nlocal) list[nlist++] = i0;
-
- if(i1 < nlocal) list[nlist++] = i1;
-
- if(i2 < nlocal) list[nlist++] = i2;
-
- v[0] = lamda01 * r01[0] * r01[0] + lamda02 * r02[0] * r02[0];
- v[1] = lamda01 * r01[1] * r01[1] + lamda02 * r02[1] * r02[1];
- v[2] = lamda01 * r01[2] * r01[2] + lamda02 * r02[2] * r02[2];
- v[3] = lamda01 * r01[0] * r01[1] + lamda02 * r02[0] * r02[1];
- v[4] = lamda01 * r01[0] * r01[2] + lamda02 * r02[0] * r02[2];
- v[5] = lamda01 * r01[1] * r01[2] + lamda02 * r02[1] * r02[2];
-
- v_tally(nlist, list, 3.0, v);
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixShakeCuda::shake4(int m)
-{
- int nlist, list[4];
- double v[6];
- double invmass0, invmass1, invmass2, invmass3;
-
- // local atom IDs and constraint distances
-
- int i0 = atom->map(shake_atom[m][0]);
- int i1 = atom->map(shake_atom[m][1]);
- int i2 = atom->map(shake_atom[m][2]);
- int i3 = atom->map(shake_atom[m][3]);
- double bond1 = bond_distance[shake_type[m][0]];
- double bond2 = bond_distance[shake_type[m][1]];
- double bond3 = bond_distance[shake_type[m][2]];
-
- // r01,r02,r03 = distance vec between atoms, with PBC
-
- double r01[3];
- r01[0] = x[i0][0] - x[i1][0];
- r01[1] = x[i0][1] - x[i1][1];
- r01[2] = x[i0][2] - x[i1][2];
- domain->minimum_image(r01);
-
- double r02[3];
- r02[0] = x[i0][0] - x[i2][0];
- r02[1] = x[i0][1] - x[i2][1];
- r02[2] = x[i0][2] - x[i2][2];
- domain->minimum_image(r02);
-
- double r03[3];
- r03[0] = x[i0][0] - x[i3][0];
- r03[1] = x[i0][1] - x[i3][1];
- r03[2] = x[i0][2] - x[i3][2];
- domain->minimum_image(r03);
-
- // s01,s02,s03 = distance vec after unconstrained update, with PBC
-
- double s01[3];
- s01[0] = xshake[i0][0] - xshake[i1][0];
- s01[1] = xshake[i0][1] - xshake[i1][1];
- s01[2] = xshake[i0][2] - xshake[i1][2];
- domain->minimum_image(s01);
-
- double s02[3];
- s02[0] = xshake[i0][0] - xshake[i2][0];
- s02[1] = xshake[i0][1] - xshake[i2][1];
- s02[2] = xshake[i0][2] - xshake[i2][2];
- domain->minimum_image(s02);
-
- double s03[3];
- s03[0] = xshake[i0][0] - xshake[i3][0];
- s03[1] = xshake[i0][1] - xshake[i3][1];
- s03[2] = xshake[i0][2] - xshake[i3][2];
- domain->minimum_image(s03);
-
- // scalar distances between atoms
-
- double r01sq = r01[0] * r01[0] + r01[1] * r01[1] + r01[2] * r01[2];
- double r02sq = r02[0] * r02[0] + r02[1] * r02[1] + r02[2] * r02[2];
- double r03sq = r03[0] * r03[0] + r03[1] * r03[1] + r03[2] * r03[2];
- double s01sq = s01[0] * s01[0] + s01[1] * s01[1] + s01[2] * s01[2];
- double s02sq = s02[0] * s02[0] + s02[1] * s02[1] + s02[2] * s02[2];
- double s03sq = s03[0] * s03[0] + s03[1] * s03[1] + s03[2] * s03[2];
-
- // matrix coeffs and rhs for lamda equations
-
- if(rmass) {
- invmass0 = 1.0 / rmass[i0];
- invmass1 = 1.0 / rmass[i1];
- invmass2 = 1.0 / rmass[i2];
- invmass3 = 1.0 / rmass[i3];
- } else {
- invmass0 = 1.0 / mass[type[i0]];
- invmass1 = 1.0 / mass[type[i1]];
- invmass2 = 1.0 / mass[type[i2]];
- invmass3 = 1.0 / mass[type[i3]];
- }
-
- double a11 = 2.0 * (invmass0 + invmass1) *
- (s01[0] * r01[0] + s01[1] * r01[1] + s01[2] * r01[2]);
- double a12 = 2.0 * invmass0 *
- (s01[0] * r02[0] + s01[1] * r02[1] + s01[2] * r02[2]);
- double a13 = 2.0 * invmass0 *
- (s01[0] * r03[0] + s01[1] * r03[1] + s01[2] * r03[2]);
- double a21 = 2.0 * invmass0 *
- (s02[0] * r01[0] + s02[1] * r01[1] + s02[2] * r01[2]);
- double a22 = 2.0 * (invmass0 + invmass2) *
- (s02[0] * r02[0] + s02[1] * r02[1] + s02[2] * r02[2]);
- double a23 = 2.0 * invmass0 *
- (s02[0] * r03[0] + s02[1] * r03[1] + s02[2] * r03[2]);
- double a31 = 2.0 * invmass0 *
- (s03[0] * r01[0] + s03[1] * r01[1] + s03[2] * r01[2]);
- double a32 = 2.0 * invmass0 *
- (s03[0] * r02[0] + s03[1] * r02[1] + s03[2] * r02[2]);
- double a33 = 2.0 * (invmass0 + invmass3) *
- (s03[0] * r03[0] + s03[1] * r03[1] + s03[2] * r03[2]);
-
- // inverse of matrix;
-
- double determ = a11 * a22 * a33 + a12 * a23 * a31 + a13 * a21 * a32 -
- a11 * a23 * a32 - a12 * a21 * a33 - a13 * a22 * a31;
-
- if(determ == 0.0) error->one(FLERR, "Shake determinant = 0.0");
-
- double determinv = 1.0 / determ;
-
- double a11inv = determinv * (a22 * a33 - a23 * a32);
- double a12inv = -determinv * (a12 * a33 - a13 * a32);
- double a13inv = determinv * (a12 * a23 - a13 * a22);
- double a21inv = -determinv * (a21 * a33 - a23 * a31);
- double a22inv = determinv * (a11 * a33 - a13 * a31);
- double a23inv = -determinv * (a11 * a23 - a13 * a21);
- double a31inv = determinv * (a21 * a32 - a22 * a31);
- double a32inv = -determinv * (a11 * a32 - a12 * a31);
- double a33inv = determinv * (a11 * a22 - a12 * a21);
-
- // quadratic correction coeffs
-
- double r0102 = (r01[0] * r02[0] + r01[1] * r02[1] + r01[2] * r02[2]);
- double r0103 = (r01[0] * r03[0] + r01[1] * r03[1] + r01[2] * r03[2]);
- double r0203 = (r02[0] * r03[0] + r02[1] * r03[1] + r02[2] * r03[2]);
-
- double quad1_0101 = (invmass0 + invmass1) * (invmass0 + invmass1) * r01sq;
- double quad1_0202 = invmass0 * invmass0 * r02sq;
- double quad1_0303 = invmass0 * invmass0 * r03sq;
- double quad1_0102 = 2.0 * (invmass0 + invmass1) * invmass0 * r0102;
- double quad1_0103 = 2.0 * (invmass0 + invmass1) * invmass0 * r0103;
- double quad1_0203 = 2.0 * invmass0 * invmass0 * r0203;
-
- double quad2_0101 = invmass0 * invmass0 * r01sq;
- double quad2_0202 = (invmass0 + invmass2) * (invmass0 + invmass2) * r02sq;
- double quad2_0303 = invmass0 * invmass0 * r03sq;
- double quad2_0102 = 2.0 * (invmass0 + invmass2) * invmass0 * r0102;
- double quad2_0103 = 2.0 * invmass0 * invmass0 * r0103;
- double quad2_0203 = 2.0 * (invmass0 + invmass2) * invmass0 * r0203;
-
- double quad3_0101 = invmass0 * invmass0 * r01sq;
- double quad3_0202 = invmass0 * invmass0 * r02sq;
- double quad3_0303 = (invmass0 + invmass3) * (invmass0 + invmass3) * r03sq;
- double quad3_0102 = 2.0 * invmass0 * invmass0 * r0102;
- double quad3_0103 = 2.0 * (invmass0 + invmass3) * invmass0 * r0103;
- double quad3_0203 = 2.0 * (invmass0 + invmass3) * invmass0 * r0203;
-
- // iterate until converged
-
- double lamda01 = 0.0;
- double lamda02 = 0.0;
- double lamda03 = 0.0;
- int niter = 0;
- int done = 0;
-
- double quad1, quad2, quad3, b1, b2, b3, lamda01_new, lamda02_new, lamda03_new;
-
- while(!done && niter < max_iter) {
- quad1 = quad1_0101 * lamda01 * lamda01 +
- quad1_0202 * lamda02 * lamda02 +
- quad1_0303 * lamda03 * lamda03 +
- quad1_0102 * lamda01 * lamda02 +
- quad1_0103 * lamda01 * lamda03 +
- quad1_0203 * lamda02 * lamda03;
-
- quad2 = quad2_0101 * lamda01 * lamda01 +
- quad2_0202 * lamda02 * lamda02 +
- quad2_0303 * lamda03 * lamda03 +
- quad2_0102 * lamda01 * lamda02 +
- quad2_0103 * lamda01 * lamda03 +
- quad2_0203 * lamda02 * lamda03;
-
- quad3 = quad3_0101 * lamda01 * lamda01 +
- quad3_0202 * lamda02 * lamda02 +
- quad3_0303 * lamda03 * lamda03 +
- quad3_0102 * lamda01 * lamda02 +
- quad3_0103 * lamda01 * lamda03 +
- quad3_0203 * lamda02 * lamda03;
-
- b1 = bond1 * bond1 - s01sq - quad1;
- b2 = bond2 * bond2 - s02sq - quad2;
- b3 = bond3 * bond3 - s03sq - quad3;
-
- lamda01_new = a11inv * b1 + a12inv * b2 + a13inv * b3;
- lamda02_new = a21inv * b1 + a22inv * b2 + a23inv * b3;
- lamda03_new = a31inv * b1 + a32inv * b2 + a33inv * b3;
-
- done = 1;
-
- if(fabs(lamda01_new - lamda01) > tolerance) done = 0;
-
- if(fabs(lamda02_new - lamda02) > tolerance) done = 0;
-
- if(fabs(lamda03_new - lamda03) > tolerance) done = 0;
-
- lamda01 = lamda01_new;
- lamda02 = lamda02_new;
- lamda03 = lamda03_new;
- niter++;
- }
-
- // update forces if atom is owned by this processor
-
- lamda01 = lamda01 / dtfsq;
- lamda02 = lamda02 / dtfsq;
- lamda03 = lamda03 / dtfsq;
-
- if(i0 < nlocal) {
- f[i0][0] += lamda01 * r01[0] + lamda02 * r02[0] + lamda03 * r03[0];
- f[i0][1] += lamda01 * r01[1] + lamda02 * r02[1] + lamda03 * r03[1];
- f[i0][2] += lamda01 * r01[2] + lamda02 * r02[2] + lamda03 * r03[2];
- }
-
- if(i1 < nlocal) {
- f[i1][0] -= lamda01 * r01[0];
- f[i1][1] -= lamda01 * r01[1];
- f[i1][2] -= lamda01 * r01[2];
- }
-
- if(i2 < nlocal) {
- f[i2][0] -= lamda02 * r02[0];
- f[i2][1] -= lamda02 * r02[1];
- f[i2][2] -= lamda02 * r02[2];
- }
-
- if(i3 < nlocal) {
- f[i3][0] -= lamda03 * r03[0];
- f[i3][1] -= lamda03 * r03[1];
- f[i3][2] -= lamda03 * r03[2];
- }
-
- if(evflag) {
- nlist = 0;
-
- if(i0 < nlocal) list[nlist++] = i0;
-
- if(i1 < nlocal) list[nlist++] = i1;
-
- if(i2 < nlocal) list[nlist++] = i2;
-
- if(i3 < nlocal) list[nlist++] = i3;
-
- v[0] = lamda01 * r01[0] * r01[0] + lamda02 * r02[0] * r02[0] + lamda03 * r03[0] * r03[0];
- v[1] = lamda01 * r01[1] * r01[1] + lamda02 * r02[1] * r02[1] + lamda03 * r03[1] * r03[1];
- v[2] = lamda01 * r01[2] * r01[2] + lamda02 * r02[2] * r02[2] + lamda03 * r03[2] * r03[2];
- v[3] = lamda01 * r01[0] * r01[1] + lamda02 * r02[0] * r02[1] + lamda03 * r03[0] * r03[1];
- v[4] = lamda01 * r01[0] * r01[2] + lamda02 * r02[0] * r02[2] + lamda03 * r03[0] * r03[2];
- v[5] = lamda01 * r01[1] * r01[2] + lamda02 * r02[1] * r02[2] + lamda03 * r03[1] * r03[2];
- //if(i0==7271) printf("%lf %lf %lf %lf %lf %lf\n",v[0],v[1],v[2],v[3],v[4],v[5]);
-
- v_tally(nlist, list, 4.0, v);
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixShakeCuda::shake3angle(int m)
-{
- int nlist, list[3];
- double v[6];
- double invmass0, invmass1, invmass2;
-
- // local atom IDs and constraint distances
-
- int i0 = atom->map(shake_atom[m][0]);
- int i1 = atom->map(shake_atom[m][1]);
- int i2 = atom->map(shake_atom[m][2]);
- double bond1 = bond_distance[shake_type[m][0]];
- double bond2 = bond_distance[shake_type[m][1]];
- double bond12 = angle_distance[shake_type[m][2]];
-
- // r01,r02,r12 = distance vec between atoms, with PBC
-
- double r01[3];
- r01[0] = x[i0][0] - x[i1][0];
- r01[1] = x[i0][1] - x[i1][1];
- r01[2] = x[i0][2] - x[i1][2];
- domain->minimum_image(r01);
-
- double r02[3];
- r02[0] = x[i0][0] - x[i2][0];
- r02[1] = x[i0][1] - x[i2][1];
- r02[2] = x[i0][2] - x[i2][2];
- domain->minimum_image(r02);
-
- double r12[3];
- r12[0] = x[i1][0] - x[i2][0];
- r12[1] = x[i1][1] - x[i2][1];
- r12[2] = x[i1][2] - x[i2][2];
- domain->minimum_image(r12);
-
- // s01,s02,s12 = distance vec after unconstrained update, with PBC
-
- double s01[3];
- s01[0] = xshake[i0][0] - xshake[i1][0];
- s01[1] = xshake[i0][1] - xshake[i1][1];
- s01[2] = xshake[i0][2] - xshake[i1][2];
- domain->minimum_image(s01);
-
- double s02[3];
- s02[0] = xshake[i0][0] - xshake[i2][0];
- s02[1] = xshake[i0][1] - xshake[i2][1];
- s02[2] = xshake[i0][2] - xshake[i2][2];
- domain->minimum_image(s02);
-
- double s12[3];
- s12[0] = xshake[i1][0] - xshake[i2][0];
- s12[1] = xshake[i1][1] - xshake[i2][1];
- s12[2] = xshake[i1][2] - xshake[i2][2];
- domain->minimum_image(s12);
-
- // scalar distances between atoms
-
- double r01sq = r01[0] * r01[0] + r01[1] * r01[1] + r01[2] * r01[2];
- double r02sq = r02[0] * r02[0] + r02[1] * r02[1] + r02[2] * r02[2];
- double r12sq = r12[0] * r12[0] + r12[1] * r12[1] + r12[2] * r12[2];
- double s01sq = s01[0] * s01[0] + s01[1] * s01[1] + s01[2] * s01[2];
- double s02sq = s02[0] * s02[0] + s02[1] * s02[1] + s02[2] * s02[2];
- double s12sq = s12[0] * s12[0] + s12[1] * s12[1] + s12[2] * s12[2];
-
- // matrix coeffs and rhs for lamda equations
-
- if(rmass) {
- invmass0 = 1.0 / rmass[i0];
- invmass1 = 1.0 / rmass[i1];
- invmass2 = 1.0 / rmass[i2];
- } else {
- invmass0 = 1.0 / mass[type[i0]];
- invmass1 = 1.0 / mass[type[i1]];
- invmass2 = 1.0 / mass[type[i2]];
- }
-
- double a11 = 2.0 * (invmass0 + invmass1) *
- (s01[0] * r01[0] + s01[1] * r01[1] + s01[2] * r01[2]);
- double a12 = 2.0 * invmass0 *
- (s01[0] * r02[0] + s01[1] * r02[1] + s01[2] * r02[2]);
- double a13 = - 2.0 * invmass1 *
- (s01[0] * r12[0] + s01[1] * r12[1] + s01[2] * r12[2]);
- double a21 = 2.0 * invmass0 *
- (s02[0] * r01[0] + s02[1] * r01[1] + s02[2] * r01[2]);
- double a22 = 2.0 * (invmass0 + invmass2) *
- (s02[0] * r02[0] + s02[1] * r02[1] + s02[2] * r02[2]);
- double a23 = 2.0 * invmass2 *
- (s02[0] * r12[0] + s02[1] * r12[1] + s02[2] * r12[2]);
- double a31 = - 2.0 * invmass1 *
- (s12[0] * r01[0] + s12[1] * r01[1] + s12[2] * r01[2]);
- double a32 = 2.0 * invmass2 *
- (s12[0] * r02[0] + s12[1] * r02[1] + s12[2] * r02[2]);
- double a33 = 2.0 * (invmass1 + invmass2) *
- (s12[0] * r12[0] + s12[1] * r12[1] + s12[2] * r12[2]);
-
- // inverse of matrix
-
- double determ = a11 * a22 * a33 + a12 * a23 * a31 + a13 * a21 * a32 -
- a11 * a23 * a32 - a12 * a21 * a33 - a13 * a22 * a31;
-
- if(determ == 0.0) error->one(FLERR, "Shake determinant = 0.0");
-
- double determinv = 1.0 / determ;
-
- double a11inv = determinv * (a22 * a33 - a23 * a32);
- double a12inv = -determinv * (a12 * a33 - a13 * a32);
- double a13inv = determinv * (a12 * a23 - a13 * a22);
- double a21inv = -determinv * (a21 * a33 - a23 * a31);
- double a22inv = determinv * (a11 * a33 - a13 * a31);
- double a23inv = -determinv * (a11 * a23 - a13 * a21);
- double a31inv = determinv * (a21 * a32 - a22 * a31);
- double a32inv = -determinv * (a11 * a32 - a12 * a31);
- double a33inv = determinv * (a11 * a22 - a12 * a21);
-
- // quadratic correction coeffs
-
- double r0102 = (r01[0] * r02[0] + r01[1] * r02[1] + r01[2] * r02[2]);
- double r0112 = (r01[0] * r12[0] + r01[1] * r12[1] + r01[2] * r12[2]);
- double r0212 = (r02[0] * r12[0] + r02[1] * r12[1] + r02[2] * r12[2]);
-
- double quad1_0101 = (invmass0 + invmass1) * (invmass0 + invmass1) * r01sq;
- double quad1_0202 = invmass0 * invmass0 * r02sq;
- double quad1_1212 = invmass1 * invmass1 * r12sq;
- double quad1_0102 = 2.0 * (invmass0 + invmass1) * invmass0 * r0102;
- double quad1_0112 = - 2.0 * (invmass0 + invmass1) * invmass1 * r0112;
- double quad1_0212 = - 2.0 * invmass0 * invmass1 * r0212;
-
- double quad2_0101 = invmass0 * invmass0 * r01sq;
- double quad2_0202 = (invmass0 + invmass2) * (invmass0 + invmass2) * r02sq;
- double quad2_1212 = invmass2 * invmass2 * r12sq;
- double quad2_0102 = 2.0 * (invmass0 + invmass2) * invmass0 * r0102;
- double quad2_0112 = 2.0 * invmass0 * invmass2 * r0112;
- double quad2_0212 = 2.0 * (invmass0 + invmass2) * invmass2 * r0212;
-
- double quad3_0101 = invmass1 * invmass1 * r01sq;
- double quad3_0202 = invmass2 * invmass2 * r02sq;
- double quad3_1212 = (invmass1 + invmass2) * (invmass1 + invmass2) * r12sq;
- double quad3_0102 = - 2.0 * invmass1 * invmass2 * r0102;
- double quad3_0112 = - 2.0 * (invmass1 + invmass2) * invmass1 * r0112;
- double quad3_0212 = 2.0 * (invmass1 + invmass2) * invmass2 * r0212;
-
- // iterate until converged
-
- double lamda01 = 0.0;
- double lamda02 = 0.0;
- double lamda12 = 0.0;
- int niter = 0;
- int done = 0;
-
- double quad1, quad2, quad3, b1, b2, b3, lamda01_new, lamda02_new, lamda12_new;
-
- while(!done && niter < max_iter) {
- quad1 = quad1_0101 * lamda01 * lamda01 +
- quad1_0202 * lamda02 * lamda02 +
- quad1_1212 * lamda12 * lamda12 +
- quad1_0102 * lamda01 * lamda02 +
- quad1_0112 * lamda01 * lamda12 +
- quad1_0212 * lamda02 * lamda12;
-
- quad2 = quad2_0101 * lamda01 * lamda01 +
- quad2_0202 * lamda02 * lamda02 +
- quad2_1212 * lamda12 * lamda12 +
- quad2_0102 * lamda01 * lamda02 +
- quad2_0112 * lamda01 * lamda12 +
- quad2_0212 * lamda02 * lamda12;
-
- quad3 = quad3_0101 * lamda01 * lamda01 +
- quad3_0202 * lamda02 * lamda02 +
- quad3_1212 * lamda12 * lamda12 +
- quad3_0102 * lamda01 * lamda02 +
- quad3_0112 * lamda01 * lamda12 +
- quad3_0212 * lamda02 * lamda12;
-
- b1 = bond1 * bond1 - s01sq - quad1;
- b2 = bond2 * bond2 - s02sq - quad2;
- b3 = bond12 * bond12 - s12sq - quad3;
-
- lamda01_new = a11inv * b1 + a12inv * b2 + a13inv * b3;
- lamda02_new = a21inv * b1 + a22inv * b2 + a23inv * b3;
- lamda12_new = a31inv * b1 + a32inv * b2 + a33inv * b3;
-
- done = 1;
-
- if(fabs(lamda01_new - lamda01) > tolerance) done = 0;
-
- if(fabs(lamda02_new - lamda02) > tolerance) done = 0;
-
- if(fabs(lamda12_new - lamda12) > tolerance) done = 0;
-
- lamda01 = lamda01_new;
- lamda02 = lamda02_new;
- lamda12 = lamda12_new;
- niter++;
- }
-
- // update forces if atom is owned by this processor
-
- lamda01 = lamda01 / dtfsq;
- lamda02 = lamda02 / dtfsq;
- lamda12 = lamda12 / dtfsq;
-
- if(i0 < nlocal) {
- f[i0][0] += lamda01 * r01[0] + lamda02 * r02[0];
- f[i0][1] += lamda01 * r01[1] + lamda02 * r02[1];
- f[i0][2] += lamda01 * r01[2] + lamda02 * r02[2];
- }
-
- if(i1 < nlocal) {
- f[i1][0] -= lamda01 * r01[0] - lamda12 * r12[0];
- f[i1][1] -= lamda01 * r01[1] - lamda12 * r12[1];
- f[i1][2] -= lamda01 * r01[2] - lamda12 * r12[2];
- }
-
- if(i2 < nlocal) {
- f[i2][0] -= lamda02 * r02[0] + lamda12 * r12[0];
- f[i2][1] -= lamda02 * r02[1] + lamda12 * r12[1];
- f[i2][2] -= lamda02 * r02[2] + lamda12 * r12[2];
- }
-
- if(evflag) {
- nlist = 0;
-
- if(i0 < nlocal) list[nlist++] = i0;
-
- if(i1 < nlocal) list[nlist++] = i1;
-
- if(i2 < nlocal) list[nlist++] = i2;
-
- v[0] = lamda01 * r01[0] * r01[0] + lamda02 * r02[0] * r02[0] + lamda12 * r12[0] * r12[0];
- v[1] = lamda01 * r01[1] * r01[1] + lamda02 * r02[1] * r02[1] + lamda12 * r12[1] * r12[1];
- v[2] = lamda01 * r01[2] * r01[2] + lamda02 * r02[2] * r02[2] + lamda12 * r12[2] * r12[2];
- v[3] = lamda01 * r01[0] * r01[1] + lamda02 * r02[0] * r02[1] + lamda12 * r12[0] * r12[1];
- v[4] = lamda01 * r01[0] * r01[2] + lamda02 * r02[0] * r02[2] + lamda12 * r12[0] * r12[2];
- v[5] = lamda01 * r01[1] * r01[2] + lamda02 * r02[1] * r02[2] + lamda12 * r12[1] * r12[2];
-
- v_tally(nlist, list, 3.0, v);
- }
-}
-
-/* ----------------------------------------------------------------------
- print-out bond & angle statistics
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::stats()
-{
- int i, j, m, n, iatom, jatom, katom;
- double delx, dely, delz;
- double r, r1, r2, r3, angle;
-
- // zero out accumulators
-
- int nb = atom->nbondtypes + 1;
- int na = atom->nangletypes + 1;
-
- for(i = 0; i < nb; i++) {
- b_count[i] = 0;
- b_ave[i] = b_max[i] = 0.0;
- b_min[i] = BIG;
- }
-
- for(i = 0; i < na; i++) {
- a_count[i] = 0;
- a_ave[i] = a_max[i] = 0.0;
- a_min[i] = BIG;
- }
-
- // log stats for each bond & angle
- // OK to double count since are just averaging
-
- double** x = atom->x;
- int nlocal = atom->nlocal;
-
- for(i = 0; i < nlocal; i++) {
- if(shake_flag[i] == 0) continue;
-
- // bond stats
-
- n = shake_flag[i];
-
- if(n == 1) n = 3;
-
- iatom = atom->map(shake_atom[i][0]);
-
- for(j = 1; j < n; j++) {
- jatom = atom->map(shake_atom[i][j]);
- delx = x[iatom][0] - x[jatom][0];
- dely = x[iatom][1] - x[jatom][1];
- delz = x[iatom][2] - x[jatom][2];
- domain->minimum_image(delx, dely, delz);
- r = sqrt(delx * delx + dely * dely + delz * delz);
-
- m = shake_type[i][j - 1];
- b_count[m]++;
- b_ave[m] += r;
- b_max[m] = MAX(b_max[m], r);
- b_min[m] = MIN(b_min[m], r);
- }
-
- // angle stats
-
- if(shake_flag[i] == 1) {
- iatom = atom->map(shake_atom[i][0]);
- jatom = atom->map(shake_atom[i][1]);
- katom = atom->map(shake_atom[i][2]);
-
- delx = x[iatom][0] - x[jatom][0];
- dely = x[iatom][1] - x[jatom][1];
- delz = x[iatom][2] - x[jatom][2];
- domain->minimum_image(delx, dely, delz);
- r1 = sqrt(delx * delx + dely * dely + delz * delz);
-
- delx = x[iatom][0] - x[katom][0];
- dely = x[iatom][1] - x[katom][1];
- delz = x[iatom][2] - x[katom][2];
- domain->minimum_image(delx, dely, delz);
- r2 = sqrt(delx * delx + dely * dely + delz * delz);
-
- delx = x[jatom][0] - x[katom][0];
- dely = x[jatom][1] - x[katom][1];
- delz = x[jatom][2] - x[katom][2];
- domain->minimum_image(delx, dely, delz);
- r3 = sqrt(delx * delx + dely * dely + delz * delz);
-
- angle = acos((r1 * r1 + r2 * r2 - r3 * r3) / (2.0 * r1 * r2));
- angle *= 180.0 / MY_PI;
- m = shake_type[i][2];
- a_count[m]++;
- a_ave[m] += angle;
- a_max[m] = MAX(a_max[m], angle);
- a_min[m] = MIN(a_min[m], angle);
- }
- }
-
- // sum across all procs
-
- MPI_Allreduce(b_count, b_count_all, nb, MPI_INT, MPI_SUM, world);
- MPI_Allreduce(b_ave, b_ave_all, nb, MPI_DOUBLE, MPI_SUM, world);
- MPI_Allreduce(b_max, b_max_all, nb, MPI_DOUBLE, MPI_MAX, world);
- MPI_Allreduce(b_min, b_min_all, nb, MPI_DOUBLE, MPI_MIN, world);
-
- MPI_Allreduce(a_count, a_count_all, na, MPI_INT, MPI_SUM, world);
- MPI_Allreduce(a_ave, a_ave_all, na, MPI_DOUBLE, MPI_SUM, world);
- MPI_Allreduce(a_max, a_max_all, na, MPI_DOUBLE, MPI_MAX, world);
- MPI_Allreduce(a_min, a_min_all, na, MPI_DOUBLE, MPI_MIN, world);
-
- // print stats only for non-zero counts
-
- if(me == 0) {
- if(screen) {
- fprintf(screen,
- "SHAKE stats (type/ave/delta) on step " BIGINT_FORMAT "\n",
- update->ntimestep);
-
- for(i = 1; i < nb; i++)
- if(b_count_all[i])
- fprintf(screen, " %d %g %g\n", i,
- b_ave_all[i] / b_count_all[i], b_max_all[i] - b_min_all[i]);
-
- for(i = 1; i < na; i++)
- if(a_count_all[i])
- fprintf(screen, " %d %g %g\n", i,
- a_ave_all[i] / a_count_all[i], a_max_all[i] - a_min_all[i]);
- }
-
- if(logfile) {
- fprintf(logfile,
- "SHAKE stats (type/ave/delta) on step " BIGINT_FORMAT "\n",
- update->ntimestep);
-
- for(i = 0; i < nb; i++)
- if(b_count_all[i])
- fprintf(logfile, " %d %g %g\n", i,
- b_ave_all[i] / b_count_all[i], b_max_all[i] - b_min_all[i]);
-
- for(i = 0; i < na; i++)
- if(a_count_all[i])
- fprintf(logfile, " %d %g %g\n", i,
- a_ave_all[i] / a_count_all[i], a_max_all[i] - a_min_all[i]);
- }
- }
-
- // next timestep for stats
-
- next_output += output_every;
-}
-
-/* ----------------------------------------------------------------------
- find a bond between global tags n1 and n2 stored with local atom i
- return -1 if don't find it
- return bond index if do find it
-------------------------------------------------------------------------- */
-
-int FixShakeCuda::bondfind(int i, int n1, int n2)
-{
- int* tag = atom->tag;
- int** bond_atom = atom->bond_atom;
- int nbonds = atom->num_bond[i];
-
- int m;
-
- for(m = 0; m < nbonds; m++) {
- if(n1 == tag[i] && n2 == bond_atom[i][m]) break;
-
- if(n1 == bond_atom[i][m] && n2 == tag[i]) break;
- }
-
- if(m < nbonds) return m;
-
- return -1;
-}
-
-/* ----------------------------------------------------------------------
- find an angle with global end atoms n1 and n2 stored with local atom i
- return -1 if don't find it
- return angle index if do find it
-------------------------------------------------------------------------- */
-
-int FixShakeCuda::anglefind(int i, int n1, int n2)
-{
- int** angle_atom1 = atom->angle_atom1;
- int** angle_atom3 = atom->angle_atom3;
- int nangles = atom->num_angle[i];
-
- int m;
-
- for(m = 0; m < nangles; m++) {
- if(n1 == angle_atom1[i][m] && n2 == angle_atom3[i][m]) break;
-
- if(n1 == angle_atom3[i][m] && n2 == angle_atom1[i][m]) break;
- }
-
- if(m < nangles) return m;
-
- return -1;
-}
-
-/* ----------------------------------------------------------------------
- memory usage of local atom-based arrays
-------------------------------------------------------------------------- */
-
-double FixShakeCuda::memory_usage()
-{
- int nmax = atom->nmax;
- double bytes = nmax * sizeof(int);
- bytes += nmax * 4 * sizeof(int);
- bytes += nmax * 3 * sizeof(int);
- bytes += nmax * 3 * sizeof(double);
- bytes += maxvatom * 6 * sizeof(double);
- return bytes;
-}
-
-/* ----------------------------------------------------------------------
- allocate local atom-based arrays
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::grow_arrays(int nmax)
-{
- memory->grow(shake_flag, nmax, "shake:shake_flag");
- memory->grow(shake_atom, nmax, 4, "shake:shake_atom");
- memory->grow(shake_type, nmax, 3, "shake:shake_type");
- memory->destroy(xshake);
- memory->create(xshake, nmax, 3, "shake:xshake");
-
- delete cu_shake_flag;
- cu_shake_flag = new cCudaData<int, int, xx > (shake_flag, nmax);
- delete cu_shake_atom;
- cu_shake_atom = new cCudaData<int, int, yx> ((int*)shake_atom, nmax, 4);
- delete cu_shake_type;
- cu_shake_type = new cCudaData<int, int, yx> ((int*)shake_type, nmax, 3);
- delete cu_xshake;
- cu_xshake = new cCudaData<double, X_CFLOAT, xy> ((double*)xshake, nmax, 3);
- cu_shake_flag->upload();
- cu_shake_atom->upload();
- cu_shake_type->upload();
-
- if(cu_bond_distance)
- Cuda_FixShakeCuda_Init(&cuda->shared_data, dtv, dtfsq,
- cu_shake_flag->dev_data(), cu_shake_atom->dev_data(), cu_shake_type->dev_data(), cu_xshake->dev_data(),
- cu_bond_distance->dev_data(), cu_angle_distance->dev_data(), cu_virial->dev_data(),
- max_iter, tolerance);
-}
-
-/* ----------------------------------------------------------------------
- copy values within local atom-based arrays
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::copy_arrays(int i, int j, int delflag)
-{
- int flag = shake_flag[j] = shake_flag[i];
-
- if(flag == 1) {
- shake_atom[j][0] = shake_atom[i][0];
- shake_atom[j][1] = shake_atom[i][1];
- shake_atom[j][2] = shake_atom[i][2];
- shake_type[j][0] = shake_type[i][0];
- shake_type[j][1] = shake_type[i][1];
- shake_type[j][2] = shake_type[i][2];
- } else if(flag == 2) {
- shake_atom[j][0] = shake_atom[i][0];
- shake_atom[j][1] = shake_atom[i][1];
- shake_type[j][0] = shake_type[i][0];
- } else if(flag == 3) {
- shake_atom[j][0] = shake_atom[i][0];
- shake_atom[j][1] = shake_atom[i][1];
- shake_atom[j][2] = shake_atom[i][2];
- shake_type[j][0] = shake_type[i][0];
- shake_type[j][1] = shake_type[i][1];
- } else if(flag == 4) {
- shake_atom[j][0] = shake_atom[i][0];
- shake_atom[j][1] = shake_atom[i][1];
- shake_atom[j][2] = shake_atom[i][2];
- shake_atom[j][3] = shake_atom[i][3];
- shake_type[j][0] = shake_type[i][0];
- shake_type[j][1] = shake_type[i][1];
- shake_type[j][2] = shake_type[i][2];
- }
-}
-
-/* ----------------------------------------------------------------------
- initialize one atom's array values, called when atom is created
-------------------------------------------------------------------------- */
-
-void FixShakeCuda::set_arrays(int i)
-{
- shake_flag[i] = 0;
-}
-
-/* ----------------------------------------------------------------------
- pack values in local atom-based arrays for exchange with another proc
-------------------------------------------------------------------------- */
-
-int FixShakeCuda::pack_exchange(int i, double* buf)
-{
- int m = 0;
- buf[m++] = shake_flag[i];
- int flag = shake_flag[i];
-
- if(flag == 1) {
- buf[m++] = shake_atom[i][0];
- buf[m++] = shake_atom[i][1];
- buf[m++] = shake_atom[i][2];
- buf[m++] = shake_type[i][0];
- buf[m++] = shake_type[i][1];
- buf[m++] = shake_type[i][2];
- } else if(flag == 2) {
- buf[m++] = shake_atom[i][0];
- buf[m++] = shake_atom[i][1];
- buf[m++] = shake_type[i][0];
- } else if(flag == 3) {
- buf[m++] = shake_atom[i][0];
- buf[m++] = shake_atom[i][1];
- buf[m++] = shake_atom[i][2];
- buf[m++] = shake_type[i][0];
- buf[m++] = shake_type[i][1];
- } else if(flag == 4) {
- buf[m++] = shake_atom[i][0];
- buf[m++] = shake_atom[i][1];
- buf[m++] = shake_atom[i][2];
- buf[m++] = shake_atom[i][3];
- buf[m++] = shake_type[i][0];
- buf[m++] = shake_type[i][1];
- buf[m++] = shake_type[i][2];
- }
-
- return m;
-}
-
-/* ----------------------------------------------------------------------
- unpack values in local atom-based arrays from exchange with another proc
-------------------------------------------------------------------------- */
-
-int FixShakeCuda::unpack_exchange(int nlocal, double* buf)
-{
- int m = 0;
- int flag = shake_flag[nlocal] = static_cast<int>(buf[m++]);
-
- if(flag == 1) {
- shake_atom[nlocal][0] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][1] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][2] = static_cast<int>(buf[m++]);
- shake_type[nlocal][0] = static_cast<int>(buf[m++]);
- shake_type[nlocal][1] = static_cast<int>(buf[m++]);
- shake_type[nlocal][2] = static_cast<int>(buf[m++]);
- } else if(flag == 2) {
- shake_atom[nlocal][0] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][1] = static_cast<int>(buf[m++]);
- shake_type[nlocal][0] = static_cast<int>(buf[m++]);
- } else if(flag == 3) {
- shake_atom[nlocal][0] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][1] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][2] = static_cast<int>(buf[m++]);
- shake_type[nlocal][0] = static_cast<int>(buf[m++]);
- shake_type[nlocal][1] = static_cast<int>(buf[m++]);
- } else if(flag == 4) {
- shake_atom[nlocal][0] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][1] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][2] = static_cast<int>(buf[m++]);
- shake_atom[nlocal][3] = static_cast<int>(buf[m++]);
- shake_type[nlocal][0] = static_cast<int>(buf[m++]);
- shake_type[nlocal][1] = static_cast<int>(buf[m++]);
- shake_type[nlocal][2] = static_cast<int>(buf[m++]);
- }
-
- return m;
-}
-
-/* ----------------------------------------------------------------------
- enforce SHAKE constraints from rRESPA
- prediction portion is different than Verlet
- rRESPA updating of atom coords is done with full v, but only portions of f
-------------------------------------------------------------------------- */
-#if 0
-void FixShakeCuda::post_force_respa(int vflag, int ilevel, int iloop)
-{
- // call stats only on outermost level
-
- if(ilevel == nlevels_respa - 1 && update->ntimestep == next_output) stats();
-
- // perform SHAKE on every loop iteration of every rRESPA level
- // except last loop iteration of inner levels
-
- if(ilevel < nlevels_respa - 1 && iloop == loop_respa[ilevel] - 1) return;
-
- // xshake = atom coords after next x update in innermost loop
- // depends on rRESPA level
- // for levels > 0 this includes more than one velocity update
- // xshake = predicted position from call to this routine at level N =
- // x + dt0 (v + dtN/m fN + 1/2 dt(N-1)/m f(N-1) + ... + 1/2 dt0/m f0)
-
- double** *f_level = ((FixRespa*) modify->fix[ifix_respa])->f_level;
- dtfsq = dtf_inner * step_respa[ilevel];
-
- double invmass, dtfmsq;
- int jlevel;
-
- if(rmass) {
- for(int i = 0; i < nlocal; i++) {
- if(shake_flag[i]) {
- invmass = 1.0 / rmass[i];
- dtfmsq = dtfsq * invmass;
- xshake[i][0] = x[i][0] + dtv * v[i][0] + dtfmsq * f[i][0];
- xshake[i][1] = x[i][1] + dtv * v[i][1] + dtfmsq * f[i][1];
- xshake[i][2] = x[i][2] + dtv * v[i][2] + dtfmsq * f[i][2];
-
- for(jlevel = 0; jlevel < ilevel; jlevel++) {
- dtfmsq = dtf_innerhalf * step_respa[jlevel] * invmass;
- xshake[i][0] += dtfmsq * f_level[i][jlevel][0];
- xshake[i][1] += dtfmsq * f_level[i][jlevel][1];
- xshake[i][2] += dtfmsq * f_level[i][jlevel][2];
- }
- } else xshake[i][2] = xshake[i][1] = xshake[i][0] = 0.0;
- }
-
- } else {
- for(int i = 0; i < nlocal; i++) {
- if(shake_flag[i]) {
- invmass = 1.0 / mass[type[i]];
- dtfmsq = dtfsq * invmass;
- xshake[i][0] = x[i][0] + dtv * v[i][0] + dtfmsq * f[i][0];
- xshake[i][1] = x[i][1] + dtv * v[i][1] + dtfmsq * f[i][1];
- xshake[i][2] = x[i][2] + dtv * v[i][2] + dtfmsq * f[i][2];
-
- for(jlevel = 0; jlevel < ilevel; jlevel++) {
- dtfmsq = dtf_innerhalf * step_respa[jlevel] * invmass;
- xshake[i][0] += dtfmsq * f_level[i][jlevel][0];
- xshake[i][1] += dtfmsq * f_level[i][jlevel][1];
- xshake[i][2] += dtfmsq * f_level[i][jlevel][2];
- }
- } else xshake[i][2] = xshake[i][1] = xshake[i][0] = 0.0;
- }
- }
-
- // communicate results if necessary
-
- if(nprocs > 1) comm->forward_comm_fix(this);
-
- // virial setup
-
- if(vflag) v_setup(vflag);
- else evflag = 0;
-
- // loop over clusters
-
- int m;
-
- for(int i = 0; i < nlist; i++) {
- m = list[i];
-
- if(shake_flag[m] == 2) shake2(m);
- else if(shake_flag[m] == 3) shake3(m);
- else if(shake_flag[m] == 4) shake4(m);
- else shake3angle(m);
- }
-}
-#endif
-
-/* ---------------------------------------------------------------------- */
-
-int FixShakeCuda::pack_forward_comm(int n, int* list, double* buf,
- int pbc_flag, int* pbc)
-{
- if(cuda->finished_setup) {
- int iswap = *list;
-
- if(iswap < 0) {
- iswap = -iswap - 1;
- int first = ((int*) buf)[0];
- Cuda_FixShakeCuda_PackComm_Self(&cuda->shared_data, n, iswap, first, pbc, pbc_flag);
- } else
- Cuda_FixShakeCuda_PackComm(&cuda->shared_data, n, iswap, (void*) buf, pbc, pbc_flag);
-
- return 3*n;
- }
-
- int i, j, m;
- double dx, dy, dz;
-
- m = 0;
-
- if(pbc_flag == 0) {
- for(i = 0; i < n; i++) {
- j = list[i];
- buf[m++] = xshake[j][0];
- buf[m++] = xshake[j][1];
- buf[m++] = xshake[j][2];
- }
- } else {
- if(domain->triclinic == 0) {
- dx = pbc[0] * domain->xprd;
- dy = pbc[1] * domain->yprd;
- dz = pbc[2] * domain->zprd;
- } else {
- dx = pbc[0] * domain->xprd + pbc[5] * domain->xy + pbc[4] * domain->xz;
- dy = pbc[1] * domain->yprd + pbc[3] * domain->yz;
- dz = pbc[2] * domain->zprd;
- }
-
- for(i = 0; i < n; i++) {
- j = list[i];
- buf[m++] = xshake[j][0] + dx;
- buf[m++] = xshake[j][1] + dy;
- buf[m++] = xshake[j][2] + dz;
- }
- }
-
- return m;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixShakeCuda::unpack_forward_comm(int n, int first, double* buf)
-{
- if(cuda->finished_setup) {
- Cuda_FixShakeCuda_UnpackComm(&cuda->shared_data, n, first, (void*)buf);
- return;
- }
-
- int i, m, last;
-
- m = 0;
- last = first + n;
-
- for(i = first; i < last; i++) {
- xshake[i][0] = buf[m++];
- xshake[i][1] = buf[m++];
- xshake[i][2] = buf[m++];
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixShakeCuda::reset_dt()
-{
- if(strstr(update->integrate_style, "verlet")) {
- dtv = update->dt;
- dtfsq = update->dt * update->dt * force->ftm2v;
- } else {
- dtv = step_respa[0];
- dtf_innerhalf = 0.5 * step_respa[0] * force->ftm2v;
- dtf_inner = step_respa[0] * force->ftm2v;
- }
-
- if(cu_shake_atom)
- Cuda_FixShakeCuda_Init(&cuda->shared_data, dtv, dtfsq,
- cu_shake_flag->dev_data(), cu_shake_atom->dev_data(), cu_shake_type->dev_data(), cu_xshake->dev_data(),
- cu_bond_distance->dev_data(), cu_angle_distance->dev_data(), cu_virial->dev_data(),
- max_iter, tolerance);
-}
diff --git a/src/USER-CUDA/fix_shake_cuda.h b/src/USER-CUDA/fix_shake_cuda.h
deleted file mode 100644
index 577ea1daa4..0000000000
--- a/src/USER-CUDA/fix_shake_cuda.h
+++ /dev/null
@@ -1,130 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(shake/cuda,FixShakeCuda)
-
-#else
-
-#ifndef LMP_FIX_SHAKE_CUDA_H
-#define LMP_FIX_SHAKE_CUDA_H
-
-#include "fix.h"
-#include "cuda_data.h"
-#include "cuda_precision.h"
-
-namespace LAMMPS_NS {
-
-class FixShakeCuda : public Fix {
- public:
- FixShakeCuda(class LAMMPS *, int, char **);
- ~FixShakeCuda();
- int setmask();
- void init();
- void setup(int);
- void pre_neighbor();
- void post_force(int);
- //void post_force_respa(int, int, int);
-
- double memory_usage();
- void grow_arrays(int);
- void copy_arrays(int, int, int);
- void set_arrays(int);
- int pack_exchange(int, double *);
- int unpack_exchange(int, double *);
- int pack_forward_comm(int, int *, double *, int, int *);
- void unpack_forward_comm(int, int, double *);
-
- int dof(int);
- void reset_dt();
-
- double time_postforce;
- private:
- class Cuda *cuda;
- int me,nprocs;
- double tolerance; // SHAKE tolerance
- int max_iter; // max # of SHAKE iterations
- int output_every; // SHAKE stat output every so often
- int next_output; // timestep for next output
-
- // settings from input command
- int *bond_flag,*angle_flag; // bond/angle types to constrain
- int *type_flag; // constrain bonds to these types
- double *mass_list; // constrain bonds to these masses
- int nmass; // # of masses in mass_list
- bool neighbor_step; // was neighboring done in this step -> need to run the Cuda_FixShake_Init
-
- double *bond_distance,*angle_distance; // constraint distances
- cCudaData<double , X_CFLOAT , xx >* cu_bond_distance;
- cCudaData<double , X_CFLOAT , xx >* cu_angle_distance;
-
- int ifix_respa; // rRESPA fix needed by SHAKE
- int nlevels_respa; // copies of needed rRESPA variables
- int *loop_respa;
- double *step_respa;
-
- double **x,**v,**f; // local ptrs to atom class quantities
- double *mass,*rmass;
- int *type;
- int nlocal;
- // atom-based arrays
- int *shake_flag; // 0 if atom not in SHAKE cluster
- // 1 = size 3 angle cluster
- // 2,3,4 = size of bond-only cluster
- int **shake_atom; // global IDs of atoms in cluster
- // central atom is 1st
- // lowest global ID is 1st for size 2
-
- int **shake_type; // bondtype of each bond in cluster
- // for angle cluster, 3rd value
- // is angletype
- double **xshake; // unconstrained atom coords
- cCudaData<int , int , xx >* cu_shake_flag;
- cCudaData<int , int , yx >* cu_shake_atom;
- cCudaData<int , int , yx >* cu_shake_type;
- cCudaData<double , X_CFLOAT , xy >* cu_xshake;
- cCudaData<int , int , xx >* cu_list;
- cCudaData<double , ENERGY_CFLOAT , xx >* cu_virial;
-
- double dtv,dtfsq; // timesteps for trial move
- double dtf_inner,dtf_innerhalf; // timesteps for rRESPA trial move
-
- int *list; // list of clusters to SHAKE
- int nlist,maxlist; // size and max-size of list
-
- // stat quantities
- int *b_count,*b_count_all; // counts for each bond type
- double *b_ave,*b_max,*b_min; // ave/max/min dist for each bond type
- double *b_ave_all,*b_max_all,*b_min_all; // MPI summing arrays
- int *a_count,*a_count_all; // ditto for angle types
- double *a_ave,*a_max,*a_min;
- double *a_ave_all,*a_max_all,*a_min_all;
-
- void find_clusters();
- void swap_clusters(int i,int j);
- int masscheck(double);
- void unconstrained_update();
- void shake2(int);
- void shake3(int);
- void shake4(int);
- void shake3angle(int);
- void stats();
- int bondfind(int, int, int);
- int anglefind(int, int, int);
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_temp_berendsen_cuda.cpp b/src/USER-CUDA/fix_temp_berendsen_cuda.cpp
deleted file mode 100644
index ee08aa3462..0000000000
--- a/src/USER-CUDA/fix_temp_berendsen_cuda.cpp
+++ /dev/null
@@ -1,219 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstring>
-#include <cstdlib>
-#include <cmath>
-#include "fix_temp_berendsen_cuda.h"
-#include "fix_temp_berendsen_cuda_cu.h"
-#include "atom.h"
-#include "force.h"
-#include "group.h"
-#include "update.h"
-#include "comm.h"
-#include "modify.h"
-#include "compute.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-enum{NOBIAS,BIAS};
-
-/* ---------------------------------------------------------------------- */
-
-FixTempBerendsenCuda::FixTempBerendsenCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg != 6) error->all(FLERR,"Illegal fix temp/berendsen/cuda command");
-
- // Berendsen thermostat should be applied every step
-
- nevery = 1;
-
- t_start = force->numeric(FLERR,arg[3]);
- t_stop = force->numeric(FLERR,arg[4]);
- t_period = force->numeric(FLERR,arg[5]);
-
- // error checks
-
- if (t_period <= 0.0) error->all(FLERR,"Fix temp/berendsen/cuda period must be > 0.0");
-
- // create a new compute temp style
- // id = fix-ID + temp, compute group = fix group
-
- int n = strlen(id) + 6;
- id_temp = new char[n];
- strcpy(id_temp,id);
- strcat(id_temp,"_temp");
-
- char **newarg = new char*[3];
- newarg[0] = id_temp;
- newarg[1] = group->names[igroup];
- newarg[2] = (char *) "temp/cuda";
- modify->add_compute(3,newarg);
- delete [] newarg;
- tflag = 1;
-}
-
-/* ---------------------------------------------------------------------- */
-
-FixTempBerendsenCuda::~FixTempBerendsenCuda()
-{
- // delete temperature if fix created it
-
- if (tflag) modify->delete_compute(id_temp);
- delete [] id_temp;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixTempBerendsenCuda::setmask()
-{
- int mask = 0;
- mask |= END_OF_STEP_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempBerendsenCuda::init()
-{
- int icompute = modify->find_compute(id_temp);
- if (icompute < 0)
- error->all(FLERR,"Temperature ID for fix temp/berendsen/cuda does not exist");
- temperature = modify->compute[icompute];
- if(not temperature->cudable)
- error->warning(FLERR,"Fix temp/berendsen/cuda uses non cudable temperature compute");
- if (temperature->tempbias) which = BIAS;
- else which = NOBIAS;
-
- //temperature->init(); //not in original berendsen possible error?
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempBerendsenCuda::end_of_step()
-{
- double t_current;
- if(not temperature->cudable) {cuda->cu_x->download();cuda->cu_v->download();}
- t_current = temperature->compute_scalar();
- if (t_current == 0.0)
- error->all(FLERR,"Computed temperature for fix temp/berendsen/cuda cannot be 0.0");
-
- double delta = update->ntimestep - update->beginstep;
- delta /= update->endstep - update->beginstep;
- t_target = t_start + delta * (t_stop-t_start);
-
- // rescale velocities by lamda
-
- double lamda = sqrt(1.0 + update->dt/t_period*(t_target/t_current - 1.0));
-
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- if (which == NOBIAS) {
- Cuda_FixTempBerendsenCuda_EndOfStep(&cuda->shared_data, groupbit,lamda);
-
- } else {
- if(not temperature->cudable)
- {
- cuda->cu_x->download();cuda->cu_v->download();
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- temperature->remove_bias(i,v[i]);
- v[i][0] *= lamda;
- v[i][1] *= lamda;
- v[i][2] *= lamda;
- temperature->restore_bias(i,v[i]);
- }
- }
- cuda->cu_v->upload();
- }
- else
- {
- temperature->remove_bias_all();
- Cuda_FixTempBerendsenCuda_EndOfStep(&cuda->shared_data, groupbit,lamda);
- temperature->restore_bias_all();
- }
- }
-
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixTempBerendsenCuda::modify_param(int narg, char **arg)
-{
- if (strcmp(arg[0],"temp") == 0) {
- if (narg < 2) error->all(FLERR,"Illegal fix_modify command");
- if (tflag) {
- modify->delete_compute(id_temp);
- tflag = 0;
- }
- delete [] id_temp;
- int n = strlen(arg[1]) + 1;
- id_temp = new char[n];
- strcpy(id_temp,arg[1]);
-
- int icompute = modify->find_compute(id_temp);
- if (icompute < 0) error->all(FLERR,"Could not find fix_modify temperature ID");
- temperature = modify->compute[icompute];
-
- if (temperature->tempflag == 0)
- error->all(FLERR,"Fix_modify temperature ID does not compute temperature");
- if (temperature->igroup != igroup && comm->me == 0)
- error->warning(FLERR,"Group for fix_modify temp != fix group");
- return 2;
- }
- return 0;
-}
-
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempBerendsenCuda::reset_target(double t_new)
-{
- t_start = t_stop = t_new;
-}
diff --git a/src/USER-CUDA/fix_temp_berendsen_cuda.h b/src/USER-CUDA/fix_temp_berendsen_cuda.h
deleted file mode 100644
index 610e5421e5..0000000000
--- a/src/USER-CUDA/fix_temp_berendsen_cuda.h
+++ /dev/null
@@ -1,58 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-#ifdef FIX_CLASS
-
-FixStyle(temp/berendsen/cuda,FixTempBerendsenCuda)
-
-#else
-
-#ifndef LMP_FIX_TEMP_BERENDSEN_CUDA_H
-#define LMP_FIX_TEMP_BERENDSEN_CUDA_H
-
-#include "fix.h"
-
-namespace LAMMPS_NS {
-class FixTempBerendsenCuda : public Fix {
- public:
- FixTempBerendsenCuda(class LAMMPS *, int, char **);
- ~FixTempBerendsenCuda();
- int setmask();
- void init();
- void end_of_step();
- int modify_param(int, char **);
- void reset_target(double);
-
- private:
- class Cuda *cuda;
- int which;
- double t_start,t_stop,t_target,t_period;
-
- char *id_temp;
- class Compute *temperature;
- int tflag;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_temp_rescale_cuda.cpp b/src/USER-CUDA/fix_temp_rescale_cuda.cpp
deleted file mode 100644
index a0ebb47d12..0000000000
--- a/src/USER-CUDA/fix_temp_rescale_cuda.cpp
+++ /dev/null
@@ -1,224 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cstring>
-#include <cstdlib>
-#include <cmath>
-#include "fix_temp_rescale_cuda.h"
-#include "fix_temp_rescale_cuda_cu.h"
-#include "atom.h"
-#include "force.h"
-#include "group.h"
-#include "update.h"
-#include "domain.h"
-#include "region.h"
-#include "comm.h"
-#include "modify.h"
-#include "compute.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-enum{NOBIAS,BIAS};
-
-/* ---------------------------------------------------------------------- */
-
-FixTempRescaleCuda::FixTempRescaleCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg < 8) error->all(FLERR,"Illegal fix temp/rescale/cuda command");
-
- nevery = force->inumeric(FLERR,arg[3]);
- if (nevery <= 0) error->all(FLERR,"Illegal fix temp/rescale/cuda command");
-
- scalar_flag = 1;
- global_freq = nevery;
- extscalar = 1;
-
- t_start = force->numeric(FLERR,arg[4]);
- t_stop = force->numeric(FLERR,arg[5]);
- t_window = force->numeric(FLERR,arg[6]);
- fraction = force->numeric(FLERR,arg[7]);
-
- // create a new compute temp
- // id = fix-ID + temp, compute group = fix group
-
- int n = strlen(id) + 6;
- id_temp = new char[n];
- strcpy(id_temp,id);
- strcat(id_temp,"_temp");
-
- char **newarg = new char*[6];
- newarg[0] = id_temp;
- newarg[1] = group->names[igroup];
- newarg[2] = (char *) "temp/cuda";
- modify->add_compute(3,newarg);
- delete [] newarg;
- tflag = 1;
-
- energy = 0.0;
-}
-
-/* ---------------------------------------------------------------------- */
-
-FixTempRescaleCuda::~FixTempRescaleCuda()
-{
- // delete temperature if fix created it
-
- if (tflag) modify->delete_compute(id_temp);
- delete [] id_temp;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixTempRescaleCuda::setmask()
-{
- int mask = 0;
- mask |= END_OF_STEP_CUDA;
- mask |= THERMO_ENERGY_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempRescaleCuda::init()
-{
- int icompute = modify->find_compute(id_temp);
- if (icompute < 0)
- error->all(FLERR,"Temperature ID for fix temp/rescale/cuda does not exist");
- temperature = modify->compute[icompute];
- if(not temperature->cudable)
- error->warning(FLERR,"Fix temp/rescale/cuda uses non cudable temperature compute");
- if (temperature->tempbias) which = BIAS;
- else which = NOBIAS;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempRescaleCuda::end_of_step()
-{
- double t_current;
- if(not temperature->cudable) {cuda->cu_x->download();cuda->cu_v->download();}
- t_current = temperature->compute_scalar();
- if (t_current == 0.0)
- error->all(FLERR,"Computed temperature for fix temp/rescale/cuda cannot be 0.0");
-
- double delta = update->ntimestep - update->beginstep;
- delta /= update->endstep - update->beginstep;
- double t_target = t_start + delta * (t_stop-t_start);
-
- // rescale velocity of appropriate atoms if outside window
-
- if (fabs(t_current-t_target) > t_window) {
- t_target = t_current - fraction*(t_current-t_target);
- double factor = sqrt(t_target/t_current);
- double efactor = 0.5 * force->boltz * temperature->dof;
-
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- if (which == NOBIAS) {
- energy += (t_current-t_target) * efactor;
-
- Cuda_FixTempRescaleCuda_EndOfStep(&cuda->shared_data, groupbit,factor);
-
- } else if (which == BIAS) {
- energy += (t_current-t_target) * efactor;
- if(not temperature->cudable)
- {
- cuda->cu_x->download();cuda->cu_v->download();
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- temperature->remove_bias(i,v[i]);
- v[i][0] *= factor;
- v[i][1] *= factor;
- v[i][2] *= factor;
- temperature->restore_bias(i,v[i]);
- }
- }
- cuda->cu_v->upload();
- }
- else
- {
- temperature->remove_bias_all();
- Cuda_FixTempRescaleCuda_EndOfStep(&cuda->shared_data, groupbit,factor);
- temperature->restore_bias_all();
- }
- }
-
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixTempRescaleCuda::modify_param(int narg, char **arg)
-{
- if (strcmp(arg[0],"temp") == 0) {
- if (narg < 2) error->all(FLERR,"Illegal fix_modify command");
- if (tflag) {
- modify->delete_compute(id_temp);
- tflag = 0;
- }
- delete [] id_temp;
- int n = strlen(arg[1]) + 1;
- id_temp = new char[n];
- strcpy(id_temp,arg[1]);
-
- int icompute = modify->find_compute(id_temp);
- if (icompute < 0) error->all(FLERR,"Could not find fix_modify temperature ID");
- temperature = modify->compute[icompute];
-
- if (temperature->tempflag == 0)
- error->all(FLERR,"Fix_modify temperature ID does not compute temperature");
- if (temperature->igroup != igroup && comm->me == 0)
- error->warning(FLERR,"Group for fix_modify temp != fix group");
- if(not temperature->cudable)
- error->warning(FLERR,"Fix temp/rescale/cuda uses non cudable temperature compute");
- return 2;
- }
- return 0;
-}
-
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempRescaleCuda::reset_target(double t_new)
-{
- t_start = t_stop = t_new;
-}
-
-/* ---------------------------------------------------------------------- */
-
-double FixTempRescaleCuda::compute_scalar()
-{
- return energy;
-}
diff --git a/src/USER-CUDA/fix_temp_rescale_cuda.h b/src/USER-CUDA/fix_temp_rescale_cuda.h
deleted file mode 100644
index 3bdc71a1a0..0000000000
--- a/src/USER-CUDA/fix_temp_rescale_cuda.h
+++ /dev/null
@@ -1,61 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(temp/rescale/cuda,FixTempRescaleCuda)
-
-#else
-
-#ifndef FIX_TEMP_RESCALE_CUDA_H
-#define FIX_TEMP_RESCALE_CUDA_H
-
-#include "fix.h"
-
-namespace LAMMPS_NS {
-class FixTempRescaleCuda : public Fix {
- public:
- FixTempRescaleCuda(class LAMMPS *, int, char **);
- ~FixTempRescaleCuda();
- int setmask();
- void init();
- void end_of_step();
- int modify_param(int, char **);
- void reset_target(double);
- double compute_scalar();
-
- private:
- class Cuda *cuda;
- int which;
- double t_start,t_stop,t_window;
- double fraction,energy,efactor;
-
- char *id_temp;
- class Compute *temperature;
- int tflag;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_temp_rescale_limit_cuda.cpp b/src/USER-CUDA/fix_temp_rescale_limit_cuda.cpp
deleted file mode 100644
index eb8cf8d948..0000000000
--- a/src/USER-CUDA/fix_temp_rescale_limit_cuda.cpp
+++ /dev/null
@@ -1,237 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cstring>
-#include <cstdlib>
-#include <cmath>
-#include "fix_temp_rescale_limit_cuda.h"
-#include "fix_temp_rescale_limit_cuda_cu.h"
-#include "atom.h"
-#include "force.h"
-#include "group.h"
-#include "update.h"
-#include "domain.h"
-#include "region.h"
-#include "comm.h"
-#include "modify.h"
-#include "compute.h"
-#include "error.h"
-#include "user_cuda.h"
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-enum{NOBIAS,BIAS};
-
-/* ---------------------------------------------------------------------- */
-
-FixTempRescaleLimitCuda::FixTempRescaleLimitCuda(LAMMPS *lmp, int narg, char **arg) :
- Fix(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if (narg < 9) error->all(FLERR,"Illegal fix temp/rescale/limit/cuda command");
-
- nevery = force->inumeric(FLERR,arg[3]);
- if (nevery <= 0) error->all(FLERR,"Illegal fix temp/rescale/limit/cuda command");
-
- scalar_flag = 1;
- global_freq = nevery;
- extscalar = 1;
-
- t_start = force->numeric(FLERR,arg[4]);
- t_stop = force->numeric(FLERR,arg[5]);
- t_window = force->numeric(FLERR,arg[6]);
- fraction = force->numeric(FLERR,arg[7]);
- limit = force->numeric(FLERR,arg[8]);
- if (limit <= 1.0) error->all(FLERR,"Illegal fix temp/rescale/limit/cuda command (limit must be > 1.0)");
-
-
- // create a new compute temp
- // id = fix-ID + temp, compute group = fix group
-
- int n = strlen(id) + 6;
- id_temp = new char[n];
- strcpy(id_temp,id);
- strcat(id_temp,"_temp");
-
- char **newarg = new char*[6];
- newarg[0] = id_temp;
- newarg[1] = group->names[igroup];
- newarg[2] = (char *) "temp/cuda";
- modify->add_compute(3,newarg);
- delete [] newarg;
- tflag = 1;
-
- energy = 0.0;
-}
-
-/* ---------------------------------------------------------------------- */
-
-FixTempRescaleLimitCuda::~FixTempRescaleLimitCuda()
-{
- // delete temperature if fix created it
-
- if (tflag) modify->delete_compute(id_temp);
- delete [] id_temp;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixTempRescaleLimitCuda::setmask()
-{
- int mask = 0;
- mask |= END_OF_STEP_CUDA;
- mask |= THERMO_ENERGY_CUDA;
- return mask;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempRescaleLimitCuda::init()
-{
- int icompute = modify->find_compute(id_temp);
- if (icompute < 0)
- error->all(FLERR,"Temperature ID for fix temp/rescale/limit/cuda does not exist");
- temperature = modify->compute[icompute];
- if(not temperature->cudable)
- error->warning(FLERR,"Fix temp/rescale/limit/cuda uses non cudable temperature compute");
- if (temperature->tempbias) which = BIAS;
- else which = NOBIAS;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempRescaleLimitCuda::end_of_step()
-{
- double t_current;
- if(not temperature->cudable) {cuda->cu_x->download();cuda->cu_v->download();}
- t_current = temperature->compute_scalar();
- if (t_current == 0.0)
- error->all(FLERR,"Computed temperature for fix temp/rescale/limit/cuda cannot be 0.0");
-
- double delta = update->ntimestep - update->beginstep;
- delta /= update->endstep - update->beginstep;
- double t_target = t_start + delta * (t_stop-t_start);
-
- // rescale velocity of appropriate atoms if outside window
-
- if (fabs(t_current-t_target) > t_window) {
- t_target = t_current - fraction*(t_current-t_target);
- double factor = sqrt(t_target/t_current);
- double efactor = 0.5 * force->boltz * temperature->dof;
-
- double **v = atom->v;
- int *mask = atom->mask;
- int nlocal = atom->nlocal;
-
- double massone;
- if(atom->rmass) massone = atom->rmass[0];
- else massone = atom->mass[0];
-
- double current_limit=sqrt(limit*force->boltz*t_target*temperature->dof/massone/force->mvv2e);
- if (which == NOBIAS) {
- energy += (t_current-t_target) * efactor;
-
-
- Cuda_FixTempRescaleLimitCuda_EndOfStep(&cuda->shared_data, groupbit,factor,current_limit);
-
- } else if (which == BIAS) {
- energy += (t_current-t_target) * efactor;
- if(not temperature->cudable)
- {
- cuda->cu_x->download();cuda->cu_v->download();
- for (int i = 0; i < nlocal; i++) {
- if (mask[i] & groupbit) {
- temperature->remove_bias(i,v[i]);
- double vx = v[i][0] * factor;
- double vy = v[i][1] * factor;
- double vz = v[i][2] * factor;
- v[i][0]=vx>0?MIN(vx,current_limit):MAX(vx,-current_limit);
- v[i][1]=vy>0?MIN(vy,current_limit):MAX(vy,-current_limit);
- v[i][2]=vz>0?MIN(vz,current_limit):MAX(vz,-current_limit);
-
- temperature->restore_bias(i,v[i]);
- }
- }
- cuda->cu_v->upload();
- }
- else
- {
- temperature->remove_bias_all();
- Cuda_FixTempRescaleLimitCuda_EndOfStep(&cuda->shared_data, groupbit,factor,current_limit);
- temperature->restore_bias_all();
- }
- }
-
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixTempRescaleLimitCuda::modify_param(int narg, char **arg)
-{
- if (strcmp(arg[0],"temp") == 0) {
- if (narg < 2) error->all(FLERR,"Illegal fix_modify command");
- if (tflag) {
- modify->delete_compute(id_temp);
- tflag = 0;
- }
- delete [] id_temp;
- int n = strlen(arg[1]) + 1;
- id_temp = new char[n];
- strcpy(id_temp,arg[1]);
-
- int icompute = modify->find_compute(id_temp);
- if (icompute < 0) error->all(FLERR,"Could not find fix_modify temperature ID");
- temperature = modify->compute[icompute];
-
- if (temperature->tempflag == 0)
- error->all(FLERR,"Fix_modify temperature ID does not compute temperature");
- if (temperature->igroup != igroup && comm->me == 0)
- error->warning(FLERR,"Group for fix_modify temp != fix group");
- if(not temperature->cudable)
- error->warning(FLERR,"Fix temp/rescale/limit/cuda uses non cudable temperature compute");
- return 2;
- }
- return 0;
-}
-
-
-/* ---------------------------------------------------------------------- */
-
-void FixTempRescaleLimitCuda::reset_target(double t_new)
-{
- t_start = t_stop = t_new;
-}
-
-/* ---------------------------------------------------------------------- */
-
-double FixTempRescaleLimitCuda::compute_scalar()
-{
- return energy;
-}
diff --git a/src/USER-CUDA/fix_temp_rescale_limit_cuda.h b/src/USER-CUDA/fix_temp_rescale_limit_cuda.h
deleted file mode 100644
index b2bba2049a..0000000000
--- a/src/USER-CUDA/fix_temp_rescale_limit_cuda.h
+++ /dev/null
@@ -1,61 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(temp/rescale/limit/cuda,FixTempRescaleLimitCuda)
-
-#else
-
-#ifndef FIX_TEMP_RESCALE_LIMIT_CUDA_H
-#define FIX_TEMP_RESCALE_LIMIT_CUDA_H
-
-#include "fix.h"
-
-namespace LAMMPS_NS {
-class FixTempRescaleLimitCuda : public Fix {
- public:
- FixTempRescaleLimitCuda(class LAMMPS *, int, char **);
- ~FixTempRescaleLimitCuda();
- int setmask();
- void init();
- void end_of_step();
- int modify_param(int, char **);
- void reset_target(double);
- double compute_scalar();
-
- private:
- class Cuda *cuda;
- int which;
- double t_start,t_stop,t_window;
- double fraction,energy,efactor;
- double limit;
- char *id_temp;
- class Compute *temperature;
- int tflag;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/fix_viscous_cuda.cpp b/src/USER-CUDA/fix_viscous_cuda.cpp
deleted file mode 100644
index 09871c86e3..0000000000
--- a/src/USER-CUDA/fix_viscous_cuda.cpp
+++ /dev/null
@@ -1,105 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdlib>
-#include <cstring>
-#include "fix_viscous_cuda.h"
-#include "fix_viscous_cuda_cu.h"
-#include "atom.h"
-#include "update.h"
-#include "respa.h"
-#include "error.h"
-#include "cuda_modify_flags.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-/* ---------------------------------------------------------------------- */
-
-FixViscousCuda::FixViscousCuda(LAMMPS *lmp, int narg, char **arg) :
- FixViscous(lmp, narg, arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- cu_gamma=NULL;
-}
-
-/* ---------------------------------------------------------------------- */
-
-FixViscousCuda::~FixViscousCuda()
-{
- delete cu_gamma;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int FixViscousCuda::setmask()
-{
- int mask = 0;
- mask |= POST_FORCE_CUDA;
- // mask |= POST_FORCE_RESPA;
- // mask |= MIN_POST_FORCE;
- return mask;
-}
-
-
-/* ---------------------------------------------------------------------- */
-
-void FixViscousCuda::setup(int vflag)
-{
- if(not cu_gamma)
- cu_gamma = new cCudaData<double, F_CFLOAT, x> (gamma,atom->ntypes+1);
- Cuda_FixViscousCuda_Init(&cuda->shared_data);
- cu_gamma->upload();
- // if (strcmp(update->integrate_style,"verlet/cuda") == 0)
- post_force(vflag);
- /* else {
- ((Respa *) update->integrate)->copy_flevel_f(nlevels_respa-1);
- post_force_respa(vflag,nlevels_respa-1,0);
- ((Respa *) update->integrate)->copy_f_flevel(nlevels_respa-1);
- }*/
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixViscousCuda::min_setup(int vflag)
-{
- Cuda_FixViscousCuda_Init(&cuda->shared_data);
- post_force(vflag);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void FixViscousCuda::post_force(int vflag)
-{
- // apply drag force to atoms in group
- // direction is opposed to velocity vector
- // magnitude depends on atom type
-
- Cuda_FixViscousCuda_PostForce(&cuda->shared_data, groupbit,cu_gamma->dev_data());
-}
diff --git a/src/USER-CUDA/fix_viscous_cuda.h b/src/USER-CUDA/fix_viscous_cuda.h
deleted file mode 100644
index e0cb6ba4b0..0000000000
--- a/src/USER-CUDA/fix_viscous_cuda.h
+++ /dev/null
@@ -1,55 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef FIX_CLASS
-
-FixStyle(viscous/cuda,FixViscousCuda)
-
-#else
-
-#ifndef LMP_FIX_VISCOUS_CUDA_H
-#define LMP_FIX_VISCOUS_CUDA_H
-
-#include "fix_viscous.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class FixViscousCuda : public FixViscous {
- public:
- FixViscousCuda(class LAMMPS *, int, char **);
- ~FixViscousCuda();
- int setmask();
- void setup(int);
- void min_setup(int);
- void post_force(int);
- cCudaData<double, F_CFLOAT, x>* cu_gamma;
-
- private:
- class Cuda *cuda;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/modify_cuda.cpp b/src/USER-CUDA/modify_cuda.cpp
deleted file mode 100644
index 82d6d92036..0000000000
--- a/src/USER-CUDA/modify_cuda.cpp
+++ /dev/null
@@ -1,437 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <cstdio>
-#include <cstring>
-#include "modify_cuda.h"
-#include "style_compute.h"
-#include "style_fix.h"
-#include "atom.h"
-#include "comm.h"
-#include "fix.h"
-#include "compute.h"
-#include "group.h"
-#include "update.h"
-#include "domain.h"
-#include "user_cuda.h"
-#include "memory.h"
-#include "error.h"
-
-#include "cuda_modify_flags.h"
-
-using namespace LAMMPS_NS;
-using namespace FixConst;
-using namespace FixConstCuda;
-
-#define DELTA 4
-
-#define BIG 1.0e20
-
-
-/* ---------------------------------------------------------------------- */
-
-ModifyCuda::ModifyCuda(LAMMPS *lmp) : Modify(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- n_initial_integrate_cuda = 0;
- n_post_integrate_cuda = 0;
- n_pre_exchange = 0;
- n_pre_neighbor_cuda = 0;
- n_pre_force_cuda = 0;
- n_post_force_cuda = 0;
- n_final_integrate_cuda = 0;
- n_end_of_step_cuda = 0;
- n_thermo_energy_cuda = 0;
-
- n_initial_integrate_host = 0;
- n_post_integrate_host = 0;
- n_pre_exchange = 0;
- n_pre_neighbor_host = 0;
- n_pre_force_host = 0;
- n_post_force_host = 0;
- n_final_integrate_host = 0;
- n_end_of_step_host = 0;
- n_thermo_energy_host = 0;
-
- list_initial_integrate_cuda = NULL;
- list_post_integrate_cuda = NULL;
- list_pre_exchange_cuda = NULL;
- list_pre_neighbor_cuda = NULL;
- list_pre_force_cuda = NULL;
- list_post_force_cuda = NULL;
- list_final_integrate_cuda = NULL;
- list_end_of_step_cuda = NULL;
- list_thermo_energy_cuda = NULL;
- end_of_step_every_cuda = NULL;
-}
-
-/* ---------------------------------------------------------------------- */
-
-ModifyCuda::~ModifyCuda()
-{
- delete [] list_initial_integrate_cuda;
- delete [] list_post_integrate_cuda;
- delete [] list_pre_exchange_cuda;
- delete [] list_pre_neighbor_cuda;
- delete [] list_pre_force_cuda;
- delete [] list_post_force_cuda;
- delete [] list_final_integrate_cuda;
- delete [] list_end_of_step_cuda;
- delete [] list_thermo_energy_cuda;
- delete [] end_of_step_every_cuda;
-}
-
-/* ----------------------------------------------------------------------
- initialize all fixes and computes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::init()
-{
- int i,j;
-
- // delete storage of restart info since it is not valid after 1st run
-
- restart_deallocate();
-
- // create lists of fixes to call at each stage of run
-
- list_init(INITIAL_INTEGRATE,n_initial_integrate,list_initial_integrate);
- list_init(POST_INTEGRATE,n_post_integrate,list_post_integrate);
- list_init(PRE_EXCHANGE,n_pre_exchange,list_pre_exchange);
- list_init(PRE_NEIGHBOR,n_pre_neighbor,list_pre_neighbor);
- list_init(PRE_FORCE,n_pre_force,list_pre_force);
- list_init(POST_FORCE,n_post_force,list_post_force);
- list_init(FINAL_INTEGRATE,n_final_integrate,list_final_integrate);
- list_init_end_of_step(END_OF_STEP,n_end_of_step,list_end_of_step);
- list_init_thermo_energy(THERMO_ENERGY,n_thermo_energy,list_thermo_energy);
-
- list_init(INITIAL_INTEGRATE_CUDA, n_initial_integrate_cuda, list_initial_integrate_cuda);
- list_init(POST_INTEGRATE_CUDA, n_post_integrate_cuda, list_post_integrate_cuda);
- list_init(PRE_EXCHANGE_CUDA, n_pre_exchange_cuda, list_pre_exchange_cuda);
- list_init(PRE_NEIGHBOR_CUDA, n_pre_neighbor_cuda, list_pre_neighbor_cuda);
- list_init(PRE_FORCE_CUDA, n_pre_force_cuda, list_pre_force_cuda);
- list_init(POST_FORCE_CUDA, n_post_force_cuda, list_post_force_cuda);
- list_init(FINAL_INTEGRATE_CUDA, n_final_integrate_cuda, list_final_integrate_cuda);
- list_init_end_of_step_cuda(END_OF_STEP_CUDA, n_end_of_step_cuda, list_end_of_step_cuda);
- list_init_thermo_energy(THERMO_ENERGY_CUDA, n_thermo_energy_cuda, list_thermo_energy_cuda);
-
- n_initial_integrate_host = n_initial_integrate;
- n_post_integrate_host = n_post_integrate;
- n_pre_exchange_host = n_pre_exchange;
- n_pre_neighbor_host = n_pre_neighbor;
- n_pre_force_host = n_pre_force;
- n_post_force_host = n_post_force;
- n_final_integrate_host = n_final_integrate;
- n_end_of_step_host = n_end_of_step;
- n_thermo_energy_host = n_thermo_energy;
-
- n_initial_integrate = n_initial_integrate_cuda+n_initial_integrate_host;
- n_post_integrate = n_post_integrate_cuda+n_post_integrate_host;
- n_pre_exchange = n_pre_exchange_cuda+n_pre_exchange_host;
- n_pre_neighbor = n_pre_neighbor_cuda+n_pre_neighbor_host;
- n_pre_force = n_pre_force_cuda+n_pre_force_host;
- n_post_force = n_post_force_cuda+n_post_force_host;
- n_final_integrate = n_final_integrate_cuda+n_final_integrate_host;
- n_end_of_step = n_end_of_step_cuda+n_end_of_step_host;
- n_thermo_energy = n_thermo_energy_cuda+n_thermo_energy_host;
-
- list_init(INITIAL_INTEGRATE_RESPA,
- n_initial_integrate_respa,list_initial_integrate_respa);
- list_init(POST_INTEGRATE_RESPA,
- n_post_integrate_respa,list_post_integrate_respa);
- list_init(POST_FORCE_RESPA,
- n_post_force_respa,list_post_force_respa);
- list_init(PRE_FORCE_RESPA,
- n_pre_force_respa,list_pre_force_respa);
- list_init(FINAL_INTEGRATE_RESPA,
- n_final_integrate_respa,list_final_integrate_respa);
-
- list_init(MIN_PRE_EXCHANGE,n_min_pre_exchange,list_min_pre_exchange);
- list_init(MIN_POST_FORCE,n_min_post_force,list_min_post_force);
- list_init(MIN_ENERGY,n_min_energy,list_min_energy);
-
- // init each fix
- // needs to come before compute init
- // this is b/c some computes call fix->dof()
- // FixRigid::dof() depends on its own init having been called
-
- for (i = 0; i < nfix; i++) fix[i]->init();
-
- // set global flag if any fix has its restart_pbc flag set
-
- restart_pbc_any = 0;
- for (i = 0; i < nfix; i++)
- if (fix[i]->restart_pbc) restart_pbc_any = 1;
-
- // create list of computes that store invocation times
-
- list_init_compute();
-
- // init each compute
- // set invoked_scalar,vector,etc to -1 to force new run to re-compute them
- // add initial timestep to all computes that store invocation times
- // since any of them may be invoked by initial thermo
- // do not clear out invocation times stored within a compute,
- // b/c some may be holdovers from previous run, like for ave fixes
-
- for (i = 0; i < ncompute; i++) {
- compute[i]->init();
- compute[i]->invoked_scalar = -1;
- compute[i]->invoked_vector = -1;
- compute[i]->invoked_array = -1;
- compute[i]->invoked_peratom = -1;
- compute[i]->invoked_local = -1;
- }
- addstep_compute_all(update->ntimestep);
-
- // warn if any particle is time integrated more than once
-
- int nlocal = atom->nlocal;
- int *mask = atom->mask;
-
- int *flag = new int[nlocal];
- for (i = 0; i < nlocal; i++) flag[i] = 0;
-
- int groupbit;
- for (i = 0; i < nfix; i++) {
- if (fix[i]->time_integrate == 0) continue;
- groupbit = fix[i]->groupbit;
- for (j = 0; j < nlocal; j++)
- if (mask[j] & groupbit) flag[j]++;
- }
-
- int check = 0;
- for (i = 0; i < nlocal; i++)
- if (flag[i] > 1) check = 1;
-
- delete [] flag;
-
- int checkall;
- MPI_Allreduce(&check,&checkall,1,MPI_INT,MPI_SUM,world);
- if (comm->me == 0 && checkall)
- error->warning(FLERR,"One or more atoms are time integrated more than once");
-}
-
-/* ----------------------------------------------------------------------
- 1st half of integrate call, only for relevant fixes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::initial_integrate(int vflag)
-{
- for(int i = 0; i < n_initial_integrate_cuda; i++)
- fix[list_initial_integrate_cuda[i]]->initial_integrate(vflag);
-
- if(n_initial_integrate_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_initial_integrate_host; i++)
- fix[list_initial_integrate[i]]->initial_integrate(vflag);
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-/* ----------------------------------------------------------------------
- post_integrate call, only for relevant fixes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::post_integrate()
-{
- for(int i = 0; i < n_post_integrate_cuda; i++)
- fix[list_post_integrate_cuda[i]]->post_integrate();
-
- if(n_post_integrate_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_post_integrate_host; i++)
- fix[list_post_integrate[i]]->post_integrate();
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-/* ----------------------------------------------------------------------
- pre_exchange call, only for relevant fixes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::pre_exchange()
-{
- for(int i = 0; i < n_pre_exchange_cuda; i++)
- fix[list_pre_exchange_cuda[i]]->pre_exchange();
-
- if(n_pre_exchange_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_pre_exchange_host; i++)
- fix[list_pre_exchange[i]]->pre_exchange();
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-/* ----------------------------------------------------------------------
- pre_neighbor call, only for relevant fixes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::pre_neighbor()
-{
- for(int i = 0; i < n_pre_neighbor_cuda; i++)
- fix[list_pre_neighbor_cuda[i]]->pre_neighbor();
-
- if(n_pre_neighbor_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_pre_neighbor_host; i++)
- fix[list_pre_neighbor[i]]->pre_neighbor();
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-/* ----------------------------------------------------------------------
- pre_force call, only for relevant fixes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::setup_pre_force(int vflag)
-{
- for(int i = 0; i < n_pre_force_cuda; i++)
- fix[list_pre_force_cuda[i]]->pre_force(vflag);
-
- if(n_pre_force_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_pre_force_host; i++)
- fix[list_pre_force[i]]->pre_force(vflag);
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-void ModifyCuda::pre_force(int vflag)
-{
- for(int i = 0; i < n_pre_force_cuda; i++)
- fix[list_pre_force_cuda[i]]->pre_force(vflag);
-
- if(n_pre_force_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_pre_force_host; i++)
- fix[list_pre_force[i]]->pre_force(vflag);
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-/* ----------------------------------------------------------------------
- post_force call, only for relevant fixes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::post_force(int vflag)
-{
- for(int i = 0; i < n_post_force_cuda; i++)
- fix[list_post_force_cuda[i]]->post_force(vflag);
-
- if(n_post_force_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_post_force_host; i++)
- fix[list_post_force[i]]->post_force(vflag);
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-/* ----------------------------------------------------------------------
- 2nd half of integrate call, only for relevant fixes
-------------------------------------------------------------------------- */
-
-void ModifyCuda::final_integrate()
-{
- for (int i = 0; i < n_final_integrate_cuda; i++)
- fix[list_final_integrate_cuda[i]]->final_integrate();
-
- if(n_final_integrate_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_final_integrate_host; i++)
- fix[list_final_integrate[i]]->final_integrate();
- cuda->uploadAll(); cuda->oncpu = false;
- }
-}
-
-/* ----------------------------------------------------------------------
- end-of-timestep call, only for relevant fixes
- only call fix->end_of_step() on timesteps that are multiples of nevery
-------------------------------------------------------------------------- */
-
-void ModifyCuda::end_of_step()
-{
- for (int i = 0; i < n_end_of_step_cuda; i++)
- if (update->ntimestep % end_of_step_every_cuda[i] == 0)
- fix[list_end_of_step_cuda[i]]->end_of_step();
-
- if(n_end_of_step_host != 0)
- {
- int do_thisstep=0;
- for (int i = 0; i < n_end_of_step_host; i++)
- if (update->ntimestep % end_of_step_every[i] == 0) do_thisstep=1;
- if(do_thisstep)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_end_of_step_host; i++)
- if (update->ntimestep % end_of_step_every[i] == 0)
- fix[list_end_of_step[i]]->end_of_step();
- cuda->uploadAll(); cuda->oncpu = false;
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- thermo energy call, only for relevant fixes
- called by Thermo class
- compute_scalar() is fix call to return energy
-------------------------------------------------------------------------- */
-
-double ModifyCuda::thermo_energy()
-{
- double energy = 0.0;
-
- for (int i = 0; i < n_thermo_energy_cuda; i++)
- energy += fix[list_thermo_energy_cuda[i]]->compute_scalar();
-
- if(n_thermo_energy_host != 0)
- {
- cuda->downloadAll(); cuda->oncpu = true;
- for (int i = 0; i < n_thermo_energy_host; i++)
- energy += fix[list_thermo_energy[i]]->compute_scalar();
- cuda->uploadAll(); cuda->oncpu = false;
- }
-
- return energy;
-}
-
-
-
-void ModifyCuda::list_init_end_of_step_cuda(int mask, int &n, int *&list)
-{
- delete [] list;
- delete [] end_of_step_every_cuda;
-
- n = 0;
- for (int i = 0; i < nfix; i++) if (fmask[i] & mask) n++;
- list = new int[n];
- end_of_step_every_cuda = new int[n];
-
- n = 0;
- for (int i = 0; i < nfix; i++)
- if (fmask[i] & mask) {
- list[n] = i;
- end_of_step_every_cuda[n++] = fix[i]->nevery;
- }
-}
diff --git a/src/USER-CUDA/modify_cuda.h b/src/USER-CUDA/modify_cuda.h
deleted file mode 100644
index bfea217046..0000000000
--- a/src/USER-CUDA/modify_cuda.h
+++ /dev/null
@@ -1,83 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifndef LMP_MODIFY_CUDA_H
-#define LMP_MODIFY_CUDA_H
-
-#include <cstdio>
-#include "modify.h"
-
-namespace LAMMPS_NS {
-
-class ModifyCuda : public Modify {
- public:
-
- int n_initial_integrate_cuda;
- int n_post_integrate_cuda;
- int n_pre_exchange_cuda;
- int n_pre_neighbor_cuda;
- int n_pre_force_cuda;
- int n_post_force_cuda;
- int n_final_integrate_cuda;
- int n_end_of_step_cuda;
- int n_thermo_energy_cuda;
-
- int n_initial_integrate_host;
- int n_post_integrate_host;
- int n_pre_exchange_host;
- int n_pre_neighbor_host;
- int n_pre_force_host;
- int n_post_force_host;
- int n_final_integrate_host;
- int n_end_of_step_host;
- int n_thermo_energy_host;
-
- ModifyCuda(class LAMMPS *);
- ~ModifyCuda();
- void init();
- void initial_integrate(int);
- void post_integrate();
- //void pre_decide();
- void pre_exchange();
- void pre_neighbor();
- void setup_pre_force(int);
- void pre_force(int);
- void post_force(int);
- void final_integrate();
- void end_of_step();
- double thermo_energy();
-
-
- protected:
- class Cuda *cuda;
-
- // lists of fixes to apply at different stages of timestep
-
- // list of cuda fixes
- int *list_initial_integrate_cuda;
- int *list_post_integrate_cuda;
- int *list_pre_exchange_cuda;
- int *list_pre_neighbor_cuda;
- int *list_pre_force_cuda;
- int *list_post_force_cuda;
- int *list_final_integrate_cuda;
- int *list_end_of_step_cuda;
- int *list_thermo_energy_cuda;
- int *end_of_step_every_cuda;
-
- void list_init_end_of_step_cuda(int, int &, int *&);
-};
-
-}
-
-#endif
diff --git a/src/USER-CUDA/neigh_full_cuda.cpp b/src/USER-CUDA/neigh_full_cuda.cpp
deleted file mode 100644
index 5fd69f1105..0000000000
--- a/src/USER-CUDA/neigh_full_cuda.cpp
+++ /dev/null
@@ -1,307 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include "neighbor_cuda.h"
-#include "neigh_list.h"
-#include "atom.h"
-#include "domain.h"
-#include "group.h"
-#include "error.h"
-#include "cuda_neigh_list.h"
-#include "user_cuda.h"
-#include "neighbor_cu.h"
-#include <cmath>
-using namespace LAMMPS_NS;
-
-/* ----------------------------------------------------------------------
- N^2 search for all neighbors
- every neighbor pair appears in list of both atoms i and j
-------------------------------------------------------------------------- */
-void NeighborCuda::full_bin_cuda(NeighList *list)
-{
- MYDBG(printf(" # CUDA::NeighFullBinCuda ... start\n");)
- if(includegroup) error->warning(FLERR,"Warning using inlcudegroup neighborbuild. This is not yet supported by CUDA neighborbuild styles.\n");
- int nlocal = atom->nlocal;
- int nall = nlocal + atom->nghost;
-
- if(nlocal==0) return;
- CudaNeighList* clist=list->cuda_list;
- cuda_shared_neighlist* slist=&clist->sneighlist;
-
- if(not clist) cuda->registerNeighborList(list);
-
- clist->build_cuda=true;
-
- if(slist->bin_extraspace<0.09)
- {
- for(int i=1;i<=atom->ntypes;i++)
- for(int j=1;j<=atom->ntypes;j++)
- {
- if(slist->maxcut<cutneighsq[i][j]) slist->maxcut=cutneighsq[i][j];
- }
- slist->maxcut=sqrt(slist->maxcut);
- }
- int bin_dim_tmp[3];
- int bin_nmax_tmp;
- do
- {
- do
- {
- bin_dim_tmp[0]=static_cast <int> ((domain->subhi[0]-domain->sublo[0])/slist->maxcut);
- bin_dim_tmp[1]=static_cast <int> ((domain->subhi[1]-domain->sublo[1])/slist->maxcut);
- bin_dim_tmp[2]=static_cast <int> ((domain->subhi[2]-domain->sublo[2])/slist->maxcut);
- if(bin_dim_tmp[0]==0) bin_dim_tmp[0]+=1;
- if(bin_dim_tmp[1]==0) bin_dim_tmp[1]+=1;
- if(bin_dim_tmp[2]==0) bin_dim_tmp[2]+=1;
- bin_nmax_tmp=static_cast <int> ((1.0+slist->bin_extraspace)*nlocal/(bin_dim_tmp[0]*bin_dim_tmp[1]*bin_dim_tmp[2]));
- bin_dim_tmp[0]+=4;
- bin_dim_tmp[1]+=4;
- bin_dim_tmp[2]+=4;
- if(bin_nmax_tmp<32) slist->maxcut*=1.2;
- // printf("slist->maxcut: %lf\n", slist->maxcut);
- } while(bin_nmax_tmp<32);
- if((slist->bin_dim[0]!=bin_dim_tmp[0])||(slist->bin_dim[1]!=bin_dim_tmp[1])||(slist->bin_dim[2]!=bin_dim_tmp[2])||(slist->bin_nmax!=bin_nmax_tmp))
- {
- if(slist->binned_id!=NULL)
- CudaWrapper_FreeCudaData(slist->binned_id,slist->bin_dim[0]*slist->bin_dim[1]*slist->bin_dim[2]*slist->bin_nmax*sizeof(int));
- slist->bin_dim[0] = bin_dim_tmp[0];
- slist->bin_dim[1] = bin_dim_tmp[1];
- slist->bin_dim[2] = bin_dim_tmp[2];
- slist->bin_nmax = bin_nmax_tmp;
- slist->binned_id=(int*) CudaWrapper_AllocCudaData(slist->bin_dim[0]*slist->bin_dim[1]*slist->bin_dim[2]*slist->bin_nmax*sizeof(int));
- //printf("slist->bin: %i %i %i %i \n", bin_dim_tmp[0],bin_dim_tmp[1],bin_dim_tmp[2],bin_nmax_tmp);
- }
- //if(list->cuda_list->sneighlist.bin_nmax>512) error->all(FLERR,"To many atoms per bin. Likely cause is very long pair cutoff. This needs major rewrite of code and is not yet scheduled to be done.\n");
- }while(Cuda_BinAtoms(&cuda->shared_data, &list->cuda_list->sneighlist));
-
- // cuda->cu_debugdata->memset_device(0);
- int maxneighbors=slist->maxneighbors;
-
- if((nex_type!=slist->nex_type)||
- (nex_group!=slist->nex_group)||
- (nex_mol!=slist->nex_mol))
- {
- slist->nex_type=nex_type;
- slist->nex_group=nex_group;
- slist->nex_mol=nex_mol;
- //printf("%i %i %i\n",nex_type,nex_group,nex_mol);
- if(nex_type)
- {
- delete clist->cu_ex_type;
- clist->cu_ex_type=new cCudaData<int , int , x> (&ex_type[0][0] , & slist->ex_type , (atom->ntypes+1)*(atom->ntypes+1) );
- clist->cu_ex_type->upload();
- }
- //printf("AA %i %i %i\n",nex_type,nex_group,nex_mol);
- if(nex_group)
- {
- delete clist->cu_ex1_bit;
- clist->cu_ex1_bit=new cCudaData<int , int , x> (ex1_bit , & slist->ex1_bit , nex_group );
- clist->cu_ex1_bit->upload();
- //printf("A %i %i %i\n",nex_type,nex_group,nex_mol);
- delete clist->cu_ex2_bit;
- clist->cu_ex2_bit=new cCudaData<int , int , x> (ex2_bit , & slist->ex2_bit , nex_group );
- clist->cu_ex2_bit->upload();
- }
- //printf("B %i %i %i\n",nex_type,nex_group,nex_mol);
- if(nex_mol)
- {
- delete clist->cu_ex_mol_bit;
- clist->cu_ex_mol_bit=new cCudaData<int , int , x> (ex_mol_bit , & slist->ex_mol_bit , nex_mol );
- clist->cu_ex_mol_bit->upload();
- }
- //printf("C %i %i %i\n",nex_type,nex_group,nex_mol);
- }
- int overflow = 0;
- do
- {
- overflow=0;
- clist->grow_device();
- slist->cutneighsq=cutneighsq;
- slist->maxneighbors=maxneighbors;
- slist->inum = list->inum = nlocal;
- //list->cuda_list->grow_device();
- if(cuda->shared_data.overlap_comm)
- {
- list->cuda_list->inum_border=0;
- list->cuda_list->cu_inum_border->upload();
- }
-
- cuda->shared_data.atom.nall=nall;
- //Cuda_NeighborReBuildFirstneigh(&cuda->shared_data, &list->cuda_list->sneighlist);
- overflow= Cuda_NeighborBuildFullBin(&cuda->shared_data, &list->cuda_list->sneighlist);
-
- /*cuda->cu_debugdata->download();
- printf("Debugdata: %i ",cuda->debugdata[0]);
- for(int i=0;i<cuda->debugdata[0];i+=3) printf("// %i %i %i",cuda->debugdata[i+1],cuda->debugdata[i+2],cuda->debugdata[i+3]);
- printf("\n");*/
- //printf("maxneighborsA: %i %i %i %i\n",maxneighbors,pgsize,oneatom,atom->nmax);
-
- if(overflow<0)
- {
- maxneighbors+=32;
- if(-overflow>maxneighbors) maxneighbors=((-overflow+37)/32)*32;
- delete list->cuda_list->cu_neighbors;
- delete [] list->cuda_list->neighbors;
- list->cuda_list->neighbors= new int[slist->maxlocal*maxneighbors];
- list->cuda_list->sneighlist.maxneighbors=maxneighbors;
- //printf("maxneighborsA1: %i %i %i %i %i\n",maxneighbors,pgsize,oneatom,atom->nmax,slist->maxlocal);
- list->cuda_list->cu_neighbors= new cCudaData<int, int, x> (list->cuda_list->neighbors , & list->cuda_list->sneighlist.neighbors, slist->maxlocal*maxneighbors );
- //printf("maxneighborsA2: %i %i %i %i\n",maxneighbors,pgsize,oneatom,atom->nmax);
-
- if(cuda->shared_data.overlap_comm)
- {
- list->cuda_list->sneighlist.maxneighbors=maxneighbors;
- list->cuda_list->dev_free();
- list->cuda_list->dev_alloc();
- }
- //printf("maxneighborsA3: %i %i %i %i\n",maxneighbors,pgsize,oneatom,atom->nmax);
- }
- //printf("maxneighborsB: %i %i %i %i\n",maxneighbors,pgsize,oneatom,atom->nmax);
- if(cuda->shared_data.overlap_comm)
- {
- list->cuda_list->cu_inum_border->download();
- list->cuda_list->sneighlist.inum_border2=list->cuda_list->inum_border;
- }
- }
- while(overflow<0);
-
- //cuda->cu_debugdata->download();
- // printf("Differences in: %i\n",cuda->debugdata[0]);
- // for(int i=0;i<20;i++) printf("%i %i %i %i// ",cuda->debugdata[4*i+1],cuda->debugdata[4*i+2],cuda->debugdata[4*i+3],cuda->debugdata[4*i+4]);
-// printf("\n");
-/*for(int i=0;i<10;i++)
-{
- printf("%i %i // ",i,numneigh[i]);
- for(int j=0;j<numneigh[i];j++)
- printf("%i ",list->cuda_list->neighbors[i+j*nlocal]);
- printf("\n");
-}*/
-/* int count=0;
- if(cuda->shared_data.overlap_comm)
- {
- list->cuda_list->cu_inum_border->download();
- list->cuda_list->cu_ilist_border->download();
- list->cuda_list->cu_numneigh_border->download();
- list->cuda_list->cu_numneigh_inner->download();
- list->cuda_list->cu_neighbors->download();
- list->cuda_list->cu_neighbors_inner->download();
- list->cuda_list->cu_neighbors_border->download();
-
- //list->cuda_list->cu_firstneigh->download();
- // list->cuda_list->nl_download();
- list->cuda_list->cu_numneigh->download();
- int diff=0;
- //for(int i=0;i<nlocal;i++)*/
- /* int i=123;
- {
- int k=-1;
- //printf("inum_border: %i\n",list->cuda_list->inum_border);
- //for(int j=0;j<list->numneigh[i];j++) printf("%i ",list->firstneigh[i][j]);printf("\n");
- for(int j=0;j<list->cuda_list->inum_border;j++)
- if(list->cuda_list->ilist_border[j]==i) k=j;
- int d=numneigh[i]-list->cuda_list->numneigh_inner[i];
- if(k>-1) d-=list->cuda_list->numneigh_border[k];
- if(d!=0) {printf("Error at %i %i %i %i %i\n",i,k,d,numneigh[i],list->cuda_list->numneigh_inner[i]); diff++;}
- if(k>-1 && count<10)
- {
- printf("Numneighs: %i %i %i Border_i: %i %i\n",numneigh[i],list->cuda_list->numneigh_inner[i],list->cuda_list->numneigh_border[k],k,(int)list->cuda_list->cu_ilist_border->dev_data());
- cuda->shared_data.me=k;
- for(int j=0;j<numneigh[i];j++)
- printf("%i ",list->cuda_list->neighbors[i+j*nlocal]);
- printf("\n");
- for(int j=0;j<list->cuda_list->numneigh_inner[i];j++)
- printf("%i ",list->cuda_list->neighbors_inner[i+j*nlocal]);
- printf(" // ");
- for(int j=0;j<list->cuda_list->numneigh_border[k];j++)
- printf("%i ",list->cuda_list->neighbors_border[k+j*nlocal]);
- printf("\n");
- count++;
- }
- }
- printf("%i\n",diff);
- }*/
- list->cuda_list->cu_numneigh->download();
- list->cuda_list->cu_ilist->download();
- cuda->shared_data.atom.update_neigh=2;
- //printf("Done\n");
-
- MYDBG(printf(" # CUDA::NeighFullBinCuda ... end\n");)
-
-}
-
-
-void NeighborCuda::full_nsq_cuda(NeighList *list)
-{
- printf("Full_Nsq cuda neighbor list build is not implemented anymore.\n");
-return;
-/*
- MYDBG(printf(" # CUDA::NeighFullNSQCuda ... start\n");)
- int nlocal = atom->nlocal;
- int nall = nlocal + atom->nghost;
-
- if(cuda->cu_xhold) cuda->cu_xhold->upload();
-
-
- if(not list->cuda_list) cuda->registerNeighborList(list);
- list->cuda_list->build_cuda=true;
- int maxneighbors=list->cuda_list->sneighlist.maxneighbors;
- int neigh_lists_per_page=pgsize/maxneighbors;
- int *ilist = list->ilist;
- int *numneigh = list->numneigh;
- int **firstneigh = list->firstneigh;
- int **pages = list->pages;
-
- int overflow = 0;
- int inum = 0;
- int npage = 0;
- int npnt = 0;
- do
- {
- npage=0;
- npnt=0;
- inum=0;
- overflow=0;
- neigh_lists_per_page=pgsize/maxneighbors;
- npage=(2*nlocal*maxneighbors-1)/pgsize;
- while(npage>list->maxpage) list->add_pages();
- pages = list->pages;
- npage=0;
- list->cuda_list->sneighlist.neigh_lists_per_page=pgsize/maxneighbors;
- list->cuda_list->grow_device();
- list->cuda_list->sneighlist.cutneighsq=cutneighsq;
- list->cuda_list->sneighlist.maxneighbors=maxneighbors;
- list->cuda_list->sneighlist.inum = list->inum = nlocal;
-
- cuda->shared_data.atom.nall=nall;
- Cuda_NeighborReBuildFirstneigh(&cuda->shared_data, &list->cuda_list->sneighlist);
- overflow= not Cuda_NeighborBuildFullNsq(&cuda->shared_data, &list->cuda_list->sneighlist);
-
-
-
- if(overflow) maxneighbors+=32;
- }
- while(overflow);
- if(not cudable) list->cuda_list->nl_download();
- MYDBG(printf(" # CUDA::NeighFullNSQCuda ... end\n");)
- */
-}
diff --git a/src/USER-CUDA/neighbor_cuda.cpp b/src/USER-CUDA/neighbor_cuda.cpp
deleted file mode 100644
index 015e85ff07..0000000000
--- a/src/USER-CUDA/neighbor_cuda.cpp
+++ /dev/null
@@ -1,240 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#include <math.h>
-#include "neighbor_cuda.h"
-#include "user_cuda.h"
-#include "atom.h"
-#include "atom_vec.h"
-#include "domain.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "force.h"
-#include "group.h"
-#include "memory.h"
-#include "error.h"
-#include "update.h"
-
-using namespace LAMMPS_NS;
-
-
-
-
-enum {NSQ, BIN, MULTI}; // also in neigh_list.cpp
-
-/* ---------------------------------------------------------------------- */
-
-NeighborCuda::NeighborCuda(LAMMPS* lmp) : Neighbor(lmp)
-{
- cuda = lmp->cuda;
-
- if(cuda == NULL)
- error->all(FLERR, "You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-}
-
-/* ---------------------------------------------------------------------- */
-
-void NeighborCuda::init()
-{
- cuda->set_neighinit(dist_check, 0.25 * skin * skin);
- cudable = 1;
-
- Neighbor::init();
-}
-
-/* ----------------------------------------------------------------------
- overwrite either full_nsq or full_bin with CUDA-equivalent methods
- any other neighbor build method is unchanged
-------------------------------------------------------------------------- */
-
-void NeighborCuda::choose_build(int index, NeighRequest* rq)
-{
- Neighbor::choose_build(index, rq);
-
- if(rq->full && style == NSQ && rq->cudable)
- pair_build[index] = (Neighbor::PairPtr) &NeighborCuda::full_nsq_cuda;
- else if(rq->full && style == BIN && rq->cudable)
- pair_build[index] = (Neighbor::PairPtr) &NeighborCuda::full_bin_cuda;
-}
-
-/* ---------------------------------------------------------------------- */
-
-int NeighborCuda::check_distance()
-{
- double delx, dely, delz, rsq;
- double delta, deltasq, delta1, delta2;
-
- if(boxcheck) {
- if(triclinic == 0) {
- delx = bboxlo[0] - boxlo_hold[0];
- dely = bboxlo[1] - boxlo_hold[1];
- delz = bboxlo[2] - boxlo_hold[2];
- delta1 = sqrt(delx * delx + dely * dely + delz * delz);
- delx = bboxhi[0] - boxhi_hold[0];
- dely = bboxhi[1] - boxhi_hold[1];
- delz = bboxhi[2] - boxhi_hold[2];
- delta2 = sqrt(delx * delx + dely * dely + delz * delz);
- delta = 0.5 * (skin - (delta1 + delta2));
- deltasq = delta * delta;
- } else {
- domain->box_corners();
- delta1 = delta2 = 0.0;
-
- for(int i = 0; i < 8; i++) {
- delx = corners[i][0] - corners_hold[i][0];
- dely = corners[i][1] - corners_hold[i][1];
- delz = corners[i][2] - corners_hold[i][2];
- delta = sqrt(delx * delx + dely * dely + delz * delz);
-
- if(delta > delta1) delta1 = delta;
- else if(delta > delta2) delta2 = delta;
- }
-
- delta = 0.5 * (skin - (delta1 + delta2));
- deltasq = delta * delta;
- }
- } else deltasq = triggersq;
-
- double** x = atom->x;
- int nlocal = atom->nlocal;
-
- if(includegroup) nlocal = atom->nfirst;
-
- int flag = 0;
-
- if(not cuda->neighbor_decide_by_integrator) {
- cuda->cu_x_download();
-
- for(int i = 0; i < nlocal; i++) {
- delx = x[i][0] - xhold[i][0];
- dely = x[i][1] - xhold[i][1];
- delz = x[i][2] - xhold[i][2];
- rsq = delx * delx + dely * dely + delz * delz;
-
- if(rsq > deltasq) flag = 1;
- }
- } else flag = cuda->shared_data.atom.reneigh_flag;
-
- int flagall;
- MPI_Allreduce(&flag, &flagall, 1, MPI_INT, MPI_MAX, world);
-
- if(flagall && ago == MAX(every, delay)) ndanger++;
-
- return flagall;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void NeighborCuda::build(int topoflag)
-{
- int i;
-
- ago = 0;
- ncalls++;
- lastcall = update->ntimestep;
- // store current atom positions and box size if needed
-
- if(dist_check) {
- if(cuda->decide_by_integrator())
- cuda->update_xhold(maxhold, &xhold[0][0]);
- else {
- if(cuda->finished_setup) cuda->cu_x_download();
-
- double** x = atom->x;
- int nlocal = atom->nlocal;
-
- if(includegroup) nlocal = atom->nfirst;
-
- if(nlocal > maxhold) {
- maxhold = atom->nmax;
- memory->destroy(xhold);
- memory->create(xhold, maxhold, 3, "neigh:xhold");
- }
-
- for(i = 0; i < nlocal; i++) {
- xhold[i][0] = x[i][0];
- xhold[i][1] = x[i][1];
- xhold[i][2] = x[i][2];
- }
-
- if(boxcheck) {
- if(triclinic == 0) {
- boxlo_hold[0] = bboxlo[0];
- boxlo_hold[1] = bboxlo[1];
- boxlo_hold[2] = bboxlo[2];
- boxhi_hold[0] = bboxhi[0];
- boxhi_hold[1] = bboxhi[1];
- boxhi_hold[2] = bboxhi[2];
- } else {
- domain->box_corners();
- corners = domain->corners;
-
- for(i = 0; i < 8; i++) {
- corners_hold[i][0] = corners[i][0];
- corners_hold[i][1] = corners[i][1];
- corners_hold[i][2] = corners[i][2];
- }
- }
- }
- }
- }
-
- if(not cudable && cuda->finished_setup && atom->avec->cudable)
- cuda->downloadAll();
-
- if(cudable && (not cuda->finished_setup)) {
- cuda->checkResize();
- cuda->uploadAll();
- }
-
- // if any lists store neighbors of ghosts:
- // invoke grow() if nlocal+nghost exceeds previous list size
- // else only invoke grow() if nlocal exceeds previous list size
- // only done for lists with growflag set and which are perpetual
-
- if(anyghostlist && atom->nlocal + atom->nghost > maxatom) {
- maxatom = atom->nmax;
-
- for(i = 0; i < nglist; i++) lists[glist[i]]->grow(maxatom);
- } else if(atom->nlocal > maxatom) {
- maxatom = atom->nmax;
-
- for(i = 0; i < nglist; i++) lists[glist[i]]->grow(maxatom);
- }
-
- // extend atom bin list if necessary
-
- if(style != NSQ && atom->nmax > maxbin) {
- maxbin = atom->nmax;
- memory->destroy(bins);
- memory->create(bins, maxbin, "bins");
- }
-
- // check that neighbor list with special bond flags will not overflow
-
- if(atom->nlocal + atom->nghost > NEIGHMASK)
- error->one(FLERR, "Too many local+ghost atoms for neighbor list");
-
- // invoke building of pair and molecular neighbor lists
- // only for pairwise lists with buildflag set
-
- for(i = 0; i < nblist; i++)
- (this->*pair_build[blist[i]])(lists[blist[i]]);
-
- if(atom->molecular && topoflag) {
- if(force->bond)(this->*bond_build)();
- if(force->angle)(this->*angle_build)();
- if(force->dihedral)(this->*dihedral_build)();
- if(force->improper)(this->*improper_build)();
- }
-}
diff --git a/src/USER-CUDA/neighbor_cuda.h b/src/USER-CUDA/neighbor_cuda.h
deleted file mode 100644
index 708f45fde2..0000000000
--- a/src/USER-CUDA/neighbor_cuda.h
+++ /dev/null
@@ -1,39 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifndef LMP_NEIGHBOR_CUDA_H
-#define LMP_NEIGHBOR_CUDA_H
-
-#include "neighbor.h"
-
-namespace LAMMPS_NS {
-
-class NeighborCuda : public Neighbor {
- public:
- NeighborCuda(class LAMMPS *);
- void init();
- int check_distance();
- void build(int do_build_bonded=1);
-
- private:
- class Cuda *cuda;
-
- void choose_build(int, class NeighRequest *);
- typedef void (NeighborCuda::*PairPtr)(class NeighList *);
- void full_nsq_cuda(class NeighList *);
- void full_bin_cuda(class NeighList *);
-};
-
-}
-
-#endif
diff --git a/src/USER-CUDA/pair_born_coul_long_cuda.cpp b/src/USER-CUDA/pair_born_coul_long_cuda.cpp
deleted file mode 100644
index d01a5bf47f..0000000000
--- a/src/USER-CUDA/pair_born_coul_long_cuda.cpp
+++ /dev/null
@@ -1,183 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_born_coul_long_cuda.h"
-#include "pair_born_coul_long_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-#define EWALD_F 1.12837917
-#define EWALD_P 0.3275911
-#define A1 0.254829592
-#define A2 -0.284496736
-#define A3 1.421413741
-#define A4 -1.453152027
-#define A5 1.061405429
-/* ---------------------------------------------------------------------- */
-
-PairBornCoulLongCuda::PairBornCoulLongCuda(LAMMPS *lmp) : PairBornCoulLong(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairBornCoulLongCuda::allocate()
-{
- if(! allocated) PairBornCoulLong::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.coeff1 = rhoinv;
- cuda->shared_data.pair.coeff2 = sigma;
- cuda->shared_data.pair.coeff3 = a;
- cuda->shared_data.pair.coeff4 = c;
- cuda->shared_data.pair.coeff5 = d;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBornCoulLongCuda::compute(int eflag, int vflag)
-{
- MYDBG( printf("PairBornCoulLongCuda compute start\n"); fflush(stdout);)
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
- #ifdef CUDA_USE_BINNING
- Cuda_PairBornCoulLongCuda(& cuda->shared_data, eflag, vflag);
- #else
- Cuda_PairBornCoulLongCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
- #endif
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
- MYDBG( printf("PairBornCoulLongCuda compute end\n"); fflush(stdout);)
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBornCoulLongCuda::settings(int narg, char **arg)
-{
- PairBornCoulLong::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBornCoulLongCuda::coeff(int narg, char **arg)
-{
- PairBornCoulLong::coeff(narg, arg);
- allocate();
-}
-
-void PairBornCoulLongCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style born/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (strstr(update->integrate_style,"respa")) error->all(FLERR,"Integrate Style Respa is not supported by pair style buck/coul/long/cuda");
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
- cut_coulsq = cut_coul * cut_coul;
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
-
- if (force->kspace == NULL)
- error->all(FLERR,"Pair style is incompatible with KSpace style");
- g_ewald = force->kspace->g_ewald;
- cuda->shared_data.pair.g_ewald=g_ewald;
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-
- if(ncoultablebits) error->warning(FLERR,"# CUDA: You asked for the usage of Coulomb Tables. This is not supported in CUDA Pair forces. Setting is ignored.\n");
-}
-
-void PairBornCoulLongCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairBornCoulLongCuda::init_list\n");)
- PairBornCoulLong::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairBornCoulLongCuda::init_list end\n");)
-}
-
-void PairBornCoulLongCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairBornCoulLong::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_born_coul_long_cuda.h b/src/USER-CUDA/pair_born_coul_long_cuda.h
deleted file mode 100644
index 6e4f42cf3b..0000000000
--- a/src/USER-CUDA/pair_born_coul_long_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(born/coul/long/cuda,PairBornCoulLongCuda)
-
-#else
-
-#ifndef LMP_PAIR_BORN_COUL_LONG_CUDA_H
-#define LMP_PAIR_BORN_COUL_LONG_CUDA_H
-
-#include "pair_born_coul_long.h"
-
-namespace LAMMPS_NS {
-
-class PairBornCoulLongCuda : public PairBornCoulLong
-{
- public:
- PairBornCoulLongCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_buck_coul_cut_cuda.cpp b/src/USER-CUDA/pair_buck_coul_cut_cuda.cpp
deleted file mode 100644
index 4291b82752..0000000000
--- a/src/USER-CUDA/pair_buck_coul_cut_cuda.cpp
+++ /dev/null
@@ -1,170 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_buck_coul_cut_cuda.h"
-#include "pair_buck_coul_cut_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairBuckCoulCutCuda::PairBuckCoulCutCuda(LAMMPS *lmp) : PairBuckCoulCut(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairBuckCoulCutCuda::allocate()
-{
- if(! allocated) PairBuckCoulCut::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut_coul = cut_coul;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.coeff1 = rhoinv;
- cuda->shared_data.pair.coeff2 = buck1;
- cuda->shared_data.pair.coeff3 = buck2;
- cuda->shared_data.pair.coeff4 = a;
- cuda->shared_data.pair.coeff5 = c;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCoulCutCuda::compute(int eflag, int vflag)
-{
- MYDBG( printf("PairBuckCoulCutCuda compute start\n"); fflush(stdout);)
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairBuckCoulCutCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
- MYDBG( printf("PairBuckCoulCutCuda compute end\n"); fflush(stdout);)
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCoulCutCuda::settings(int narg, char **arg)
-{
- PairBuckCoulCut::settings(narg, arg);
- cuda->shared_data.pair.cut_coul_global = (F_CFLOAT) cut_coul_global;
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCoulCutCuda::coeff(int narg, char **arg)
-{
- PairBuckCoulCut::coeff(narg, arg);
- allocate();
-}
-
-void PairBuckCoulCutCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style buck/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (strstr(update->integrate_style,"respa")) error->all(FLERR,"Integrate Style Respa is not supported by pair style buck/coul/long/cuda");
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
- cuda->shared_data.pair.cut_coulsq_global=cut_coul_global * cut_coul_global;
-
- if(ncoultablebits) error->warning(FLERR,"# CUDA: You asked for the usage of Coulomb Tables. This is not supported in CUDA Pair forces. Setting is ignored.\n");
-}
-
-void PairBuckCoulCutCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairBuckCoulCutCuda::init_list\n");)
- PairBuckCoulCut::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairBuckCoulCutCuda::init_list end\n");)
-}
-
-void PairBuckCoulCutCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairBuckCoulCut::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_buck_coul_cut_cuda.h b/src/USER-CUDA/pair_buck_coul_cut_cuda.h
deleted file mode 100644
index f66b70fb00..0000000000
--- a/src/USER-CUDA/pair_buck_coul_cut_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(buck/coul/cut/cuda,PairBuckCoulCutCuda)
-
-#else
-
-#ifndef LMP_PAIR_BUCK_COUL_CUT_CUDA_H
-#define LMP_PAIR_BUCK_COUL_CUT_CUDA_H
-
-#include "pair_buck_coul_cut.h"
-
-namespace LAMMPS_NS {
-
-class PairBuckCoulCutCuda : public PairBuckCoulCut
-{
- public:
- PairBuckCoulCutCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_buck_coul_long_cuda.cpp b/src/USER-CUDA/pair_buck_coul_long_cuda.cpp
deleted file mode 100644
index 8c8d667165..0000000000
--- a/src/USER-CUDA/pair_buck_coul_long_cuda.cpp
+++ /dev/null
@@ -1,181 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_buck_coul_long_cuda.h"
-#include "pair_buck_coul_long_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-#define EWALD_F 1.12837917
-#define EWALD_P 0.3275911
-#define A1 0.254829592
-#define A2 -0.284496736
-#define A3 1.421413741
-#define A4 -1.453152027
-#define A5 1.061405429
-/* ---------------------------------------------------------------------- */
-
-PairBuckCoulLongCuda::PairBuckCoulLongCuda(LAMMPS *lmp) : PairBuckCoulLong(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairBuckCoulLongCuda::allocate()
-{
- if(! allocated) PairBuckCoulLong::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.coeff1 = rhoinv;
- cuda->shared_data.pair.coeff2 = buck1;
- cuda->shared_data.pair.coeff3 = buck2;
- cuda->shared_data.pair.coeff4 = a;
- cuda->shared_data.pair.coeff5 = c;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCoulLongCuda::compute(int eflag, int vflag)
-{
- MYDBG( printf("PairBuckCoulLongCuda compute start\n"); fflush(stdout);)
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairBuckCoulLongCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
- MYDBG( printf("PairBuckCoulLongCuda compute end\n"); fflush(stdout);)
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCoulLongCuda::settings(int narg, char **arg)
-{
- PairBuckCoulLong::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCoulLongCuda::coeff(int narg, char **arg)
-{
- PairBuckCoulLong::coeff(narg, arg);
- allocate();
-}
-
-void PairBuckCoulLongCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style buck/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (strstr(update->integrate_style,"respa")) error->all(FLERR,"Integrate Style Respa is not supported by pair style buck/coul/long/cuda");
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
- cut_coulsq = cut_coul * cut_coul;
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
-
- if (force->kspace == NULL)
- error->all(FLERR,"Pair style is incompatible with KSpace style");
- g_ewald = force->kspace->g_ewald;
- cuda->shared_data.pair.g_ewald=g_ewald;
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-
- if(ncoultablebits) error->warning(FLERR,"# CUDA: You asked for the usage of Coulomb Tables. This is not supported in CUDA Pair forces. Setting is ignored.\n");
-}
-
-void PairBuckCoulLongCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairBuckCoulLongCuda::init_list\n");)
- PairBuckCoulLong::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairBuckCoulLongCuda::init_list end\n");)
-}
-
-void PairBuckCoulLongCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairBuckCoulLong::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_buck_coul_long_cuda.h b/src/USER-CUDA/pair_buck_coul_long_cuda.h
deleted file mode 100644
index 41d4637d9a..0000000000
--- a/src/USER-CUDA/pair_buck_coul_long_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(buck/coul/long/cuda,PairBuckCoulLongCuda)
-
-#else
-
-#ifndef LMP_PAIR_BUCK_COUL_LONG_CUDA_H
-#define LMP_PAIR_BUCK_COUL_LONG_CUDA_H
-
-#include "pair_buck_coul_long.h"
-
-namespace LAMMPS_NS {
-
-class PairBuckCoulLongCuda : public PairBuckCoulLong
-{
- public:
- PairBuckCoulLongCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_buck_cuda.cpp b/src/USER-CUDA/pair_buck_cuda.cpp
deleted file mode 100644
index bcb9314c5f..0000000000
--- a/src/USER-CUDA/pair_buck_cuda.cpp
+++ /dev/null
@@ -1,166 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_buck_cuda.h"
-#include "pair_buck_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairBuckCuda::PairBuckCuda(LAMMPS *lmp) : PairBuck(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairBuckCuda::allocate()
-{
- if(! allocated) PairBuck::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.coeff1 = rhoinv;
- cuda->shared_data.pair.coeff2 = buck1;
- cuda->shared_data.pair.coeff3 = buck2;
- cuda->shared_data.pair.coeff4 = a;
- cuda->shared_data.pair.coeff5 = c;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCuda::compute(int eflag, int vflag)
-{
- MYDBG( printf("PairBuckCuda compute start\n"); fflush(stdout);)
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairBuckCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
- MYDBG( printf("PairBuckCuda compute end\n"); fflush(stdout);)
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCuda::settings(int narg, char **arg)
-{
- PairBuck::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairBuckCuda::coeff(int narg, char **arg)
-{
- PairBuck::coeff(narg, arg);
- allocate();
-}
-
-void PairBuckCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style buck/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (strstr(update->integrate_style,"respa")) error->all(FLERR,"Integrate Style Respa is not supported by pair style buck/coul/long/cuda");
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-
- if(ncoultablebits) error->warning(FLERR,"# CUDA: You asked for the usage of Coulomb Tables. This is not supported in CUDA Pair forces. Setting is ignored.\n");
-}
-
-void PairBuckCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairBuckCuda::init_list\n");)
- PairBuck::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairBuckCuda::init_list end\n");)
-}
-
-void PairBuckCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairBuck::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_buck_cuda.h b/src/USER-CUDA/pair_buck_cuda.h
deleted file mode 100644
index 9dfb742ed0..0000000000
--- a/src/USER-CUDA/pair_buck_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(buck/cuda,PairBuckCuda)
-
-#else
-
-#ifndef LMP_PAIR_BUCK_CUDA_H
-#define LMP_PAIR_BUCK_CUDA_H
-
-#include "pair_buck.h"
-
-namespace LAMMPS_NS {
-
-class PairBuckCuda : public PairBuck
-{
- public:
- PairBuckCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_eam_alloy_cuda.cpp b/src/USER-CUDA/pair_eam_alloy_cuda.cpp
deleted file mode 100644
index c0b76c7e11..0000000000
--- a/src/USER-CUDA/pair_eam_alloy_cuda.cpp
+++ /dev/null
@@ -1,326 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing authors: Stephen Foiles (SNL), Murray Daw (SNL)
-------------------------------------------------------------------------- */
-
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_eam_alloy_cuda.h"
-#include "atom.h"
-#include "comm.h"
-#include "memory.h"
-#include "error.h"
-
-using namespace LAMMPS_NS;
-
-#define MAXLINE 1024
-
-/* ---------------------------------------------------------------------- */
-
-PairEAMAlloyCuda::PairEAMAlloyCuda(LAMMPS *lmp) : PairEAMCuda(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- one_coeff = 1;
-}
-
-/* ----------------------------------------------------------------------
- set coeffs for one or more type pairs
- read DYNAMO setfl file
-------------------------------------------------------------------------- */
-
-void PairEAMAlloyCuda::coeff(int narg, char **arg)
-{
- int i,j;
-
- if (!allocated) allocate();
-
- if (narg != 3 + atom->ntypes)
- error->all(FLERR,"Incorrect args for pair coefficients");
-
- // insure I,J args are * *
-
- if (strcmp(arg[0],"*") != 0 || strcmp(arg[1],"*") != 0)
- error->all(FLERR,"Incorrect args for pair coefficients");
-
- // read EAM setfl file
-
- if (setfl) {
- for (i = 0; i < setfl->nelements; i++) delete [] setfl->elements[i];
- delete [] setfl->elements;
- delete [] setfl->mass;
- memory->destroy(setfl->frho);
- memory->destroy(setfl->rhor);
- memory->destroy(setfl->z2r);
- delete setfl;
- }
- setfl = new Setfl();
- read_file(arg[2]);
-
- // read args that map atom types to elements in potential file
- // map[i] = which element the Ith atom type is, -1 if NULL
-
- for (i = 3; i < narg; i++) {
- if (strcmp(arg[i],"NULL") == 0) {
- map[i-2] = -1;
- continue;
- }
- for (j = 0; j < setfl->nelements; j++)
- if (strcmp(arg[i],setfl->elements[j]) == 0) break;
- if (j < setfl->nelements) map[i-2] = j;
- else error->all(FLERR,"No matching element in EAM potential file");
- }
-
- // clear setflag since coeff() called once with I,J = * *
-
- int n = atom->ntypes;
- for (i = 1; i <= n; i++)
- for (j = i; j <= n; j++)
- setflag[i][j] = 0;
-
- // set setflag i,j for type pairs where both are mapped to elements
- // set mass of atom type if i = j
-
- int count = 0;
- for (i = 1; i <= n; i++) {
- for (j = i; j <= n; j++) {
- if (map[i] >= 0 && map[j] >= 0) {
- setflag[i][j] = 1;
- if (i == j) atom->set_mass(i,setfl->mass[map[i]]);
- count++;
- }
- }
- }
-
- if (count == 0) error->all(FLERR,"Incorrect args for pair coefficients");
-}
-
-/* ----------------------------------------------------------------------
- read a multi-element DYNAMO setfl file
-------------------------------------------------------------------------- */
-
-void PairEAMAlloyCuda::read_file(char *filename)
-{
- Setfl *file = setfl;
-
- // open potential file
-
- int me = comm->me;
- FILE *fptr;
- char line[MAXLINE];
-
- if (me == 0) {
- fptr = fopen(filename,"r");
- if (fptr == NULL) {
- char str[128];
- sprintf(str,"Cannot open EAM potential file %s",filename);
- error->one(FLERR,str);
- }
- }
-
- // read and broadcast header
- // extract element names from nelements line
-
- int n;
- if (me == 0) {
- fgets(line,MAXLINE,fptr);
- fgets(line,MAXLINE,fptr);
- fgets(line,MAXLINE,fptr);
- fgets(line,MAXLINE,fptr);
- n = strlen(line) + 1;
- }
- MPI_Bcast(&n,1,MPI_INT,0,world);
- MPI_Bcast(line,n,MPI_CHAR,0,world);
-
- sscanf(line,"%d",&file->nelements);
- int nwords = atom->count_words(line);
- if (nwords != file->nelements + 1)
- error->all(FLERR,"Incorrect element names in EAM potential file");
-
- char **words = new char*[file->nelements+1];
- nwords = 0;
- strtok(line," \t\n\r\f");
- while ((words[nwords++] = strtok(NULL," \t\n\r\f"))) continue;
-
- file->elements = new char*[file->nelements];
- for (int i = 0; i < file->nelements; i++) {
- n = strlen(words[i]) + 1;
- file->elements[i] = new char[n];
- strcpy(file->elements[i],words[i]);
- }
- delete [] words;
-
- if (me == 0) {
- fgets(line,MAXLINE,fptr);
- sscanf(line,"%d %lg %d %lg %lg",
- &file->nrho,&file->drho,&file->nr,&file->dr,&file->cut);
- }
-
- MPI_Bcast(&file->nrho,1,MPI_INT,0,world);
- MPI_Bcast(&file->drho,1,MPI_DOUBLE,0,world);
- MPI_Bcast(&file->nr,1,MPI_INT,0,world);
- MPI_Bcast(&file->dr,1,MPI_DOUBLE,0,world);
- MPI_Bcast(&file->cut,1,MPI_DOUBLE,0,world);
-
- file->mass = new double[file->nelements];
- memory->create(file->frho,file->nelements,file->nrho+1,"pair:frho");
- memory->create(file->rhor,file->nelements,file->nr+1,"pair:rhor");
- memory->create(file->z2r,file->nelements,file->nelements,file->nr+1,
- "pair:z2r");
- int i,j,tmp;
- for (i = 0; i < file->nelements; i++) {
- if (me == 0) {
- fgets(line,MAXLINE,fptr);
- sscanf(line,"%d %lg",&tmp,&file->mass[i]);
- }
- MPI_Bcast(&file->mass[i],1,MPI_DOUBLE,0,world);
-
- if (me == 0) grab(fptr,file->nrho,&file->frho[i][1]);
- MPI_Bcast(&file->frho[i][1],file->nrho,MPI_DOUBLE,0,world);
- if (me == 0) grab(fptr,file->nr,&file->rhor[i][1]);
- MPI_Bcast(&file->rhor[i][1],file->nr,MPI_DOUBLE,0,world);
- }
-
- for (i = 0; i < file->nelements; i++)
- for (j = 0; j <= i; j++) {
- if (me == 0) grab(fptr,file->nr,&file->z2r[i][j][1]);
- MPI_Bcast(&file->z2r[i][j][1],file->nr,MPI_DOUBLE,0,world);
- }
-
- // close the potential file
-
- if (me == 0) fclose(fptr);
-}
-
-/* ----------------------------------------------------------------------
- copy read-in setfl potential to standard array format
-------------------------------------------------------------------------- */
-
-void PairEAMAlloyCuda::file2array()
-{
- int i,j,m,n;
- int ntypes = atom->ntypes;
-
- // set function params directly from setfl file
-
- nrho = setfl->nrho;
- nr = setfl->nr;
- drho = setfl->drho;
- dr = setfl->dr;
-
- // ------------------------------------------------------------------
- // setup frho arrays
- // ------------------------------------------------------------------
-
- // allocate frho arrays
- // nfrho = # of setfl elements + 1 for zero array
-
- nfrho = setfl->nelements + 1;
- memory->destroy(frho);
- memory->create(frho,nfrho,nrho+1,"pair:frho");
-
- // copy each element's frho to global frho
-
- for (i = 0; i < setfl->nelements; i++)
- for (m = 1; m <= nrho; m++) frho[i][m] = setfl->frho[i][m];
-
- // add extra frho of zeroes for non-EAM types to point to (pair hybrid)
- // this is necessary b/c fp is still computed for non-EAM atoms
-
- for (m = 1; m <= nrho; m++) frho[nfrho-1][m] = 0.0;
-
- // type2frho[i] = which frho array (0 to nfrho-1) each atom type maps to
- // if atom type doesn't point to element (non-EAM atom in pair hybrid)
- // then map it to last frho array of zeroes
-
- for (i = 1; i <= ntypes; i++)
- if (map[i] >= 0) type2frho[i] = map[i];
- else type2frho[i] = nfrho-1;
-
- // ------------------------------------------------------------------
- // setup rhor arrays
- // ------------------------------------------------------------------
-
- // allocate rhor arrays
- // nrhor = # of setfl elements
-
- nrhor = setfl->nelements;
- memory->destroy(rhor);
- memory->create(rhor,nrhor,nr+1,"pair:rhor");
-
- // copy each element's rhor to global rhor
-
- for (i = 0; i < setfl->nelements; i++)
- for (m = 1; m <= nr; m++) rhor[i][m] = setfl->rhor[i][m];
-
- // type2rhor[i][j] = which rhor array (0 to nrhor-1) each type pair maps to
- // for setfl files, I,J mapping only depends on I
- // OK if map = -1 (non-EAM atom in pair hybrid) b/c type2rhor not used
-
- for (i = 1; i <= ntypes; i++)
- for (j = 1; j <= ntypes; j++)
- type2rhor[i][j] = map[i];
-
- // ------------------------------------------------------------------
- // setup z2r arrays
- // ------------------------------------------------------------------
-
- // allocate z2r arrays
- // nz2r = N*(N+1)/2 where N = # of setfl elements
-
- nz2r = setfl->nelements * (setfl->nelements+1) / 2;
- memory->destroy(z2r);
- memory->create(z2r,nz2r,nr+1,"pair:z2r");
-
- // copy each element pair z2r to global z2r, only for I >= J
-
- n = 0;
- for (i = 0; i < setfl->nelements; i++)
- for (j = 0; j <= i; j++) {
- for (m = 1; m <= nr; m++) z2r[n][m] = setfl->z2r[i][j][m];
- n++;
- }
-
- // type2z2r[i][j] = which z2r array (0 to nz2r-1) each type pair maps to
- // set of z2r arrays only fill lower triangular Nelement matrix
- // value = n = sum over rows of lower-triangular matrix until reach irow,icol
- // swap indices when irow < icol to stay lower triangular
- // if map = -1 (non-EAM atom in pair hybrid):
- // type2z2r is not used by non-opt
- // but set type2z2r to 0 since accessed by opt
-
- int irow,icol;
- for (i = 1; i <= ntypes; i++) {
- for (j = 1; j <= ntypes; j++) {
- irow = map[i];
- icol = map[j];
- if (irow == -1 || icol == -1) {
- type2z2r[i][j] = 0;
- continue;
- }
- if (irow < icol) {
- irow = map[j];
- icol = map[i];
- }
- n = 0;
- for (m = 0; m < irow; m++) n += m + 1;
- n += icol;
- type2z2r[i][j] = n;
- }
- }
-}
diff --git a/src/USER-CUDA/pair_eam_alloy_cuda.h b/src/USER-CUDA/pair_eam_alloy_cuda.h
deleted file mode 100644
index c46755d0f8..0000000000
--- a/src/USER-CUDA/pair_eam_alloy_cuda.h
+++ /dev/null
@@ -1,44 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(eam/alloy/cuda,PairEAMAlloyCuda)
-
-#else
-
-#ifndef LMP_PAIR_EAM_CUDA_ALLOY_H
-#define LMP_PAIR_EAM_CUDA_ALLOY_H
-
-#include "pair_eam_cuda.h"
-
-namespace LAMMPS_NS {
-
-// use virtual public since this class is parent in multiple inheritance
-
-class PairEAMAlloyCuda : virtual public PairEAMCuda {
- public:
- PairEAMAlloyCuda(class LAMMPS *);
- virtual ~PairEAMAlloyCuda() {}
- void coeff(int, char **);
-
- protected:
- class Cuda *cuda;
- void read_file(char *);
- void file2array();
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_eam_cuda.cpp b/src/USER-CUDA/pair_eam_cuda.cpp
deleted file mode 100644
index 3db0c66cd6..0000000000
--- a/src/USER-CUDA/pair_eam_cuda.cpp
+++ /dev/null
@@ -1,265 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_eam_cuda.h"
-#include "pair_eam_cuda_cu.h"
-#include "pair_virial_compute_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairEAMCuda::PairEAMCuda(LAMMPS* lmp) : PairEAM(lmp)
-{
- cuda = lmp->cuda;
-
- if(cuda == NULL)
- error->all(FLERR, "You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.override_block_per_atom = 0;
-
- cuda->setSystemParams();
- cu_rho = NULL;
- cu_fp = NULL;
- cu_frho_spline = NULL;
- cu_z2r_spline = NULL;
- cu_rhor_spline = NULL;
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairEAMCuda::allocate()
-{
- if(! allocated) PairEAM::allocate();
-
- cuda->shared_data.pair.cutsq = cutsq;
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cutforcesq;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairEAMCuda::compute(int eflag, int vflag)
-{
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cutforcesq;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->shared_data.pair.collect_forces_later = 0;
-
- if(atom->nmax > nmax || cuda->finished_setup == false) {
- memory->destroy(rho);
- memory->destroy(fp);
- nmax = atom->nmax;
- memory->create(rho, nmax, "pair:rho");
- memory->create(fp, nmax, "pair:fp");
- delete cu_rho;
- delete cu_fp;
- cu_rho = new cCudaData<double, F_CFLOAT, x> (rho, atom->nmax);
- cu_fp = new cCudaData<double, F_CFLOAT, x> (fp, atom->nmax);
- Cuda_PairEAMCuda_Init(&cuda->shared_data, rdr, rdrho, nfrho, nrhor, nr, nrho, nz2r,
- cu_frho_spline->dev_data(), cu_rhor_spline->dev_data(), cu_z2r_spline->dev_data(),
- cu_rho->dev_data(), cu_fp->dev_data(), type2frho, type2z2r, type2rhor);
- }
-
-
-
- if(eflag || vflag) ev_setup(eflag, vflag);
-
- if(eflag) cuda->cu_eng_vdwl->upload();
-
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairEAM1Cuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- comm->forward_comm_pair(this);
-
- Cuda_PairEAM2Cuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(eflag) cuda->cu_eng_vdwl->download();
-
- if(vflag) cuda->cu_virial->download();
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairEAMCuda::settings(int narg, char** arg)
-{
- PairEAM::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cutforcesq;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairEAMCuda::coeff(int narg, char** arg)
-{
- PairEAM::coeff(narg, arg);
- allocate();
-}
-
-void PairEAMCuda::init_style()
-{
- MYDBG(printf("# CUDA PairEAMCuda::init_style start\n");)
- // request regular or rRESPA neighbor lists
- file2array();
- array2spline();
- int irequest;
-
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- delete cu_rhor_spline;
- delete cu_z2r_spline;
- delete cu_frho_spline;
-
- cu_rhor_spline = new cCudaData<double, F_CFLOAT, xyz>((double*)rhor_spline, nrhor, nr + 1, EAM_COEFF_LENGTH);
- cu_z2r_spline = new cCudaData<double, F_CFLOAT, xyz>((double*)z2r_spline, nz2r, nr + 1, EAM_COEFF_LENGTH);
- cu_frho_spline = new cCudaData<double, F_CFLOAT, xyz>((double*)frho_spline, nfrho, nrho + 1, EAM_COEFF_LENGTH);
-
- cu_rhor_spline->upload();
- cu_z2r_spline->upload();
- cu_frho_spline->upload();
-
- MYDBG(printf("# CUDA PairEAMCuda::init_style end\n");)
-}
-
-void PairEAMCuda::init_list(int id, NeighList* ptr)
-{
- MYDBG(printf("# CUDA PairEAMCuda::init_list\n");)
- PairEAM::init_list(id, ptr);
-
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
-
- // see Neighbor::init() for details on lammps lists' logic
- MYDBG(printf("# CUDA PairEAMCuda::init_list end\n");)
-}
-
-void PairEAMCuda::array2spline()
-{
- rdr = 1.0 / dr;
- rdrho = 1.0 / drho;
-
- memory->destroy(frho_spline);
- memory->destroy(rhor_spline);
- memory->destroy(z2r_spline);
-
- memory->create(frho_spline, nfrho, nrho + 1, 8, "pair:frho");
- memory->create(rhor_spline, nrhor, nr + 1, 8, "pair:rhor");
- memory->create(z2r_spline, nz2r, nr + 1, 8, "pair:z2r");
-
- for(int i = 0; i < nfrho; i++) {
- interpolate(nrho, drho, frho[i], frho_spline[i]);
-
- for(int j = 0; j < nrho + 1; j++)
- frho_spline[i][j][7] = frho_spline[i][j][3];
- }
-
- for(int i = 0; i < nrhor; i++) {
- interpolate(nr, dr, rhor[i], rhor_spline[i]);
-
- for(int j = 0; j < nr + 1; j++)
- rhor_spline[i][j][7] = rhor_spline[i][j][3];
- }
-
- for(int i = 0; i < nz2r; i++) {
- interpolate(nr, dr, z2r[i], z2r_spline[i]);
-
- for(int j = 0; j < nr + 1; j++)
- z2r_spline[i][j][7] = z2r_spline[i][j][3];
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-int PairEAMCuda::pack_forward_comm(int n, int* iswap, double* buf,
- int pbc_flag, int* pbc)
-{
- Cuda_PairEAMCuda_PackComm(&cuda->shared_data, n, *iswap, buf);
-
- if(sizeof(F_CFLOAT) < sizeof(double)) return n;
- else return n;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairEAMCuda::unpack_forward_comm(int n, int first, double* buf)
-{
- Cuda_PairEAMCuda_UnpackComm(&cuda->shared_data, n, first, buf, cu_fp->dev_data());
-}
-
-void PairEAMCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold = maxeatom;
- PairEAM::ev_setup(eflag, vflag);
-
- if(eflag_atom && atom->nmax > maxeatomold) {
- delete cuda->cu_eatom;
- cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax);
- }
-
- if(vflag_atom && atom->nmax > maxeatomold) {
- delete cuda->cu_vatom;
- cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6);
- }
-
-}
diff --git a/src/USER-CUDA/pair_eam_cuda.h b/src/USER-CUDA/pair_eam_cuda.h
deleted file mode 100644
index 973fc20a45..0000000000
--- a/src/USER-CUDA/pair_eam_cuda.h
+++ /dev/null
@@ -1,80 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-#ifdef PAIR_CLASS
-
-PairStyle(eam/cuda,PairEAMCuda)
-
-#else
-
-#ifndef PAIR_EAM_CUDA_H
-#define PAIR_EAM_CUDA_H
-
-#include "cuda_data.h"
-#include "pair_eam.h"
-
-namespace LAMMPS_NS {
-
-class PairEAMCuda : public PairEAM
-{
- public:
- PairEAMCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void array2spline();
- int pack_forward_comm(int n, int *iswap, double *buf,
- int pbc_flag, int *pbc);
- void unpack_forward_comm(int n, int first, double *buf);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- virtual void ev_setup(int eflag, int vflag);
- class CudaNeighList* cuda_neigh_list;
- cCudaData<double, F_CFLOAT, x>* cu_rho;
- cCudaData<double, F_CFLOAT, x>* cu_fp;
- cCudaData<double, F_CFLOAT, xyz>* cu_rhor_spline;
- cCudaData<double, F_CFLOAT, xyz>* cu_z2r_spline;
- cCudaData<double, F_CFLOAT, xyz>* cu_frho_spline;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_eam_fs_cuda.cpp b/src/USER-CUDA/pair_eam_fs_cuda.cpp
deleted file mode 100644
index 6190213402..0000000000
--- a/src/USER-CUDA/pair_eam_fs_cuda.cpp
+++ /dev/null
@@ -1,335 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing authors: Tim Lau (MIT)
-------------------------------------------------------------------------- */
-
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_eam_fs_cuda.h"
-#include "atom.h"
-#include "comm.h"
-#include "memory.h"
-#include "error.h"
-
-using namespace LAMMPS_NS;
-
-#define MAXLINE 1024
-
-/* ---------------------------------------------------------------------- */
-
-PairEAMFSCuda::PairEAMFSCuda(LAMMPS *lmp) : PairEAMCuda(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- one_coeff = 1;
-}
-
-/* ----------------------------------------------------------------------
- set coeffs for one or more type pairs
- read EAM Finnis-Sinclair file
-------------------------------------------------------------------------- */
-
-void PairEAMFSCuda::coeff(int narg, char **arg)
-{
- int i,j;
-
- if (!allocated) allocate();
-
- if (narg != 3 + atom->ntypes)
- error->all(FLERR,"Incorrect args for pair coefficients");
-
- // insure I,J args are * *
-
- if (strcmp(arg[0],"*") != 0 || strcmp(arg[1],"*") != 0)
- error->all(FLERR,"Incorrect args for pair coefficients");
-
- // read EAM Finnis-Sinclair file
-
- if (fs) {
- for (i = 0; i < fs->nelements; i++) delete [] fs->elements[i];
- delete [] fs->elements;
- delete [] fs->mass;
- memory->destroy(fs->frho);
- memory->destroy(fs->rhor);
- memory->destroy(fs->z2r);
- delete fs;
- }
- fs = new Fs();
- read_file(arg[2]);
-
- // read args that map atom types to elements in potential file
- // map[i] = which element the Ith atom type is, -1 if NULL
-
- for (i = 3; i < narg; i++) {
- if (strcmp(arg[i],"NULL") == 0) {
- map[i-2] = -1;
- continue;
- }
- for (j = 0; j < fs->nelements; j++)
- if (strcmp(arg[i],fs->elements[j]) == 0) break;
- if (j < fs->nelements) map[i-2] = j;
- else error->all(FLERR,"No matching element in EAM potential file");
- }
-
- // clear setflag since coeff() called once with I,J = * *
-
- int n = atom->ntypes;
- for (i = 1; i <= n; i++)
- for (j = i; j <= n; j++)
- setflag[i][j] = 0;
-
- // set setflag i,j for type pairs where both are mapped to elements
- // set mass of atom type if i = j
-
- int count = 0;
- for (i = 1; i <= n; i++) {
- for (j = i; j <= n; j++) {
- if (map[i] >= 0 && map[j] >= 0) {
- setflag[i][j] = 1;
- if (i == j) atom->set_mass(i,fs->mass[map[i]]);
- count++;
- }
- }
- }
-
- if (count == 0) error->all(FLERR,"Incorrect args for pair coefficients");
-}
-
-/* ----------------------------------------------------------------------
- read a multi-element DYNAMO setfl file
-------------------------------------------------------------------------- */
-
-void PairEAMFSCuda::read_file(char *filename)
-{
- Fs *file = fs;
-
- // open potential file
-
- int me = comm->me;
- FILE *fptr;
- char line[MAXLINE];
-
- if (me == 0) {
- fptr = fopen(filename,"r");
- if (fptr == NULL) {
- char str[128];
- sprintf(str,"Cannot open EAM potential file %s",filename);
- error->one(FLERR,str);
- }
- }
-
- // read and broadcast header
- // extract element names from nelements line
-
- int n;
- if (me == 0) {
- fgets(line,MAXLINE,fptr);
- fgets(line,MAXLINE,fptr);
- fgets(line,MAXLINE,fptr);
- fgets(line,MAXLINE,fptr);
- n = strlen(line) + 1;
- }
- MPI_Bcast(&n,1,MPI_INT,0,world);
- MPI_Bcast(line,n,MPI_CHAR,0,world);
-
- sscanf(line,"%d",&file->nelements);
- int nwords = atom->count_words(line);
- if (nwords != file->nelements + 1)
- error->all(FLERR,"Incorrect element names in EAM potential file");
-
- char **words = new char*[file->nelements+1];
- nwords = 0;
- strtok(line," \t\n\r\f");
- while ((words[nwords++] = strtok(NULL," \t\n\r\f"))) continue;
-
- file->elements = new char*[file->nelements];
- for (int i = 0; i < file->nelements; i++) {
- n = strlen(words[i]) + 1;
- file->elements[i] = new char[n];
- strcpy(file->elements[i],words[i]);
- }
- delete [] words;
-
- if (me == 0) {
- fgets(line,MAXLINE,fptr);
- sscanf(line,"%d %lg %d %lg %lg",
- &file->nrho,&file->drho,&file->nr,&file->dr,&file->cut);
- }
-
- MPI_Bcast(&file->nrho,1,MPI_INT,0,world);
- MPI_Bcast(&file->drho,1,MPI_DOUBLE,0,world);
- MPI_Bcast(&file->nr,1,MPI_INT,0,world);
- MPI_Bcast(&file->dr,1,MPI_DOUBLE,0,world);
- MPI_Bcast(&file->cut,1,MPI_DOUBLE,0,world);
-
- file->mass = new double[file->nelements];
- memory->create(file->frho,file->nelements,file->nrho+1,
- "pair:frho");
- memory->create(file->rhor,file->nelements,file->nelements,
- file->nr+1,"pair:rhor");
- memory->create(file->z2r,file->nelements,file->nelements,
- file->nr+1,"pair:z2r");
- int i,j,tmp;
- for (i = 0; i < file->nelements; i++) {
- if (me == 0) {
- fgets(line,MAXLINE,fptr);
- sscanf(line,"%d %lg",&tmp,&file->mass[i]);
- }
- MPI_Bcast(&file->mass[i],1,MPI_DOUBLE,0,world);
-
- if (me == 0) grab(fptr,file->nrho,&file->frho[i][1]);
- MPI_Bcast(&file->frho[i][1],file->nrho,MPI_DOUBLE,0,world);
-
- for (j = 0; j < file->nelements; j++) {
- if (me == 0) grab(fptr,file->nr,&file->rhor[i][j][1]);
- MPI_Bcast(&file->rhor[i][j][1],file->nr,MPI_DOUBLE,0,world);
- }
- }
-
- for (i = 0; i < file->nelements; i++)
- for (j = 0; j <= i; j++) {
- if (me == 0) grab(fptr,file->nr,&file->z2r[i][j][1]);
- MPI_Bcast(&file->z2r[i][j][1],file->nr,MPI_DOUBLE,0,world);
- }
-
- // close the potential file
-
- if (me == 0) fclose(fptr);
-}
-
-/* ----------------------------------------------------------------------
- copy read-in setfl potential to standard array format
-------------------------------------------------------------------------- */
-
-void PairEAMFSCuda::file2array()
-{
- int i,j,m,n;
- int ntypes = atom->ntypes;
-
- // set function params directly from fs file
-
- nrho = fs->nrho;
- nr = fs->nr;
- drho = fs->drho;
- dr = fs->dr;
-
- // ------------------------------------------------------------------
- // setup frho arrays
- // ------------------------------------------------------------------
-
- // allocate frho arrays
- // nfrho = # of fs elements + 1 for zero array
-
- nfrho = fs->nelements + 1;
- memory->destroy(frho);
- memory->create(frho,nfrho,nrho+1,"pair:frho");
-
- // copy each element's frho to global frho
-
- for (i = 0; i < fs->nelements; i++)
- for (m = 1; m <= nrho; m++) frho[i][m] = fs->frho[i][m];
-
- // add extra frho of zeroes for non-EAM types to point to (pair hybrid)
- // this is necessary b/c fp is still computed for non-EAM atoms
-
- for (m = 1; m <= nrho; m++) frho[nfrho-1][m] = 0.0;
-
- // type2frho[i] = which frho array (0 to nfrho-1) each atom type maps to
- // if atom type doesn't point to element (non-EAM atom in pair hybrid)
- // then map it to last frho array of zeroes
-
- for (i = 1; i <= ntypes; i++)
- if (map[i] >= 0) type2frho[i] = map[i];
- else type2frho[i] = nfrho-1;
-
- // ------------------------------------------------------------------
- // setup rhor arrays
- // ------------------------------------------------------------------
-
- // allocate rhor arrays
- // nrhor = square of # of fs elements
-
- nrhor = fs->nelements * fs->nelements;
- memory->destroy(rhor);
- memory->create(rhor,nrhor,nr+1,"pair:rhor");
-
- // copy each element pair rhor to global rhor
-
- n = 0;
- for (i = 0; i < fs->nelements; i++)
- for (j = 0; j < fs->nelements; j++) {
- for (m = 1; m <= nr; m++) rhor[n][m] = fs->rhor[i][j][m];
- n++;
- }
-
- // type2rhor[i][j] = which rhor array (0 to nrhor-1) each type pair maps to
- // for fs files, there is a full NxN set of rhor arrays
- // OK if map = -1 (non-EAM atom in pair hybrid) b/c type2rhor not used
-
- for (i = 1; i <= ntypes; i++)
- for (j = 1; j <= ntypes; j++)
- type2rhor[i][j] = map[i] * fs->nelements + map[j];
-
- // ------------------------------------------------------------------
- // setup z2r arrays
- // ------------------------------------------------------------------
-
- // allocate z2r arrays
- // nz2r = N*(N+1)/2 where N = # of fs elements
-
- nz2r = fs->nelements * (fs->nelements+1) / 2;
- memory->destroy(z2r);
- memory->create(z2r,nz2r,nr+1,"pair:z2r");
-
- // copy each element pair z2r to global z2r, only for I >= J
-
- n = 0;
- for (i = 0; i < fs->nelements; i++)
- for (j = 0; j <= i; j++) {
- for (m = 1; m <= nr; m++) z2r[n][m] = fs->z2r[i][j][m];
- n++;
- }
-
- // type2z2r[i][j] = which z2r array (0 to nz2r-1) each type pair maps to
- // set of z2r arrays only fill lower triangular Nelement matrix
- // value = n = sum over rows of lower-triangular matrix until reach irow,icol
- // swap indices when irow < icol to stay lower triangular
- // if map = -1 (non-EAM atom in pair hybrid):
- // type2z2r is not used by non-opt
- // but set type2z2r to 0 since accessed by opt
-
- int irow,icol;
- for (i = 1; i <= ntypes; i++) {
- for (j = 1; j <= ntypes; j++) {
- irow = map[i];
- icol = map[j];
- if (irow == -1 || icol == -1) {
- type2z2r[i][j] = 0;
- continue;
- }
- if (irow < icol) {
- irow = map[j];
- icol = map[i];
- }
- n = 0;
- for (m = 0; m < irow; m++) n += m + 1;
- n += icol;
- type2z2r[i][j] = n;
- }
- }
-}
diff --git a/src/USER-CUDA/pair_eam_fs_cuda.h b/src/USER-CUDA/pair_eam_fs_cuda.h
deleted file mode 100644
index 698b485875..0000000000
--- a/src/USER-CUDA/pair_eam_fs_cuda.h
+++ /dev/null
@@ -1,44 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(eam/fs/cuda,PairEAMFSCuda)
-
-#else
-
-#ifndef LMP_PAIR_EAM_FS_CUDA_H
-#define LMP_PAIR_EAM_FS_CUDA_H
-
-#include "pair_eam_cuda.h"
-
-namespace LAMMPS_NS {
-
-// use virtual public since this class is parent in multiple inheritance
-
-class PairEAMFSCuda : virtual public PairEAMCuda {
- public:
- PairEAMFSCuda(class LAMMPS *);
- virtual ~PairEAMFSCuda() {}
- void coeff(int, char **);
-
- protected:
- class Cuda *cuda;
- void read_file(char *);
- void file2array();
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_gran_hooke_cuda.cpp b/src/USER-CUDA/pair_gran_hooke_cuda.cpp
deleted file mode 100644
index 3f60475ad5..0000000000
--- a/src/USER-CUDA/pair_gran_hooke_cuda.cpp
+++ /dev/null
@@ -1,250 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_gran_hooke_cuda.h"
-#include "pair_gran_hooke_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "modify.h"
-#include "fix_pour.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairGranHookeCuda::PairGranHookeCuda(LAMMPS *lmp) : PairGranHooke(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairGranHookeCuda::allocate()
-{
- if(! allocated) PairGranHooke::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- int n = atom->ntypes;
- cuda->shared_data.pair.cutsq = cutsq;
- memory->create(cuda->shared_data.pair.coeff1,n+1,n+1,
- "pair:cuda_coeff1");
- memory->create(cuda->shared_data.pair.coeff2,
- n+1,n+1,"pair:cuda_coeff2");
- cuda->shared_data.pair.coeff1[0][0]=kn;
- cuda->shared_data.pair.coeff1[0][1]=kt;
- cuda->shared_data.pair.coeff1[1][0]=gamman;
- cuda->shared_data.pair.coeff1[1][1]=gammat;
- cuda->shared_data.pair.coeff2[0][0]=xmu;
- cuda->shared_data.pair.coeff2[0][1]=dampflag;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairGranHookeCuda::compute(int eflag, int vflag)
-{
- cuda->shared_data.pair.use_block_per_atom = 0;
- //cuda->cu_debugdata->memset_device(0);
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairGranHookeCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
- //cuda->cu_debugdata->download();
- //printf("%lf %lf %lf %lf %lf %lf\n",1.0e-6*cuda->debugdata[0],1.0e-6*cuda->debugdata[1],1.0e-6*cuda->debugdata[2],1.0e-6*cuda->debugdata[3],1.0e-6*cuda->debugdata[4],1.0e-6*cuda->debugdata[5]);
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairGranHookeCuda::settings(int narg, char **arg)
-{
- PairGranHooke::settings(narg, arg);
- }
-
-/* ---------------------------------------------------------------------- */
-
-void PairGranHookeCuda::coeff(int narg, char **arg)
-{
- PairGranHooke::coeff(narg, arg);
- allocate();
-}
-
-void PairGranHookeCuda::init_style()
-{
- int i;
- MYDBG(printf("# CUDA PairGranHookeCuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (update->whichflag == 0 && strstr(update->integrate_style,"respa")) {
-
- }
- else
- {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->gran = 1;
- neighbor->requests[irequest]->cudable = 1;
- //neighbor->style=0; //0=NSQ neighboring
- }
-
- if (!atom->radius_flag || !atom->omega_flag || !atom->torque_flag)
- error->all(FLERR,"Pair granular requires atom attributes radius, omega, torque");
- if (comm->ghost_velocity == 0)
- error->all(FLERR,"Pair granular requires ghost atoms store velocity");
-
- // need a half neigh list and optionally a granular history neigh list
-
- dt = update->dt;
-
- // check for Fix freeze and set freeze_group_bit
-
- for (i = 0; i < modify->nfix; i++)
- if (strcmp(modify->fix[i]->style,"freeze") == 0) break;
- if (i < modify->nfix) freeze_group_bit = modify->fix[i]->groupbit;
- else freeze_group_bit = 0;
-
- cuda->shared_data.pair.freeze_group_bit=freeze_group_bit;
-
- // check for FixPour and FixDeposit so can extract particle radii
-
- int ipour;
- for (ipour = 0; ipour < modify->nfix; ipour++)
- if (strcmp(modify->fix[ipour]->style,"pour") == 0) break;
- if (ipour == modify->nfix) ipour = -1;
-
- int idep;
- for (idep = 0; idep < modify->nfix; idep++)
- if (strcmp(modify->fix[idep]->style,"deposit") == 0) break;
- if (idep == modify->nfix) idep = -1;
-
- // set maxrad_dynamic and maxrad_frozen for each type
- // include future FixPour and FixDeposit particles as dynamic
-
- int itype;
- for (i = 1; i <= atom->ntypes; i++) {
- onerad_dynamic[i] = onerad_frozen[i] = 0.0;
- if (ipour >= 0) {
- itype = i;
- onerad_dynamic[i] =
- *((double *) modify->fix[ipour]->extract("radius",itype));
- }
- if (idep >= 0) {
- itype = i;
- onerad_dynamic[i] =
- *((double *) modify->fix[idep]->extract("radius",itype));
- }
- }
-
- double *radius = atom->radius;
- int *mask = atom->mask;
- int *type = atom->type;
- int nlocal = atom->nlocal;
-
- for (i = 0; i < nlocal; i++)
- if (mask[i] & freeze_group_bit)
- onerad_frozen[type[i]] = MAX(onerad_frozen[type[i]],radius[i]);
- else
- onerad_dynamic[type[i]] = MAX(onerad_dynamic[type[i]],radius[i]);
-
- MPI_Allreduce(&onerad_dynamic[1],&maxrad_dynamic[1],atom->ntypes,
- MPI_DOUBLE,MPI_MAX,world);
- MPI_Allreduce(&onerad_frozen[1],&maxrad_frozen[1],atom->ntypes,
- MPI_DOUBLE,MPI_MAX,world);
-}
-
-void PairGranHookeCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairGranHookeCuda::init_list\n");)
- PairGranHooke::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairGranHookeCuda::init_list end\n");)
-}
-
-void PairGranHookeCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairGranHooke::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.eatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_gran_hooke_cuda.h b/src/USER-CUDA/pair_gran_hooke_cuda.h
deleted file mode 100644
index 6fa622ab40..0000000000
--- a/src/USER-CUDA/pair_gran_hooke_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(gran/hooke/cuda,PairGranHookeCuda)
-
-#else
-
-#ifndef PAIR_GRAN_HOOKE_CUDA_H
-#define PAIR_GRAN_HOOKE_CUDA_H
-
-#include "pair_gran_hooke.h"
-
-namespace LAMMPS_NS {
-
-class PairGranHookeCuda : public PairGranHooke
-{
- public:
- PairGranHookeCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj96_cut_cuda.cpp b/src/USER-CUDA/pair_lj96_cut_cuda.cpp
deleted file mode 100644
index 7edb722d36..0000000000
--- a/src/USER-CUDA/pair_lj96_cut_cuda.cpp
+++ /dev/null
@@ -1,179 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj96_cut_cuda.h"
-#include "pair_lj96_cut_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJ96CutCuda::PairLJ96CutCuda(LAMMPS *lmp) : PairLJ96Cut(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJ96CutCuda::allocate()
-{
- if(! allocated) PairLJ96Cut::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJ96CutCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJ96CutCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJ96CutCuda::settings(int narg, char **arg)
-{
- PairLJ96Cut::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJ96CutCuda::coeff(int narg, char **arg)
-{
- PairLJ96Cut::coeff(narg, arg);
- allocate();
-}
-
-void PairLJ96CutCuda::init_style()
-{
- MYDBG(printf("# CUDA PairLJ96CutCuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (update->whichflag == 0 && strstr(update->integrate_style,"respa")) {
-
- }
- else
- {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- //neighbor->style=0; //0=NSQ neighboring
- }
-
-
- cut_respa = NULL;
- MYDBG(printf("# CUDA PairLJ96CutCuda::init_style end\n"); )
-}
-
-void PairLJ96CutCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJ96CutCuda::init_list\n");)
- PairLJ96Cut::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJ96CutCuda::init_list end\n");)
-}
-
-void PairLJ96CutCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJ96Cut::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj96_cut_cuda.h b/src/USER-CUDA/pair_lj96_cut_cuda.h
deleted file mode 100644
index 8a8f36e504..0000000000
--- a/src/USER-CUDA/pair_lj96_cut_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj96/cut/cuda,PairLJ96CutCuda)
-
-#else
-
-#ifndef PAIR_LJ96_CUT_CUDA_H
-#define PAIR_LJ96_CUT_CUDA_H
-
-#include "pair_lj96_cut.h"
-
-namespace LAMMPS_NS {
-
-class PairLJ96CutCuda : public PairLJ96Cut
-{
- public:
- PairLJ96CutCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.cpp b/src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.cpp
deleted file mode 100644
index 3a0ad0e288..0000000000
--- a/src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.cpp
+++ /dev/null
@@ -1,188 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_charmm_coul_charmm_cuda.h"
-#include "pair_lj_charmm_coul_charmm_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJCharmmCoulCharmmCuda::PairLJCharmmCoulCharmmCuda(LAMMPS *lmp) : PairLJCharmmCoulCharmm(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmCuda::allocate()
-{
- if(! allocated) PairLJCharmmCoulCharmm::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- cu_lj1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj1, &cuda->shared_data.pair.coeff1_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj2, &cuda->shared_data.pair.coeff2_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj3, &cuda->shared_data.pair.coeff3_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj4, &cuda->shared_data.pair.coeff4_gm, (atom->ntypes+1)*(atom->ntypes+1));
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
- }
-
- Cuda_PairLJCharmmCoulCharmmCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom,denom_lj,cut_coul_innersq,denom_coul);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmCuda::settings(int narg, char **arg)
-{
- PairLJCharmmCoulCharmm::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (X_CFLOAT) cut_lj;
- cuda->shared_data.pair.cut_coulsq_global = (X_CFLOAT) cut_coulsq;
- cuda->shared_data.pair.cut_inner_global = (F_CFLOAT) cut_lj_inner;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmCuda::coeff(int narg, char **arg)
-{
- PairLJCharmmCoulCharmm::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCharmmCoulCharmmCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/charmm/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- if(atom->molecular)
- {
- cuda->shared_data.pair.collect_forces_later = 1;
- }
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- if (cut_lj_inner >= cut_lj || cut_coul_inner >= cut_coul)
- error->all(FLERR,"Pair inner cutoff >= Pair outer cutoff");
-
- cut_lj_innersq = cut_lj_inner * cut_lj_inner;
- cut_ljsq = cut_lj * cut_lj;
- cut_coul_innersq = cut_coul_inner * cut_coul_inner;
- cut_coulsq = cut_coul * cut_coul;
- cut_bothsq = MAX(cut_ljsq,cut_coulsq);
-
- denom_lj = (cut_ljsq-cut_lj_innersq) * (cut_ljsq-cut_lj_innersq) *
- (cut_ljsq-cut_lj_innersq);
- denom_coul = (cut_coulsq-cut_coul_innersq) * (cut_coulsq-cut_coul_innersq) *
- (cut_coulsq-cut_coul_innersq);
-
- cut_coulsq = cut_coul * cut_coul;
-
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-}
-
-void PairLJCharmmCoulCharmmCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCharmmCoulCharmmCuda::init_list\n");)
- PairLJCharmmCoulCharmm::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCharmmCoulCharmmCuda::init_list end\n");)
-}
-
-void PairLJCharmmCoulCharmmCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCharmmCoulCharmm::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.h b/src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.h
deleted file mode 100644
index c19411f03c..0000000000
--- a/src/USER-CUDA/pair_lj_charmm_coul_charmm_cuda.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/charmm/coul/charmm/cuda,PairLJCharmmCoulCharmmCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CHARMM_COUL_CHARMM_CUDA_H
-#define LMP_PAIR_LJ_CHARMM_COUL_CHARMM_CUDA_H
-
-#include "pair_lj_charmm_coul_charmm.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCharmmCoulCharmmCuda : public PairLJCharmmCoulCharmm
-{
- public:
- PairLJCharmmCoulCharmmCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- cCudaData<double , F_CFLOAT , x >* cu_lj1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj4_gm;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.cpp b/src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.cpp
deleted file mode 100644
index c2f2ca871f..0000000000
--- a/src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.cpp
+++ /dev/null
@@ -1,183 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_charmm_coul_charmm_implicit_cuda.h"
-#include "pair_lj_charmm_coul_charmm_implicit_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJCharmmCoulCharmmImplicitCuda::PairLJCharmmCoulCharmmImplicitCuda(LAMMPS *lmp) : PairLJCharmmCoulCharmmImplicit(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.collect_forces_later = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmImplicitCuda::allocate()
-{
- if(! allocated) PairLJCharmmCoulCharmmImplicit::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- cu_lj1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj1, &cuda->shared_data.pair.coeff1_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj2, &cuda->shared_data.pair.coeff2_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj3, &cuda->shared_data.pair.coeff3_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj4, &cuda->shared_data.pair.coeff4_gm, (atom->ntypes+1)*(atom->ntypes+1));
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmImplicitCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
- }
-
- Cuda_PairLJCharmmCoulCharmmImplicitCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom,denom_lj,cut_coul_innersq,denom_coul);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmImplicitCuda::settings(int narg, char **arg)
-{
- PairLJCharmmCoulCharmmImplicit::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (X_CFLOAT) cut_lj;
- cuda->shared_data.pair.cut_coulsq_global = (X_CFLOAT) cut_coulsq;
- cuda->shared_data.pair.cut_inner_global = (F_CFLOAT) cut_lj_inner;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulCharmmImplicitCuda::coeff(int narg, char **arg)
-{
- PairLJCharmmCoulCharmmImplicit::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCharmmCoulCharmmImplicitCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/charmm/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- if (cut_lj_inner >= cut_lj || cut_coul_inner >= cut_coul)
- error->all(FLERR,"Pair inner cutoff >= Pair outer cutoff");
-
- cut_lj_innersq = cut_lj_inner * cut_lj_inner;
- cut_ljsq = cut_lj * cut_lj;
- cut_coul_innersq = cut_coul_inner * cut_coul_inner;
- cut_coulsq = cut_coul * cut_coul;
- cut_bothsq = MAX(cut_ljsq,cut_coulsq);
-
- denom_lj = (cut_ljsq-cut_lj_innersq) * (cut_ljsq-cut_lj_innersq) *
- (cut_ljsq-cut_lj_innersq);
- denom_coul = (cut_coulsq-cut_coul_innersq) * (cut_coulsq-cut_coul_innersq) *
- (cut_coulsq-cut_coul_innersq);
-
- cut_coulsq = cut_coul * cut_coul;
-
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-}
-
-void PairLJCharmmCoulCharmmImplicitCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCharmmCoulCharmmImplicitCuda::init_list\n");)
- PairLJCharmmCoulCharmmImplicit::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCharmmCoulCharmmImplicitCuda::init_list end\n");)
-}
-
-void PairLJCharmmCoulCharmmImplicitCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCharmmCoulCharmmImplicit::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.h b/src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.h
deleted file mode 100644
index b3cc8c9336..0000000000
--- a/src/USER-CUDA/pair_lj_charmm_coul_charmm_implicit_cuda.h
+++ /dev/null
@@ -1,62 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/charmm/coul/charmm/implicit/cuda,PairLJCharmmCoulCharmmImplicitCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CHARMM_COUL_CHARMM_IMPLICIT_CUDA_H
-#define LMP_PAIR_LJ_CHARMM_COUL_CHARMM_IMPLICIT_CUDA_H
-
-#include "pair_lj_charmm_coul_charmm_implicit.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCharmmCoulCharmmImplicitCuda : public PairLJCharmmCoulCharmmImplicit
-{
- public:
- PairLJCharmmCoulCharmmImplicitCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- cCudaData<double , F_CFLOAT , x >* cu_lj1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj4_gm;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_charmm_coul_long_cuda.cpp b/src/USER-CUDA/pair_lj_charmm_coul_long_cuda.cpp
deleted file mode 100644
index b228bd6f41..0000000000
--- a/src/USER-CUDA/pair_lj_charmm_coul_long_cuda.cpp
+++ /dev/null
@@ -1,196 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_charmm_coul_long_cuda.h"
-#include "pair_lj_charmm_coul_long_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-#define EWALD_F 1.12837917
-#define EWALD_P 0.3275911
-#define A1 0.254829592
-#define A2 -0.284496736
-#define A3 1.421413741
-#define A4 -1.453152027
-#define A5 1.061405429
-/* ---------------------------------------------------------------------- */
-
-PairLJCharmmCoulLongCuda::PairLJCharmmCoulLongCuda(LAMMPS *lmp) : PairLJCharmmCoulLong(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.collect_forces_later = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCharmmCoulLongCuda::allocate()
-{
- if(! allocated) PairLJCharmmCoulLong::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- //cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- cu_lj1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj1, &cuda->shared_data.pair.coeff1_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj2, &cuda->shared_data.pair.coeff2_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj3, &cuda->shared_data.pair.coeff3_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj4, &cuda->shared_data.pair.coeff4_gm, (atom->ntypes+1)*(atom->ntypes+1));
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulLongCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
- }
-
- Cuda_PairLJCharmmCoulLongCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom,denom_lj);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulLongCuda::settings(int narg, char **arg)
-{
- PairLJCharmmCoulLong::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (X_CFLOAT) cut_lj;
- cuda->shared_data.pair.cut_coulsq_global = (X_CFLOAT) cut_coulsq;
- cuda->shared_data.pair.cut_inner_global = (F_CFLOAT) cut_lj_inner;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCharmmCoulLongCuda::coeff(int narg, char **arg)
-{
- PairLJCharmmCoulLong::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCharmmCoulLongCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/charmm/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- if (cut_lj_inner >= cut_lj)
- error->all(FLERR,"Pair inner cutoff >= Pair outer cutoff");
-
- cut_lj_innersq = cut_lj_inner * cut_lj_inner;
- cut_ljsq = cut_lj * cut_lj;
- cut_coulsq = cut_coul * cut_coul;
- cut_bothsq = MAX(cut_ljsq,cut_coulsq);
-
- denom_lj = (cut_ljsq-cut_lj_innersq) * (cut_ljsq-cut_lj_innersq) *
- (cut_ljsq-cut_lj_innersq);
-
- cut_coulsq = cut_coul * cut_coul;
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
-
- if (force->kspace == NULL)
- error->all(FLERR,"Pair style is incompatible with KSpace style");
- g_ewald = force->kspace->g_ewald;
- cuda->shared_data.pair.g_ewald=g_ewald;
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-
- if(ncoultablebits) error->warning(FLERR,"# CUDA: You asked for the usage of Coulomb Tables. This is not supported in CUDA Pair forces. Setting is ignored.\n");
-}
-
-void PairLJCharmmCoulLongCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCharmmCoulLongCuda::init_list\n");)
- PairLJCharmmCoulLong::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCharmmCoulLongCuda::init_list end\n");)
-}
-
-void PairLJCharmmCoulLongCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCharmmCoulLong::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_charmm_coul_long_cuda.h b/src/USER-CUDA/pair_lj_charmm_coul_long_cuda.h
deleted file mode 100644
index 8d9048a341..0000000000
--- a/src/USER-CUDA/pair_lj_charmm_coul_long_cuda.h
+++ /dev/null
@@ -1,62 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/charmm/coul/long/cuda,PairLJCharmmCoulLongCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CHARMM_COUL_LONG_CUDA_H
-#define LMP_PAIR_LJ_CHARMM_COUL_LONG_CUDA_H
-
-#include "pair_lj_charmm_coul_long.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCharmmCoulLongCuda : public PairLJCharmmCoulLong
-{
- public:
- PairLJCharmmCoulLongCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- cCudaData<double , F_CFLOAT , x >* cu_lj1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj4_gm;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_class2_coul_cut_cuda.cpp b/src/USER-CUDA/pair_lj_class2_coul_cut_cuda.cpp
deleted file mode 100644
index 01b6dc071f..0000000000
--- a/src/USER-CUDA/pair_lj_class2_coul_cut_cuda.cpp
+++ /dev/null
@@ -1,162 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_class2_coul_cut_cuda.h"
-#include "pair_lj_class2_coul_cut_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJClass2CoulCutCuda::PairLJClass2CoulCutCuda(LAMMPS *lmp) : PairLJClass2CoulCut(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJClass2CoulCutCuda::allocate()
-{
- if(! allocated) PairLJClass2CoulCut::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.cut_coul= cut_coul;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2CoulCutCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJClass2CoulCutCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2CoulCutCuda::settings(int narg, char **arg)
-{
- PairLJClass2CoulCut::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
- cuda->shared_data.pair.cut_coul_global = (F_CFLOAT) cut_coul_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2CoulCutCuda::coeff(int narg, char **arg)
-{
- PairLJClass2CoulCut::coeff(narg, arg);
- allocate();
-}
-
-void PairLJClass2CoulCutCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/cut/coul/cut/cuda requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-}
-
-void PairLJClass2CoulCutCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJClass2CoulCutCuda::init_list\n");)
- PairLJClass2CoulCut::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJClass2CoulCutCuda::init_list end\n");)
-}
-
-void PairLJClass2CoulCutCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJClass2CoulCut::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_class2_coul_cut_cuda.h b/src/USER-CUDA/pair_lj_class2_coul_cut_cuda.h
deleted file mode 100644
index 6601e2797c..0000000000
--- a/src/USER-CUDA/pair_lj_class2_coul_cut_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/class2/coul/cut/cuda,PairLJClass2CoulCutCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CLASS2_COUL_CUT_CUDA_H
-#define LMP_PAIR_LJ_CLASS2_COUL_CUT_CUDA_H
-
-#include "pair_lj_class2_coul_cut.h"
-
-namespace LAMMPS_NS {
-
-class PairLJClass2CoulCutCuda : public PairLJClass2CoulCut
-{
- public:
- PairLJClass2CoulCutCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_class2_coul_long_cuda.cpp b/src/USER-CUDA/pair_lj_class2_coul_long_cuda.cpp
deleted file mode 100644
index 20f257ffea..0000000000
--- a/src/USER-CUDA/pair_lj_class2_coul_long_cuda.cpp
+++ /dev/null
@@ -1,175 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_class2_coul_long_cuda.h"
-#include "pair_lj_class2_coul_long_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-#define EWALD_F 1.12837917
-#define EWALD_P 0.3275911
-#define A1 0.254829592
-#define A2 -0.284496736
-#define A3 1.421413741
-#define A4 -1.453152027
-#define A5 1.061405429
-/* ---------------------------------------------------------------------- */
-
-PairLJClass2CoulLongCuda::PairLJClass2CoulLongCuda(LAMMPS *lmp) : PairLJClass2CoulLong(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJClass2CoulLongCuda::allocate()
-{
- if(! allocated) PairLJClass2CoulLong::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2CoulLongCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJClass2CoulLongCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2CoulLongCuda::settings(int narg, char **arg)
-{
- PairLJClass2CoulLong::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2CoulLongCuda::coeff(int narg, char **arg)
-{
- PairLJClass2CoulLong::coeff(narg, arg);
- allocate();
-}
-
-void PairLJClass2CoulLongCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/cut/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- cut_coulsq = cut_coul * cut_coul;
- cuda->shared_data.pair.cut_coul_global=cut_coul;
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
- // set rRESPA cutoffs
-
- if (force->newton) error->warning(FLERR,"Pair style uses does not use \"newton\" setting. You might test if \"newton off\" makes the simulation run faster.");
- if (force->kspace == NULL)
- error->all(FLERR,"Pair style is incompatible with KSpace style");
- g_ewald = force->kspace->g_ewald;
- cuda->shared_data.pair.g_ewald=g_ewald;
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-
- if(ncoultablebits) error->warning(FLERR,"# CUDA: You asked for the usage of Coulomb Tables. This is not supported in CUDA Pair forces. Setting is ignored.\n");
-}
-
-void PairLJClass2CoulLongCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJClass2CoulLongCuda::init_list\n");)
- PairLJClass2CoulLong::init_list(id, ptr);
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- MYDBG(printf("# CUDA PairLJClass2CoulLongCuda::init_list end\n");)
-}
-
-void PairLJClass2CoulLongCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJClass2CoulLong::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_class2_coul_long_cuda.h b/src/USER-CUDA/pair_lj_class2_coul_long_cuda.h
deleted file mode 100644
index 43af51b4ed..0000000000
--- a/src/USER-CUDA/pair_lj_class2_coul_long_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/class2/coul/long/cuda,PairLJClass2CoulLongCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CLASS2_COUL_LONG_CUDA_H
-#define LMP_PAIR_LJ_CLASS2_COUL_LONG_CUDA_H
-
-#include "pair_lj_class2_coul_long.h"
-
-namespace LAMMPS_NS {
-
-class PairLJClass2CoulLongCuda : public PairLJClass2CoulLong
-{
- public:
- PairLJClass2CoulLongCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_class2_cuda.cpp b/src/USER-CUDA/pair_lj_class2_cuda.cpp
deleted file mode 100644
index 6b9f686c13..0000000000
--- a/src/USER-CUDA/pair_lj_class2_cuda.cpp
+++ /dev/null
@@ -1,167 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_class2_cuda.h"
-#include "pair_lj_class2_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJClass2Cuda::PairLJClass2Cuda(LAMMPS *lmp) : PairLJClass2(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJClass2Cuda::allocate()
-{
- if(! allocated) PairLJClass2::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2Cuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJClass2Cuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2Cuda::settings(int narg, char **arg)
-{
- PairLJClass2::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJClass2Cuda::coeff(int narg, char **arg)
-{
- PairLJClass2::coeff(narg, arg);
- allocate();
-}
-
-void PairLJClass2Cuda::init_style()
-{
- MYDBG(printf("# CUDA PairLJClass2Cuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- //neighbor->style=0; //0=NSQ neighboring
- MYDBG(printf("# CUDA PairLJClass2Cuda::init_style end\n"); )
-}
-
-void PairLJClass2Cuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJClass2Cuda::init_list\n");)
- PairLJClass2::init_list(id, ptr);
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- MYDBG(printf("# CUDA PairLJClass2Cuda::init_list end\n");)
-}
-
-void PairLJClass2Cuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJClass2::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_class2_cuda.h b/src/USER-CUDA/pair_lj_class2_cuda.h
deleted file mode 100644
index 6f2673c4a3..0000000000
--- a/src/USER-CUDA/pair_lj_class2_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/class2/cuda,PairLJClass2Cuda)
-
-#else
-
-#ifndef PAIR_LJ_CLASS2_CUDA_H
-#define PAIR_LJ_CLASS2_CUDA_H
-
-#include "pair_lj_class2.h"
-
-namespace LAMMPS_NS {
-
-class PairLJClass2Cuda : public PairLJClass2
-{
- public:
- PairLJClass2Cuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_cut_coul_cut_cuda.cpp b/src/USER-CUDA/pair_lj_cut_coul_cut_cuda.cpp
deleted file mode 100644
index 3872be0d0e..0000000000
--- a/src/USER-CUDA/pair_lj_cut_coul_cut_cuda.cpp
+++ /dev/null
@@ -1,162 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_cut_coul_cut_cuda.h"
-#include "pair_lj_cut_coul_cut_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJCutCoulCutCuda::PairLJCutCoulCutCuda(LAMMPS *lmp) : PairLJCutCoulCut(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCutCoulCutCuda::allocate()
-{
- if(! allocated) PairLJCutCoulCut::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.cut_coul= cut_coul;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulCutCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJCutCoulCutCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulCutCuda::settings(int narg, char **arg)
-{
- PairLJCutCoulCut::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
- cuda->shared_data.pair.cut_coul_global = (F_CFLOAT) cut_coul_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulCutCuda::coeff(int narg, char **arg)
-{
- PairLJCutCoulCut::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCutCoulCutCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/cut/coul/cut/cuda requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-}
-
-void PairLJCutCoulCutCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCutCoulCutCuda::init_list\n");)
- PairLJCutCoulCut::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCutCoulCutCuda::init_list end\n");)
-}
-
-void PairLJCutCoulCutCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCutCoulCut::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_cut_coul_cut_cuda.h b/src/USER-CUDA/pair_lj_cut_coul_cut_cuda.h
deleted file mode 100644
index 10f44c76da..0000000000
--- a/src/USER-CUDA/pair_lj_cut_coul_cut_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/cut/coul/cut/cuda,PairLJCutCoulCutCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CUT_COUL_CUT_CUDA_H
-#define LMP_PAIR_LJ_CUT_COUL_CUT_CUDA_H
-
-#include "pair_lj_cut_coul_cut.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCutCoulCutCuda : public PairLJCutCoulCut
-{
- public:
- PairLJCutCoulCutCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_cut_coul_debye_cuda.cpp b/src/USER-CUDA/pair_lj_cut_coul_debye_cuda.cpp
deleted file mode 100644
index 43bcce68c9..0000000000
--- a/src/USER-CUDA/pair_lj_cut_coul_debye_cuda.cpp
+++ /dev/null
@@ -1,163 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_cut_coul_debye_cuda.h"
-#include "pair_lj_cut_coul_debye_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJCutCoulDebyeCuda::PairLJCutCoulDebyeCuda(LAMMPS *lmp) : PairLJCutCoulDebye(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCutCoulDebyeCuda::allocate()
-{
- if(! allocated) PairLJCutCoulDebye::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.cut_coul= cut_coul;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulDebyeCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJCutCoulDebyeCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulDebyeCuda::settings(int narg, char **arg)
-{
- PairLJCutCoulDebye::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
- cuda->shared_data.pair.cut_coul_global = (F_CFLOAT) cut_coul_global;
- cuda->shared_data.pair.kappa = (F_CFLOAT) kappa;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulDebyeCuda::coeff(int narg, char **arg)
-{
- PairLJCutCoulDebye::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCutCoulDebyeCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/cut/coul/debye/cuda requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-}
-
-void PairLJCutCoulDebyeCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCutCoulDebyeCuda::init_list\n");)
- PairLJCutCoulDebye::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCutCoulDebyeCuda::init_list end\n");)
-}
-
-void PairLJCutCoulDebyeCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCutCoulDebye::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_cut_coul_debye_cuda.h b/src/USER-CUDA/pair_lj_cut_coul_debye_cuda.h
deleted file mode 100644
index aea3a42f66..0000000000
--- a/src/USER-CUDA/pair_lj_cut_coul_debye_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/cut/coul/debye/cuda,PairLJCutCoulDebyeCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CUT_COUL_DEBYE_CUDA_H
-#define LMP_PAIR_LJ_CUT_COUL_DEBYE_CUDA_H
-
-#include "pair_lj_cut_coul_debye.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCutCoulDebyeCuda : public PairLJCutCoulDebye
-{
- public:
- PairLJCutCoulDebyeCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_cut_coul_long_cuda.cpp b/src/USER-CUDA/pair_lj_cut_coul_long_cuda.cpp
deleted file mode 100644
index 52397f9429..0000000000
--- a/src/USER-CUDA/pair_lj_cut_coul_long_cuda.cpp
+++ /dev/null
@@ -1,216 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_cut_coul_long_cuda.h"
-#include "pair_lj_cut_coul_long_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-#define EWALD_F 1.12837917
-#define EWALD_P 0.3275911
-#define A1 0.254829592
-#define A2 -0.284496736
-#define A3 1.421413741
-#define A4 -1.453152027
-#define A5 1.061405429
-/* ---------------------------------------------------------------------- */
-
-PairLJCutCoulLongCuda::PairLJCutCoulLongCuda(LAMMPS *lmp) : PairLJCutCoulLong(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCutCoulLongCuda::allocate()
-{
- if(! allocated) PairLJCutCoulLong::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulLongCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJCutCoulLongCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulLongCuda::settings(int narg, char **arg)
-{
- PairLJCutCoulLong::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCoulLongCuda::coeff(int narg, char **arg)
-{
- PairLJCutCoulLong::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCutCoulLongCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/cut/coul/long requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (update->whichflag == 0 && strstr(update->integrate_style,"respa")) {
- int respa = 0;
- if (((Respa *) update->integrate)->level_inner >= 0) respa = 1;
- if (((Respa *) update->integrate)->level_middle >= 0) respa = 2;
-
- if (respa == 0) irequest = neighbor->request(this,instance_me);
- else if (respa == 1) {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->id = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->respainner = 1;
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->id = 3;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->respaouter = 1;
- } else {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->id = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->respainner = 1;
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->id = 2;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->respamiddle = 1;
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->id = 3;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->respaouter = 1;
- }
-
- }
- else
- {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- }
-
- cut_coulsq = cut_coul * cut_coul;
- cuda->shared_data.pair.cut_coul_global=cut_coul;
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
- // set rRESPA cutoffs
-
- if (strstr(update->integrate_style,"respa") &&
- ((Respa *) update->integrate)->level_inner >= 0)
- cut_respa = ((Respa *) update->integrate)->cutoff;
- else cut_respa = NULL;
-
- if (force->newton) error->warning(FLERR,"Pair style uses does not use \"newton\" setting. You might test if \"newton off\" makes the simulation run faster.");
- if (force->kspace == NULL)
- error->all(FLERR,"Pair style is incompatible with KSpace style");
- g_ewald = force->kspace->g_ewald;
- cuda->shared_data.pair.g_ewald=g_ewald;
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-
-
- if(ncoultablebits) error->warning(FLERR,"# CUDA: You asked for the usage of Coulomb Tables. This is not supported in CUDA Pair forces. Setting is ignored.\n");
-}
-
-void PairLJCutCoulLongCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCutCoulLongCuda::init_list\n");)
- PairLJCutCoulLong::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCutCoulLongCuda::init_list end\n");)
-}
-
-void PairLJCutCoulLongCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCutCoulLong::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_cut_coul_long_cuda.h b/src/USER-CUDA/pair_lj_cut_coul_long_cuda.h
deleted file mode 100644
index 2c7e55eb1e..0000000000
--- a/src/USER-CUDA/pair_lj_cut_coul_long_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/cut/coul/long/cuda,PairLJCutCoulLongCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_CUT_COUL_LONG_CUDA_H
-#define LMP_PAIR_LJ_CUT_COUL_LONG_CUDA_H
-
-#include "pair_lj_cut_coul_long.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCutCoulLongCuda : public PairLJCutCoulLong
-{
- public:
- PairLJCutCoulLongCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_cut_cuda.cpp b/src/USER-CUDA/pair_lj_cut_cuda.cpp
deleted file mode 100644
index a5d4f47a51..0000000000
--- a/src/USER-CUDA/pair_lj_cut_cuda.cpp
+++ /dev/null
@@ -1,179 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_cut_cuda.h"
-#include "pair_lj_cut_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJCutCuda::PairLJCutCuda(LAMMPS *lmp) : PairLJCut(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCutCuda::allocate()
-{
- if(! allocated) PairLJCut::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJCutCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCuda::settings(int narg, char **arg)
-{
- PairLJCut::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutCuda::coeff(int narg, char **arg)
-{
- PairLJCut::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCutCuda::init_style()
-{
- MYDBG(printf("# CUDA PairLJCutCuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (update->whichflag == 0 && strstr(update->integrate_style,"respa")) {
-
- }
- else
- {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- //neighbor->style=0; //0=NSQ neighboring
- }
-
-
- cut_respa = NULL;
- MYDBG(printf("# CUDA PairLJCutCuda::init_style end\n"); )
-}
-
-void PairLJCutCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCutCuda::init_list\n");)
- PairLJCut::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCutCuda::init_list end\n");)
-}
-
-void PairLJCutCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCut::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_cut_cuda.h b/src/USER-CUDA/pair_lj_cut_cuda.h
deleted file mode 100644
index f42c7d04c0..0000000000
--- a/src/USER-CUDA/pair_lj_cut_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/cut/cuda,PairLJCutCuda)
-
-#else
-
-#ifndef PAIR_LJ_CUT_CUDA_H
-#define PAIR_LJ_CUT_CUDA_H
-
-#include "pair_lj_cut.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCutCuda : public PairLJCut
-{
- public:
- PairLJCutCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_cut_experimental_cuda.cpp b/src/USER-CUDA/pair_lj_cut_experimental_cuda.cpp
deleted file mode 100644
index f60aaa6f38..0000000000
--- a/src/USER-CUDA/pair_lj_cut_experimental_cuda.cpp
+++ /dev/null
@@ -1,178 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_cut_experimental_cuda.h"
-#include "pair_lj_cut_experimental_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJCutExperimentalCuda::PairLJCutExperimentalCuda(LAMMPS *lmp) : PairLJCut(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJCutExperimentalCuda::allocate()
-{
- if(! allocated) PairLJCut::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutExperimentalCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
- Cuda_PairLJCutExperimentalCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- CudaWrapper_Sync();
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
- }
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutExperimentalCuda::settings(int narg, char **arg)
-{
- PairLJCut::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJCutExperimentalCuda::coeff(int narg, char **arg)
-{
- PairLJCut::coeff(narg, arg);
- allocate();
-}
-
-void PairLJCutExperimentalCuda::init_style()
-{
- MYDBG(printf("# CUDA PairLJCutExperimentalCuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (update->whichflag == 0 && strstr(update->integrate_style,"respa")) {
-
- }
- else
- {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- //neighbor->style=0; //0=NSQ neighboring
- }
-
-
- cut_respa = NULL;
- MYDBG(printf("# CUDA PairLJCutExperimentalCuda::init_style end\n"); )
-}
-
-void PairLJCutExperimentalCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJCutExperimentalCuda::init_list\n");)
- PairLJCut::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJCutExperimentalCuda::init_list end\n");)
-}
-
-void PairLJCutExperimentalCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJCut::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_cut_experimental_cuda.h b/src/USER-CUDA/pair_lj_cut_experimental_cuda.h
deleted file mode 100644
index fafb2d63d4..0000000000
--- a/src/USER-CUDA/pair_lj_cut_experimental_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/cut/experimental/cuda,PairLJCutExperimentalCuda)
-
-#else
-
-#ifndef PAIR_LJ_CUT_EXPERIMENTAL_CUDA_H
-#define PAIR_LJ_CUT_EXPERIMENTAL_CUDA_H
-
-#include "pair_lj_cut.h"
-
-namespace LAMMPS_NS {
-
-class PairLJCutExperimentalCuda : public PairLJCut
-{
- public:
- PairLJCutExperimentalCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_expand_cuda.cpp b/src/USER-CUDA/pair_lj_expand_cuda.cpp
deleted file mode 100644
index a102dea75a..0000000000
--- a/src/USER-CUDA/pair_lj_expand_cuda.cpp
+++ /dev/null
@@ -1,180 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_expand_cuda.h"
-#include "pair_lj_expand_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJExpandCuda::PairLJExpandCuda(LAMMPS *lmp) : PairLJExpand(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJExpandCuda::allocate()
-{
- if(! allocated) PairLJExpand::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.cutsq = cutsq;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.coeff5 = shift;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJExpandCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJExpandCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJExpandCuda::settings(int narg, char **arg)
-{
- PairLJExpand::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJExpandCuda::coeff(int narg, char **arg)
-{
- PairLJExpand::coeff(narg, arg);
- allocate();
-}
-
-void PairLJExpandCuda::init_style()
-{
- MYDBG(printf("# CUDA PairLJExpandCuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (update->whichflag == 0 && strstr(update->integrate_style,"respa")) {
-
- }
- else
- {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- //neighbor->style=0; //0=NSQ neighboring
- }
-
-
- MYDBG(printf("# CUDA PairLJExpandCuda::init_style end\n"); )
-}
-
-void PairLJExpandCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJExpandCuda::init_list\n");)
- PairLJExpand::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJExpandCuda::init_list end\n");)
-}
-
-void PairLJExpandCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJExpand::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_expand_cuda.h b/src/USER-CUDA/pair_lj_expand_cuda.h
deleted file mode 100644
index b61578c295..0000000000
--- a/src/USER-CUDA/pair_lj_expand_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/expand/cuda,PairLJExpandCuda)
-
-#else
-
-#ifndef PAIR_LJ_EXPAND_CUDA_H
-#define PAIR_LJ_EXPAND_CUDA_H
-
-#include "pair_lj_expand.h"
-
-namespace LAMMPS_NS {
-
-class PairLJExpandCuda : public PairLJExpand
-{
- public:
- PairLJExpandCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.cpp b/src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.cpp
deleted file mode 100644
index 73df6a66cb..0000000000
--- a/src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.cpp
+++ /dev/null
@@ -1,194 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_gromacs_coul_gromacs_cuda.h"
-#include "pair_lj_gromacs_coul_gromacs_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJGromacsCoulGromacsCuda::PairLJGromacsCoulGromacsCuda(LAMMPS *lmp) : PairLJGromacsCoulGromacs(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJGromacsCoulGromacsCuda::allocate()
-{
- if(! allocated) PairLJGromacsCoulGromacs::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.coeff5 = ljsw1;
- cuda->shared_data.pair.coeff6 = ljsw2;
- cuda->shared_data.pair.coeff7 = ljsw3;
- cuda->shared_data.pair.coeff8 = ljsw4;
- cuda->shared_data.pair.coeff9 = ljsw5;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- cu_lj1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj1, &cuda->shared_data.pair.coeff1_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj2, &cuda->shared_data.pair.coeff2_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj3, &cuda->shared_data.pair.coeff3_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj4, &cuda->shared_data.pair.coeff4_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw1, &cuda->shared_data.pair.coeff5_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw2, &cuda->shared_data.pair.coeff6_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw3, &cuda->shared_data.pair.coeff7_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw4, &cuda->shared_data.pair.coeff8_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw5_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw5, &cuda->shared_data.pair.coeff9_gm, (atom->ntypes+1)*(atom->ntypes+1));
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJGromacsCoulGromacsCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
- }
-
- Cuda_PairLJGromacsCoulGromacsCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom,cut_coul_inner,coulsw1,coulsw2,coulsw5);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJGromacsCoulGromacsCuda::settings(int narg, char **arg)
-{
- PairLJGromacsCoulGromacs::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (X_CFLOAT) cut_lj;
- cuda->shared_data.pair.cut_coulsq_global = (X_CFLOAT) cut_coulsq;
- cuda->shared_data.pair.cut_inner_global = (F_CFLOAT) cut_lj_inner;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJGromacsCoulGromacsCuda::coeff(int narg, char **arg)
-{
- PairLJGromacsCoulGromacs::coeff(narg, arg);
- allocate();
-}
-
-void PairLJGromacsCoulGromacsCuda::init_style()
-{
- if (!atom->q_flag)
- error->all(FLERR,"Pair style lj/gromacs/coul/gromacs requires atom attribute q");
- // request regular or rRESPA neighbor lists
-
- if(atom->molecular)
- {
- cuda->shared_data.pair.collect_forces_later = 1;
- }
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- if (cut_lj_inner >= cut_lj || cut_coul_inner >= cut_coul)
- error->all(FLERR,"Pair inner cutoff >= Pair outer cutoff");
-
- cut_lj_innersq = cut_lj_inner * cut_lj_inner;
- cut_ljsq = cut_lj * cut_lj;
- cut_coul_innersq = cut_coul_inner * cut_coul_inner;
- cut_coulsq = cut_coul * cut_coul;
- cut_bothsq = MAX(cut_ljsq,cut_coulsq);
-
-
- cut_coulsq = cut_coul * cut_coul;
-
- cuda->shared_data.pair.cut_coulsq_global=cut_coulsq;
-
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
-}
-
-void PairLJGromacsCoulGromacsCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJGromacsCoulGromacsCuda::init_list\n");)
- PairLJGromacsCoulGromacs::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJGromacsCoulGromacsCuda::init_list end\n");)
-}
-
-void PairLJGromacsCoulGromacsCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJGromacsCoulGromacs::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.h b/src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.h
deleted file mode 100644
index 6e48df1931..0000000000
--- a/src/USER-CUDA/pair_lj_gromacs_coul_gromacs_cuda.h
+++ /dev/null
@@ -1,68 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/gromacs/coul/gromacs/cuda,PairLJGromacsCoulGromacsCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_GROMACS_COUL_GROMACS_CUDA_H
-#define LMP_PAIR_LJ_GROMACS_COUL_GROMACS_CUDA_H
-
-#include "pair_lj_gromacs_coul_gromacs.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairLJGromacsCoulGromacsCuda : public PairLJGromacsCoulGromacs
-{
- public:
- PairLJGromacsCoulGromacsCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- cCudaData<double , F_CFLOAT , x >* cu_lj1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj4_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw4_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw5_gm;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_gromacs_cuda.cpp b/src/USER-CUDA/pair_lj_gromacs_cuda.cpp
deleted file mode 100644
index b2786d81ab..0000000000
--- a/src/USER-CUDA/pair_lj_gromacs_cuda.cpp
+++ /dev/null
@@ -1,177 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_gromacs_cuda.h"
-#include "pair_lj_gromacs_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJGromacsCuda::PairLJGromacsCuda(LAMMPS *lmp) : PairLJGromacs(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJGromacsCuda::allocate()
-{
- if(! allocated) PairLJGromacs::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.cut_inner = cut_inner;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.coeff5 = ljsw1;
- cuda->shared_data.pair.coeff6 = ljsw2;
- cuda->shared_data.pair.coeff7 = ljsw3;
- cuda->shared_data.pair.coeff8 = ljsw4;
- cuda->shared_data.pair.coeff9 = ljsw5;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cu_lj1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj1, &cuda->shared_data.pair.coeff1_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj2, &cuda->shared_data.pair.coeff2_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj3, &cuda->shared_data.pair.coeff3_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj4, &cuda->shared_data.pair.coeff4_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw1, &cuda->shared_data.pair.coeff5_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw2, &cuda->shared_data.pair.coeff6_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw3, &cuda->shared_data.pair.coeff7_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw4, &cuda->shared_data.pair.coeff8_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw5_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw5, &cuda->shared_data.pair.coeff9_gm, (atom->ntypes+1)*(atom->ntypes+1));
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJGromacsCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
- }
-
- Cuda_PairLJGromacsCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJGromacsCuda::settings(int narg, char **arg)
-{
- PairLJGromacs::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
- cuda->shared_data.pair.cut_inner_global = (F_CFLOAT) cut_inner_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJGromacsCuda::coeff(int narg, char **arg)
-{
- PairLJGromacs::coeff(narg, arg);
- allocate();
-}
-
-void PairLJGromacsCuda::init_style()
-{
- // request regular or rRESPA neighbor lists
-
- if(atom->molecular)
- {
- cuda->shared_data.pair.collect_forces_later = 1;
- }
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
-
-}
-
-void PairLJGromacsCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJGromacsCuda::init_list\n");)
- PairLJGromacs::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJGromacsCuda::init_list end\n");)
-}
-
-void PairLJGromacsCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJGromacs::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_gromacs_cuda.h b/src/USER-CUDA/pair_lj_gromacs_cuda.h
deleted file mode 100644
index b4bbc15c6d..0000000000
--- a/src/USER-CUDA/pair_lj_gromacs_cuda.h
+++ /dev/null
@@ -1,68 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/gromacs/cuda,PairLJGromacsCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_GROMACS_CUDA_H
-#define LMP_PAIR_LJ_GROMACS_CUDA_H
-
-#include "pair_lj_gromacs.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairLJGromacsCuda : public PairLJGromacs
-{
- public:
- PairLJGromacsCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- cCudaData<double , F_CFLOAT , x >* cu_lj1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj4_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw4_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw5_gm;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_sdk_coul_long_cuda.cpp b/src/USER-CUDA/pair_lj_sdk_coul_long_cuda.cpp
deleted file mode 100644
index aca0f6d013..0000000000
--- a/src/USER-CUDA/pair_lj_sdk_coul_long_cuda.cpp
+++ /dev/null
@@ -1,193 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_sdk_coul_long_cuda.h"
-#include "pair_lj_sdk_coul_long_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJSDKCoulLongCuda::PairLJSDKCoulLongCuda(LAMMPS *lmp) : PairLJSDKCoulLong(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- lj_type_double = NULL;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJSDKCoulLongCuda::allocate()
-{
- if(! allocated) PairLJSDKCoulLong::allocate();
- int n = atom->ntypes;
- if(! allocated2)
- {
- allocated2 = true;
-
-
- memory->create(lj_type_double,n+1,n+1,"pairlj:ljtypedouble");
-
- cuda->shared_data.pair.cut = cut_lj;
- cuda->shared_data.pair.cut_coul= NULL;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.coeff5 = lj_type_double;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
-
- }
- for (int i = 1; i <= n; i++) {
- for (int j = i; j <= n; j++) {
- lj_type_double[i][j] = lj_type[i][j];
- lj_type_double[j][i] = lj_type[i][j];
- }
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSDKCoulLongCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(eflag) cuda->cu_eng_coul->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJSDKCoulLongCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(eflag) cuda->cu_eng_coul->download();
- if(vflag) cuda->cu_virial->download();
- }
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSDKCoulLongCuda::settings(int narg, char **arg)
-{
- PairLJSDKCoulLong::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_lj_global;
- cuda->shared_data.pair.cut_coul_global = (F_CFLOAT) cut_coul;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSDKCoulLongCuda::coeff(int narg, char **arg)
-{
- PairLJSDKCoulLong::coeff(narg, arg);
- allocate();
-}
-
-void PairLJSDKCoulLongCuda::init_style()
-{
- MYDBG(printf("# CUDA PairLJSDKCoulLongCuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- g_ewald = force->kspace->g_ewald;
- cuda->shared_data.pair.g_ewald=g_ewald;
- cuda->shared_data.pppm.qqrd2e=force->qqrd2e;
- if (force->newton) error->warning(FLERR,"Pair style uses does not use \"newton\" setting. You might test if \"newton off\" makes the simulation run faster.");
- MYDBG(printf("# CUDA PairLJSDKCoulLongCuda::init_style end\n"); )
-}
-
-void PairLJSDKCoulLongCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJSDKCoulLongCuda::init_list\n");)
- PairLJSDKCoulLong::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJSDKCoulLongCuda::init_list end\n");)
-}
-
-void PairLJSDKCoulLongCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJSDKCoulLong::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_sdk_coul_long_cuda.h b/src/USER-CUDA/pair_lj_sdk_coul_long_cuda.h
deleted file mode 100644
index 4b5c07c79d..0000000000
--- a/src/USER-CUDA/pair_lj_sdk_coul_long_cuda.h
+++ /dev/null
@@ -1,59 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(cg/cmm/coul/long/cuda,PairLJSDKCoulLongCuda)
-PairStyle(lj/sdk/coul/long/cuda,PairLJSDKCoulLongCuda)
-
-#else
-
-#ifndef PAIR_LJ_SDK_COUL_LONG_CUDA_H
-#define PAIR_LJ_SDK_COUL_LONG_CUDA_H
-
-#include "pair_lj_sdk_coul_long.h"
-
-namespace LAMMPS_NS {
-
-class PairLJSDKCoulLongCuda : public PairLJSDKCoulLong
-{
- public:
- PairLJSDKCoulLongCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- double** lj_type_double;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_sdk_cuda.cpp b/src/USER-CUDA/pair_lj_sdk_cuda.cpp
deleted file mode 100644
index f6eba1ba7b..0000000000
--- a/src/USER-CUDA/pair_lj_sdk_cuda.cpp
+++ /dev/null
@@ -1,184 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_sdk_cuda.h"
-#include "pair_lj_sdk_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJSDKCuda::PairLJSDKCuda(LAMMPS *lmp) : PairLJSDK(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- lj_type_double = NULL;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJSDKCuda::allocate()
-{
- if(! allocated) PairLJSDK::allocate();
- int n = atom->ntypes;
- if(! allocated2)
- {
- allocated2 = true;
-
-
- memory->create(lj_type_double,n+1,n+1,"pairlj:ljtypedouble");
-
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.coeff5 = lj_type_double;
- /*cu_lj1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj1, &cuda->shared_data.pair.coeff1_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj2, &cuda->shared_data.pair.coeff2_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj3, &cuda->shared_data.pair.coeff3_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj4, &cuda->shared_data.pair.coeff4_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj_type_double_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj_type_double, &cuda->shared_data.pair.coeff5_gm, (atom->ntypes+1)*(atom->ntypes+1));*/
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- }
- for (int i = 1; i <= n; i++) {
- for (int j = i; j <= n; j++) {
- lj_type_double[i][j] = lj_type[i][j];
- lj_type_double[j][i] = lj_type[i][j];
- }
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSDKCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairLJSDKCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSDKCuda::settings(int narg, char **arg)
-{
- PairLJSDK::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSDKCuda::coeff(int narg, char **arg)
-{
- PairLJSDK::coeff(narg, arg);
- allocate();
-}
-
-void PairLJSDKCuda::init_style()
-{
- MYDBG(printf("# CUDA PairLJSDKCuda::init_style start\n"); )
-
- int irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
- MYDBG(printf("# CUDA PairLJSDKCuda::init_style end\n"); )
-}
-
-void PairLJSDKCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJSDKCuda::init_list\n");)
- PairLJSDK::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJSDKCuda::init_list end\n");)
-}
-
-void PairLJSDKCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJSDK::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_sdk_cuda.h b/src/USER-CUDA/pair_lj_sdk_cuda.h
deleted file mode 100644
index 5e7807cbd7..0000000000
--- a/src/USER-CUDA/pair_lj_sdk_cuda.h
+++ /dev/null
@@ -1,65 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/sdk/cuda,PairLJSDKCuda)
-PairStyle(cg/cmm/cuda,PairLJSDKCuda)
-
-#else
-
-#ifndef PAIR_LJ_SDK_CUDA_H
-#define PAIR_LJ_SDK_CUDA_H
-
-#include "pair_lj_sdk.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairLJSDKCuda : public PairLJSDK
-{
- public:
- PairLJSDKCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- double** lj_type_double;
- cCudaData<double , F_CFLOAT , x >* cu_lj1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj4_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj_type_double_gm;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_lj_smooth_cuda.cpp b/src/USER-CUDA/pair_lj_smooth_cuda.cpp
deleted file mode 100644
index 3a51e94fef..0000000000
--- a/src/USER-CUDA/pair_lj_smooth_cuda.cpp
+++ /dev/null
@@ -1,177 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- Contributing author: Paul Crozier (SNL)
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_lj_smooth_cuda.h"
-#include "pair_lj_smooth_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "kspace.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairLJSmoothCuda::PairLJSmoothCuda(LAMMPS *lmp) : PairLJSmooth(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.use_block_per_atom = 0;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairLJSmoothCuda::allocate()
-{
- if(! allocated) PairLJSmooth::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.cut_inner = cut_inner;
- cuda->shared_data.pair.coeff1 = lj1;
- cuda->shared_data.pair.coeff2 = lj2;
- cuda->shared_data.pair.coeff3 = lj3;
- cuda->shared_data.pair.coeff4 = lj4;
- cuda->shared_data.pair.coeff5 = ljsw1;
- cuda->shared_data.pair.coeff6 = ljsw2;
- cuda->shared_data.pair.coeff7 = ljsw3;
- cuda->shared_data.pair.coeff8 = ljsw4;
- cuda->shared_data.pair.coeff9 = ljsw0;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cu_lj1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj1, &cuda->shared_data.pair.coeff1_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj2, &cuda->shared_data.pair.coeff2_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj3, &cuda->shared_data.pair.coeff3_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_lj4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)lj4, &cuda->shared_data.pair.coeff4_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw0_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw0, &cuda->shared_data.pair.coeff9_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw1_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw1, &cuda->shared_data.pair.coeff5_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw2_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw2, &cuda->shared_data.pair.coeff6_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw3_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw3, &cuda->shared_data.pair.coeff7_gm, (atom->ntypes+1)*(atom->ntypes+1));
- cu_ljsw4_gm = new cCudaData<double, F_CFLOAT, x> ((double*)ljsw4, &cuda->shared_data.pair.coeff8_gm, (atom->ntypes+1)*(atom->ntypes+1));
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSmoothCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
- }
-
- Cuda_PairLJSmoothCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSmoothCuda::settings(int narg, char **arg)
-{
- PairLJSmooth::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
- cuda->shared_data.pair.cut_inner_global = (F_CFLOAT) cut_inner_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairLJSmoothCuda::coeff(int narg, char **arg)
-{
- PairLJSmooth::coeff(narg, arg);
- allocate();
-}
-
-void PairLJSmoothCuda::init_style()
-{
- // request regular or rRESPA neighbor lists
-
- if(atom->molecular)
- {
- cuda->shared_data.pair.collect_forces_later = 1;
- }
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
-
-
-
-}
-
-void PairLJSmoothCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairLJSmoothCuda::init_list\n");)
- PairLJSmooth::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairLJSmoothCuda::init_list end\n");)
-}
-
-void PairLJSmoothCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairLJSmooth::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_lj_smooth_cuda.h b/src/USER-CUDA/pair_lj_smooth_cuda.h
deleted file mode 100644
index 0a57e6f663..0000000000
--- a/src/USER-CUDA/pair_lj_smooth_cuda.h
+++ /dev/null
@@ -1,68 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(lj/smooth/cuda,PairLJSmoothCuda)
-
-#else
-
-#ifndef LMP_PAIR_LJ_SMOOTH_CUDA_H
-#define LMP_PAIR_LJ_SMOOTH_CUDA_H
-
-#include "pair_lj_smooth.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairLJSmoothCuda : public PairLJSmooth
-{
- public:
- PairLJSmoothCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- cCudaData<double , F_CFLOAT , x >* cu_lj1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_lj4_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw0_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw1_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw2_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw3_gm;
- cCudaData<double , F_CFLOAT , x >* cu_ljsw4_gm;
-
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_morse_cuda.cpp b/src/USER-CUDA/pair_morse_cuda.cpp
deleted file mode 100644
index a38712aabe..0000000000
--- a/src/USER-CUDA/pair_morse_cuda.cpp
+++ /dev/null
@@ -1,177 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_morse_cuda.h"
-#include "pair_morse_cuda_cu.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-/* ---------------------------------------------------------------------- */
-
-PairMorseCuda::PairMorseCuda(LAMMPS *lmp) : PairMorse(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- cuda->shared_data.pair.cudable_force = 1;
- cuda->setSystemParams();
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairMorseCuda::allocate()
-{
- if(! allocated) PairMorse::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cut = cut;
- cuda->shared_data.pair.coeff1 = r0;
- cuda->shared_data.pair.coeff2 = alpha;
- cuda->shared_data.pair.coeff3 = morse1;
- cuda->shared_data.pair.coeff4 = d0;
- cuda->shared_data.pair.offset = offset;
- cuda->shared_data.pair.special_lj = force->special_lj;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairMorseCuda::compute(int eflag, int vflag)
-{
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairMorseCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);
-
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairMorseCuda::settings(int narg, char **arg)
-{
- PairMorse::settings(narg, arg);
- cuda->shared_data.pair.cut_global = (F_CFLOAT) cut_global;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairMorseCuda::coeff(int narg, char **arg)
-{
- PairMorse::coeff(narg, arg);
- allocate();
-}
-
-void PairMorseCuda::init_style()
-{
- MYDBG(printf("# CUDA PairMorseCuda::init_style start\n"); )
- // request regular or rRESPA neighbor lists
-
- int irequest;
-
- if (update->whichflag == 0 && strstr(update->integrate_style,"respa")) {
-
- }
- else
- {
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- //neighbor->style=0; //0=NSQ neighboring
- }
-
-
- MYDBG(printf("# CUDA PairMorseCuda::init_style end\n"); )
-}
-
-void PairMorseCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairMorseCuda::init_list\n");)
- PairMorse::init_list(id, ptr);
- #ifndef CUDA_USE_BINNING
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- #endif
- MYDBG(printf("# CUDA PairMorseCuda::init_list end\n");)
-}
-
-void PairMorseCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairMorse::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-
-}
diff --git a/src/USER-CUDA/pair_morse_cuda.h b/src/USER-CUDA/pair_morse_cuda.h
deleted file mode 100644
index f76e687527..0000000000
--- a/src/USER-CUDA/pair_morse_cuda.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(morse/cuda,PairMorseCuda)
-
-#else
-
-#ifndef PAIR_MORSE_CUDA_H
-#define PAIR_MORSE_CUDA_H
-
-#include "pair_morse.h"
-
-namespace LAMMPS_NS {
-
-class PairMorseCuda : public PairMorse
-{
- public:
- PairMorseCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_sw_cuda.cpp b/src/USER-CUDA/pair_sw_cuda.cpp
deleted file mode 100644
index 1dfccc4e7b..0000000000
--- a/src/USER-CUDA/pair_sw_cuda.cpp
+++ /dev/null
@@ -1,207 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_sw_cuda.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-
-
-
-/* ---------------------------------------------------------------------- */
-
-PairSWCuda::PairSWCuda(LAMMPS *lmp) : PairSW(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- params_f = NULL;
- cuda->setSystemParams();
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.override_block_per_atom = 0;
- cuda->shared_data.pair.neighall = true;
- init = false;
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairSWCuda::allocate()
-{
- if(! allocated) PairSW::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cutsq = cutsq;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairSWCuda::compute(int eflag, int vflag)
-{
- if(!init) {Cuda_PairSWCuda_Init(&cuda->shared_data,params_f,map, &elem2param[0][0][0],nelements); init=true;}
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairSWCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);//,&elem2param[0][0][0],map
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairSWCuda::settings(int narg, char **arg)
-{
- PairSW::settings(narg, arg);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairSWCuda::coeff(int narg, char **arg)
-{
- PairSW::coeff(narg, arg);
- allocate();
- params_f = (ParamSW_Float *) memory->srealloc(params_f,maxparam*sizeof(ParamSW_Float),
- "pair:params_f");
- for(int i=0;i<maxparam;i++)
- {
- printf("%e %e\n",params[i].cut,params[i].cutsq);
- params_f[i].cut = params[i].cut;
- params_f[i].cutsq = params[i].cutsq;
- params_f[i].c1 = params[i].c1;
- params_f[i].c2 = params[i].c2;
- params_f[i].c3 = params[i].c3;
- params_f[i].c4 = params[i].c4;
- params_f[i].c5 = params[i].c5;
- params_f[i].c6 = params[i].c6;
- params_f[i].ielement = params[i].ielement;
- params_f[i].jelement = params[i].jelement;
- params_f[i].kelement = params[i].kelement;
- params_f[i].epsilon = params[i].epsilon;
- params_f[i].sigma = params[i].sigma;
- params_f[i].littlea = params[i].littlea;
- params_f[i].lambda = params[i].lambda;
- params_f[i].costheta = params[i].costheta;
- params_f[i].tol = params[i].tol;
- params_f[i].sigma_gamma = params[i].sigma_gamma;
- params_f[i].lambda_epsilon = params[i].lambda_epsilon;
- params_f[i].lambda_epsilon2 = params[i].lambda_epsilon2;
- params_f[i].gamma = params[i].gamma;
-
- params_f[i].biga = params[i].biga;
- params_f[i].bigb = params[i].bigb;
- params_f[i].gamma = params[i].gamma;
- params_f[i].powerp = params[i].powerp;
- params_f[i].powerq = params[i].powerq;
- }
- cuda->shared_data.pair.cut_global = cutmax;
-}
-
-void PairSWCuda::init_style()
-{
- MYDBG(printf("# CUDA PairSWCuda::init_style start\n"); )
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- neighbor->requests[irequest]->ghost = 1;
-
-
- MYDBG(printf("# CUDA PairSWCuda::init_style end\n"); )
-}
-
-void PairSWCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairSWCuda::init_list\n");)
- PairSW::init_list(id, ptr);
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- MYDBG(printf("# CUDA PairSWCuda::init_list end\n");)
- cu_params_f = (ParamSW_Float*) CudaWrapper_AllocCudaData(sizeof(ParamSW_Float)*maxparam);
- CudaWrapper_UploadCudaData((void*) params_f,(void*) cu_params_f,sizeof(ParamSW_Float)*maxparam);
- cu_elem2param = new cCudaData<int, int, xyz > ((int*) elem2param, nelements,nelements,nelements);
- cu_elem2param->upload();
- cu_map = new cCudaData<int, int, x > ( map,atom->ntypes+1 );
- cu_map->upload();
-}
-
-void PairSWCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairSW::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-}
diff --git a/src/USER-CUDA/pair_sw_cuda.h b/src/USER-CUDA/pair_sw_cuda.h
deleted file mode 100644
index c61f849a1f..0000000000
--- a/src/USER-CUDA/pair_sw_cuda.h
+++ /dev/null
@@ -1,66 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(sw/cuda,PairSWCuda)
-
-#else
-
-#ifndef PAIR_SW_CUDA_H
-#define PAIR_SW_CUDA_H
-
-#include "pair_sw_cuda_cu.h"
-#include "pair_sw.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairSWCuda : public PairSW
-{
- public:
- PairSWCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
-
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- ParamSW_Float* params_f;
- ParamSW_Float* cu_params_f;
- cCudaData<int, int, xyz >* cu_elem2param;
- cCudaData<int, int, x >* cu_map;
- bool init;
- bool iszbl;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_tersoff_cuda.cpp b/src/USER-CUDA/pair_tersoff_cuda.cpp
deleted file mode 100644
index f22b551284..0000000000
--- a/src/USER-CUDA/pair_tersoff_cuda.cpp
+++ /dev/null
@@ -1,204 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <cmath>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include "pair_tersoff_cuda.h"
-#include "cuda_data.h"
-#include "atom.h"
-#include "comm.h"
-#include "force.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "cuda_neigh_list.h"
-#include "update.h"
-#include "integrate.h"
-#include "respa.h"
-#include "memory.h"
-#include "error.h"
-#include "user_cuda.h"
-
-using namespace LAMMPS_NS;
-
-
-
-
-/* ---------------------------------------------------------------------- */
-
-PairTersoffCuda::PairTersoffCuda(LAMMPS *lmp) : PairTersoff(lmp)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- allocated2 = false;
- params_f = NULL;
- cuda->setSystemParams();
- cuda->shared_data.pair.cudable_force = 1;
- cuda->shared_data.pair.override_block_per_atom = 0;
- cuda->shared_data.pair.neighall = true;
- init = false;
- iszbl = false;
-}
-
-/* ----------------------------------------------------------------------
- remember pointer to arrays in cuda shared data
-------------------------------------------------------------------------- */
-
-void PairTersoffCuda::allocate()
-{
- if(! allocated) PairTersoff::allocate();
- if(! allocated2)
- {
- allocated2 = true;
- cuda->shared_data.pair.cutsq = cutsq;
- cuda->shared_data.pair.special_lj = force->special_lj;
- cuda->shared_data.pair.special_coul = force->special_coul;
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairTersoffCuda::compute(int eflag, int vflag)
-{
- if(!init) {Cuda_PairTersoffCuda_Init(&cuda->shared_data,params_f,map, &elem2param[0][0][0],nelements,iszbl); init=true;}
- if (eflag || vflag) ev_setup(eflag,vflag);
- if(eflag) cuda->cu_eng_vdwl->upload();
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_PairTersoffCuda(& cuda->shared_data, & cuda_neigh_list->sneighlist, eflag, vflag, eflag_atom, vflag_atom);//,&elem2param[0][0][0],map
- if(not cuda->shared_data.pair.collect_forces_later)
- {
- if(eflag) cuda->cu_eng_vdwl->download();
- if(vflag) cuda->cu_virial->download();
- }
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairTersoffCuda::settings(int narg, char **arg)
-{
- PairTersoff::settings(narg, arg);
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairTersoffCuda::coeff(int narg, char **arg)
-{
- PairTersoff::coeff(narg, arg);
- allocate();
- params_f = (Param_Float *) memory->srealloc(params_f,maxparam*sizeof(Param_Float),
- "pair:params_f");
- for(int i=0;i<maxparam;i++)
- {
- params_f[i].lam1 = params[i].lam1;
- params_f[i].lam2 = params[i].lam2;
- params_f[i].lam3 = params[i].lam3;
- params_f[i].c = params[i].c;
- params_f[i].d = params[i].d;
- params_f[i].h = params[i].h;
- params_f[i].gamma = params[i].gamma;
- params_f[i].powerm = params[i].powerm;
- params_f[i].powern = params[i].powern;
- params_f[i].beta = params[i].beta;
- params_f[i].biga = params[i].biga;
- params_f[i].bigb = params[i].bigb;
- params_f[i].bigd = params[i].bigd;
- params_f[i].bigr = params[i].bigr;
- params_f[i].cut = params[i].cut;
- params_f[i].cutsq = params[i].cutsq;
- params_f[i].c1 = params[i].c1;
- params_f[i].c2 = params[i].c2;
- params_f[i].c3 = params[i].c3;
- params_f[i].c4 = params[i].c4;
- params_f[i].ielement = params[i].ielement;
- params_f[i].jelement = params[i].jelement;
- params_f[i].kelement = params[i].kelement;
- params_f[i].powermint = params[i].powermint;
- }
- cuda->shared_data.pair.cut_global = cutmax;
-}
-
-void PairTersoffCuda::init_style()
-{
- MYDBG(printf("# CUDA PairTersoffCuda::init_style start\n"); )
-
- int irequest;
-
- irequest = neighbor->request(this,instance_me);
- neighbor->requests[irequest]->full = 1;
- neighbor->requests[irequest]->half = 0;
- neighbor->requests[irequest]->cudable = 1;
- neighbor->requests[irequest]->ghost = 1;
-
-
- MYDBG(printf("# CUDA PairTersoffCuda::init_style end\n"); )
-}
-
-void PairTersoffCuda::init_list(int id, NeighList *ptr)
-{
- MYDBG(printf("# CUDA PairTersoffCuda::init_list\n");)
- PairTersoff::init_list(id, ptr);
- // right now we can only handle verlet (id 0), not respa
- if(id == 0) cuda_neigh_list = cuda->registerNeighborList(ptr);
- // see Neighbor::init() for details on lammps lists' logic
- MYDBG(printf("# CUDA PairTersoffCuda::init_list end\n");)
- cu_params_f = (Param_Float*) CudaWrapper_AllocCudaData(sizeof(Param_Float)*maxparam);
- CudaWrapper_UploadCudaData((void*) params_f,(void*) cu_params_f,sizeof(Param_Float)*maxparam);
- cu_elem2param = new cCudaData<int, int, xyz > ((int*) elem2param, nelements,nelements,nelements);
- cu_elem2param->upload();
- cu_map = new cCudaData<int, int, x > ( map,atom->ntypes+1 );
- cu_map->upload();
-}
-
-void PairTersoffCuda::ev_setup(int eflag, int vflag)
-{
- int maxeatomold=maxeatom;
- PairTersoff::ev_setup(eflag,vflag);
-
- if (eflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_eatom; cuda->cu_eatom = new cCudaData<double, ENERGY_CFLOAT, x > ((double*)eatom, & cuda->shared_data.atom.eatom , atom->nmax );}
-
- if (vflag_atom && atom->nmax > maxeatomold)
- {delete cuda->cu_vatom; cuda->cu_vatom = new cCudaData<double, ENERGY_CFLOAT, yx > ((double*)vatom, & cuda->shared_data.atom.vatom , atom->nmax, 6 );}
-}
diff --git a/src/USER-CUDA/pair_tersoff_cuda.h b/src/USER-CUDA/pair_tersoff_cuda.h
deleted file mode 100644
index 5b829114f2..0000000000
--- a/src/USER-CUDA/pair_tersoff_cuda.h
+++ /dev/null
@@ -1,66 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(tersoff/cuda,PairTersoffCuda)
-
-#else
-
-#ifndef PAIR_TERSOFF_CUDA_H
-#define PAIR_TERSOFF_CUDA_H
-
-#include "pair_tersoff_cuda_cu.h"
-#include "pair_tersoff.h"
-#include "cuda_data.h"
-
-namespace LAMMPS_NS {
-
-class PairTersoffCuda : public PairTersoff
-{
- public:
- PairTersoffCuda(class LAMMPS *);
- void compute(int, int);
- void settings(int, char **);
- void coeff(int, char **);
- void init_list(int, class NeighList *);
- void init_style();
- void ev_setup(int eflag, int vflag);
- protected:
-
- class Cuda *cuda;
- void allocate();
- bool allocated2;
- class CudaNeighList* cuda_neigh_list;
- Param_Float* params_f;
- Param_Float* cu_params_f;
- cCudaData<int, int, xyz >* cu_elem2param;
- cCudaData<int, int, x >* cu_map;
- bool init;
- bool iszbl;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pair_tersoff_zbl_cuda.cpp b/src/USER-CUDA/pair_tersoff_zbl_cuda.cpp
deleted file mode 100644
index 91dcf6189e..0000000000
--- a/src/USER-CUDA/pair_tersoff_zbl_cuda.cpp
+++ /dev/null
@@ -1,220 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing author: Aidan Thompson (SNL) - original Tersoff implementation
- David Farrell (NWU) - ZBL addition
-------------------------------------------------------------------------- */
-
-#include <math.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include "pair_tersoff_zbl_cuda.h"
-#include "atom.h"
-#include "update.h"
-#include "neighbor.h"
-#include "neigh_list.h"
-#include "neigh_request.h"
-#include "force.h"
-#include "comm.h"
-#include "memory.h"
-#include "error.h"
-#include "math_const.h"
-
-using namespace LAMMPS_NS;
-using namespace MathConst;
-
-#define MAXLINE 1024
-#define DELTA 4
-
-/* ---------------------------------------------------------------------- */
-
-PairTersoffZBLCuda::PairTersoffZBLCuda(LAMMPS *lmp) : PairTersoffCuda(lmp)
-{
- // hard-wired constants in metal or real units
- // a0 = Bohr radius
- // epsilon0 = permittivity of vacuum = q / energy-distance units
- // e = unit charge
- // 1 Kcal/mole = 0.043365121 eV
-
- if (strcmp(update->unit_style,"metal") == 0) {
- global_a_0 = 0.529;
- global_epsilon_0 = 0.00552635;
- global_e = 1.0;
- } else if (strcmp(update->unit_style,"real") == 0) {
- global_a_0 = 0.529;
- global_epsilon_0 = 0.00552635 * 0.043365121;
- global_e = 1.0;
- } else error->all(FLERR,"Pair tersoff/zbl requires metal or real units");
- iszbl = true;
-}
-
-/* ---------------------------------------------------------------------- */
-
-void PairTersoffZBLCuda::read_file(char *file)
-{
- int params_per_line = 21;
- char **words = new char*[params_per_line+1];
-
- delete [] params;
- params = NULL;
- nparams = 0;
-
- // open file on proc 0
-
- FILE *fp;
- if (comm->me == 0) {
- fp = fopen(file,"r");
- if (fp == NULL) {
- char str[128];
- sprintf(str,"Cannot open Tersoff potential file %s",file);
- error->one(FLERR,str);
- }
- }
-
- // read each line out of file, skipping blank lines or leading '#'
- // store line of params if all 3 element tags are in element list
-
- int n,nwords,ielement,jelement,kelement;
- char line[MAXLINE],*ptr;
- int eof = 0;
-
- while (1) {
- if (comm->me == 0) {
- ptr = fgets(line,MAXLINE,fp);
- if (ptr == NULL) {
- eof = 1;
- fclose(fp);
- } else n = strlen(line) + 1;
- }
- MPI_Bcast(&eof,1,MPI_INT,0,world);
- if (eof) break;
- MPI_Bcast(&n,1,MPI_INT,0,world);
- MPI_Bcast(line,n,MPI_CHAR,0,world);
-
- // strip comment, skip line if blank
-
- if ((ptr = strchr(line,'#'))) *ptr = '\0';
- nwords = atom->count_words(line);
- if (nwords == 0) continue;
-
- // concatenate additional lines until have params_per_line words
-
- while (nwords < params_per_line) {
- n = strlen(line);
- if (comm->me == 0) {
- ptr = fgets(&line[n],MAXLINE-n,fp);
- if (ptr == NULL) {
- eof = 1;
- fclose(fp);
- } else n = strlen(line) + 1;
- }
- MPI_Bcast(&eof,1,MPI_INT,0,world);
- if (eof) break;
- MPI_Bcast(&n,1,MPI_INT,0,world);
- MPI_Bcast(line,n,MPI_CHAR,0,world);
- if ((ptr = strchr(line,'#'))) *ptr = '\0';
- nwords = atom->count_words(line);
- }
-
- if (nwords != params_per_line)
- error->all(FLERR,"Incorrect format in Tersoff potential file");
-
- // words = ptrs to all words in line
-
- nwords = 0;
- words[nwords++] = strtok(line," \t\n\r\f");
- while ((words[nwords++] = strtok(NULL," \t\n\r\f"))) continue;
-
- // ielement,jelement,kelement = 1st args
- // if all 3 args are in element list, then parse this line
- // else skip to next line
-
- for (ielement = 0; ielement < nelements; ielement++)
- if (strcmp(words[0],elements[ielement]) == 0) break;
- if (ielement == nelements) continue;
- for (jelement = 0; jelement < nelements; jelement++)
- if (strcmp(words[1],elements[jelement]) == 0) break;
- if (jelement == nelements) continue;
- for (kelement = 0; kelement < nelements; kelement++)
- if (strcmp(words[2],elements[kelement]) == 0) break;
- if (kelement == nelements) continue;
-
- // load up parameter settings and error check their values
-
- if (nparams == maxparam) {
- maxparam += DELTA;
- params = (Param *) memory->srealloc(params,maxparam*sizeof(Param),
- "pair:params");
- }
-
- params[nparams].ielement = ielement;
- params[nparams].jelement = jelement;
- params[nparams].kelement = kelement;
- params[nparams].powerm = atof(words[3]);
- params[nparams].gamma = atof(words[4]);
- params[nparams].lam3 = atof(words[5]);
- params[nparams].c = atof(words[6]);
- params[nparams].d = atof(words[7]);
- params[nparams].h = atof(words[8]);
- params[nparams].powern = atof(words[9]);
- params[nparams].beta = atof(words[10]);
- params[nparams].lam2 = atof(words[11]);
- params[nparams].bigb = atof(words[12]);
- params[nparams].bigr = atof(words[13]);
- params[nparams].bigd = atof(words[14]);
- params[nparams].lam1 = atof(words[15]);
- params[nparams].biga = atof(words[16]);
- params[nparams].Z_i = atof(words[17]);
- params[nparams].Z_j = atof(words[18]);
- params[nparams].ZBLcut = atof(words[19]);
- params[nparams].ZBLexpscale = atof(words[20]);
-
- // currently only allow m exponent of 1 or 3
-
- params[nparams].powermint = int(params[nparams].powerm);
-
- if (
- params[nparams].lam3 < 0.0 || params[nparams].c < 0.0 ||
- params[nparams].d < 0.0 || params[nparams].powern < 0.0 ||
- params[nparams].beta < 0.0 || params[nparams].lam2 < 0.0 ||
- params[nparams].bigb < 0.0 || params[nparams].bigr < 0.0 ||
- params[nparams].bigd < 0.0 ||
- params[nparams].bigd > params[nparams].bigr ||
- params[nparams].lam3 < 0.0 || params[nparams].biga < 0.0 ||
- params[nparams].powerm - params[nparams].powermint != 0.0 ||
- (params[nparams].powermint != 3 && params[nparams].powermint != 1) ||
- params[nparams].gamma < 0.0 ||
- params[nparams].Z_i < 1.0 || params[nparams].Z_j < 1.0 ||
- params[nparams].ZBLcut < 0.0 || params[nparams].ZBLexpscale < 0.0)
- error->all(FLERR,"Illegal Tersoff parameter");
-
- nparams++;
- }
-
- delete [] words;
-}
-
-void PairTersoffZBLCuda::coeff(int narg, char **arg)
-{
- PairTersoffCuda::coeff(narg, arg);
- for(int i=0;i<maxparam;i++)
- {
- params_f[i].a_ij = (0.8854*global_a_0) /
- (pow(params[i].Z_i,0.23) + pow(params[i].Z_j,0.23));
- params_f[i].premult = (params[i].Z_i * params[i].Z_j * pow(global_e,2.0))/(4.0*MY_PI*global_epsilon_0);
- params_f[i].ZBLcut = params[i].ZBLcut;
- params_f[i].ZBLexpscale = params[i].ZBLexpscale;
- }
-}
diff --git a/src/USER-CUDA/pair_tersoff_zbl_cuda.h b/src/USER-CUDA/pair_tersoff_zbl_cuda.h
deleted file mode 100644
index 4b935b86b2..0000000000
--- a/src/USER-CUDA/pair_tersoff_zbl_cuda.h
+++ /dev/null
@@ -1,53 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef PAIR_CLASS
-
-PairStyle(tersoff/zbl/cuda,PairTersoffZBLCuda)
-
-#else
-
-#ifndef PAIR_TERSOFF_ZBL_CUDA_H
-#define PAIR_TERSOFF_ZBL_CUDA_H
-
-#include "pair_tersoff_cuda.h"
-
-namespace LAMMPS_NS {
-
-class PairTersoffZBLCuda : public PairTersoffCuda
-{
- public:
- PairTersoffZBLCuda(class LAMMPS *);
- private:
- double global_a_0; // Bohr radius for Coulomb repulsion
- double global_epsilon_0; // permittivity of vacuum for Coulomb repulsion
- double global_e; // proton charge (negative of electron charge)
-
- void read_file(char *);
- void coeff(int narg, char **arg);
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pppm_cuda.cpp b/src/USER-CUDA/pppm_cuda.cpp
deleted file mode 100644
index b322c9dd17..0000000000
--- a/src/USER-CUDA/pppm_cuda.cpp
+++ /dev/null
@@ -1,1420 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing authors: Roy Pollock (LLNL), Paul Crozier (SNL)
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <cstring>
-#include <cstdio>
-#include <cstdlib>
-#include <cmath>
-#include "pppm_cuda.h"
-#include "atom.h"
-#include "comm.h"
-#include "neighbor.h"
-#include "force.h"
-#include "fft3d_wrap_cuda.h" // has to come before pair.h to avoid clash with kokkos
-#include "pair.h"
-#include "bond.h"
-#include "angle.h"
-#include "domain.h"
-#include "remap_wrap.h"
-#include "memory.h"
-#include "error.h"
-#include "update.h"
-#include <ctime> //crmadd
-#include "cuda_wrapper_cu.h"
-#include "pppm_cuda_cu.h"
-#include "user_cuda.h"
-#include "math_const.h"
-
-using namespace LAMMPS_NS;
-using namespace MathConst;
-
-#define MAXORDER 7
-#define OFFSET 4096
-#define SMALL 0.00001
-#define LARGE 10000.0
-#define EPS_HOC 1.0e-7
-
-void printArray(double* data,int nx, int ny, int nz)
-{
- for(int i=0;i<nz;i++)
- for(int j=0;j<ny;j++)
- {
- printf("%i %i\n",i,j);
- for(int k=0;k<nx;k++)
- printf("%e ",data[2*(i*ny*nx+j*nx+k)]);
- printf("\n\n");
- }
-}
-void printArray(double*** data,int nx, int ny, int nz)
-{
- for(int i=0;i<nx;i++)
- for(int j=0;j<ny;j++)
- {
- printf("%i %i\n",i,j);
- for(int k=0;k<nz;k++)
- printf("%e ",data[i][j][k]);
- printf("\n\n");
- }
-}
-/* ---------------------------------------------------------------------- */
-
-PPPMCuda::PPPMCuda(LAMMPS *lmp, int narg, char **arg) :
- PPPMOld(lmp, (narg==2?1:narg), arg)
-{
- cuda = lmp->cuda;
- if(cuda == NULL)
- error->all(FLERR,"You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- if ((narg > 3)||(narg<1)) error->all(FLERR,"Illegal kspace_style pppm/cuda command");
- #ifndef FFT_CUFFT
- error->all(FLERR,"Using kspace_style pppm/cuda without cufft is not possible. Compile with cufft=1 to include cufft. Aborting.");
- #endif
-
- triclinic_support = 0;
- accuracy_relative = fabs(force->numeric(FLERR,arg[0]));
-
- nfactors = 3;
- factors = new int[nfactors];
- factors[0] = 2;
- factors[1] = 3;
- factors[2] = 5;
-
- MPI_Comm_rank(world,&me);
- MPI_Comm_size(world,&nprocs);
-
- density_brick = vdx_brick = vdy_brick = vdz_brick = vdx_brick_tmp = NULL;
- density_fft = NULL;
- greensfn = NULL;
- work1 = work2 = NULL;
- vg = NULL;
- fkx = fky = fkz = NULL;
- buf1 = buf2 = NULL;
-
- gf_b = NULL;
- rho1d = rho_coeff = NULL;
-
- fft1c = fft2c = NULL;
- remap = NULL;
-
- density_brick_int=NULL;
- density_intScale=1000000;
- cu_vdx_brick = cu_vdy_brick = cu_vdz_brick = NULL;
- cu_density_brick = NULL;
- cu_density_brick_int = NULL;
- cu_density_fft = NULL;
- cu_energy=NULL;
- cu_greensfn = NULL;
- cu_work1 = cu_work2 = cu_work3 = NULL;
- cu_vg = NULL;
- cu_fkx = cu_fky = cu_fkz = NULL;
-
- cu_flag = NULL;
- cu_debugdata = NULL;
- cu_rho_coeff = NULL;
- cu_virial = NULL;
-
- cu_gf_b = NULL;
-
- cu_slabbuf = NULL;
- slabbuf = NULL;
-
- nmax = 0;
- part2grid = NULL;
- cu_part2grid = NULL;
- adev_data_array=NULL;
- poissontime=0;
- old_nmax=0;
- cu_pppm_grid_n=NULL;
- cu_pppm_grid_ids=NULL;
-
- pppm_grid_nmax=0;
- pppm2partgrid=new int[3];
- pppm_grid=new int[3];
- firstpass=true;
- scale = 1.0;
-}
-
-
-/* ----------------------------------------------------------------------
- free all memory
-------------------------------------------------------------------------- */
-
-PPPMCuda::~PPPMCuda()
-{
- delete [] slabbuf;
- delete cu_slabbuf;
-
- delete [] factors;
- factors=NULL;
- deallocate();
- delete cu_part2grid;
- cu_part2grid=NULL;
- memory->destroy(part2grid);
- part2grid = NULL;
-}
-
-/* ----------------------------------------------------------------------
- called once before run
-------------------------------------------------------------------------- */
-
-void PPPMCuda::init()
-{
- cuda->shared_data.pppm.cudable_force=1;
-
- //if(cuda->finished_run) {PPPM::init(); return;}
-
- if (me == 0) {
- if (screen) fprintf(screen,"PPPMCuda initialization ...\n");
- if (logfile) fprintf(logfile,"PPPMCuda initialization ...\n");
- }
-
- // error check
-
- if (domain->dimension == 2) error->all(FLERR,"Cannot use PPPMCuda with 2d simulation");
- if (comm->style != 0)
- error->universe_all(FLERR,"PPPMCuda can only currently be used with "
- "comm_style brick");
-
- if (!atom->q_flag) error->all(FLERR,"Kspace style requires atom attribute q");
-
- if (slabflag == 0 && domain->nonperiodic > 0)
- error->all(FLERR,"Cannot use nonperiodic boundaries with PPPMCuda");
- if (slabflag == 1) {
- if (domain->xperiodic != 1 || domain->yperiodic != 1 ||
- domain->boundary[2][0] != 1 || domain->boundary[2][1] != 1)
- error->all(FLERR,"Incorrect boundaries with slab PPPMCuda");
- }
-
- if (order < 2 || order > MAXORDER) {
- char str[128];
- sprintf(str,"PPPMCuda order cannot be smaller than 2 or greater than %d",MAXORDER);
- error->all(FLERR,str);
- }
- // free all arrays previously allocated
-
- deallocate();
-
- // extract short-range Coulombic cutoff from pair style
-
- triclinic_check();
-
- if (force->pair == NULL)
- error->all(FLERR,"KSpace style is incompatible with Pair style");
- int itmp=0;
- double *p_cutoff = (double *) force->pair->extract("cut_coul",itmp);
- if (p_cutoff == NULL)
- error->all(FLERR,"KSpace style is incompatible with Pair style");
- cutoff = *p_cutoff;
-
- // if kspace is TIP4P, extract TIP4P params from pair style
-
- qdist = 0.0;
-
- if (strcmp(force->kspace_style,"pppm/tip4p") == 0) {
- if (force->pair == NULL)
- error->all(FLERR,"KSpace style is incompatible with Pair style");
- double *p_qdist = (double *) force->pair->extract("qdist",itmp);
- int *p_typeO = (int *) force->pair->extract("typeO",itmp);
- int *p_typeH = (int *) force->pair->extract("typeH",itmp);
- int *p_typeA = (int *) force->pair->extract("typeA",itmp);
- int *p_typeB = (int *) force->pair->extract("typeB",itmp);
- if (!p_qdist || !p_typeO || !p_typeH || !p_typeA || !p_typeB)
- error->all(FLERR,"KSpace style is incompatible with Pair style");
- qdist = *p_qdist;
- typeO = *p_typeO;
- typeH = *p_typeH;
- int typeA = *p_typeA;
- int typeB = *p_typeB;
-
- if (force->angle == NULL || force->bond == NULL)
- error->all(FLERR,"Bond and angle potentials must be defined for TIP4P");
- double theta = force->angle->equilibrium_angle(typeA);
- double blen = force->bond->equilibrium_distance(typeB);
- alpha = qdist / (2.0 * cos(0.5*theta) * blen);
- }
-
- // compute qsum & qsqsum and warn if not charge-neutral
-
- scale = 1.0;
- qqrd2e = force->qqrd2e;
- qsum_qsq();
- natoms_original = atom->natoms;
-
- // set accuracy (force units) from accuracy_relative or accuracy_absolute
-
- if (accuracy_absolute >= 0.0) accuracy = accuracy_absolute;
- else accuracy = accuracy_relative * two_charge_force;
-
- // setup FFT grid resolution and g_ewald
- // normally one iteration thru while loop is all that is required
- // if grid stencil extends beyond neighbor proc, reduce order and try again
-
- int iteration = 0;
-
- while (order > 1) {
- if (iteration && me == 0)
- error->warning(FLERR,"Reducing PPPMCuda order b/c stencil extends "
- "beyond neighbor processor");
- iteration++;
-
- set_grid();
-
- if (nx_pppm >= OFFSET || ny_pppm >= OFFSET || nz_pppm >= OFFSET)
- error->all(FLERR,"PPPMCuda grid is too large");
-
- // global indices of PPPMCuda grid range from 0 to N-1
- // nlo_in,nhi_in = lower/upper limits of the 3d sub-brick of
- // global PPPMCuda grid that I own without ghost cells
- // for slab PPPMCuda, assign z grid as if it were not extended
-
- nxlo_in = comm->myloc[0]*nx_pppm / comm->procgrid[0];
- nxhi_in = (comm->myloc[0]+1)*nx_pppm / comm->procgrid[0] - 1;
- nylo_in = comm->myloc[1]*ny_pppm / comm->procgrid[1];
- nyhi_in = (comm->myloc[1]+1)*ny_pppm / comm->procgrid[1] - 1;
- nzlo_in = comm->myloc[2] *
- (static_cast<int> (nz_pppm/slab_volfactor)) / comm->procgrid[2];
- nzhi_in = (comm->myloc[2]+1) *
- (static_cast<int> (nz_pppm/slab_volfactor)) / comm->procgrid[2] - 1;
-
- // nlower,nupper = stencil size for mapping particles to PPPMCuda grid
-
- nlower = -(order-1)/2;
- nupper = order/2;
-
- // shift values for particle <-> grid mapping
- // add/subtract OFFSET to avoid int(-0.75) = 0 when want it to be -1
-
- if (order % 2) shift = OFFSET + 0.5;
- else shift = OFFSET;
- if (order % 2) shiftone = 0.0;
- else shiftone = 0.5;
-
- // nlo_out,nhi_out = lower/upper limits of the 3d sub-brick of
- // global PPPMCuda grid that my particles can contribute charge to
- // effectively nlo_in,nhi_in + ghost cells
- // nlo,nhi = global coords of grid pt to "lower left" of smallest/largest
- // position a particle in my box can be at
- // dist[3] = particle position bound = subbox + skin/2.0 + qdist
- // qdist = offset due to TIP4P fictitious charge
- // convert to triclinic if necessary
- // nlo_out,nhi_out = nlo,nhi + stencil size for particle mapping
- // for slab PPPMCuda, assign z grid as if it were not extended
-
-
- triclinic = domain->triclinic;
- double *prd,*sublo,*subhi;
-
- if (triclinic == 0) {
- prd = domain->prd;
- boxlo = domain->boxlo;
- sublo = domain->sublo;
- subhi = domain->subhi;
- } else {
- prd = domain->prd_lamda;
- boxlo = domain->boxlo_lamda;
- sublo = domain->sublo_lamda;
- subhi = domain->subhi_lamda;
- }
-
- double xprd = prd[0];
- double yprd = prd[1];
- double zprd = prd[2];
- double zprd_slab = zprd*slab_volfactor;
-
- double dist[3];
- double cuthalf = 0.5*neighbor->skin + qdist;
- if (triclinic == 0) dist[0] = dist[1] = dist[2] = cuthalf;
- else {
- dist[0] = cuthalf/domain->prd[0];
- dist[1] = cuthalf/domain->prd[1];
- dist[2] = cuthalf/domain->prd[2];
- }
-
- int nlo,nhi;
-
- nlo = static_cast<int> ((sublo[0]-dist[0]-boxlo[0]) *
- nx_pppm/xprd + shift) - OFFSET;
- nhi = static_cast<int> ((subhi[0]+dist[0]-boxlo[0]) *
- nx_pppm/xprd + shift) - OFFSET;
- nxlo_out = nlo + nlower;
- nxhi_out = nhi + nupper;
-
- nlo = static_cast<int> ((sublo[1]-dist[1]-boxlo[1]) *
- ny_pppm/yprd + shift) - OFFSET;
- nhi = static_cast<int> ((subhi[1]+dist[1]-boxlo[1]) *
- ny_pppm/yprd + shift) - OFFSET;
- nylo_out = nlo + nlower;
- nyhi_out = nhi + nupper;
-
- nlo = static_cast<int> ((sublo[2]-dist[2]-boxlo[2]) *
- nz_pppm/zprd_slab + shift) - OFFSET;
- nhi = static_cast<int> ((subhi[2]+dist[2]-boxlo[2]) *
- nz_pppm/zprd_slab + shift) - OFFSET;
- nzlo_out = nlo + nlower;
- nzhi_out = nhi + nupper;
-
- // for slab PPPMCuda, change the grid boundary for processors at +z end
- // to include the empty volume between periodically repeating slabs
- // for slab PPPMCuda, want charge data communicated from -z proc to +z proc,
- // but not vice versa, also want field data communicated from +z proc to
- // -z proc, but not vice versa
- // this is accomplished by nzhi_in = nzhi_out on +z end (no ghost cells)
-
- if (slabflag && ((comm->myloc[2]+1) == (comm->procgrid[2]))) {
- nzhi_in = nz_pppm - 1;
- nzhi_out = nz_pppm - 1;
- }
-
- // nlo_ghost,nhi_ghost = # of planes I will recv from 6 directions
- // that overlay domain I own
- // proc in that direction tells me via sendrecv()
- // if no neighbor proc, value is from self since I have ghosts regardless
-
- int nplanes;
-
- nplanes = nxlo_in - nxlo_out;
- if (comm->procneigh[0][0] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[0][0],0,
- &nxhi_ghost,1,MPI_INT,comm->procneigh[0][1],0,
- world,MPI_STATUS_IGNORE);
- else nxhi_ghost = nplanes;
-
- nplanes = nxhi_out - nxhi_in;
- if (comm->procneigh[0][1] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[0][1],0,
- &nxlo_ghost,1,MPI_INT,comm->procneigh[0][0],
- 0,world,MPI_STATUS_IGNORE);
- else nxlo_ghost = nplanes;
-
- nplanes = nylo_in - nylo_out;
- if (comm->procneigh[1][0] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[1][0],0,
- &nyhi_ghost,1,MPI_INT,comm->procneigh[1][1],0,
- world,MPI_STATUS_IGNORE);
- else nyhi_ghost = nplanes;
-
- nplanes = nyhi_out - nyhi_in;
- if (comm->procneigh[1][1] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[1][1],0,
- &nylo_ghost,1,MPI_INT,comm->procneigh[1][0],0,
- world,MPI_STATUS_IGNORE);
- else nylo_ghost = nplanes;
-
- nplanes = nzlo_in - nzlo_out;
- if (comm->procneigh[2][0] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[2][0],0,
- &nzhi_ghost,1,MPI_INT,comm->procneigh[2][1],0,
- world,MPI_STATUS_IGNORE);
- else nzhi_ghost = nplanes;
-
- nplanes = nzhi_out - nzhi_in;
- if (comm->procneigh[2][1] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[2][1],0,
- &nzlo_ghost,1,MPI_INT,comm->procneigh[2][0],0,
- world,MPI_STATUS_IGNORE);
- else nzlo_ghost = nplanes;
-
- // test that ghost overlap is not bigger than my sub-domain
-
- int flag = 0;
- if (nxlo_ghost > nxhi_in-nxlo_in+1) flag = 1;
- if (nxhi_ghost > nxhi_in-nxlo_in+1) flag = 1;
- if (nylo_ghost > nyhi_in-nylo_in+1) flag = 1;
- if (nyhi_ghost > nyhi_in-nylo_in+1) flag = 1;
- if (nzlo_ghost > nzhi_in-nzlo_in+1) flag = 1;
- if (nzhi_ghost > nzhi_in-nzlo_in+1) flag = 1;
-
- int flag_all;
- MPI_Allreduce(&flag,&flag_all,1,MPI_INT,MPI_SUM,world);
-
- if (flag_all == 0) break;
- order--;
- }
-
- if (order == 0) error->all(FLERR,"PPPMCuda order has been reduced to 0");
-
- // decomposition of FFT mesh
- // global indices range from 0 to N-1
- // proc owns entire x-dimension, clump of columns in y,z dimensions
- // npey_fft,npez_fft = # of procs in y,z dims
- // if nprocs is small enough, proc can own 1 or more entire xy planes,
- // else proc owns 2d sub-blocks of yz plane
- // me_y,me_z = which proc (0-npe_fft-1) I am in y,z dimensions
- // nlo_fft,nhi_fft = lower/upper limit of the section
- // of the global FFT mesh that I own
-
- int npey_fft,npez_fft;
- if (nz_pppm >= nprocs) {
- npey_fft = 1;
- npez_fft = nprocs;
- } else procs2grid2d(nprocs,ny_pppm,nz_pppm,&npey_fft,&npez_fft);
-
- int me_y = me % npey_fft;
- int me_z = me / npey_fft;
-
- nxlo_fft = 0;
- nxhi_fft = nx_pppm - 1;
- nylo_fft = me_y*ny_pppm/npey_fft;
- nyhi_fft = (me_y+1)*ny_pppm/npey_fft - 1;
- nzlo_fft = me_z*nz_pppm/npez_fft;
- nzhi_fft = (me_z+1)*nz_pppm/npez_fft - 1;
-
- // PPPMCuda grid for this proc, including ghosts
-
- ngrid = (nxhi_out-nxlo_out+1) * (nyhi_out-nylo_out+1) *
- (nzhi_out-nzlo_out+1);
-
- // FFT arrays on this proc, without ghosts
- // nfft = FFT points in FFT decomposition on this proc
- // nfft_brick = FFT points in 3d brick-decomposition on this proc
- // nfft_both = greater of 2 values
-
- nfft = (nxhi_fft-nxlo_fft+1) * (nyhi_fft-nylo_fft+1) *
- (nzhi_fft-nzlo_fft+1);
- int nfft_brick = (nxhi_in-nxlo_in+1) * (nyhi_in-nylo_in+1) *
- (nzhi_in-nzlo_in+1);
- nfft_both = MAX(nfft,nfft_brick);
-
- // buffer space for use in brick2fft and fillbrick
- // idel = max # of ghost planes to send or recv in +/- dir of each dim
- // nx,ny,nz = owned planes (including ghosts) in each dim
- // nxx,nyy,nzz = max # of grid cells to send in each dim
- // nbuf = max in any dim, augment by 3x for components of vd_xyz in fillbrick
-
- int idelx,idely,idelz,nx,ny,nz,nxx,nyy,nzz;
-
- idelx = MAX(nxlo_ghost,nxhi_ghost);
- idelx = MAX(idelx,nxhi_out-nxhi_in);
- idelx = MAX(idelx,nxlo_in-nxlo_out);
-
- idely = MAX(nylo_ghost,nyhi_ghost);
- idely = MAX(idely,nyhi_out-nyhi_in);
- idely = MAX(idely,nylo_in-nylo_out);
-
- idelz = MAX(nzlo_ghost,nzhi_ghost);
- idelz = MAX(idelz,nzhi_out-nzhi_in);
- idelz = MAX(idelz,nzlo_in-nzlo_out);
-
- nx = nxhi_out - nxlo_out + 1;
- ny = nyhi_out - nylo_out + 1;
- nz = nzhi_out - nzlo_out + 1;
-
- nxx = idelx * ny * nz;
- nyy = idely * nx * nz;
- nzz = idelz * nx * ny;
-
- nbuf = MAX(nxx,nyy);
- nbuf = MAX(nbuf,nzz);
- nbuf *= 3;
-
- // print stats
-
- int ngrid_max,nfft_both_max,nbuf_max;
- MPI_Allreduce(&ngrid,&ngrid_max,1,MPI_INT,MPI_MAX,world);
- MPI_Allreduce(&nfft_both,&nfft_both_max,1,MPI_INT,MPI_MAX,world);
- MPI_Allreduce(&nbuf,&nbuf_max,1,MPI_INT,MPI_MAX,world);
-
- if (me == 0) {
- if (screen) fprintf(screen," brick FFT buffer size/proc = %d %d %d\n",
- ngrid_max,nfft_both_max,nbuf_max);
- if (logfile) fprintf(logfile," brick FFT buffer size/proc = %d %d %d\n",
- ngrid_max,nfft_both_max,nbuf_max);
- }
- cuda_shared_pppm* ap=&(cuda->shared_data.pppm);
-
- ap->density_intScale=density_intScale;
- ap->nxlo_in=nxlo_in;
- ap->nxhi_in=nxhi_in;
- ap->nxlo_out=nxlo_out;
- ap->nxhi_out=nxhi_out;
- ap->nylo_in=nylo_in;
- ap->nyhi_in=nyhi_in;
- ap->nylo_out=nylo_out;
- ap->nyhi_out=nyhi_out;
- ap->nzlo_in=nzlo_in;
- ap->nzhi_in=nzhi_in;
- ap->nzlo_out=nzlo_out;
- ap->nzhi_out=nzhi_out;
- ap->nxlo_in=nxlo_fft;
- ap->nxhi_in=nxhi_fft;
- ap->nylo_in=nylo_fft;
- ap->nyhi_in=nyhi_fft;
- ap->nzlo_in=nzlo_fft;
- ap->nzhi_in=nzhi_fft;
- ap->nx_pppm=nx_pppm;
- ap->ny_pppm=ny_pppm;
- ap->nz_pppm=nz_pppm;
- ap->qqrd2e=qqrd2e;
- ap->order=order;
- ap->nmax=nmax;
- ap->nlocal=atom->nlocal;
- ap->delxinv=delxinv;
- ap->delyinv=delyinv;
- ap->delzinv=delzinv;
- ap->nlower=nlower;
- ap->nupper=nupper;
- ap->shiftone=shiftone;
-
- // allocate K-space dependent memory
-
-
- allocate();
-
- // pre-compute Green's function denomiator expansion
- // pre-compute 1d charge distribution coefficients
-
- compute_gf_denom();
- compute_rho_coeff();
-}
-
-/* ----------------------------------------------------------------------
- adjust PPPMCuda coeffs, called initially and whenever volume has changed
-------------------------------------------------------------------------- */
-
-void PPPMCuda::setup()
-{
- double *prd;
- cu_gf_b->upload();
- // volume-dependent factors
- // adjust z dimension for 2d slab PPPMCuda
- // z dimension for 3d PPPMCuda is zprd since slab_volfactor = 1.0
-
- if (triclinic == 0) prd = domain->prd;
- else prd = domain->prd_lamda;
-
- double xprd = prd[0];
- double yprd = prd[1];
- double zprd = prd[2];
- double zprd_slab = zprd*slab_volfactor;
- volume = xprd * yprd * zprd_slab;
-
- delxinv = nx_pppm/xprd;
- delyinv = ny_pppm/yprd;
- delzinv = nz_pppm/zprd_slab;
-
- delvolinv = delxinv*delyinv*delzinv;
-
- double unitkx = (2.0*MY_PI/xprd);
- double unitky = (2.0*MY_PI/yprd);
- double unitkz = (2.0*MY_PI/zprd_slab);
-
- // fkx,fky,fkz for my FFT grid pts
- Cuda_PPPM_Setup_fkxyz_vg(nx_pppm, ny_pppm,nz_pppm,unitkx,unitky,unitkz,g_ewald);
-
-
-
- // modified (Hockney-Eastwood) Coulomb Green's function
-
- int nbx = static_cast<int> ((g_ewald*xprd/(MY_PI*nx_pppm)) *
- pow(-log(EPS_HOC),0.25));
- int nby = static_cast<int> ((g_ewald*yprd/(MY_PI*ny_pppm)) *
- pow(-log(EPS_HOC),0.25));
- int nbz = static_cast<int> ((g_ewald*zprd_slab/(MY_PI*nz_pppm)) *
- pow(-log(EPS_HOC),0.25));
- Cuda_PPPM_setup_greensfn(nx_pppm,ny_pppm,nz_pppm,unitkx,unitky,unitkz,g_ewald,
-nbx,nby,nbz,xprd,yprd,zprd_slab);
-
-
-#ifdef FFT_CUFFT
- cu_vdx_brick->upload();
- cu_vdy_brick->upload();
- cu_vdz_brick->upload();
-#endif
- cu_rho_coeff->upload();
- cu_density_brick->memset_device(0);
- pppm_device_init_setup(&cuda->shared_data,shiftone,delxinv,delyinv,delzinv,nlower,nupper);
-}
-
-/* ----------------------------------------------------------------------
- compute the PPPMCuda long-range force, energy, virial
-------------------------------------------------------------------------- */
-
-void PPPMCuda::compute(int eflag, int vflag)
-{
- cuda_shared_atom* cu_atom = & cuda->shared_data.atom;
-
- int i;
- my_times starttime;
- my_times endtime;
- my_times starttotal;
- my_times endtotal;
- // convert atoms from box to lamda coords
-
- if (triclinic == 0) boxlo = domain->boxlo;
- else {
- boxlo = domain->boxlo_lamda;
- domain->x2lamda(atom->nlocal);
- }
-
- // extend size of PPPM per-atom arrays if necessary
- // force update of device data, if arrays resized
-
-
- if (cu_atom->update_nmax || old_nmax == 0) {
- memory->destroy(part2grid);
- nmax = atom->nmax;
- memory->create(part2grid,nmax,3,"pppm:part2grid");
- delete cu_part2grid;
- delete [] adev_data_array;
- adev_data_array=new dev_array[1];
- cu_part2grid = new cCudaData<int , int , yx >
- ((int*)part2grid,adev_data_array, nmax,3);
-
- pppm_device_update(&cuda->shared_data,cu_part2grid->dev_data(),
- atom->nlocal,atom->nmax);
- old_nmax=nmax;
- }
- if(cu_atom->update_nlocal) {pppm_update_nlocal(cu_atom->nlocal);}
-
- energy = 0.0;
- if (vflag)
- {
- for (i = 0; i < 6; i++) virial[i] = 0.0;
- cu_virial->memset_device(0);
- }
- if(eflag) cu_energy->memset_device(0);
- my_gettime(CLOCK_REALTIME,&starttotal);
-
- // find grid points for all my particles
- // map my particle charge onto my local 3d density grid
-
- my_gettime(CLOCK_REALTIME,&starttime);
-
- particle_map();
-
- my_gettime(CLOCK_REALTIME,&endtime);
- cuda->shared_data.cuda_timings.pppm_particle_map+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- //cu_part2grid->download();
- my_gettime(CLOCK_REALTIME,&starttime);
- make_rho();
- my_gettime(CLOCK_REALTIME,&endtime);
- cuda->shared_data.cuda_timings.pppm_make_rho+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- // all procs communicate density values from their ghost cells
- // to fully sum contribution in their 3d bricks
- // remap from 3d decomposition to FFT decomposition
-
- int nprocs=comm->nprocs;
-
- my_gettime(CLOCK_REALTIME,&starttime);
-
- if(nprocs>1)
- {
- cu_density_brick->download();
- brick2fft();
- }
- else
- {
- #ifdef FFT_CUFFT
- pppm_initfftdata(&cuda->shared_data,(PPPM_CFLOAT*)cu_density_brick->dev_data(),(FFT_CFLOAT*)cu_work2->dev_data());
- #endif
- }
-
- my_gettime(CLOCK_REALTIME,&endtime);
- cuda->shared_data.cuda_timings.pppm_brick2fft+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- // compute potential gradient on my FFT grid and
- // portion of e_long on this proc's FFT grid
- // return gradients (electric fields) in 3d brick decomposition
-
- my_gettime(CLOCK_REALTIME,&starttime);
- poisson(eflag,vflag);
- my_gettime(CLOCK_REALTIME,&endtime);
- cuda->shared_data.cuda_timings.pppm_poisson+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- // all procs communicate E-field values to fill ghost cells
- // surrounding their 3d bricks
-
- // not necessary since all the calculations are done on one proc
-
- // calculate the force on my particles
-
- my_gettime(CLOCK_REALTIME,&starttime);
- fieldforce();
- my_gettime(CLOCK_REALTIME,&endtime);
- cuda->shared_data.cuda_timings.pppm_fieldforce+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- // sum energy across procs and add in volume-dependent term
- // reset qsum and qsqsum if atom count has changed
-
- my_gettime(CLOCK_REALTIME,&endtotal);
- cuda->shared_data.cuda_timings.pppm_compute+=(endtotal.tv_sec-starttotal.tv_sec+1.0*(endtotal.tv_nsec-starttotal.tv_nsec)/1000000000);
-
- if (eflag) {
- double energy_all;
- MPI_Allreduce(&energy,&energy_all,1,MPI_DOUBLE,MPI_SUM,world);
- energy = energy_all;
-
- if (atom->natoms != natoms_original) {
- qsum_qsq();
- natoms_original = atom->natoms;
- }
-
- energy *= 0.5*volume;
- energy -= g_ewald*qsqsum/1.772453851 +
- MY_PI2*qsum*qsum / (g_ewald*g_ewald*volume);
- energy *= qqrd2e;
- }
-
- // sum virial across procs
-
- if (vflag) {
- double virial_all[6];
- MPI_Allreduce(virial,virial_all,6,MPI_DOUBLE,MPI_SUM,world);
- for (i = 0; i < 6; i++) virial[i] = 0.5*qqrd2e*volume*virial_all[i];
- }
-
- // 2d slab correction
-
- if (slabflag) slabcorr(eflag);
-
- // convert atoms back from lamda to box coords
-
- if (triclinic) domain->lamda2x(atom->nlocal);
-
- if(firstpass) firstpass=false;
-}
-
-
-/* ----------------------------------------------------------------------
- allocate memory that depends on # of K-vectors and order
-------------------------------------------------------------------------- */
-
-
-void PPPMCuda::allocate()
-{
-
- struct dev_array* dev_tmp=new struct dev_array[20];
- int n_cudata=0;
-
-
- memory->create3d_offset(density_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:density_brick");
- memory->create3d_offset(density_brick_int,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:density_brick_int");
-
-
- cu_density_brick = new cCudaData<double, PPPM_CFLOAT, x> ((double*) &(density_brick[nzlo_out][nylo_out][nxlo_out]), & (dev_tmp[n_cudata++]),
- (nzhi_out-nzlo_out+1)*(nyhi_out-nylo_out+1)*(nxhi_out-nxlo_out+1));
-
- cu_density_brick_int = new cCudaData<int, int, x> ((int*) &(density_brick_int[nzlo_out][nylo_out][nxlo_out]), & (dev_tmp[n_cudata++]),
- (nzhi_out-nzlo_out+1)*(nyhi_out-nylo_out+1)*(nxhi_out-nxlo_out+1));
-
- memory->create3d_offset(vdx_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:vdx_brick");
- memory->create3d_offset(vdx_brick_tmp,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:vdx_brick_tmp");
-
- cu_vdx_brick = new cCudaData<double, PPPM_CFLOAT, x> ((double*) &(vdx_brick[nzlo_out][nylo_out][nxlo_out]), & (dev_tmp[n_cudata++]),
- (nzhi_out-nzlo_out+1)*(nyhi_out-nylo_out+1)*(nxhi_out-nxlo_out+1));
-
- memory->create3d_offset(vdy_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:vdy_brick");
- cu_vdy_brick = new cCudaData<double, PPPM_CFLOAT, x> ((double*) &(vdy_brick[nzlo_out][nylo_out][nxlo_out]), & (dev_tmp[n_cudata++]),
- (nzhi_out-nzlo_out+1)*(nyhi_out-nylo_out+1)*(nxhi_out-nxlo_out+1));
-
- memory->create3d_offset(vdz_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:vdz_brick");
- cu_vdz_brick = new cCudaData<double, PPPM_CFLOAT, x> ((double*) &(vdz_brick[nzlo_out][nylo_out][nxlo_out]), & (dev_tmp[n_cudata++]),
- (nzhi_out-nzlo_out+1)*(nyhi_out-nylo_out+1)*(nxhi_out-nxlo_out+1));
-
- memory->create(density_fft,nfft_both,"pppm:density_fft");
-
- cu_density_fft = new cCudaData<double, PPPM_CFLOAT, x> (density_fft, & (dev_tmp[n_cudata++]),nfft_both);
-
- cu_energy = new cCudaData<double, ENERGY_CFLOAT, x> (NULL, &(dev_tmp[n_cudata++]),ny_pppm*nz_pppm);
- cu_virial = new cCudaData<double, ENERGY_CFLOAT, x> (NULL, &(dev_tmp[n_cudata++]),ny_pppm*nz_pppm*6);
-
- memory->create(greensfn,nfft_both,"pppm:greensfn");
- cu_greensfn = new cCudaData<double, PPPM_CFLOAT, x> (greensfn, & (dev_tmp[n_cudata++]) , nx_pppm*ny_pppm*nz_pppm);
-
- memory->create(work1,2*nx_pppm*ny_pppm*nz_pppm,"pppm:work1");
- memory->create(work2,2*nx_pppm*ny_pppm*nz_pppm,"pppm:work2");
- memory->create(work3,2*nx_pppm*ny_pppm*nz_pppm,"pppm:work3");
-
- cu_work1 = new cCudaData<double, FFT_CFLOAT, x> (work1, & (dev_tmp[n_cudata++]) , 2*nx_pppm*ny_pppm*nz_pppm);
- cu_work2 = new cCudaData<double, FFT_CFLOAT, x> (work2, & (dev_tmp[n_cudata++]) , 2*nx_pppm*ny_pppm*nz_pppm);
- cu_work3 = new cCudaData<double, FFT_CFLOAT, x> (work3, & (dev_tmp[n_cudata++]) , 2*nx_pppm*ny_pppm*nz_pppm);
-
-
- memory->create(fkx,nx_pppm,"pppmcuda:fkx");
- cu_fkx = new cCudaData<double, PPPM_CFLOAT, x> (fkx, & (dev_tmp[n_cudata++]) , nx_pppm);
- memory->create(fky,ny_pppm,"pppmcuda:fky");
- cu_fky = new cCudaData<double, PPPM_CFLOAT, x> (fky, & (dev_tmp[n_cudata++]) , ny_pppm);
- memory->create(fkz,nz_pppm,"pppmcuda:fkz");
- cu_fkz = new cCudaData<double, PPPM_CFLOAT, x> (fkz, & (dev_tmp[n_cudata++]) , nz_pppm);
-
- memory->create(vg,nfft_both,6,"pppm:vg");
-
- cu_vg = new cCudaData<double, PPPM_CFLOAT, xy> ((double*)vg, & (dev_tmp[n_cudata++]) , nfft_both,6);
-
- memory->create(buf1,nbuf,"pppm:buf1");
- memory->create(buf2,nbuf,"pppm:buf2");
-
-
- // summation coeffs
-
-
- gf_b = new double[order];
- cu_gf_b = new cCudaData<double,PPPM_CFLOAT,x> (gf_b, &(dev_tmp[n_cudata++]) , order);
- memory->create2d_offset(rho1d,3,-order/2,order/2,"pppm:rho1d");
- memory->create2d_offset(rho_coeff,order,(1-order)/2,order/2,"pppm:rho_coeff");
-
- cu_rho_coeff = new cCudaData<double, PPPM_CFLOAT, x> ((double*) &(rho_coeff[0][(1-order)/2]), & (dev_tmp[n_cudata++]) , order*(order/2-(1-order)/2+1));
-
- debugdata=new PPPM_CFLOAT[100];
- cu_debugdata = new cCudaData<PPPM_CFLOAT, PPPM_CFLOAT, x> (debugdata,& (dev_tmp[n_cudata++]),100);
- cu_flag = new cCudaData<int, int, x> (&global_flag,& (dev_tmp[n_cudata++]),3);
-
- // create 2 FFTs and a Remap
- // 1st FFT keeps data in FFT decompostion
- // 2nd FFT returns data in 3d brick decomposition
- // remap takes data from 3d brick to FFT decomposition
-
- int tmp;
-
-
-
-
- fft1c = new FFT3dCuda(lmp,world,nx_pppm,ny_pppm,nz_pppm,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- 0,0,&tmp,true);
-
- fft2c = new FFT3dCuda(lmp,world,nx_pppm,ny_pppm,nz_pppm,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- nxlo_in,nxhi_in,nylo_in,nyhi_in,nzlo_in,nzhi_in,
- 0,0,&tmp,false);
-
-
-#ifdef FFT_CUFFT
- fft1c->set_cudata(cu_work2->dev_data(),cu_work1->dev_data());
- fft2c->set_cudata(cu_work2->dev_data(),cu_work3->dev_data());
-#endif
-
- remap = new Remap(lmp,world,
- nxlo_in,nxhi_in,nylo_in,nyhi_in,nzlo_in,nzhi_in,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- 1,0,0,2,0);
-
-
-pppm_device_init(cu_density_brick->dev_data(), cu_vdx_brick->dev_data(), cu_vdy_brick->dev_data(), cu_vdz_brick->dev_data(), cu_density_fft->dev_data(),cu_energy->dev_data(),cu_virial->dev_data()
- , cu_work1->dev_data(), cu_work2->dev_data(), cu_work3->dev_data(), cu_greensfn->dev_data(), cu_fkx->dev_data(), cu_fky->dev_data(), cu_fkz->dev_data(), cu_vg->dev_data()
- ,nxlo_in,nxhi_in,nylo_in,nyhi_in,nzlo_in,nzhi_in,nxlo_out,nxhi_out,nylo_out,nyhi_out,nzlo_out,nzhi_out,nx_pppm,ny_pppm,nz_pppm
- ,nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,cu_gf_b->dev_data()
- ,qqrd2e,order,cu_rho_coeff->dev_data(),cu_debugdata->dev_data(),cu_density_brick_int->dev_data(),slabflag
- );
-}
-
-
-
-/* ----------------------------------------------------------------------
- deallocate memory that depends on # of K-vectors and order
- ---------------------------------------------------------------------- */
-
-void PPPMCuda::deallocate()
-{
- memory->destroy3d_offset(density_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(vdx_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(vdy_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(vdz_brick,nzlo_out,nylo_out,nxlo_out);
-
- density_brick = vdx_brick = vdy_brick = vdz_brick = NULL;
-
- memory->destroy(density_fft);
- memory->destroy(greensfn);
- memory->destroy(work1);
- memory->destroy(work2);
- memory->destroy(vg);
-
- density_fft = NULL;
- greensfn = NULL;
- work1 = NULL;
- work2 = NULL;
- vg = NULL;
-
- memory->destroy(fkx);
- memory->destroy(fky);
- memory->destroy(fkz);
-
- fkx = NULL;
- fky = NULL;
- fkz = NULL;
-
- delete cu_density_brick;
- delete cu_density_brick_int;
- delete cu_vdx_brick;
- delete cu_vdy_brick;
- delete cu_vdz_brick;
- delete cu_density_fft;
- delete cu_energy;
- delete cu_virial;
-#ifdef FFT_CUFFT
- delete cu_greensfn;
- delete cu_gf_b;
- delete cu_vg;
- delete cu_work1;
- delete cu_work2;
- delete cu_work3;
- delete cu_fkx;
- delete cu_fky;
- delete cu_fkz;
-#endif
-
- delete cu_flag;
- delete cu_debugdata;
- delete cu_rho_coeff;
-
-
- cu_vdx_brick = cu_vdy_brick = cu_vdz_brick = NULL;
- cu_density_brick = NULL;
- cu_density_brick_int = NULL;
- cu_density_fft = NULL;
- cu_energy=NULL;
- cu_virial=NULL;
-#ifdef FFT_CUFFT
- cu_greensfn = NULL;
- cu_gf_b = NULL;
- cu_work1 = cu_work2 = cu_work3 = NULL;
- cu_vg = NULL;
- cu_fkx = cu_fky = cu_fkz = NULL;
-#endif
-
- cu_flag = NULL;
- cu_debugdata = NULL;
- cu_rho_coeff = NULL;
- cu_part2grid = NULL;
-
- memory->destroy(buf1);
- memory->destroy(buf2);
-
- delete [] gf_b;
- gf_b = NULL;
- memory->destroy2d_offset(rho1d,-order/2); rho1d = NULL;
- memory->destroy2d_offset(rho_coeff,(1-order)/2); rho_coeff = NULL;
-
- delete fft1c;
- fft1c = NULL;
-
- delete fft2c;
- fft2c = NULL;
- delete remap;
- remap = NULL;
- buf1 = NULL;
- buf2 = NULL;
-}
-
-/* ----------------------------------------------------------------------
- set size of FFT grid (nx,ny,nz_pppm) and g_ewald
--------------------------------------------------------------------------*/
-
-void PPPMCuda::set_grid()
-{
- // see JCP 109, pg 7698 for derivation of coefficients
- // higher order coefficients may be computed if needed
-
- double **acons;
- memory->create(acons,8,7,"pppm:acons");
-
- acons[1][0] = 2.0 / 3.0;
- acons[2][0] = 1.0 / 50.0;
- acons[2][1] = 5.0 / 294.0;
- acons[3][0] = 1.0 / 588.0;
- acons[3][1] = 7.0 / 1440.0;
- acons[3][2] = 21.0 / 3872.0;
- acons[4][0] = 1.0 / 4320.0;
- acons[4][1] = 3.0 / 1936.0;
- acons[4][2] = 7601.0 / 2271360.0;
- acons[4][3] = 143.0 / 28800.0;
- acons[5][0] = 1.0 / 23232.0;
- acons[5][1] = 7601.0 / 13628160.0;
- acons[5][2] = 143.0 / 69120.0;
- acons[5][3] = 517231.0 / 106536960.0;
- acons[5][4] = 106640677.0 / 11737571328.0;
- acons[6][0] = 691.0 / 68140800.0;
- acons[6][1] = 13.0 / 57600.0;
- acons[6][2] = 47021.0 / 35512320.0;
- acons[6][3] = 9694607.0 / 2095994880.0;
- acons[6][4] = 733191589.0 / 59609088000.0;
- acons[6][5] = 326190917.0 / 11700633600.0;
- acons[7][0] = 1.0 / 345600.0;
- acons[7][1] = 3617.0 / 35512320.0;
- acons[7][2] = 745739.0 / 838397952.0;
- acons[7][3] = 56399353.0 / 12773376000.0;
- acons[7][4] = 25091609.0 / 1560084480.0;
- acons[7][5] = 1755948832039.0 / 36229939200000.0;
- acons[7][6] = 4887769399.0 / 37838389248.0;
-
- bigint natoms = atom->natoms;
-
- // use xprd,yprd,zprd even if triclinic so grid size is the same
- // adjust z dimension for 2d slab PPPMCuda
- // 3d PPPMCuda just uses zprd since slab_volfactor = 1.0
-
- double xprd = domain->xprd;
- double yprd = domain->yprd;
- double zprd = domain->zprd;
- double zprd_slab = zprd*slab_volfactor;
-
- // make initial g_ewald estimate
- // based on desired error and real space cutoff
- // fluid-occupied volume used to estimate real-space error
- // zprd used rather than zprd_slab
-
- double h_x,h_y,h_z;
-
- if (!gewaldflag)
- g_ewald = sqrt(-log(accuracy*sqrt(natoms*cutoff*xprd*yprd*zprd) /
- (2.0*q2))) / cutoff;
-
- // set optimal nx_pppm,ny_pppm,nz_pppm based on order and precision
- // nz_pppm uses extended zprd_slab instead of zprd
- // h = 1/g_ewald is upper bound on h such that h*g_ewald <= 1
- // reduce it until precision target is met
-
- if (!gridflag) {
- double err;
- h_x = h_y = h_z = 1/g_ewald;
-
- nx_pppm = static_cast<int> (xprd/h_x + 1);
- ny_pppm = static_cast<int> (yprd/h_y + 1);
- nz_pppm = static_cast<int> (zprd_slab/h_z + 1);
-
- err = rms(h_x,xprd,natoms,q2,acons);
- while (err > accuracy) {
- err = rms(h_x,xprd,natoms,q2,acons);
- nx_pppm++;
- h_x = xprd/nx_pppm;
- }
-
- err = rms(h_y,yprd,natoms,q2,acons);
- while (err > accuracy) {
- err = rms(h_y,yprd,natoms,q2,acons);
- ny_pppm++;
- h_y = yprd/ny_pppm;
- }
-
- err = rms(h_z,zprd_slab,natoms,q2,acons);
- while (err > accuracy) {
- err = rms(h_z,zprd_slab,natoms,q2,acons);
- nz_pppm++;
- h_z = zprd_slab/nz_pppm;
- }
- }
-
- // boost grid size until it is factorable
-
- while (!factorable(nx_pppm)) nx_pppm++;
- while (!factorable(ny_pppm)) ny_pppm++;
- while (!factorable(nz_pppm)) nz_pppm++;
-
-
- // adjust g_ewald for new grid size
-
- h_x = xprd/nx_pppm;
- h_y = yprd/ny_pppm;
- h_z = zprd_slab/nz_pppm;
-
- if (!gewaldflag) {
- double gew1,gew2,dgew,f,fmid,hmin,rtb;
- int ncount;
-
- gew1 = 0.0;
- g_ewald = gew1;
- f = diffpr(h_x,h_y,h_z,q2,acons);
-
- hmin = MIN(h_x,MIN(h_y,h_z));
- gew2 = 10/hmin;
- g_ewald = gew2;
- fmid = diffpr(h_x,h_y,h_z,q2,acons);
-
- if (f*fmid >= 0.0) error->all(FLERR,"Cannot compute PPPMCuda G");
- rtb = f < 0.0 ? (dgew=gew2-gew1,gew1) : (dgew=gew1-gew2,gew2);
- ncount = 0;
- while (fabs(dgew) > SMALL && fmid != 0.0) {
- dgew *= 0.5;
- g_ewald = rtb + dgew;
- fmid = diffpr(h_x,h_y,h_z,q2,acons);
- if (fmid <= 0.0) rtb = g_ewald;
- ncount++;
- if (ncount > LARGE) error->all(FLERR,"Cannot compute PPPMCuda G");
- }
- }
-
- // final RMS precision
-
- double lprx = rms(h_x,xprd,natoms,q2,acons);
- double lpry = rms(h_y,yprd,natoms,q2,acons);
- double lprz = rms(h_z,zprd_slab,natoms,q2,acons);
- double lpr = sqrt(lprx*lprx + lpry*lpry + lprz*lprz) / sqrt(3.0);
- double spr = 2.0*q2 * exp(-g_ewald*g_ewald*cutoff*cutoff) /
- sqrt(natoms*cutoff*xprd*yprd*zprd_slab);
-
- // free local memory
-
- memory->destroy(acons);
-
- // print info
-
- if (me == 0) {
- if (screen) {
- fprintf(screen," G vector = %g\n",g_ewald);
- fprintf(screen," grid = %d %d %d\n",nx_pppm,ny_pppm,nz_pppm);
- fprintf(screen," stencil order = %d\n",order);
- fprintf(screen," absolute RMS force accuracy = %g\n",MAX(lpr,spr));
- fprintf(screen," relative force accuracy = %g\n",
- MAX(lpr,spr)/two_charge_force);
- }
- if (logfile) {
- fprintf(logfile," G vector = %g\n",g_ewald);
- fprintf(logfile," grid = %d %d %d\n",nx_pppm,ny_pppm,nz_pppm);
- fprintf(logfile," stencil order = %d\n",order);
- fprintf(logfile," absolute RMS force accuracy = %g\n",MAX(lpr,spr));
- fprintf(logfile," relative force accuracy = %g\n",
- MAX(lpr,spr)/two_charge_force);
- }
- }
-}
-
-
-/* ----------------------------------------------------------------------
- find center grid pt for each of my particles
- check that full stencil for the particle will fit in my 3d brick
- store central grid pt indices in part2grid array
-------------------------------------------------------------------------- */
-
-
-void PPPMCuda::particle_map()
-{
- MYDBG(printf("# CUDA PPPMCuda::particle_map() ... start\n");)
- int flag = 0;
-
- cu_flag->memset_device(0);
- flag=cuda_particle_map(&cuda->shared_data,cu_flag->dev_data());
- if(flag)
- {
- cu_debugdata->download();
- printf("Out of range atom: ");
- printf("ID: %i ",atom->tag[int(debugdata[0])]);
- printf("x: %e ",debugdata[7]);
- printf("y: %e ",debugdata[8]);
- printf("z: %e ",debugdata[9]);
- printf("nx: %e ",debugdata[4]);
- printf("ny: %e ",debugdata[5]);
-
- printf("\n");
- //printf("debugdata: cpu: %e %e %e %i\n",boxlo[0],boxlo[1],boxlo[2],atom->nlocal);
- cuda->cu_x->download();
- int nx,ny,nz;
-
- double **x = atom->x;
- int nlocal = atom->nlocal;
- for (int i = 0; i < nlocal; i++) {
- nx = static_cast<int> ((x[i][0]-boxlo[0])*delxinv+shift) - OFFSET;
- ny = static_cast<int> ((x[i][1]-boxlo[1])*delyinv+shift) - OFFSET;
- nz = static_cast<int> ((x[i][2]-boxlo[2])*delzinv+shift) - OFFSET;
-
- if(i==1203)printf("Outside Atom: %i %e %e %e (%i %i %i)\n",i,x[i][0],x[i][1],x[i][2],nx,ny,nz);
- if (nx+nlower < nxlo_out || nx+nupper > nxhi_out ||
- ny+nlower < nylo_out || ny+nupper > nyhi_out ||
- nz+nlower < nzlo_out || nz+nupper > nzhi_out || i==1203) {printf("Outside Atom: %i %e %e %e (%i %i %i)\n",i,x[i][0],x[i][1],x[i][2],nx,ny,nz); }
- }
-
- }
-
- int flag_all;
- MPI_Allreduce(&flag,&flag_all,1,MPI_INT,MPI_SUM,world);
- if (flag_all) error->all(FLERR,"Out of range atoms - cannot compute PPPMCuda!");
-}
-
-/* ----------------------------------------------------------------------
- create discretized "density" on section of global grid due to my particles
- density(x,y,z) = charge "density" at grid points of my 3d brick
- (nxlo:nxhi,nylo:nyhi,nzlo:nzhi) is extent of my brick (including ghosts)
- in global grid
-------------------------------------------------------------------------- */
-
-
-void PPPMCuda::make_rho()
-{
- cuda_make_rho(&cuda->shared_data,cu_flag->dev_data(),&density_intScale,nxhi_out,nxlo_out,nyhi_out,nylo_out,nzhi_out,nzlo_out,cu_density_brick->dev_data(),cu_density_brick_int->dev_data());
-}
-
-
-/* ----------------------------------------------------------------------
- FFT-based Poisson solver
-------------------------------------------------------------------------- */
-void PPPMCuda::poisson(int eflag, int vflag)
-{
-
-#ifndef FFT_CUFFT
- PPPMOld::poisson(eflag,vflag);
- return;
-#endif
-#ifdef FFT_CUFFT
- my_times starttime;
- my_times endtime;
-
-
- my_gettime(CLOCK_REALTIME,&starttime);
- fft1c->compute(density_fft,work1,1);
-
- my_gettime(CLOCK_REALTIME,&endtime);
- poissontime+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
-
-
- if (eflag || vflag) {
- poisson_energy(nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,vflag);
- ENERGY_CFLOAT gpuvirial[6];
- energy+=sum_energy(cu_virial->dev_data(),cu_energy->dev_data(),nx_pppm,ny_pppm,nz_pppm,vflag,gpuvirial);
- if(vflag)
- {
- for(int j=0;j<6;j++) virial[j]+=gpuvirial[j];
- }
- }
-
-
- // scale by 1/total-grid-pts to get rho(k)
- // multiply by Green's function to get V(k)
-
- poisson_scale(nx_pppm,ny_pppm,nz_pppm);
-
- // compute gradients of V(r) in each of 3 dims by transformimg -ik*V(k)
- // FFT leaves data in 3d brick decomposition
- // copy it into inner portion of vdx,vdy,vdz arrays
-
- // x direction gradient
-
-
- poisson_xgrad(nx_pppm,ny_pppm,nz_pppm);
-
-
- my_gettime(CLOCK_REALTIME,&starttime);
- fft2c->compute(work2,work2,-1);
- my_gettime(CLOCK_REALTIME,&endtime);
- poissontime+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- poisson_vdx_brick(nxhi_out,nxlo_out,nyhi_out,nylo_out,nzhi_out,nzlo_out,nx_pppm,ny_pppm,nz_pppm);
-
-
- // y direction gradient
-
- poisson_ygrad(nx_pppm,ny_pppm,nz_pppm);
-
- my_gettime(CLOCK_REALTIME,&starttime);
- fft2c->compute(work2,work2,-1);
- my_gettime(CLOCK_REALTIME,&endtime);
- poissontime+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- poisson_vdy_brick(nxhi_out,nxlo_out,nyhi_out,nylo_out,nzhi_out,nzlo_out,nx_pppm,ny_pppm,nz_pppm);
-
- // z direction gradient
-
- poisson_zgrad(nx_pppm,ny_pppm,nz_pppm);
-
- my_gettime(CLOCK_REALTIME,&starttime);
- fft2c->compute(work2,work2,-1);
- my_gettime(CLOCK_REALTIME,&endtime);
- poissontime+=(endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000);
-
- poisson_vdz_brick(nxhi_out,nxlo_out,nyhi_out,nylo_out,nzhi_out,nzlo_out,nx_pppm,ny_pppm,nz_pppm);
- #endif
-}
-
-/*----------------------------------------------------------------------
- interpolate from grid to get electric field & force on my particles
--------------------------------------------------------------------------*/
-
-void PPPMCuda::fieldforce()
-{
- cuda_fieldforce(& cuda->shared_data,cu_flag);
- return;
-}
-
-/* ----------------------------------------------------------------------
- perform and time the 4 FFTs required for N timesteps
-------------------------------------------------------------------------- */
-
-int PPPMCuda::timing_1d(int n, double &time1d)
-{
- time1d = cuda->shared_data.cuda_timings.pppm_poisson/update->nsteps/4*n;
- return 4;
-}
-
-int PPPMCuda::timing_3d(int n, double &time3d)
-{
- time3d = cuda->shared_data.cuda_timings.pppm_poisson/update->nsteps*n;
- return 4;
-}
-
-void PPPMCuda::slabcorr(int eflag)
-{
- // compute local contribution to global dipole moment
- if(slabbuf==NULL)
- {
- slabbuf=new ENERGY_CFLOAT[(atom->nmax+31)/32];
- cu_slabbuf = new cCudaData<ENERGY_CFLOAT,ENERGY_CFLOAT, x> (slabbuf, (atom->nmax+31)/32);
- }
- if(unsigned((atom->nlocal+31)/32)*sizeof(ENERGY_CFLOAT)>=unsigned(cu_slabbuf->dev_size()))
- {
- delete [] slabbuf;
- delete cu_slabbuf;
- slabbuf=new ENERGY_CFLOAT[(atom->nmax+31)/32];
- cu_slabbuf = new cCudaData<ENERGY_CFLOAT,ENERGY_CFLOAT, x> (slabbuf, (atom->nmax+31)/32);
- }
-
-
- double dipole = cuda_slabcorr_energy(&cuda->shared_data,slabbuf,(ENERGY_CFLOAT*) cu_slabbuf->dev_data());
-
- double dipole_all;
- MPI_Allreduce(&dipole,&dipole_all,1,MPI_DOUBLE,MPI_SUM,world);
-
- //if (eflag) energy += qqrd2e*scale * e_slabcorr;
- // need to add a correction to make non-neutral systems and per-atom energy translationally invariant
- if (eflag || fabs(qsum) > SMALL)
- error->all(FLERR,"Cannot (yet) use slab correction with kspace_style pppm/cuda for non-neutral systems or to get per-atom energy. Aborting.");
-
- double ffact = -4.0*MY_PI*dipole_all/volume;
-
- cuda_slabcorr_force(&cuda->shared_data,ffact);
-}
diff --git a/src/USER-CUDA/pppm_cuda.h b/src/USER-CUDA/pppm_cuda.h
deleted file mode 100644
index cd22aa1d5d..0000000000
--- a/src/USER-CUDA/pppm_cuda.h
+++ /dev/null
@@ -1,113 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef KSPACE_CLASS
-
-KSpaceStyle(pppm/cuda,PPPMCuda)
-
-#else
-
-#ifndef LMP_PPPM_CUDA_H
-#define LMP_PPPM_CUDA_H
-
-#include "pppm_old.h"
-#include "cuda_data.h"
-#include "cuda_precision.h"
-
-namespace LAMMPS_NS {
-
-class PPPMCuda : public PPPMOld {
- public:
- PPPMCuda(class LAMMPS *, int, char **);
- ~PPPMCuda();
- void init();
- void setup();
- void compute(int, int);
- int timing_1d(int, double &);
- int timing_3d(int, double &);
-
- double poissontime;
-
- protected:
- class Cuda *cuda;
- class FFT3dCuda *fft1c,*fft2c;
- double* work3;
-
- cCudaData<double , FFT_CFLOAT , x >* cu_work1;
- cCudaData<double , FFT_CFLOAT , x >* cu_work2;
- cCudaData<double , FFT_CFLOAT , x >* cu_work3;
- cCudaData<double , PPPM_CFLOAT , x >* cu_greensfn;
- cCudaData<double , PPPM_CFLOAT , x >* cu_gf_b;
- cCudaData<double , PPPM_CFLOAT , x >* cu_fkx;
- cCudaData<double , PPPM_CFLOAT , x >* cu_fky;
- cCudaData<double , PPPM_CFLOAT , x >* cu_fkz;
- cCudaData<double , PPPM_CFLOAT , xy>* cu_vg;
- cCudaData<double , PPPM_CFLOAT , x >* cu_density_brick;
- cCudaData<int , int , x >* cu_density_brick_int;
- cCudaData<double , PPPM_CFLOAT , x >* cu_vdx_brick;
- cCudaData<double , PPPM_CFLOAT , x >* cu_vdy_brick;
- cCudaData<double , PPPM_CFLOAT , x >* cu_vdz_brick;
- cCudaData<double , PPPM_CFLOAT , x >* cu_density_fft;
- cCudaData<double , ENERGY_CFLOAT , x >* cu_energy;
- cCudaData<double , ENERGY_CFLOAT , x >* cu_virial;
- cCudaData<double , X_CFLOAT , yx>* cu_x;
- cCudaData<double , V_CFLOAT , yx>* cu_v;
- cCudaData<double , F_CFLOAT , yx>* cu_f;
- cCudaData<double , F_CFLOAT , yx>* cu_q;
- cCudaData<int , int , yx>* cu_part2grid;
- cCudaData<double , PPPM_CFLOAT , x >* cu_rho_coeff;
- cCudaData<PPPM_CFLOAT , PPPM_CFLOAT , x >* cu_debugdata;
- cCudaData<int , int , x >* cu_flag;
- cCudaData<int , int , x >* cu_pppm_grid_n;
- cCudaData<int , int , x >* cu_pppm_grid_ids;
-
- ENERGY_CFLOAT* slabbuf;
- cCudaData<ENERGY_CFLOAT, ENERGY_CFLOAT, x >* cu_slabbuf;
-
- int*** density_brick_int;
- PPPM_CFLOAT density_intScale;
- int pppm_grid_nmax;
- int* pppm2partgrid;
- int* pppm_grid;
- PPPM_CFLOAT* debugdata;
- bool firstpass;
-
- void set_grid();
- void allocate();
- void deallocate();
-
- virtual void particle_map();
- virtual void make_rho();
- virtual void poisson(int, int);
- virtual void fieldforce();
- virtual void slabcorr(int);
- double*** vdx_brick_tmp;
- int old_nmax;
- int global_flag;
- dev_array* adev_data_array;
-};
-
-}
-
-#endif
-#endif
diff --git a/src/USER-CUDA/pppm_old.cpp b/src/USER-CUDA/pppm_old.cpp
deleted file mode 100755
index 2cc4c18626..0000000000
--- a/src/USER-CUDA/pppm_old.cpp
+++ /dev/null
@@ -1,2839 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- Contributing authors: Roy Pollock (LLNL), Paul Crozier (SNL)
- per-atom energy/virial & group/group energy/force added by Stan Moore (BYU)
-------------------------------------------------------------------------- */
-
-#include <mpi.h>
-#include <string.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <math.h>
-#include "pppm_old.h"
-#include "math_const.h"
-#include "atom.h"
-#include "comm.h"
-#include "neighbor.h"
-#include "force.h"
-#include "pair.h"
-#include "bond.h"
-#include "angle.h"
-#include "domain.h"
-#include "fft3d_wrap.h"
-#include "remap_wrap.h"
-#include "memory.h"
-#include "error.h"
-
-using namespace LAMMPS_NS;
-using namespace MathConst;
-
-#define MAXORDER 7
-#define OFFSET 16384
-#define SMALL 0.00001
-#define LARGE 10000.0
-#define EPS_HOC 1.0e-7
-
-#ifdef FFT_SINGLE
-#define ZEROF 0.0f
-#define ONEF 1.0f
-#else
-#define ZEROF 0.0
-#define ONEF 1.0
-#endif
-
-/* ---------------------------------------------------------------------- */
-
-PPPMOld::PPPMOld(LAMMPS *lmp, int narg, char **arg) : KSpace(lmp, narg, arg)
-{
- if (narg < 1) error->all(FLERR,"Illegal kspace_style pppm command");
-
- triclinic_support = 0;
- pppmflag = 1;
- group_group_enable = 0;
-
- accuracy_relative = fabs(force->numeric(FLERR,arg[0]));
-
- nfactors = 3;
- factors = new int[nfactors];
- factors[0] = 2;
- factors[1] = 3;
- factors[2] = 5;
-
- MPI_Comm_rank(world,&me);
- MPI_Comm_size(world,&nprocs);
-
- density_brick = vdx_brick = vdy_brick = vdz_brick = NULL;
- density_fft = NULL;
- u_brick = NULL;
- v0_brick = v1_brick = v2_brick = v3_brick = v4_brick = v5_brick = NULL;
- greensfn = NULL;
- work1 = work2 = NULL;
- vg = NULL;
- fkx = fky = fkz = NULL;
- buf1 = buf2 = buf3 = buf4 = NULL;
-
- density_A_brick = density_B_brick = NULL;
- density_A_fft = density_B_fft = NULL;
-
- gf_b = NULL;
- rho1d = rho_coeff = NULL;
-
- fft1 = fft2 = NULL;
- remap = NULL;
-
- nmax = 0;
- part2grid = NULL;
-}
-
-/* ----------------------------------------------------------------------
- free all memory
-------------------------------------------------------------------------- */
-
-PPPMOld::~PPPMOld()
-{
- delete [] factors;
- deallocate();
- deallocate_peratom();
- deallocate_groups();
- memory->destroy(part2grid);
-}
-
-/* ----------------------------------------------------------------------
- called once before run
-------------------------------------------------------------------------- */
-
-void PPPMOld::init()
-{
- if (me == 0) {
- if (screen) fprintf(screen,"PPPM initialization ...\n");
- if (logfile) fprintf(logfile,"PPPM initialization ...\n");
- }
-
- // error check
-
- triclinic_check();
- if (domain->dimension == 2) error->all(FLERR,
- "Cannot use PPPM with 2d simulation");
-
- if (!atom->q_flag) error->all(FLERR,"Kspace style requires atom attribute q");
-
- if (slabflag == 0 && domain->nonperiodic > 0)
- error->all(FLERR,"Cannot use nonperiodic boundaries with PPPM");
- if (slabflag) {
- if (domain->xperiodic != 1 || domain->yperiodic != 1 ||
- domain->boundary[2][0] != 1 || domain->boundary[2][1] != 1)
- error->all(FLERR,"Incorrect boundaries with slab PPPM");
- }
-
- if (order < 2 || order > MAXORDER) {
- char str[128];
- sprintf(str,"PPPM order cannot be < 2 or > than %d",MAXORDER);
- error->all(FLERR,str);
- }
-
- // free all arrays previously allocated
-
- deallocate();
- deallocate_peratom();
- peratom_allocate_flag = 0;
- deallocate_groups();
- group_allocate_flag = 0;
-
- // extract short-range Coulombic cutoff from pair style
-
- pair_check();
-
- int itmp=0;
- double *p_cutoff = (double *) force->pair->extract("cut_coul",itmp);
- if (p_cutoff == NULL)
- error->all(FLERR,"KSpace style is incompatible with Pair style");
- cutoff = *p_cutoff;
-
- // if kspace is TIP4P, extract TIP4P params from pair style
- // bond/angle are not yet init(), so insure equilibrium request is valid
-
- qdist = 0.0;
-
- if (tip4pflag) {
- double *p_qdist = (double *) force->pair->extract("qdist",itmp);
- int *p_typeO = (int *) force->pair->extract("typeO",itmp);
- int *p_typeH = (int *) force->pair->extract("typeH",itmp);
- int *p_typeA = (int *) force->pair->extract("typeA",itmp);
- int *p_typeB = (int *) force->pair->extract("typeB",itmp);
- if (!p_qdist || !p_typeO || !p_typeH || !p_typeA || !p_typeB)
- error->all(FLERR,"KSpace style is incompatible with Pair style");
- qdist = *p_qdist;
- typeO = *p_typeO;
- typeH = *p_typeH;
- int typeA = *p_typeA;
- int typeB = *p_typeB;
-
- if (force->angle == NULL || force->bond == NULL)
- error->all(FLERR,"Bond and angle potentials must be defined for TIP4P");
- if (typeA < 1 || typeA > atom->nangletypes ||
- force->angle->setflag[typeA] == 0)
- error->all(FLERR,"Bad TIP4P angle type for PPPM/TIP4P");
- if (typeB < 1 || typeB > atom->nbondtypes ||
- force->bond->setflag[typeB] == 0)
- error->all(FLERR,"Bad TIP4P bond type for PPPM/TIP4P");
- double theta = force->angle->equilibrium_angle(typeA);
- double blen = force->bond->equilibrium_distance(typeB);
- alpha = qdist / (cos(0.5*theta) * blen);
- }
-
- // compute qsum & qsqsum and warn if not charge-neutral
-
- scale = 1.0;
- qqrd2e = force->qqrd2e;
- qsum_qsq();
- natoms_original = atom->natoms;
-
- // set accuracy (force units) from accuracy_relative or accuracy_absolute
-
- if (accuracy_absolute >= 0.0) accuracy = accuracy_absolute;
- else accuracy = accuracy_relative * two_charge_force;
-
- // setup FFT grid resolution and g_ewald
- // normally one iteration thru while loop is all that is required
- // if grid stencil extends beyond neighbor proc, reduce order and try again
-
- int iteration = 0;
-
- while (order > 1) {
- if (iteration && me == 0)
- error->warning(FLERR,"Reducing PPPM order b/c stencil extends "
- "beyond neighbor processor");
- iteration++;
-
- set_grid();
-
- if (nx_pppm >= OFFSET || ny_pppm >= OFFSET || nz_pppm >= OFFSET)
- error->all(FLERR,"PPPM grid is too large");
-
- // global indices of PPPM grid range from 0 to N-1
- // nlo_in,nhi_in = lower/upper limits of the 3d sub-brick of
- // global PPPM grid that I own without ghost cells
- // for slab PPPM, assign z grid as if it were not extended
-
- nxlo_in = static_cast<int> (comm->xsplit[comm->myloc[0]] * nx_pppm);
- nxhi_in = static_cast<int> (comm->xsplit[comm->myloc[0]+1] * nx_pppm) - 1;
-
- nylo_in = static_cast<int> (comm->ysplit[comm->myloc[1]] * ny_pppm);
- nyhi_in = static_cast<int> (comm->ysplit[comm->myloc[1]+1] * ny_pppm) - 1;
-
- nzlo_in = static_cast<int>
- (comm->zsplit[comm->myloc[2]] * nz_pppm/slab_volfactor);
- nzhi_in = static_cast<int>
- (comm->zsplit[comm->myloc[2]+1] * nz_pppm/slab_volfactor) - 1;
-
- // nlower,nupper = stencil size for mapping particles to PPPM grid
-
- nlower = -(order-1)/2;
- nupper = order/2;
-
- // shift values for particle <-> grid mapping
- // add/subtract OFFSET to avoid int(-0.75) = 0 when want it to be -1
-
- if (order % 2) shift = OFFSET + 0.5;
- else shift = OFFSET;
- if (order % 2) shiftone = 0.0;
- else shiftone = 0.5;
-
- // nlo_out,nhi_out = lower/upper limits of the 3d sub-brick of
- // global PPPM grid that my particles can contribute charge to
- // effectively nlo_in,nhi_in + ghost cells
- // nlo,nhi = global coords of grid pt to "lower left" of smallest/largest
- // position a particle in my box can be at
- // dist[3] = particle position bound = subbox + skin/2.0 + qdist
- // qdist = offset due to TIP4P fictitious charge
- // convert to triclinic if necessary
- // nlo_out,nhi_out = nlo,nhi + stencil size for particle mapping
- // for slab PPPM, assign z grid as if it were not extended
-
- triclinic = domain->triclinic;
- double *prd,*sublo,*subhi;
-
- if (triclinic == 0) {
- prd = domain->prd;
- boxlo = domain->boxlo;
- sublo = domain->sublo;
- subhi = domain->subhi;
- } else {
- prd = domain->prd_lamda;
- boxlo = domain->boxlo_lamda;
- sublo = domain->sublo_lamda;
- subhi = domain->subhi_lamda;
- }
-
- double xprd = prd[0];
- double yprd = prd[1];
- double zprd = prd[2];
- double zprd_slab = zprd*slab_volfactor;
-
- double dist[3];
- double cuthalf = 0.5*neighbor->skin + qdist;
- if (triclinic == 0) dist[0] = dist[1] = dist[2] = cuthalf;
- else {
- dist[0] = cuthalf/domain->prd[0];
- dist[1] = cuthalf/domain->prd[1];
- dist[2] = cuthalf/domain->prd[2];
- }
-
- int nlo,nhi;
-
- nlo = static_cast<int> ((sublo[0]-dist[0]-boxlo[0]) *
- nx_pppm/xprd + shift) - OFFSET;
- nhi = static_cast<int> ((subhi[0]+dist[0]-boxlo[0]) *
- nx_pppm/xprd + shift) - OFFSET;
- nxlo_out = nlo + nlower;
- nxhi_out = nhi + nupper;
-
- nlo = static_cast<int> ((sublo[1]-dist[1]-boxlo[1]) *
- ny_pppm/yprd + shift) - OFFSET;
- nhi = static_cast<int> ((subhi[1]+dist[1]-boxlo[1]) *
- ny_pppm/yprd + shift) - OFFSET;
- nylo_out = nlo + nlower;
- nyhi_out = nhi + nupper;
-
- nlo = static_cast<int> ((sublo[2]-dist[2]-boxlo[2]) *
- nz_pppm/zprd_slab + shift) - OFFSET;
- nhi = static_cast<int> ((subhi[2]+dist[2]-boxlo[2]) *
- nz_pppm/zprd_slab + shift) - OFFSET;
- nzlo_out = nlo + nlower;
- nzhi_out = nhi + nupper;
-
- // for slab PPPM, change the grid boundary for processors at +z end
- // to include the empty volume between periodically repeating slabs
- // for slab PPPM, want charge data communicated from -z proc to +z proc,
- // but not vice versa, also want field data communicated from +z proc to
- // -z proc, but not vice versa
- // this is accomplished by nzhi_in = nzhi_out on +z end (no ghost cells)
-
- if (slabflag == 1 && (comm->myloc[2] == comm->procgrid[2]-1)) {
- nzhi_in = nz_pppm - 1;
- nzhi_out = nz_pppm - 1;
- }
-
- // nlo_ghost,nhi_ghost = # of planes I will recv from 6 directions
- // that overlay domain I own
- // proc in that direction tells me via sendrecv()
- // if no neighbor proc, value is from self since I have ghosts regardless
-
- int nplanes;
-
- nplanes = nxlo_in - nxlo_out;
- if (comm->procneigh[0][0] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[0][0],0,
- &nxhi_ghost,1,MPI_INT,comm->procneigh[0][1],0,
- world,MPI_STATUS_IGNORE);
- else nxhi_ghost = nplanes;
-
- nplanes = nxhi_out - nxhi_in;
- if (comm->procneigh[0][1] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[0][1],0,
- &nxlo_ghost,1,MPI_INT,comm->procneigh[0][0],
- 0,world,MPI_STATUS_IGNORE);
- else nxlo_ghost = nplanes;
-
- nplanes = nylo_in - nylo_out;
- if (comm->procneigh[1][0] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[1][0],0,
- &nyhi_ghost,1,MPI_INT,comm->procneigh[1][1],0,
- world,MPI_STATUS_IGNORE);
- else nyhi_ghost = nplanes;
-
- nplanes = nyhi_out - nyhi_in;
- if (comm->procneigh[1][1] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[1][1],0,
- &nylo_ghost,1,MPI_INT,comm->procneigh[1][0],0,
- world,MPI_STATUS_IGNORE);
- else nylo_ghost = nplanes;
-
- nplanes = nzlo_in - nzlo_out;
- if (comm->procneigh[2][0] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[2][0],0,
- &nzhi_ghost,1,MPI_INT,comm->procneigh[2][1],0,
- world,MPI_STATUS_IGNORE);
- else nzhi_ghost = nplanes;
-
- nplanes = nzhi_out - nzhi_in;
- if (comm->procneigh[2][1] != me)
- MPI_Sendrecv(&nplanes,1,MPI_INT,comm->procneigh[2][1],0,
- &nzlo_ghost,1,MPI_INT,comm->procneigh[2][0],0,
- world,MPI_STATUS_IGNORE);
- else nzlo_ghost = nplanes;
-
- // test that ghost overlap is not bigger than my sub-domain
-
- int flag = 0;
- if (nxlo_ghost > nxhi_in-nxlo_in+1) flag = 1;
- if (nxhi_ghost > nxhi_in-nxlo_in+1) flag = 1;
- if (nylo_ghost > nyhi_in-nylo_in+1) flag = 1;
- if (nyhi_ghost > nyhi_in-nylo_in+1) flag = 1;
- if (nzlo_ghost > nzhi_in-nzlo_in+1) flag = 1;
- if (nzhi_ghost > nzhi_in-nzlo_in+1) flag = 1;
-
- int flag_all;
- MPI_Allreduce(&flag,&flag_all,1,MPI_INT,MPI_SUM,world);
-
- if (flag_all == 0) break;
- order--;
- }
-
- if (order == 0) error->all(FLERR,"PPPM order has been reduced to 0");
-
- // decomposition of FFT mesh
- // global indices range from 0 to N-1
- // proc owns entire x-dimension, clump of columns in y,z dimensions
- // npey_fft,npez_fft = # of procs in y,z dims
- // if nprocs is small enough, proc can own 1 or more entire xy planes,
- // else proc owns 2d sub-blocks of yz plane
- // me_y,me_z = which proc (0-npe_fft-1) I am in y,z dimensions
- // nlo_fft,nhi_fft = lower/upper limit of the section
- // of the global FFT mesh that I own
-
- int npey_fft,npez_fft;
- if (nz_pppm >= nprocs) {
- npey_fft = 1;
- npez_fft = nprocs;
- } else procs2grid2d(nprocs,ny_pppm,nz_pppm,&npey_fft,&npez_fft);
-
- int me_y = me % npey_fft;
- int me_z = me / npey_fft;
-
- nxlo_fft = 0;
- nxhi_fft = nx_pppm - 1;
- nylo_fft = me_y*ny_pppm/npey_fft;
- nyhi_fft = (me_y+1)*ny_pppm/npey_fft - 1;
- nzlo_fft = me_z*nz_pppm/npez_fft;
- nzhi_fft = (me_z+1)*nz_pppm/npez_fft - 1;
-
- // PPPM grid for this proc, including ghosts
-
- ngrid = (nxhi_out-nxlo_out+1) * (nyhi_out-nylo_out+1) *
- (nzhi_out-nzlo_out+1);
-
- // FFT arrays on this proc, without ghosts
- // nfft = FFT points in FFT decomposition on this proc
- // nfft_brick = FFT points in 3d brick-decomposition on this proc
- // nfft_both = greater of 2 values
-
- nfft = (nxhi_fft-nxlo_fft+1) * (nyhi_fft-nylo_fft+1) *
- (nzhi_fft-nzlo_fft+1);
- int nfft_brick = (nxhi_in-nxlo_in+1) * (nyhi_in-nylo_in+1) *
- (nzhi_in-nzlo_in+1);
- nfft_both = MAX(nfft,nfft_brick);
-
- // buffer space for use in brick2fft and fillbrick
- // idel = max # of ghost planes to send or recv in +/- dir of each dim
- // nx,ny,nz = owned planes (including ghosts) in each dim
- // nxx,nyy,nzz = max # of grid cells to send in each dim
- // nbuf = max in any dim, augment by 3x for components of vd_xyz in fillbrick
-
- int idelx,idely,idelz,nx,ny,nz,nxx,nyy,nzz;
-
- idelx = MAX(nxlo_ghost,nxhi_ghost);
- idelx = MAX(idelx,nxhi_out-nxhi_in);
- idelx = MAX(idelx,nxlo_in-nxlo_out);
-
- idely = MAX(nylo_ghost,nyhi_ghost);
- idely = MAX(idely,nyhi_out-nyhi_in);
- idely = MAX(idely,nylo_in-nylo_out);
-
- idelz = MAX(nzlo_ghost,nzhi_ghost);
- idelz = MAX(idelz,nzhi_out-nzhi_in);
- idelz = MAX(idelz,nzlo_in-nzlo_out);
-
- nx = nxhi_out - nxlo_out + 1;
- ny = nyhi_out - nylo_out + 1;
- nz = nzhi_out - nzlo_out + 1;
-
- nxx = idelx * ny * nz;
- nyy = idely * nx * nz;
- nzz = idelz * nx * ny;
-
- nbuf = MAX(nxx,nyy);
- nbuf = MAX(nbuf,nzz);
-
- nbuf_peratom = 7*nbuf;
- nbuf *= 3;
-
- // print stats
-
- int ngrid_max,nfft_both_max,nbuf_max;
- MPI_Allreduce(&ngrid,&ngrid_max,1,MPI_INT,MPI_MAX,world);
- MPI_Allreduce(&nfft_both,&nfft_both_max,1,MPI_INT,MPI_MAX,world);
- MPI_Allreduce(&nbuf,&nbuf_max,1,MPI_INT,MPI_MAX,world);
-
- if (me == 0) {
- if (screen) fprintf(screen," brick FFT buffer size/proc = %d %d %d\n",
- ngrid_max,nfft_both_max,nbuf_max);
- if (logfile) fprintf(logfile," brick FFT buffer size/proc = %d %d %d\n",
- ngrid_max,nfft_both_max,nbuf_max);
- }
-
- // allocate K-space dependent memory
- // don't invoke allocate_peratom() here, wait to see if needed
-
- allocate();
-
- // pre-compute Green's function denomiator expansion
- // pre-compute 1d charge distribution coefficients
-
- compute_gf_denom();
- compute_rho_coeff();
-}
-
-/* ----------------------------------------------------------------------
- adjust PPPM coeffs, called initially and whenever volume has changed
-------------------------------------------------------------------------- */
-
-void PPPMOld::setup()
-{
- int i,j,k,l,m,n;
- double *prd;
-
- // volume-dependent factors
- // adjust z dimension for 2d slab PPPM
- // z dimension for 3d PPPM is zprd since slab_volfactor = 1.0
-
- if (triclinic == 0) prd = domain->prd;
- else prd = domain->prd_lamda;
-
- double xprd = prd[0];
- double yprd = prd[1];
- double zprd = prd[2];
- double zprd_slab = zprd*slab_volfactor;
- volume = xprd * yprd * zprd_slab;
-
- delxinv = nx_pppm/xprd;
- delyinv = ny_pppm/yprd;
- delzinv = nz_pppm/zprd_slab;
-
- delvolinv = delxinv*delyinv*delzinv;
-
- double unitkx = (2.0*MY_PI/xprd);
- double unitky = (2.0*MY_PI/yprd);
- double unitkz = (2.0*MY_PI/zprd_slab);
-
- // fkx,fky,fkz for my FFT grid pts
-
- double per;
-
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- per = i - nx_pppm*(2*i/nx_pppm);
- fkx[i] = unitkx*per;
- }
-
- for (i = nylo_fft; i <= nyhi_fft; i++) {
- per = i - ny_pppm*(2*i/ny_pppm);
- fky[i] = unitky*per;
- }
-
- for (i = nzlo_fft; i <= nzhi_fft; i++) {
- per = i - nz_pppm*(2*i/nz_pppm);
- fkz[i] = unitkz*per;
- }
-
- // virial coefficients
-
- double sqk,vterm;
-
- n = 0;
- for (k = nzlo_fft; k <= nzhi_fft; k++) {
- for (j = nylo_fft; j <= nyhi_fft; j++) {
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- sqk = fkx[i]*fkx[i] + fky[j]*fky[j] + fkz[k]*fkz[k];
- if (sqk == 0.0) {
- vg[n][0] = 0.0;
- vg[n][1] = 0.0;
- vg[n][2] = 0.0;
- vg[n][3] = 0.0;
- vg[n][4] = 0.0;
- vg[n][5] = 0.0;
- } else {
- vterm = -2.0 * (1.0/sqk + 0.25/(g_ewald*g_ewald));
- vg[n][0] = 1.0 + vterm*fkx[i]*fkx[i];
- vg[n][1] = 1.0 + vterm*fky[j]*fky[j];
- vg[n][2] = 1.0 + vterm*fkz[k]*fkz[k];
- vg[n][3] = vterm*fkx[i]*fky[j];
- vg[n][4] = vterm*fkx[i]*fkz[k];
- vg[n][5] = vterm*fky[j]*fkz[k];
- }
- n++;
- }
- }
- }
-
- // modified (Hockney-Eastwood) Coulomb Green's function
-
- int nx,ny,nz,kper,lper,mper;
- double snx,sny,snz,snx2,sny2,snz2;
- double argx,argy,argz,wx,wy,wz,sx,sy,sz,qx,qy,qz;
- double sum1,dot1,dot2;
- double numerator,denominator;
-
- int nbx = static_cast<int> ((g_ewald*xprd/(MY_PI*nx_pppm)) *
- pow(-log(EPS_HOC),0.25));
- int nby = static_cast<int> ((g_ewald*yprd/(MY_PI*ny_pppm)) *
- pow(-log(EPS_HOC),0.25));
- int nbz = static_cast<int> ((g_ewald*zprd_slab/(MY_PI*nz_pppm)) *
- pow(-log(EPS_HOC),0.25));
-
- double form = 1.0;
-
- n = 0;
- for (m = nzlo_fft; m <= nzhi_fft; m++) {
- mper = m - nz_pppm*(2*m/nz_pppm);
- snz = sin(0.5*unitkz*mper*zprd_slab/nz_pppm);
- snz2 = snz*snz;
-
- for (l = nylo_fft; l <= nyhi_fft; l++) {
- lper = l - ny_pppm*(2*l/ny_pppm);
- sny = sin(0.5*unitky*lper*yprd/ny_pppm);
- sny2 = sny*sny;
-
- for (k = nxlo_fft; k <= nxhi_fft; k++) {
- kper = k - nx_pppm*(2*k/nx_pppm);
- snx = sin(0.5*unitkx*kper*xprd/nx_pppm);
- snx2 = snx*snx;
-
- sqk = pow(unitkx*kper,2.0) + pow(unitky*lper,2.0) +
- pow(unitkz*mper,2.0);
-
- if (sqk != 0.0) {
- numerator = form*12.5663706/sqk;
- denominator = gf_denom(snx2,sny2,snz2);
- sum1 = 0.0;
- const double dorder = static_cast<double>(order);
- for (nx = -nbx; nx <= nbx; nx++) {
- qx = unitkx*(kper+nx_pppm*nx);
- sx = exp(-0.25*pow(qx/g_ewald,2.0));
- wx = 1.0;
- argx = 0.5*qx*xprd/nx_pppm;
- if (argx != 0.0) wx = pow(sin(argx)/argx,dorder);
- for (ny = -nby; ny <= nby; ny++) {
- qy = unitky*(lper+ny_pppm*ny);
- sy = exp(-0.25*pow(qy/g_ewald,2.0));
- wy = 1.0;
- argy = 0.5*qy*yprd/ny_pppm;
- if (argy != 0.0) wy = pow(sin(argy)/argy,dorder);
- for (nz = -nbz; nz <= nbz; nz++) {
- qz = unitkz*(mper+nz_pppm*nz);
- sz = exp(-0.25*pow(qz/g_ewald,2.0));
- wz = 1.0;
- argz = 0.5*qz*zprd_slab/nz_pppm;
- if (argz != 0.0) wz = pow(sin(argz)/argz,dorder);
-
- dot1 = unitkx*kper*qx + unitky*lper*qy + unitkz*mper*qz;
- dot2 = qx*qx+qy*qy+qz*qz;
- sum1 += (dot1/dot2) * sx*sy*sz * pow(wx*wy*wz,2.0);
- }
- }
- }
- greensfn[n++] = numerator*sum1/denominator;
- } else greensfn[n++] = 0.0;
- }
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- compute the PPPM long-range force, energy, virial
-------------------------------------------------------------------------- */
-
-void PPPMOld::compute(int eflag, int vflag)
-{
- int i,j;
-
- // set energy/virial flags
- // invoke allocate_peratom() if needed for first time
-
- if (eflag || vflag) ev_setup(eflag,vflag);
- else evflag = evflag_atom = eflag_global = vflag_global =
- eflag_atom = vflag_atom = 0;
-
- if (evflag_atom && !peratom_allocate_flag) {
- allocate_peratom();
- peratom_allocate_flag = 1;
- }
-
- // convert atoms from box to lamda coords
-
- if (triclinic == 0) boxlo = domain->boxlo;
- else {
- boxlo = domain->boxlo_lamda;
- domain->x2lamda(atom->nlocal);
- }
-
- // extend size of per-atom arrays if necessary
-
- if (atom->nlocal > nmax) {
- memory->destroy(part2grid);
- nmax = atom->nmax;
- memory->create(part2grid,nmax,3,"pppm:part2grid");
- }
-
- // find grid points for all my particles
- // map my particle charge onto my local 3d density grid
-
- particle_map();
- make_rho();
-
- // all procs communicate density values from their ghost cells
- // to fully sum contribution in their 3d bricks
- // remap from 3d decomposition to FFT decomposition
-
- brick2fft();
-
- // compute potential gradient on my FFT grid and
- // portion of e_long on this proc's FFT grid
- // return gradients (electric fields) in 3d brick decomposition
- // also performs per-atom calculations via poisson_peratom()
-
- poisson(eflag,vflag);
-
- // all procs communicate E-field values
- // to fill ghost cells surrounding their 3d bricks
-
- fillbrick();
-
- // extra per-atom energy/virial communication
-
- if (evflag_atom) fillbrick_peratom();
-
- // calculate the force on my particles
-
- fieldforce();
-
- // extra per-atom energy/virial communication
-
- if (evflag_atom) fieldforce_peratom();
-
- // update qsum and qsqsum, if atom count has changed and energy needed
-
- if ((eflag_global || eflag_atom) && atom->natoms != natoms_original) {
- qsum_qsq();
- natoms_original = atom->natoms;
- }
-
- // sum global energy across procs and add in volume-dependent term
-
- const double qscale = qqrd2e * scale;
-
- if (eflag_global) {
- double energy_all;
- MPI_Allreduce(&energy,&energy_all,1,MPI_DOUBLE,MPI_SUM,world);
- energy = energy_all;
-
- energy *= 0.5*volume;
- energy -= g_ewald*qsqsum/MY_PIS +
- MY_PI2*qsum*qsum / (g_ewald*g_ewald*volume);
- energy *= qscale;
- }
-
- // sum global virial across procs
-
- if (vflag_global) {
- double virial_all[6];
- MPI_Allreduce(virial,virial_all,6,MPI_DOUBLE,MPI_SUM,world);
- for (i = 0; i < 6; i++) virial[i] = 0.5*qscale*volume*virial_all[i];
- }
-
- // per-atom energy/virial
- // energy includes self-energy correction
-
- if (evflag_atom) {
- double *q = atom->q;
- int nlocal = atom->nlocal;
-
- if (eflag_atom) {
- for (i = 0; i < nlocal; i++) {
- eatom[i] *= 0.5;
- eatom[i] -= g_ewald*q[i]*q[i]/MY_PIS + MY_PI2*q[i]*qsum /
- (g_ewald*g_ewald*volume);
- eatom[i] *= qscale;
- }
- }
-
- if (vflag_atom) {
- for (i = 0; i < nlocal; i++)
- for (j = 0; j < 6; j++) vatom[i][j] *= 0.5*q[i]*qscale;
- }
- }
-
- // 2d slab correction
-
- if (slabflag == 1) slabcorr();
-
- // convert atoms back from lamda to box coords
-
- if (triclinic) domain->lamda2x(atom->nlocal);
-}
-
-/* ----------------------------------------------------------------------
- allocate memory that depends on # of K-vectors and order
-------------------------------------------------------------------------- */
-
-void PPPMOld::allocate()
-{
- memory->create3d_offset(density_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:density_brick");
- memory->create3d_offset(vdx_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:vdx_brick");
- memory->create3d_offset(vdy_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:vdy_brick");
- memory->create3d_offset(vdz_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:vdz_brick");
-
- memory->create(density_fft,nfft_both,"pppm:density_fft");
- memory->create(greensfn,nfft_both,"pppm:greensfn");
- memory->create(work1,2*nfft_both,"pppm:work1");
- memory->create(work2,2*nfft_both,"pppm:work2");
- memory->create(vg,nfft_both,6,"pppm:vg");
-
- memory->create1d_offset(fkx,nxlo_fft,nxhi_fft,"pppm:fkx");
- memory->create1d_offset(fky,nylo_fft,nyhi_fft,"pppm:fky");
- memory->create1d_offset(fkz,nzlo_fft,nzhi_fft,"pppm:fkz");
-
- memory->create(buf1,nbuf,"pppm:buf1");
- memory->create(buf2,nbuf,"pppm:buf2");
-
- // summation coeffs
-
- memory->create(gf_b,order,"pppm:gf_b");
- memory->create2d_offset(rho1d,3,-order/2,order/2,"pppm:rho1d");
- memory->create2d_offset(rho_coeff,order,(1-order)/2,order/2,"pppm:rho_coeff");
-
- // create 2 FFTs and a Remap
- // 1st FFT keeps data in FFT decompostion
- // 2nd FFT returns data in 3d brick decomposition
- // remap takes data from 3d brick to FFT decomposition
-
- int tmp;
-
- fft1 = new FFT3d(lmp,world,nx_pppm,ny_pppm,nz_pppm,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- 0,0,&tmp,collective_flag);
-
- fft2 = new FFT3d(lmp,world,nx_pppm,ny_pppm,nz_pppm,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- nxlo_in,nxhi_in,nylo_in,nyhi_in,nzlo_in,nzhi_in,
- 0,0,&tmp,collective_flag);
-
- remap = new Remap(lmp,world,
- nxlo_in,nxhi_in,nylo_in,nyhi_in,nzlo_in,nzhi_in,
- nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
- 1,0,0,FFT_PRECISION,collective_flag);
-}
-
-/* ----------------------------------------------------------------------
- allocate per-atom memory that depends on # of K-vectors and order
-------------------------------------------------------------------------- */
-
-void PPPMOld::allocate_peratom()
-{
- memory->create3d_offset(u_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:u_brick");
-
- memory->create3d_offset(v0_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:v0_brick");
- memory->create3d_offset(v1_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:v1_brick");
- memory->create3d_offset(v2_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:v2_brick");
- memory->create3d_offset(v3_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:v3_brick");
- memory->create3d_offset(v4_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:v4_brick");
- memory->create3d_offset(v5_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:v5_brick");
-
- memory->create(buf3,nbuf_peratom,"pppm:buf3");
- memory->create(buf4,nbuf_peratom,"pppm:buf4");
-}
-
-/* ----------------------------------------------------------------------
- deallocate memory that depends on # of K-vectors and order
-------------------------------------------------------------------------- */
-
-void PPPMOld::deallocate()
-{
- memory->destroy3d_offset(density_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(vdx_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(vdy_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(vdz_brick,nzlo_out,nylo_out,nxlo_out);
-
- memory->destroy(density_fft);
- memory->destroy(greensfn);
- memory->destroy(work1);
- memory->destroy(work2);
- memory->destroy(vg);
-
- memory->destroy1d_offset(fkx,nxlo_fft);
- memory->destroy1d_offset(fky,nylo_fft);
- memory->destroy1d_offset(fkz,nzlo_fft);
-
- memory->destroy(buf1);
- memory->destroy(buf2);
-
- memory->destroy(gf_b);
- memory->destroy2d_offset(rho1d,-order/2);
- memory->destroy2d_offset(rho_coeff,(1-order)/2);
-
- delete fft1;
- delete fft2;
- delete remap;
-}
-
-/* ----------------------------------------------------------------------
- deallocate per-atom memory that depends on # of K-vectors and order
-------------------------------------------------------------------------- */
-
-void PPPMOld::deallocate_peratom()
-{
- memory->destroy3d_offset(u_brick,nzlo_out,nylo_out,nxlo_out);
-
- memory->destroy3d_offset(v0_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(v1_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(v2_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(v3_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(v4_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(v5_brick,nzlo_out,nylo_out,nxlo_out);
-
- memory->destroy(buf3);
- memory->destroy(buf4);
-}
-
-/* ----------------------------------------------------------------------
- set size of FFT grid (nx,ny,nz_pppm) and g_ewald
-------------------------------------------------------------------------- */
-
-void PPPMOld::set_grid()
-{
- // see JCP 109, pg 7698 for derivation of coefficients
- // higher order coefficients may be computed if needed
-
- double **acons;
- memory->create(acons,8,7,"pppm:acons");
-
- acons[1][0] = 2.0 / 3.0;
- acons[2][0] = 1.0 / 50.0;
- acons[2][1] = 5.0 / 294.0;
- acons[3][0] = 1.0 / 588.0;
- acons[3][1] = 7.0 / 1440.0;
- acons[3][2] = 21.0 / 3872.0;
- acons[4][0] = 1.0 / 4320.0;
- acons[4][1] = 3.0 / 1936.0;
- acons[4][2] = 7601.0 / 2271360.0;
- acons[4][3] = 143.0 / 28800.0;
- acons[5][0] = 1.0 / 23232.0;
- acons[5][1] = 7601.0 / 13628160.0;
- acons[5][2] = 143.0 / 69120.0;
- acons[5][3] = 517231.0 / 106536960.0;
- acons[5][4] = 106640677.0 / 11737571328.0;
- acons[6][0] = 691.0 / 68140800.0;
- acons[6][1] = 13.0 / 57600.0;
- acons[6][2] = 47021.0 / 35512320.0;
- acons[6][3] = 9694607.0 / 2095994880.0;
- acons[6][4] = 733191589.0 / 59609088000.0;
- acons[6][5] = 326190917.0 / 11700633600.0;
- acons[7][0] = 1.0 / 345600.0;
- acons[7][1] = 3617.0 / 35512320.0;
- acons[7][2] = 745739.0 / 838397952.0;
- acons[7][3] = 56399353.0 / 12773376000.0;
- acons[7][4] = 25091609.0 / 1560084480.0;
- acons[7][5] = 1755948832039.0 / 36229939200000.0;
- acons[7][6] = 4887769399.0 / 37838389248.0;
-
- // use xprd,yprd,zprd even if triclinic so grid size is the same
- // adjust z dimension for 2d slab PPPM
- // 3d PPPM just uses zprd since slab_volfactor = 1.0
-
- double xprd = domain->xprd;
- double yprd = domain->yprd;
- double zprd = domain->zprd;
- double zprd_slab = zprd*slab_volfactor;
-
- // make initial g_ewald estimate
- // based on desired accuracy and real space cutoff
- // fluid-occupied volume used to estimate real-space error
- // zprd used rather than zprd_slab
-
- double h_x,h_y,h_z;
- bigint natoms = atom->natoms;
-
- if (!gewaldflag) {
- if (accuracy <= 0.0)
- error->all(FLERR,"KSpace accuracy must be > 0");
- g_ewald = accuracy*sqrt(natoms*cutoff*xprd*yprd*zprd) / (2.0*q2);
- if (g_ewald >= 1.0) g_ewald = (1.35 - 0.15*log(accuracy))/cutoff;
- else g_ewald = sqrt(-log(g_ewald)) / cutoff;
- }
-
- // set optimal nx_pppm,ny_pppm,nz_pppm based on order and accuracy
- // nz_pppm uses extended zprd_slab instead of zprd
- // h = 1/g_ewald is upper bound on h such that h*g_ewald <= 1
- // reduce it until accuracy target is met
-
- if (!gridflag) {
- double err;
- h_x = h_y = h_z = 1.0/g_ewald;
-
- nx_pppm = static_cast<int> (xprd/h_x) + 1;
- ny_pppm = static_cast<int> (yprd/h_y) + 1;
- nz_pppm = static_cast<int> (zprd_slab/h_z) + 1;
-
- err = rms(h_x,xprd,natoms,q2,acons);
- while (err > accuracy) {
- err = rms(h_x,xprd,natoms,q2,acons);
- nx_pppm++;
- h_x = xprd/nx_pppm;
- }
-
- err = rms(h_y,yprd,natoms,q2,acons);
- while (err > accuracy) {
- err = rms(h_y,yprd,natoms,q2,acons);
- ny_pppm++;
- h_y = yprd/ny_pppm;
- }
-
- err = rms(h_z,zprd_slab,natoms,q2,acons);
- while (err > accuracy) {
- err = rms(h_z,zprd_slab,natoms,q2,acons);
- nz_pppm++;
- h_z = zprd_slab/nz_pppm;
- }
- }
-
- // boost grid size until it is factorable
-
- while (!factorable(nx_pppm)) nx_pppm++;
- while (!factorable(ny_pppm)) ny_pppm++;
- while (!factorable(nz_pppm)) nz_pppm++;
-
- // adjust g_ewald for new grid size
-
- h_x = xprd/static_cast<double>(nx_pppm);
- h_y = yprd/static_cast<double>(ny_pppm);
- h_z = zprd_slab/static_cast<double>(nz_pppm);
-
- if (!gewaldflag) {
- double gew1,gew2,dgew,f,fmid,hmin,rtb;
- int ncount;
-
- gew1 = 0.0;
- g_ewald = gew1;
- f = diffpr(h_x,h_y,h_z,q2,acons);
-
- hmin = MIN(h_x,MIN(h_y,h_z));
- gew2 = 10.0/hmin;
- g_ewald = gew2;
- fmid = diffpr(h_x,h_y,h_z,q2,acons);
-
- if (f*fmid >= 0.0) error->all(FLERR,"Cannot compute PPPM G");
- rtb = f < 0.0 ? (dgew=gew2-gew1,gew1) : (dgew=gew1-gew2,gew2);
- ncount = 0;
- while (fabs(dgew) > SMALL && fmid != 0.0) {
- dgew *= 0.5;
- g_ewald = rtb + dgew;
- fmid = diffpr(h_x,h_y,h_z,q2,acons);
- if (fmid <= 0.0) rtb = g_ewald;
- ncount++;
- if (ncount > LARGE) error->all(FLERR,"Cannot compute PPPM G");
- }
- }
-
- // final RMS accuracy
-
- double lprx = rms(h_x,xprd,natoms,q2,acons);
- double lpry = rms(h_y,yprd,natoms,q2,acons);
- double lprz = rms(h_z,zprd_slab,natoms,q2,acons);
- double lpr = sqrt(lprx*lprx + lpry*lpry + lprz*lprz) / sqrt(3.0);
- double q2_over_sqrt = q2 / sqrt(natoms*cutoff*xprd*yprd*zprd_slab);
- double spr = 2.0 *q2_over_sqrt * exp(-g_ewald*g_ewald*cutoff*cutoff);
- double tpr = estimate_table_accuracy(q2_over_sqrt,spr);
- double accuracy = sqrt(lpr*lpr + spr*spr + tpr*tpr);
-
- // free local memory
-
- memory->destroy(acons);
-
- // print info
-
- if (me == 0) {
-#ifdef FFT_SINGLE
- const char fft_prec[] = "single";
-#else
- const char fft_prec[] = "double";
-#endif
- if (screen) {
- fprintf(screen," G vector (1/distance)= %g\n",g_ewald);
- fprintf(screen," grid = %d %d %d\n",nx_pppm,ny_pppm,nz_pppm);
- fprintf(screen," stencil order = %d\n",order);
- fprintf(screen," estimated absolute RMS force accuracy = %g\n",
- accuracy);
- fprintf(screen," estimated relative force accuracy = %g\n",
- accuracy/two_charge_force);
- fprintf(screen," using %s precision FFTs\n",fft_prec);
- }
- if (logfile) {
- fprintf(logfile," G vector (1/distance) = %g\n",g_ewald);
- fprintf(logfile," grid = %d %d %d\n",nx_pppm,ny_pppm,nz_pppm);
- fprintf(logfile," stencil order = %d\n",order);
- fprintf(logfile," estimated absolute RMS force accuracy = %g\n",
- accuracy);
- fprintf(logfile," estimated relative force accuracy = %g\n",
- accuracy/two_charge_force);
- fprintf(logfile," using %s precision FFTs\n",fft_prec);
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- check if all factors of n are in list of factors
- return 1 if yes, 0 if no
-------------------------------------------------------------------------- */
-
-int PPPMOld::factorable(int n)
-{
- int i;
-
- while (n > 1) {
- for (i = 0; i < nfactors; i++) {
- if (n % factors[i] == 0) {
- n /= factors[i];
- break;
- }
- }
- if (i == nfactors) return 0;
- }
-
- return 1;
-}
-
-/* ----------------------------------------------------------------------
- compute RMS accuracy for a dimension
-------------------------------------------------------------------------- */
-
-double PPPMOld::rms(double h, double prd, bigint natoms,
- double q2, double **acons)
-{
- double sum = 0.0;
- for (int m = 0; m < order; m++)
- sum += acons[order][m] * pow(h*g_ewald,2.0*m);
- double value = q2 * pow(h*g_ewald,(double)order) *
- sqrt(g_ewald*prd*sqrt(2.0*MY_PI)*sum/natoms) / (prd*prd);
- return value;
-}
-
-/* ----------------------------------------------------------------------
- compute difference in real-space and KSpace RMS accuracy
-------------------------------------------------------------------------- */
-
-double PPPMOld::diffpr(double h_x, double h_y, double h_z, double q2,
- double **acons)
-{
- double lprx,lpry,lprz,kspace_prec,real_prec;
- double xprd = domain->xprd;
- double yprd = domain->yprd;
- double zprd = domain->zprd;
- bigint natoms = atom->natoms;
-
- lprx = rms(h_x,xprd,natoms,q2,acons);
- lpry = rms(h_y,yprd,natoms,q2,acons);
- lprz = rms(h_z,zprd*slab_volfactor,natoms,q2,acons);
- kspace_prec = sqrt(lprx*lprx + lpry*lpry + lprz*lprz) / sqrt(3.0);
- real_prec = 2.0*q2 * exp(-g_ewald*g_ewald*cutoff*cutoff) /
- sqrt(static_cast<double>(natoms)*cutoff*xprd*yprd*zprd);
- double value = kspace_prec - real_prec;
- return value;
-}
-
-/* ----------------------------------------------------------------------
- pre-compute Green's function denominator expansion coeffs, Gamma(2n)
-------------------------------------------------------------------------- */
-
-void PPPMOld::compute_gf_denom()
-{
- int k,l,m;
-
- for (l = 1; l < order; l++) gf_b[l] = 0.0;
- gf_b[0] = 1.0;
-
- for (m = 1; m < order; m++) {
- for (l = m; l > 0; l--)
- gf_b[l] = 4.0 * (gf_b[l]*(l-m)*(l-m-0.5)-gf_b[l-1]*(l-m-1)*(l-m-1));
- gf_b[0] = 4.0 * (gf_b[0]*(l-m)*(l-m-0.5));
- }
-
- bigint ifact = 1;
- for (k = 1; k < 2*order; k++) ifact *= k;
- double gaminv = 1.0/ifact;
- for (l = 0; l < order; l++) gf_b[l] *= gaminv;
-}
-
-/* ----------------------------------------------------------------------
- ghost-swap to accumulate full density in brick decomposition
- remap density from 3d brick decomposition to FFT decomposition
-------------------------------------------------------------------------- */
-
-void PPPMOld::brick2fft()
-{
- int i,n,ix,iy,iz;
- MPI_Request request;
-
- // pack my ghosts for +x processor
- // pass data to self or +x processor
- // unpack and sum recv data into my real cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxhi_in+1; ix <= nxhi_out; ix++)
- buf1[n++] = density_brick[iz][iy][ix];
-
- if (comm->procneigh[0][1] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[0][0],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[0][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxlo_in; ix < nxlo_in+nxlo_ghost; ix++)
- density_brick[iz][iy][ix] += buf2[n++];
-
- // pack my ghosts for -x processor
- // pass data to self or -x processor
- // unpack and sum recv data into my real cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxlo_out; ix < nxlo_in; ix++)
- buf1[n++] = density_brick[iz][iy][ix];
-
- if (comm->procneigh[0][0] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[0][1],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[0][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxhi_in-nxhi_ghost+1; ix <= nxhi_in; ix++)
- density_brick[iz][iy][ix] += buf2[n++];
-
- // pack my ghosts for +y processor
- // pass data to self or +y processor
- // unpack and sum recv data into my real cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nyhi_in+1; iy <= nyhi_out; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- buf1[n++] = density_brick[iz][iy][ix];
-
- if (comm->procneigh[1][1] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[1][0],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[1][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_in; iy < nylo_in+nylo_ghost; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- density_brick[iz][iy][ix] += buf2[n++];
-
- // pack my ghosts for -y processor
- // pass data to self or -y processor
- // unpack and sum recv data into my real cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy < nylo_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- buf1[n++] = density_brick[iz][iy][ix];
-
- if (comm->procneigh[1][0] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[1][1],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[1][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nyhi_in-nyhi_ghost+1; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- density_brick[iz][iy][ix] += buf2[n++];
-
- // pack my ghosts for +z processor
- // pass data to self or +z processor
- // unpack and sum recv data into my real cells
-
- n = 0;
- for (iz = nzhi_in+1; iz <= nzhi_out; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- buf1[n++] = density_brick[iz][iy][ix];
-
- if (comm->procneigh[2][1] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[2][0],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[2][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_in; iz < nzlo_in+nzlo_ghost; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- density_brick[iz][iy][ix] += buf2[n++];
-
- // pack my ghosts for -z processor
- // pass data to self or -z processor
- // unpack and sum recv data into my real cells
-
- n = 0;
- for (iz = nzlo_out; iz < nzlo_in; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- buf1[n++] = density_brick[iz][iy][ix];
-
- if (comm->procneigh[2][0] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[2][1],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[2][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzhi_in-nzhi_ghost+1; iz <= nzhi_in; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- density_brick[iz][iy][ix] += buf2[n++];
-
- // remap from 3d brick decomposition to FFT decomposition
- // copy grabs inner portion of density from 3d brick
- // remap could be done as pre-stage of FFT,
- // but this works optimally on only double values, not complex values
-
- n = 0;
- for (iz = nzlo_in; iz <= nzhi_in; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++)
- density_fft[n++] = density_brick[iz][iy][ix];
-
- remap->perform(density_fft,density_fft,work1);
-}
-
-/* ----------------------------------------------------------------------
- ghost-swap to fill ghost cells of my brick with field values
-------------------------------------------------------------------------- */
-
-void PPPMOld::fillbrick()
-{
- int i,n,ix,iy,iz;
- MPI_Request request;
-
- // pack my real cells for +z processor
- // pass data to self or +z processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzhi_in-nzhi_ghost+1; iz <= nzhi_in; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- buf1[n++] = vdx_brick[iz][iy][ix];
- buf1[n++] = vdy_brick[iz][iy][ix];
- buf1[n++] = vdz_brick[iz][iy][ix];
- }
-
- if (comm->procneigh[2][1] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[2][0],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[2][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz < nzlo_in; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- vdx_brick[iz][iy][ix] = buf2[n++];
- vdy_brick[iz][iy][ix] = buf2[n++];
- vdz_brick[iz][iy][ix] = buf2[n++];
- }
-
- // pack my real cells for -z processor
- // pass data to self or -z processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_in; iz < nzlo_in+nzlo_ghost; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- buf1[n++] = vdx_brick[iz][iy][ix];
- buf1[n++] = vdy_brick[iz][iy][ix];
- buf1[n++] = vdz_brick[iz][iy][ix];
- }
-
- if (comm->procneigh[2][0] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[2][1],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[2][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzhi_in+1; iz <= nzhi_out; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- vdx_brick[iz][iy][ix] = buf2[n++];
- vdy_brick[iz][iy][ix] = buf2[n++];
- vdz_brick[iz][iy][ix] = buf2[n++];
- }
-
- // pack my real cells for +y processor
- // pass data to self or +y processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nyhi_in-nyhi_ghost+1; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- buf1[n++] = vdx_brick[iz][iy][ix];
- buf1[n++] = vdy_brick[iz][iy][ix];
- buf1[n++] = vdz_brick[iz][iy][ix];
- }
-
- if (comm->procneigh[1][1] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[1][0],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[1][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy < nylo_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- vdx_brick[iz][iy][ix] = buf2[n++];
- vdy_brick[iz][iy][ix] = buf2[n++];
- vdz_brick[iz][iy][ix] = buf2[n++];
- }
-
- // pack my real cells for -y processor
- // pass data to self or -y processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_in; iy < nylo_in+nylo_ghost; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- buf1[n++] = vdx_brick[iz][iy][ix];
- buf1[n++] = vdy_brick[iz][iy][ix];
- buf1[n++] = vdz_brick[iz][iy][ix];
- }
-
- if (comm->procneigh[1][0] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[1][1],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[1][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nyhi_in+1; iy <= nyhi_out; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- vdx_brick[iz][iy][ix] = buf2[n++];
- vdy_brick[iz][iy][ix] = buf2[n++];
- vdz_brick[iz][iy][ix] = buf2[n++];
- }
-
- // pack my real cells for +x processor
- // pass data to self or +x processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxhi_in-nxhi_ghost+1; ix <= nxhi_in; ix++) {
- buf1[n++] = vdx_brick[iz][iy][ix];
- buf1[n++] = vdy_brick[iz][iy][ix];
- buf1[n++] = vdz_brick[iz][iy][ix];
- }
-
- if (comm->procneigh[0][1] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[0][0],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[0][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxlo_out; ix < nxlo_in; ix++) {
- vdx_brick[iz][iy][ix] = buf2[n++];
- vdy_brick[iz][iy][ix] = buf2[n++];
- vdz_brick[iz][iy][ix] = buf2[n++];
- }
-
- // pack my real cells for -x processor
- // pass data to self or -x processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxlo_in; ix < nxlo_in+nxlo_ghost; ix++) {
- buf1[n++] = vdx_brick[iz][iy][ix];
- buf1[n++] = vdy_brick[iz][iy][ix];
- buf1[n++] = vdz_brick[iz][iy][ix];
- }
-
- if (comm->procneigh[0][0] == me)
- for (i = 0; i < n; i++) buf2[i] = buf1[i];
- else {
- MPI_Irecv(buf2,nbuf,MPI_FFT_SCALAR,comm->procneigh[0][1],0,world,&request);
- MPI_Send(buf1,n,MPI_FFT_SCALAR,comm->procneigh[0][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxhi_in+1; ix <= nxhi_out; ix++) {
- vdx_brick[iz][iy][ix] = buf2[n++];
- vdy_brick[iz][iy][ix] = buf2[n++];
- vdz_brick[iz][iy][ix] = buf2[n++];
- }
-}
-
-/* ----------------------------------------------------------------------
- ghost-swap to fill ghost cells of my brick with per-atom field values
-------------------------------------------------------------------------- */
-
-void PPPMOld::fillbrick_peratom()
-{
- int i,n,ix,iy,iz;
- MPI_Request request;
-
- // pack my real cells for +z processor
- // pass data to self or +z processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzhi_in-nzhi_ghost+1; iz <= nzhi_in; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) buf3[n++] = u_brick[iz][iy][ix];
- if (vflag_atom) {
- buf3[n++] = v0_brick[iz][iy][ix];
- buf3[n++] = v1_brick[iz][iy][ix];
- buf3[n++] = v2_brick[iz][iy][ix];
- buf3[n++] = v3_brick[iz][iy][ix];
- buf3[n++] = v4_brick[iz][iy][ix];
- buf3[n++] = v5_brick[iz][iy][ix];
- }
- }
-
- if (comm->procneigh[2][1] == me)
- for (i = 0; i < n; i++) buf4[i] = buf3[i];
- else {
- MPI_Irecv(buf4,nbuf_peratom,MPI_FFT_SCALAR,
- comm->procneigh[2][0],0,world,&request);
- MPI_Send(buf3,n,MPI_FFT_SCALAR,comm->procneigh[2][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz < nzlo_in; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) u_brick[iz][iy][ix] = buf4[n++];
- if (vflag_atom) {
- v0_brick[iz][iy][ix] = buf4[n++];
- v1_brick[iz][iy][ix] = buf4[n++];
- v2_brick[iz][iy][ix] = buf4[n++];
- v3_brick[iz][iy][ix] = buf4[n++];
- v4_brick[iz][iy][ix] = buf4[n++];
- v5_brick[iz][iy][ix] = buf4[n++];
- }
- }
-
- // pack my real cells for -z processor
- // pass data to self or -z processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_in; iz < nzlo_in+nzlo_ghost; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) buf3[n++] = u_brick[iz][iy][ix];
- if (vflag_atom) {
- buf3[n++] = v0_brick[iz][iy][ix];
- buf3[n++] = v1_brick[iz][iy][ix];
- buf3[n++] = v2_brick[iz][iy][ix];
- buf3[n++] = v3_brick[iz][iy][ix];
- buf3[n++] = v4_brick[iz][iy][ix];
- buf3[n++] = v5_brick[iz][iy][ix];
- }
- }
-
- if (comm->procneigh[2][0] == me)
- for (i = 0; i < n; i++) buf4[i] = buf3[i];
- else {
- MPI_Irecv(buf4,nbuf_peratom,MPI_FFT_SCALAR,
- comm->procneigh[2][1],0,world,&request);
- MPI_Send(buf3,n,MPI_FFT_SCALAR,comm->procneigh[2][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzhi_in+1; iz <= nzhi_out; iz++)
- for (iy = nylo_in; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) u_brick[iz][iy][ix] = buf4[n++];
- if (vflag_atom) {
- v0_brick[iz][iy][ix] = buf4[n++];
- v1_brick[iz][iy][ix] = buf4[n++];
- v2_brick[iz][iy][ix] = buf4[n++];
- v3_brick[iz][iy][ix] = buf4[n++];
- v4_brick[iz][iy][ix] = buf4[n++];
- v5_brick[iz][iy][ix] = buf4[n++];
- }
- }
-
- // pack my real cells for +y processor
- // pass data to self or +y processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nyhi_in-nyhi_ghost+1; iy <= nyhi_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) buf3[n++] = u_brick[iz][iy][ix];
- if (vflag_atom) {
- buf3[n++] = v0_brick[iz][iy][ix];
- buf3[n++] = v1_brick[iz][iy][ix];
- buf3[n++] = v2_brick[iz][iy][ix];
- buf3[n++] = v3_brick[iz][iy][ix];
- buf3[n++] = v4_brick[iz][iy][ix];
- buf3[n++] = v5_brick[iz][iy][ix];
- }
- }
-
- if (comm->procneigh[1][1] == me)
- for (i = 0; i < n; i++) buf4[i] = buf3[i];
- else {
- MPI_Irecv(buf4,nbuf_peratom,MPI_FFT_SCALAR,
- comm->procneigh[1][0],0,world,&request);
- MPI_Send(buf3,n,MPI_FFT_SCALAR,comm->procneigh[1][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy < nylo_in; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) u_brick[iz][iy][ix] = buf4[n++];
- if (vflag_atom) {
- v0_brick[iz][iy][ix] = buf4[n++];
- v1_brick[iz][iy][ix] = buf4[n++];
- v2_brick[iz][iy][ix] = buf4[n++];
- v3_brick[iz][iy][ix] = buf4[n++];
- v4_brick[iz][iy][ix] = buf4[n++];
- v5_brick[iz][iy][ix] = buf4[n++];
- }
- }
-
- // pack my real cells for -y processor
- // pass data to self or -y processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_in; iy < nylo_in+nylo_ghost; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) buf3[n++] = u_brick[iz][iy][ix];
- if (vflag_atom) {
- buf3[n++] = v0_brick[iz][iy][ix];
- buf3[n++] = v1_brick[iz][iy][ix];
- buf3[n++] = v2_brick[iz][iy][ix];
- buf3[n++] = v3_brick[iz][iy][ix];
- buf3[n++] = v4_brick[iz][iy][ix];
- buf3[n++] = v5_brick[iz][iy][ix];
- }
- }
-
- if (comm->procneigh[1][0] == me)
- for (i = 0; i < n; i++) buf4[i] = buf3[i];
- else {
- MPI_Irecv(buf4,nbuf_peratom,MPI_FFT_SCALAR,
- comm->procneigh[1][1],0,world,&request);
- MPI_Send(buf3,n,MPI_FFT_SCALAR,comm->procneigh[1][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nyhi_in+1; iy <= nyhi_out; iy++)
- for (ix = nxlo_in; ix <= nxhi_in; ix++) {
- if (eflag_atom) u_brick[iz][iy][ix] = buf4[n++];
- if (vflag_atom) {
- v0_brick[iz][iy][ix] = buf4[n++];
- v1_brick[iz][iy][ix] = buf4[n++];
- v2_brick[iz][iy][ix] = buf4[n++];
- v3_brick[iz][iy][ix] = buf4[n++];
- v4_brick[iz][iy][ix] = buf4[n++];
- v5_brick[iz][iy][ix] = buf4[n++];
- }
- }
-
- // pack my real cells for +x processor
- // pass data to self or +x processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxhi_in-nxhi_ghost+1; ix <= nxhi_in; ix++) {
- if (eflag_atom) buf3[n++] = u_brick[iz][iy][ix];
- if (vflag_atom) {
- buf3[n++] = v0_brick[iz][iy][ix];
- buf3[n++] = v1_brick[iz][iy][ix];
- buf3[n++] = v2_brick[iz][iy][ix];
- buf3[n++] = v3_brick[iz][iy][ix];
- buf3[n++] = v4_brick[iz][iy][ix];
- buf3[n++] = v5_brick[iz][iy][ix];
- }
- }
-
- if (comm->procneigh[0][1] == me)
- for (i = 0; i < n; i++) buf4[i] = buf3[i];
- else {
- MPI_Irecv(buf4,nbuf_peratom,MPI_FFT_SCALAR,
- comm->procneigh[0][0],0,world,&request);
- MPI_Send(buf3,n,MPI_FFT_SCALAR,comm->procneigh[0][1],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxlo_out; ix < nxlo_in; ix++) {
- if (eflag_atom) u_brick[iz][iy][ix] = buf4[n++];
- if (vflag_atom) {
- v0_brick[iz][iy][ix] = buf4[n++];
- v1_brick[iz][iy][ix] = buf4[n++];
- v2_brick[iz][iy][ix] = buf4[n++];
- v3_brick[iz][iy][ix] = buf4[n++];
- v4_brick[iz][iy][ix] = buf4[n++];
- v5_brick[iz][iy][ix] = buf4[n++];
- }
- }
-
- // pack my real cells for -x processor
- // pass data to self or -x processor
- // unpack and sum recv data into my ghost cells
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxlo_in; ix < nxlo_in+nxlo_ghost; ix++) {
- if (eflag_atom) buf3[n++] = u_brick[iz][iy][ix];
- if (vflag_atom) {
- buf3[n++] = v0_brick[iz][iy][ix];
- buf3[n++] = v1_brick[iz][iy][ix];
- buf3[n++] = v2_brick[iz][iy][ix];
- buf3[n++] = v3_brick[iz][iy][ix];
- buf3[n++] = v4_brick[iz][iy][ix];
- buf3[n++] = v5_brick[iz][iy][ix];
- }
- }
-
- if (comm->procneigh[0][0] == me)
- for (i = 0; i < n; i++) buf4[i] = buf3[i];
- else {
- MPI_Irecv(buf4,nbuf_peratom,MPI_FFT_SCALAR,
- comm->procneigh[0][1],0,world,&request);
- MPI_Send(buf3,n,MPI_FFT_SCALAR,comm->procneigh[0][0],0,world);
- MPI_Wait(&request,MPI_STATUS_IGNORE);
- }
-
- n = 0;
- for (iz = nzlo_out; iz <= nzhi_out; iz++)
- for (iy = nylo_out; iy <= nyhi_out; iy++)
- for (ix = nxhi_in+1; ix <= nxhi_out; ix++) {
- if (eflag_atom) u_brick[iz][iy][ix] = buf4[n++];
- if (vflag_atom) {
- v0_brick[iz][iy][ix] = buf4[n++];
- v1_brick[iz][iy][ix] = buf4[n++];
- v2_brick[iz][iy][ix] = buf4[n++];
- v3_brick[iz][iy][ix] = buf4[n++];
- v4_brick[iz][iy][ix] = buf4[n++];
- v5_brick[iz][iy][ix] = buf4[n++];
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- find center grid pt for each of my particles
- check that full stencil for the particle will fit in my 3d brick
- store central grid pt indices in part2grid array
-------------------------------------------------------------------------- */
-
-void PPPMOld::particle_map()
-{
- int nx,ny,nz;
-
- double **x = atom->x;
- int nlocal = atom->nlocal;
-
- int flag = 0;
- for (int i = 0; i < nlocal; i++) {
-
- // (nx,ny,nz) = global coords of grid pt to "lower left" of charge
- // current particle coord can be outside global and local box
- // add/subtract OFFSET to avoid int(-0.75) = 0 when want it to be -1
-
- nx = static_cast<int> ((x[i][0]-boxlo[0])*delxinv+shift) - OFFSET;
- ny = static_cast<int> ((x[i][1]-boxlo[1])*delyinv+shift) - OFFSET;
- nz = static_cast<int> ((x[i][2]-boxlo[2])*delzinv+shift) - OFFSET;
-
- part2grid[i][0] = nx;
- part2grid[i][1] = ny;
- part2grid[i][2] = nz;
-
- // check that entire stencil around nx,ny,nz will fit in my 3d brick
-
- if (nx+nlower < nxlo_out || nx+nupper > nxhi_out ||
- ny+nlower < nylo_out || ny+nupper > nyhi_out ||
- nz+nlower < nzlo_out || nz+nupper > nzhi_out)
- flag = 1;
- }
-
- if (flag) error->one(FLERR,"Out of range atoms - cannot compute PPPM");
-}
-
-/* ----------------------------------------------------------------------
- create discretized "density" on section of global grid due to my particles
- density(x,y,z) = charge "density" at grid points of my 3d brick
- (nxlo:nxhi,nylo:nyhi,nzlo:nzhi) is extent of my brick (including ghosts)
- in global grid
-------------------------------------------------------------------------- */
-
-void PPPMOld::make_rho()
-{
- int l,m,n,nx,ny,nz,mx,my,mz;
- FFT_SCALAR dx,dy,dz,x0,y0,z0;
-
- // clear 3d density array
-
- memset(&(density_brick[nzlo_out][nylo_out][nxlo_out]),0,
- ngrid*sizeof(FFT_SCALAR));
-
- // loop over my charges, add their contribution to nearby grid points
- // (nx,ny,nz) = global coords of grid pt to "lower left" of charge
- // (dx,dy,dz) = distance to "lower left" grid pt
- // (mx,my,mz) = global coords of moving stencil pt
-
- double *q = atom->q;
- double **x = atom->x;
- int nlocal = atom->nlocal;
-
- for (int i = 0; i < nlocal; i++) {
-
- nx = part2grid[i][0];
- ny = part2grid[i][1];
- nz = part2grid[i][2];
- dx = nx+shiftone - (x[i][0]-boxlo[0])*delxinv;
- dy = ny+shiftone - (x[i][1]-boxlo[1])*delyinv;
- dz = nz+shiftone - (x[i][2]-boxlo[2])*delzinv;
-
- compute_rho1d(dx,dy,dz);
-
- z0 = delvolinv * q[i];
- for (n = nlower; n <= nupper; n++) {
- mz = n+nz;
- y0 = z0*rho1d[2][n];
- for (m = nlower; m <= nupper; m++) {
- my = m+ny;
- x0 = y0*rho1d[1][m];
- for (l = nlower; l <= nupper; l++) {
- mx = l+nx;
- density_brick[mz][my][mx] += x0*rho1d[0][l];
- }
- }
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- FFT-based Poisson solver
-------------------------------------------------------------------------- */
-
-void PPPMOld::poisson(int,int)
-{
- int i,j,k,n;
- double eng;
-
- // transform charge density (r -> k)
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work1[n++] = density_fft[i];
- work1[n++] = ZEROF;
- }
-
- fft1->compute(work1,work1,1);
-
- // global energy and virial contribution
-
- double scaleinv = 1.0/(nx_pppm*ny_pppm*nz_pppm);
- double s2 = scaleinv*scaleinv;
-
- if (eflag_global || vflag_global) {
- if (vflag_global) {
- n = 0;
- for (i = 0; i < nfft; i++) {
- eng = s2 * greensfn[i] * (work1[n]*work1[n] + work1[n+1]*work1[n+1]);
- for (j = 0; j < 6; j++) virial[j] += eng*vg[i][j];
- if (eflag_global) energy += eng;
- n += 2;
- }
- } else {
- n = 0;
- for (i = 0; i < nfft; i++) {
- energy +=
- s2 * greensfn[i] * (work1[n]*work1[n] + work1[n+1]*work1[n+1]);
- n += 2;
- }
- }
- }
-
- // scale by 1/total-grid-pts to get rho(k)
- // multiply by Green's function to get V(k)
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work1[n++] *= scaleinv * greensfn[i];
- work1[n++] *= scaleinv * greensfn[i];
- }
-
- // extra FFTs for per-atom energy/virial
-
- if (evflag_atom) poisson_peratom();
-
- // compute gradients of V(r) in each of 3 dims by transformimg -ik*V(k)
- // FFT leaves data in 3d brick decomposition
- // copy it into inner portion of vdx,vdy,vdz arrays
-
- // x direction gradient
-
- n = 0;
- for (k = nzlo_fft; k <= nzhi_fft; k++)
- for (j = nylo_fft; j <= nyhi_fft; j++)
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- work2[n] = fkx[i]*work1[n+1];
- work2[n+1] = -fkx[i]*work1[n];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- vdx_brick[k][j][i] = work2[n];
- n += 2;
- }
-
- // y direction gradient
-
- n = 0;
- for (k = nzlo_fft; k <= nzhi_fft; k++)
- for (j = nylo_fft; j <= nyhi_fft; j++)
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- work2[n] = fky[j]*work1[n+1];
- work2[n+1] = -fky[j]*work1[n];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- vdy_brick[k][j][i] = work2[n];
- n += 2;
- }
-
- // z direction gradient
-
- n = 0;
- for (k = nzlo_fft; k <= nzhi_fft; k++)
- for (j = nylo_fft; j <= nyhi_fft; j++)
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- work2[n] = fkz[k]*work1[n+1];
- work2[n+1] = -fkz[k]*work1[n];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- vdz_brick[k][j][i] = work2[n];
- n += 2;
- }
-}
-
-/* ----------------------------------------------------------------------
- FFT-based Poisson solver for per-atom energy/virial
-------------------------------------------------------------------------- */
-
-void PPPMOld::poisson_peratom()
-{
- int i,j,k,n;
-
- // energy
-
- if (eflag_atom) {
- n = 0;
- for (i = 0; i < nfft; i++) {
- work2[n] = work1[n];
- work2[n+1] = work1[n+1];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- u_brick[k][j][i] = work2[n];
- n += 2;
- }
- }
-
- // 6 components of virial in v0 thru v5
-
- if (!vflag_atom) return;
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work2[n] = work1[n]*vg[i][0];
- work2[n+1] = work1[n+1]*vg[i][0];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- v0_brick[k][j][i] = work2[n];
- n += 2;
- }
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work2[n] = work1[n]*vg[i][1];
- work2[n+1] = work1[n+1]*vg[i][1];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- v1_brick[k][j][i] = work2[n];
- n += 2;
- }
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work2[n] = work1[n]*vg[i][2];
- work2[n+1] = work1[n+1]*vg[i][2];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- v2_brick[k][j][i] = work2[n];
- n += 2;
- }
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work2[n] = work1[n]*vg[i][3];
- work2[n+1] = work1[n+1]*vg[i][3];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- v3_brick[k][j][i] = work2[n];
- n += 2;
- }
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work2[n] = work1[n]*vg[i][4];
- work2[n+1] = work1[n+1]*vg[i][4];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- v4_brick[k][j][i] = work2[n];
- n += 2;
- }
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work2[n] = work1[n]*vg[i][5];
- work2[n+1] = work1[n+1]*vg[i][5];
- n += 2;
- }
-
- fft2->compute(work2,work2,-1);
-
- n = 0;
- for (k = nzlo_in; k <= nzhi_in; k++)
- for (j = nylo_in; j <= nyhi_in; j++)
- for (i = nxlo_in; i <= nxhi_in; i++) {
- v5_brick[k][j][i] = work2[n];
- n += 2;
- }
-}
-
-/* ----------------------------------------------------------------------
- interpolate from grid to get electric field & force on my particles
-------------------------------------------------------------------------- */
-
-void PPPMOld::fieldforce()
-{
- int i,l,m,n,nx,ny,nz,mx,my,mz;
- FFT_SCALAR dx,dy,dz,x0,y0,z0;
- FFT_SCALAR ekx,eky,ekz;
-
- // loop over my charges, interpolate electric field from nearby grid points
- // (nx,ny,nz) = global coords of grid pt to "lower left" of charge
- // (dx,dy,dz) = distance to "lower left" grid pt
- // (mx,my,mz) = global coords of moving stencil pt
- // ek = 3 components of E-field on particle
-
- double *q = atom->q;
- double **x = atom->x;
- double **f = atom->f;
-
- int nlocal = atom->nlocal;
-
- for (i = 0; i < nlocal; i++) {
- nx = part2grid[i][0];
- ny = part2grid[i][1];
- nz = part2grid[i][2];
- dx = nx+shiftone - (x[i][0]-boxlo[0])*delxinv;
- dy = ny+shiftone - (x[i][1]-boxlo[1])*delyinv;
- dz = nz+shiftone - (x[i][2]-boxlo[2])*delzinv;
-
- compute_rho1d(dx,dy,dz);
-
- ekx = eky = ekz = ZEROF;
- for (n = nlower; n <= nupper; n++) {
- mz = n+nz;
- z0 = rho1d[2][n];
- for (m = nlower; m <= nupper; m++) {
- my = m+ny;
- y0 = z0*rho1d[1][m];
- for (l = nlower; l <= nupper; l++) {
- mx = l+nx;
- x0 = y0*rho1d[0][l];
- ekx -= x0*vdx_brick[mz][my][mx];
- eky -= x0*vdy_brick[mz][my][mx];
- ekz -= x0*vdz_brick[mz][my][mx];
- }
- }
- }
-
- // convert E-field to force
-
- const double qfactor = qqrd2e * scale * q[i];
- f[i][0] += qfactor*ekx;
- f[i][1] += qfactor*eky;
- if (slabflag != 2) f[i][2] += qfactor*ekz;
- }
-}
-
-/* ----------------------------------------------------------------------
- interpolate from grid to get per-atom energy/virial
-------------------------------------------------------------------------- */
-
-void PPPMOld::fieldforce_peratom()
-{
- int i,l,m,n,nx,ny,nz,mx,my,mz;
- FFT_SCALAR dx,dy,dz,x0,y0,z0;
- FFT_SCALAR u,v0,v1,v2,v3,v4,v5;
-
- // loop over my charges, interpolate from nearby grid points
- // (nx,ny,nz) = global coords of grid pt to "lower left" of charge
- // (dx,dy,dz) = distance to "lower left" grid pt
- // (mx,my,mz) = global coords of moving stencil pt
-
- double *q = atom->q;
- double **x = atom->x;
-
- int nlocal = atom->nlocal;
-
- for (i = 0; i < nlocal; i++) {
- nx = part2grid[i][0];
- ny = part2grid[i][1];
- nz = part2grid[i][2];
- dx = nx+shiftone - (x[i][0]-boxlo[0])*delxinv;
- dy = ny+shiftone - (x[i][1]-boxlo[1])*delyinv;
- dz = nz+shiftone - (x[i][2]-boxlo[2])*delzinv;
-
- compute_rho1d(dx,dy,dz);
-
- u = v0 = v1 = v2 = v3 = v4 = v5 = ZEROF;
- for (n = nlower; n <= nupper; n++) {
- mz = n+nz;
- z0 = rho1d[2][n];
- for (m = nlower; m <= nupper; m++) {
- my = m+ny;
- y0 = z0*rho1d[1][m];
- for (l = nlower; l <= nupper; l++) {
- mx = l+nx;
- x0 = y0*rho1d[0][l];
- if (eflag_atom) u += x0*u_brick[mz][my][mx];
- if (vflag_atom) {
- v0 += x0*v0_brick[mz][my][mx];
- v1 += x0*v1_brick[mz][my][mx];
- v2 += x0*v2_brick[mz][my][mx];
- v3 += x0*v3_brick[mz][my][mx];
- v4 += x0*v4_brick[mz][my][mx];
- v5 += x0*v5_brick[mz][my][mx];
- }
- }
- }
- }
-
- if (eflag_atom) eatom[i] += q[i]*u;
- if (vflag_atom) {
- vatom[i][0] += v0;
- vatom[i][1] += v1;
- vatom[i][2] += v2;
- vatom[i][3] += v3;
- vatom[i][4] += v4;
- vatom[i][5] += v5;
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- map nprocs to NX by NY grid as PX by PY procs - return optimal px,py
-------------------------------------------------------------------------- */
-
-void PPPMOld::procs2grid2d(int nprocs, int nx, int ny, int *px, int *py)
-{
- // loop thru all possible factorizations of nprocs
- // surf = surface area of largest proc sub-domain
- // innermost if test minimizes surface area and surface/volume ratio
-
- int bestsurf = 2 * (nx + ny);
- int bestboxx = 0;
- int bestboxy = 0;
-
- int boxx,boxy,surf,ipx,ipy;
-
- ipx = 1;
- while (ipx <= nprocs) {
- if (nprocs % ipx == 0) {
- ipy = nprocs/ipx;
- boxx = nx/ipx;
- if (nx % ipx) boxx++;
- boxy = ny/ipy;
- if (ny % ipy) boxy++;
- surf = boxx + boxy;
- if (surf < bestsurf ||
- (surf == bestsurf && boxx*boxy > bestboxx*bestboxy)) {
- bestsurf = surf;
- bestboxx = boxx;
- bestboxy = boxy;
- *px = ipx;
- *py = ipy;
- }
- }
- ipx++;
- }
-}
-
-/* ----------------------------------------------------------------------
- charge assignment into rho1d
- dx,dy,dz = distance of particle from "lower left" grid point
-------------------------------------------------------------------------- */
-
-void PPPMOld::compute_rho1d(const FFT_SCALAR &dx, const FFT_SCALAR &dy,
- const FFT_SCALAR &dz)
-{
- int k,l;
- FFT_SCALAR r1,r2,r3;
-
- for (k = (1-order)/2; k <= order/2; k++) {
- r1 = r2 = r3 = ZEROF;
-
- for (l = order-1; l >= 0; l--) {
- r1 = rho_coeff[l][k] + r1*dx;
- r2 = rho_coeff[l][k] + r2*dy;
- r3 = rho_coeff[l][k] + r3*dz;
- }
- rho1d[0][k] = r1;
- rho1d[1][k] = r2;
- rho1d[2][k] = r3;
- }
-}
-
-/* ----------------------------------------------------------------------
- generate coeffients for the weight function of order n
-
- (n-1)
- Wn(x) = Sum wn(k,x) , Sum is over every other integer
- k=-(n-1)
- For k=-(n-1),-(n-1)+2, ....., (n-1)-2,n-1
- k is odd integers if n is even and even integers if n is odd
- ---
- | n-1
- | Sum a(l,j)*(x-k/2)**l if abs(x-k/2) < 1/2
- wn(k,x) = < l=0
- |
- | 0 otherwise
- ---
- a coeffients are packed into the array rho_coeff to eliminate zeros
- rho_coeff(l,((k+mod(n+1,2))/2) = a(l,k)
-------------------------------------------------------------------------- */
-
-void PPPMOld::compute_rho_coeff()
-{
- int j,k,l,m;
- FFT_SCALAR s;
-
- FFT_SCALAR **a;
- memory->create2d_offset(a,order,-order,order,"pppm:a");
-
- for (k = -order; k <= order; k++)
- for (l = 0; l < order; l++)
- a[l][k] = 0.0;
-
- a[0][0] = 1.0;
- for (j = 1; j < order; j++) {
- for (k = -j; k <= j; k += 2) {
- s = 0.0;
- for (l = 0; l < j; l++) {
- a[l+1][k] = (a[l][k+1]-a[l][k-1]) / (l+1);
-#ifdef FFT_SINGLE
- s += powf(0.5,(float) l+1) *
- (a[l][k-1] + powf(-1.0,(float) l) * a[l][k+1]) / (l+1);
-#else
- s += pow(0.5,(double) l+1) *
- (a[l][k-1] + pow(-1.0,(double) l) * a[l][k+1]) / (l+1);
-#endif
- }
- a[0][k] = s;
- }
- }
-
- m = (1-order)/2;
- for (k = -(order-1); k < order; k += 2) {
- for (l = 0; l < order; l++)
- rho_coeff[l][m] = a[l][k];
- m++;
- }
-
- memory->destroy2d_offset(a,-order);
-}
-
-/* ----------------------------------------------------------------------
- Slab-geometry correction term to dampen inter-slab interactions between
- periodically repeating slabs. Yields good approximation to 2D Ewald if
- adequate empty space is left between repeating slabs (J. Chem. Phys.
- 111, 3155). Slabs defined here to be parallel to the xy plane. Also
- extended to non-neutral systems (J. Chem. Phys. 131, 094107).
-------------------------------------------------------------------------- */
-
-void PPPMOld::slabcorr()
-{
- // compute local contribution to global dipole moment
-
- double *q = atom->q;
- double **x = atom->x;
- double zprd = domain->zprd;
- int nlocal = atom->nlocal;
-
- double dipole = 0.0;
- for (int i = 0; i < nlocal; i++) dipole += q[i]*x[i][2];
-
- // sum local contributions to get global dipole moment
-
- double dipole_all;
- MPI_Allreduce(&dipole,&dipole_all,1,MPI_DOUBLE,MPI_SUM,world);
-
- // need to make non-neutral systems and/or
- // per-atom energy translationally invariant
-
- double dipole_r2 = 0.0;
- if (eflag_atom || fabs(qsum) > SMALL) {
- for (int i = 0; i < nlocal; i++)
- dipole_r2 += q[i]*x[i][2]*x[i][2];
-
- // sum local contributions
-
- double tmp;
- MPI_Allreduce(&dipole_r2,&tmp,1,MPI_DOUBLE,MPI_SUM,world);
- dipole_r2 = tmp;
- }
-
- // compute corrections
-
- const double e_slabcorr = MY_2PI*(dipole_all*dipole_all -
- qsum*dipole_r2 - qsum*qsum*zprd*zprd/12.0)/volume;
- const double qscale = qqrd2e * scale;
-
- if (eflag_global) energy += qscale * e_slabcorr;
-
- // per-atom energy
-
- if (eflag_atom) {
- double efact = qscale * MY_2PI/volume;
- for (int i = 0; i < nlocal; i++)
- eatom[i] += efact * q[i]*(x[i][2]*dipole_all - 0.5*(dipole_r2 +
- qsum*x[i][2]*x[i][2]) - qsum*zprd*zprd/12.0);
- }
-
- // add on force corrections
-
- double ffact = qscale * (-4.0*MY_PI/volume);
- double **f = atom->f;
-
- for (int i = 0; i < nlocal; i++) f[i][2] += ffact * q[i]*(dipole_all - qsum*x[i][2]);
-}
-
-
-/* ----------------------------------------------------------------------
- perform and time the 1d FFTs required for N timesteps
-------------------------------------------------------------------------- */
-
-int PPPMOld::timing_1d(int n, double &time1d)
-{
- double time1,time2;
-
- for (int i = 0; i < 2*nfft_both; i++) work1[i] = ZEROF;
-
- MPI_Barrier(world);
- time1 = MPI_Wtime();
-
- for (int i = 0; i < n; i++) {
- fft1->timing1d(work1,nfft_both,1);
- fft2->timing1d(work1,nfft_both,-1);
- fft2->timing1d(work1,nfft_both,-1);
- fft2->timing1d(work1,nfft_both,-1);
- }
-
- MPI_Barrier(world);
- time2 = MPI_Wtime();
- time1d = time2 - time1;
-
- return 4;
-}
-
-/* ----------------------------------------------------------------------
- perform and time the 3d FFTs required for N timesteps
-------------------------------------------------------------------------- */
-
-int PPPMOld::timing_3d(int n, double &time3d)
-{
- double time1,time2;
-
- for (int i = 0; i < 2*nfft_both; i++) work1[i] = ZEROF;
-
- MPI_Barrier(world);
- time1 = MPI_Wtime();
-
- for (int i = 0; i < n; i++) {
- fft1->compute(work1,work1,1);
- fft2->compute(work1,work1,-1);
- fft2->compute(work1,work1,-1);
- fft2->compute(work1,work1,-1);
- }
-
- MPI_Barrier(world);
- time2 = MPI_Wtime();
- time3d = time2 - time1;
-
- return 4;
-}
-
-/* ----------------------------------------------------------------------
- memory usage of local arrays
-------------------------------------------------------------------------- */
-
-double PPPMOld::memory_usage()
-{
- double bytes = nmax*3 * sizeof(double);
- int nbrick = (nxhi_out-nxlo_out+1) * (nyhi_out-nylo_out+1) *
- (nzhi_out-nzlo_out+1);
- bytes += 4 * nbrick * sizeof(FFT_SCALAR);
- bytes += 6 * nfft_both * sizeof(double);
- bytes += nfft_both * sizeof(double);
- bytes += nfft_both*5 * sizeof(FFT_SCALAR);
- bytes += 2 * nbuf * sizeof(FFT_SCALAR);
-
- if (peratom_allocate_flag) {
- bytes += 7 * nbrick * sizeof(FFT_SCALAR);
- bytes += 2 * nbuf_peratom * sizeof(FFT_SCALAR);
- }
-
- if (group_allocate_flag) {
- bytes += 2 * nbrick * sizeof(FFT_SCALAR);
- bytes += 2 * nfft_both * sizeof(FFT_SCALAR);;
- }
-
- return bytes;
-}
-
-/* ----------------------------------------------------------------------
- group-group interactions
- ------------------------------------------------------------------------- */
-
-/* ----------------------------------------------------------------------
- compute the PPPM total long-range force and energy for groups A and B
- ------------------------------------------------------------------------- */
-
-void PPPMOld::compute_group_group(int groupbit_A, int groupbit_B, int BA_flag)
-{
- if (slabflag)
- error->all(FLERR,"Cannot (yet) use K-space slab "
- "correction with compute group/group");
-
- int i;
-
- if (!group_allocate_flag) {
- allocate_groups();
- group_allocate_flag = 1;
- }
-
- e2group = 0; //energy
- f2group[0] = 0; //force in x-direction
- f2group[1] = 0; //force in y-direction
- f2group[2] = 0; //force in z-direction
-
- // map my particle charge onto my local 3d density grid
-
- make_rho_groups(groupbit_A,groupbit_B,BA_flag);
-
- // all procs communicate density values from their ghost cells
- // to fully sum contribution in their 3d bricks
- // remap from 3d decomposition to FFT decomposition
-
- // temporarily store and switch pointers so we can
- // use brick2fft() for groups A and B (without
- // writing an additional function)
-
- FFT_SCALAR ***density_brick_real = density_brick;
- FFT_SCALAR *density_fft_real = density_fft;
-
- // group A
-
- density_brick = density_A_brick;
- density_fft = density_A_fft;
-
- brick2fft();
-
- // group B
-
- density_brick = density_B_brick;
- density_fft = density_B_fft;
-
- brick2fft();
-
- // switch back pointers
-
- density_brick = density_brick_real;
- density_fft = density_fft_real;
-
- // compute potential gradient on my FFT grid and
- // portion of group-group energy/force on this proc's FFT grid
-
- poisson_groups(BA_flag);
-
- const double qscale = qqrd2e * scale;
-
- // total group A <--> group B energy
- // self and boundary correction terms are in compute_group_group.cpp
-
- double e2group_all;
- MPI_Allreduce(&e2group,&e2group_all,1,MPI_DOUBLE,MPI_SUM,world);
- e2group = e2group_all;
-
- e2group *= qscale*0.5*volume;
-
- // total group A <--> group B force
-
- double f2group_all[3];
- MPI_Allreduce(f2group,f2group_all,3,MPI_DOUBLE,MPI_SUM,world);
-
- for (i = 0; i < 3; i++) f2group[i] = qscale*volume*f2group_all[i];
-}
-
-/* ----------------------------------------------------------------------
- allocate group-group memory that depends on # of K-vectors and order
- ------------------------------------------------------------------------- */
-
-void PPPMOld::allocate_groups()
-{
- memory->create3d_offset(density_A_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:density_A_brick");
- memory->create3d_offset(density_B_brick,nzlo_out,nzhi_out,nylo_out,nyhi_out,
- nxlo_out,nxhi_out,"pppm:density_B_brick");
- memory->create(density_A_fft,nfft_both,"pppm:density_A_fft");
- memory->create(density_B_fft,nfft_both,"pppm:density_B_fft");
-}
-
-/* ----------------------------------------------------------------------
- deallocate group-group memory that depends on # of K-vectors and order
- ------------------------------------------------------------------------- */
-
-void PPPMOld::deallocate_groups()
-{
- memory->destroy3d_offset(density_A_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy3d_offset(density_B_brick,nzlo_out,nylo_out,nxlo_out);
- memory->destroy(density_A_fft);
- memory->destroy(density_B_fft);
-}
-
-/* ----------------------------------------------------------------------
- create discretized "density" on section of global grid due to my particles
- density(x,y,z) = charge "density" at grid points of my 3d brick
- (nxlo:nxhi,nylo:nyhi,nzlo:nzhi) is extent of my brick (including ghosts)
- in global grid for group-group interactions
- ------------------------------------------------------------------------- */
-
-void PPPMOld::make_rho_groups(int groupbit_A, int groupbit_B, int BA_flag)
-{
- int l,m,n,nx,ny,nz,mx,my,mz;
- FFT_SCALAR dx,dy,dz,x0,y0,z0;
-
- // clear 3d density arrays
-
- memset(&(density_A_brick[nzlo_out][nylo_out][nxlo_out]),0,
- ngrid*sizeof(FFT_SCALAR));
-
- memset(&(density_B_brick[nzlo_out][nylo_out][nxlo_out]),0,
- ngrid*sizeof(FFT_SCALAR));
-
- // loop over my charges, add their contribution to nearby grid points
- // (nx,ny,nz) = global coords of grid pt to "lower left" of charge
- // (dx,dy,dz) = distance to "lower left" grid pt
- // (mx,my,mz) = global coords of moving stencil pt
-
- double *q = atom->q;
- double **x = atom->x;
- int nlocal = atom->nlocal;
- int *mask = atom->mask;
-
- for (int i = 0; i < nlocal; i++) {
-
- if ((mask[i] & groupbit_A) && (mask[i] & groupbit_B))
- if (BA_flag) continue;
-
- if ((mask[i] & groupbit_A) || (mask[i] & groupbit_B)) {
-
- nx = part2grid[i][0];
- ny = part2grid[i][1];
- nz = part2grid[i][2];
- dx = nx+shiftone - (x[i][0]-boxlo[0])*delxinv;
- dy = ny+shiftone - (x[i][1]-boxlo[1])*delyinv;
- dz = nz+shiftone - (x[i][2]-boxlo[2])*delzinv;
-
- compute_rho1d(dx,dy,dz);
-
- z0 = delvolinv * q[i];
- for (n = nlower; n <= nupper; n++) {
- mz = n+nz;
- y0 = z0*rho1d[2][n];
- for (m = nlower; m <= nupper; m++) {
- my = m+ny;
- x0 = y0*rho1d[1][m];
- for (l = nlower; l <= nupper; l++) {
- mx = l+nx;
-
- // group A
-
- if (mask[i] & groupbit_A)
- density_A_brick[mz][my][mx] += x0*rho1d[0][l];
-
- // group B
-
- if (mask[i] & groupbit_B)
- density_B_brick[mz][my][mx] += x0*rho1d[0][l];
- }
- }
- }
- }
- }
-}
-
-/* ----------------------------------------------------------------------
- FFT-based Poisson solver for group-group interactions
- ------------------------------------------------------------------------- */
-
-void PPPMOld::poisson_groups(int BA_flag)
-{
- int i,j,k,n;
-
- // reuse memory (already declared)
-
- FFT_SCALAR *work_A = work1;
- FFT_SCALAR *work_B = work2;
-
- // transform charge density (r -> k)
-
- // group A
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work_A[n++] = density_A_fft[i];
- work_A[n++] = ZEROF;
- }
-
- fft1->compute(work_A,work_A,1);
-
- // group B
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work_B[n++] = density_B_fft[i];
- work_B[n++] = ZEROF;
- }
-
- fft1->compute(work_B,work_B,1);
-
- // group-group energy and force contribution,
- // keep everything in reciprocal space so
- // no inverse FFTs needed
-
- double scaleinv = 1.0/(nx_pppm*ny_pppm*nz_pppm);
- double s2 = scaleinv*scaleinv;
-
- // energy
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- e2group += s2 * greensfn[i] *
- (work_A[n]*work_B[n] + work_A[n+1]*work_B[n+1]);
- n += 2;
- }
-
- if (BA_flag) return;
-
-
- // multiply by Green's function and s2
- // (only for work_A so it is not squared below)
-
- n = 0;
- for (i = 0; i < nfft; i++) {
- work_A[n++] *= s2 * greensfn[i];
- work_A[n++] *= s2 * greensfn[i];
- }
-
- double partial_group;
-
- // force, x direction
-
- n = 0;
- for (k = nzlo_fft; k <= nzhi_fft; k++)
- for (j = nylo_fft; j <= nyhi_fft; j++)
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- partial_group = work_A[n+1]*work_B[n] - work_A[n]*work_B[n+1];
- f2group[0] += fkx[i] * partial_group;
- n += 2;
- }
-
- // force, y direction
-
- n = 0;
- for (k = nzlo_fft; k <= nzhi_fft; k++)
- for (j = nylo_fft; j <= nyhi_fft; j++)
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- partial_group = work_A[n+1]*work_B[n] - work_A[n]*work_B[n+1];
- f2group[1] += fky[j] * partial_group;
- n += 2;
- }
-
- // force, z direction
-
- n = 0;
- for (k = nzlo_fft; k <= nzhi_fft; k++)
- for (j = nylo_fft; j <= nyhi_fft; j++)
- for (i = nxlo_fft; i <= nxhi_fft; i++) {
- partial_group = work_A[n+1]*work_B[n] - work_A[n]*work_B[n+1];
- f2group[2] += fkz[k] * partial_group;
- n += 2;
- }
-}
diff --git a/src/USER-CUDA/pppm_old.h b/src/USER-CUDA/pppm_old.h
deleted file mode 100644
index 57a92e1202..0000000000
--- a/src/USER-CUDA/pppm_old.h
+++ /dev/null
@@ -1,271 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- Copyright (2003) Sandia Corporation. Under the terms of Contract
- DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
- certain rights in this software. This software is distributed under
- the GNU General Public License.
-
- See the README file in the top-level LAMMPS directory.
-------------------------------------------------------------------------- */
-
-#ifdef KSPACE_CLASS
-
-KSpaceStyle(pppm/old,PPPMOld)
-
-#else
-
-#ifndef LMP_PPPM_OLD_H
-#define LMP_PPPM_OLD_H
-
-#include "lmptype.h"
-#include <mpi.h>
-
-#ifdef FFT_SINGLE
-typedef float FFT_SCALAR;
-#define MPI_FFT_SCALAR MPI_CFLOAT
-#else
-typedef double FFT_SCALAR;
-#define MPI_FFT_SCALAR MPI_DOUBLE
-#endif
-
-#include "kspace.h"
-
-namespace LAMMPS_NS {
-
-class PPPMOld : public KSpace {
- public:
- PPPMOld(class LAMMPS *, int, char **);
- virtual ~PPPMOld();
- virtual void init();
- virtual void setup();
- virtual void compute(int, int);
- virtual int timing_1d(int, double &);
- virtual int timing_3d(int, double &);
- virtual double memory_usage();
-
- virtual void compute_group_group(int, int, int);
-
- protected:
- int me,nprocs;
- int nfactors;
- int *factors;
- double cutoff;
- double volume;
- double delxinv,delyinv,delzinv,delvolinv;
- double shift,shiftone;
- int peratom_allocate_flag;
-
- int nxlo_in,nylo_in,nzlo_in,nxhi_in,nyhi_in,nzhi_in;
- int nxlo_out,nylo_out,nzlo_out,nxhi_out,nyhi_out,nzhi_out;
- int nxlo_ghost,nxhi_ghost,nylo_ghost,nyhi_ghost,nzlo_ghost,nzhi_ghost;
- int nxlo_fft,nylo_fft,nzlo_fft,nxhi_fft,nyhi_fft,nzhi_fft;
- int nlower,nupper;
- int ngrid,nfft,nfft_both;
- int nbuf,nbuf_peratom;
-
- FFT_SCALAR ***density_brick;
- FFT_SCALAR ***vdx_brick,***vdy_brick,***vdz_brick;
- FFT_SCALAR ***u_brick;
- FFT_SCALAR ***v0_brick,***v1_brick,***v2_brick;
- FFT_SCALAR ***v3_brick,***v4_brick,***v5_brick;
- double *greensfn;
- double **vg;
- double *fkx,*fky,*fkz;
- FFT_SCALAR *density_fft;
- FFT_SCALAR *work1,*work2;
- FFT_SCALAR *buf1,*buf2,*buf3,*buf4;
-
- double *gf_b;
- FFT_SCALAR **rho1d,**rho_coeff;
-
- // group-group interactions
-
- int group_allocate_flag;
- FFT_SCALAR ***density_A_brick,***density_B_brick;
- FFT_SCALAR *density_A_fft,*density_B_fft;
-
-
- class FFT3d *fft1,*fft2;
- class Remap *remap;
-
- int **part2grid; // storage for particle -> grid mapping
- int nmax;
-
- int triclinic; // domain settings, orthog or triclinic
- double *boxlo;
- // TIP4P settings
- int typeH,typeO; // atom types of TIP4P water H and O atoms
- double qdist; // distance from O site to negative charge
- double alpha; // geometric factor
-
- void set_grid();
- virtual void allocate();
- virtual void allocate_peratom();
- virtual void deallocate();
- virtual void deallocate_peratom();
- int factorable(int);
- double rms(double, double, bigint, double, double **);
- double diffpr(double, double, double, double, double **);
- void compute_gf_denom();
-
- virtual void particle_map();
- virtual void make_rho();
- virtual void brick2fft();
- virtual void fillbrick();
- virtual void fillbrick_peratom();
- virtual void poisson(int,int);
- virtual void poisson_peratom();
- virtual void fieldforce();
- virtual void fieldforce_peratom();
- void procs2grid2d(int,int,int,int *, int*);
- void compute_rho1d(const FFT_SCALAR &, const FFT_SCALAR &,
- const FFT_SCALAR &);
- void compute_rho_coeff();
- void slabcorr();
-
- // group-group interactions
-
- virtual void allocate_groups();
- virtual void deallocate_groups();
- virtual void make_rho_groups(int, int, int);
- virtual void poisson_groups(int);
-
-/* ----------------------------------------------------------------------
- denominator for Hockney-Eastwood Green's function
- of x,y,z = sin(kx*deltax/2), etc
-
- inf n-1
- S(n,k) = Sum W(k+pi*j)**2 = Sum b(l)*(z*z)**l
- j=-inf l=0
-
- = -(z*z)**n /(2n-1)! * (d/dx)**(2n-1) cot(x) at z = sin(x)
- gf_b = denominator expansion coeffs
-------------------------------------------------------------------------- */
-
- inline double gf_denom(const double &x, const double &y,
- const double &z) const {
- double sx,sy,sz;
- sz = sy = sx = 0.0;
- for (int l = order-1; l >= 0; l--) {
- sx = gf_b[l] + sx*x;
- sy = gf_b[l] + sy*y;
- sz = gf_b[l] + sz*z;
- }
- double s = sx*sy*sz;
- return s*s;
- };
-};
-
-}
-
-#endif
-#endif
-
-/* ERROR/WARNING messages:
-
-E: Illegal ... command
-
-Self-explanatory. Check the input script syntax and compare to the
-documentation for the command. You can use -echo screen as a
-command-line option when running LAMMPS to see the offending line.
-
-E: Cannot use PPPM with 2d simulation
-
-The kspace style pppm cannot be used in 2d simulations. You can use
-2d PPPM in a 3d simulation; see the kspace_modify command.
-
-E: Kspace style requires atom attribute q
-
-The atom style defined does not have these attributes.
-
-E: Cannot use nonperiodic boundaries with PPPM
-
-For kspace style pppm, all 3 dimensions must have periodic boundaries
-unless you use the kspace_modify command to define a 2d slab with a
-non-periodic z dimension.
-
-E: Incorrect boundaries with slab PPPM
-
-Must have periodic x,y dimensions and non-periodic z dimension to use
-2d slab option with PPPM.
-
-E: PPPM order cannot be < 2 or > than %d
-
-This is a limitation of the PPPM implementation in LAMMPS.
-
-E: KSpace style is incompatible with Pair style
-
-Setting a kspace style requires that a pair style with a long-range
-Coulombic or dispersion component be used.
-
-E: Bond and angle potentials must be defined for TIP4P
-
-Cannot use TIP4P pair potential unless bond and angle potentials
-are defined.
-
-E: Bad TIP4P angle type for PPPM/TIP4P
-
-Specified angle type is not valid.
-
-E: Bad TIP4P bond type for PPPM/TIP4P
-
-Specified bond type is not valid.
-
-E: Cannot use kspace solver on system with no charge
-
-No atoms in system have a non-zero charge.
-
-W: System is not charge neutral, net charge = %g
-
-The total charge on all atoms on the system is not 0.0, which
-is not valid for the long-range Coulombic solvers.
-
-W: Reducing PPPM order b/c stencil extends beyond neighbor processor
-
-This may lead to a larger grid than desired. See the kspace_modify overlap
-command to prevent changing of the PPPM order.
-
-E: PPPM grid is too large
-
-The global PPPM grid is larger than OFFSET in one or more dimensions.
-OFFSET is currently set to 4096. You likely need to decrease the
-requested accuracy.
-
-E: PPPM order has been reduced to 0
-
-The auto-adjust of the order failed. You will need to
-set the grid size and order directly via kspace_modify.
-
-E: KSpace accuracy must be > 0
-
-The kspace accuracy designated in the input must be greater than zero.
-
-E: Cannot compute PPPM G
-
-The Ewald factor could not be computed for the current choice of
-grid size, cutoff, accuracy.
-
-E: Out of range atoms - cannot compute PPPM
-
-One or more atoms are attempting to map their charge to a PPPM grid
-point that is not owned by a processor. This is likely for one of two
-reasons, both of them bad. First, it may mean that an atom near the
-boundary of a processor's sub-domain has moved more than 1/2 the
-"neighbor skin distance"_neighbor.html without neighbor lists being
-rebuilt and atoms being migrated to new processors. This also means
-you may be missing pairwise interactions that need to be computed.
-The solution is to change the re-neighboring criteria via the
-"neigh_modify"_neigh_modify command. The safest settings are "delay 0
-every 1 check yes". Second, it may mean that an atom has moved far
-outside a processor's sub-domain or even the entire simulation box.
-This indicates bad physics, e.g. due to highly overlapping atoms, too
-large a timestep, etc.
-
-E: Cannot (yet) use K-space slab correction with compute group/group
-
-This option is not yet supported.
-
-*/
diff --git a/src/USER-CUDA/user_cuda.h b/src/USER-CUDA/user_cuda.h
deleted file mode 100644
index dbcc41ab3b..0000000000
--- a/src/USER-CUDA/user_cuda.h
+++ /dev/null
@@ -1,159 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifndef CUDA_H
-#define CUDA_H
-
-#include "pointers.h"
-#include "cuda_shared.h"
-#include "cuda_data.h"
-#include "cuda_precision.h"
-#include <map>
-
-#ifdef _DEBUG
-#define MYDBG(a) a
-#else
-#define MYDBG(a)
-#endif
-
-namespace LAMMPS_NS
-{
-class Cuda : protected Pointers
-{
- public:
- Cuda(class LAMMPS*);
- ~Cuda();
- //static void setDevice(class LAMMPS*);
- void allocate();
-
- void accelerator(int, char**);
- void activate();
-
- void setSharedDataZero();
- void setSystemParams();
-
- void setDomainParams();
-
- void checkResize();
- void evsetup_eatom_vatom(int eflag_atom, int vflag_atom);
- void uploadAll();
- void downloadAll();
- void upload(int datamask);
- void download(int datamask);
- void downloadX();
-
- class CudaNeighList* registerNeighborList(class NeighList* neigh_list);
- void uploadAllNeighborLists();
- void downloadAllNeighborLists();
- void set_neighinit(int dist_check, double triggerneighsq) {
- shared_data.atom.dist_check = dist_check;
- shared_data.atom.triggerneighsq = triggerneighsq;
- }
- bool decide_by_integrator() {
- return neighbor_decide_by_integrator && cu_xhold && finished_setup;
- }
- void update_xhold(int &maxhold, double* xhold);
-
- void setTimingsZero();
- void print_timings();
-
- void cu_x_download() {
- cu_x->download();
- }
- bool device_set;
- bool dotiming;
- bool dotestatom;
- int testatom;
-
- double uploadtime, downloadtime;
- bool finished_setup, begin_setup;
- bool oncpu;
- bool finished_run;
-
- int self_comm;
-
- int cuda_exists;
-
- double extent[6];
- int* debugdata;
- // data shared between host code and device code
- // (number of atoms, device pointers for up- & download)
- cuda_shared_data shared_data;
-
- cCudaData<double , F_CFLOAT , x >* cu_q;
- cCudaData<double , F_CFLOAT , yx>* cu_f;
- cCudaData<double , V_CFLOAT , x >* cu_mass;
- cCudaData<double , V_CFLOAT , x >* cu_rmass;
- cCudaData<double , V_CFLOAT , yx>* cu_v;
- cCudaData<double , X_CFLOAT , yx>* cu_x;
- cCudaData<double , X_CFLOAT , yx>* cu_xhold;
- cCudaData<int , int , x >* cu_mask;
- cCudaData<int , int , x >* cu_tag;
- cCudaData<int , int , x >* cu_type;
- cCudaData<int , int , x >* cu_image;
- cCudaData<double , ENERGY_CFLOAT, x >* cu_eatom;
- cCudaData<double , ENERGY_CFLOAT, yx>* cu_vatom;
- cCudaData<double , ENERGY_CFLOAT, x >* cu_virial;
- cCudaData<double , ENERGY_CFLOAT, x >* cu_eng_vdwl;
- cCudaData<double , ENERGY_CFLOAT, x >* cu_eng_coul;
- cCudaData<double , double , x >* cu_extent;
- int* binned_id;
- cCudaData<int , int , xx >* cu_binned_id;
- int* binned_idnew;
- cCudaData<int , int , xx >* cu_binned_idnew;
- cCudaData<int , int , x >* cu_debugdata;
- cCudaData<double , X_CFLOAT , x>* cu_radius;
- cCudaData<double , F_CFLOAT , x>* cu_density;
- cCudaData<double , V_CFLOAT , yx>* cu_omega;
- cCudaData<double , F_CFLOAT , yx>* cu_torque;
- cCudaData<int , int , yx >* cu_special;
- cCudaData<int , int , yx >* cu_nspecial;
- cCudaData<int , int , x >* cu_molecule;
-
-
- cCudaData<X_CFLOAT , X_CFLOAT , x>* cu_x_type;
- X_CFLOAT* x_type;
-
- cCudaData<V_CFLOAT , V_CFLOAT , x>* cu_v_radius;
- V_CFLOAT* v_radius;
-
- cCudaData<V_CFLOAT , V_CFLOAT , x>* cu_omega_rmass;
- V_CFLOAT* omega_rmass;
-
- cCudaData<int , int , x >* cu_map_array;
- int neighbor_decide_by_integrator;
-
- bool pinned;
-
- void* copy_buffer;
- int copy_buffersize;
-
- private:
- int pppn; // number of GPUs/node
- int *devicelist; // IDs of GPUs
-
- std::map<class NeighList*, class CudaNeighList*> neigh_lists;
-};
-}
-
-#endif // CUDA_H
diff --git a/src/USER-CUDA/verlet_cuda.cpp b/src/USER-CUDA/verlet_cuda.cpp
deleted file mode 100644
index 8bb4419cda..0000000000
--- a/src/USER-CUDA/verlet_cuda.cpp
+++ /dev/null
@@ -1,1230 +0,0 @@
-/* ----------------------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-
-#include <cstdlib>
-#include <cstdio>
-#include <cstring>
-#include "verlet_cuda.h"
-#include "neighbor.h"
-#include "domain.h"
-#include "comm.h"
-#include "atom.h"
-#include "atom_vec.h"
-#include "force.h"
-#include "pair.h"
-#include "bond.h"
-#include "angle.h"
-#include "dihedral.h"
-#include "improper.h"
-#include "kspace.h"
-#include "output.h"
-#include "update.h"
-#include "modify_cuda.h"
-#include "compute.h"
-#include "fix.h"
-#include "timer.h"
-#include "memory.h"
-#include "error.h"
-#include "cuda_wrapper_cu.h"
-#include "thermo.h"
-#include "cuda_pair_cu.h"
-#include "user_cuda.h"
-#include <ctime>
-#include <cmath>
-#ifdef _OPENMP
-#include <omp.h>
-#endif
-
-using namespace LAMMPS_NS;
-
-#define MAKETIMEING
-
-
-VerletCuda::VerletCuda(LAMMPS* lmp, int narg, char** arg) : Verlet(lmp, narg, arg)
-{
- if (comm->me == 0)
- error->warning(FLERR,"The USER-CUDA pacakge will be deprecated "
- "soon - users should switch to the GPU or KOKKOS packages");
-
- cuda = lmp->cuda;
-
- if(cuda == NULL)
- error->all(FLERR, "You cannot use a /cuda class, without activating 'cuda' acceleration. Provide '-c on' as command-line argument to LAMMPS..");
-
- modify_cuda = (ModifyCuda*) modify;
- int ifix = modify->find_fix("package_omp");
-
- if(ifix >= 0) external_force_clear = 1;
-}
-
-/* ----------------------------------------------------------------------
- setup before run
-------------------------------------------------------------------------- */
-
-void VerletCuda::setup()
-{
- //debug related variables
- cuda->debugdata[0] = 0;
- cuda->cu_debugdata->upload();
- dotestatom = cuda->dotestatom;
- int testatom = cuda->testatom; //48267;
-
- if(atom->nlocal == 0)
- error->warning(FLERR, "# CUDA: There are currently no atoms on one of the MPI processes. This is known to cause errors with the USER-CUDA package. Please use the 'processors' keyword to enforce more balanced processor layout.");
-
- MYDBG(printf("# CUDA VerletCuda::setup start\n");)
-
- cuda->oncpu = true;
- cuda->begin_setup = true;
- cuda->finished_setup = false;
- cuda->finished_run = false;
-
- time_pair = 0;
- time_kspace = 0;
- time_comm = 0;
- time_modify = 0;
- time_fulliterate = 0;
-
- atom->setup();
-
- cuda_shared_atom* cu_atom = & cuda->shared_data.atom;
- cu_atom->update_nlocal = 1;
- cu_atom->update_nmax = 1;
-
- if(atom->molecular || (force->kspace && (not cuda->shared_data.pppm.cudable_force))) cuda->shared_data.pair.collect_forces_later = true;
-
- cuda->setDomainParams();
-
-
- if(cuda->shared_data.me == 0)
- printf("# CUDA: Using precision: Global: %u X: %u V: %u F: %u PPPM: %u \n", CUDA_PRECISION == 1 ? 4 : 8, (int) sizeof(X_CFLOAT), (int) sizeof(V_CFLOAT), (int) sizeof(F_CFLOAT), (int) sizeof(PPPM_CFLOAT));
-
- cuda->allocate();
-
- if (comm->me == 0 && screen) {
- fprintf(screen,"Setting up Verlet run ...\n");
- fprintf(screen," Unit style : %s\n", update->unit_style);
- fprintf(screen," Current step: " BIGINT_FORMAT "\n", update->ntimestep);
- fprintf(screen," Time step : %g\n", update->dt);
- }
-
- // setup domain, communication and neighboring
- // acquire ghosts
- // build neighbor lists
- modify->setup_pre_exchange();
-
- if(triclinic) domain->x2lamda(atom->nlocal);
-
- domain->pbc();
- domain->reset_box();
- comm->setup();
-
- if(neighbor->style) neighbor->setup_bins();
-
- comm->exchange();
-
- if(atom->sortfreq > 0) atom->sort();
-
- comm->borders();
-
- if(triclinic) domain->lamda2x(atom->nlocal + atom->nghost);
-
- cuda->setSystemParams();
- cuda->checkResize();
-
- if(cuda->shared_data.me == 0)
- printf("# CUDA: VerletCuda::setup: Upload data...\n");
-
- cuda->uploadAll();
- neighbor->build();
- neighbor->ncalls = 0;
-
- if(atom->mass)
- cuda->cu_mass->upload();
-
- if(cuda->cu_map_array)
- cuda->cu_map_array->upload();
-
- // compute all forces
-
- ev_set(update->ntimestep);
-
- if(elist_atom) cuda->shared_data.atom.need_eatom = 1;
-
- if(vlist_atom) cuda->shared_data.atom.need_vatom = 1;
-
- if(elist_atom || vlist_atom) cuda->checkResize();
-
- int test_BpA_vs_TpA = true;
- my_times starttime;
- my_times endtime;
-#ifdef NO_PREC_TIMING
- double startsec, endsec;
-#endif
-
- //if(atom->molecular||(force->kspace&&(not cuda->shared_data.pppm.cudable_force))) cuda->shared_data.pair.collect_forces_later = false;
- if(test_BpA_vs_TpA && cuda->shared_data.pair.cudable_force && force->pair && (cuda->shared_data.pair.override_block_per_atom < 0)) {
- int StyleLoops = 10;
-
- if(cuda->shared_data.me == 0)
- printf("Test TpA\n");
-
- cuda->shared_data.pair.use_block_per_atom = 0;
- neighbor->build();
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
-
- force->pair->compute(eflag, vflag);
- CudaWrapper_Sync();
-#ifdef NO_PREC_TIMING
- startsec = 1.0 * clock() / CLOCKS_PER_SEC;
-#endif
- my_gettime(CLOCK_REALTIME, &starttime);
-
- for(int i = 0; i < StyleLoops; i++) {
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
-
- force->pair->compute(eflag, vflag);
- CudaWrapper_Sync();
- }
-
- my_gettime(CLOCK_REALTIME, &endtime);
-
- double TpAtime = endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
-#ifdef NO_PREC_TIMING
- endsec = 1.0 * clock() / CLOCKS_PER_SEC;
- TpAtime = endsec - startsec;
-#endif
-
- if(cuda->shared_data.me == 0)
- printf("Test BpA\n");
-
- cuda->shared_data.pair.use_block_per_atom = 1;
- neighbor->build();
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
-
- force->pair->compute(eflag, vflag);
- CudaWrapper_Sync();
-
- my_gettime(CLOCK_REALTIME, &starttime);
-#ifdef NO_PREC_TIMING
- startsec = 1.0 * clock() / CLOCKS_PER_SEC;
-#endif
-
- for(int i = 0; i < StyleLoops; i++) {
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
-
- force->pair->compute(eflag, vflag);
- CudaWrapper_Sync();
- }
-
- my_gettime(CLOCK_REALTIME, &endtime);
- double BpAtime = endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
-#ifdef NO_PREC_TIMING
- endsec = 1.0 * clock() / CLOCKS_PER_SEC;
- BpAtime = endsec - startsec;
-#endif
-
- if(cuda->shared_data.me == 0)
- printf("\n# CUDA: Timing of parallelisation layout with %i loops:\n", StyleLoops);
-
- if(cuda->shared_data.me == 0)
- printf("# CUDA: BpA TpA\n %lf %lf\n", BpAtime, TpAtime);
-
- if(BpAtime > TpAtime) cuda->shared_data.pair.use_block_per_atom = 0;
- } else
- cuda->shared_data.pair.use_block_per_atom = cuda->shared_data.pair.override_block_per_atom;
-
- //cuda->shared_data.pair.use_block_per_atom = 0;
- if(atom->molecular || (force->kspace && (not cuda->shared_data.pppm.cudable_force))) cuda->shared_data.pair.collect_forces_later = true;
-
- neighbor->build();
- neighbor->ncalls = 0;
-
- force_clear();
-
- modify->setup_pre_force(vflag);
-
- cuda->cu_f->download();
-
- if(cuda->cu_torque)
- cuda->cu_torque->download();
-
- //printf("# Verlet::setup: g f[0] = (%f, %f, %f)\n", atom->f[0][0], atom->f[0][1], atom->f[0][2]);
-
- MYDBG(printf("# CUDA: VerletCuda::setup: initial force compute\n");)
-
- //test_atom(testatom,"pre pair force");
-
- if(cuda->shared_data.pair.cudable_force) {
- cuda->uploadAll();
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
- }
-
- if(force->pair) force->pair->compute(eflag, vflag);
-
- if(cuda->shared_data.pair.cudable_force) {
- if(cuda->shared_data.pair.collect_forces_later) {
- if(eflag) cuda->cu_eng_vdwl->upload();
-
- if(eflag) cuda->cu_eng_coul->upload();
-
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_Pair_CollectForces(&cuda->shared_data, eflag, vflag);
-
- if(eflag) cuda->cu_eng_vdwl->download();
-
- if(eflag) cuda->cu_eng_coul->download();
-
- if(vflag) cuda->cu_virial->download();
- }
-
- cuda->downloadAll();
- }
-
- test_atom(testatom, "post pair force");
-
- MYDBG(printf("# CUDA: VerletCuda::setup: initial force compute done\n");)
- //printf("# Verlet::setup: h f[0] = (%f, %f, %f)\n", atom->f[0][0], atom->f[0][1], atom->f[0][2]);
-
- if(atom->molecular) {
- if(force->bond) force->bond->compute(eflag, vflag);
-
- if(force->angle) force->angle->compute(eflag, vflag);
-
- if(force->dihedral) force->dihedral->compute(eflag, vflag);
-
- if(force->improper) force->improper->compute(eflag, vflag);
- }
-
-
- if(cuda->shared_data.pppm.cudable_force) {
- cuda->cu_tag ->upload();
- cuda->cu_type->upload();
- cuda->cu_x ->upload();
- cuda->cu_v ->upload();
- cuda->cu_f ->upload();
-
- if(cu_atom->q_flag) cuda->cu_q->upload();
- }
-
- if(force->kspace) {
- force->kspace->setup();
- force->kspace->compute(eflag, vflag);
- }
-
- if(cuda->shared_data.pppm.cudable_force) {
- cuda->cu_f ->download();
- }
-
- test_atom(testatom, "post kspace");
-
- cuda->uploadAll();
-
- if(force->newton) comm->reverse_comm();
-
- cuda->downloadAll();
-
- test_atom(testatom, "post reverse comm");
-
- if(cuda->shared_data.me == 0)
- printf("# CUDA: Total Device Memory usage post setup: %lf MB\n", 1.0 * CudaWrapper_CheckMemUsage() / 1024 / 1024);
-
- MYDBG(printf("# CUDA: VerletCuda::setup: call modify setup\n");)
- modify->setup(vflag);
-
- MYDBG(printf("# CUDA: VerletCuda::setup: call modify setup done\n");)
- output->setup(1);
-
- test_atom(testatom, "post setup");
-
- MYDBG(printf("# CUDA: VerletCuda::setup: done\n");)
- cuda->finished_setup = true;
- cuda->oncpu = false;
-}
-
-
-//this routine is in a messy state
-void VerletCuda::setup_minimal(int flag)
-{
-
- printf("SetupMinimal\n");
- dotestatom = 0;
- int testatom = 104;
- cuda->oncpu = true;
- cuda->begin_setup = true;
- cuda->finished_run = false;
- MYDBG(printf("# CUDA VerletCuda::setup start\n");)
- time_pair = 0;
- time_kspace = 0;
- time_comm = 0;
- time_modify = 0;
- time_fulliterate = 0;
-
- //cuda->allocate();
-
- cuda_shared_atom* cu_atom = & cuda->shared_data.atom;
- cu_atom->update_nlocal = 1;
- cu_atom->update_nmax = 1;
-
- if(atom->molecular) cuda->shared_data.pair.collect_forces_later = true;
-
- cuda->setDomainParams();
-
-
-
- if(cuda->shared_data.me == 0)
- printf("# CUDA: VerletCuda::setup: Allocate memory on device for maximum of %i atoms...\n", atom->nmax);
-
- cuda->allocate();
-
-
-
-
- // setup domain, communication and neighboring
- // acquire ghosts
- // build neighbor lists
-
- if(flag) {
- if(triclinic) domain->x2lamda(atom->nlocal);
-
- domain->pbc();
- domain->reset_box();
- comm->setup();
-
- if(neighbor->style) neighbor->setup_bins();
-
- comm->exchange();
- comm->borders();
-
- if(triclinic) domain->lamda2x(atom->nlocal + atom->nghost);
-
- cuda->setSystemParams();
- cuda->checkResize();
- neighbor->build();
- neighbor->ncalls = 0;
- }
-
- if(cuda->shared_data.me == 0)
- printf("# CUDA: VerletCuda::setup: Upload data...\n");
-
- cuda->uploadAll();
- cuda->uploadAllNeighborLists();
-
- if(atom->mass)
- cuda->cu_mass->upload();
-
- if(cuda->cu_map_array)
- cuda->cu_map_array->upload();
-
- // compute all forces
-
- ev_set(update->ntimestep);
-
- if(elist_atom) cuda->shared_data.atom.need_eatom = 1;
-
- if(vlist_atom) cuda->shared_data.atom.need_vatom = 1;
-
- if(elist_atom || vlist_atom) cuda->checkResize();
-
- force_clear();
- cuda->cu_f->download();
-
- //printf("# Verlet::setup: g f[0] = (%f, %f, %f)\n", atom->f[0][0], atom->f[0][1], atom->f[0][2]);
-
- cuda->cu_mass->upload();
- MYDBG(printf("# CUDA: VerletCuda::setup: initial force compute\n");)
-
- test_atom(testatom, "pre pair force");
-
- if(cuda->shared_data.pair.cudable_force) {
- cuda->uploadAll();
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
- }
-
- if(force->pair) force->pair->compute(eflag, vflag);
-
- if(cuda->shared_data.pair.cudable_force) {
- if(cuda->shared_data.pair.collect_forces_later) {
- if(eflag) cuda->cu_eng_vdwl->upload();
-
- if(eflag) cuda->cu_eng_coul->upload();
-
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_Pair_CollectForces(&cuda->shared_data, eflag, vflag);
-
- if(eflag) cuda->cu_eng_vdwl->download();
-
- if(eflag) cuda->cu_eng_coul->download();
-
- if(vflag) cuda->cu_virial->download();
- }
-
- cuda->downloadAll();
- }
-
- test_atom(testatom, "post pair force");
-
- MYDBG(printf("# CUDA: VerletCuda::setup: initial force compute done\n");)
- //printf("# Verlet::setup: h f[0] = (%f, %f, %f)\n", atom->f[0][0], atom->f[0][1], atom->f[0][2]);
-
- if(atom->molecular) {
- if(force->bond) force->bond->compute(eflag, vflag);
-
- if(force->angle) force->angle->compute(eflag, vflag);
-
- if(force->dihedral) force->dihedral->compute(eflag, vflag);
-
- if(force->improper) force->improper->compute(eflag, vflag);
- }
-
-
- if(cuda->shared_data.pppm.cudable_force) {
- cuda->cu_tag ->upload();
- cuda->cu_type->upload();
- cuda->cu_x ->upload();
- cuda->cu_v ->upload();
- cuda->cu_f ->upload();
-
- if(cu_atom->q_flag) cuda->cu_q->upload();
- }
-
- if(force->kspace) {
- force->kspace->setup();
- force->kspace->compute(eflag, vflag);
- }
-
- if(cuda->shared_data.pppm.cudable_force) {
- cuda->cu_f ->download();
- }
-
- test_atom(testatom, "post kspace");
-
- cuda->uploadAll();
-
- if(force->newton) comm->reverse_comm();
-
- cuda->downloadAll();
-
- test_atom(testatom, "post reverse comm");
-
- if(cuda->shared_data.me == 0)
- printf("# CUDA: Total Device Memory usage post setup: %lf MB\n", 1.0 * CudaWrapper_CheckMemUsage() / 1024 / 1024);
-
- MYDBG(printf("# CUDA: VerletCuda::setup: call modify setup\n");)
- modify->setup(vflag);
-
- MYDBG(printf("# CUDA: VerletCuda::setup: done\n");)
- cuda->finished_setup = true;
- cuda->oncpu = false;
-}
-
-//#define TESTATOM
-/* ----------------------------------------------------------------------
- iterate for n steps
-------------------------------------------------------------------------- */
-
-void VerletCuda::run(int n)
-{
- dotestatom = cuda->dotestatom;
- int testatom = cuda->testatom; //48267;
-
-
- my_times starttime;
- my_times endtime;
- my_times starttotal;
- my_times endtotal;
-
- cuda->setTimingsZero();
-
- int nflag, ntimestep, sortflag;
-
- int n_post_integrate = modify_cuda->n_post_integrate;
- int n_pre_exchange = modify_cuda->n_pre_exchange;
- int n_pre_neighbor = modify_cuda->n_pre_neighbor;
- int n_pre_force = modify_cuda->n_pre_force;
- int n_post_force = modify_cuda->n_post_force;
- int n_end_of_step = modify_cuda->n_end_of_step;
- MYDBG(printf("# CUDA: Fixes: i_int: %i p_int: %i f_int: %i pr_exc: %i pr_neigh: %i pr_f: %i p_f: %i eos: %i\n",
- n_initial_integrate, n_post_integrate, n_final_integrate, n_pre_exchange, n_pre_neighbor, n_pre_force, n_post_force, n_end_of_step);)
-
- if(atom->sortfreq > 0) sortflag = 1;
- else sortflag = 0;
-
-
- if(cuda->shared_data.me == 0) {
- if((not cuda->shared_data.pair.cudable_force) && (force->pair))
- error->warning(FLERR, "# CUDA: You asked for a Verlet integration using Cuda, "
- "but selected a pair force which has not yet been ported to Cuda");
-
- if((not cuda->shared_data.pppm.cudable_force) && (force->kspace))
- error->warning(FLERR, "# CUDA: You asked for a Verlet integration using Cuda, "
- "but selected a kspace force which has not yet been ported to Cuda");
-
- if(modify_cuda->n_post_integrate_host + modify_cuda->n_pre_exchange_host + modify_cuda->n_pre_neighbor_host + modify_cuda->n_pre_force_host + modify_cuda->n_post_force_host + modify_cuda->n_end_of_step_host + modify_cuda->n_initial_integrate_host + modify_cuda->n_final_integrate_host)
- error->warning(FLERR, "# CUDA: You asked for a Verlet integration using Cuda, "
- "but several fixes have not yet been ported to Cuda.\n"
- "This can cause a severe speed penalty due to frequent data synchronization between host and GPU.");
-
- if(atom->firstgroupname)
- error->warning(FLERR, "Warning: firstgroupname is used, this will cause additional data transfers.");
- }
-
- cuda->uploadAll();
-
- if(cuda->neighbor_decide_by_integrator && cuda->cu_xhold) {
- const int n = cuda->shared_data.atom.maxhold;
- CudaWrapper_CopyData(cuda->cu_xhold->dev_data(), cuda->cu_x->dev_data(), n * sizeof(X_CFLOAT));
- CudaWrapper_CopyData((void*) & ((X_CFLOAT*)cuda->cu_xhold->dev_data())[n], (void*) & ((X_CFLOAT*)cuda->cu_x->dev_data())[atom->nmax], n * sizeof(X_CFLOAT));
- CudaWrapper_CopyData((void*) & ((X_CFLOAT*)cuda->cu_xhold->dev_data())[2 * n], (void*) & ((X_CFLOAT*)cuda->cu_x->dev_data())[2 * atom->nmax], n * sizeof(X_CFLOAT));
- }
-
- cuda->shared_data.atom.reneigh_flag = 0;
- cuda->shared_data.atom.update_nlocal = 1;
- cuda->shared_data.atom.update_nmax = 1;
- cuda->shared_data.atom.update_neigh = 1;
- cuda->shared_data.domain.update = 1;
- cuda->shared_data.buffer_new = 1;
- cuda->uploadtime = 0;
- cuda->downloadtime = 0;
- int firstreneigh = 1;
-
- for(int i = 0; i < n; i++) {
- if(atom->nlocal == 0)
- error->warning(FLERR, "# CUDA: There are currently no atoms on one of the MPI processes. This is currently prone to encountering errors with USER-CUDA package. Please use the 'processors' keyword to use a more balanced processor layout.");
-
- ntimestep = ++update->ntimestep;
- ev_set(ntimestep);
-
- // initial time integration
-
- test_atom(testatom, "Pre initial");
-
- MYDBG(printf("# CUDA VerletCuda::iterate: before initial_integrate\n");)
-
- modify->initial_integrate(vflag);
-
- MYDBG(printf("# CUDA VerletCuda::iterate: after initial_integrate\n");)
-
- if(n_post_integrate) modify->post_integrate();
-
-
-
- // regular communication vs neighbor list rebuild
-
- test_atom(testatom, "Pre Exchange");
-
- MYDBG(printf("# CUDA VerletCuda::iterate: before neighbor decide\n");)
- nflag = neighbor->decide();
-
- if(nflag == 0) {
- MYDBG(printf("# CUDA VerletCuda::iterate: communicate\n");)
- timer->stamp();
-
- if((not(eflag || vflag)) && (cuda->shared_data.overlap_comm)) {
- //overlap forward communication of ghost atom positions with inner force calculation (interactions between local atoms)
- //build communication buffers
- // printf("Pre forward_comm(1)\n");
- my_gettime(CLOCK_REALTIME, &starttotal);
- cuda->shared_data.atom.reneigh_flag = 0;
- my_gettime(CLOCK_REALTIME, &starttime);
- timer->stamp();
- comm->forward_comm(1);
- timer->stamp(Timer::COMM);
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.comm_forward_total +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
-
- //prepare force calculation
- // printf("Pre force_clear\n");
- force_clear();
- // printf("Pre Generate XType\n");
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
-
- //start force calculation asynchronus
- cuda->shared_data.comm.comm_phase = 1;
- force->pair->compute(eflag, vflag);
- timer->stamp(Timer::PAIR);
- //CudaWrapper_Sync();
-
- //download comm buffers from GPU, perform MPI communication and upload buffers again
- my_gettime(CLOCK_REALTIME, &starttime);
- comm->forward_comm(2);
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.comm_forward_total +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
- timer->stamp(Timer::COMM);
-
- //wait for force calculation
- CudaWrapper_Sync();
- timer->stamp(Timer::PAIR);
-
- //unpack communication buffers
- my_gettime(CLOCK_REALTIME, &starttime);
- comm->forward_comm(3);
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.comm_forward_total +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
-
- timer->stamp(Timer::COMM);
- MYDBG(printf("# CUDA VerletCuda::iterate: communicate done\n");)
- cuda->shared_data.cuda_timings.test1 +=
- endtotal.tv_sec - starttotal.tv_sec + 1.0 * (endtotal.tv_nsec - starttotal.tv_nsec) / 1000000000;
- } else {
- //perform standard forward communication
- my_gettime(CLOCK_REALTIME, &starttime);
- comm->forward_comm();
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.comm_forward_total +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
- timer->stamp(Timer::COMM);
- MYDBG(printf("# CUDA VerletCuda::iterate: communicate done\n");)
- }
- } else {
- int nlocalold = cuda->shared_data.atom.nlocal;
-
- if(firstreneigh) {
- cuda->shared_data.atom.update_nlocal = 1;
- cuda->shared_data.atom.update_nmax = 1;
- firstreneigh = 0;
- }
-
- cuda->shared_data.buffer_new = 1;
- MYDBG(printf("# CUDA VerletCuda::iterate: neighbor\n");)
- cuda->setDomainParams();
-
- if(n_pre_exchange) modify->pre_exchange();
-
- if(atom->nlocal != cuda->shared_data.atom.nlocal) { //did someone add atoms during pre_exchange?
- cuda->checkResize();
- cuda->uploadAll();
- }
-
- //check domain changes
- if(domain->triclinic) domain->x2lamda(atom->nlocal);
-
- MYDBG(printf("# CUDA VerletCuda::iterate: neighbor pbc\n");)
- domain->pbc();
-
- if(domain->box_change) {
- domain->reset_box();
- comm->setup();
-
- if(neighbor->style) neighbor->setup_bins();
-
- }
-
- timer->stamp();
- MYDBG(printf("# CUDA VerletCuda::iterate: neighbor exchange\n");)
-
- //perform exchange of local atoms
- my_gettime(CLOCK_REALTIME, &starttime);
- comm->exchange();
- my_gettime(CLOCK_REALTIME, &endtime);
-
- //special and nspecial fields of the atom data are not currently transfered via the GPU buffer might be changed in the future
- if(comm->nprocs > 1) {
- my_gettime(CLOCK_REALTIME, &starttime);
-
- if(atom->special)
- cuda->cu_special->upload();
-
- if(atom->nspecial)
- cuda->cu_nspecial->upload();
-
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.test1 +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
- }
-
- cuda->shared_data.cuda_timings.comm_exchange_total +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
-
- if(nlocalold != cuda->shared_data.atom.nlocal) cuda->shared_data.atom.update_nlocal = 2;
-
- //sort atoms
- if(sortflag && ntimestep >= atom->nextsort) atom->sort();
-
- MYDBG(printf("# CUDA VerletCuda::iterate: neighbor borders\n");)
-
- //generate ghost atom lists, and transfer ghost atom data
- my_gettime(CLOCK_REALTIME, &starttime);
- comm->borders();
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.comm_border_total +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
-
- my_gettime(CLOCK_REALTIME, &starttime);
- //atom index maps are generated on CPU, and need to be transfered to GPU if they are used
- if(cuda->cu_map_array)
- cuda->cu_map_array->upload();
-
-
- if(domain->triclinic) domain->lamda2x(atom->nlocal + atom->nghost);
-
- if(n_pre_neighbor) modify->pre_neighbor();
-
- cuda->shared_data.buffer_new = 2;
-
- MYDBG(printf("# CUDA VerletCuda::iterate: neighbor build\n");)
- timer->stamp(Timer::COMM);
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.test2 +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
-
- //rebuild neighbor list
- test_atom(testatom, "Pre Neighbor");
- neighbor->build(0);
- timer->stamp(Timer::NEIGH);
- MYDBG(printf("# CUDA VerletCuda::iterate: neighbor done\n");)
- //if bonded interactions are used (in this case collect_forces_later is true), transfer data which only changes upon exchange/border routines from GPU to CPU
- if(cuda->shared_data.pair.collect_forces_later) {
- if(cuda->cu_molecule) cuda->cu_molecule->downloadAsync(2);
-
- cuda->cu_tag->downloadAsync(2);
- cuda->cu_type->downloadAsync(2);
- cuda->cu_mask->downloadAsync(2);
-
- if(cuda->cu_q) cuda->cu_q->downloadAsync(2);
- }
- cuda->shared_data.comm.comm_phase = 3;
- }
-
- test_atom(testatom, "Post Exchange");
-
- // force computations
-
- //only do force_clear if it has not been done during overlap of communication with local interactions
- if(not((not(eflag || vflag)) && (cuda->shared_data.overlap_comm) && (cuda->shared_data.comm.comm_phase < 3)))
- force_clear();
-
- if(n_pre_force) modify->pre_force(vflag);
-
- timer->stamp();
-
- //if overlap of bonded interactions with nonbonded interactions takes place, download forces and positions
- /* if(cuda->shared_data.pair.collect_forces_later)
- {
- cuda->cu_x->downloadAsync(2);
- cuda->cu_f->downloadAsync(2);
- }*/
-
- if(force->pair) {
- if((not(eflag || vflag)) && (cuda->shared_data.overlap_comm) && (cuda->shared_data.comm.comm_phase < 3) && cuda->shared_data.pair.cudable_force) {
- //second part of force calculations in case of overlaping it with commuincation. Only interactions between local and ghost atoms are done now
- //regenerate data layout for force computations, its actually only needed for the ghost atoms
- cuda->shared_data.comm.comm_phase = 2;
-
- my_times atime1, atime2;
- my_gettime(CLOCK_REALTIME, &atime1);
-
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
-
- my_gettime(CLOCK_REALTIME, &atime2);
- cuda->shared_data.cuda_timings.pair_xtype_conversion +=
- atime2.tv_sec - atime1.tv_sec + 1.0 * (atime2.tv_nsec - atime1.tv_nsec) / 1000000000;
- force->pair->compute(eflag, vflag);
-
- } else {
- //calculate complete pair interactions
- if(not cuda->shared_data.pair.cudable_force) cuda->downloadAll();
- else {
- //regenerate data layout for force computations, its actually only needed for the ghost atoms
- my_times atime1, atime2;
- my_gettime(CLOCK_REALTIME, &atime1);
-
- Cuda_Pair_GenerateXType(&cuda->shared_data);
-
- if(cuda->cu_v_radius)
- Cuda_Pair_GenerateVRadius(&cuda->shared_data);
-
- if(cuda->cu_omega_rmass)
- Cuda_Pair_GenerateOmegaRmass(&cuda->shared_data);
-
- my_gettime(CLOCK_REALTIME, &atime2);
- cuda->shared_data.cuda_timings.pair_xtype_conversion +=
- atime2.tv_sec - atime1.tv_sec + 1.0 * (atime2.tv_nsec - atime1.tv_nsec) / 1000000000;
- }
-
- cuda->shared_data.comm.comm_phase = 0;
- force->pair->compute(eflag, vflag);
- }
-
- if(not cuda->shared_data.pair.cudable_force) cuda->uploadAll();
-
- //wait for force calculation in case of not using overlap with bonded interactions
- if(not cuda->shared_data.pair.collect_forces_later)
- CudaWrapper_Sync();
-
- timer->stamp(Timer::PAIR);
- }
-
- //calculate bonded interactions
- if(atom->molecular) {
- cuda->cu_x->downloadAsync(2);
-
- if(n_pre_force == 0) Verlet::force_clear();
- else cuda->cu_f->downloadAsync(2);
-
- timer->stamp(Timer::PAIR);
-
- if(neighbor->lastcall == update->ntimestep) {
- neighbor->build_topology();
- timer->stamp(Timer::NEIGH);
- }
-
- test_atom(testatom, "pre bond force");
-
- if(force->bond) force->bond->compute(eflag, vflag);
-
- if(force->angle) force->angle->compute(eflag, vflag);
-
- if(force->dihedral) force->dihedral->compute(eflag, vflag);
-
- if(force->improper) force->improper->compute(eflag, vflag);
-
- timer->stamp(Timer::BOND);
- }
-
- //collect forces in case pair force and bonded interactions were overlapped, and either no KSPACE or a GPU KSPACE style is used
- if(cuda->shared_data.pair.collect_forces_later && cuda->shared_data.pair.cudable_force && (not(force->kspace && (not cuda->shared_data.pppm.cudable_force)))) {
- my_gettime(CLOCK_REALTIME, &starttime);
- cuda->cu_f->uploadAsync(2);
-
- test_atom(testatom, "post molecular force");
-
-
- if(eflag) cuda->cu_eng_vdwl->upload();
-
- if(eflag) cuda->cu_eng_coul->upload();
-
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_Pair_CollectForces(&cuda->shared_data, eflag, vflag);
-
- if(eflag) cuda->cu_eng_vdwl->download();
-
- if(eflag) cuda->cu_eng_coul->download();
-
- if(vflag) cuda->cu_virial->download();
-
- timer->stamp(Timer::PAIR);
-
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.pair_force_collection +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
- }
-
- //compute kspace force
- if(force->kspace) {
- if((not cuda->shared_data.pppm.cudable_force) && (not cuda->shared_data.pair.collect_forces_later))
- cuda->downloadAll();
-
- if((not cuda->shared_data.pppm.cudable_force) && (cuda->shared_data.pair.collect_forces_later) && (not atom->molecular)) {
- cuda->cu_x->downloadAsync(2);
-
- if(n_pre_force == 0) Verlet::force_clear();
- else cuda->cu_f->downloadAsync(2);
-
- timer->stamp(Timer::PAIR);
- }
-
- force->kspace->compute(eflag, vflag);
-
- if((not cuda->shared_data.pppm.cudable_force) && (not cuda->shared_data.pair.collect_forces_later))
- cuda->uploadAll();
-
- timer->stamp(Timer::KSPACE);
- }
-
- //collect forces in case pair forces and kspace was overlaped
- if(cuda->shared_data.pair.collect_forces_later && cuda->shared_data.pair.cudable_force && ((force->kspace && (not cuda->shared_data.pppm.cudable_force)))) {
- cuda->cu_f->uploadAsync(2);
-
- my_gettime(CLOCK_REALTIME, &starttime);
-
- if(eflag) cuda->cu_eng_vdwl->upload();
-
- if(eflag) cuda->cu_eng_coul->upload();
-
- if(vflag) cuda->cu_virial->upload();
-
- Cuda_Pair_CollectForces(&cuda->shared_data, eflag, vflag);
-
- if(eflag) cuda->cu_eng_vdwl->download();
-
- if(eflag) cuda->cu_eng_coul->download();
-
- if(vflag) cuda->cu_virial->download();
-
- timer->stamp(Timer::PAIR);
-
- my_gettime(CLOCK_REALTIME, &endtime);
- cuda->shared_data.cuda_timings.pair_force_collection +=
- endtime.tv_sec - starttime.tv_sec + 1.0 * (endtime.tv_nsec - starttime.tv_nsec) / 1000000000;
- }
-
- //send forces on ghost atoms back to other GPU: THIS SHOULD NEVER HAPPEN
- if(force->newton) {
- comm->reverse_comm();
- timer->stamp(Timer::COMM);
- }
-
- test_atom(testatom, "post force");
- // force modifications, final time integration, diagnostics
-
- if(n_post_force) modify->post_force(vflag);
-
- test_atom(testatom, "pre final");
-
- modify->final_integrate();
-
- test_atom(testatom, "post final");
-
- if(n_end_of_step) modify->end_of_step();
-
- // all output
-
- test_atom(testatom, "pre output");
-
- if(ntimestep == output->next) {
- if(not output->thermo->cudable)
- cuda->downloadAll();
-
- timer->stamp();
- output->write(ntimestep);
- timer->stamp(Timer::OUTPUT);
- }
-
-
- test_atom(testatom, "post output");
-
- if(cuda->shared_data.atom.update_nlocal > 0)
- cuda->shared_data.atom.update_nlocal--;
-
- if(cuda->shared_data.atom.update_nmax > 0)
- cuda->shared_data.atom.update_nmax--;
-
- if(cuda->shared_data.atom.update_neigh > 0)
- cuda->shared_data.atom.update_neigh--;
-
- if(cuda->shared_data.domain.update > 0)
- cuda->shared_data.domain.update--;
-
- if(cuda->shared_data.buffer_new > 0)
- cuda->shared_data.buffer_new--;
-
- cuda->shared_data.atom.reneigh_flag = 0;
- }
-
-
- cuda->downloadAll();
- cuda->downloadAllNeighborLists();
- cuda->shared_data.atom.update_nlocal = 1;
- cuda->shared_data.atom.update_nmax = 1;
- cuda->shared_data.atom.update_neigh = 1;
- cuda->shared_data.buffer_new = 1;
- cuda->shared_data.domain.update = 1;
- cuda->oncpu = true;
- cuda->finished_run = true;
-}
-
-
-/* ----------------------------------------------------------------------
- clear force on own & ghost atoms
- setup and clear other arrays as needed
-------------------------------------------------------------------------- */
-
-void VerletCuda::force_clear()
-{
- cuda->cu_f->memset_device(0);
-
- if(cuda->cu_torque) cuda->cu_torque->memset_device(0);
-
-#if 0
- //The rest should not be necessary
- int i;
-
- for(i = 0; i < atom->nlocal; i++) {
- atom->f[i][0] = 0.0;
- atom->f[i][1] = 0.0;
- atom->f[i][2] = 0.0;
- }
-
- // clear force on all particles
- // if either newton flag is set, also include ghosts
-
- if(neighbor->includegroup == 0) {
- int nall;
-
- if(force->newton) nall = atom->nlocal + atom->nghost;
- else nall = atom->nlocal;
-
- if(torqueflag) {
- double** torque = atom->torque;
-
- for(i = 0; i < nall; i++) {
- torque[i][0] = 0.0;
- torque[i][1] = 0.0;
- torque[i][2] = 0.0;
- }
- }
-
- // neighbor includegroup flag is set
- // clear force only on initial nfirst particles
- // if either newton flag is set, also include ghosts
-
- } else {
- int nall = atom->nfirst;
-
-
- if(torqueflag) {
- double** torque = atom->torque;
-
- for(i = 0; i < nall; i++) {
- torque[i][0] = 0.0;
- torque[i][1] = 0.0;
- torque[i][2] = 0.0;
- }
- }
-
- if(force->newton) {
- nall = atom->nlocal + atom->nghost;
-
- if(torqueflag) {
- double** torque = atom->torque;
-
- for(i = atom->nlocal; i < nall; i++) {
- torque[i][0] = 0.0;
- torque[i][1] = 0.0;
- torque[i][2] = 0.0;
- }
- }
- }
- }
-#endif
-}
-
-void VerletCuda::test_atom(int aatom, const char* string) //printing properties of one atom for test purposes
-{
- if(not dotestatom) return;
-
- bool check = false;
-
- if(cuda->finished_setup) cuda->downloadAll();
-
- for(int i = 0; i < atom->nlocal + atom->nghost; i++) {
- if((atom->tag[i] == aatom) && (i < atom->nlocal)) {
-
- printf("%i # CUDA %s: " BIGINT_FORMAT " %i %e %e %e %i ",
- comm->me, string, update->ntimestep, atom->tag[i],
- atom->x[i][0], atom->v[i][0], atom->f[i][0], i);
-
- if(atom->molecular && (i < atom->nlocal)) {
- printf(" // %i %i %i ", atom->num_bond[i], atom->num_angle[i], atom->num_dihedral[i]);
-
- for(int k = 0; k < atom->num_bond[i]; k++)
- printf("// %i %i ", atom->bond_type[i][k], atom->bond_atom[i][k]);
- }
-
- printf("\n");
- }
-
- if(i < atom->nlocal) {
- if((atom->v[i][0] < -100 || atom->v[i][0] > 100) ||
- (atom->v[i][1] < -100 || atom->v[i][1] > 100) ||
- (atom->v[i][2] < -100 || atom->v[i][2] > 100) ||
- (atom->v[i][0] != atom->v[i][0]) ||
- (atom->v[i][1] != atom->v[i][1]) ||
- (atom->v[i][2] != atom->v[i][2])) {
- printf("%i # CUDA %s velocity: %i %e %e %e %i\n", comm->me, string, atom->tag[i], atom->x[i][0], atom->v[i][0], atom->f[i][0], i);
- check = true;
- }
-
- if((atom->f[i][0] < -10000 || atom->f[i][0] > 10000) ||
- (atom->f[i][1] < -10000 || atom->f[i][1] > 10000) ||
- (atom->f[i][2] < -10000 || atom->f[i][2] > 10000) ||
- (atom->f[i][0] != atom->f[i][0]) ||
- (atom->f[i][1] != atom->f[i][1]) ||
- (atom->f[i][2] != atom->f[i][2])) {
- printf("%i # CUDA %s force: %i %e %e %e %i\n", comm->me, string, atom->tag[i], atom->x[i][0], atom->v[i][0], atom->f[i][0], i);
- check = true;
- }
-
- if(atom->tag[i] <= 0)
- printf("%i # CUDA %s tag: %i %e %e %e %i\n", comm->me, string, atom->tag[i], atom->x[i][0], atom->v[i][0], atom->f[i][0], i);
- }
- }
-
- if(check) exit(0);
-}
diff --git a/src/USER-CUDA/verlet_cuda.h b/src/USER-CUDA/verlet_cuda.h
deleted file mode 100644
index 6760828010..0000000000
--- a/src/USER-CUDA/verlet_cuda.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/* -*- c++ -*- ----------------------------------------------------------
- LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
-
- Original Version:
- http://lammps.sandia.gov, Sandia National Laboratories
- Steve Plimpton, sjplimp@sandia.gov
-
- See the README file in the top-level LAMMPS directory.
-
- -----------------------------------------------------------------------
-
- USER-CUDA Package and associated modifications:
- https://sourceforge.net/projects/lammpscuda/
-
- Christian Trott, christian.trott@tu-ilmenau.de
- Lars Winterfeld, lars.winterfeld@tu-ilmenau.de
- Theoretical Physics II, University of Technology Ilmenau, Germany
-
- See the README file in the USER-CUDA directory.
-
- This software is distributed under the GNU General Public License.
-------------------------------------------------------------------------- */
-
-#ifdef INTEGRATE_CLASS
-
-IntegrateStyle(verlet/cuda,VerletCuda)
-
-#else
-
-
-#ifndef LMP_VERLET_CUDA_H
-#define LMP_VERLET_CUDA_H
-#include "verlet.h"
-#include "modify_cuda.h"
-
-namespace LAMMPS_NS {
-
-class VerletCuda : public Verlet
-{
- public:
- VerletCuda(class LAMMPS *, int, char **);
- void setup();
- void setup_minimal(int);
- void run(int);
-
- void test_atom(int atom,const char* astring); //debugging purpose
- int dotestatom; //debugging purpose
-
- protected:
- class Cuda *cuda;
- void force_clear();
- double time_pair;
- double time_kspace;
- double time_comm;
- double time_modify;
- double time_fulliterate;
- ModifyCuda* modify_cuda;
-};
-
-}
-
-#endif
-#endif
--
GitLab