From e337db40593ce56685be1dfdc20fd513680b65d5 Mon Sep 17 00:00:00 2001
From: Stan Moore <stamoor@sandia.gov>
Date: Mon, 6 Nov 2017 11:31:43 -0700
Subject: [PATCH] Replicate bbox from Chris Knight
---
doc/src/replicate.txt | 12 +-
src/replicate.cpp | 480 +++++++++++++++++++++++++++++++++++-------
2 files changed, 417 insertions(+), 75 deletions(-)
diff --git a/doc/src/replicate.txt b/doc/src/replicate.txt
index 291558e0e5..08523ecdd8 100644
--- a/doc/src/replicate.txt
+++ b/doc/src/replicate.txt
@@ -10,9 +10,11 @@ replicate command :h3
[Syntax:]
-replicate nx ny nz :pre
+replicate nx ny nz {keyword} :pre
-nx,ny,nz = replication factors in each dimension :ul
+nx,ny,nz = replication factors in each dimension :ulb
+optional {keyword} = {bbox} :l
+ {bbox} = only check atoms in replicas that overlap with a processor's subdomain :ule
[Examples:]
@@ -43,6 +45,12 @@ file that crosses a periodic boundary should be between two atoms with
image flags that differ by 1. This will allow the bond to be
unwrapped appropriately.
+The optional keyword {bbox} uses a bounding box to only check atoms
+in replicas that overlap with a processor's subdomain when assigning
+atoms to processors, and thus can result in substantial speedups for
+calculations using a large number of processors. It does require
+temporarily using more memory.
+
[Restrictions:]
A 2d simulation cannot be replicated in the z dimension.
diff --git a/src/replicate.cpp b/src/replicate.cpp
index f3d1964169..1251688211 100644
--- a/src/replicate.cpp
+++ b/src/replicate.cpp
@@ -44,7 +44,7 @@ void Replicate::command(int narg, char **arg)
if (domain->box_exist == 0)
error->all(FLERR,"Replicate command before simulation box is defined");
- if (narg != 3) error->all(FLERR,"Illegal replicate command");
+ if (narg < 3 || narg > 4) error->all(FLERR,"Illegal replicate command");
int me = comm->me;
int nprocs = comm->nprocs;
@@ -58,6 +58,10 @@ void Replicate::command(int narg, char **arg)
int nz = force->inumeric(FLERR,arg[2]);
int nrep = nx*ny*nz;
+ int bbox_flag = 0;
+ if (narg == 4)
+ if (strcmp(arg[3],"bbox") == 0) bbox_flag = 1;
+
// error and warning checks
if (nx <= 0 || ny <= 0 || nz <= 0)
@@ -99,6 +103,37 @@ void Replicate::command(int narg, char **arg)
maxmol = maxmol_all;
}
+ // check image flags maximum extent; only efficient small image flags compared to new system
+
+ int _imagelo[3], _imagehi[3];
+ _imagelo[0] = 0;
+ _imagelo[1] = 0;
+ _imagelo[2] = 0;
+ _imagehi[0] = 0;
+ _imagehi[1] = 0;
+ _imagehi[2] = 0;
+
+ if (bbox_flag) {
+
+ for (i=0; i<atom->nlocal; ++i) {
+ imageint image = atom->image[i];
+ int xbox = (image & IMGMASK) - IMGMAX;
+ int ybox = (image >> IMGBITS & IMGMASK) - IMGMAX;
+ int zbox = (image >> IMG2BITS) - IMGMAX;
+
+ if (xbox < _imagelo[0]) _imagelo[0] = xbox;
+ if (ybox < _imagelo[1]) _imagelo[1] = ybox;
+ if (zbox < _imagelo[2]) _imagelo[2] = zbox;
+
+ if (xbox > _imagehi[0]) _imagehi[0] = xbox;
+ if (ybox > _imagehi[1]) _imagehi[1] = ybox;
+ if (zbox > _imagehi[2]) _imagehi[2] = zbox;
+ }
+
+ MPI_Allreduce(MPI_IN_PLACE, &(_imagelo[0]), 3, MPI_INT, MPI_MIN, world);
+ MPI_Allreduce(MPI_IN_PLACE, &(_imagehi[0]), 3, MPI_INT, MPI_MAX, world);
+ }
+
// unmap existing atoms via image flags
for (i = 0; i < atom->nlocal; i++)
@@ -280,93 +315,392 @@ void Replicate::command(int narg, char **arg)
double *coord;
int tag_enable = atom->tag_enable;
- for (int iproc = 0; iproc < nprocs; iproc++) {
- if (me == iproc) {
- n = 0;
- for (i = 0; i < old->nlocal; i++) n += old_avec->pack_restart(i,&buf[n]);
+ if (bbox_flag) {
+
+ // allgather size of buf on each proc
+
+ n = 0;
+ for (i = 0; i < old->nlocal; i++) n += old_avec->pack_restart(i,&buf[n]);
+
+ int * size_buf_rnk;
+ memory->create(size_buf_rnk, nprocs, "replicate:size_buf_rnk");
+
+ MPI_Allgather(&n, 1, MPI_INT, size_buf_rnk, 1, MPI_INT, world);
+
+ // size of buf_all
+
+ int size_buf_all = 0;
+ MPI_Allreduce(&n, &size_buf_all, 1, MPI_INT, MPI_SUM, world);
+
+ if (me == 0 && screen) {
+ fprintf(screen," bounding box image = (%i %i %i) to (%i %i %i)\n",
+ _imagelo[0],_imagelo[1],_imagelo[2],_imagehi[0],_imagehi[1],_imagehi[2]);
+ fprintf(screen," bounding box extra memory = %.2f MB\n",
+ (double)size_buf_all*sizeof(double)/1024/1024);
}
- MPI_Bcast(&n,1,MPI_INT,iproc,world);
- MPI_Bcast(buf,n,MPI_DOUBLE,iproc,world);
+
+ // rnk offsets
+
+ int * disp_buf_rnk;
+ memory->create(disp_buf_rnk, nprocs, "replicate:disp_buf_rnk");
+ disp_buf_rnk[0] = 0;
+ for (i=1; i<nprocs; ++i) disp_buf_rnk[i] = disp_buf_rnk[i-1] + size_buf_rnk[i-1];
+
+ // allgather buf_all
+
+ double * buf_all;
+ memory->create(buf_all, size_buf_all, "replicate:buf_all");
+
+ MPI_Allgatherv(buf, n, MPI_DOUBLE, buf_all, size_buf_rnk, disp_buf_rnk, MPI_DOUBLE, world);
+
+ // bounding box of original unwrapped system
+
+ double _orig_lo[3], _orig_hi[3];
+ if (triclinic) {
+ _orig_lo[0] = domain->boxlo[0] + _imagelo[0] * old_xprd + _imagelo[1] * old_xy + _imagelo[2] * old_xz;
+ _orig_lo[1] = domain->boxlo[1] + _imagelo[1] * old_yprd + _imagelo[2] * old_yz;
+ _orig_lo[2] = domain->boxlo[2] + _imagelo[2] * old_zprd;
+
+ _orig_hi[0] = domain->boxlo[0] + (_imagehi[0]+1) * old_xprd + (_imagehi[1]+1) * old_xy + (_imagehi[2]+1) * old_xz;
+ _orig_hi[1] = domain->boxlo[1] + (_imagehi[1]+1) * old_yprd + (_imagehi[2]+1) * old_yz;
+ _orig_hi[2] = domain->boxlo[2] + (_imagehi[2]+1) * old_zprd;
+ } else {
+ _orig_lo[0] = domain->boxlo[0] + _imagelo[0] * old_xprd;
+ _orig_lo[1] = domain->boxlo[1] + _imagelo[1] * old_yprd;
+ _orig_lo[2] = domain->boxlo[2] + _imagelo[2] * old_zprd;
+
+ _orig_hi[0] = domain->boxlo[0] + (_imagehi[0]+1) * old_xprd;
+ _orig_hi[1] = domain->boxlo[1] + (_imagehi[1]+1) * old_yprd;
+ _orig_hi[2] = domain->boxlo[2] + (_imagehi[2]+1) * old_zprd;
+ }
+
+ double _lo[3], _hi[3];
+
+ int num_replicas_added = 0;
for (ix = 0; ix < nx; ix++) {
for (iy = 0; iy < ny; iy++) {
for (iz = 0; iz < nz; iz++) {
- // while loop over one proc's atom list
+ // domain->remap() overwrites coordinates, so always recompute here
+
+ if (triclinic) {
+ _lo[0] = _orig_lo[0] + ix * old_xprd + iy * old_xy + iz * old_xz;
+ _hi[0] = _orig_hi[0] + ix * old_xprd + iy * old_xy + iz * old_xz;
+
+ _lo[1] = _orig_lo[1] + iy * old_yprd + iz * old_yz;
+ _hi[1] = _orig_hi[1] + iy * old_yprd + iz * old_yz;
+
+ _lo[2] = _orig_lo[2] + iz * old_zprd;
+ _hi[2] = _orig_hi[2] + iz * old_zprd;
+ } else {
+ _lo[0] = _orig_lo[0] + ix * old_xprd;
+ _hi[0] = _orig_hi[0] + ix * old_xprd;
+
+ _lo[1] = _orig_lo[1] + iy * old_yprd;
+ _hi[1] = _orig_hi[1] + iy * old_yprd;
- m = 0;
- while (m < n) {
- image = ((imageint) IMGMAX << IMG2BITS) |
+ _lo[2] = _orig_lo[2] + iz * old_zprd;
+ _hi[2] = _orig_hi[2] + iz * old_zprd;
+ }
+
+ // test if bounding box of shifted replica overlaps sub-domain of proc
+ // if not, then skip testing atoms
+
+ int xoverlap = 1;
+ int yoverlap = 1;
+ int zoverlap = 1;
+ if (triclinic) {
+ double _llo[3];
+ domain->x2lamda(_lo,_llo);
+ double _lhi[3];
+ domain->x2lamda(_hi,_lhi);
+
+ if (_llo[0] > (subhi[0] - EPSILON)
+ || _lhi[0] < (sublo[0] + EPSILON) ) xoverlap = 0;
+ if (_llo[1] > (subhi[1] - EPSILON)
+ || _lhi[1] < (sublo[1] + EPSILON) ) yoverlap = 0;
+ if (_llo[2] > (subhi[2] - EPSILON)
+ || _lhi[2] < (sublo[2] + EPSILON) ) zoverlap = 0;
+ } else {
+ if (_lo[0] > (subhi[0] - EPSILON)
+ || _hi[0] < (sublo[0] + EPSILON) ) xoverlap = 0;
+ if (_lo[1] > (subhi[1] - EPSILON)
+ || _hi[1] < (sublo[1] + EPSILON) ) yoverlap = 0;
+ if (_lo[2] > (subhi[2] - EPSILON)
+ || _hi[2] < (sublo[2] + EPSILON) ) zoverlap = 0;
+ }
+
+ int overlap = 0;
+ if (xoverlap && yoverlap && zoverlap) overlap = 1;
+
+ // if no overlap, test if bounding box wrapped back into new system
+
+ if (!overlap) {
+
+ // wrap back into cell
+
+ imageint imagelo = ((imageint) IMGMAX << IMG2BITS) |
((imageint) IMGMAX << IMGBITS) | IMGMAX;
- if (triclinic == 0) {
- x[0] = buf[m+1] + ix*old_xprd;
- x[1] = buf[m+2] + iy*old_yprd;
- x[2] = buf[m+3] + iz*old_zprd;
- } else {
- x[0] = buf[m+1] + ix*old_xprd + iy*old_xy + iz*old_xz;
- x[1] = buf[m+2] + iy*old_yprd + iz*old_yz;
- x[2] = buf[m+3] + iz*old_zprd;
- }
- domain->remap(x,image);
+ domain->remap(&(_lo[0]), imagelo);
+ int xboxlo = (imagelo & IMGMASK) - IMGMAX;
+ int yboxlo = (imagelo >> IMGBITS & IMGMASK) - IMGMAX;
+ int zboxlo = (imagelo >> IMG2BITS) - IMGMAX;
+
+ imageint imagehi = ((imageint) IMGMAX << IMG2BITS) |
+ ((imageint) IMGMAX << IMGBITS) | IMGMAX;
+ domain->remap(&(_hi[0]), imagehi);
+ int xboxhi = (imagehi & IMGMASK) - IMGMAX;
+ int yboxhi = (imagehi >> IMGBITS & IMGMASK) - IMGMAX;
+ int zboxhi = (imagehi >> IMG2BITS) - IMGMAX;
+
if (triclinic) {
- domain->x2lamda(x,lamda);
- coord = lamda;
- } else coord = x;
-
- if (coord[0] >= sublo[0] && coord[0] < subhi[0] &&
- coord[1] >= sublo[1] && coord[1] < subhi[1] &&
- coord[2] >= sublo[2] && coord[2] < subhi[2]) {
-
- m += avec->unpack_restart(&buf[m]);
-
- i = atom->nlocal - 1;
- if (tag_enable)
- atom_offset = iz*ny*nx*maxtag + iy*nx*maxtag + ix*maxtag;
- else atom_offset = 0;
- mol_offset = iz*ny*nx*maxmol + iy*nx*maxmol + ix*maxmol;
-
- atom->x[i][0] = x[0];
- atom->x[i][1] = x[1];
- atom->x[i][2] = x[2];
-
- atom->tag[i] += atom_offset;
- atom->image[i] = image;
-
- if (atom->molecular) {
- if (atom->molecule[i] > 0)
- atom->molecule[i] += mol_offset;
- if (atom->molecular == 1) {
- if (atom->avec->bonds_allow)
- for (j = 0; j < atom->num_bond[i]; j++)
- atom->bond_atom[i][j] += atom_offset;
- if (atom->avec->angles_allow)
- for (j = 0; j < atom->num_angle[i]; j++) {
- atom->angle_atom1[i][j] += atom_offset;
- atom->angle_atom2[i][j] += atom_offset;
- atom->angle_atom3[i][j] += atom_offset;
- }
- if (atom->avec->dihedrals_allow)
- for (j = 0; j < atom->num_dihedral[i]; j++) {
- atom->dihedral_atom1[i][j] += atom_offset;
- atom->dihedral_atom2[i][j] += atom_offset;
- atom->dihedral_atom3[i][j] += atom_offset;
- atom->dihedral_atom4[i][j] += atom_offset;
- }
- if (atom->avec->impropers_allow)
- for (j = 0; j < atom->num_improper[i]; j++) {
- atom->improper_atom1[i][j] += atom_offset;
- atom->improper_atom2[i][j] += atom_offset;
- atom->improper_atom3[i][j] += atom_offset;
- atom->improper_atom4[i][j] += atom_offset;
- }
+ double _llo[3];
+ _llo[0] = _lo[0]; _llo[1] = _lo[1]; _llo[2] = _lo[2];
+ domain->x2lamda(_llo,_lo);
+
+ double _lhi[3];
+ _lhi[0] = _hi[0]; _lhi[1] = _hi[1]; _lhi[2] = _hi[2];
+ domain->x2lamda(_lhi,_hi);
+ }
+
+ // test all fragments for any overlap; ok to include false positives
+
+ int _xoverlap1 = 0;
+ int _xoverlap2 = 0;
+ if (!xoverlap) {
+ if (xboxlo < 0) {
+ _xoverlap1 = 1;
+ if ( _lo[0] > (subhi[0] - EPSILON) ) _xoverlap1 = 0;
+ }
+
+ if (xboxhi > 0) {
+ _xoverlap2 = 1;
+ if ( _hi[0] < (sublo[0] + EPSILON) ) _xoverlap2 = 0;
+ }
+
+ if (_xoverlap1 || _xoverlap2) xoverlap = 1;
+ }
+
+ int _yoverlap1 = 0;
+ int _yoverlap2 = 0;
+ if (!yoverlap) {
+ if (yboxlo < 0) {
+ _yoverlap1 = 1;
+ if ( _lo[1] > (subhi[1] - EPSILON) ) _yoverlap1 = 0;
+ }
+
+ if (yboxhi > 0) {
+ _yoverlap2 = 1;
+ if ( _hi[1] < (sublo[1] + EPSILON) ) _yoverlap2 = 0;
+ }
+
+ if (_yoverlap1 || _yoverlap2) yoverlap = 1;
+ }
+
+
+ int _zoverlap1 = 0;
+ int _zoverlap2 = 0;
+ if (!zoverlap) {
+ if (zboxlo < 0) {
+ _zoverlap1 = 1;
+ if ( _lo[2] > (subhi[2] - EPSILON) ) _zoverlap1 = 0;
+ }
+
+ if (zboxhi > 0) {
+ _zoverlap2 = 1;
+ if ( _hi[2] < (sublo[2] + EPSILON) ) _zoverlap2 = 0;
+ }
+
+ if (_zoverlap1 || _zoverlap2) zoverlap = 1;
+ }
+
+ // does either fragment overlap w/ sub-domain
+
+ if (xoverlap && yoverlap && zoverlap) overlap = 1;
+ }
+
+ // while loop over one proc's atom list
+
+ if (overlap) {
+ num_replicas_added++;
+
+ m = 0;
+ while (m < size_buf_all) {
+ image = ((imageint) IMGMAX << IMG2BITS) |
+ ((imageint) IMGMAX << IMGBITS) | IMGMAX;
+ if (triclinic == 0) {
+ x[0] = buf_all[m+1] + ix*old_xprd;
+ x[1] = buf_all[m+2] + iy*old_yprd;
+ x[2] = buf_all[m+3] + iz*old_zprd;
+ } else {
+ x[0] = buf_all[m+1] + ix*old_xprd + iy*old_xy + iz*old_xz;
+ x[1] = buf_all[m+2] + iy*old_yprd + iz*old_yz;
+ x[2] = buf_all[m+3] + iz*old_zprd;
+ }
+ domain->remap(x,image);
+ if (triclinic) {
+ domain->x2lamda(x,lamda);
+ coord = lamda;
+ } else coord = x;
+
+ if (coord[0] >= sublo[0] && coord[0] < subhi[0] &&
+ coord[1] >= sublo[1] && coord[1] < subhi[1] &&
+ coord[2] >= sublo[2] && coord[2] < subhi[2]) {
+
+ m += avec->unpack_restart(&buf_all[m]);
+
+ i = atom->nlocal - 1;
+ if (tag_enable)
+ atom_offset = iz*ny*nx*maxtag + iy*nx*maxtag + ix*maxtag;
+ else atom_offset = 0;
+ mol_offset = iz*ny*nx*maxmol + iy*nx*maxmol + ix*maxmol;
+
+ atom->x[i][0] = x[0];
+ atom->x[i][1] = x[1];
+ atom->x[i][2] = x[2];
+
+ atom->tag[i] += atom_offset;
+ atom->image[i] = image;
+
+ if (atom->molecular) {
+ if (atom->molecule[i] > 0)
+ atom->molecule[i] += mol_offset;
+ if (atom->molecular == 1) {
+ if (atom->avec->bonds_allow)
+ for (j = 0; j < atom->num_bond[i]; j++)
+ atom->bond_atom[i][j] += atom_offset;
+ if (atom->avec->angles_allow)
+ for (j = 0; j < atom->num_angle[i]; j++) {
+ atom->angle_atom1[i][j] += atom_offset;
+ atom->angle_atom2[i][j] += atom_offset;
+ atom->angle_atom3[i][j] += atom_offset;
+ }
+ if (atom->avec->dihedrals_allow)
+ for (j = 0; j < atom->num_dihedral[i]; j++) {
+ atom->dihedral_atom1[i][j] += atom_offset;
+ atom->dihedral_atom2[i][j] += atom_offset;
+ atom->dihedral_atom3[i][j] += atom_offset;
+ atom->dihedral_atom4[i][j] += atom_offset;
+ }
+ if (atom->avec->impropers_allow)
+ for (j = 0; j < atom->num_improper[i]; j++) {
+ atom->improper_atom1[i][j] += atom_offset;
+ atom->improper_atom2[i][j] += atom_offset;
+ atom->improper_atom3[i][j] += atom_offset;
+ atom->improper_atom4[i][j] += atom_offset;
+ }
+ }
}
+ } else m += static_cast<int> (buf_all[m]);
+ }
+ } // if (overlap)
+
+ }
+ }
+ }
+
+ memory->destroy(size_buf_rnk);
+ memory->destroy(disp_buf_rnk);
+ memory->destroy(buf_all);
+
+ int sum = 0;
+ MPI_Reduce(&num_replicas_added, &sum, 1, MPI_INT, MPI_SUM, 0, world);
+ double avg = (double) sum / nprocs;
+ if (me == 0 && screen)
+ fprintf(screen," average # of replicas added to proc = %.2f out of %i (%.2f %%)\n",
+ avg,nx*ny*nz,avg/(nx*ny*nz)*100.0);
+
+ } else {
+
+ for (int iproc = 0; iproc < nprocs; iproc++) {
+ if (me == iproc) {
+ n = 0;
+ for (i = 0; i < old->nlocal; i++) n += old_avec->pack_restart(i,&buf[n]);
+ }
+ MPI_Bcast(&n,1,MPI_INT,iproc,world);
+ MPI_Bcast(buf,n,MPI_DOUBLE,iproc,world);
+
+ for (ix = 0; ix < nx; ix++) {
+ for (iy = 0; iy < ny; iy++) {
+ for (iz = 0; iz < nz; iz++) {
+
+ // while loop over one proc's atom list
+
+ m = 0;
+ while (m < n) {
+ image = ((imageint) IMGMAX << IMG2BITS) |
+ ((imageint) IMGMAX << IMGBITS) | IMGMAX;
+ if (triclinic == 0) {
+ x[0] = buf[m+1] + ix*old_xprd;
+ x[1] = buf[m+2] + iy*old_yprd;
+ x[2] = buf[m+3] + iz*old_zprd;
+ } else {
+ x[0] = buf[m+1] + ix*old_xprd + iy*old_xy + iz*old_xz;
+ x[1] = buf[m+2] + iy*old_yprd + iz*old_yz;
+ x[2] = buf[m+3] + iz*old_zprd;
}
- } else m += static_cast<int> (buf[m]);
+ domain->remap(x,image);
+ if (triclinic) {
+ domain->x2lamda(x,lamda);
+ coord = lamda;
+ } else coord = x;
+
+ if (coord[0] >= sublo[0] && coord[0] < subhi[0] &&
+ coord[1] >= sublo[1] && coord[1] < subhi[1] &&
+ coord[2] >= sublo[2] && coord[2] < subhi[2]) {
+
+ m += avec->unpack_restart(&buf[m]);
+
+ i = atom->nlocal - 1;
+ if (tag_enable)
+ atom_offset = iz*ny*nx*maxtag + iy*nx*maxtag + ix*maxtag;
+ else atom_offset = 0;
+ mol_offset = iz*ny*nx*maxmol + iy*nx*maxmol + ix*maxmol;
+
+ atom->x[i][0] = x[0];
+ atom->x[i][1] = x[1];
+ atom->x[i][2] = x[2];
+
+ atom->tag[i] += atom_offset;
+ atom->image[i] = image;
+
+ if (atom->molecular) {
+ if (atom->molecule[i] > 0)
+ atom->molecule[i] += mol_offset;
+ if (atom->molecular == 1) {
+ if (atom->avec->bonds_allow)
+ for (j = 0; j < atom->num_bond[i]; j++)
+ atom->bond_atom[i][j] += atom_offset;
+ if (atom->avec->angles_allow)
+ for (j = 0; j < atom->num_angle[i]; j++) {
+ atom->angle_atom1[i][j] += atom_offset;
+ atom->angle_atom2[i][j] += atom_offset;
+ atom->angle_atom3[i][j] += atom_offset;
+ }
+ if (atom->avec->dihedrals_allow)
+ for (j = 0; j < atom->num_dihedral[i]; j++) {
+ atom->dihedral_atom1[i][j] += atom_offset;
+ atom->dihedral_atom2[i][j] += atom_offset;
+ atom->dihedral_atom3[i][j] += atom_offset;
+ atom->dihedral_atom4[i][j] += atom_offset;
+ }
+ if (atom->avec->impropers_allow)
+ for (j = 0; j < atom->num_improper[i]; j++) {
+ atom->improper_atom1[i][j] += atom_offset;
+ atom->improper_atom2[i][j] += atom_offset;
+ atom->improper_atom3[i][j] += atom_offset;
+ atom->improper_atom4[i][j] += atom_offset;
+ }
+ }
+ }
+ } else m += static_cast<int> (buf[m]);
+ }
}
}
}
}
- }
+ } // if (bbox_flag)
// free communication buffer and old atom class
--
GitLab