/* ---------------------------------------------------------------------- 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 Coffman (IBM) ------------------------------------------------------------------------- */ #include "math.h" #include "stdlib.h" #include "string.h" #include "dump_cfg_mpiio.h" #include "atom.h" #include "domain.h" #include "comm.h" #include "modify.h" #include "compute.h" #include "input.h" #include "fix.h" #include "variable.h" #include "update.h" #include "memory.h" #include "error.h" #ifdef LMP_USER_IO_TIMER #include #include #include long dumpCFGTimestamps[10]; #endif #if defined(_OPENMP) #include #endif using namespace LAMMPS_NS; #define MAX_TEXT_HEADER_SIZE 4096 #define DUMP_BUF_CHUNK_SIZE 16384 #define DUMP_BUF_INCREMENT_SIZE 4096 enum{INT,DOUBLE,STRING,BIGINT}; // same as in DumpCustom #define UNWRAPEXPAND 10.0 #define ONEFIELD 32 #define DELTA 1048576 /* ---------------------------------------------------------------------- */ DumpCFGMPIIO::DumpCFGMPIIO(LAMMPS *lmp, int narg, char **arg) : DumpCFG(lmp, narg, arg) {} /* ---------------------------------------------------------------------- */ DumpCFGMPIIO::~DumpCFGMPIIO() { if (multifile == 0) MPI_File_close(&mpifh); } /* ---------------------------------------------------------------------- */ void DumpCFGMPIIO::openfile() { if (singlefile_opened) { // single file already opened, so just return after resetting filesize mpifo = currentFileSize; MPI_File_set_size(mpifh,mpifo+headerSize+sumFileSize); currentFileSize = mpifo+headerSize+sumFileSize; return; } if (multifile == 0) singlefile_opened = 1; // if one file per timestep, replace '*' with current timestep filecurrent = filename; if (multifile) { char *filestar = filecurrent; filecurrent = new char[strlen(filestar) + 16]; char *ptr = strchr(filestar,'*'); *ptr = '\0'; if (padflag == 0) sprintf(filecurrent,"%s" BIGINT_FORMAT "%s", filestar,update->ntimestep,ptr+1); else { char bif[8],pad[16]; strcpy(bif,BIGINT_FORMAT); sprintf(pad,"%%s%%0%d%s%%s",padflag,&bif[1]); sprintf(filecurrent,pad,filestar,update->ntimestep,ptr+1); } *ptr = '*'; } if (append_flag) { // append open int err = MPI_File_open( world, filecurrent, MPI_MODE_CREATE | MPI_MODE_APPEND | MPI_MODE_WRONLY , MPI_INFO_NULL, &mpifh); if (err != MPI_SUCCESS) { char str[128]; sprintf(str,"Cannot open dump file %s",filecurrent); error->one(FLERR,str); } int myrank; MPI_Comm_rank(world,&myrank); if (myrank == 0) MPI_File_get_size(mpifh,&mpifo); MPI_Bcast(&mpifo, 1, MPI_LMP_BIGINT, 0, world); MPI_File_set_size(mpifh,mpifo+headerSize+sumFileSize); currentFileSize = mpifo+headerSize+sumFileSize; } else { // replace open int err = MPI_File_open( world, filecurrent, MPI_MODE_CREATE | MPI_MODE_WRONLY , MPI_INFO_NULL, &mpifh); if (err != MPI_SUCCESS) { char str[128]; sprintf(str,"Cannot open dump file %s",filecurrent); error->one(FLERR,str); } mpifo = 0; MPI_File_set_size(mpifh,(MPI_Offset) (headerSize+sumFileSize)); currentFileSize = (headerSize+sumFileSize); } } /* ---------------------------------------------------------------------- */ void DumpCFGMPIIO::write() { #ifdef LMP_USER_IO_TIMER long startTimeBase, endTimeBase; MPI_Barrier(world); // timestamp barrier if (me == 0) startTimeBase = GetTimeBase(); #endif if (domain->triclinic == 0) { boxxlo = domain->boxlo[0]; boxxhi = domain->boxhi[0]; boxylo = domain->boxlo[1]; boxyhi = domain->boxhi[1]; boxzlo = domain->boxlo[2]; boxzhi = domain->boxhi[2]; } else { boxxlo = domain->boxlo_bound[0]; boxxhi = domain->boxhi_bound[0]; boxylo = domain->boxlo_bound[1]; boxyhi = domain->boxhi_bound[1]; boxzlo = domain->boxlo_bound[2]; boxzhi = domain->boxhi_bound[2]; boxxy = domain->xy; boxxz = domain->xz; boxyz = domain->yz; } // nme = # of dump lines this proc contributes to dump nme = count(); // ntotal = total # of dump lines in snapshot // nmax = max # of dump lines on any proc bigint bnme = nme; MPI_Allreduce(&bnme,&ntotal,1,MPI_LMP_BIGINT,MPI_SUM,world); int nmax; MPI_Allreduce(&nme,&nmax,1,MPI_INT,MPI_MAX,world); // write timestep header // for multiproc, // nheader = # of lines in this file via Allreduce on clustercomm bigint nheader = ntotal; // insure filewriter proc can receive everyone's info // limit nmax*size_one to int since used as arg in MPI_Rsend() below // pack my data into buf // if sorting on IDs also request ID list from pack() // sort buf as needed if (nmax > maxbuf) { if ((bigint) nmax * size_one > MAXSMALLINT) error->all(FLERR,"Too much per-proc info for dump"); maxbuf = nmax; memory->destroy(buf); memory->create(buf,(maxbuf*size_one),"dump:buf"); } if (sort_flag && sortcol == 0 && nmax > maxids) { maxids = nmax; memory->destroy(ids); memory->create(ids,maxids,"dump:ids"); } if (sort_flag && sortcol == 0) pack(ids); else pack(NULL); if (sort_flag) sort(); // determine how much data needs to be written for setting the file size and prepocess it prior to writing performEstimate = 1; write_header(nheader); write_data(nme,buf); MPI_Bcast(&sumFileSize, 1, MPI_LMP_BIGINT, (nprocs-1), world); #ifdef LMP_USER_IO_TIMER MPI_Barrier(world); // timestamp barrier dumpCFGTimestamps[0] = GetTimeBase(); #endif openfile(); #ifdef LMP_USER_IO_TIMER MPI_Barrier(world); // timestamp barrier dumpCFGTimestamps[1] = GetTimeBase(); #endif performEstimate = 0; write_header(nheader); // mpifo now points to end of header info #ifdef LMP_USER_IO_TIMER MPI_Barrier(world); // timestamp barrier dumpCFGTimestamps[2] = GetTimeBase(); #endif // now actually write the data performEstimate = 0; write_data(nme,buf); #ifdef LMP_USER_IO_TIMER MPI_Barrier(world); // timestamp barrier dumpCFGTimestamps[3] = GetTimeBase(); #endif if (multifile) MPI_File_close(&mpifh); if (multifile) delete [] filecurrent; #ifdef LMP_USER_IO_TIMER MPI_Barrier(world); // timestamp barrier dumpCFGTimestamps[4] = GetTimeBase(); if (me == 0) { endTimeBase = GetTimeBase(); printf("total dump cycles: %ld - estimates and setup: %ld openfile: %ld write header: %ld write data: %ld close file: %ld\n",(long) (endTimeBase-startTimeBase),(long) (dumpCFGTimestamps[0]-startTimeBase),(long) (dumpCFGTimestamps[1]-dumpCFGTimestamps[0]),(long) (dumpCFGTimestamps[2]-dumpCFGTimestamps[1]),(long) (dumpCFGTimestamps[3]-dumpCFGTimestamps[2]),(long) (dumpCFGTimestamps[4]-dumpCFGTimestamps[3])); } #endif } /* ---------------------------------------------------------------------- */ void DumpCFGMPIIO::init_style() { if (multifile == 0 && !multifile_override) error->all(FLERR,"Dump cfg requires one snapshot per file"); DumpCustom::init_style(); // setup function ptrs write_choice = &DumpCFGMPIIO::write_string; } /* ---------------------------------------------------------------------- */ void DumpCFGMPIIO::write_header(bigint n) { // set scale factor used by AtomEye for CFG viz // default = 1.0 // for peridynamics, set to pre-computed PD scale factor // so PD particles mimic C atoms // for unwrapped coords, set to UNWRAPEXPAND (10.0) // so molecules are not split across periodic box boundaries if (performEstimate) { headerBuffer = (char *) malloc(MAX_TEXT_HEADER_SIZE); headerSize = 0; double scale = 1.0; if (atom->peri_flag) scale = atom->pdscale; else if (unwrapflag == 1) scale = UNWRAPEXPAND; char str[64]; sprintf(str,"Number of particles = %s\n",BIGINT_FORMAT); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),str,n); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"A = %g Angstrom (basic length-scale)\n",scale); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(1,1) = %g A\n",domain->xprd); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(1,2) = 0 A \n"); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(1,3) = 0 A \n"); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(2,1) = %g A \n",domain->xy); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(2,2) = %g A\n",domain->yprd); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(2,3) = 0 A \n"); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(3,1) = %g A \n",domain->xz); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(3,2) = %g A \n",domain->yz); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"H0(3,3) = %g A\n",domain->zprd); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),".NO_VELOCITY.\n"); headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"entry_count = %d\n",nfield-2); for (int i = 0; i < nfield-5; i++) headerSize += sprintf(((char*)&((char*)headerBuffer)[headerSize]),"auxiliary[%d] = %s\n",i,auxname[i]); } else { // write data if (me == 0) MPI_File_write_at(mpifh,mpifo,headerBuffer,headerSize,MPI_CHAR,MPI_STATUS_IGNORE); mpifo += headerSize; free(headerBuffer); } } #if defined(_OPENMP) /* ---------------------------------------------------------------------- convert mybuf of doubles to one big formatted string in sbuf return -1 if strlen exceeds an int, since used as arg in MPI calls in Dump ------------------------------------------------------------------------- */ int DumpCFGMPIIO::convert_string_omp(int n, double *mybuf) { char **mpifh_buffer_line_per_thread; int mpifhStringCount; int *mpifhStringCountPerThread, *bufOffset, *bufRange, *bufLength; mpifhStringCount = 0; int nthreads = omp_get_max_threads(); if (nthreads > n) { // call serial version convert_string(n,mybuf); } else { memory->create(mpifhStringCountPerThread,nthreads,"dump:mpifhStringCountPerThread"); mpifh_buffer_line_per_thread = (char **) malloc(nthreads*sizeof(char*)); memory->create(bufOffset,nthreads,"dump:bufOffset"); memory->create(bufRange,nthreads,"dump:bufRange"); memory->create(bufLength,nthreads,"dump:bufLength"); int i=0; for (i=0;i<(nthreads-1);i++) { mpifhStringCountPerThread[i] = 0; bufOffset[i] = (int) (i*(int)(floor((double)n/(double)nthreads))*size_one); bufRange[i] = (int)(floor((double)n/(double)nthreads)); bufLength[i] = DUMP_BUF_CHUNK_SIZE; mpifh_buffer_line_per_thread[i] = (char *) malloc(DUMP_BUF_CHUNK_SIZE * sizeof(char)); mpifh_buffer_line_per_thread[i][0] = '\0'; } mpifhStringCountPerThread[i] = 0; bufOffset[i] = (int) (i*(int)(floor((double)n/(double)nthreads))*size_one); bufRange[i] = n-(i*(int)(floor((double)n/(double)nthreads))); bufLength[i] = DUMP_BUF_CHUNK_SIZE; mpifh_buffer_line_per_thread[i] = (char *) malloc(DUMP_BUF_CHUNK_SIZE * sizeof(char)); mpifh_buffer_line_per_thread[i][0] = '\0'; #pragma omp parallel default(none) shared(bufOffset, bufRange, bufLength, mpifhStringCountPerThread, mpifh_buffer_line_per_thread, mybuf) { int tid = omp_get_thread_num(); int m=0; if (unwrapflag == 0) { for (int i = 0; i < bufRange[tid]; i++) { if ((bufLength[tid] - mpifhStringCountPerThread[tid]) < DUMP_BUF_INCREMENT_SIZE) { mpifh_buffer_line_per_thread[tid] = (char *) realloc(mpifh_buffer_line_per_thread[tid],(mpifhStringCountPerThread[tid]+DUMP_BUF_CHUNK_SIZE) * sizeof(char)); bufLength[tid] = (mpifhStringCountPerThread[tid]+DUMP_BUF_CHUNK_SIZE) * sizeof(char); } for (int j = 0; j < size_one; j++) { if (j == 0) { mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),"%f \n",(mybuf[bufOffset[tid]+m])); } else if (j == 1) { mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),"%s \n",typenames[(int) mybuf[bufOffset[tid]+m]]); } else if (j >= 2) { if (vtype[j] == INT) mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],static_cast (mybuf[bufOffset[tid]+m])); else if (vtype[j] == DOUBLE) mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],mybuf[bufOffset[tid]+m]); else if (vtype[j] == STRING) mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],typenames[(int) mybuf[bufOffset[tid]+m]]); else if (vtype[j] == BIGINT) mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],static_cast (mybuf[bufOffset[tid]+m])); } m++; } // for j mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),"\n"); } // for i } // wrap flag else if (unwrapflag == 1) { for (int i = 0; i < bufRange[tid]; i++) { if ((bufLength[tid] - mpifhStringCountPerThread[tid]) < DUMP_BUF_INCREMENT_SIZE) { mpifh_buffer_line_per_thread[tid] = (char *) realloc(mpifh_buffer_line_per_thread[tid],(mpifhStringCountPerThread[tid]+DUMP_BUF_CHUNK_SIZE) * sizeof(char)); bufLength[tid] = (mpifhStringCountPerThread[tid]+DUMP_BUF_CHUNK_SIZE) * sizeof(char); } for (int j = 0; j < size_one; j++) { double unwrap_coord; if (j == 0) { //offset += sprintf(&sbuf[offset],"%f \n",mybuf[m]); mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),"%f \n",mybuf[bufOffset[tid]+m]); } else if (j == 1) { // offset += sprintf(&sbuf[offset],"%s \n",typenames[(int) mybuf[m]]); mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),"%s \n",typenames[(int) mybuf[bufOffset[tid]+m]]); } else if (j >= 2 && j <= 4) { unwrap_coord = (mybuf[bufOffset[tid]+m] - 0.5)/UNWRAPEXPAND + 0.5; //offset += sprintf(&sbuf[offset],vformat[j],unwrap_coord); mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],unwrap_coord); } else if (j >= 5 ) { if (vtype[j] == INT) //offset += // sprintf(&sbuf[offset],vformat[j],static_cast (mybuf[m])); mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],static_cast (mybuf[bufOffset[tid]+m])); else if (vtype[j] == DOUBLE) // offset += sprintf(&sbuf[offset],vformat[j],mybuf[m]); mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],mybuf[bufOffset[tid]+m]); else if (vtype[j] == STRING) // offset += // sprintf(&sbuf[offset],vformat[j],typenames[(int) mybuf[m]]); mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],typenames[(int) mybuf[bufOffset[tid]+m]]); else if (vtype[j] == BIGINT) // offset += // sprintf(&sbuf[offset],vformat[j],static_cast (mybuf[m])); mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),vformat[j],static_cast (mybuf[bufOffset[tid]+m])); } m++; } // for j mpifhStringCountPerThread[tid] += sprintf(&(mpifh_buffer_line_per_thread[tid][mpifhStringCountPerThread[tid]]),"\n"); } // for i } // unwrap flag } // pragma omp parallel #pragma omp barrier mpifhStringCount = 0; for (i=0;idestroy(bufOffset); memory->destroy(bufRange); memory->destroy(bufLength); if (mpifhStringCount > 0) { if (mpifhStringCount > maxsbuf) { if (mpifhStringCount > MAXSMALLINT) return -1; maxsbuf = mpifhStringCount; memory->grow(sbuf,maxsbuf,"dump:sbuf"); } sbuf[0] = '\0'; } for (int i=0;idestroy(mpifhStringCountPerThread); free(mpifh_buffer_line_per_thread); } // else omp return mpifhStringCount; } #endif /* ---------------------------------------------------------------------- */ void DumpCFGMPIIO::write_data(int n, double *mybuf) { (this->*write_choice)(n,mybuf); } /* ---------------------------------------------------------------------- */ void DumpCFGMPIIO::write_string(int n, double *mybuf) { if (performEstimate) { #if defined(_OPENMP) int nthreads = omp_get_max_threads(); if (nthreads > 1) nsme = convert_string_omp(n,mybuf); else nsme = convert_string(n,mybuf); #else nsme = convert_string(n,mybuf); #endif bigint incPrefix = 0; bigint bigintNsme = (bigint) nsme; MPI_Scan(&bigintNsme,&incPrefix,1,MPI_LMP_BIGINT,MPI_SUM,world); sumFileSize = (incPrefix*sizeof(char)); offsetFromHeader = ((incPrefix-bigintNsme)*sizeof(char)); } else { MPI_File_write_at_all(mpifh,mpifo+offsetFromHeader,sbuf,nsme,MPI_CHAR,MPI_STATUS_IGNORE); if (flush_flag) MPI_File_sync(mpifh); } }