Write a simple 1D-array example filled with one variable

examples/1D-Array/Makefile

+MPICXX = mpixlcxx
+
+MPI_CFLAGS  = -g -I$(HOME)/install_bgq/include/ -DBGQ -DDEBUG -DTIMING
+MPI_LDFLAGS = -L$(HOME)/install_bgq/lib/ -ltapioca
+
+all: main_tapioca
+
+main_tapioca:main_tapioca.cpp
+	$(MPICXX) $(MPI_CFLAGS) $? -o $@ $(MPI_LDFLAGS)
+
+clean:
+	rm main_tapioca

examples/1D-Array/Makefile.bgq

+MPICXX = mpixlcxx
+
+MPI_CFLAGS  = -g -I$(HOME)/install_bgq/include/ -DBGQ -DDEBUG -DTIMING
+MPI_LDFLAGS = -L$(HOME)/install_bgq/lib/ -ltapioca
+
+all: simple_io_tapioca simple_io_mpiio
+
+simple_io_tapioca:simple_io_tapioca.cpp
+	$(MPICXX) $(MPI_CFLAGS) $? -o $@ $(MPI_LDFLAGS)
+
+simple_io_mpiio:simple_io_mpiio.cpp
+	$(MPICXX) $(MPI_CFLAGS) $? -o $@ $(MPI_LDFLAGS)
+
+clean:
+	rm simple_io_tapioca simple_io_mpiio
+

examples/1D-Array/Makefile.xc40

+MPICXX = CC
+
+MPI_CFLAGS  = -g -I$(HOME)/install/include/ -DXC40 -DDEBUG -DTIMING
+MPI_LDFLAGS = -L$(HOME)/install/lib/ -ltapioca -llustreapi -Wl,--whole-archive,-ldmapp,--no-whole-archive
+
+all: simple_io_tapioca simple_io_mpiio
+
+simple_io_tapioca:simple_io_tapioca.cpp
+	$(MPICXX) $(MPI_CFLAGS) $? -o $@ $(MPI_LDFLAGS)
+
+simple_io_mpiio:simple_io_mpiio.cpp
+	$(MPICXX) $(MPI_CFLAGS) $? -o $@ $(MPI_LDFLAGS)
+
+clean:
+	rm simple_io_tapioca simple_io_mpiio
+

examples/1D-Array/main_tapioca.cpp

+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <stdint.h>
+#include <mpi.h>
+
+#include "tapioca.hpp"
+
+#define RED     "\x1b[31m"
+#define GREEN   "\x1b[32m"
+#define BLUE    "\x1b[34m"
+#define RESET   "\x1b[0m"
+
+int64_t particles (int64_t max, int rank) {
+  int64_t n_part;
+
+  srand(time(NULL) + rank);
+  n_part = (rand () % max) + 10;
+
+  return n_part;
+}
+
+int main (int argc, char * argv[])
+{
+  int world_numtasks, world_myrank, mycolor, mykey, sub_numtasks, sub_myrank, i, file_id;
+  int64_t num_particles;
+  int64_t sub_particles, tot_particles, particle_size, file_size, tot_size;
+  int64_t scan_size = 0, offset, hdr = 0;
+  double start_time, end_time, tot_time, max_time;
+  double io_bw;
+  MPI_Comm sub_comm;
+  MPI_File file_handle;
+  MPI_Status status;
+  char output[100];
+  Tapioca tp, tp2;
+  int64_t chunkCount[1], chunkOffset[1];
+  int chunkSize[1];
+  
+  MPI_Init(&argc, &argv);
+  MPI_Comm_size(MPI_COMM_WORLD, &world_numtasks);
+  MPI_Comm_rank(MPI_COMM_WORLD, &world_myrank);
+
+  //mycolor = tp.topology.BridgeNodeId ();
+  mycolor = 42;
+  mykey = world_myrank;
+
+  MPI_Comm_split (MPI_COMM_WORLD, mycolor, mykey, &sub_comm);
+  MPI_Comm_size(sub_comm, &sub_numtasks);
+  MPI_Comm_rank(sub_comm, &sub_myrank);
+
+  if ( argv[2] != NULL )
+    num_particles = particles (atoi ( argv[1] ), world_myrank);
+  else
+    num_particles = atoi ( argv[1] );
+
+#ifdef BGQ
+  snprintf (output, 100, "/projects/visualization/ftessier/debug/SimpleIO-%08d.dat", mycolor);
+#elif XC40
+  snprintf (output, 100, "/lus/theta-fs0/projects/Performance/ftessier/HACC/SimpleIO-%08d.dat", mycolor);
+#else
+  snprintf (output, 100, "./SimpleIO-%08d.dat", mycolor);
+#endif
+
+
+  /*****************/
+  /*     WRITE     */
+  /*****************/
+  float *xx;
+  xx = new float[num_particles];
+
+  for (uint64_t i = 0; i< num_particles; i++)
+    xx[i] = (float)i;
+  
+  MPI_Allreduce(&num_particles, &sub_particles, 1, MPI_LONG_LONG, MPI_SUM, sub_comm);
+  MPI_Allreduce(&num_particles, &tot_particles, 1, MPI_LONG_LONG, MPI_SUM, MPI_COMM_WORLD);
+  
+  particle_size = sizeof(float);
+  file_size     = particle_size * sub_particles;
+  tot_size      = particle_size * tot_particles;
+
+  if (sub_myrank == 0) {
+    MPI_File_open(MPI_COMM_SELF, output,
+		  MPI_MODE_WRONLY | MPI_MODE_CREATE, MPI_INFO_NULL, &file_handle);
+    MPI_File_set_size(file_handle, file_size);
+    MPI_File_close (&file_handle);
+  }
+
+  MPI_Exscan (&num_particles, &scan_size, 1, MPI_LONG_LONG, MPI_SUM, sub_comm);
+    
+  if (0 == sub_myrank) {
+    fprintf (stdout, GREEN "[INFO]" RESET " [%08d] TAPIOCA Simple I/O\n", mycolor);
+    fprintf (stdout, GREEN "[INFO]" RESET " [%08d] Write output file\n", mycolor);
+    fprintf (stdout, GREEN "[INFO]" RESET " [%08d] --> %lld particles per rank\n", mycolor, num_particles);
+    fprintf (stdout, GREEN "[INFO]" RESET " [%08d] --> File size: %.2f MB (%lld particles)\n", 
+	     mycolor, (double)file_size/(1024*1024), sub_particles);
+  }
+
+  /*****************/
+  /* INIT TAPIOCA  */
+  /*****************/
+  chunkCount[0]  = num_particles;
+  chunkSize[0]   = sizeof(float);
+  chunkOffset[0] = hdr + scan_size * particle_size;
+  
+  tp.WriteInitialize (output, chunkCount, chunkSize, chunkOffset, 1, hdr, ARRAY_OF_STRUCTURES, sub_comm);
+  /*****************/
+  
+  MPI_Barrier (MPI_COMM_WORLD);
+  start_time = MPI_Wtime();
+
+  MPI_File_open(sub_comm, output,
+		MPI_MODE_WRONLY, MPI_INFO_NULL, &file_handle);
+  
+  offset = scan_size * particle_size;
+
+  tp.Write (file_handle, offset, xx, num_particles, MPI_FLOAT, &status);
+  
+  MPI_File_close (&file_handle);
+
+  end_time = MPI_Wtime();
+  tot_time = end_time - start_time;
+  MPI_Reduce (&tot_time, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
+  
+  if (0 == world_myrank) {
+    io_bw = (double)tot_size / max_time / (1024 * 1024);
+    fprintf (stdout, BLUE "[TIMING]" RESET " Write I/O bandwidth: %.2f MBps (%.2f MB in %.2f ms)\n",
+	     io_bw, (double)tot_size/(1024*1024), max_time * 1000);
+  }
+
+  MPI_Barrier (MPI_COMM_WORLD);
+
+  /*****************/
+  /*     READ      */
+  /*****************/
+  float *xx_r;
+  xx_r = new float[num_particles];
+
+  /*****************/
+  /* INIT TAPIOCA  */
+  /*****************/
+  tp2.ReadInitialize (output, chunkCount, chunkSize, chunkOffset, 1, hdr, ARRAY_OF_STRUCTURES, sub_comm);
+  /*****************/
+
+  start_time = MPI_Wtime();
+
+  MPI_File_open(sub_comm, output,
+		MPI_MODE_RDONLY, MPI_INFO_NULL, &file_handle);
+
+  if (0 == sub_myrank)
+    fprintf (stdout, GREEN "[INFO]" RESET " [%08d] Read output file\n", mycolor);
+
+  offset = scan_size * particle_size;
+
+  tp2.Read (file_handle, offset, xx_r, num_particles, MPI_FLOAT, &status);
+  
+  MPI_File_close (&file_handle);
+
+  end_time = MPI_Wtime();
+  tot_time = end_time - start_time;
+  MPI_Reduce (&tot_time, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
+  
+  if (0 == world_myrank) {
+    io_bw = (double)tot_size / max_time / (1024 * 1024);
+    fprintf (stdout, BLUE "[TIMING]" RESET " Read I/O bandwidth: %.2f MBps (%.2f MB in %.2f ms)\n",
+	     io_bw, (double)tot_size/(1024*1024), max_time * 1000);
+  }
+
+  /*****************/
+  /* VERIFICATION  */
+  /*****************/
+  for (uint64_t i = 0; i< num_particles; i++) {
+    if (xx[i] != xx_r[i])
+      {
+	fprintf (stdout, RED "[ERROR]" RESET " Wrong value for particle %d\n", i);
+	MPI_Abort (MPI_COMM_WORLD, -1);
+      }
+  }
+
+  if (0 == sub_myrank)
+    fprintf (stdout, GREEN "[INFO]" RESET " [%08d] Content verified and consistent\n", mycolor);
+
+  /*****************/
+  /*      FREE     */
+  /*****************/
+  delete [] xx;
+  delete [] xx_r;
+  
+  MPI_Finalize ();
+}
+

examples/1D-Array/run_bgq.sh

+#!/bin/bash
+
+NODES=4
+PPN=2
+NPROCS=$((NODES*PPN))
+TARGET="/projects/visualization/ftessier/debug"
+
+cd $HOME/TAPIOCA/examples/1D-Array
+
+export TAPIOCA_DEVNULL=false
+export TAPIOCA_COMMSPLIT=true
+export TAPIOCA_STRATEGY=TOPOLOGY_AWARE
+export TAPIOCA_NBAGGR=1
+export TAPIOCA_BUFFERSIZE=1048576
+export TAPIOCA_PIPELINING=true
+
+function updateSettings()
+{
+    printenv | egrep "TAPIOCA_"
+    SETTINGS="TAPIOCA_DEVNULL=$TAPIOCA_DEVNULL"
+    SETTINGS="$SETTINGS TAPIOCA_COMMSPLIT=$TAPIOCA_COMMSPLIT"
+    SETTINGS="$SETTINGS TAPIOCA_STRATEGY=$TAPIOCA_STRATEGY"
+    SETTINGS="$SETTINGS TAPIOCA_NBAGGR=$TAPIOCA_NBAGGR"
+    SETTINGS="$SETTINGS TAPIOCA_BUFFERSIZE=$TAPIOCA_BUFFERSIZE"
+    SETTINGS="$SETTINGS TAPIOCA_PIPELINING=$TAPIOCA_PIPELINING"
+    
+    SETTINGS="$SETTINGS PAMID_VERBOSE=1"
+    SETTINGS="$SETTINGS PAMID_COLLECTIVES_MEMORY_OPTIMIZED=1"
+    SETTINGS="$SETTINGS PAMID_COLLECTIVES=1"
+    SETTINGS="$SETTINGS BG_SHAREDMEMSIZE=64"
+    SETTINGS="$SETTINGS BGLOCKLESSMPIO_F_TYPE=0x47504653"
+    SETTINGS="$SETTINGS GPFSMPIO_BALANCECONTIG=1"
+    SETTINGS="$SETTINGS GPFSMPIO_P2PCONTIG=1"
+}
+
+function outputFile ()
+{
+    OUTPUT="$HOME/TAPIOCA/examples/1D-Array/$1"
+    OUTPUT="${OUTPUT}_${NODES}_${PPN}_${PARTICLES}_${TAPIOCA_NBAGGR}_${TAPIOCA_BUFFERSIZE}_BGQ"
+}
+
+
+PARTICLES=50000
+rm $TARGET/*
+updateSettings
+outputFile "TAPIOCA"
+runjob --block $COBALT_PARTNAME --envs $SETTINGS -p $PPN --np $NPROCS : ./main_tapioca $PARTICLES > $OUTPUT