/ - Diff - Bench4GPU - Forge du Centre Blaise Pascal

     //
     // Estimation of Pi using Monte Carlo exploration process
     // gcc -std=c99 -O3 -o Pi Pi.c -lm
     // Emmanuel Quemener <emmanuel.quemener@ens-lyon.fr>
     // Cecill v2 Emmanuel QUEMENER <emmanuel.quemener@gmail.com>
     // gcc -std=c99 -O3 -o Pi_MPI Pi_MPI.c -lm
     // Needed for gethostname
     #define _BSD_SOURCE
     #define _DEFAULT_SOURCE
     #include <sys/unistd.h>
     #include <math.h>
-...
     #include <limits.h>
     #include <mpi.h>
     #include <stddef.h>
     #include <sys/time.h>
     // Marsaglia RNG very simple implementation
-...
       unsigned int seed_z=362436069,seed_w=52128862;
       LENGTH iterations=ITERATIONS,inside[8192],insides,part_inside,part_iterations;
       int numtasks,rank,rc,tag=1,i;
       float pi;
       int NumberProcesses,rank,rc,tag=1,i;
       char hostname[128];
-...
         MPI_Abort(MPI_COMM_WORLD, rc);
+      }
       MPI_Comm_size(MPI_COMM_WORLD,&numtasks);
       MPI_Comm_size(MPI_COMM_WORLD,&NumberProcesses);
       MPI_Comm_rank(MPI_COMM_WORLD,&rank);
       const int nitems=2;
-...
       MPI_Type_commit(&mpi_result_type);
       if (rank==0) {
         struct timeval tv1,tv2;
         if (argc > 1) {
           iterations=(LENGTH)atoll(argv[1]);
+        }
-...
         printf ("\tMax long = %ld\n", LONG_MAX);
         printf ("\tMax long long = %lld\n\n", LLONG_MAX);
         part_iterations=((iterations%numtasks) == 0) ? iterations/numtasks:iterations/numtasks+1 ;
         part_iterations=((iterations%NumberProcesses) == 0) ? iterations/NumberProcesses:iterations/NumberProcesses+1 ;
         gettimeofday(&tv1, NULL);
         // Split part of code
         for (i=1;i<numtasks;i++) {
         for (i=1;i<NumberProcesses;i++) {
     #ifdef LONG
           rc = MPI_Send(&part_iterations, 1, MPI_LONG_LONG, i, tag,
-...
     	     hostname,rank,(long long)insides,useconds);
         // Join part of code
           for (i=1;i<numtasks;i++) {
           for (i=1;i<NumberProcesses;i++) {
     	result recv;
-...
     	insides+=inside[i];
+          }
           pi=4.*(float)insides/(float)(part_iterations*numtasks);
           float pi=4.*(float)insides/(float)(part_iterations*NumberProcesses);
           gettimeofday(&tv2, NULL);
           printf("\n\tPi=%.40f\n\twith error %.40f\n\twith %lld iterations\n\n",pi,
     	     fabs(pi-4*atan(1.))/pi,(long long)iterations);
           double elapsed=(double)((tv2.tv_sec-tv1.tv_sec) * 1000000L +
     			      (tv2.tv_usec-tv1.tv_usec))/1000000;
           double itops=(double)(part_iterations*NumberProcesses)/elapsed;
           printf("\nParallelRate %i\nElapsed Time %.2f\nItops %.0f\nLogItops %.2f\n",NumberProcesses,elapsed,itops,log10(itops));
           LENGTH total=((iterations%NumberProcesses)==0)?iterations:(iterations/NumberProcesses+1)*NumberProcesses;
           printf("Inside/Total %ld %ld\nPi estimation %f\n\n",(long int)insides,(long int)total,(4.*(float)insides/total));
+      }
       else
+        {
-...
       MPI_Type_free(&mpi_result_type);
       MPI_Finalize();
+    }

     //
     // Estimation of Pi using Monte Carlo exploration process
     // Cecill v2 Emmanuel QUEMENER <emmanuel.quemener@gmail.com>
     // gcc -std=c99 -O3 -o Pi_aMPI Pi_aMPI.c -lm
     // Emmanuel Quemener <emmanuel.quemener@ens-lyon.fr>
     // Needed for gethostname
     #define _BSD_SOURCE
     #define _DEFAULT_SOURCE
     #include <sys/unistd.h>
     #include <math.h>
-...
     #include <limits.h>
     #include <mpi.h>
     #include <stddef.h>
     #include <sys/time.h>
     // Marsaglia RNG very simple implementation
-...
       unsigned int seed_z=362436069,seed_w=52128862;
       LENGTH iterations=ITERATIONS,inside[8192],insides,part_inside,part_iterations;
       int numtasks,rank,rc,tag=1,i;
       float pi;
       int NumberProcesses,rank,rc,tag=1,i;
       char hostname[128];
-...
         MPI_Abort(MPI_COMM_WORLD, rc);
+      }
       MPI_Comm_size(MPI_COMM_WORLD,&numtasks);
       MPI_Comm_size(MPI_COMM_WORLD,&NumberProcesses);
       MPI_Comm_rank(MPI_COMM_WORLD,&rank);
       MPI_Comm_size(MPI_COMM_WORLD,&numtasks);
       MPI_Comm_size(MPI_COMM_WORLD,&NumberProcesses);
       MPI_Comm_rank(MPI_COMM_WORLD,&rank);
       const int nitems=2;
-...
       MPI_Type_commit(&mpi_result_type);
       if (rank==0) {
         struct timeval tv1,tv2;
         if (argc > 1) {
           iterations=(LENGTH)atoll(argv[1]);
-...
         printf ("\tMax int = %u\n", INT_MAX);
         printf ("\tMax long = %ld\n", LONG_MAX);
         printf ("\tMax long long = %lld\n", LLONG_MAX);
         printf ("\tMax long long = %lld\n\n", LLONG_MAX);
         part_iterations=((iterations%numtasks) == 0) ? iterations/numtasks:iterations/numtasks+1 ;
         part_iterations=((iterations%NumberProcesses) == 0) ? iterations/NumberProcesses:iterations/NumberProcesses+1 ;
         gettimeofday(&tv1, NULL);
         // Split part of code
         for (i=1;i<numtasks;i++) {
         for (i=1;i<NumberProcesses;i++) {
     #ifdef LONG
           rc = MPI_Isend(&part_iterations, 1, MPI_LONG_LONG, i, tag,
-...
     		     MPI_COMM_WORLD,&RequestSend);
     #endif
+        }
         if (numtasks>1) {
         if (NumberProcesses>1) {
           MPI_Wait(&RequestSend, &Stat);
+        }
-...
     	     hostname,rank,(long long)insides,useconds);
         // Join part of code
           for (i=1;i<numtasks;i++) {
           for (i=1;i<NumberProcesses;i++) {
     	result recv;
     	rc = MPI_Irecv(&recv, 1, mpi_result_type, i, tag,
     		      MPI_COMM_WORLD, &RequestRecv2);
     	if (numtasks>1) {
     	if (NumberProcesses>1) {
     	    MPI_Wait(&RequestRecv2, &Stat);
+    	}
-...
     	insides+=inside[i];
+          }
           pi=4.*(float)insides/(float)(part_iterations*numtasks);
           gettimeofday(&tv2, NULL);
           printf("\n\tPi=%.40f\n\twith error %.40f\n\twith %lld iterations\n\n",pi,
     	     fabs(pi-4*atan(1.))/pi,(long long)iterations);
           double elapsed=(double)((tv2.tv_sec-tv1.tv_sec) * 1000000L +
     			      (tv2.tv_usec-tv1.tv_usec))/1000000;
           double itops=(double)(part_iterations*NumberProcesses)/elapsed;
           printf("\nParallelRate %i\nElapsed Time %.2f\nItops %.0f\nLogItops %.2f\n",NumberProcesses,elapsed,itops,log10(itops));
           LENGTH total=((iterations%NumberProcesses)==0)?iterations:(iterations/NumberProcesses+1)*NumberProcesses;
           printf("Inside/Total %ld %ld\nPi estimation %f\n\n",(long int)insides,(long int)total,(4.*(float)insides/total));
+      }
       else
+        {

       double itops=(double)(iterations)/elapsed;
       printf("Inside/Total %lld %lld\nElapsed Time %.2f\nItops %.0f\nPi estimation %f\n\n",inside,iterations,elapsed,itops,(4.*(float)inside/(float)(iterations)));
       printf("Inside/Total %lld %lld\nElapsed Time %.2f\nItops %.0f\nLogItops %.2f\nPi estimation %f\n\n",inside,iterations,elapsed,itops,log10(itops),(4.*(float)inside/(float)(iterations)));
       printf("\n\n");
       return 0;
+    }

     //
     // Estimation of Pi using Monte Carlo exploration process
     // Cecill v2 Emmanuel QUEMENER <emmanuel.quemener@gmail.com>
     // gcc -std=c99 -O3 -o Pi Pi.c -lm
     // gcc -std=c99 -O3 -o Pi_OpenMP Pi_OpenMP.c -lm
     //
     #include <math.h>
-...
       double itops=(double)(ParallelRate*IterationsEach)/elapsed;
       printf("ParallelRate %i\nElapsed Time %.2f\nItops %.0f\n",ParallelRate,elapsed,itops);
       printf("ParallelRate %i\nElapsed Time %.2f\nItops %.0f\nLogItops %.2f\n",ParallelRate,elapsed,itops,log10(itops));
       free(inside);
-...
       if (argc > 1) {
         iterations=(LENGTH)atoll(argv[1]);
         ParallelRate=atoi(argv[2]);
         if (argc > 2) {
           ParallelRate=atoi(argv[2]);
+        }
+      }
       else {
         printf("\n\tPi : Estimate Pi with Monte Carlo exploration\n\n");

     //
     // Estimation of Pi using Monte Carlo exploration process
     // gcc -std=c99 -O3 -o Pi Pi.c -lm
     // Emmanuel Quemener <emmanuel.quemener@ens-lyon.fr>
     // Cecill v2
     // Estimation of Pi using Monte Carlo exploration process using OpenMP & MPI
     // Cecill v2 Emmanuel QUEMENER <emmanuel.quemener@gmail.com>
     // gcc -std=c99 -O3 -o Pi_Hybrid Pi_Hybrid.c -lm
     // Needed for gethostname
     #define _BSD_SOURCE
     #define _DEFAULT_SOURCE
     #include <sys/unistd.h>
     #include <math.h>
-...
       LENGTH iterations=ITERATIONS,insideMPI[8192],insideOpenMP[1024],
         part_inside=0,part_iterations,insides=0;
       int numtasks,rank,rc,tag=1,i;
       float pi;
       // Hostname supposed to be <128 characters
       char hostname[128];
-...
       MPI_Type_commit(&mpi_result_type);
       if (rank==0) {
         struct timeval tv1,tv2;
         if (argc > 1) {
           iterations=(LENGTH)atoll(argv[1]);
-...
         send.iterations=part_iterations;
         send.process=process;
         gettimeofday(&tv1, NULL);
         // Split part of code
         for (i=1;i<numtasks;i++) {
           rc = MPI_Send(&send, 1, mpi_node_type, i, tag, MPI_COMM_WORLD);
-...
           insides+=insideMPI[i];
+        }
         pi=4.*(float)insides/(float)(part_iterations*numtasks*process);
         gettimeofday(&tv2, NULL);
         double elapsed=(double)((tv2.tv_sec-tv1.tv_sec) * 1000000L +
     			      (tv2.tv_usec-tv1.tv_usec))/1000000;
         printf("\n\tPi=%.40f\n\twith error %.40f\n\twith %lld iterations\n\n",pi,
     	   fabs(pi-4*atan(1.))/pi,(long long)(part_iterations*numtasks*process));
         double itops=(double)(part_iterations*numtasks)/elapsed;
         printf("\nParallelRate %i\nElapsed Time %.2f\nItops %.0f\nLogItops %.2f\n",numtasks*process,elapsed,itops,log10(itops));
         LENGTH total=((iterations%numtasks)==0)?iterations:(iterations/numtasks+1)*numtasks;
         printf("Inside/Total %ld %ld\nPi estimation %f\n\n",(long int)insides,(long int)total,(4.*(float)insides/total));
+      }
       else
+        {

     #include <acml.h>
     #else
     #include <cblas.h>
     // #include <blaswrap.h>
     //#include <blaswrap.h>
     #endif
     #ifdef CLBLAS

Centre Blaise Pascal » Bench4GPU

Révision 247