Centre Blaise Pascal » Bench4GPU

//

// 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

// Needed for gethostname

#define _BSD_SOURCE

#include <sys/unistd.h>

#include <math.h>

#include <stdio.h>

#include <stdlib.h>

#include <limits.h>

#include <mpi.h>

#include <stddef.h>

#ifdef TIME

#include <sys/time.h>

#endif

// Marsaglia RNG very simple implementation

#define znew  ((z=36969*(z&65535)+(z>>16))<<16)

#define wnew  ((w=18000*(w&65535)+(w>>16))&65535)

#define MWC   (znew+wnew)

#define SHR3  (jsr=(jsr=(jsr=jsr^(jsr<<17))^(jsr>>13))^(jsr<<5))

#define CONG  (jcong=69069*jcong+1234567)

#define KISS  ((MWC^CONG)+SHR3)

#define ITERATIONS 1000000000

#define MWCfp MWC * 2.328306435454494e-10f

#define KISSfp KISS * 2.328306435454494e-10f

#define SHR3fp SHR3 * 2.328306435454494e-10f

#define CONGfp CONG * 2.328306435454494e-10f

#define PROCESS 1

#ifdef LONG

#define LENGTH long long

#else

#define LENGTH int

#endif

typedef struct compute_node {

        LENGTH iterations;

        int process;

} node;

unsigned int rotl(unsigned int value, int shift) {

    return (value << shift) | (value >> (sizeof(value) * CHAR_BIT - shift));

}

unsigned int rotr(unsigned int value, int shift) {

    return (value >> shift) | (value << (sizeof(value) * CHAR_BIT - shift));

}

LENGTH MainLoopGlobal(LENGTH iterations,unsigned int seed_w,unsigned int seed_z)

{

#if defined TCONG

   unsigned int jcong=seed_z;

#elif defined TSHR3

   unsigned int jsr=seed_w;

#elif defined TMWC

   unsigned int z=seed_z;

   unsigned int w=seed_w;

#elif defined TKISS

   unsigned int jcong=seed_z;

   unsigned int jsr=seed_w;

   unsigned int z=seed_z;

   unsigned int w=seed_w;

#endif

   LENGTH total=0;

   for (LENGTH i=0;i<iterations;i++) {

#if defined TINT32

    #define THEONE 1073741824

    #if defined TCONG

        unsigned int x=CONG>>17 ;

        unsigned int y=CONG>>17 ;

    #elif defined TSHR3

        unsigned int x=SHR3>>17 ;

        unsigned int y=SHR3>>17 ;

    #elif defined TMWC

        unsigned int x=MWC>>17 ;

        unsigned int y=MWC>>17 ;

    #elif defined TKISS

        unsigned int x=KISS>>17 ;

        unsigned int y=KISS>>17 ;

    #endif

#elif defined TINT64

    #define THEONE 4611686018427387904

    #if defined TCONG

        unsigned long x=(unsigned long)(CONG>>1) ;

        unsigned long y=(unsigned long)(CONG>>1) ;

    #elif defined TSHR3

        unsigned long x=(unsigned long)(SHR3>>1) ;

        unsigned long y=(unsigned long)(SHR3>>1) ;

    #elif defined TMWC

        unsigned long x=(unsigned long)(MWC>>1) ;

        unsigned long y=(unsigned long)(MWC>>1) ;

    #elif defined TKISS

        unsigned long x=(unsigned long)(KISS>>1) ;

        unsigned long y=(unsigned long)(KISS>>1) ;

    #endif

#elif defined TFP32

    #define THEONE 1.0f

    #if defined TCONG

        float x=CONGfp ;

        float y=CONGfp ;

    #elif defined TSHR3

        float x=SHR3fp ;

        float y=SHR3fp ;

    #elif defined TMWC

        float x=MWCfp ;

        float y=MWCfp ;

    #elif defined TKISS

      float x=KISSfp ;

      float y=KISSfp ;

    #endif

#elif defined TFP64

    #define THEONE 1.0f

    #if defined TCONG

        double x=(double)CONGfp ;

        double y=(double)CONGfp ;

    #elif defined TSHR3

        double x=(double)SHR3fp ;

        double y=(double)SHR3fp ;

    #elif defined TMWC

        double x=(double)MWCfp ;

        double y=(double)MWCfp ;

    #elif defined TKISS

        double x=(double)KISSfp ;

        double y=(double)KISSfp ;

    #endif

#endif

      // Matching test

      unsigned long inside=((x*x+y*y) < THEONE) ? 1:0;

      total+=inside;

   }

   return(total);

}

int main(int argc, char *argv[]) {

  unsigned int seed_z=362436069,seed_w=52128862,process=PROCESS;

  // Number of NP or OpenMP processes <1024

  LENGTH iterations=ITERATIONS,insideMPI[1024],insideOpenMP[1024],

    part_inside,part_iterations,insides;

  int numtasks,rank,rc,tag=1,i;

  float pi;

  // Hostname supposed to be <128 characters

  char hostname[128];

  gethostname(hostname, sizeof hostname);

#ifdef TIME

  struct timeval start,end;

  long int useconds;

#endif

  MPI_Status Stat;

  MPI_Request RequestSend,RequestRecv,RequestSend2,RequestRecv2;

  rc = MPI_Init(&argc,&argv);

  if (rc != MPI_SUCCESS) {

    printf ("Error starting MPI program. Terminating.\n");

    MPI_Abort(MPI_COMM_WORLD, rc);

  }

  MPI_Comm_size(MPI_COMM_WORLD,&numtasks);

    const int nitems=2;

    int blocklengths[2] = {1,1};

#ifdef LONG

    MPI_Datatype types[2] = {MPI_LONG, MPI_INT};

#else

    MPI_Datatype types[2] = {MPI_INT, MPI_INT};

#endif

    MPI_Datatype mpi_node_type;

    MPI_Aint     offsets[2];

    offsets[0] = offsetof(node, iterations);

    offsets[1] = offsetof(node, process);

    MPI_Type_create_struct(nitems, blocklengths, offsets, types, &mpi_node_type);

    MPI_Type_commit(&mpi_node_type);

    MPI_Comm_rank(MPI_COMM_WORLD,&rank);

  if (rank==0) {

    if (argc > 1) {

      iterations=(LENGTH)atoll(argv[1]);

      process=atoi(argv[2]);

    }

    else {

      printf("\n\tPi : Estimate Pi with Monte Carlo exploration\n\n");

      printf("\t\t#1 : number of iterations (default 1 billion)\n\n");

      printf("\t\t#1 : number of OpenMP processes (default 1)\n\n");

    }

    printf ("\n\tInformation about architecture:\n\n");

    printf ("Sizeof int = %lld bytes.\n", (long long)sizeof(int));

    printf ("Sizeof long = %lld bytes.\n", (long long)sizeof(long));

    printf ("Sizeof long long = %lld bytes.\n", (long long)sizeof(long long));

    printf ("Max int = %u\n", INT_MAX);

    printf ("Max long = %ld\n", LONG_MAX);

    printf ("Max long long = %lld\n\n", LLONG_MAX);

    part_iterations=iterations/(numtasks*process)+1;

    node send;

    send.iterations=part_iterations;

    send.process=process;

    // Split part of code

    for (i=1;i<numtasks;i++) {      

      rc = MPI_Isend(&send, 1, mpi_node_type, i, tag, MPI_COMM_WORLD, &RequestSend);

    }

    MPI_Wait(&RequestSend,&Stat);

#ifdef TIME

    gettimeofday(&start,(struct timezone *)0);

#endif

#pragma omp parallel for

  for (int i=0 ; i<process; i++) {

    insideOpenMP[i]=MainLoopGlobal(part_iterations,

                                   rotr(seed_w,process),

                                   rotl(seed_z,process));

    printf("\t(%s,%i) found %lld for process %i\n",hostname,0,

           (long long)insideOpenMP[i],i);

  }

  printf("\n");

  insides=0;

  for (int i=0 ; i<process; i++) {

    insides+=insideOpenMP[i];

  }

#ifdef TIME

    gettimeofday(&end,(struct timezone *)0);

    useconds=(end.tv_sec-start.tv_sec)*1000000+end.tv_usec-start.tv_usec;

      printf("\tOn %s with %i find %lld inside in %lu useconds.\n",

             hostname,rank,(long long)insides,useconds);

#else

      printf("\tOn %s with %i find %lld inside\n",hostname,rank,

             (long long)insides);

#endif

    // Join part of code

      for (i=1;i<numtasks;i++) {

#ifdef LONG

        rc = MPI_Irecv(&insideMPI[i], 1, MPI_LONG_LONG, i, tag, 

                       MPI_COMM_WORLD, &RequestRecv2);

#else

        rc = MPI_Irecv(&insideMPI[i], 1, MPI_INT, i, tag, 

                       MPI_COMM_WORLD, &RequestRecv2);

#endif

        MPI_Wait(&RequestRecv2,&Stat);

        printf("\tReceive %lu inside from rank %i\n",(unsigned long)insideMPI[i],i);

        insides+=insideMPI[i];

      }

      pi=4.*(float)insides/(float)((iterations/numtasks)*numtasks);

      printf("\n\tPi=%.40f\n\twith error %.40f\n\twith %lld iterations\n\n",pi,

             fabs(pi-4*atan(1.))/pi,(long long)iterations);

  }

  else

    {

      // Receive information from master

      node recv;

      rc = MPI_Irecv(&recv, 1, mpi_node_type, 0, tag, MPI_COMM_WORLD, &RequestRecv);

      MPI_Wait(&RequestRecv,&Stat);

      printf("\t(%s,%i) receive from master %lld with %i process\n",

             hostname,rank,(long long)recv.iterations,recv.process);

#ifdef TIME

      gettimeofday(&start,(struct timezone *)0);

#endif

#pragma omp parallel for

  for (int i=0 ; i<recv.process; i++) {

      insideOpenMP[i]=MainLoopGlobal(recv.iterations,rotr(seed_w,rank+process),rotl(seed_z,rank+process));

      printf("\t(%s,%i) found %lld for process %i\n",hostname,rank,

             (long long)insideOpenMP[i],i);

  }

  printf("\n");

  part_inside=0;

  for (int i=0 ; i<recv.process; i++) {

    part_inside+=insideOpenMP[i];

  }

#ifdef TIME

      gettimeofday(&end,(struct timezone *)0);

      useconds=(end.tv_sec-start.tv_sec)*1000000+end.tv_usec-start.tv_usec;

      printf("\tOn %s rank %i find %lld inside in %lu useconds.\n",

             hostname,rank,(long long)part_inside,useconds);

#else

      printf("\tOn %s rank %i find %lld inside\n",hostname,rank,

             (long long)part_inside);

#endif

#ifdef LONG

      rc = MPI_Isend(&part_inside, 1, MPI_LONG_LONG, 0, tag, MPI_COMM_WORLD, &RequestSend2);

#else

      rc = MPI_Isend(&part_inside, 1, MPI_INT, 0, tag, MPI_COMM_WORLD, &RequestSend2);

#endif

      MPI_Wait(&RequestSend2, &Stat);

    }

  MPI_Type_free(&mpi_node_type);

  MPI_Finalize();

}