Centre Blaise Pascal » Bench4GPU

/*

   Performs matrix multiply on several BLAS implementations

   Copyleft Emmanuel QUEMENER <emmanuel.quemener@gmail.com> under GPLv3

   2014-03-14 : Add clBLAS implementation

   Thanks for help from aurel32@debian.org

*/

#include <stdio.h>

#include <math.h>

#include <stdlib.h>

#include <sys/time.h>

#include <string.h>

#ifdef CLBLAS

#include <clBLAS.h>

/* Precise here to avoid new specific bench function */

int MyPlatform;

int MyDevice;

#elif CUBLAS

#include <cublas.h>

#define CUBLAS_WRAPPER_ERROR_NOERR      0

#define CUBLAS_WRAPPER_ERROR_ALLOC      1

#define CUBLAS_WRAPPER_ERROR_SET        2

#define CUBLAS_WRAPPER_ERROR_GET        3

#define CUBLAS_WRAPPER_ERROR_STUB       4

#elif THUNKING

#include <cublas.h>

#include "fortran_common.h"

#include "fortran_thunking.h"

#elif FBLAS

#include <f77blas.h>

#elif GSL

#include <gsl_cblas.h>

#elif ACML

#include <acml.h>

#else

#include <cblas.h>

//#include <blaswrap.h>

#endif

#ifdef CLBLAS

#ifdef FP64

#define LENGTH cl_double

#else

#define LENGTH cl_float

#endif

#else

#ifdef FP64

#define LENGTH double

#else

#define LENGTH float

#endif

#endif

/* #ifdef FBLAS */

/* #ifdef FP64 */

/* void F77_dgemm(FCHAR, FCHAR, FINT, FINT, FINT, const double *, const double *, FINT,  */

/*                const double *, FINT, const double *, double *, FINT); */

/* #else */

/* void F77_sgemm(FCHAR, FCHAR, FINT, FINT, FINT, const float *, const float *, FINT,  */

/*                const float *, FINT, const float *, float *, FINT); */

/* #endif */

/* #endif */

/* Matrix with only defined triangular terms */

/* Even if there are 0 in matrix, must be defined at all ! */

/* Get from fortran.c */

#ifdef CUBLAS

static char *errMsg[5] =

{

    "no error",

    "allocation error",

    "setVector/setMatrix error",

    "getVector/getMatrix error",

    "not implemented"

};

static void wrapperError (const char *funcName, int error)

{

    printf ("cublas%s wrapper: %s\n", funcName, errMsg[error]);

    fflush (stdout);

}

#endif

int printVector(const int dimVector,const LENGTH *dataVector,

                char *nameVector,char *mesgVector)

{

#ifndef QUIET

  int i;

  printf("\n%s of %s, size %i:\n",mesgVector,nameVector,dimVector);

  for (i=0;i<dimVector;i++)

    {

      printf("%s[%i]=%2.10e\n",nameVector,i,dataVector[i]);

    }

#endif

  return 0;

}

int printResults(const int dimVector,const LENGTH *dataVector,

                 char *nameVector,char *mesgVector)

{

#ifdef RESULTS

  int i;

  printf("\n%s of %s, size %i:\n",mesgVector,nameVector,dimVector);

  for (i=0;i<dimVector;i++)

    {

      printf("%s[%i]=%2.10e\n",nameVector,i,dataVector[i]);

    }

#endif

  return 0;

}

#ifdef CUBLAS

int printVectorGPU(const int dimVector,const LENGTH *dataVector,

                   char *nameVector,char *mesgVector)

{

#ifndef QUIET

  int i;

  cublasStatus stat;

  LENGTH *P=0;

  int incx=1;

  P=malloc(dimVector*sizeof(LENGTH));

  stat=cublasGetVector(dimVector,sizeof(P[0]),dataVector,incx,P,incx);

  if (stat != CUBLAS_STATUS_SUCCESS) {

    wrapperError ("ToGet", CUBLAS_WRAPPER_ERROR_GET);

  }

  printf("\n%s of %s, size %i:\n",mesgVector,nameVector,dimVector);

  for (i=0;i<dimVector;i++)

    {

      printf("%s[%i]=%2.10e\n",nameVector,i,P[i]);

    }

  free(P);

#endif

  return 0;

}

#endif

int bench(int dim,int RUNS)

{

  /*

  int dim=1000;

  int RUNS=100;

  int incx=1;

  */

#ifdef PRINT

  LENGTH factor=1.;

#endif

  LENGTH alpha=1.,beta=0.;

  LENGTH *A,*B,*C,*D;

  /* checkBefore checkAfter checks */

  LENGTH *checksA,*checksB;

  int i=0, j=0;

  double duration;

  struct timeval tv1,tv2;

  struct timezone tz;

  /* Create 4 Matrix of dimension dim by dim  */

  A=malloc(dim*dim*sizeof(LENGTH));

  B=malloc(dim*dim*sizeof(LENGTH));

  C=malloc(dim*dim*sizeof(LENGTH));

  D=malloc(dim*dim*sizeof(LENGTH));

  /* Create 2 vectors for checker Before and After */

  checksA=malloc(RUNS*sizeof(LENGTH));

  checksB=malloc(RUNS*sizeof(LENGTH));

  /* Initialize elements with random numbers */

  /* Initialize the seed for rand() */

  /* srand(time()); */

  for (i=0; i<dim; i++) {

    for (j=0; j<dim; j++) {

        A[i*dim+j]=(LENGTH)rand()/(RAND_MAX+1.)

          *(LENGTH)(i+1.)/(LENGTH)(j+1.);

        B[i*dim+j]=(LENGTH)rand()/(RAND_MAX+1.)

          *(LENGTH)(i+1.)/(LENGTH)(j+1.);

        C[i*dim+j]=0.;

        D[i*dim+j]=0.;

    }

  }

  /*

  A[0]=1;

  A[1]=2;

  A[2]=3;

  A[3]=4;

  B[0]=5;

  B[1]=6;

  B[2]=7;

  B[3]=8;

  */

  /* Print the matrix */

#ifdef QUIET

#else

  for (i=0; i<dim; i++) {

    for (j=0; j<dim; j++) printf("A[%i,%i]=%1.5f ", i,j,A[i*dim+j]);

    putchar('\n');

  }

  putchar('\n');

  for (i=0; i<dim; i++) {

    for (j=0; j<dim; j++) printf("B[%i,%i]=%1.5f ", i,j,B[i*dim+j]);

    putchar('\n');

  }

  putchar('\n');

#endif

 /* Get first timer before launching */

  gettimeofday(&tv1, &tz);

  /* Compute with CLBLAS library  */

#ifdef CLBLAS

  cl_uint platformCount;

  cl_platform_id* platforms;

  cl_uint deviceCount;

  cl_device_id* devices;

  cl_int err,errA,errB,errC,errD;

  cl_context_properties props[3] = { CL_CONTEXT_PLATFORM, 0, 0 };

  cl_context ctx = 0;

  cl_command_queue queue = 0;

  cl_mem bufA, bufB, bufC, bufD;

  cl_event event = NULL;

  char* value;

  size_t valueSize;

  // tv3 Put on Device: Allocate & Write buffer

  // tv4 Compute

  struct timeval tv3,tv4;

  printf("Using CLBLAS: %i iterations for %ix%i matrix on (%d,%d)\n",

         RUNS,dim,dim,MyPlatform,MyDevice);

  /* Setup OpenCL environment. */

  /* - get all platforms and select MyPlatform */

  /* - get all devices from MyPlatform and select MyDevice */

  // Get all platforms

  err = clGetPlatformIDs(0, NULL, &platformCount);

  platforms = (cl_platform_id*) malloc(sizeof(cl_platform_id) * platformCount);

  err = clGetPlatformIDs(platformCount, platforms, NULL);

  // Get Device defined

  err = clGetDeviceIDs(platforms[MyPlatform], CL_DEVICE_TYPE_ALL, 0, NULL, &deviceCount);

  devices = (cl_device_id*) malloc(sizeof(cl_device_id) * deviceCount);

  err = clGetDeviceIDs(platforms[MyPlatform], CL_DEVICE_TYPE_ALL, deviceCount, devices, NULL);

  // print device name

  err = clGetDeviceInfo(devices[MyDevice], CL_DEVICE_NAME, 0, NULL, &valueSize);

  value = (char*) malloc(valueSize);

  err = clGetDeviceInfo(devices[MyDevice], CL_DEVICE_NAME, valueSize, value, NULL);

  printf("Device (%d,%d): %s\n",MyPlatform,MyDevice, value);

  free(value);

  props[1] = (cl_context_properties)platforms[MyPlatform];

  /* Initialize Context */

  ctx = clCreateContext( props, 1, &devices[MyDevice], NULL, NULL, &err );

  queue = clCreateCommandQueue( ctx, devices[MyDevice], 0, &err );

  /* Setup clBLAS */

  err = clblasSetup( );

  /* Prepare OpenCL memory objects and place matrices inside them. */

  bufA = clCreateBuffer(ctx,CL_MEM_READ_ONLY,dim*dim*sizeof(*A),NULL,&errA );

  bufB = clCreateBuffer(ctx,CL_MEM_READ_ONLY,dim*dim*sizeof(*B),NULL,&errB );

  bufC = clCreateBuffer(ctx,CL_MEM_READ_WRITE,dim*dim*sizeof(*C),NULL,&errC );

  bufD = clCreateBuffer(ctx,CL_MEM_READ_WRITE,dim*dim*sizeof(*D),NULL,&errD );

  errA = clEnqueueWriteBuffer( queue,bufA,CL_TRUE,0,

                               dim*dim*sizeof(*A),A,0,NULL,NULL );

  errB = clEnqueueWriteBuffer( queue, bufB, CL_TRUE,0,

                               dim*dim*sizeof(*B),B,0,NULL,NULL );

  errC = clEnqueueWriteBuffer( queue, bufC, CL_TRUE,0,

                                 dim*dim*sizeof(*C),C,0,NULL,NULL );

  errD = clEnqueueWriteBuffer( queue, bufD, CL_TRUE,0,

                                 dim*dim*sizeof(*D),D,0,NULL,NULL );

  /* Get third timer after memory operation */

  gettimeofday(&tv3, &tz);

#ifdef FP64

  for (i=0;i<RUNS;i++)

    {

      err = clblasDgemm( clblasRowMajor,clblasNoTrans,clblasNoTrans,

                         dim,dim,dim,alpha,bufA,0,dim,bufB,0,dim,beta,

                         bufC,0,dim,1,&queue,0,NULL,&event );

      err = clblasDgemm( clblasRowMajor,clblasTrans,clblasTrans,

                         dim,dim,dim,alpha,bufB,0,dim,bufA,0,dim,beta,

                         bufD,0,dim,1,&queue,0,NULL,&event );

    }

  if (err != CL_SUCCESS) {

    printf("clblasDgemm() failed with %d\n", err);

  }

#else

  for (i=0;i<RUNS;i++)

    {

      err = clblasSgemm( clblasRowMajor,clblasNoTrans,clblasNoTrans,

                         dim,dim,dim,alpha,bufA,0,dim,bufB,0,dim,beta,

                         bufC,0,dim,1,&queue,0,NULL,&event );

      err = clblasSgemm( clblasRowMajor,clblasTrans,clblasTrans,

                         dim,dim,dim,alpha,bufB,0,dim,bufA,0,dim,beta,

                         bufD,0,dim,1,&queue,0,NULL,&event );

    }

  if (err != CL_SUCCESS) {

    printf("clblasSgemm() failed with %d\n", err);

  }

#endif

  /* Wait for calculations to be finished. */

  err = clWaitForEvents( 1, &event );

  /* Get fourth timer after memory free */

  gettimeofday(&tv4, &tz);

  /* Fetch results of calculations from GPU memory. */

  errC = clEnqueueReadBuffer( queue,bufC,CL_TRUE,0,dim*dim * sizeof(*C),

                             C,0,NULL,NULL );

  /* Fetch results of calculations from GPU memory. */

  errD = clEnqueueReadBuffer( queue,bufD,CL_TRUE,0,dim*dim*sizeof(*D),

                             D,0,NULL,NULL );

  /* Release OpenCL memory objects. */

  clReleaseMemObject( bufD );

  clReleaseMemObject( bufC );

  clReleaseMemObject( bufB );

  clReleaseMemObject( bufA );

  /* Finalize work with clBLAS */

  clblasTeardown( );

  /* Release OpenCL working objects. */

  clReleaseCommandQueue( queue );

  clReleaseContext( ctx );

  /* Compute with CuBLAS library  */

#elif CUBLAS

  LENGTH *devPtrA=0, *devPtrB=0, *devPtrC=0, *devPtrD=0;

  cublasStatus stat1, stat2, stat3, stat4;

  struct timeval tv3,tv4;

  /* Order is Row */

  /* Have to swap uplo and trans */

  char transa='N',transb='T';

  printf("Using CuBLAS: %i iterations for %ix%i matrix\n",

         RUNS,dim,dim);

  stat1=cublasAlloc(dim*dim,sizeof(devPtrA[0]),(void**)&devPtrA);

  stat2=cublasAlloc(dim*dim,sizeof(devPtrB[0]),(void**)&devPtrB);

  stat3=cublasAlloc(dim*dim,sizeof(devPtrC[0]),(void**)&devPtrC);

  stat4=cublasAlloc(dim*dim,sizeof(devPtrD[0]),(void**)&devPtrD);

  if ((stat1 != CUBLAS_STATUS_SUCCESS) ||

      (stat2 != CUBLAS_STATUS_SUCCESS) ||

      (stat3 != CUBLAS_STATUS_SUCCESS) ||

      (stat4 != CUBLAS_STATUS_SUCCESS) ) {

    wrapperError ("xGEMM", CUBLAS_WRAPPER_ERROR_ALLOC);

    cublasFree (devPtrA);

    cublasFree (devPtrB);

    cublasFree (devPtrC);

    cublasFree (devPtrD);

    return 1;

  }

  stat1=cublasSetMatrix(dim,dim,sizeof(A[0]),A,dim,devPtrA,dim);

  stat2=cublasSetMatrix(dim,dim,sizeof(B[0]),B,dim,devPtrB,dim);

  stat3=cublasSetMatrix(dim,dim,sizeof(C[0]),C,dim,devPtrC,dim);

  stat4=cublasSetMatrix(dim,dim,sizeof(D[0]),D,dim,devPtrD,dim);

  if ((stat1 != CUBLAS_STATUS_SUCCESS) ||

      (stat2 != CUBLAS_STATUS_SUCCESS) ||

      (stat3 != CUBLAS_STATUS_SUCCESS) ||

      (stat4 != CUBLAS_STATUS_SUCCESS) ) {

    wrapperError ("xGEMM", CUBLAS_WRAPPER_ERROR_SET);

    cublasFree (devPtrA);

    cublasFree (devPtrB);

    cublasFree (devPtrC);

    cublasFree (devPtrD);

    return 1;

  }

  /* Get third timer after memory operation */

  gettimeofday(&tv3, &tz);

#ifdef FP64

  for (i=0;i<RUNS;i++)

    {

      cublasDgemm(transa,transa,dim,dim,dim,alpha,devPtrB,dim,

                  devPtrA,dim,beta,devPtrC,dim);

      cublasDgemm(transb,transb,dim,dim,dim,alpha,devPtrA,dim,

                  devPtrB,dim,beta,devPtrD,dim);

    }

#else

  for (i=0;i<RUNS;i++)

    {

      cublasSgemm(transa,transa,dim,dim,dim,alpha,devPtrB,dim,

                  devPtrA,dim,beta,devPtrC,dim);

      cublasSgemm(transb,transb,dim,dim,dim,alpha,devPtrA,dim,

                  devPtrB,dim,beta,devPtrD,dim);

    }

#endif

  stat3=cublasGetMatrix(dim,dim,sizeof(C[0]),devPtrC,dim,C,dim);

  stat4=cublasGetMatrix(dim,dim,sizeof(D[0]),devPtrD,dim,D,dim);

  /* Get fourth timer before memory free */

  gettimeofday(&tv4, &tz);

  cublasFree (devPtrA);

  cublasFree (devPtrB);

  cublasFree (devPtrC);

  cublasFree (devPtrD);

  if ((stat1 != CUBLAS_STATUS_SUCCESS) ) {

    wrapperError ("xGEMM", CUBLAS_WRAPPER_ERROR_GET);

  }

#elif THUNKING

  /* Order is Row : Have to swap uplo='U' and trans='N' */

  char transa='N',transb='T';

  printf("Using CuBLAS/Thunking: %i iterations for %ix%i matrix\n",

         RUNS,dim,dim);

#ifdef FP64

  for (i=0;i<RUNS;i++)

    {

      CUBLAS_DGEMM(&transa,&transa,

                         &dim,&dim,&dim,&alpha,B,&dim,A,&dim,&beta,C,&dim);

      CUBLAS_DGEMM(&transb,&transb,

                         &dim,&dim,&dim,&alpha,A,&dim,B,&dim,&beta,D,&dim);

    }

#else

  for (i=0;i<RUNS;i++)

    {

      CUBLAS_SGEMM(&transa,&transa,

                         &dim,&dim,&dim,&alpha,B,&dim,A,&dim,&beta,C,&dim);

      CUBLAS_SGEMM(&transb,&transb,

                         &dim,&dim,&dim,&alpha,A,&dim,B,&dim,&beta,D,&dim);

    }

#endif

#elif FBLAS

  /* Order is Row : Have to swap uplo='U' and trans='N' */

      char transa='N',transb='T';

  printf("Using FBLAS: %i iterations for %ix%i matrix\n",

         RUNS,dim,dim);

#ifdef FP64

  for (i=0;i<RUNS;i++)

    {

      dgemm_(&transa,&transa,&dim,&dim,&dim,&alpha,B,&dim,A,&dim,&beta,C,&dim);

      dgemm_(&transb,&transb,&dim,&dim,&dim,&alpha,A,&dim,B,&dim,&beta,D,&dim);

    }

#else

  for (i=0;i<RUNS;i++)

    {

      sgemm_(&transa,&transa,&dim,&dim,&dim,&alpha,B,&dim,A,&dim,&beta,C,&dim);

      sgemm_(&transb,&transb,&dim,&dim,&dim,&alpha,A,&dim,B,&dim,&beta,D,&dim);

    }

#endif

#elif ACML

  /* Order is Row : Have to swap uplo='U' and trans='N' */

      char transa='N',transb='T';

  printf("Using ACML: %i iterations for %ix%i matrix\n",

         RUNS,dim,dim);

#ifdef FP64

  for (i=0;i<RUNS;i++)

    {

      dgemm(transa,transa,dim,dim,dim,alpha,B,dim,A,dim,beta,C,dim);

      dgemm(transb,transb,dim,dim,dim,alpha,A,dim,B,dim,beta,D,dim);

    }

#else

  for (i=0;i<RUNS;i++)

    {

      sgemm(transa,transa,dim,dim,dim,alpha,B,dim,A,dim,beta,C,dim);

      sgemm(transb,transb,dim,dim,dim,alpha,A,dim,B,dim,beta,D,dim);

    }

#endif

#elif GSL

  printf("Using GSL: %i iterations for %ix%i matrix\n",RUNS,dim,dim);

  /*

     RowMajor : Matrix is read row by row

     Upper : the no null elements are on top

     NoTrans : no transposition before estimation

     NonUnit : Matrix is not unit

   */

#ifdef FP64

  for (i=0;i<RUNS;i++)

    {

      cblas_dgemm(CblasRowMajor,CblasNoTrans,CblasNoTrans,

                  dim,dim,dim,alpha,A,dim,B,dim,beta,C,dim);

      cblas_dgemm(CblasRowMajor,CblasTrans,CblasTrans,

                  dim,dim,dim,alpha,B,dim,A,dim,beta,D,dim);

    }

#else

  for (i=0;i<RUNS;i++)

    {

      cblas_sgemm(CblasRowMajor,CblasNoTrans,CblasNoTrans,

                  dim,dim,dim,alpha,A,dim,B,dim,beta,C,dim);

      cblas_sgemm(CblasRowMajor,CblasTrans,CblasTrans,

                  dim,dim,dim,alpha,B,dim,A,dim,beta,D,dim);

    }

#endif

#else

  printf("Using CBLAS: %i iterations for %ix%i matrix\n",RUNS,dim,dim);

  /*

     RowMajor : Matrix is read row bu row

     Upper : the no null elements are on top

     NoTrans : no transposition before estimation

     NonUnit : Matrix is not unit

   */

#ifdef FP64

  for (i=0;i<RUNS;i++)

    {

      cblas_dgemm(CblasRowMajor,CblasNoTrans,CblasNoTrans,

                  dim,dim,dim,alpha,A,dim,B,dim,beta,C,dim);

      cblas_dgemm(CblasRowMajor,CblasTrans,CblasTrans,

                  dim,dim,dim,alpha,B,dim,A,dim,beta,D,dim);

    }

#else

  for (i=0;i<RUNS;i++)

    {

      cblas_sgemm(CblasRowMajor,CblasNoTrans,CblasNoTrans,

                  dim,dim,dim,alpha,A,dim,B,dim,beta,C,dim);

      cblas_sgemm(CblasRowMajor,CblasTrans,CblasTrans,

                  dim,dim,dim,alpha,B,dim,A,dim,beta,D,dim);

    }

#endif

#endif

  /* Get second timer after launching */

  gettimeofday(&tv2, &tz);

  /* Store the checker of errors */

  checksA[0]=0.;

  for (i=0; i<dim; i++) {

    for (j=0; j<dim; j++) {

      checksA[0]=checksA[0]+fabs(D[i*dim+j]-C[j*dim+i]);

    }

  }

  /* Print the matrix */

#ifdef QUIET

#else

  for (i=0; i<dim; i++) {

    for (j=0; j<dim; j++) printf("C[%i,%i]=%1.5f ", i,j,C[i*dim+j]);

    putchar('\n');

  }

  putchar('\n');

  for (i=0; i<dim; i++) {

    for (j=0; j<dim; j++) printf("D[%i,%i]=%1.5f ", i,j,D[i*dim+j]);

    putchar('\n');

  }

  putchar('\n');

#endif

  /* Free 1 Matrix and 2 Vectors of dimension dim  */

  free(A);

  free(B);

  free(C);

  free(D);

  putchar('\n');

#ifdef CLBLAS

  double memoryIn,memoryOut;

  memoryIn=(double)((tv3.tv_sec-tv1.tv_sec) * 1000000L +        \

                    (tv3.tv_usec-tv1.tv_usec))/1000000.;

  memoryOut=(double)((tv2.tv_sec-tv4.tv_sec) * 1000000L +        \

                    (tv2.tv_usec-tv4.tv_usec))/1000000.;

  duration=(double)((tv4.tv_sec-tv3.tv_sec) * 1000000L +        \

                    (tv4.tv_usec-tv3.tv_usec))/1000000./RUNS;

  printf("Duration of memory allocation : %2.10f s\n",memoryIn);

  printf("Duration of memory free : %2.10f s\n",memoryOut);

#elif CUBLAS

  double memoryIn,memoryOut;

  memoryIn=(double)((tv3.tv_sec-tv1.tv_sec) * 1000000L +        \

                    (tv3.tv_usec-tv1.tv_usec))/1000000.;

  memoryOut=(double)((tv2.tv_sec-tv4.tv_sec) * 1000000L +        \

                    (tv2.tv_usec-tv4.tv_usec))/1000000.;

  duration=(double)((tv4.tv_sec-tv3.tv_sec) * 1000000L +        \

                    (tv4.tv_usec-tv3.tv_usec))/1000000./RUNS;

  printf("Duration of memory allocation : %2.10f s\n",memoryIn);

  printf("Duration of memory free : %2.10f s\n",memoryOut);

#else

  duration=(double)((tv2.tv_sec-tv1.tv_sec) * 1000000L +        \

                    (tv2.tv_usec-tv1.tv_usec))/1000000./RUNS;

#endif

  printf("Duration of each cycle : %2.10f s\n",duration);

  printf("Number of GFlops : %2.3f \n",

         dim*dim*2.*(2.*dim-1)/duration/1000000000.);

  printf("Error %1.10f\n",checksA[0]);

  printResults(RUNS,checksA,"C","Errors cumulated");

  putchar('\n');

  /* Free 2 vectors for checker Before and After */

  free(checksA);

  free(checksB);

  return 0;

}

#ifdef CLBLAS

int DelectOpenCLDevices()

{

  /* */

  /* Not needed to import CL.h, already done in CLBLAS.h */

  int i, j;

  char* value;

  size_t valueSize;

  cl_uint platformCount;

  cl_platform_id* platforms;

  cl_uint deviceCount;

  cl_device_id* devices;

  cl_uint maxComputeUnits;

  cl_int maxWorkGroupSize;

  cl_int maxWorkItemSizes;

  cl_device_type dev_type;

  // get all platforms

  clGetPlatformIDs(0, NULL, &platformCount);

  platforms = (cl_platform_id*) malloc(sizeof(cl_platform_id) * platformCount);

  clGetPlatformIDs(platformCount, platforms, NULL);

  printf("OpenCL statistics: %d platform(s) detected\n\n",platformCount);

  for (i = 0; i < platformCount; i++) {

    // get all devices

    clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, 0, NULL, &deviceCount);

    devices = (cl_device_id*) malloc(sizeof(cl_device_id) * deviceCount);

    clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, deviceCount, devices, NULL);

    // for each device print critical attributes

    for (j = 0; j < deviceCount; j++) {

      // print device name

      clGetDeviceInfo(devices[j], CL_DEVICE_NAME, 0, NULL, &valueSize);

      value = (char*) malloc(valueSize);

      clGetDeviceInfo(devices[j], CL_DEVICE_NAME, valueSize, value, NULL);

      printf("Device (%d,%d): %s\n",i, j, value);

      free(value);

      // print type device CPU/GPU/ACCELERATOR

      clGetDeviceInfo(devices[j], CL_DEVICE_TYPE, sizeof(dev_type), &dev_type, NULL);

      printf("\tDevice Type: ");

      if(dev_type & CL_DEVICE_TYPE_GPU)

        printf("CL_DEVICE_TYPE_GPU ");

      if(dev_type & CL_DEVICE_TYPE_CPU)

        printf("CL_DEVICE_TYPE_CPU ");

      if(dev_type & CL_DEVICE_TYPE_ACCELERATOR)

        printf("CL_DEVICE_TYPE_ACCELERATOR ");

      if(dev_type & CL_DEVICE_TYPE_DEFAULT)

        printf("CL_DEVICE_TYPE_DEFAULT ");

      printf("\n");

      // print device vendor

      clGetDeviceInfo(devices[j], CL_DEVICE_VENDOR, 0, NULL, &valueSize);

      value = (char*) malloc(valueSize);

      clGetDeviceInfo(devices[j], CL_DEVICE_VENDOR, valueSize, value, NULL);

      printf("\tDevice vendor: %s\n", value);

      free(value);

      // print hardware device version

      clGetDeviceInfo(devices[j], CL_DEVICE_VERSION, 0, NULL, &valueSize);

      value = (char*) malloc(valueSize);

      clGetDeviceInfo(devices[j], CL_DEVICE_VERSION, valueSize, value, NULL);

      printf("\tHardware version: %s\n", value);

      free(value);

      // print software driver version

      clGetDeviceInfo(devices[j], CL_DRIVER_VERSION, 0, NULL, &valueSize);

      value = (char*) malloc(valueSize);

      clGetDeviceInfo(devices[j], CL_DRIVER_VERSION, valueSize, value, NULL);

      printf("\tSoftware version: %s\n", value);

      free(value);

      // print c version supported by compiler for device

      clGetDeviceInfo(devices[j], CL_DEVICE_OPENCL_C_VERSION, 0, NULL, &valueSize);

      value = (char*) malloc(valueSize);

      clGetDeviceInfo(devices[j], CL_DEVICE_OPENCL_C_VERSION, valueSize, value, NULL);

      printf("\tOpenCL C version: %s\n", value);

      free(value);

      // print parallel compute units

      clGetDeviceInfo(devices[j], CL_DEVICE_MAX_COMPUTE_UNITS,

                      sizeof(maxComputeUnits), &maxComputeUnits, NULL);

      printf("\tParallel compute units: %d\n", maxComputeUnits);

      // print max work group size

      clGetDeviceInfo(devices[j], CL_DEVICE_MAX_WORK_GROUP_SIZE,

                      sizeof(maxWorkGroupSize), &maxWorkGroupSize, NULL);

      printf("\tMaximum Work Group Size: %d\n", maxWorkGroupSize);

      // print max work items size

      clGetDeviceInfo(devices[j], CL_DEVICE_MAX_WORK_ITEM_SIZES,

                      sizeof(maxWorkItemSizes), &maxWorkItemSizes, NULL);

      printf("\tMaximum Work Item Sizes: %d\n", maxWorkItemSizes);

    }

    printf("\n");

    free(devices);

  }

  free(platforms);

  return 0;

}

#endif

int main(int argc,char **argv)

{

  if ((argc==1)||

      (strcmp(argv[1],"-h")==0)||

      (strcmp(argv[1],"--help")==0))

    {

#ifdef CLBLAS

      printf("\nPerforms a bench using BLAS library implementation:\n\n"

             "\t#1 Size of square matrices \n"

             "\t#2 Number of iterations \n"

             "\t#3 OpenCL Plateform ID\n"

             "\t#4 OpenCL Device ID\n\n");

      DelectOpenCLDevices();

#else

      printf("\nPerforms a bench using BLAS library implementation:\n\n"

             "\t#1 Size of square matrices \n"

             "\t#2 Number of iterations\n\n");

#endif

    }

  else if ((atoi(argv[1])>=2)&&

           (atoi(argv[2])>=1))

    {

#ifdef CLBLAS

      MyPlatform=atoi(argv[3]);

      MyDevice=atoi(argv[4]);

#endif

      bench(atoi(argv[1]),atoi(argv[2]));

    }

  return 0;

}