Centre Blaise Pascal » Bench4GPU

// Pi Dart Dash in OpenCL in C, illustrative example

//

// Emmanuel Quemener <emmanuel.quemener@ens-lyon.fr

// Copyleft Cecill v2

//

// -h : print the documentation and detect devices as (platform,device)

//

// classical use:

// #1 OpenCL Plateform ID: get this information with -h option

// #2 OpenCL Device ID: get this information with -h option

// #3 Minimal number of iterations:

// #4 Parallel Rate: scattering global work in parts executed //ly

// #5 Loops (to improve statistics)

//

#define CL_USE_DEPRECATED_OPENCL_1_2_APIS

#include <stdio.h>

#include <stdlib.h>

#include <string.h>

#include <CL/cl.h>

#include <stdint.h>

#include <sys/time.h>

#define TINT32 0

#define TINT64 1

#define TFP32 2

#define TFP64 3

int DetectOpenCLDevices(void)

{

  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;

}

const char* OpenCLSource[] = {

  "#pragma OPENCL EXTENSION cl_khr_fp64: enable \n",

  "// Marsaglia RNG very simple implementation \n",

  "#define znew  ((z=36969*(z&65535)+(z>>16))<<16) \n",

  "#define wnew  ((w=18000*(w&65535)+(w>>16))&65535) \n",

  "#define MWC   (znew+wnew) \n",

  "#define SHR3  (jsr=(jsr=(jsr=jsr^(jsr<<17))^(jsr>>13))^(jsr<<5)) \n",

  "#define CONG  (jcong=69069*jcong+1234567) \n",

  "#define KISS  ((MWC^CONG)+SHR3) \n",

  "#define MWCfp MWC * 2.328306435454494e-10f \n",

  "#define KISSfp KISS * 2.328306435454494e-10f \n",

  "#define CONGfp CONG * 2.328306435454494e-10f \n",

  "#define SHR3fp SHR3 * 2.328306435454494e-10f \n",

  "#define TINT32 0 \n",

  "#define TINT64 1 \n",

  "#define TFP32 2 \n",

  "#define TFP64 3 \n",

  "#define THEONE32I 1073741824 \n",

  "#define THEONE32F 1.e0f \n",

  "#define THEONE64I 4611686018427387904 \n",

  "#define THEONE64F (double)1.e0f \n",

  "ulong MainLoop32I(ulong iterations,uint seed_z,uint seed_w,size_t work)",

  "{",

  "   uint z=seed_z+work;",

  "   uint w=seed_w+work;",

  "   ulong total=0;",

  "   for (ulong i=0;i<iterations;i++)",

  "   {",

  "      uint x= MWC>>17;",

  "      uint y= MWC>>17;",

  "      ulong inside=((x*x+y*y) <= THEONE32I) ? 1:0;",

  "      total+=inside;",

  "   }",

  "   return(total);",

  "}",

  "ulong MainLoop32F(ulong iterations,uint seed_z,uint seed_w,size_t work)",

  "{",

  "   uint z=seed_z+work;",

  "   uint w=seed_w+work;",

  "   ulong total=0;",

  "   for (ulong i=0;i<iterations;i++)",

  "   {",

  "      float x=(float)MWCfp ;",

  "      float y=(float)MWCfp ;",

  "      ulong inside=((x*x+y*y) <= THEONE32F) ? 1:0;",

  "      total+=inside;",

  "   }",

  "   return(total);",

  "}",

  "ulong MainLoop64I(ulong iterations,uint seed_z,uint seed_w,size_t work)",

  "{",

  "   uint z=seed_z+work;",

  "   uint w=seed_w+work;",

  "   ulong total=0;",

  "   for (ulong i=0;i<iterations;i++)",

  "   {",

  "      ulong x=(ulong)(MWC>>1);",

  "      ulong y=(ulong)(MWC>>1);",

  "      ulong inside=((x*x+y*y) <= THEONE64I) ? 1:0;",

  "      total+=inside;",

  "   }",

  "   return(total);",

  "}",

  "ulong MainLoop64F(ulong iterations,uint seed_z,uint seed_w,size_t work)",

  "{",

  "   uint z=seed_z+work;",

  "   uint w=seed_w+work;",

  "   ulong total=0;",

  "   for (ulong i=0;i<iterations;i++)",

  "{",

  "        double x=(double)MWCfp ;",

  "        double y=(double)MWCfp ;",

  "      ulong inside=((x*x+y*y) <= THEONE64F) ? 1:0;",

  "      total+=inside;",

  "}",

  "   return(total);",

  "}",

  "__kernel void MainLoopGlobal(__global ulong *s,ulong iterations,uint seed_w,uint seed_z,uint MyType)",

  "{",

  "   ulong total;",

  "   if (MyType==TFP32) {",

  "      total=(ulong)MainLoop32F(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   else if (MyType==TFP64) {",

  "      total=(ulong)MainLoop64F(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   else if (MyType==TINT32) {",

  "      total=(ulong)MainLoop32I(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   else if (MyType==TINT64) {",

  "      total=(ulong)MainLoop64I(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   barrier(CLK_GLOBAL_MEM_FENCE);",

  "   s[get_global_id(0)]=(ulong)total;",

  "}",

  "__kernel void MainLoopLocal(__global ulong *s,ulong iterations,uint seed_w,uint seed_z,uint MyType)",

  "{",

  "   ulong total;",

  "   if (MyType==TFP32) {",

  "      total=(ulong)MainLoop32F(iterations,seed_z,seed_w,get_local_id(0));",

  "   }",

  "   else if (MyType==TFP64) {",

  "      total=(ulong)MainLoop64F(iterations,seed_z,seed_w,get_local_id(0));",

  "   }",

  "   else if (MyType==TINT32) {",

  "      total=(ulong)MainLoop32I(iterations,seed_z,seed_w,get_local_id(0));",

  "   }",

  "   else if (MyType==TINT64) {",

  "      total=(ulong)MainLoop64I(iterations,seed_z,seed_w,get_local_id(0));",

  "   }",

  "   barrier(CLK_LOCAL_MEM_FENCE);",

  "   s[get_local_id(0)]=(ulong)total;",

  "}",

  "__kernel void MainLoopHybrid(__global ulong *s,ulong iterations,uint seed_w,uint seed_z,uint MyType)",

  "{",

  "   ulong total;",

  "   if (MyType==TFP32) {",

  "      total=(ulong)MainLoop32F(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   else if (MyType==TFP64) {",

  "      total=(ulong)MainLoop64F(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   else if (MyType==TINT32) {",

  "      total=(ulong)MainLoop32I(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   else if (MyType==TINT64) {",

  "      total=(ulong)MainLoop64I(iterations,seed_z,seed_w,get_global_id(0));",

  "   }",

  "   barrier(CLK_GLOBAL_MEM_FENCE || CLK_LOCAL_MEM_FENCE);",

  "   s[get_global_id(0)]=total;",

  "}"

};

int main(int argc, char **argv)

{

  if ((argc==1)||

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

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

    {

      printf("\nPerforms a Pi estimation by Dart Dash:\n\n"

             "\t#1 OpenCL Plateform ID (default 0)\n"

             "\t#2 OpenCL Device ID (default 0)\n"

             "\t#3 Minimal number of iterations (default 1000000)\n"

             "\t#4 Parallel Rate (default 1024)\n"

             "\t#5 Loops (default 1)\n"

             "\t#6 Type of variable: INT32, INT64, FP32, FP64 (default FP32)\n\n");

      DetectOpenCLDevices();

    }

  else

    {

      int MyPlatform=atoi(argv[1]);

      int MyDevice=atoi(argv[2]);

      struct timeval tv1,tv2;

      struct timezone tz;

      uint64_t Iterations=1000000;

      if (argc>3) {

        Iterations=(uint64_t)atoll(argv[3]);

      }

      uint32_t ParallelRate=64;

      if (argc>4) {

        ParallelRate=(uint32_t)atoi(argv[4]);

      }

      uint32_t Loops=1;

      if (argc>5) {

        Loops=(uint32_t)atoi(argv[5]);

      }

      uint32_t MyType=TFP32;

      if (argc>6) {

        if (strcmp(argv[6],"INT32")==0) {

          MyType=(uint32_t)TINT32;

        }

        else if (strcmp(argv[6],"INT64")==0) {

          MyType=(uint32_t)TINT64;

        }

        else if (strcmp(argv[6],"FP32")==0) {

          MyType=(uint32_t)TFP32;

        }

        else if (strcmp(argv[6],"FP64")==0) {

          MyType=(uint32_t)TFP64;

        }

      }

      printf("MyType %d\n",MyType);

      cl_int err;

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

      // Detect, scan, get & initialize platform and device

      cl_uint platformCount;

      cl_platform_id* platforms;

      cl_uint deviceCount;

      cl_device_id* devices;

      size_t valueSize;

      /* Setup OpenCL environment. */

      // 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);

      char* deviceName=(char*)malloc(valueSize);

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

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

      char* vendorName=(char*)malloc(valueSize);

      err = clGetDeviceInfo(devices[MyDevice], CL_DEVICE_VENDOR, valueSize, vendorName, NULL);

      printf("\nDevice (%d,%d):\n\t- vendor: %s\n\t- device: %s\n\n",MyPlatform,MyDevice, vendorName,deviceName);

      free(deviceName);

      free(vendorName);

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

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

      cl_command_queue cqCommandQueue = clCreateCommandQueue(GPUContext,devices[MyDevice], 0, &err);

      cl_mem GPUInside = clCreateBuffer(GPUContext, CL_MEM_WRITE_ONLY,

                                              sizeof(uint64_t) * ParallelRate, NULL, NULL);

      // 51 is the number of line for OpenCL code

      // 66, sans test

      cl_program OpenCLProgram = clCreateProgramWithSource(GPUContext, 130 ,OpenCLSource,NULL,NULL);

      clBuildProgram(OpenCLProgram, 0, NULL, NULL, NULL, NULL);

      cl_kernel OpenCLMainLoopGlobal = clCreateKernel(OpenCLProgram, "MainLoopGlobal", NULL);

      cl_kernel OpenCLMainLoopLocal = clCreateKernel(OpenCLProgram, "MainLoopLocal", NULL);

      cl_kernel OpenCLMainLoopHybrid = clCreateKernel(OpenCLProgram, "MainLoopHybrid", NULL);

      // Divide the total number of iterations by the parallel rate

      // Add +1 to the number of per work iterations if division not integer

      uint64_t IterationsEach=((Iterations%ParallelRate)==0)?Iterations/ParallelRate:Iterations/ParallelRate+1;

      // Initialize seeds for MWC RNG generator from Marsaglia

      uint32_t seed_w=110271;

      uint32_t seed_z=101008;

      // Set the values of arguments for OpenCL function call

      clSetKernelArg(OpenCLMainLoopGlobal, 0, sizeof(cl_mem),&GPUInside);

      clSetKernelArg(OpenCLMainLoopGlobal, 1, sizeof(uint64_t),&IterationsEach);

      clSetKernelArg(OpenCLMainLoopGlobal, 2, sizeof(uint32_t),&seed_w);

      clSetKernelArg(OpenCLMainLoopGlobal, 3, sizeof(uint32_t),&seed_z);

      clSetKernelArg(OpenCLMainLoopGlobal, 4, sizeof(uint32_t),&MyType);

      size_t WorkSize[1] = {ParallelRate}; // one dimensional Range

      uint64_t HostInside[ParallelRate];

      for (uint32_t loop=0;loop<Loops;loop++) {

        // Set start timer

        gettimeofday(&tv1, &tz);

            // Execute the OpenCL kernel with datas

        clEnqueueNDRangeKernel(cqCommandQueue, OpenCLMainLoopGlobal, 1, NULL,

                               WorkSize, NULL, 0, NULL, NULL);

        // Copy each result for each PR from Device to Host

        clEnqueueReadBuffer(cqCommandQueue, GPUInside, CL_TRUE, 0,

                            ParallelRate * sizeof(uint64_t), HostInside, 0, NULL, NULL);

        uint64_t inside=0;

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

          inside+=HostInside[i];

        }

        // Set stop timer

        gettimeofday(&tv2, &tz);

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

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

        double itops=(double)(ParallelRate*IterationsEach)/elapsed;

        printf("Inside/Total %ld %ld\nParallelRate %i\nElapsed Time %.2f\nItops %.0f\nPi estimation %f\n\n",inside,ParallelRate*IterationsEach,ParallelRate,elapsed,itops,(4.*(float)inside/((float)(ParallelRate)*(float)(IterationsEach))));

      }

      printf("\n\n");

      clReleaseKernel(OpenCLMainLoopGlobal);

      clReleaseProgram(OpenCLProgram);

      clReleaseCommandQueue(cqCommandQueue);

      clReleaseContext(GPUContext);

      clReleaseMemObject(GPUInside);

      return 0;

    }

}