如何使用OpenCL编写程序

代码功能为计算2个数相加

#include <CL/CL.h>
#include <iostream>

#pragma comment( lib, "opencl.lib" )

int main(int argc, char** argv)
{
    cl_context hContext;
    cl_context_properties prop = CL_CONTEXT_PLATFORM;
    cl_platform_id platform;
    clGetPlatformIDs( 1, &platform, NULL );
    cl_device_id device;
    clGetDeviceIDs( platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL );
    cl_int err;
    hContext = clCreateContext( NULL, 1, &device, NULL, NULL , &err );
    size_t nCtxDescSize;
    clGetContextInfo( hContext, CL_CONTEXT_DEVICES, 0, 0, &nCtxDescSize );
    cl_device_id* pDevices = (cl_device_id*)new char[nCtxDescSize];
    clGetContextInfo( hContext, CL_CONTEXT_DEVICES, nCtxDescSize, pDevices, NULL );

    const char* sources =
    {
        "__kernel void VectorAdd(__global const float* a, __global const float* b, __global float* c)\n"
        "{\n"
        "    // get index into global data array\n"
        "    int iGID = get_global_id(0);\n"
        "\n"
        "    // bound check (equivalent to the limit on a 'for' loop for standard/serial C code\n"
        "    if (iGID >= 1)\n"
        "    {   \n"
        "        return; \n"
        "    }\n"
        "    c[iGID] = a[iGID] + b[iGID];\n"
        "}\n"
    };

    const char** sSources = (const char**)&sources;
    cl_command_queue hCmdQueue;
    hCmdQueue = clCreateCommandQueue( hContext, pDevices[0], 0, 0 );
    cl_program hProgram;
    hProgram = clCreateProgramWithSource( hContext, 1, sSources, 0, 0 );
    clBuildProgram( hProgram, NULL, NULL, NULL, NULL, NULL );

    cl_kernel hKernel;
    hKernel = clCreateKernel( hProgram, "VectorAdd", 0 );

    int cnBlockSize = 512;
    int cnBlocks = 3;
    size_t cnDimension = cnBlocks * cnBlockSize;

    float* pA = new float[cnDimension];
    float* pB = new float[cnDimension];
    float* pC = new float[cnDimension];
    memset( pA, 0, 4 * cnDimension );
    memset( pB, 0, 4 * cnDimension );
    *pA = 1.0f;
    *pB = 2.0f;
    cl_mem MemA, MemB, MemC;
    MemA = clCreateBuffer( hContext, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR,
        cnDimension * sizeof(float), pA, 0 );
    MemB = clCreateBuffer( hContext, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR,
        cnDimension * sizeof(float), pB, 0 );
    MemC = clCreateBuffer( hContext, CL_MEM_WRITE_ONLY | CL_MEM_USE_HOST_PTR,
        cnDimension * sizeof(float), pC, 0 );

    clSetKernelArg( hKernel, 0, sizeof(cl_mem), (void*)&MemA );
    clSetKernelArg( hKernel, 1, sizeof(cl_mem), (void*)&MemB );
    clSetKernelArg( hKernel, 2, sizeof(cl_mem), (void*)&MemC );
    clEnqueueNDRangeKernel( hCmdQueue, hKernel, 1, 0, &cnDimension, 0, 0, 0, 0 );
    clEnqueueReadBuffer( hCmdQueue, MemC, CL_TRUE, 0, cnBlockSize * sizeof(float),
        pC, NULL, NULL, NULL );

    delete[] pA;
    delete[] pB;
    delete[] pC;
    delete[] pDevices;

    clReleaseMemObject( MemA );
    clReleaseMemObject( MemB );
    clReleaseMemObject( MemC );

  return 0;
}