Jungsoo Nam A BlockChain developer

OpenCL example

OpenCL 예제이다. 이 예제에서는 GPU와 CPU를 동시에 사용하고 있다.

인텔 CPU에서 테스트를 했는데 인텔 OpenCL 드라이버를 설치하지 않으면, GPU만 가지고 수행을 하기 때문에 ret_num_platforms=1이 된다.

인텔 i5-2500과 지포스 GTX970에서 테스트 하였을때 ret_num_platforms=3이 되었다.

#include <stdio.h>
#include <stdlib.h>

#ifdef __APPLE__
#include <OpenCL/opencl.h>
#else
#include <CL/cl.h>
#endif

#define MEM_SIZE (128)
#define MAX_SOURCE_SIZE (0x100000)

int main()
{
    cl_device_id device_cpu = NULL;
    cl_device_id device_gpu = NULL;
    cl_context context_cpu = NULL;
    cl_context context_gpu = NULL;
    cl_command_queue command_queue_cpu = NULL;
    cl_command_queue command_queue_gpu = NULL;
    cl_mem memobj_cpu = NULL;
    cl_mem memobj_gpu = NULL;
    cl_program program_cpu = NULL;
    cl_program program_gpu = NULL;
    cl_kernel kernel_cpu = NULL;
    cl_kernel kernel_gpu = NULL;
    cl_platform_id platform[2];
    cl_uint ret_num_devices;
    cl_uint ret_num_platforms;
    cl_int ret;

    char string_cpu[MEM_SIZE];
    char string_gpu[MEM_SIZE];

    FILE *fp;
    char fileName[] = "./hello.cl";
    char *source_str;
    size_t source_size;

    /* 커널을 포함한 소스 코드를 로드 */
    fp = fopen(fileName, "r");
    if (!fp) {
        fprintf(stderr, "Failed to load kernel.\n");
        exit(1);
    }
    source_str = (char*)malloc(MAX_SOURCE_SIZE);
    source_size = fread(source_str, 1, MAX_SOURCE_SIZE, fp);
    fclose(fp);

    /* 플랫폼, 디바이스 정보를 얻음 */
    ret = clGetPlatformIDs(2, platform, &ret_num_platforms);
    ret = clGetDeviceIDs(platform[0], CL_DEVICE_TYPE_ALL, 1, &device_gpu, &ret_num_devices);
    ret = clGetDeviceIDs(platform[1], CL_DEVICE_TYPE_ALL, 1, &device_cpu, &ret_num_devices);

    /* OpenCL 컨텍스트 생성 */
    context_gpu = clCreateContext(NULL, 1, &device_gpu, NULL, NULL, &ret);
    context_cpu = clCreateContext(NULL, 1, &device_cpu, NULL, NULL, &ret);

    /* 커맨드 큐 생성 */
    command_queue_cpu = clCreateCommandQueue(context_cpu, device_cpu, 0, &ret);
    command_queue_gpu = clCreateCommandQueue(context_gpu, device_gpu, 0, &ret);

    /* 메모리 버퍼 생성 */
    memobj_cpu = clCreateBuffer(context_cpu, CL_MEM_READ_WRITE, MEM_SIZE * sizeof(char), NULL, &ret);
    memobj_gpu = clCreateBuffer(context_gpu, CL_MEM_READ_WRITE, MEM_SIZE * sizeof(char), NULL, &ret);

    /* 미리 로드한 소스 코드로 커널 프로그램을 생성 */
    program_cpu = clCreateProgramWithSource(context_cpu, 1, (const char **)&source_str, (const size_t *)&source_size, &ret);
    program_gpu = clCreateProgramWithSource(context_gpu, 1, (const char **)&source_str, (const size_t *)&source_size, &ret);

    /* 커널 프로그램 빌드 */
    ret = clBuildProgram(program_cpu, 1, &device_cpu, NULL, NULL, NULL);
    ret = clBuildProgram(program_gpu, 1, &device_gpu, NULL, NULL, NULL);

    /* OpenCL 커널 생성*/
    kernel_cpu = clCreateKernel(program_cpu, "hello", &ret);
    kernel_gpu = clCreateKernel(program_gpu, "hello", &ret);

    /* OpenCL 커널 파라미터 설정 */
    ret = clSetKernelArg(kernel_cpu, 0, sizeof(cl_mem), (void *)&memobj_cpu);
    ret = clSetKernelArg(kernel_gpu, 0, sizeof(cl_mem), (void *)&memobj_gpu);

    /* OpenCL 커널 실행 */
    ret = clEnqueueTask(command_queue_cpu, kernel_cpu, 0, NULL, NULL);
    ret = clEnqueueTask(command_queue_gpu, kernel_gpu, 0, NULL, NULL);

    /* 실행 결과를 메모리 버퍼에서 얻음 */
    ret = clEnqueueReadBuffer(command_queue_cpu, memobj_cpu, CL_TRUE, 0, MEM_SIZE * sizeof(char), string_cpu, 0, NULL, NULL);
    ret = clEnqueueReadBuffer(command_queue_gpu, memobj_gpu, CL_TRUE, 0, MEM_SIZE * sizeof(char), string_gpu, 0, NULL, NULL);

    /* 결과 출력 */
    puts(string_cpu);
    puts(string_gpu);

    /* 종료 처리 */
    ret = clFlush(command_queue_cpu);
    ret = clFlush(command_queue_gpu);
    ret = clFinish(command_queue_cpu);
    ret = clFinish(command_queue_gpu);
    ret = clReleaseKernel(kernel_cpu);
    ret = clReleaseKernel(kernel_gpu);
    ret = clReleaseProgram(program_cpu);
    ret = clReleaseProgram(program_gpu);
    ret = clReleaseMemObject(memobj_cpu);
    ret = clReleaseMemObject(memobj_gpu);
    ret = clReleaseCommandQueue(command_queue_cpu);
    ret = clReleaseCommandQueue(command_queue_gpu);
    ret = clReleaseContext(context_cpu);
    ret = clReleaseContext(context_gpu);

    free(source_str);

    // while (1);
    return 0;
}
comments powered by Disqus