opencl 对双数据类型给出错误的结果？

Question

我对 OpenCl 很陌生。

问题：我正在初始化一个包含 10 个元素的数组，其值为 -1 在 Host 中。我将相同的数组作为输入传递给设备并递增每个元素，然后通过缓冲区接收回数组作为输出。再一次，将相同的输出数组作为输入发送回主机并接收增加的输出缓冲区。这是循环完成的。

问题：当我将数组的数据类型设为浮点/整数时，程序运行正常。但是当我将数据类型作为数组的两倍时，它并没有给我想要的结果。 请建议我在哪里失踪？任何帮助/建议/关键字都会有很大帮助。提前致谢。

数组浮点数据类型代码：

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


#define MEM_SIZE (10)
#define MAX_SOURCE_SIZE (0x100000)


int main() {
    float input[MEM_SIZE], output[MEM_SIZE];
    int go, i;
    for (i = 0; i < MEM_SIZE; i++) {
        input[i] = -1.0;
        output[i] = -1.0;
    }
    FILE *fp;
    cl_device_id device_id = NULL;
    cl_context context = NULL;
    cl_command_queue command_queue = NULL;
    cl_program program = NULL;
    cl_kernel kernel = NULL;
    cl_platform_id platform_id = NULL;
    cl_uint ret_num_devices;
    cl_uint ret_num_platforms;
    cl_int ret;
    size_t source_size;
    char *source_str;
    fp = fopen("calc_float.cl", "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);


    /*Initialization*/
    /* Get Platform and Device Info */
    ret = clGetPlatformIDs(1, &platform_id, &ret_num_platforms);
    ret = clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id, &ret_num_devices);

    /* Create OpenCL context */
    context = clCreateContext(NULL, 1, &device_id, NULL, NULL, &ret);

    /* Create Command Queue */
    command_queue = clCreateCommandQueue(context, device_id, 0, &ret);

    /*Initialization complete*/

    cl_mem inputBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, MEM_SIZE * sizeof(float),(void *) input, NULL);
    cl_mem outputBuffer = clCreateBuffer(context, CL_MEM_WRITE_ONLY , MEM_SIZE * sizeof(float), NULL, NULL);


    /* Create Kernel Program from the source */
    program = clCreateProgramWithSource(context, 1, (const char **)&source_str,(const size_t *)&source_size, &ret);

    /* Build Kernel Program */
    ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);

    /* Create OpenCL Kernel */
    kernel = clCreateKernel(program, "calc", &ret);

    /* Set OpenCL Kernel Parameters */
    ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&inputBuffer);
    ret = clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&outputBuffer);

    /* Execute OpenCL Kernel */
    ret = clEnqueueTask(command_queue, kernel, 0, NULL,NULL);
    double x = 10, io;
    size_t global_work_size[1] = {MEM_SIZE};
    for (io = 0; io < x; io++) {
        inputBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, MEM_SIZE * sizeof(float),(void *) output, NULL);
        ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&inputBuffer);
        ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL, global_work_size, NULL, 0, NULL, NULL);

        ret = clEnqueueReadBuffer(command_queue, outputBuffer, CL_TRUE, 0, MEM_SIZE * sizeof(float), output, 0, NULL, NULL);
        for (go = 0; go < MEM_SIZE; go++) {
            printf("output[%d] = %f\n",io, go, output[go]);
        }
        printf("\n\n");

    }


    /* Finalization */
    ret = clFlush(command_queue);
    ret = clFinish(command_queue);
    ret = clReleaseKernel(kernel);
    ret = clReleaseProgram(program);
    ret = clReleaseMemObject(inputBuffer);
    ret = clReleaseMemObject(outputBuffer);
    ret = clReleaseCommandQueue(command_queue);
    ret = clReleaseContext(context);

    return 1;
}

以上代码的 calc_float.cl 文件：

__kernel void calc(__global float* in, __global float* out)
{
    int i;
    for (i = 0; i < 10; i++) {
        out[i] = in[i] + 1;
    }
}

数组双精度数据类型代码：

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

#define MEM_SIZE (10)
#define MAX_SOURCE_SIZE (0x100000)


int main() {
    double input[MEM_SIZE], output[MEM_SIZE];
    int go, i;
    for (i = 0; i < MEM_SIZE; i++) {
        input[i] = -1.0;
        output[i] = -1.0;
    }
    FILE *fp;
    cl_device_id device_id = NULL;
    cl_context context = NULL;
    cl_command_queue command_queue = NULL;
    cl_program program = NULL;
    cl_kernel kernel = NULL;
    cl_platform_id platform_id = NULL;
    cl_uint ret_num_devices;
    cl_uint ret_num_platforms;
    cl_int ret;
    size_t source_size;
    char *source_str;
    fp = fopen("calc_double.cl", "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);


    /*Initialization*/
    /* Get Platform and Device Info */
    ret = clGetPlatformIDs(1, &platform_id, &ret_num_platforms);
    ret = clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id, &ret_num_devices);

    /* Create OpenCL context */
    context = clCreateContext(NULL, 1, &device_id, NULL, NULL, &ret);

    /* Create Command Queue */
    command_queue = clCreateCommandQueue(context, device_id, 0, &ret);

    /*Initialization complete*/

    cl_mem inputBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, MEM_SIZE * sizeof(double),(void *) input, NULL);
    cl_mem outputBuffer = clCreateBuffer(context, CL_MEM_WRITE_ONLY , MEM_SIZE * sizeof(double), NULL, NULL);


    /* Create Kernel Program from the source */
    program = clCreateProgramWithSource(context, 1, (const char **)&source_str,(const size_t *)&source_size, &ret);

    /* Build Kernel Program */
    ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);

    /* Create OpenCL Kernel */
    kernel = clCreateKernel(program, "calc", &ret);

    /* Set OpenCL Kernel Parameters */
    ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&inputBuffer);
    ret = clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&outputBuffer);

    /* Execute OpenCL Kernel */
    ret = clEnqueueTask(command_queue, kernel, 0, NULL,NULL);
    double x = 10, io;
    size_t global_work_size[1] = {MEM_SIZE};
    for (io = 0; io < x; io++) {
        inputBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, MEM_SIZE * sizeof(double),(void *) output, NULL);
        ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&inputBuffer);
        ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL, global_work_size, NULL, 0, NULL, NULL);

        ret = clEnqueueReadBuffer(command_queue, outputBuffer, CL_TRUE, 0, MEM_SIZE * sizeof(double), output, 0, NULL, NULL);
        for (go = 0; go < MEM_SIZE; go++) {
            printf("output[%d] = %lf\n",io, go, output[go]);
        }
        printf("\n\n");

    }


    /* Finalization */
    ret = clFlush(command_queue);
    ret = clFinish(command_queue);
    ret = clReleaseKernel(kernel);
    ret = clReleaseProgram(program);
    ret = clReleaseMemObject(inputBuffer);
    ret = clReleaseMemObject(outputBuffer);
    ret = clReleaseCommandQueue(command_queue);
    ret = clReleaseContext(context);

    return 1;
}

以上代码的calc_double.cl文件：

__kernel void calc(__global double* in, __global double* out)
{
    int i;
    for (i = 0; i < 10; i++) {
        out[i] = in[i] + 1;
    }
}

程序编译：

gcc program.c -o doublesimulation -l OpenCL -I /usr/local/test/AMD-APP-SDK-v2.9-RC-lnx32/include/ -L /usr/local/test/AMD-APP-SDK-v2.9-RC-lnx32/lib/x86

我也尝试添加以下检查，但没有帮助：

#ifdef cl_khr_fp64
    #pragma OPENCL EXTENSION cl_khr_fp64 : enable
#elif defined(cl_amd_fp64)
    #pragma OPENCL EXTENSION cl_amd_fp64 : enable
#else
    #error "Double precision floating point not supported by OpenCL implementation."
#endif

操作系统：Centos5 32bit AMD显卡

Answer 1

检查您的 GPU 是否支持双精度：

cl_uint native_double_width;    
clGetDeviceInfo(device_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_DOUBLE, sizeof(cl_uint), &native_double_width, NULL);

if(native_double_width == 0){
    printf("No double precision support.\n");
}

如果原生双向量宽度等于 0，则不支持双精度 (clGetDeviceInfo description)