Question

我正在尝试使用存储在aprox文件中的SURF点训练神经网络。 50GB。我当时使用python生成器和熊猫来训练我的网络，但我发现我的GPU的内存已满，但是在那里没有任何处理。

#include <stdio.h>
#include <stdlib.h>
#include <vector>
#include "glm/glm.hpp"

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

#define MAX_SOURCE_SIZE (0x100000)

int main(void) 
{
    std::vector<glm::vec3> values;
    values.resize(256);

    // Load the kernel source code into the array source_str
    FILE *fp;
    char *source_str;
    size_t source_size;

    fp = fopen("E:/Dev/fill_array_kernel.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 );

    // Get platform and device information
    cl_platform_id platform_id = NULL;
    cl_device_id device_id = NULL;   
    cl_uint ret_num_devices;
    cl_uint ret_num_platforms;
    cl_int ret = clGetPlatformIDs(1, &platform_id, &ret_num_platforms);
    ret = clGetDeviceIDs( platform_id, CL_DEVICE_TYPE_ALL, 1, 
            &device_id, &ret_num_devices);

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

    // Create a command queue
    cl_command_queue command_queue = clCreateCommandQueue(context, device_id, 0, &ret);

    // Create memory buffers on the device for each vector 
    cl_mem output_mem = clCreateBuffer(context, CL_MEM_WRITE_ONLY, values.size() * sizeof(glm::vec3), NULL, &ret);

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

    // Build the program
    ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);
    if(ret != CL_SUCCESS)
    {
        cl_build_status build_status;
        ret = clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_STATUS, sizeof(cl_build_status), &build_status, NULL);
        size_t ret_val_size;
        ret = clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, 0, NULL, &ret_val_size);
        char *build_log = (char*)malloc(sizeof(char)*(ret_val_size + 1));
        ret = clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, ret_val_size, build_log, NULL);
        build_log[ret_val_size] = '\0';
        printf("%s\n", build_log);
        free(build_log);
        return -1;
    }

    // Create the OpenCL kernel
    cl_kernel kernel = clCreateKernel(program, "fill_array", &ret);

    // Set the arguments of the kernel
    ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&output_mem);

    // Execute the OpenCL kernel on the list
    size_t global_item_size = values.size(); // Process the entire lists
    size_t local_item_size = 64; // Process in groups of 64
    ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL, 
            &global_item_size, &local_item_size, 0, NULL, NULL);

    // Read the memory buffer C on the device to the local variable C
    ret = clEnqueueReadBuffer(command_queue, output_mem, CL_TRUE, 0, values.size() * sizeof(glm::vec3), values.data(), 0, NULL, NULL);

    // Clean up
    ret = clFlush(command_queue);
    ret = clFinish(command_queue);
    ret = clReleaseKernel(kernel);
    ret = clReleaseProgram(program);
    ret = clReleaseMemObject(output_mem);
    ret = clReleaseCommandQueue(command_queue);
    ret = clReleaseContext(context);
    return 0;
}

IIRC，我应该使用Sequence子类，但是我没有找到一种以块方式加载文件的方法，这样就不会出现Memory异常，并且可以使用GPU而不是CPU。

使用fit_generator和大型数据集在GPU上进行培训

0 个答案: