OpenCL矩陣相乘輸入形式

一．核函式（運行在gpu上）

__kernel void matrix_mult(
    const int Ndim,
    const int Mdim,
    const int Pdim,
    __global const float* A,
    __global const float* B,
    __global float* C)
{
//兩個執行緒同時執行
    int i = get_global_id(0);
    int j = get_global_id(1);

    int k;
    float tmp;

    if ((i < Ndim) && (j < Mdim)) {
        tmp = 0.0;
        for (k = 0; k < Pdim; k++)
            tmp += A[i * Pdim + k] * B[k * Mdim + j];
        C[i * Mdim + j] = tmp;
    }
}

二． 輸入矩陣形式（博主列舉了三種形式的矩陣輸入，代碼中通過注釋給出：（1）自動生成（2）直接給定（3）動態輸入）

```c
#include <CL/cl.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <iostream>
#include <fstream>

using namespace std;

#define NWITEMS 6

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

//把文本檔案讀入一個 string 中
int convertToString(const char* filename, std::string& s)
{
    size_t size;
    char* str;
    std::fstream f(filename, (std::fstream::in | std::fstream::binary));
    if (f.is_open())
    {
        size_t fileSize;
        f.seekg(0, std::fstream::end);
        size = fileSize = (size_t)f.tellg();
        f.seekg(0, std::fstream::beg);
        str = new char[size + 1];
        if (!str)
        {
            f.close();
            return NULL;
        }
        f.read(str, fileSize);
        f.close();
        str[size] = '\0';
        s = str;
        delete[] str;
        return 0;
    }
    printf("Error: Failed to open file %s\n", filename);
    return 1;
}

int main()
{
    cl_uint status;
    cl_platform_id platform;

    //創建平臺物件
    status = clGetPlatformIDs(1, &platform, NULL);
    cl_device_id device;
    //創建 GPU 設備
    clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL);
    //創建context
    cl_context context = clCreateContext(NULL, 1, &device, NULL, NULL, NULL);
    //創建命令佇列
    cl_command_queue commandQueue = clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, NULL);

    if (commandQueue == NULL)
        perror("Failed to create commandQueue for device 0.");

    //建立要傳入從機的資料
    /********  創建內核和記憶體物件 ********/

    //輸入定義矩陣的長寬
    int Ndim, Pdim, Mdim;
    cout << "請輸入A矩陣長" << endl;
    cin >> Ndim;
    cout << "請輸入A矩陣寬" << endl;
    cin >> Pdim;
    cout << "請輸入B矩陣寬" << endl;
    cin >> Mdim;
    int szA = Ndim * Pdim;
    int szB = Pdim * Mdim;
    int szC = Ndim * Mdim;

    float* A;
    float* B;
    float* C;

    A = (float*)malloc(szA * sizeof(float));
    B = (float*)malloc(szB * sizeof(float));
    C = (float*)malloc(szC * sizeof(float));
    int i, j;
    /* （1）回圈矩陣
          for (i = 0; i < szA; i++)
                A[i] = (float)((float)i + 1.0);
            for (i = 0; i < szB; i++)
                B[i] = (float)((float)i + 1.0);
      （2）測驗手動輸入的矩陣
                A[0] = 1;
                A[1] = 6;
                A[2] = 2;
                A[3] = 4;
                B[0] = 2;
                B[1] = 4;
                B[2] = 1;
                B[3] = 4;
下面是第三種，手動輸入：
        */

    cout << "請輸入矩陣A,按陣列形式單個輸入" << endl;
    for (int i = 0; i < szA; i++)
        cin >> A[i];
    cout << "請輸入矩陣B,按陣列形式單個輸入" << endl;
    for (int i = 0; i < szB; i++)
        cin >> B[i];




    //創建三個 OpenCL 記憶體物件，并把buf1 的內容通過隱式拷貝的方式
    //拷貝到clbuf1, buf2 的內容通過顯示拷貝的方式拷貝到clbuf2
    cl_mem memObjects[3] = { 0, 0, 0 };
    memObjects[0] = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
        sizeof(float) * szA, A, NULL);
    memObjects[1] = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
        sizeof(float) * szB, B, NULL);
    memObjects[2] = clCreateBuffer(context, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR,
        sizeof(float) * szC, C, NULL);
    if (memObjects[0] == NULL || memObjects[1] == NULL || memObjects[2] == NULL)
        perror("Error in clCreateBuffer.\n");

    const char* filename = "lhl.cl";
    std::string sourceStr;
    status = convertToString(filename, sourceStr);
    if (status)
        cout << status << "  !!!!!!!!" << endl;
    const char* source = sourceStr.c_str();
    size_t sourceSize[] = { strlen(source) };
    //創建程式物件
    cl_program program = clCreateProgramWithSource(context, 1, &source, sourceSize, NULL);
    //編譯程式物件
    status = clBuildProgram(program, 1, &device, NULL, NULL, NULL);
    if (status)
        cout << status << "  !!!!!!!!" << endl;
    if (status != 0)
    {
        printf("clBuild failed:%d\n", status);
        char tbuf[0x10000];
        clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0x10000, tbuf,
            NULL);
        printf("\n%s\n", tbuf);
        //return ?1;
    }

    //創建 Kernel 物件
    cl_kernel kernel = clCreateKernel(program, "matrix_mult", NULL);

    //設定 Kernel 引數
    cl_int clnum = NWITEMS;
    status = clSetKernelArg(kernel, 0, sizeof(int), &Ndim);
    status = clSetKernelArg(kernel, 1, sizeof(int), &Mdim);
    status = clSetKernelArg(kernel, 2, sizeof(int), &Pdim);
    status = clSetKernelArg(kernel, 3, sizeof(cl_mem), &memObjects[0]);
    status = clSetKernelArg(kernel, 4, sizeof(cl_mem), &memObjects[1]);
    status = clSetKernelArg(kernel, 5, sizeof(cl_mem), &memObjects[2]);
    if (status)
        cout << "引數設定錯誤" << endl;

    //執行 kernel
    size_t global[2];
    cl_event prof_event;
    cl_ulong ev_start_time = (cl_ulong)0;
    cl_ulong ev_end_time = (cl_ulong)0;
    double rum_time;
    global[0] = (size_t)Ndim;
    global[1] = (size_t)Mdim;
    status = clEnqueueNDRangeKernel(commandQueue, kernel, 2, NULL,
        global, NULL, 0, NULL, &prof_event);
    if (status)
        cout << "執行內核時錯誤" << endl;
    clFinish(commandQueue);

    //讀取時間
    status = clGetEventProfilingInfo(prof_event, CL_PROFILING_COMMAND_QUEUED,
        sizeof(cl_ulong), &ev_start_time, NULL);
    status = clGetEventProfilingInfo(prof_event, CL_PROFILING_COMMAND_END,
        sizeof(cl_ulong), &ev_end_time, NULL);
    if (status)
        perror("讀取時間的時候發生錯誤\n");
    rum_time = (double)(ev_end_time - ev_start_time);
    cout << "執行時間為:" << rum_time << endl;

    //資料拷回 host 記憶體
    status = clEnqueueReadBuffer(commandQueue, memObjects[2], CL_TRUE, 0,
        sizeof(float) * szC, C, 0, NULL, NULL);
    if (status)
        perror("讀回資料的時候發生錯誤\n");

    //結果顯示
    printf("\nArray A:\n");
    for (i = 0; i < Ndim; i++) {
        for (j = 0; j < Pdim; j++)
            printf("%.1f\t", A[i * Pdim + j]);
        printf("\n");
    }
    printf("\nArray B:\n");
    for (i = 0; i < Pdim; i++) {
        for (j = 0; j < Mdim; j++)
            printf("%.1f\t", B[i * Mdim + j]);
        printf("\n");
    }
    printf("\nArray C:\n");
    for (i = 0; i < Ndim; i++) {
        for (j = 0; j < Mdim; j++)
            printf("%.1f\t", C[i * Mdim + j]);
        printf("\n");
    }




    cout << endl;

    if (A)
        free(A);
    if (B)
        free(B);
    if (C)
        free(C);

    //洗掉 OpenCL 資源物件
    clReleaseMemObject(memObjects[2]);
    clReleaseMemObject(memObjects[1]);
    clReleaseMemObject(memObjects[0]);
    clReleaseProgram(program);
    clReleaseCommandQueue(commandQueue);
    clReleaseContext(context);
    system("pause");

    return 0;
}