使用 std::vector 的 OpenCL 矩阵乘法

Question

我正在尝试使用 OpenCl 执行向量乘法，而代码本身似乎有效，但返回的结果要么是垃圾，要么是零。据我所知，似乎内核没有收到正确的值，对我来说有些不明显的东西是我在这里遗漏的，它是什么？我以为这是我分配缓冲区的方式，但我不确定。

#define CL_USE_DEPRECATED_OPENCL_2_0_APIS

#include <iostream>
#include <fstream>
#include <sstream>
#include "./cl.hpp"

void populate_vector(std::vector<float> &vect, std::stringstream &readStream) {

    std::string x;
    std::string fStripped;
    float readFloat;

    while(std::getline(readStream, x, ',')){
        std::stringstream elementStream;
        elementStream << x;
        std::getline(elementStream, fStripped, 'f');
        elementStream << fStripped;
        elementStream >> readFloat;
        vect.push_back(readFloat);
    }

}

int main()
{
    std::vector<cl::Platform> platforms;
    cl::Platform::get(&platforms);
    if(platforms.empty()){
        throw std::runtime_error("No Platforms found, check OpenCL installation.");
    }

    cl::Platform platform = platforms[0];
    std::cout << "Using Platform: " << platform.getInfo<CL_PLATFORM_NAME>() << std::endl;
    std::vector<cl::Device> devices;
    platform.getDevices(CL_DEVICE_TYPE_GPU, &devices);
    if(devices.empty()){
        throw std::runtime_error ("No Devices Found, check installation.");
    }
    cl::Device device = devices[0];

    // Create an execusion context
    cl::Context context(device);

    cl::CommandQueue queue(context,device);


    // Load the kernel sources, use global memory
    std::ifstream fs("mCrossProd.cl");
    if(!fs.is_open()){
        throw  std::runtime_error("Cannot open kernel source file.");
    }

    // Extract kernel code
    std::stringstream ss;
    ss << fs.rdbuf();
    auto code = ss.str();
    cl::Program::Sources sources;
    sources.push_back({code.c_str(), code.length()});
    fs.close();

    // Build the kernel
    cl::Program program(context, sources);
    if(program.build({device})!= CL_SUCCESS){
        std::cout << " Error building: " << program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(device) << "\n";
        exit(1);
    }

    // Output matrix dimensions
    int M = 4, N = 3, K = 6;
    int A_dims = M * K;
    int B_dims = N * K;
    int C_dims = M * N;

    // Create buffers for device
    cl::Buffer buffer_A(context,CL_MEM_READ_WRITE,sizeof(float)*A_dims);
    cl::Buffer buffer_B(context,CL_MEM_READ_WRITE,sizeof(float)*B_dims);
    cl::Buffer buffer_C(context,CL_MEM_READ_WRITE,sizeof(float)*C_dims);

    std::string s;
    std::vector<float> A;
    std::vector<float> B;
    std::vector<float> C;
    std::ifstream infile("matrices.txt");
    std::cout << "Opened file to read" << '\n';


    std::getline(infile,s);
    //std::cout << s;
    std::stringstream mss(s);
    populate_vector(A, mss);   
    std::copy(A.begin(), A.end(), std::ostream_iterator<float>(std::cout, ", "));
    std::cout << '\n';

    mss.str("");
    mss.clear();
    std::getline(infile,s);
    mss << s;
    populate_vector(B, mss);  
    std::copy(B.begin(), B.end(), std::ostream_iterator<float>(std::cout, ", "));
    std::cout << '\n';

    mss.str("");
    mss.clear();
    std::getline(infile,s);
    mss << s;
    populate_vector(C, mss);
    std::copy(C.begin(), C.end(), std::ostream_iterator<float>(std::cout, ", "));
    std::cout << '\n';




    //write arrays A and B to the device
    queue.enqueueWriteBuffer(buffer_A,CL_TRUE,0,A.size()*sizeof(float),&A);
    queue.enqueueWriteBuffer(buffer_B,CL_TRUE,0,B.size()*sizeof(float),&B);

    std::cout << A.size() * sizeof(float) << '\n';
    std::cout << B.size() * sizeof(float) << '\n';
    std::cout << C.size() * sizeof(float) << '\n';

    // Select kernel, pass arguments
    cl::Kernel kernel = cl::Kernel(program, "mCrossProd");
    kernel.setArg(0, M);
    kernel.setArg(1, N);
    kernel.setArg(2, K);
    kernel.setArg(3, buffer_A);
    kernel.setArg(4, buffer_B);
    kernel.setArg(5, buffer_C);

    // Execute kernel
    if( queue.enqueueNDRangeKernel(kernel,cl::NullRange,cl::NDRange(M,N),cl::NDRange(1,1)) != CL_SUCCESS )
    {
        std::cout << "Failed to launch kernel" << std::endl;
        exit(1);
    }
    queue.finish();

    // read result C from the device to array C
    queue.enqueueReadBuffer(buffer_C,CL_TRUE,0,C.size(),&C[0]);
    std::cout << C.size() << std::endl;
    std::cout << C_dims << std::endl;
    std::cout << M << " " << N << std::endl;
    std::cout << "\nThe solution is" << std::endl;
    std::copy(C.begin(), C.end(), std::ostream_iterator<float>(std::cout, ", "));
    std::cout << '\n';

     for(int i = 0; i < M; i++) {
        for(int j = 0; j < N; j++) {
            std::cout << "C[" + std::to_string(i*N+j) + "] = ";
            std::cout << C[i*N+j] << " ";
        }
        std::cout << std::endl;
    }
}

测试内核

    __kernel void mCrossProd(const int M, const int N, const int K, __global float* A, __global float* B, __global float* C) {
    int const i = get_global_id(0);
    int const j = get_global_id(1);
    int const debug_elem_id = 3; // purely for debug purposes.

    for(int k = 0; k < K; k++){
        C[i*N+j] += A[i*K+k] * B[N*k+j];
        if((i*N+j)==debug_elem_id)
        {   
            //printf("PROD, i = %d, j = %d, k = %d, N = %d\n", i,j,k,N);
            printf("PROD, %.2f\n", A[i*K+k] * B[N*k+j]);
            printf("SUM: %.2f\n", C[i*N+j]);
        }
    }
}

matrices.txt的内容

1.5f, 1.0f, 2.0f, 2.0f, 4.0f, 1.0f, 4.0f, 2.0f, 1.0f, 1.0f, 0.0f, 0.0f, 3.0f, 2.0f, 5.0f, 1.0f, 1.0f, 1.0f, 0.0f, 0.0f, 0.0f, 2.0f, 1.0f, 1.0f
1.5f, 2.0f, 4.0f, 1.0f, 1.0f, 2.0f, 4.0f, 2.0f, 1.0f, 0.0f, 0.0f, 1.0f, 9.0f, 2.0f, 1.0f, 2.0f, 1.0f, 0.0f
0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f

Answer 1

    queue.enqueueWriteBuffer(buffer_A,CL_TRUE,0,A.size()*sizeof(float),&A);
    queue.enqueueWriteBuffer(buffer_B,CL_TRUE,0,B.size()*sizeof(float),&B);

&A应该是A.data()或者&A[0]，我推荐第一个

使用 std::vector 的 OpenCL 矩阵乘法

OpenCL Matrix Multiplication Using std::vector

c++

opencl

matrix-multiplication