c++ 不转置的行优先数据CUBLAS矩阵乘法

dy1byipe  于 2023-05-30  发布在  其他
关注(0)|答案(1)|浏览(174)

我目前正在尝试在我的GPU上使用CUBLAS实现矩阵乘法。

***对于方阵和特定大小的输入,它工作得很好,但对于其他人,最后一行不会返回(并且包含0,因为它是我实现的方式)。

我假设这是cublasSgemm的分配或语法的问题,但我找不到它在哪里。

注意:如果您不熟悉CUBLAS:它是column-majored,这就是为什么看起来操作是以另一种方式执行的原因。

任何帮助将不胜感激。

乘法源代码

  • 请注意,gpuErrchkcublasErrchk在这里当然无关。
#include <cuda.h>
#include <cuda_runtime.h>
#include <cublas_v2.h>

#include <vector>

std::vector<float> CUDA_mult_MAT(const std::vector<float> &data_1 , const uint64_t data_1_rows, const uint64_t data_1_columns,
                                 const std::vector<float> &data_2 , const uint64_t data_2_rows, const uint64_t data_2_columns){

    cublasHandle_t handle;

    cublasErrchk(cublasCreate(&handle));

    std::vector<float> result(data_1_rows * data_2_columns); //Vector holding the result of the multiplication

    /*----------------------------------------------------------------------------------------------*/

    float* GPU_data_1 = NULL;
    gpuErrchk(cudaMalloc((void**)&GPU_data_1 , data_1.size()*sizeof(float))); //Allocate memory on the GPU
    gpuErrchk(cudaMemcpy(GPU_data_1, data_1.data(), data_1.size()*sizeof(float), cudaMemcpyHostToDevice)); //Copy data from data_1 to GPU_data_1

    float* GPU_data_2 = NULL;
    gpuErrchk(cudaMalloc((void**)&GPU_data_2 ,data_2.size()*sizeof(float))); //Allocate memory on the GPU
    gpuErrchk(cudaMemcpy(GPU_data_2, data_2.data(), data_2.size()*sizeof(float), cudaMemcpyHostToDevice));//Copy data from data_2 to GPU_data_2

    float* GPU_result = NULL;
    gpuErrchk(cudaMalloc((void**)&GPU_result , result.size()*sizeof(float))); //Allocate memory on the GPU

    /*----------------------------------------------------------------------------------------------*/

    const float alpha = 1.f; 
    const float beta = 0.f;

    cublasErrchk(
               cublasSgemm(handle , CUBLAS_OP_N , CUBLAS_OP_N,
                           data_2_columns , data_2_rows ,data_1_columns,
                           &alpha , GPU_data_2 , data_2_columns,
                           GPU_data_1 , data_1_columns,
                           &beta , GPU_result , data_1_rows)
             ); //Perform multiplication 


    gpuErrchk(cudaMemcpy(result.data() , GPU_result , result.size() * sizeof(float) , cudaMemcpyDeviceToHost)); //Copy back to the vector 'result'

    gpuErrchk(cudaFree(GPU_data_1)); //Free GPU memory
    gpuErrchk(cudaFree(GPU_data_2)); //Free GPU memory
    gpuErrchk(cudaFree(GPU_result)); //Free GPU memory

    cublasErrchk(cublasDestroy_v2(handle)); 

    return result;

}

测试输入

#include <iostream>

#include <vector>

int main(){

    const auto r1 = CUDA_mult_MAT({1 , 2 , 3 , 4 , 5 , 6} , 2 , 3 ,
                                  {7 , 8 , 9 , 10 , 11 , 12} , 3 , 2);
    /*
    Product:
              7  8
    1 2 3  x  9  10
    4 5 6     11 12
    
    */

    for(const auto& value: r1){std::cout << value << " " ;}
    std::cout << std::endl;

    const auto r2 = CUDA_mult_MAT({7 , 8 , 9 , 10 , 11 , 12} , 3 , 2 ,
                                  {1 , 2 , 3 , 4 , 5 , 6} , 2 , 3);
    /*
    Product:
    7  8   
    9  10  x  1  2  3
    11 12     4  5  6
    */

    for(const auto& value: r2){std::cout << value << " " ;}
    std::cout << std::endl;

    return 0;
}

输出

程序打印:

58 64 139 154 
39 54 69 49 68 87 0 0 0
                  ^~~~~~~

预期值:

58 64 139 154 
39 54 69 49 68 87 59 82 105
                  ^~~~~~~
1zmg4dgp

1zmg4dgp1#

我们可以通过不同的方式观察您的CUBLAS使用问题。
首先,研究CUBLAS Sgemm文档,我们看到3个参数mnk出现,按此顺序紧接在转置说明符之后:

cublasStatus_t cublasSgemm(cublasHandle_t handle,
                       cublasOperation_t transa, cublasOperation_t transb,
                       int m, int n, int k, 
                           ^      ^      ^

我们还观察到矩阵维数由下式给出:
A、B和C是以列为主格式存储的矩阵,维度为op(A)m × k、op(B)k × n和Cm × n,
因此,第一输入矩阵的维度为m x k,第二输入矩阵的维度为k x n,并且输出矩阵的维度为m x n
让我们先关注一下输出矩阵。考虑到它的尺寸是使用mn参数指定的,它不可能是正确的(假设在非正方形的情况下)只传递**data_2尺寸:

cublasSgemm(handle , CUBLAS_OP_N , CUBLAS_OP_N,
                       data_2_columns , data_2_rows ,data_1_columns,
                       ^^^^^^^^^^^^^^   ^^^^^^^^^^^

其次,从错误检查的Angular 来看,通过使用cuda-memcheck运行代码,可以快速估计CUBLAS调用是否有问题。报告的第一个错误如下:

$ cuda-memcheck ./t23
========= CUDA-MEMCHECK
========= Invalid __global__ read of size 4
=========     at 0x000006f0 in void gemmSN_NN_kernel<float, int=256, int=4, int=2, int=8, int=3, int=4, bool=0, cublasGemvTensorStridedBatched<float const >, cublasGemvTensorStridedBatched<float>>(cublasGemmSmallNParams<float const , cublasGemvTensorStridedBatched<float const >, float>)
=========     by thread (64,0,0) in block (0,0,0)
=========     Address 0x7f9c30a2061c is out of bounds
=========     Device Frame:void gemmSN_NN_kernel<float, int=256, int=4, int=2, int=8, int=3, int=4, bool=0, cublasGemvTensorStridedBatched<float const >, cublasGemvTensorStridedBatched<float>>(cublasGemmSmallNParams<float const , cublasGemvTensorStridedBatched<float const >, float>) (void gemmSN_NN_kernel<float, int=256, int=4, int=2, int=8, int=3, int=4, bool=0, cublasGemvTensorStridedBatched<float const >, cublasGemvTensorStridedBatched<float>>(cublasGemmSmallNParams<float const , cublasGemvTensorStridedBatched<float const >, float>) : 0x6f0)
=========     Saved host backtrace up to driver entry point at kernel launch time
=========     Host Frame:/usr/lib/x86_64-linux-gnu/libcuda.so.1 (cuLaunchKernel + 0x2b8) [0x1e5cc8]
=========     Host Frame:/usr/local/cuda/lib64/libcublasLt.so.11 [0x1063c8b]
=========     Host Frame:/usr/local/cuda/lib64/libcublasLt.so.11 [0x10a9965]
=========     Host Frame:/usr/local/cuda/lib64/libcublasLt.so.11 [0x6bfacc]
=========     Host Frame:/usr/local/cuda/lib64/libcublasLt.so.11 [0x5fc7af]
=========     Host Frame:/usr/local/cuda/lib64/libcublasLt.so.11 [0x436c35]
=========     Host Frame:/usr/local/cuda/lib64/libcublasLt.so.11 (cublasLtMatmul + 0x60f) [0x43484f]
=========     Host Frame:/usr/local/cuda/lib64/libcublas.so.11 [0x9ef6db]
=========     Host Frame:/usr/local/cuda/lib64/libcublas.so.11 [0x50e4f0]
=========     Host Frame:/usr/local/cuda/lib64/libcublas.so.11 (cublasSgemm_v2 + 0x1ee) [0x50f29e]
=========     Host Frame:./t23 [0x7986]
=========     Host Frame:./t23 [0x7b4c]
=========     Host Frame:/lib/x86_64-linux-gnu/libc.so.6 (__libc_start_main + 0xe7) [0x21b97]
=========     Host Frame:./t23 [0x744a]
=========

当然,一个可能的解决方案是转置输入矩阵,因此它们是以列为主的顺序,CUBLAS提供了Sgemm选项来完成这一点(见上文)。然而,在我看来,你试图做的是C风格的行优先乘法,而不转置输入数组。有一篇文章here给出了如何做到这一点的描述。
当我将启发式应用到cublasSgemm()调用时,我得到了这个:

cublasSgemm(handle , CUBLAS_OP_N , CUBLAS_OP_N,
                       data_2_columns , data_1_rows ,data_1_columns,
                       &alpha , GPU_data_2 , data_2_columns,
                       GPU_data_1 , data_1_columns,
                       &beta , GPU_result , data_2_columns)

当我编译并运行你的代码时,我会得到这样的结果:

$ cuda-memcheck ./t23
========= CUDA-MEMCHECK
58 64 139 154
39 54 69 49 68 87 59 82 105
========= ERROR SUMMARY: 0 errors

相关问题