cuBLAS cublasSgemv“分段错误” [英] cuBLAS cublasSgemv “Segmentation fault"

问题描述

运行cublasSegmv时出现分段错误，我的GPU是K20Xm，这是我的代码。

I have gotten a segmentation fault when running cublasSegmv.My GPU is K20Xm.Here is my code.

float *a, *x, *y;
int NUM_VEC = 8;
y = (float*)malloc(sizeof(float) * rows * NUM_VEC);
a = (float*)malloc(sizeof(float) * rows * cols);
x = (float*)malloc(sizeof(float) * cols * NUM_VEC);
get_mat_random(a, rows, cols);
get_vec_random(x, cols * NUM_VEC);

float *d_a = 0;
float *d_x = 0;
float *d_y = 0;

cudaMalloc((void **)&d_a, rows * cols * sizeof(float);
cudaMalloc((void **)&d_x, cols * NUM_VEC * sizeof(float);
cudaMalloc((void **)&d_y, rows * NUM_VEC * sizeof(float);
cublasSetVector(rows * cols, sizeof(float), a, 1, d_a, 1);
cublasSetVector(NUM_VEC * cols, sizeof(float), x, 1, d_x, 1);
cublasSetVector(NUM_VEC * rows, sizeof(float), y, 1, d_y, 1);
float alpha = 1.0f;
for (int i = 0; i < NUM_VEC; i++) {
  cublasSgemv(handle, CUBLAS_OP_T, cols, rows, &alpha, d_a, rows, d_x + i * cols, 1,0, d_y + i * rows, 1);
}

推荐答案

在我的有限测试中，该错误是因为 cublasSgemv beta 参数c>不能为 NULL 。您应在主机或设备上为 beta 变量分配内存。

In my limited testing, the error is because the beta parameter of cublasSgemv cannot be NULL. You should allocate memory for the beta variable either on host or device. Following is the code which I used to reproduce and fix the error.

#include <cstdio>
#include <iostream>
#include <cuda_runtime.h>
#include <cublas_v2.h>
#include <cstdlib>

using namespace std;

void get_vec_random(float* a, int count)
{
    for(int i=0; i<count; i++)
        a[i] = rand() / float(RAND_MAX);    
}

void get_mat_random(float* a, int rows, int cols)
{
    get_vec_random(a, rows * cols);
}

int main(int argc, char** argv)
{
    int rows = 10, cols = 10;

    cublasHandle_t handle;
    cublasCreate(&handle);


    float *a, *x, *y;
    int NUM_VEC = 8;
    y = (float*)malloc(sizeof(float) * rows * NUM_VEC);
    a = (float*)malloc(sizeof(float) * rows * cols);
    x = (float*)malloc(sizeof(float) * cols * NUM_VEC);
    get_mat_random(a, rows, cols);
    get_vec_random(x, cols * NUM_VEC);

    float *d_a = 0;
    float *d_x = 0;
    float *d_y = 0;

    cudaMalloc((void **)&d_a, rows * cols * sizeof(float));
    cudaMalloc((void **)&d_x, cols * NUM_VEC * sizeof(float));
    cudaMalloc((void **)&d_y, rows * NUM_VEC * sizeof(float));


    cublasSetVector(rows * cols, sizeof(float), a, 1, d_a, 1);
    cublasSetVector(NUM_VEC * cols, sizeof(float), x, 1, d_x, 1);
    cublasSetVector(NUM_VEC * rows, sizeof(float), y, 1, d_y, 1);

    float alpha = 1.0f, beta = 1.0f;

    cublasSetPointerMode(handle, CUBLAS_POINTER_MODE_HOST);
    for (int i = 0; i < NUM_VEC; i++) 
    {
        cublasSgemv(handle, 
                    CUBLAS_OP_T, 
                    cols, 
                    rows, 
                    &alpha, 
                    d_a, 
                    rows, 
                    d_x + i * cols, 
                    1, 
                    &beta, 
                    d_y + i * rows, 
                    1);
    }
    return 0;
}

希望这可以解决问题。

这篇关于cuBLAS cublasSgemv“分段错误”的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！