Question

我正在尝试使用CUDA从较小的矩阵构建更大的矩阵。我的任务是，给定两个矩阵，一个矩阵应该复制到顶部，第二个矩阵应该复制到底部。当我尝试时，我可以复制顶部矩阵但底部无法复制（或给出意想不到的结果，即0）。我不想改变块的数量请帮帮我

这是我的代码和

#include <stdio.h>
#include <stdlib.h>

#define N 5
#define BLOCK_DIM 3

__global__ void matrixCombine (int *a, int *b, int *c) {
    int col = blockIdx.x * blockDim.x + threadIdx.x;
    int row = blockIdx.y * blockDim.y + threadIdx.y;

    int index = col + row * N;

    if (col < N && row < N)     //copying upper matrix is working
        c[index] = a[index];
    if (col >= N && row >= N)  //copying lower matrix is NOT working
        c[index] = b[index];
}

void printMatrix(int a[N][N] )
{
    for(int i=0; i<N; i++){
        for (int j=0; j<N; j++){
            printf("%d\t", a[i][j] );
        }
        printf("\n");
    }
}

void printMatrixAns(int a[N][N] )
{
    for(int i=0; i<2*N; i++){
        for (int j=0; j<N; j++){
            printf("%d\t", a[i][j] );
        }
        printf("\n");
    }
}

int main() {
    int a[N][N], b[N][N], c[2*N][N];
    int *dev_a, *dev_b, *dev_c;

    int size = N * N * sizeof(int);
    int sizeofc = 2 * N * N * sizeof(int);

    for(int i=0; i<N; i++)
        for (int j=0; j<N; j++){
            a[i][j] = rand() % 256;
            b[i][j] = rand() % 256;
        }

    printf("Matrix A\n");
    printMatrix(a);
    printf("Matrix B\n");
    printMatrix(b);

    cudaMalloc((void**)&dev_a, size);
    cudaMalloc((void**)&dev_b, size);
    cudaMalloc((void**)&dev_c, sizeofc);

    cudaMemcpy(dev_a, a, size, cudaMemcpyHostToDevice);
    cudaMemcpy(dev_b, b, size, cudaMemcpyHostToDevice);

    dim3 dimBlock(BLOCK_DIM, BLOCK_DIM);
    dim3 dimGrid((N+dimBlock.x-1)/dimBlock.x, (N+dimBlock.y-1)/dimBlock.y);

    printf("dimGrid.x = %d, dimGrid.y = %d\n", dimGrid.x, dimGrid.y);

    matrixCombine<<<dimGrid,dimBlock>>>(dev_a,dev_b,dev_c);
    cudaDeviceSynchronize();
    cudaMemcpy(c, dev_c, sizeofc, cudaMemcpyDeviceToHost);

    printf("Matrix c\n");
    printMatrixAns(c);

    cudaFree(dev_a);
    cudaFree(dev_b);
    cudaFree(dev_c);
}

Answer 1

在你的内核中，这个：

int index = col + row * N;
if (col < N && row < N)     //copying upper matrix is working
    c[index] = a[index];
if (col >= N && row >= N)  //copying lower matrix is NOT working
    c[index] = b[index];

错了。 a和b都只有NxN元素和c 2NxN元素，因此会导致b和c中的超出内存访问权限，并且你只需要比NxN线程稍微启动一样，所以无法保证网格大小足以覆盖2NxN输出元素。如果您将a叠加在b上的2NxN矩阵中，您需要这样的内容：

c[row][col] = a[row][col]
c[row+N][col] = b[row][col]

row<N和col<N。对于稍微超过NxN个线程的网格，代码可能看起来像这样：

int index = col + row * N;

if (col < N && row < N) {
    c[index] = a[index];
    c[index + N*N] = b[index];
}

即。每个线程将a元素和b元素复制到c，a元素写入c[row][col]，b写入c[row+N][col]

的元素

使用CUDA将两个矩阵组合成更大的矩阵

1 个答案: