CUDA三维数组

http://hpcbbs.it168.com/forum.php?mod=viewthread&tid=1643

根据上面链接的帖子研究了下三维数组，就像他自己说的一样是有问题的，我自己修改了下，结果终于正确了。大家有兴趣的可以对照着看看。

整个过程关键参考了这篇文章http://www.xuebuyuan.com/685353.html

#include <stdio.h>
//#include <cutil.h>
#include <helper_cuda.h>//替代cutil.h
#include <cuda_runtime.h>

#define Width  2
#define Height 2
#define Depth  2
int bmp1[Width][Height][Depth]={11,22,33,44,55,66,77,88};
int bmp2[Width][Height][Depth];
int main(){

    cudaPitchedPtr devPitchedPtr;
    cudaExtent extent = make_cudaExtent(Width * sizeof(int), Height, Depth);
    cudaMalloc3D(&devPitchedPtr, extent);
    cudaMemset3D(devPitchedPtr, 0, extent);


    cudaError status;
    cudaMemcpy3DParms HostToDev = { 0 };
    HostToDev.srcPtr = make_cudaPitchedPtr((void*)bmp1,Width * sizeof(int), Width, Height);
    HostToDev.dstPtr = devPitchedPtr;
    HostToDev.extent = extent;
    HostToDev.kind = cudaMemcpyHostToDevice;
    status = cudaMemcpy3D(&HostToDev);
    if (status != cudaSuccess){
        fprintf(stderr, "MemcpyHtD: %s
", cudaGetErrorString(status));
    }

    cudaMemcpy3DParms DevToHost = { 0 };
    DevToHost.srcPtr = devPitchedPtr;
    DevToHost.dstPtr = make_cudaPitchedPtr((void*)bmp2, Width * sizeof(int), Width, Height);
    DevToHost.extent = extent;
    DevToHost.kind = cudaMemcpyDeviceToHost;
    status = cudaMemcpy3D(&DevToHost);
    if (status != cudaSuccess){
        fprintf(stderr, "MemcpyHtD: %s
", cudaGetErrorString(status));
    }
    cudaFree(&devPitchedPtr);

    int i, j, k;
    for (i = 0; i<2; i++)
    for (j = 0; j<2; j++)
    for (k = 0; k<2; k++)
        printf("bmp2[%d][%d][%d]=%d
", i, j, k, bmp2[i][j][k]);

    return 0;
}