我试图复制从主机到设备的图像的RGB数据(假设每个都是int)。这是我的代码的一部分
int *img_redd,*img_greend,*img_blued;//d denotes device
int **img_redh,**img_greenh,**img_blueh;// h denotes host
//Initialize+ copy values into the arrays pointed by img_redh,img_greenh etc
// then Copy the values of RGB into host array <here>
//Allocating memory on device below
cudaMallocPitch((void**)&img_redd,&pitch1,img_width*sizeof(int),img_height);
cudaMallocPitch((void**)&img_greend,&pitch2,img_width*sizeof(int),img_height);
cudaMallocPitch((void**)&img_blued,&pitch3,img_width*sizeof(int),img_height);
// copy it to CUDA device
cudaMemcpy2D(img_redd,pitch1,img_redh[0],img_width*sizeof(int),img_width*sizeof(int),img_height,cudaMemcpyHostToDevice);
//I even tried with just img_redh above
//Similarly for green and blue
cudaMallocpitch工作正常但它在cudamemcpy2d行崩溃并打开host_runtime.h并指向
static void __cudaUnregisterBinaryUtil(void)
{
__cudaUnregisterFatBinary(__cudaFatCubinHandle);
}
我觉得内存分配背后的逻辑很好。任何评论可能导致崩溃的原因是什么?
答案 0 :(得分:1)
听起来你正在为img_redh
的多维数组使用Iliffe向量。尝试使用常规多维数组(int* img_redh = (int*)malloc(img_width*img_height*sizeof(int)
)