CUDA 6.5 (第1.1章)结构的cudaMemcpy数组导致“无效参数”

Question

代码：

Octree_t* tarr = (Octree*)malloc(nodes * sizeof(Octree_t));
//nodearray[2]->test = 42;
for (int i = 0; i < cnodes; i++)
{
    tarr[i] = *nodearray[i];
}
//printf(" test: %d\n",tarr[2].test); //returns 42

cudaError_t err;
Octree_t*     gpu_nodearray;

//allocate storage on gpu
err = cudaMalloc( &gpu_nodearray, cnodes * sizeof(Octree_t) != cudaSuccess  );
if ( err != cudaSuccess)
{
    printf("1: %s\n", cudaGetErrorString(err));
    return;
}
err = cudaMemcpy(gpu_nodearray, tarr,  cnodes * sizeof(Octree_t), cudaMemcpyHostToDevice);
if ( err != cudaSuccess)
{
    printf("3: %s\n", cudaGetErrorString(err));
    return;
}

它编译得很好，但是当我运行它时，它返回"3:无效参数“。所以很明显，cudaMempy有问题。我已经在这里看过类似的帖子了，我找不到我犯的错误。

谢谢你的帮助。

Answer 1

线

err = cudaMalloc( &gpu_nodearray, cnodes * sizeof(Octree_t) != cudaSuccess  );

应该是

err = cudaMalloc( &gpu_nodearray, cnodes * sizeof(Octree_t));