我正在尝试使用PyCUDA处理2D数组,我需要每个线程的x,y坐标.
这里和这里已经提出并回答了这个问题,但链接的解决方案对我来说对于超过块大小的2D数据不起作用.为什么？
这是我用来帮助解决这个问题的SourceModule:
mod = SourceModule("""
__global__ void kIndexTest(float *M, float *X, float*Y)
{
int bIdx = blockIdx.x + blockIdx.y * gridDim.x;
int idx = bIdx * (blockDim.x * blockDim.y) + (threadIdx.y * blockDim.x) + threadIdx.x;
/* this array shows me the unique thread indices */
M[idx] = idx;
/* these arrays should capture x, y for each unique index */
X[idx] = (blockDim.x * blockIdx.x) + threadIdx.x;
Y[idx] = (blockDim.y * blockIdx.y) + threadIdx.y; …