when and where blockIdx.x, blockDim.x and threadIdx.x are defined and assigned?

[seven...@gmail.com]

there's a piece of code in caffe/util/device_alternate.hpp:

// CUDA: grid stride looping
#define CUDA_KERNEL_LOOP(i, n) \
  for (int i = blockIdx.x * blockDim.x + threadIdx.x; \
       i < (n); \
       i += blockDim.x * gridDim.x)

I do not know when and where blockIdx.x, blockDim.x and threadIdx.x are defined and assigned. 

Anyone can give some advises? thank you very much~

[Przemek D]

This is specific to CUDA, you should check out some basics tutorial on it, like this one.