CUDA Snippets.

snippets.cuh

// Calculate the thread global position, taken it's thread position within the block.
__device__ __inline__ uint3 __getGlobalPosition (uint3 blockIdx, dim3 blockDim, uint3 threadIdx) {
	return make_uint3(blockIdx.x * blockDim.x + threadIdx.x, blockIdx.y * blockDim.y + threadIdx.y, blockIdx.z * blockDim.z + threadIdx.z);
};

// Device function to return the index of a given thread within a block.
__device__ __inline__ unsigned int __tIndex(uint3 b, dim3 bd, uint3 t) {
	return t.x + t.y*bd.x + t.z*bd.x*bd.z;
};

// Return thread global index on a 2D space, given its global position and number of rows.
__device__ __inline__ unsigned int __tIndex2D(uint3 globalPosition, unsigned int rows){
	return globalPosition.y * rows + globalPosition.x;
};