Pscf::ThreadMesh Namespace Reference

Management of multidimensional GPU thread execution configurations. More...

Functions
template<int D>
void	setConfig (IntVec< D > const &meshDims, bool invert, int blockSize=-1)
	Given a multidimensional grid of threads, set execution configuration.
void	checkConfig ()
	Check the execution configuration (grid and block dimensions).
void	setThreadsPerBlock (int blockSize)
	Manually set the block size that should be used by default.
dim3	gridDims ()
	Get the multidimensional grid of blocks determined by setConfig.
dim3	blockDims ()
	Get the dimensions of a single block determined by setConfig.
dim3	meshDims ()
	Return last requested multidimensional grid of threads.
int	warpSize ()
	Get the warp size.
bool	hasUnusedThreads ()
	Will there be unused threads?

Detailed Description

Management of multidimensional GPU thread execution configurations.

For calculations involving data defined on a multidimensional mesh (e.g. a field on a real-space grid), it is often useful to use a multidimensional grid of CUDA threads defined on that same mesh (or on a slightly larger mesh if the mesh dimensions are not nicely divisible by powers of 2). This way, within the CUDA kernel, it is easy to identify the grid point that corresponds to a given thread using threadIdx and blockIdx.