viskores::cont::cuda Namespace Reference

CUDA implementation for Control Environment. More...

Classes
class	ErrorCuda
	This error is thrown whenever an unidentified CUDA runtime error is encountered. More...
struct	ScheduleParameters
	Represents how to schedule 1D, 2D, and 3D Cuda kernels. More...

Functions
void	InitScheduleParameters (viskores::cont::cuda::ScheduleParameters(*)(char const *name, int major, int minor, int multiProcessorCount, int maxThreadsPerMultiProcessor, int maxThreadsPerBlock))
	Specify the custom scheduling to use for Viskores CUDA kernel launches. More...

Detailed Description

CUDA implementation for Control Environment.

viskores::cont::cuda includes the code to implement the Viskores Control Environment for the CUDA-based device adapter.

Function Documentation

InitScheduleParameters()

void viskores::cont::cuda::InitScheduleParameters

(

viskores::cont::cuda::ScheduleParameters(*)(char const *name, int major, int minor, int multiProcessorCount, int maxThreadsPerMultiProcessor, int maxThreadsPerBlock)

)

Specify the custom scheduling to use for Viskores CUDA kernel launches.

By default Viskores uses a preset table based on the GPU's found at runtime to determine the best scheduling parameters for a worklet. When these defaults are insufficient for certain projects it is possible to override the defaults by binding a custom function to InitScheduleParameters.

Note: The this function must be called before any invocation of any worklets by Viskores.

Note: This function will be called for each GPU on a machine.

ScheduleParameters CustomScheduleValues(char const* name,
                                        int major,
                                        int minor,
                                        int multiProcessorCount,
                                        int maxThreadsPerMultiProcessor,
                                        int maxThreadsPerBlock)
{
 
  ScheduleParameters params  {
      64 * multiProcessorCount,  //1d blocks
      64,                        //1d threads per block
      64 * multiProcessorCount,  //2d blocks
      { 8, 8, 1 },               //2d threads per block
      64 * multiProcessorCount,  //3d blocks
      { 4, 4, 4 } };             //3d threads per block
  return params;
}