Define the 'flavor' of a PxGpuTask. More...
#include <PxGpuTask.h>
Public Types | |
enum | Enum { HostToDevice, Kernel, DeviceToHost, NUM_GPU_TASK_HINTS } |
Enums for the type of GPU task. More... | |
Define the 'flavor' of a PxGpuTask.
Each PxGpuTask should have a specific function; either copying data to the device, running kernels on that data, or copying data from the device.
For optimal performance, the dispatcher should run all available HtoD tasks before running all Kernel tasks, and all Kernel tasks before running any DtoH tasks. This provides maximal kernel overlap and the least number of CUDA flushes.