portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-24 10:42:03 +01:00
Code Releases Activity

Copy stream HIP improvements

Browse Source
This commit is contained in:
Peter Boyle
2022-08-04 15:24:03 -04:00
parent 806b02bddf
commit 2cb5bedc15
2 changed files with 15 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
Grid/threads/Accelerator.h
View File

@ -370,7 +370,8 @@ accelerator_inline int acceleratorSIMTlane(int Nsimd) {
num1,num2,nsimd, lambda); \
} \
}
// Works with MPI if barrier here
// accelerator_barrier();
template<typename lambda> __global__
__launch_bounds__(64,1)
@ -400,7 +401,7 @@ void LambdaApply(uint64_t numx, uint64_t numy, uint64_t numz, lambda Lambda)
#define accelerator_barrier(dummy) \
{ \
hipStreamSynchronize(cpuStream); \
hipDeviceSynchronize(); \
auto err = hipGetLastError(); \
if ( err != hipSuccess ) { \
printf("After hipDeviceSynchronize() : HIP error %s \n", hipGetErrorString( err )); \
@ -443,7 +444,7 @@ inline void acceleratorMemSet(void *base,int value,size_t bytes) { hipMemset(bas
inline void acceleratorCopyDeviceToDeviceAsynch(void *from,void *to,size_t bytes) // Asynch
{
hipMemcpy(to,from,bytes, hipMemcpyDeviceToDevice);
hipMemcpyDtoDAsync(to,from,bytes, copyStream);
}
inline void acceleratorCopySynchronise(void) { hipStreamSynchronize(copyStream); };