gallo/atrip
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Lint and tidy up Equations

Browse Source
This commit is contained in:
Alejandro Gallo 2022-09-08 13:51:49 +02:00
parent 368c5619cc
commit 2c5a4620ca
1 changed files with 81 additions and 70 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
src/atrip/Equations.cxx
View File

@ -132,9 +132,8 @@ namespace cuda {
#endif #endif
#if defined(HAVE_CUDA) #if defined(HAVE_CUDA)
#define LIMS_KS \ #define LIMS_KS() \
size_t \ size_t kmin = blockIdx.x * blockDim.x + threadIdx.x, \
kmin = blockIdx.x * blockDim.x + threadIdx.x, \
k = kmin, \ k = kmin, \
idx = kmin * size * size * size \ idx = kmin * size * size * size \
; \ ; \
@ -144,7 +143,7 @@ namespace cuda {
#endif #endif
#define _IJK_(i, j, k) i + j*size + k*size*size #define _IJK_(i, j, k) i + j*size + k*size*size
#define _REORDER_BODY_(...) \ #define _REORDER_BODY_(...) \
for (LIMS_KS ; k++) \ for (LIMS_KS() ; k++) \
for (size_t j = 0; j < size; j++) \ for (size_t j = 0; j < size; j++) \
for (size_t i = 0; i < size; i++, idx++) { \ for (size_t i = 0; i < size; i++, idx++) { \
__VA_ARGS__ \ __VA_ARGS__ \
@ -447,9 +446,8 @@ double getEnergySame
#if defined(ATRIP_USE_DGEMM) #if defined(ATRIP_USE_DGEMM)
#if defined(HAVE_CUDA) #if defined(HAVE_CUDA)
#define REORDER(__II, __JJ, __KK) \ #define REORDER(__II, __JJ, __KK) \
reorder<<< \ reorder<<<bs, ths>>>(reorder_proxy<DataFieldType<F>, \
bs, ths \ __II ## __JJ ## __KK >{}, \
>>>(reorder_proxy<DataFieldType<F>, __II ## __JJ ## __KK >{}, \
No, Tijk, _t_buffer); No, Tijk, _t_buffer);
#define DGEMM_PARTICLES(__A, __B) \ #define DGEMM_PARTICLES(__A, __B) \
atrip::xgemm<F>("T", \ atrip::xgemm<F>("T", \
@ -485,10 +483,20 @@ double getEnergySame
Atrip::kernelDimensions.ooo.blocks, \ Atrip::kernelDimensions.ooo.blocks, \
Atrip::kernelDimensions.ooo.threads \ Atrip::kernelDimensions.ooo.threads \
>>>((DataFieldType<F>*)_conj, (DataFieldType<F>*)_buffer, NoNoNo); >>>((DataFieldType<F>*)_conj, (DataFieldType<F>*)_buffer, NoNoNo);
// END CUDA ////////////////////////////////////////////////////////////////////
#else #else
// NONCUDA //////////////////////////////////////////////////////////////////////
// NONCUDA /////////////////////////////////////////////////////////////////////
#define REORDER(__II, __JJ, __KK) \ #define REORDER(__II, __JJ, __KK) \
reorder(reorder_proxy<DataFieldType<F>, __II ## __JJ ## __KK >{}, \ reorder(reorder_proxy<DataFieldType<F>, \
__II ## __JJ ## __KK >{}, \
No, Tijk, _t_buffer); No, Tijk, _t_buffer);
#define DGEMM_PARTICLES(__A, __B) \ #define DGEMM_PARTICLES(__A, __B) \
atrip::xgemm<F>("T", \ atrip::xgemm<F>("T", \
@ -550,7 +558,8 @@ double getEnergySame
// Set Tijk to zero // Set Tijk to zero
#ifdef HAVE_CUDA #ifdef HAVE_CUDA
WITH_CHRONO("double:reorder", WITH_CHRONO("double:reorder",
cuda::zeroing<<<bs, ths>>>((DataFieldType<F>*)Tijk, NoNoNo); cuda::zeroing<<<bs, ths>>>((DataFieldType<F>*)Tijk,
NoNoNo);
// synchronize all initializations to zero // synchronize all initializations to zero
) )
#else #else
@ -640,6 +649,8 @@ double getEnergySame
{ // free resources { // free resources
#ifdef HAVE_CUDA #ifdef HAVE_CUDA
// we need to synchronize here since we need
// the Tijk for next process in the pipeline
cuCtxSynchronize(); cuCtxSynchronize();
cuMemFree((CUdeviceptr)_vhhh); cuMemFree((CUdeviceptr)_vhhh);
cuMemFree((CUdeviceptr)_t_buffer); cuMemFree((CUdeviceptr)_t_buffer);