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

Fix the reordering kernel in Equations

Browse Source
This commit is contained in:
Ania Brown 2023-01-25 13:53:22 +01:00 committed by Alejandro Gallo
parent c7e3fa45bd
commit 933d556c84
1 changed files with 3 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
src/atrip/Equations.cxx
View File

@ -25,11 +25,8 @@ namespace atrip {
#if defined(HAVE_CUDA) #if defined(HAVE_CUDA)
#define FOR_K() \ #define FOR_K() \
for (size_t kmin = blockIdx.x * blockDim.x + threadIdx.x, \ const size_t k = blockIdx.x * blockDim.x + threadIdx.x; \
k = kmin, \ size_t idx = 0;
idx = kmin * size * size * size; \
k < (kmin < size) ? kmin + 1 : size; \
k++)
#else #else
#define FOR_K() for (size_t k=0, idx=0; k < size; k++) #define FOR_K() for (size_t k=0, idx=0; k < size; k++)
#endif #endif
@ -545,7 +542,7 @@ void 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<<<bs, ths>>>(reorder_proxy< \ reorder<<<1, No>>>(reorder_proxy< \
DataFieldType<F>, \ DataFieldType<F>, \
__II ## __JJ ## __KK \ __II ## __JJ ## __KK \
>{}, \ >{}, \