TREX/qmckl
1
0
Fork 0
mirror of https://github.com/TREX-CoE/qmckl.git synced 2024-12-22 20:36:01 +01:00
Code Issues Projects Releases Wiki Activity

Memory alignment in qmckl_malloc

Browse Source
This commit is contained in:
Anthony Scemama 2022-06-29 13:59:09 +02:00
parent d5fcd2e0fe
commit cd6de216b8
2 changed files with 7 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
org/qmckl_ao.org
View File

@ -6536,7 +6536,7 @@ qmckl_compute_ao_vgl_hpc_gaussian (
/* --- */ /* --- */
switch (512) { switch (8) {
case(5): case(5):
for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) { for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) {
@ -6556,7 +6556,7 @@ qmckl_compute_ao_vgl_hpc_gaussian (
for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) { for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) {
#ifdef HAVE_OPENMP #ifdef HAVE_OPENMP
#pragma omp simd simdlen(8) #pragma omp simd
#endif #endif
for (int j=0 ; j<8 ; ++j) { for (int j=0 ; j<8 ; ++j) {
ce_mat[i][j] = 0.; ce_mat[i][j] = 0.;
@ -6564,7 +6564,7 @@ qmckl_compute_ao_vgl_hpc_gaussian (
for (int k=0 ; k<nidx; ++k) { for (int k=0 ; k<nidx; ++k) {
const double cm = coef_mat[inucl][i][k]; const double cm = coef_mat[inucl][i][k];
#ifdef HAVE_OPENMP #ifdef HAVE_OPENMP
#pragma omp simd simdlen(8) #pragma omp simd
#endif #endif
for (int j=0 ; j<8 ; ++j) { for (int j=0 ; j<8 ; ++j) {
ce_mat[i][j] += cm * exp_mat[k][j]; ce_mat[i][j] += cm * exp_mat[k][j];

4
org/qmckl_memory.org
View File

@ -119,7 +119,11 @@ void* qmckl_malloc(qmckl_context context, const qmckl_memory_info_struct info) {
qmckl_context_struct* const ctx = (qmckl_context_struct*) context; qmckl_context_struct* const ctx = (qmckl_context_struct*) context;
/* Allocate memory and zero it */ /* Allocate memory and zero it */
#ifdef HAVE_HPC
void * pointer = aligned_alloc(64, ((info.size+64) >> 6) << 6 );
#else
void * pointer = malloc(info.size); void * pointer = malloc(info.size);
#endif
if (pointer == NULL) { if (pointer == NULL) {
return NULL; return NULL;
} }