TREX/qmckl
1
0
Fork 0
mirror of https://github.com/TREX-CoE/qmckl.git synced 2024-12-23 04:44:03 +01:00
Code Issues Projects Releases Wiki Activity

Switch for asm in AOs

Browse Source
This commit is contained in:
Anthony Scemama 2022-06-14 22:30:33 +02:00
parent 07e1e44f05
commit c0e82939ac
1 changed files with 26 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
org/qmckl_ao.org
View File

@ -6523,6 +6523,23 @@ qmckl_compute_ao_vgl_hpc_gaussian (
/* --- */ /* --- */
switch (8) {
case(5):
for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) {
for (int j=0 ; j<5 ; ++j) {
ce_mat[i][j] = 0.;
}
for (int k=0 ; k<nidx; ++k) {
for (int j=0 ; j<5 ; ++j) {
ce_mat[i][j] += coef_mat[inucl][i][k] * exp_mat[k][j];
}
}
}
break;
case(8):
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 simdlen(8)
@ -6540,9 +6557,11 @@ qmckl_compute_ao_vgl_hpc_gaussian (
} }
} }
/* break;
for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) { case(256):
// Following loop is the assembly version AVX2 // Following loop is the assembly version AVX2
for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) {
__asm__ volatile ( __asm__ volatile (
"mov %[k],%%RSI" "\n\t" // &(nidx) "mov %[k],%%RSI" "\n\t" // &(nidx)
"mov %[a],%%RAX" "\n\t" // &(coef_mat[inucl][i][k]) "mov %[a],%%RAX" "\n\t" // &(coef_mat[inucl][i][k])
@ -6577,10 +6596,10 @@ qmckl_compute_ao_vgl_hpc_gaussian (
"ymm0", "ymm1", "ymm2", "memory" ); "ymm0", "ymm1", "ymm2", "memory" );
} }
}
,*/
/* break;
case(512):
// Following loop is the assembly version AVX512 // Following loop is the assembly version AVX512
for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) { for (int i=0 ; i<nucleus_shell_num[inucl] ; ++i) {
__asm__ volatile ( __asm__ volatile (
@ -6608,7 +6627,7 @@ qmckl_compute_ao_vgl_hpc_gaussian (
: : [k] "m"(nidx), [c] "m"(&(ce_mat[i][0])), [a] "m"(&(coef_mat[inucl][i][0])), [b] "m"(&(exp_mat[0][0])) : "rax", "rbx", "rcx", "rsi", "zmm0", "zmm1", "memory" ); : : [k] "m"(nidx), [c] "m"(&(ce_mat[i][0])), [a] "m"(&(coef_mat[inucl][i][0])), [b] "m"(&(exp_mat[0][0])) : "rax", "rbx", "rcx", "rsi", "zmm0", "zmm1", "memory" );
} }
,*/ }
const int64_t ishell_start = nucleus_index[inucl]; const int64_t ishell_start = nucleus_index[inucl];
const int64_t ishell_end = nucleus_index[inucl] + nucleus_shell_num[inucl]; const int64_t ishell_end = nucleus_index[inucl] + nucleus_shell_num[inucl];