Parallelize on the level of variables rather than grid points.HEADmaster

The latter has higher overhead.

1 files changed, 0 insertions, 12 deletions

diff --git a/src/Operators.c b/src/Operators.c
index 2fed466..ed318f7 100644
--- a/src/Operators.c
+++ b/src/Operators.c
@@ -202,21 +202,18 @@ MoL_LinearCombination(cGH const *const cctkGH,
       // performance
       switch (nsrcs) {
       case 0:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_set_0(varptr + offset, ash[0]);
           }
         break;
       case 1:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_set_1(varptr + offset, srcptrs[0] + offset, facts[0], ash[0]);
           }
         break;
       case 2:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_set_2(varptr + offset, srcptrs[0] + offset, facts[0],
@@ -224,7 +221,6 @@ MoL_LinearCombination(cGH const *const cctkGH,
           }
         break;
       case 3:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_set_3(varptr + offset,
@@ -234,7 +230,6 @@ MoL_LinearCombination(cGH const *const cctkGH,
         break;
       default:
         // Loop over all grid points
-#pragma omp parallel for
         for (i = 0; i < ash[2]; i++) {
             for (j = 0; j < ash[0]; j++) {
                 int idx = CCTK_GFINDEX3D(cctkGH, j, y_idx, i);
@@ -255,21 +250,18 @@ MoL_LinearCombination(cGH const *const cctkGH,
       // performance
       switch (nsrcs) {
       case 0:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_update_0(varptr + offset, scale, ash[0]);
           }
         break;
       case 1:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_update_1(varptr + offset, scale, srcptrs[0] + offset, facts[0], ash[0]);
           }
         break;
       case 2:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_update_2(varptr + offset, scale,
@@ -278,7 +270,6 @@ MoL_LinearCombination(cGH const *const cctkGH,
           }
         break;
       case 3:
-#pragma omp parallel for
           for (i = 0; i < ash[2]; i++) {
               int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, i);
               op_real_update_3(varptr + offset, scale,
@@ -288,7 +279,6 @@ MoL_LinearCombination(cGH const *const cctkGH,
         break;
       default:
         // Loop over all grid points
-#pragma omp parallel for
         for (i = 0; i < ash[2]; i++) {
             for (j = 0; j < ash[0]; j++) {
                 int idx = CCTK_GFINDEX3D(cctkGH, j, y_idx, i);
@@ -320,7 +310,6 @@ MoL_LinearCombination(cGH const *const cctkGH,
     if (scale == 0.0) {
       // Set (overwrite) target variable
       // Loop over all grid points
-#pragma omp parallel for
       for (ptrdiff_t i=0; i<npoints; ++i) {
         CCTK_COMPLEX tmp = CCTK_Cmplx(0.0, 0.0);
         for (int n=0; n<nsrcs; ++n) {
@@ -331,7 +320,6 @@ MoL_LinearCombination(cGH const *const cctkGH,
       }
     } else {
       // Update (add to) target variable
-#pragma omp parallel for
       for (ptrdiff_t i=0; i<npoints; ++i) {
         CCTK_COMPLEX tmp = CCTK_CmplxMul(CCTK_Cmplx(scale, 0.0), varptr[i]);
         for (int n=0; n<nsrcs; ++n) {