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

The latter has higher overhead.

1 files changed, 6 insertions, 3 deletions

diff --git a/src/InitialCopy.c b/src/InitialCopy.c
index 4c0ee87..176e2de 100644
--- a/src/InitialCopy.c
+++ b/src/InitialCopy.c
@@ -122,6 +122,7 @@ void MoL_InitialCopy(CCTK_ARGUMENTS)
     totalsize *= cctk_ash[arraydim];
   }
 
+#pragma omp parallel for
   for (var = 0; var < MoLNumEvolvedVariables; var++)
   {
     const int       nsrc = 1;
@@ -236,6 +237,7 @@ void MoL_InitialCopy(CCTK_ARGUMENTS)
      current level to the scratch space, then do the copy 
   */
 
+#pragma omp parallel for
   for (var = 0; var < MoLNumSandRVariables; var++)
   {
 
@@ -280,6 +282,7 @@ void MoL_InitialCopy(CCTK_ARGUMENTS)
     Now do the constrained variables.
   */
 
+#pragma omp parallel for
   for (var = 0; var < MoLNumConstrainedVariables; var++)
   {
 
@@ -316,7 +319,6 @@ void MoL_InitialCopy(CCTK_ARGUMENTS)
 
       if (PreviousVar && CurrentVar)
       {     
-#pragma omp parallel for
         for (int k = 0; k < cctk_ash[2]; k++) {
             int offset = CCTK_GFINDEX3D(cctkGH, 0, y_idx, k);
             memcpy(CurrentVar + offset, PreviousVar + offset, cctk_ash[0] * sizeof(double));
@@ -389,6 +391,7 @@ void MoL_InitRHS(CCTK_ARGUMENTS)
     totalsize *= cctk_ash[arraydim];
   }
 
+#pragma omp parallel for
   for (var = 0; var < MoLNumEvolvedVariables; var++)
   {
     StorageOn = CCTK_QueryGroupStorageI(cctkGH,
@@ -415,6 +418,7 @@ void MoL_InitRHS(CCTK_ARGUMENTS)
                           NULL, NULL, NULL, 0);
   }
     
+#pragma omp parallel for
   for (var = 0; var < MoLNumEvolvedArrayVariables; var++)
   {
     RHSVar = (CCTK_REAL*)CCTK_VarDataPtrI(cctkGH, 0,
@@ -440,7 +444,6 @@ void MoL_InitRHS(CCTK_ARGUMENTS)
     {
       if (RHSVar)
       {
-#pragma omp parallel for
         for (index = 0; index < arraytotalsize; index++)
         {
           RHSVar[index] = 0;
@@ -461,6 +464,7 @@ void MoL_InitRHS(CCTK_ARGUMENTS)
 
 #ifdef MOLDOESCOMPLEX
 
+#pragma omp parallel for
   for (var = 0; var < MoLNumEvolvedComplexVariables; var++)
   {
     
@@ -483,7 +487,6 @@ void MoL_InitRHS(CCTK_ARGUMENTS)
                                                     RHSComplexVariableIndex[var]);
     if (RHSVar)
     {
-#pragma omp parallel for
       for (index = 0; index < totalsize; index++)
       {
         RHSVar[index] = 0;