Merge pull request #101 from Adrian-Diaz/Adrian's-Branch

ENH: make FOR_FIRST arguments consistent with other FORs
lanl · Oct 3, 2024 · d5bf5c7 · d5bf5c7
2 parents d7ecb7f + 1e26051
commit d5bf5c7
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 11 deletions.
diff --git a/examples/main_kokkos.cpp b/examples/main_kokkos.cpp
@@ -875,7 +875,7 @@ int main(int argc, char* argv[])
         FOR_ALL(i_i, 0, hiersize, j_j, 0, hiersize, k_k, 0, hiersize, {
             hierTest3D(i_i, j_j, k_k) = 0.0;
         });
-        FOR_FIRST(i_i,hiersize, {
+        FOR_FIRST(i_i, 0, hiersize, {
             // Kokkos::parallel_for( \
             //Kokkos::TeamPolicy<>( 32, Kokkos::AUTO, 32 ), \
             //KOKKOS_LAMBDA ( const Kokkos::TeamPolicy<>::member_type &teamMember ) {
@@ -908,7 +908,7 @@ int main(int argc, char* argv[])
 
         printf("\n\n\nHierarchical Reduce\n");
         //2D nesting
-        FOR_FIRST(i_i,hiersize, {
+        FOR_FIRST(i_i, 0, hiersize, {
             // Kokkos::parallel_for( \
             //Kokkos::TeamPolicy<>( 32, Kokkos::AUTO, 32 ), \
             //KOKKOS_LAMBDA ( const Kokkos::TeamPolicy<>::member_type &teamMember ) {
@@ -936,29 +936,29 @@ int main(int argc, char* argv[])
 
         printf("\n\n\nHierarchical Vectorized Reduce\n");
         //3D vector nesting
-        FOR_FIRST(i_i,hiersize, {
+        FOR_FIRST(i_i, 1, hiersize+1, {
             // Kokkos::parallel_for( \
             //Kokkos::TeamPolicy<>( 32, Kokkos::AUTO, 32 ), \
             //KOKKOS_LAMBDA ( const Kokkos::TeamPolicy<>::member_type &teamMember ) {
             //const int i_i = TEAM_ID;
             double result = 0;
             double lsum;
-            FOR_SECOND(j_j, i_i, hiersize, {
+            FOR_SECOND(j_j, i_i-1, hiersize, {
                 // Kokkos::parallel_for( \
                 //Kokkos::TeamThreadRange( teamMember, istart, iend ), [&] ( const int (j_j) ) {
                 // hierTest2D(i_i,j_j) = i_i * (j_j+1);
                 //    int jstart = j_j*32;
                 //    int jend = (j_j+1)*32;
-                FOR_REDUCE_SUM_THIRD(k_k, i_i, j_j, lsum, {
-                lsum += hierTest3D(i_i,j_j,k_k);
+                FOR_REDUCE_SUM_THIRD(k_k, i_i-1, j_j, lsum, {
+                lsum += hierTest3D(i_i-1,j_j,k_k);
                 // Kokkos::parallel_for( \
                 //Kokkos::TeamThreadRange( teamMember, istart, iend ), [&] ( const int (j_j) ) {
                 // hierTest2D(i_i,j_j) = i_i * (j_j+1);
                 //    int jstart = j_j*32;
                 //    int jend = (j_j+1)*32;
                 }, result);
-                hierTest2D(i_i,j_j)= result;
-                //printf("value at %d , %d is %f\n", i_i, j_j, hierTest2D(i_i,j_j));
+                hierTest2D(i_i-1,j_j)= result;
+                printf("value at %d , %d is %f\n", i_i, j_j, hierTest2D(i_i-1,j_j));
             });
         });
         Kokkos::fence();

diff --git a/src/include/macros.h b/src/include/macros.h
@@ -445,11 +445,11 @@ THREAD_ID \
 teamMember.team_rank()
 
 #define \
-FOR_FIRST(i, x1, fcn) \
+FOR_FIRST(i, x0, x1, fcn) \
 Kokkos::parallel_for( \
-                        Kokkos::TeamPolicy<>( x1, Kokkos::AUTO, 32 ), \
+                        Kokkos::TeamPolicy<>( x1-x0, Kokkos::AUTO, 32 ), \
                         KOKKOS_LAMBDA ( const Kokkos::TeamPolicy<>::member_type &teamMember ) \
-                        { const int i = TEAM_ID; fcn} )
+                        { const int i = TEAM_ID + x0; fcn} )
 
 #define \
 FOR_SECOND(j, y0, y1, fcn) \