diff options
Diffstat (limited to 'openmp/runtime/src/kmp_csupport.cpp')
| -rw-r--r-- | openmp/runtime/src/kmp_csupport.cpp | 23 |
1 files changed, 20 insertions, 3 deletions
diff --git a/openmp/runtime/src/kmp_csupport.cpp b/openmp/runtime/src/kmp_csupport.cpp index d39bf9af4334..ac9a93590ad0 100644 --- a/openmp/runtime/src/kmp_csupport.cpp +++ b/openmp/runtime/src/kmp_csupport.cpp @@ -18,10 +18,7 @@ #include "kmp_itt.h" #include "kmp_lock.h" #include "kmp_stats.h" - -#if OMPT_SUPPORT #include "ompt-specific.h" -#endif #define MAX_MESSAGE 512 @@ -3429,13 +3426,18 @@ __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck); __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method); + OMPT_REDUCTION_DECL(th, global_tid); if (packed_reduction_method == critical_reduce_block) { + OMPT_REDUCTION_BEGIN; + __kmp_enter_critical_section_reduce_block(loc, global_tid, lck); retval = 1; } else if (packed_reduction_method == empty_reduce_block) { + OMPT_REDUCTION_BEGIN; + // usage: if team size == 1, no synchronization is required ( Intel // platforms only ) retval = 1; @@ -3536,15 +3538,20 @@ void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid, packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid); + OMPT_REDUCTION_DECL(__kmp_thread_from_gtid(global_tid), global_tid); + if (packed_reduction_method == critical_reduce_block) { __kmp_end_critical_section_reduce_block(loc, global_tid, lck); + OMPT_REDUCTION_END; } else if (packed_reduction_method == empty_reduce_block) { // usage: if team size == 1, no synchronization is required ( on Intel // platforms only ) + OMPT_REDUCTION_END; + } else if (packed_reduction_method == atomic_reduce_block) { // neither master nor other workers should get here @@ -3556,6 +3563,7 @@ void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid, tree_reduce_block)) { // only master gets here + // OMPT: tree reduction is annotated in the barrier code } else { @@ -3629,13 +3637,17 @@ kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck); __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method); + OMPT_REDUCTION_DECL(th, global_tid); + if (packed_reduction_method == critical_reduce_block) { + OMPT_REDUCTION_BEGIN; __kmp_enter_critical_section_reduce_block(loc, global_tid, lck); retval = 1; } else if (packed_reduction_method == empty_reduce_block) { + OMPT_REDUCTION_BEGIN; // usage: if team size == 1, no synchronization is required ( Intel // platforms only ) retval = 1; @@ -3723,10 +3735,13 @@ void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid, // this barrier should be visible to a customer and to the threading profile // tool (it's a terminating barrier on constructs if NOWAIT not specified) + OMPT_REDUCTION_DECL(th, global_tid); if (packed_reduction_method == critical_reduce_block) { __kmp_end_critical_section_reduce_block(loc, global_tid, lck); + OMPT_REDUCTION_END; + // TODO: implicit barrier: should be exposed #if OMPT_SUPPORT ompt_frame_t *ompt_frame; @@ -3749,6 +3764,8 @@ void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid, } else if (packed_reduction_method == empty_reduce_block) { + OMPT_REDUCTION_END; + // usage: if team size==1, no synchronization is required (Intel platforms only) // TODO: implicit barrier: should be exposed |