diff options
Diffstat (limited to 'openmp/runtime/src/kmp_dispatch.cpp')
| -rw-r--r-- | openmp/runtime/src/kmp_dispatch.cpp | 29 |
1 files changed, 20 insertions, 9 deletions
diff --git a/openmp/runtime/src/kmp_dispatch.cpp b/openmp/runtime/src/kmp_dispatch.cpp index 65abcf7fc43..ba87a5407cb 100644 --- a/openmp/runtime/src/kmp_dispatch.cpp +++ b/openmp/runtime/src/kmp_dispatch.cpp @@ -670,6 +670,7 @@ __kmp_dispatch_init( } else { pr->ordered = FALSE; } + if ( schedule == kmp_sch_static ) { schedule = __kmp_static; } else { @@ -761,6 +762,19 @@ __kmp_dispatch_init( tc = 0; // zero-trip } + // Any half-decent optimizer will remove this test when the blocks are empty since the macros expand to nothing + // when statistics are disabled. + if (schedule == __kmp_static) + { + KMP_COUNT_BLOCK(OMP_FOR_static); + KMP_COUNT_VALUE(FOR_static_iterations, tc); + } + else + { + KMP_COUNT_BLOCK(OMP_FOR_dynamic); + KMP_COUNT_VALUE(FOR_dynamic_iterations, tc); + } + pr->u.p.lb = lb; pr->u.p.ub = ub; pr->u.p.st = st; @@ -1384,6 +1398,11 @@ __kmp_dispatch_next( static const int ___kmp_size_type = sizeof( UT ); #endif + // This is potentially slightly misleading, schedule(runtime) will appear here even if the actual runtme schedule + // is static. (Which points out a disadavantage of schedule(runtime): even when static scheduling is used it costs + // more than a compile time choice to use static scheduling would.) + KMP_TIME_BLOCK(FOR_dynamic_scheduling); + int status; dispatch_private_info_template< T > * pr; kmp_info_t * th = __kmp_threads[ gtid ]; @@ -2164,7 +2183,6 @@ __kmp_dist_get_bounds( T *pupper, typename traits_t< T >::signed_t incr ) { - KMP_COUNT_BLOCK(OMP_DISTR_FOR_dynamic); typedef typename traits_t< T >::unsigned_t UT; typedef typename traits_t< T >::signed_t ST; register kmp_uint32 team_id; @@ -2222,6 +2240,7 @@ __kmp_dist_get_bounds( } else { trip_count = (ST)(*pupper - *plower) / incr + 1; // cast to signed to cover incr<0 case } + if( trip_count <= nteams ) { KMP_DEBUG_ASSERT( __kmp_static == kmp_sch_static_greedy || \ @@ -2297,7 +2316,6 @@ void __kmpc_dispatch_init_4( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int32 lb, kmp_int32 ub, kmp_int32 st, kmp_int32 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dispatch_init< kmp_int32 >( loc, gtid, schedule, lb, ub, st, chunk, true ); } @@ -2308,7 +2326,6 @@ void __kmpc_dispatch_init_4u( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_uint32 lb, kmp_uint32 ub, kmp_int32 st, kmp_int32 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dispatch_init< kmp_uint32 >( loc, gtid, schedule, lb, ub, st, chunk, true ); } @@ -2321,7 +2338,6 @@ __kmpc_dispatch_init_8( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int64 lb, kmp_int64 ub, kmp_int64 st, kmp_int64 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dispatch_init< kmp_int64 >( loc, gtid, schedule, lb, ub, st, chunk, true ); } @@ -2334,7 +2350,6 @@ __kmpc_dispatch_init_8u( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_uint64 lb, kmp_uint64 ub, kmp_int64 st, kmp_int64 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dispatch_init< kmp_uint64 >( loc, gtid, schedule, lb, ub, st, chunk, true ); } @@ -2352,7 +2367,6 @@ void __kmpc_dist_dispatch_init_4( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int32 *p_last, kmp_int32 lb, kmp_int32 ub, kmp_int32 st, kmp_int32 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dist_get_bounds< kmp_int32 >( loc, gtid, p_last, &lb, &ub, st ); __kmp_dispatch_init< kmp_int32 >( loc, gtid, schedule, lb, ub, st, chunk, true ); @@ -2362,7 +2376,6 @@ void __kmpc_dist_dispatch_init_4u( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int32 *p_last, kmp_uint32 lb, kmp_uint32 ub, kmp_int32 st, kmp_int32 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dist_get_bounds< kmp_uint32 >( loc, gtid, p_last, &lb, &ub, st ); __kmp_dispatch_init< kmp_uint32 >( loc, gtid, schedule, lb, ub, st, chunk, true ); @@ -2372,7 +2385,6 @@ void __kmpc_dist_dispatch_init_8( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int32 *p_last, kmp_int64 lb, kmp_int64 ub, kmp_int64 st, kmp_int64 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dist_get_bounds< kmp_int64 >( loc, gtid, p_last, &lb, &ub, st ); __kmp_dispatch_init< kmp_int64 >( loc, gtid, schedule, lb, ub, st, chunk, true ); @@ -2382,7 +2394,6 @@ void __kmpc_dist_dispatch_init_8u( ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int32 *p_last, kmp_uint64 lb, kmp_uint64 ub, kmp_int64 st, kmp_int64 chunk ) { - KMP_COUNT_BLOCK(OMP_FOR_dynamic); KMP_DEBUG_ASSERT( __kmp_init_serial ); __kmp_dist_get_bounds< kmp_uint64 >( loc, gtid, p_last, &lb, &ub, st ); __kmp_dispatch_init< kmp_uint64 >( loc, gtid, schedule, lb, ub, st, chunk, true ); |
