diff options
| -rw-r--r-- | gcc/c-family/c-omp.c | 169 | ||||
| -rw-r--r-- | gcc/c-family/c-pragma.h | 1 | ||||
| -rw-r--r-- | gcc/c/c-parser.c | 74 | ||||
| -rw-r--r-- | gcc/c/c-typeck.c | 62 | ||||
| -rw-r--r-- | gcc/cp/parser.c | 59 | ||||
| -rw-r--r-- | gcc/cp/pt.c | 1 | ||||
| -rw-r--r-- | gcc/cp/semantics.c | 93 | ||||
| -rw-r--r-- | gcc/gimplify.c | 25 | ||||
| -rw-r--r-- | gcc/omp-low.c | 2 | ||||
| -rw-r--r-- | gcc/testsuite/c-c++-common/gomp/allocate-1.c | 76 | ||||
| -rw-r--r-- | gcc/testsuite/c-c++-common/gomp/allocate-2.c | 45 | ||||
| -rw-r--r-- | gcc/testsuite/c-c++-common/gomp/clauses-1.c | 138 | ||||
| -rw-r--r-- | gcc/tree-core.h | 3 | ||||
| -rw-r--r-- | gcc/tree-nested.c | 22 | ||||
| -rw-r--r-- | gcc/tree-pretty-print.c | 13 | ||||
| -rw-r--r-- | gcc/tree.c | 3 | ||||
| -rw-r--r-- | gcc/tree.h | 10 |
17 files changed, 739 insertions, 57 deletions
diff --git a/gcc/c-family/c-omp.c b/gcc/c-family/c-omp.c index d7cff0f..bce2f8e 100644 --- a/gcc/c-family/c-omp.c +++ b/gcc/c-family/c-omp.c @@ -35,6 +35,7 @@ along with GCC; see the file COPYING3. If not see #include "attribs.h" #include "gimplify.h" #include "langhooks.h" +#include "bitmap.h" /* Complete a #pragma oacc wait construct. LOC is the location of @@ -1575,6 +1576,7 @@ c_omp_split_clauses (location_t loc, enum tree_code code, tree next, c; enum c_omp_clause_split s; int i; + bool has_dup_allocate = false; for (i = 0; i < C_OMP_CLAUSE_SPLIT_COUNT; i++) cclauses[i] = NULL; @@ -2198,6 +2200,71 @@ c_omp_split_clauses (location_t loc, enum tree_code code, else s = C_OMP_CLAUSE_SPLIT_FOR; break; + /* Allocate clause is allowed on target, teams, distribute, parallel, + for, sections and taskloop. Distribute it to all. */ + case OMP_CLAUSE_ALLOCATE: + s = C_OMP_CLAUSE_SPLIT_COUNT; + for (i = 0; i < C_OMP_CLAUSE_SPLIT_COUNT; i++) + { + switch (i) + { + case C_OMP_CLAUSE_SPLIT_TARGET: + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_MAP)) == 0) + continue; + break; + case C_OMP_CLAUSE_SPLIT_TEAMS: + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_NUM_TEAMS)) == 0) + continue; + break; + case C_OMP_CLAUSE_SPLIT_DISTRIBUTE: + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) == 0) + continue; + break; + case C_OMP_CLAUSE_SPLIT_PARALLEL: + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_NUM_THREADS)) == 0) + continue; + break; + case C_OMP_CLAUSE_SPLIT_FOR: + STATIC_ASSERT (C_OMP_CLAUSE_SPLIT_SECTIONS + == C_OMP_CLAUSE_SPLIT_FOR + && (C_OMP_CLAUSE_SPLIT_TASKLOOP + == C_OMP_CLAUSE_SPLIT_FOR) + && (C_OMP_CLAUSE_SPLIT_LOOP + == C_OMP_CLAUSE_SPLIT_FOR)); + if (code == OMP_SECTIONS) + break; + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_SCHEDULE)) != 0) + break; + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_NOGROUP)) != 0) + break; + continue; + case C_OMP_CLAUSE_SPLIT_SIMD: + continue; + default: + gcc_unreachable (); + } + if (s != C_OMP_CLAUSE_SPLIT_COUNT) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_ALLOCATE); + OMP_CLAUSE_DECL (c) + = OMP_CLAUSE_DECL (clauses); + OMP_CLAUSE_ALLOCATE_ALLOCATOR (c) + = OMP_CLAUSE_ALLOCATE_ALLOCATOR (clauses); + OMP_CLAUSE_CHAIN (c) = cclauses[s]; + cclauses[s] = c; + has_dup_allocate = true; + } + s = (enum c_omp_clause_split) i; + } + gcc_assert (s != C_OMP_CLAUSE_SPLIT_COUNT); + break; default: gcc_unreachable (); } @@ -2205,6 +2272,108 @@ c_omp_split_clauses (location_t loc, enum tree_code code, cclauses[s] = clauses; } + if (has_dup_allocate) + { + bool need_prune = false; + bitmap_obstack_initialize (NULL); + for (i = 0; i < C_OMP_CLAUSE_SPLIT_SIMD - (code == OMP_LOOP); i++) + if (cclauses[i]) + { + bitmap_head allocate_head; + bitmap_initialize (&allocate_head, &bitmap_default_obstack); + for (c = cclauses[i]; c; c = OMP_CLAUSE_CHAIN (c)) + if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ALLOCATE + && DECL_P (OMP_CLAUSE_DECL (c))) + bitmap_set_bit (&allocate_head, + DECL_UID (OMP_CLAUSE_DECL (c))); + for (c = cclauses[i]; c; c = OMP_CLAUSE_CHAIN (c)) + switch (OMP_CLAUSE_CODE (c)) + { + case OMP_CLAUSE_PRIVATE: + case OMP_CLAUSE_FIRSTPRIVATE: + case OMP_CLAUSE_LASTPRIVATE: + case OMP_CLAUSE_LINEAR: + case OMP_CLAUSE_REDUCTION: + case OMP_CLAUSE_IN_REDUCTION: + case OMP_CLAUSE_TASK_REDUCTION: + if (DECL_P (OMP_CLAUSE_DECL (c))) + bitmap_clear_bit (&allocate_head, + DECL_UID (OMP_CLAUSE_DECL (c))); + break; + default: + break; + } + for (c = cclauses[i]; c; c = OMP_CLAUSE_CHAIN (c)) + if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ALLOCATE + && DECL_P (OMP_CLAUSE_DECL (c)) + && bitmap_bit_p (&allocate_head, + DECL_UID (OMP_CLAUSE_DECL (c)))) + { + /* Mark allocate clauses which don't have corresponding + explicit data sharing clause. */ + OMP_CLAUSE_ALLOCATE_COMBINED (c) = 1; + need_prune = true; + } + } + bitmap_obstack_release (NULL); + if (need_prune) + { + /* At least one allocate clause has been marked. Walk all the + duplicated allocate clauses in sync. If it is marked in all + constituent constructs, diagnose it as invalid and remove + them. Otherwise, remove all marked inner clauses inside + a construct that doesn't have them marked. Keep the outer + marked ones, because some clause duplication is done only + during gimplification. */ + tree *p[C_OMP_CLAUSE_SPLIT_COUNT]; + for (i = 0; i < C_OMP_CLAUSE_SPLIT_COUNT; i++) + if (cclauses[i] == NULL_TREE + || i == C_OMP_CLAUSE_SPLIT_SIMD + || (i == C_OMP_CLAUSE_SPLIT_LOOP && code == OMP_LOOP)) + p[i] = NULL; + else + p[i] = &cclauses[i]; + do + { + int j = -1; + tree seen = NULL_TREE; + for (i = C_OMP_CLAUSE_SPLIT_COUNT - 1; i >= 0; i--) + if (p[i]) + { + while (*p[i] + && OMP_CLAUSE_CODE (*p[i]) != OMP_CLAUSE_ALLOCATE) + p[i] = &OMP_CLAUSE_CHAIN (*p[i]); + if (*p[i] == NULL_TREE) + { + i = C_OMP_CLAUSE_SPLIT_COUNT; + break; + } + if (!OMP_CLAUSE_ALLOCATE_COMBINED (*p[i]) && j == -1) + j = i; + seen = *p[i]; + } + if (i == C_OMP_CLAUSE_SPLIT_COUNT) + break; + if (j == -1) + error_at (OMP_CLAUSE_LOCATION (seen), + "%qD specified in %<allocate%> clause but not in " + "an explicit privatization clause", + OMP_CLAUSE_DECL (seen)); + for (i = 0; i < C_OMP_CLAUSE_SPLIT_COUNT; i++) + if (p[i]) + { + if (i > j) + /* Remove. */ + *p[i] = OMP_CLAUSE_CHAIN (*p[i]); + else + /* Keep. */ + p[i] = &OMP_CLAUSE_CHAIN (*p[i]); + } + } + while (1); + } + } + if (!flag_checking) return; diff --git a/gcc/c-family/c-pragma.h b/gcc/c-family/c-pragma.h index 3e9695fa..5a493fe 100644 --- a/gcc/c-family/c-pragma.h +++ b/gcc/c-family/c-pragma.h @@ -86,6 +86,7 @@ enum pragma_omp_clause { PRAGMA_OMP_CLAUSE_NONE = 0, PRAGMA_OMP_CLAUSE_ALIGNED, + PRAGMA_OMP_CLAUSE_ALLOCATE, PRAGMA_OMP_CLAUSE_BIND, PRAGMA_OMP_CLAUSE_COLLAPSE, PRAGMA_OMP_CLAUSE_COPYIN, diff --git a/gcc/c/c-parser.c b/gcc/c/c-parser.c index d49e508..b921c4e 100644 --- a/gcc/c/c-parser.c +++ b/gcc/c/c-parser.c @@ -12603,6 +12603,8 @@ c_parser_omp_clause_name (c_parser *parser) case 'a': if (!strcmp ("aligned", p)) result = PRAGMA_OMP_CLAUSE_ALIGNED; + else if (!strcmp ("allocate", p)) + result = PRAGMA_OMP_CLAUSE_ALLOCATE; else if (!strcmp ("async", p)) result = PRAGMA_OACC_CLAUSE_ASYNC; else if (!strcmp ("attach", p)) @@ -15112,6 +15114,62 @@ c_parser_omp_clause_aligned (c_parser *parser, tree list) return nl; } +/* OpenMP 5.0: + allocate ( variable-list ) + allocate ( expression : variable-list ) */ + +static tree +c_parser_omp_clause_allocate (c_parser *parser, tree list) +{ + location_t clause_loc = c_parser_peek_token (parser)->location; + tree nl, c; + tree allocator = NULL_TREE; + + matching_parens parens; + if (!parens.require_open (parser)) + return list; + + if ((c_parser_next_token_is_not (parser, CPP_NAME) + && c_parser_next_token_is_not (parser, CPP_KEYWORD)) + || (c_parser_peek_2nd_token (parser)->type != CPP_COMMA + && c_parser_peek_2nd_token (parser)->type != CPP_CLOSE_PAREN)) + { + location_t expr_loc = c_parser_peek_token (parser)->location; + c_expr expr = c_parser_expr_no_commas (parser, NULL); + expr = convert_lvalue_to_rvalue (expr_loc, expr, false, true); + allocator = expr.value; + allocator = c_fully_fold (allocator, false, NULL); + tree orig_type + = expr.original_type ? expr.original_type : TREE_TYPE (allocator); + orig_type = TYPE_MAIN_VARIANT (orig_type); + if (!INTEGRAL_TYPE_P (TREE_TYPE (allocator)) + || TREE_CODE (orig_type) != ENUMERAL_TYPE + || TYPE_NAME (orig_type) != get_identifier ("omp_allocator_handle_t")) + { + error_at (clause_loc, "%<allocate%> clause allocator expression " + "has type %qT rather than " + "%<omp_allocator_handle_t%>", + TREE_TYPE (allocator)); + allocator = NULL_TREE; + } + if (!c_parser_require (parser, CPP_COLON, "expected %<:%>")) + { + parens.skip_until_found_close (parser); + return list; + } + } + + nl = c_parser_omp_variable_list (parser, clause_loc, + OMP_CLAUSE_ALLOCATE, list); + + if (allocator) + for (c = nl; c != list; c = OMP_CLAUSE_CHAIN (c)) + OMP_CLAUSE_ALLOCATE_ALLOCATOR (c) = allocator; + + parens.skip_until_found_close (parser); + return nl; +} + /* OpenMP 4.0: linear ( variable-list ) linear ( variable-list : expression ) @@ -16354,6 +16412,10 @@ c_parser_omp_all_clauses (c_parser *parser, omp_clause_mask mask, clauses = c_parser_omp_clause_aligned (parser, clauses); c_name = "aligned"; break; + case PRAGMA_OMP_CLAUSE_ALLOCATE: + clauses = c_parser_omp_clause_allocate (parser, clauses); + c_name = "allocate"; + break; case PRAGMA_OMP_CLAUSE_LINEAR: clauses = c_parser_omp_clause_linear (parser, clauses); c_name = "linear"; @@ -18534,6 +18596,7 @@ c_parser_omp_simd (location_t loc, c_parser *parser, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_SCHEDULE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COLLAPSE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ORDER)) static tree @@ -18825,6 +18888,7 @@ c_parser_omp_sections_scope (location_t sections_loc, c_parser *parser) | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_LASTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT)) static tree @@ -18879,6 +18943,7 @@ c_parser_omp_sections (location_t loc, c_parser *parser, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COPYIN) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_THREADS) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PROC_BIND)) static tree @@ -19020,6 +19085,7 @@ c_parser_omp_parallel (location_t loc, c_parser *parser, ( (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COPYPRIVATE) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT)) static tree @@ -19054,6 +19120,7 @@ c_parser_omp_single (location_t loc, c_parser *parser, bool *if_p) | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_MERGEABLE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DEPEND) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIORITY) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_IN_REDUCTION)) static tree @@ -19124,7 +19191,8 @@ c_parser_omp_taskyield (c_parser *parser) */ #define OMP_TASKGROUP_CLAUSE_MASK \ - ( (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_TASK_REDUCTION)) + ( (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_TASK_REDUCTION)) static tree c_parser_omp_taskgroup (location_t loc, c_parser *parser, bool *if_p) @@ -19225,6 +19293,7 @@ c_parser_omp_cancellation_point (c_parser *parser, enum pragma_context context) | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_LASTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)\ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COLLAPSE)) static tree @@ -19314,6 +19383,7 @@ c_parser_omp_distribute (location_t loc, c_parser *parser, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_TEAMS) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_THREAD_LIMIT) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DEFAULT)) static tree @@ -19701,6 +19771,7 @@ c_parser_omp_target_exit_data (location_t loc, c_parser *parser, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DEFAULTMAP) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_IS_DEVICE_PTR)) @@ -21331,6 +21402,7 @@ c_finish_taskloop_clauses (tree clauses) | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_MERGEABLE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOGROUP) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIORITY) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_IN_REDUCTION)) diff --git a/gcc/c/c-typeck.c b/gcc/c/c-typeck.c index 459090e..cc9a262 100644 --- a/gcc/c/c-typeck.c +++ b/gcc/c/c-typeck.c @@ -13795,6 +13795,7 @@ c_finish_omp_clauses (tree clauses, enum c_omp_region_type ort) /* 1 if normal/task reduction has been seen, -1 if inscan reduction has been seen, -2 if mixed inscan/normal reduction diagnosed. */ int reduction_seen = 0; + bool allocate_seen = false; bitmap_obstack_initialize (NULL); bitmap_initialize (&generic_head, &bitmap_default_obstack); @@ -14344,6 +14345,29 @@ c_finish_omp_clauses (tree clauses, enum c_omp_region_type ort) bitmap_set_bit (&oacc_reduction_head, DECL_UID (t)); break; + case OMP_CLAUSE_ALLOCATE: + t = OMP_CLAUSE_DECL (c); + if (!VAR_P (t) && TREE_CODE (t) != PARM_DECL) + { + error_at (OMP_CLAUSE_LOCATION (c), + "%qE is not a variable in %<allocate%> clause", t); + remove = true; + } + else if (bitmap_bit_p (&aligned_head, DECL_UID (t))) + { + warning_at (OMP_CLAUSE_LOCATION (c), 0, + "%qE appears more than once in %<allocate%> clauses", + t); + remove = true; + } + else + { + bitmap_set_bit (&aligned_head, DECL_UID (t)); + if (!OMP_CLAUSE_ALLOCATE_COMBINED (c)) + allocate_seen = true; + } + break; + case OMP_CLAUSE_DEPEND: t = OMP_CLAUSE_DECL (c); if (t == NULL_TREE) @@ -15041,10 +15065,27 @@ c_finish_omp_clauses (tree clauses, enum c_omp_region_type ort) reduction_seen = -2; } - if (linear_variable_step_check || reduction_seen == -2) + if (linear_variable_step_check || reduction_seen == -2 || allocate_seen) for (pc = &clauses, c = clauses; c ; c = *pc) { bool remove = false; + if (allocate_seen) + switch (OMP_CLAUSE_CODE (c)) + { + case OMP_CLAUSE_PRIVATE: + case OMP_CLAUSE_FIRSTPRIVATE: + case OMP_CLAUSE_LASTPRIVATE: + case OMP_CLAUSE_LINEAR: + case OMP_CLAUSE_REDUCTION: + case OMP_CLAUSE_IN_REDUCTION: + case OMP_CLAUSE_TASK_REDUCTION: + if (DECL_P (OMP_CLAUSE_DECL (c))) + bitmap_clear_bit (&aligned_head, + DECL_UID (OMP_CLAUSE_DECL (c))); + break; + default: + break; + } if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LINEAR && OMP_CLAUSE_LINEAR_VARIABLE_STRIDE (c) && !bitmap_bit_p (&map_head, @@ -15065,6 +15106,25 @@ c_finish_omp_clauses (tree clauses, enum c_omp_region_type ort) pc = &OMP_CLAUSE_CHAIN (c); } + if (allocate_seen) + for (pc = &clauses, c = clauses; c ; c = *pc) + { + bool remove = false; + if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ALLOCATE + && !OMP_CLAUSE_ALLOCATE_COMBINED (c) + && bitmap_bit_p (&aligned_head, DECL_UID (OMP_CLAUSE_DECL (c)))) + { + error_at (OMP_CLAUSE_LOCATION (c), + "%qD specified in %<allocate%> clause but not in " + "an explicit privatization clause", OMP_CLAUSE_DECL (c)); + remove = true; + } + if (remove) + *pc = OMP_CLAUSE_CHAIN (c); + else + pc = &OMP_CLAUSE_CHAIN (c); + } + if (nogroup_seen && reduction_seen) { error_at (OMP_CLAUSE_LOCATION (*nogroup_seen), diff --git a/gcc/cp/parser.c b/gcc/cp/parser.c index 6a5469b..03780fa 100644 --- a/gcc/cp/parser.c +++ b/gcc/cp/parser.c @@ -34127,6 +34127,8 @@ cp_parser_omp_clause_name (cp_parser *parser) case 'a': if (!strcmp ("aligned", p)) result = PRAGMA_OMP_CLAUSE_ALIGNED; + else if (!strcmp ("allocate", p)) + result = PRAGMA_OMP_CLAUSE_ALLOCATE; else if (!strcmp ("async", p)) result = PRAGMA_OACC_CLAUSE_ASYNC; else if (!strcmp ("attach", p)) @@ -36282,6 +36284,47 @@ cp_parser_omp_clause_aligned (cp_parser *parser, tree list) return nlist; } +/* OpenMP 5.0: + allocate ( variable-list ) + allocate ( expression : variable-list ) */ + +static tree +cp_parser_omp_clause_allocate (cp_parser *parser, tree list) +{ + tree nlist, c, allocator = NULL_TREE; + bool colon; + + matching_parens parens; + if (!parens.require_open (parser)) + return list; + + cp_parser_parse_tentatively (parser); + bool saved_colon_corrects_to_scope_p = parser->colon_corrects_to_scope_p; + parser->colon_corrects_to_scope_p = false; + allocator = cp_parser_assignment_expression (parser); + parser->colon_corrects_to_scope_p = saved_colon_corrects_to_scope_p; + if (cp_lexer_next_token_is (parser->lexer, CPP_COLON)) + { + cp_parser_parse_definitely (parser); + cp_lexer_consume_token (parser->lexer); + if (allocator == error_mark_node) + allocator = NULL_TREE; + } + else + { + cp_parser_abort_tentative_parse (parser); + allocator = NULL_TREE; + } + + nlist = cp_parser_omp_var_list_no_open (parser, OMP_CLAUSE_ALLOCATE, list, + &colon); + + for (c = nlist; c != list; c = OMP_CLAUSE_CHAIN (c)) + OMP_CLAUSE_ALLOCATE_ALLOCATOR (c) = allocator; + + return nlist; +} + /* OpenMP 2.5: lastprivate ( variable-list ) @@ -37599,6 +37642,10 @@ cp_parser_omp_all_clauses (cp_parser *parser, omp_clause_mask mask, clauses = cp_parser_omp_clause_aligned (parser, clauses); c_name = "aligned"; break; + case PRAGMA_OMP_CLAUSE_ALLOCATE: + clauses = cp_parser_omp_clause_allocate (parser, clauses); + c_name = "allocate"; + break; case PRAGMA_OMP_CLAUSE_LINEAR: { bool declare_simd = false; @@ -39618,6 +39665,7 @@ cp_parser_omp_simd (cp_parser *parser, cp_token *pragma_tok, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_SCHEDULE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COLLAPSE) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ORDER)) static tree @@ -39902,6 +39950,7 @@ cp_parser_omp_sections_scope (cp_parser *parser) | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_LASTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT)) static tree @@ -39952,6 +40001,7 @@ cp_parser_omp_sections (cp_parser *parser, cp_token *pragma_tok, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COPYIN) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_THREADS) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PROC_BIND)) static tree @@ -40097,6 +40147,7 @@ cp_parser_omp_parallel (cp_parser *parser, cp_token *pragma_tok, ( (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COPYPRIVATE) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT)) static tree @@ -40129,6 +40180,7 @@ cp_parser_omp_single (cp_parser *parser, cp_token *pragma_tok, bool *if_p) | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_MERGEABLE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DEPEND) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIORITY) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_IN_REDUCTION)) static tree @@ -40193,7 +40245,8 @@ cp_parser_omp_taskyield (cp_parser *parser, cp_token *pragma_tok) # pragma omp taskgroup taskgroup-clause[optseq] new-line */ #define OMP_TASKGROUP_CLAUSE_MASK \ - ( (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_TASK_REDUCTION)) + ( (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_TASK_REDUCTION)) static tree cp_parser_omp_taskgroup (cp_parser *parser, cp_token *pragma_tok, bool *if_p) @@ -40302,6 +40355,7 @@ cp_parser_omp_cancellation_point (cp_parser *parser, cp_token *pragma_tok, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_LASTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)\ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COLLAPSE)) static tree @@ -40400,6 +40454,7 @@ cp_parser_omp_distribute (cp_parser *parser, cp_token *pragma_tok, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_TEAMS) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_THREAD_LIMIT) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DEFAULT)) static tree @@ -40806,6 +40861,7 @@ cp_parser_omp_target_update (cp_parser *parser, cp_token *pragma_tok, | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_FIRSTPRIVATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DEFAULTMAP) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_IS_DEVICE_PTR)) static bool @@ -42955,6 +43011,7 @@ cp_parser_omp_requires (cp_parser *parser, cp_token *pragma_tok) | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_MERGEABLE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOGROUP) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_PRIORITY) \ + | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ALLOCATE) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_REDUCTION) \ | (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_IN_REDUCTION)) diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c index 7e401e1..3c0f254 100644 --- a/gcc/cp/pt.c +++ b/gcc/cp/pt.c @@ -17306,6 +17306,7 @@ tsubst_omp_clauses (tree clauses, enum c_omp_region_type ort, break; case OMP_CLAUSE_GANG: case OMP_CLAUSE_ALIGNED: + case OMP_CLAUSE_ALLOCATE: OMP_CLAUSE_DECL (nc) = tsubst_omp_clause_decl (OMP_CLAUSE_DECL (oc), args, complain, in_decl, NULL); diff --git a/gcc/cp/semantics.c b/gcc/cp/semantics.c index ac48847..d8d3baf92 100644 --- a/gcc/cp/semantics.c +++ b/gcc/cp/semantics.c @@ -6373,6 +6373,7 @@ finish_omp_clauses (tree clauses, enum c_omp_region_type ort) /* 1 if normal/task reduction has been seen, -1 if inscan reduction has been seen, -2 if mixed inscan/normal reduction diagnosed. */ int reduction_seen = 0; + bool allocate_seen = false; bitmap_obstack_initialize (NULL); bitmap_initialize (&generic_head, &bitmap_default_obstack); @@ -7198,6 +7199,73 @@ finish_omp_clauses (tree clauses, enum c_omp_region_type ort) bitmap_set_bit (&oacc_reduction_head, DECL_UID (t)); break; + case OMP_CLAUSE_ALLOCATE: + t = OMP_CLAUSE_DECL (c); + if (t == current_class_ptr) + { + error_at (OMP_CLAUSE_LOCATION (c), + "%<this%> not allowed in %<allocate%> clause"); + remove = true; + break; + } + if (!VAR_P (t) && TREE_CODE (t) != PARM_DECL) + { + if (processing_template_decl && TREE_CODE (t) != OVERLOAD) + break; + if (DECL_P (t)) + error_at (OMP_CLAUSE_LOCATION (c), + "%qD is not a variable in %<allocate%> clause", t); + else + error_at (OMP_CLAUSE_LOCATION (c), + "%qE is not a variable in %<allocate%> clause", t); + remove = true; + } + else if (bitmap_bit_p (&aligned_head, DECL_UID (t))) + { + warning_at (OMP_CLAUSE_LOCATION (c), 0, + "%qD appears more than once in %<allocate%> clauses", + t); + remove = true; + } + else + { + bitmap_set_bit (&aligned_head, DECL_UID (t)); + allocate_seen = true; + } + t = OMP_CLAUSE_ALLOCATE_ALLOCATOR (c); + if (error_operand_p (t)) + { + remove = true; + break; + } + if (t == NULL_TREE) + break; + tree allocatort; + allocatort = TYPE_MAIN_VARIANT (TREE_TYPE (t)); + if (!type_dependent_expression_p (t) + && (TREE_CODE (allocatort) != ENUMERAL_TYPE + || TYPE_NAME (allocatort) == NULL_TREE + || TREE_CODE (TYPE_NAME (allocatort)) != TYPE_DECL + || (DECL_NAME (TYPE_NAME (allocatort)) + != get_identifier ("omp_allocator_handle_t")) + || (TYPE_CONTEXT (allocatort) + != DECL_CONTEXT (global_namespace)))) + { + error_at (OMP_CLAUSE_LOCATION (c), + "%<allocate%> clause allocator expression has " + "type %qT rather than %<omp_allocator_handle_t%>", + TREE_TYPE (t)); + remove = true; + } + else + { + t = mark_rvalue_use (t); + if (!processing_template_decl) + t = maybe_constant_value (t); + OMP_CLAUSE_ALLOCATE_ALLOCATOR (c) = t; + } + break; + case OMP_CLAUSE_DEPEND: t = OMP_CLAUSE_DECL (c); if (t == NULL_TREE) @@ -8169,6 +8237,12 @@ finish_omp_clauses (tree clauses, enum c_omp_region_type ort) { const char *share_name = NULL; + if (allocate_seen + && OMP_CLAUSE_CODE (c) != OMP_CLAUSE_SHARED + && DECL_P (t)) + bitmap_clear_bit (&aligned_head, + DECL_UID (OMP_CLAUSE_DECL (c))); + if (VAR_P (t) && CP_DECL_THREAD_LOCAL_P (t)) share_name = "threadprivate"; else switch (cxx_omp_predetermined_sharing_1 (t)) @@ -8262,6 +8336,25 @@ finish_omp_clauses (tree clauses, enum c_omp_region_type ort) pc = &OMP_CLAUSE_CHAIN (c); } + if (allocate_seen) + for (pc = &clauses, c = clauses; c ; c = *pc) + { + bool remove = false; + if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ALLOCATE + && !OMP_CLAUSE_ALLOCATE_COMBINED (c) + && bitmap_bit_p (&aligned_head, DECL_UID (OMP_CLAUSE_DECL (c)))) + { + error_at (OMP_CLAUSE_LOCATION (c), + "%qD specified in %<allocate%> clause but not in " + "an explicit privatization clause", OMP_CLAUSE_DECL (c)); + remove = true; + } + if (remove) + *pc = OMP_CLAUSE_CHAIN (c); + else + pc = &OMP_CLAUSE_CHAIN (c); + } + bitmap_obstack_release (NULL); return clauses; } diff --git a/gcc/gimplify.c b/gcc/gimplify.c index 29f385c..f766fee 100644 --- a/gcc/gimplify.c +++ b/gcc/gimplify.c @@ -9708,6 +9708,21 @@ gimplify_scan_omp_clauses (tree *list_p, gimple_seq *pre_p, omp_add_variable (ctx, decl, GOVD_NONTEMPORAL); break; + case OMP_CLAUSE_ALLOCATE: + decl = OMP_CLAUSE_DECL (c); + if (error_operand_p (decl)) + { + remove = true; + break; + } + if (gimplify_expr (&OMP_CLAUSE_ALLOCATE_ALLOCATOR (c), pre_p, NULL, + is_gimple_val, fb_rvalue) == GS_ERROR) + { + remove = true; + break; + } + break; + case OMP_CLAUSE_DEFAULT: ctx->default_kind = OMP_CLAUSE_DEFAULT_KIND (c); break; @@ -10618,6 +10633,7 @@ gimplify_adjust_omp_clauses (gimple_seq *pre_p, gimple_seq body, tree *list_p, case OMP_CLAUSE_FINALIZE: case OMP_CLAUSE_INCLUSIVE: case OMP_CLAUSE_EXCLUSIVE: + case OMP_CLAUSE_ALLOCATE: break; default: @@ -12149,7 +12165,7 @@ gimplify_omp_for (tree *expr_p, gimple_seq *pre_p) *gforo_clauses_ptr = c; gforo_clauses_ptr = &OMP_CLAUSE_CHAIN (c); break; - /* Taskloop clause we duplicate on both taskloops. */ + /* Collapse clause we duplicate on both taskloops. */ case OMP_CLAUSE_COLLAPSE: *gfor_clauses_ptr = c; gfor_clauses_ptr = &OMP_CLAUSE_CHAIN (c); @@ -12191,6 +12207,13 @@ gimplify_omp_for (tree *expr_p, gimple_seq *pre_p) gtask_clauses_ptr = &OMP_CLAUSE_CHAIN (*gtask_clauses_ptr); break; + /* Allocate clause we duplicate on task and inner taskloop. */ + case OMP_CLAUSE_ALLOCATE: + *gfor_clauses_ptr = c; + gfor_clauses_ptr = &OMP_CLAUSE_CHAIN (c); + *gtask_clauses_ptr = copy_node (c); + gtask_clauses_ptr = &OMP_CLAUSE_CHAIN (*gtask_clauses_ptr); + break; default: gcc_unreachable (); } diff --git a/gcc/omp-low.c b/gcc/omp-low.c index 6d0aa8d..5392fa7 100644 --- a/gcc/omp-low.c +++ b/gcc/omp-low.c @@ -1486,6 +1486,7 @@ scan_sharing_clauses (tree clauses, omp_context *ctx) case OMP_CLAUSE_IF_PRESENT: case OMP_CLAUSE_FINALIZE: case OMP_CLAUSE_TASK_REDUCTION: + case OMP_CLAUSE_ALLOCATE: break; case OMP_CLAUSE_ALIGNED: @@ -1653,6 +1654,7 @@ scan_sharing_clauses (tree clauses, omp_context *ctx) case OMP_CLAUSE_SIMDLEN: case OMP_CLAUSE_ALIGNED: case OMP_CLAUSE_DEPEND: + case OMP_CLAUSE_ALLOCATE: case OMP_CLAUSE__LOOPTEMP_: case OMP_CLAUSE__REDUCTEMP_: case OMP_CLAUSE_TO: diff --git a/gcc/testsuite/c-c++-common/gomp/allocate-1.c b/gcc/testsuite/c-c++-common/gomp/allocate-1.c new file mode 100644 index 0000000..29ebdf1 --- /dev/null +++ b/gcc/testsuite/c-c++-common/gomp/allocate-1.c @@ -0,0 +1,76 @@ +typedef enum omp_allocator_handle_t +#if __cplusplus >= 201103L +: __UINTPTR_TYPE__ +#endif +{ + omp_null_allocator = 0, + omp_default_mem_alloc = 1, + omp_large_cap_mem_alloc = 2, + omp_const_mem_alloc = 3, + omp_high_bw_mem_alloc = 4, + omp_low_lat_mem_alloc = 5, + omp_cgroup_mem_alloc = 6, + omp_pteam_mem_alloc = 7, + omp_thread_mem_alloc = 8, + __omp_allocator_handle_t_max__ = __UINTPTR_MAX__ +} omp_allocator_handle_t; + +int bar (int, int *, int); +omp_allocator_handle_t baz (void); + +void +foo (int x, int z) +{ + int y[16] = { 0 }, r = 0, i; + omp_allocator_handle_t h = baz (); + #pragma omp parallel allocate (x) allocate (omp_default_mem_alloc : y) \ + allocate ((omp_allocator_handle_t) omp_default_mem_alloc:z) firstprivate (x, y, z) + bar (x, y, z); + #pragma omp task private (x) firstprivate (z) allocate (omp_low_lat_mem_alloc:x,z) + bar (0, &x, z); + #pragma omp taskwait + #pragma omp target teams distribute parallel for private (x) firstprivate (y) \ + allocate ((omp_allocator_handle_t)(omp_default_mem_alloc + 0):z) \ + allocate (omp_default_mem_alloc: x, y) allocate (omp_low_lat_mem_alloc: r) \ + lastprivate (z) reduction(+:r) + for (i = 0; i < 64; i++) + { + z = bar (0, &x, 0); + r += bar (1, y, 0); + } + #pragma omp single private (x) allocate (h:x) + ; + #pragma omp single allocate (*&h : x) private (x) + ; + #pragma omp parallel shared (r, x, z) + #pragma omp single firstprivate (r) allocate (x, r, z) private (x, z) + ; + #pragma omp for allocate (x) private (x) + for (i = 0; i < 64; i++) + x = 1; + #pragma omp sections private (x) allocate (omp_low_lat_mem_alloc: x) + { + x = 1; + #pragma omp section + x = 2; + #pragma omp section + x = 3; + } + #pragma omp taskgroup task_reduction(+:r) allocate (omp_default_mem_alloc : r) + #pragma omp task in_reduction(+:r) allocate (omp_default_mem_alloc : r) + r += bar (r, &r, 0); + #pragma omp teams private (x) firstprivate (y) allocate (h : x, y) + bar (x, y, 0); + #pragma omp taskloop lastprivate (x) reduction (+:r) allocate (h : x, r) + for (i = 0; i < 16; i++) + { + r += bar (0, &r, 0); + x = i; + } + #pragma omp taskgroup task_reduction(+:r) allocate (omp_default_mem_alloc : r) + #pragma omp taskloop firstprivate (x) in_reduction (+:r) \ + allocate (omp_default_mem_alloc : x, r) + for (i = 0; i < 16; i++) + r += bar (x, &r, 0); + #pragma omp taskwait +} diff --git a/gcc/testsuite/c-c++-common/gomp/allocate-2.c b/gcc/testsuite/c-c++-common/gomp/allocate-2.c new file mode 100644 index 0000000..cc77efc --- /dev/null +++ b/gcc/testsuite/c-c++-common/gomp/allocate-2.c @@ -0,0 +1,45 @@ +typedef enum omp_allocator_handle_t +#if __cplusplus >= 201103L +: __UINTPTR_TYPE__ +#endif +{ + omp_null_allocator = 0, + omp_default_mem_alloc = 1, + omp_large_cap_mem_alloc = 2, + omp_const_mem_alloc = 3, + omp_high_bw_mem_alloc = 4, + omp_low_lat_mem_alloc = 5, + omp_cgroup_mem_alloc = 6, + omp_pteam_mem_alloc = 7, + omp_thread_mem_alloc = 8, + __omp_allocator_handle_t_max__ = __UINTPTR_MAX__ +} omp_allocator_handle_t; + +int bar (int, int *, int); +omp_allocator_handle_t baz (void); + +void +foo (int x, int z) +{ + int i; + #pragma omp task allocate (x) /* { dg-error "'x' specified in 'allocate' clause but not in an explicit privatization clause" } */ + bar (x, &x, 0); + #pragma omp taskwait + #pragma omp parallel allocate (x) /* { dg-error "'x' specified in 'allocate' clause but not in an explicit privatization clause" } */ + bar (x, &x, 0); + #pragma omp parallel for simd private (x) allocate (x) /* { dg-error "'x' specified in 'allocate' clause but not in an explicit privatization clause" } */ + for (i = 0; i < 16; i++) + x = i; + #pragma omp parallel allocate (foo) /* { dg-error "'\[^\n\r]*foo\[^\n\r]*' is not a variable in 'allocate' clause" } */ + ; + #pragma omp parallel allocate (x) shared (x) /* { dg-error "'x' specified in 'allocate' clause but not in an explicit privatization clause" } */ + bar (x, &x, 0); + #pragma omp parallel private (x) allocate (x) allocate (x) /* { dg-warning "'x' appears more than once in 'allocate' clauses" } */ + bar (x, &x, 0); + #pragma omp parallel private (x) allocate (x, x) /* { dg-warning "'x' appears more than once in 'allocate' clauses" } */ + bar (x, &x, 0); + #pragma omp parallel private (x) allocate (0.0 : x) /* { dg-error "'allocate' clause allocator expression has type 'double' rather than 'omp_allocator_handle_t'" } */ + bar (x, &x, 0); + #pragma omp parallel private (x) allocate (0 : x) /* { dg-error "'allocate' clause allocator expression has type 'int' rather than 'omp_allocator_handle_t'" } */ + bar (x, &x, 0); +} diff --git a/gcc/testsuite/c-c++-common/gomp/clauses-1.c b/gcc/testsuite/c-c++-common/gomp/clauses-1.c index be42797..105288e 100644 --- a/gcc/testsuite/c-c++-common/gomp/clauses-1.c +++ b/gcc/testsuite/c-c++-common/gomp/clauses-1.c @@ -1,6 +1,23 @@ /* { dg-do compile } */ /* { dg-additional-options "-std=c99" { target c } } */ +typedef enum omp_allocator_handle_t +#if __cplusplus >= 201103L +: __UINTPTR_TYPE__ +#endif +{ + omp_null_allocator = 0, + omp_default_mem_alloc = 1, + omp_large_cap_mem_alloc = 2, + omp_const_mem_alloc = 3, + omp_high_bw_mem_alloc = 4, + omp_low_lat_mem_alloc = 5, + omp_cgroup_mem_alloc = 6, + omp_pteam_mem_alloc = 7, + omp_thread_mem_alloc = 8, + __omp_allocator_handle_t_max__ = __UINTPTR_MAX__ +} omp_allocator_handle_t; + int t; #pragma omp threadprivate (t) @@ -14,20 +31,20 @@ foo (int d, int m, int i1, int i2, int p, int *idp, int s, #pragma omp distribute parallel for \ private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) schedule(static, 4) order(concurrent) + lastprivate (l) schedule(static, 4) order(concurrent) allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp distribute parallel for simd \ private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \ if (parallel: i2) if(simd: i1) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ lastprivate (l) schedule(static, 4) nontemporal(ntm) \ - safelen(8) simdlen(4) aligned(q: 32) order(concurrent) + safelen(8) simdlen(4) aligned(q: 32) order(concurrent) allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp distribute simd \ private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \ safelen(8) simdlen(4) aligned(q: 32) reduction(+:r) if(i1) nontemporal(ntm) \ - order(concurrent) + order(concurrent) allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; } @@ -49,20 +66,20 @@ baz (int d, int m, int i1, int i2, int p, int *idp, int s, #pragma omp distribute parallel for \ private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) schedule(static, 4) copyin(t) order(concurrent) + lastprivate (l) schedule(static, 4) copyin(t) order(concurrent) allocate (p) for (int i = 0; i < 64; i++) ll++; #pragma omp distribute parallel for simd \ private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \ if (parallel: i2) if(simd: i1) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ lastprivate (l) schedule(static, 4) nontemporal(ntm) \ - safelen(8) simdlen(4) aligned(q: 32) copyin(t) order(concurrent) + safelen(8) simdlen(4) aligned(q: 32) copyin(t) order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp distribute simd \ private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \ safelen(8) simdlen(4) aligned(q: 32) reduction(+:r) if(i1) nontemporal(ntm) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp loop bind(parallel) order(concurrent) \ @@ -77,28 +94,28 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, { #pragma omp for simd \ private (p) firstprivate (f) lastprivate (l) linear (ll:1) reduction(+:r) schedule(static, 4) collapse(1) nowait \ - safelen(8) simdlen(4) aligned(q: 32) nontemporal(ntm) if(i1) order(concurrent) + safelen(8) simdlen(4) aligned(q: 32) nontemporal(ntm) if(i1) order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel for \ private (p) firstprivate (f) if (parallel: i2) default(shared) shared(s) copyin(t) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) linear (ll:1) ordered schedule(static, 4) collapse(1) + lastprivate (l) linear (ll:1) ordered schedule(static, 4) collapse(1) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel for \ private (p) firstprivate (f) if (parallel: i2) default(shared) shared(s) copyin(t) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) linear (ll:1) schedule(static, 4) collapse(1) order(concurrent) + lastprivate (l) linear (ll:1) schedule(static, 4) collapse(1) order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel for simd \ private (p) firstprivate (f) if (i2) default(shared) shared(s) copyin(t) reduction(+:r) num_threads (nth) proc_bind(spread) \ lastprivate (l) linear (ll:1) schedule(static, 4) collapse(1) \ - safelen(8) simdlen(4) aligned(q: 32) nontemporal(ntm) order(concurrent) + safelen(8) simdlen(4) aligned(q: 32) nontemporal(ntm) order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel sections \ private (p) firstprivate (f) if (parallel: i2) default(shared) shared(s) copyin(t) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) + lastprivate (l) allocate (f) { #pragma omp section {} @@ -108,35 +125,39 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, #pragma omp target parallel \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ - nowait depend(inout: dd[0]) + nowait depend(inout: dd[0]) allocate (omp_default_mem_alloc:f) ; #pragma omp target parallel for \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) linear (ll:1) ordered schedule(static, 4) collapse(1) nowait depend(inout: dd[0]) + lastprivate (l) linear (ll:1) ordered schedule(static, 4) collapse(1) nowait depend(inout: dd[0]) \ + allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp target parallel for \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) linear (ll:1) schedule(static, 4) collapse(1) nowait depend(inout: dd[0]) order(concurrent) + lastprivate (l) linear (ll:1) schedule(static, 4) collapse(1) nowait depend(inout: dd[0]) order(concurrent) \ + allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp target parallel for simd \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ lastprivate (l) linear (ll:1) schedule(static, 4) collapse(1) \ - safelen(8) simdlen(4) aligned(q: 32) nowait depend(inout: dd[0]) nontemporal(ntm) if (simd: i3) order(concurrent) + safelen(8) simdlen(4) aligned(q: 32) nowait depend(inout: dd[0]) nontemporal(ntm) if (simd: i3) order(concurrent) \ + allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp target teams \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ - shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) nowait depend(inout: dd[0]) + shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) nowait depend(inout: dd[0]) \ + allocate (omp_default_mem_alloc:f) ; #pragma omp target teams distribute \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ - collapse(1) dist_schedule(static, 16) nowait depend(inout: dd[0]) + collapse(1) dist_schedule(static, 16) nowait depend(inout: dd[0]) allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ; #pragma omp target teams distribute parallel for \ @@ -144,7 +165,8 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ collapse(1) dist_schedule(static, 16) \ if (parallel: i2) num_threads (nth) proc_bind(spread) \ - lastprivate (l) schedule(static, 4) nowait depend(inout: dd[0]) order(concurrent) + lastprivate (l) schedule(static, 4) nowait depend(inout: dd[0]) order(concurrent) \ + allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp target teams distribute parallel for simd \ @@ -153,47 +175,50 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, collapse(1) dist_schedule(static, 16) \ if (parallel: i2) num_threads (nth) proc_bind(spread) \ lastprivate (l) schedule(static, 4) order(concurrent) \ - safelen(8) simdlen(4) aligned(q: 32) nowait depend(inout: dd[0]) nontemporal(ntm) if (simd: i3) + safelen(8) simdlen(4) aligned(q: 32) nowait depend(inout: dd[0]) nontemporal(ntm) if (simd: i3) \ + allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp target teams distribute simd \ device(d) map (tofrom: m) if (i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ collapse(1) dist_schedule(static, 16) order(concurrent) \ - safelen(8) simdlen(4) aligned(q: 32) nowait depend(inout: dd[0]) nontemporal(ntm) + safelen(8) simdlen(4) aligned(q: 32) nowait depend(inout: dd[0]) nontemporal(ntm) \ + allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; #pragma omp target simd \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ safelen(8) simdlen(4) lastprivate (l) linear(ll: 1) aligned(q: 32) reduction(+:r) \ - nowait depend(inout: dd[0]) nontemporal(ntm) if(simd:i3) order(concurrent) + nowait depend(inout: dd[0]) nontemporal(ntm) if(simd:i3) order(concurrent) \ + allocate (omp_default_mem_alloc:f) for (int i = 0; i < 64; i++) ll++; - #pragma omp taskgroup task_reduction(+:r2) + #pragma omp taskgroup task_reduction(+:r2) allocate (r2) #pragma omp taskloop simd \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) grainsize (g) collapse(1) untied if(taskloop: i1) if(simd: i2) final(fi) mergeable priority (pp) \ safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) reduction(default, +:r) in_reduction(+:r2) nontemporal(ntm) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; - #pragma omp taskgroup task_reduction(+:r) + #pragma omp taskgroup task_reduction(+:r) allocate (r) #pragma omp taskloop simd \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) grainsize (g) collapse(1) untied if(i1) final(fi) mergeable nogroup priority (pp) \ safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) in_reduction(+:r) nontemporal(ntm) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp taskwait #pragma omp taskloop simd \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) num_tasks (nta) collapse(1) if(taskloop: i1) final(fi) priority (pp) \ safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) reduction(+:r) if (simd: i3) nontemporal(ntm) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp target nowait depend(inout: dd[0]) #pragma omp teams distribute \ private(p) firstprivate (f) shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ - collapse(1) dist_schedule(static, 16) + collapse(1) dist_schedule(static, 16) allocate (omp_default_mem_alloc: f) for (int i = 0; i < 64; i++) ; #pragma omp target @@ -201,7 +226,7 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, private(p) firstprivate (f) shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ collapse(1) dist_schedule(static, 16) \ if (parallel: i2) num_threads (nth) proc_bind(spread) \ - lastprivate (l) schedule(static, 4) order(concurrent) + lastprivate (l) schedule(static, 4) order(concurrent) allocate (omp_default_mem_alloc: f) for (int i = 0; i < 64; i++) ll++; #pragma omp target @@ -210,21 +235,23 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, collapse(1) dist_schedule(static, 16) \ if (parallel: i2) num_threads (nth) proc_bind(spread) \ lastprivate (l) schedule(static, 4) order(concurrent) \ - safelen(8) simdlen(4) aligned(q: 32) if (simd: i3) nontemporal(ntm) + safelen(8) simdlen(4) aligned(q: 32) if (simd: i3) nontemporal(ntm) \ + allocate (omp_default_mem_alloc: f) for (int i = 0; i < 64; i++) ll++; #pragma omp target #pragma omp teams distribute simd \ private(p) firstprivate (f) shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ collapse(1) dist_schedule(static, 16) order(concurrent) \ - safelen(8) simdlen(4) aligned(q: 32) if(i3) nontemporal(ntm) + safelen(8) simdlen(4) aligned(q: 32) if(i3) nontemporal(ntm) \ + allocate (omp_default_mem_alloc: f) for (int i = 0; i < 64; i++) ll++; #pragma omp teams distribute parallel for \ private(p) firstprivate (f) shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ collapse(1) dist_schedule(static, 16) order(concurrent) \ if (parallel: i2) num_threads (nth) proc_bind(spread) \ - lastprivate (l) schedule(static, 4) copyin(t) + lastprivate (l) schedule(static, 4) copyin(t) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp teams distribute parallel for simd \ @@ -232,65 +259,66 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, collapse(1) dist_schedule(static, 16) \ if (parallel: i2) num_threads (nth) proc_bind(spread) \ lastprivate (l) schedule(static, 4) order(concurrent) \ - safelen(8) simdlen(4) aligned(q: 32) if (simd: i3) nontemporal(ntm) copyin(t) + safelen(8) simdlen(4) aligned(q: 32) if (simd: i3) nontemporal(ntm) copyin(t) \ + allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp teams distribute simd \ private(p) firstprivate (f) shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ collapse(1) dist_schedule(static, 16) order(concurrent) \ - safelen(8) simdlen(4) aligned(q: 32) if(i3) nontemporal(ntm) + safelen(8) simdlen(4) aligned(q: 32) if(i3) nontemporal(ntm) allocate(f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel master \ private (p) firstprivate (f) if (parallel: i2) default(shared) shared(s) reduction(+:r) \ - num_threads (nth) proc_bind(spread) copyin(t) + num_threads (nth) proc_bind(spread) copyin(t) allocate (f) ; - #pragma omp taskgroup task_reduction (+:r2) + #pragma omp taskgroup task_reduction (+:r2) allocate (r2) #pragma omp master taskloop \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) grainsize (g) collapse(1) untied if(taskloop: i1) final(fi) mergeable priority (pp) \ - reduction(default, +:r) in_reduction(+:r2) + reduction(default, +:r) in_reduction(+:r2) allocate (f) for (int i = 0; i < 64; i++) ll++; - #pragma omp taskgroup task_reduction (+:r2) + #pragma omp taskgroup task_reduction (+:r2) allocate (r2) #pragma omp master taskloop simd \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) grainsize (g) collapse(1) untied if(taskloop: i1) if(simd: i2) final(fi) mergeable priority (pp) \ safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) reduction(default, +:r) in_reduction(+:r2) nontemporal(ntm) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel master taskloop \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) grainsize (g) collapse(1) untied if(taskloop: i1) final(fi) mergeable priority (pp) \ - reduction(default, +:r) if (parallel: i2) num_threads (nth) proc_bind(spread) copyin(t) + reduction(default, +:r) if (parallel: i2) num_threads (nth) proc_bind(spread) copyin(t) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel master taskloop simd \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) grainsize (g) collapse(1) untied if(taskloop: i1) if(simd: i2) final(fi) mergeable priority (pp) \ safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) reduction(default, +:r) nontemporal(ntm) if (parallel: i2) num_threads (nth) proc_bind(spread) copyin(t) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; - #pragma omp taskgroup task_reduction (+:r2) + #pragma omp taskgroup task_reduction (+:r2) allocate (r2) #pragma omp master taskloop \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) num_tasks (nta) collapse(1) untied if(i1) final(fi) mergeable priority (pp) \ reduction(default, +:r) in_reduction(+:r2) for (int i = 0; i < 64; i++) ll++; - #pragma omp taskgroup task_reduction (+:r2) + #pragma omp taskgroup task_reduction (+:r2) allocate (r2) #pragma omp master taskloop simd \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) num_tasks (nta) collapse(1) untied if(i1) final(fi) mergeable priority (pp) \ safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) reduction(default, +:r) in_reduction(+:r2) nontemporal(ntm) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel master taskloop \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) num_tasks (nta) collapse(1) untied if(i1) final(fi) mergeable priority (pp) \ - reduction(default, +:r) num_threads (nth) proc_bind(spread) copyin(t) + reduction(default, +:r) num_threads (nth) proc_bind(spread) copyin(t) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp parallel master taskloop simd \ private (p) firstprivate (f) lastprivate (l) shared (s) default(shared) num_tasks (nta) collapse(1) untied if(i1) final(fi) mergeable priority (pp) \ safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) reduction(default, +:r) nontemporal(ntm) num_threads (nth) proc_bind(spread) copyin(t) \ - order(concurrent) + order(concurrent) allocate (f) for (int i = 0; i < 64; i++) ll++; #pragma omp loop bind(thread) order(concurrent) \ @@ -299,46 +327,50 @@ bar (int d, int m, int i1, int i2, int i3, int p, int *idp, int s, ll++; #pragma omp parallel loop \ private (p) firstprivate (f) if (parallel: i2) default(shared) shared(s) copyin(t) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) collapse(1) bind(parallel) order(concurrent) + lastprivate (l) collapse(1) bind(parallel) order(concurrent) allocate (f) for (l = 0; l < 64; l++) ll++; #pragma omp parallel loop \ private (p) firstprivate (f) if (parallel: i2) default(shared) shared(s) copyin(t) reduction(+:r) num_threads (nth) proc_bind(spread) \ - lastprivate (l) collapse(1) + lastprivate (l) collapse(1) allocate (f) for (l = 0; l < 64; l++) ll++; #pragma omp teams loop \ private(p) firstprivate (f) shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ - collapse(1) lastprivate (l) bind(teams) + collapse(1) lastprivate (l) bind(teams) allocate (f) for (l = 0; l < 64; ++l) ; #pragma omp teams loop \ private(p) firstprivate (f) shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \ - collapse(1) lastprivate (l) order(concurrent) + collapse(1) lastprivate (l) order(concurrent) allocate (f) for (l = 0; l < 64; ++l) ; #pragma omp target parallel loop \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ - nowait depend(inout: dd[0]) lastprivate (l) bind(parallel) order(concurrent) collapse(1) + nowait depend(inout: dd[0]) lastprivate (l) bind(parallel) order(concurrent) collapse(1) \ + allocate (omp_default_mem_alloc: f) for (l = 0; l < 64; ++l) ; #pragma omp target parallel loop \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads (nth) proc_bind(spread) \ - nowait depend(inout: dd[0]) lastprivate (l) order(concurrent) collapse(1) + nowait depend(inout: dd[0]) lastprivate (l) order(concurrent) collapse(1) \ + allocate (omp_default_mem_alloc: f) for (l = 0; l < 64; ++l) ; #pragma omp target teams loop \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) nowait depend(inout: dd[0]) \ - lastprivate (l) bind(teams) collapse(1) + lastprivate (l) bind(teams) collapse(1) \ + allocate (omp_default_mem_alloc: f) for (l = 0; l < 64; ++l) ; #pragma omp target teams loop \ device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f) defaultmap(tofrom: scalar) is_device_ptr (idp) \ shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) nowait depend(inout: dd[0]) \ - lastprivate (l) order(concurrent) collapse(1) + lastprivate (l) order(concurrent) collapse(1) \ + allocate (omp_default_mem_alloc: f) for (l = 0; l < 64; ++l) ; } diff --git a/gcc/tree-core.h b/gcc/tree-core.h index 752bec3..c9280a8 100644 --- a/gcc/tree-core.h +++ b/gcc/tree-core.h @@ -276,6 +276,9 @@ enum omp_clause_code { /* OpenMP clause: aligned (variable-list[:alignment]). */ OMP_CLAUSE_ALIGNED, + /* OpenMP clause: allocate ([allocator:]variable-list). */ + OMP_CLAUSE_ALLOCATE, + /* OpenMP clause: depend ({in,out,inout}:variable-list). */ OMP_CLAUSE_DEPEND, diff --git a/gcc/tree-nested.c b/gcc/tree-nested.c index 9cb4a08..a4d82174 100644 --- a/gcc/tree-nested.c +++ b/gcc/tree-nested.c @@ -1441,6 +1441,7 @@ convert_nonlocal_omp_clauses (tree *pclauses, struct walk_stmt_info *wi) } /* FALLTHRU */ case OMP_CLAUSE_NONTEMPORAL: + do_decl_clause_no_supp: /* Like do_decl_clause, but don't add any suppression. */ decl = OMP_CLAUSE_DECL (clause); if (VAR_P (decl) @@ -1453,6 +1454,16 @@ convert_nonlocal_omp_clauses (tree *pclauses, struct walk_stmt_info *wi) } break; + case OMP_CLAUSE_ALLOCATE: + if (OMP_CLAUSE_ALLOCATE_ALLOCATOR (clause)) + { + wi->val_only = true; + wi->is_lhs = false; + convert_nonlocal_reference_op + (&OMP_CLAUSE_ALLOCATE_ALLOCATOR (clause), &dummy, wi); + } + goto do_decl_clause_no_supp; + case OMP_CLAUSE_NOWAIT: case OMP_CLAUSE_ORDERED: case OMP_CLAUSE_DEFAULT: @@ -2203,6 +2214,7 @@ convert_local_omp_clauses (tree *pclauses, struct walk_stmt_info *wi) } /* FALLTHRU */ case OMP_CLAUSE_NONTEMPORAL: + do_decl_clause_no_supp: /* Like do_decl_clause, but don't add any suppression. */ decl = OMP_CLAUSE_DECL (clause); if (VAR_P (decl) @@ -2221,6 +2233,16 @@ convert_local_omp_clauses (tree *pclauses, struct walk_stmt_info *wi) } break; + case OMP_CLAUSE_ALLOCATE: + if (OMP_CLAUSE_ALLOCATE_ALLOCATOR (clause)) + { + wi->val_only = true; + wi->is_lhs = false; + convert_local_reference_op + (&OMP_CLAUSE_ALLOCATE_ALLOCATOR (clause), &dummy, wi); + } + goto do_decl_clause_no_supp; + case OMP_CLAUSE_NOWAIT: case OMP_CLAUSE_ORDERED: case OMP_CLAUSE_DEFAULT: diff --git a/gcc/tree-pretty-print.c b/gcc/tree-pretty-print.c index 64ac5ab..bd60d74 100644 --- a/gcc/tree-pretty-print.c +++ b/gcc/tree-pretty-print.c @@ -712,6 +712,19 @@ dump_omp_clause (pretty_printer *pp, tree clause, int spc, dump_flags_t flags) pp_right_paren (pp); break; + case OMP_CLAUSE_ALLOCATE: + pp_string (pp, "allocate("); + if (OMP_CLAUSE_ALLOCATE_ALLOCATOR (clause)) + { + dump_generic_node (pp, OMP_CLAUSE_ALLOCATE_ALLOCATOR (clause), + spc, flags, false); + pp_colon (pp); + } + dump_generic_node (pp, OMP_CLAUSE_DECL (clause), + spc, flags, false); + pp_right_paren (pp); + break; + case OMP_CLAUSE_DEPEND: pp_string (pp, "depend("); switch (OMP_CLAUSE_DEPEND_KIND (clause)) @@ -291,6 +291,7 @@ unsigned const char omp_clause_num_ops[] = 1, /* OMP_CLAUSE_COPYPRIVATE */ 3, /* OMP_CLAUSE_LINEAR */ 2, /* OMP_CLAUSE_ALIGNED */ + 2, /* OMP_CLAUSE_ALLOCATE */ 1, /* OMP_CLAUSE_DEPEND */ 1, /* OMP_CLAUSE_NONTEMPORAL */ 1, /* OMP_CLAUSE_UNIFORM */ @@ -375,6 +376,7 @@ const char * const omp_clause_code_name[] = "copyprivate", "linear", "aligned", + "allocate", "depend", "nontemporal", "uniform", @@ -12213,6 +12215,7 @@ walk_tree_1 (tree *tp, walk_tree_fn func, void *data, WALK_SUBTREE_TAIL (OMP_CLAUSE_CHAIN (*tp)); case OMP_CLAUSE_ALIGNED: + case OMP_CLAUSE_ALLOCATE: case OMP_CLAUSE_FROM: case OMP_CLAUSE_TO: case OMP_CLAUSE_MAP: @@ -1731,6 +1731,16 @@ class auto_suppress_location_wrappers #define OMP_CLAUSE_ALIGNED_ALIGNMENT(NODE) \ OMP_CLAUSE_OPERAND (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_ALIGNED), 1) +#define OMP_CLAUSE_ALLOCATE_ALLOCATOR(NODE) \ + OMP_CLAUSE_OPERAND (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_ALLOCATE), 1) + +/* True if an ALLOCATE clause was present on a combined or composite + construct and the code for splitting the clauses has already performed + checking if the listed variable has explicit privatization on the + construct. */ +#define OMP_CLAUSE_ALLOCATE_COMBINED(NODE) \ + (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_ALLOCATE)->base.public_flag) + #define OMP_CLAUSE_NUM_TEAMS_EXPR(NODE) \ OMP_CLAUSE_OPERAND (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_NUM_TEAMS), 0) |