openmp: Add omp_all_memory support (C/C++ only so far)

The ugly part is that OpenMP 5.1 made omp_all_memory a reserved identifier
which isn't allowed to be used anywhere but in the depend clause, this is
against how everything else has been handled in OpenMP so far (where
some identifiers could have special meaning in some OpenMP clauses or
pragmas but not elsewhere).
The patch handles it by making it a conditional keyword (for -fopenmp
only) and emitting a better diagnostics when it is used in a primary
expression.  Having a nicer diagnostics when e.g. trying to do
int omp_all_memory;
or
int *omp_all_memory[10];
etc. would mean changing too many spots and hooking into name lookups
to reject declaring any such symbols would be too ugly and I'm afraid
there are way too many spots where one can introduce a name
(variables, functions, namespaces, struct, enum, enumerators, template
arguments, ...).

Otherwise, the handling is quite simple, normal depend clauses lower
into addresses of variables being handed over to the library, for
omp_all_memory I'm using NULL pointers.  omp_all_memory can only be
used with inout or out depend kinds and means that a task is dependent
on all previously created sibling tasks that have any dependency (of
any depend kind) and that any later created sibling tasks will be
dependent on it if they have any dependency.

2022-05-12  Jakub Jelinek  <jakub@redhat.com>

gcc/
	* gimplify.cc (gimplify_omp_depend): Don't build_fold_addr_expr
	if null_pointer_node.
	(gimplify_scan_omp_clauses): Likewise.
	* tree-pretty-print.cc (dump_omp_clause): Print null_pointer_node
	as omp_all_memory.
gcc/c-family/
	* c-common.h (enum rid): Add RID_OMP_ALL_MEMORY.
	* c-omp.cc (c_finish_omp_depobj): Don't build_fold_addr_expr
	if null_pointer_node.
gcc/c/
	* c-parser.cc (c_parse_init): Register omp_all_memory as keyword
	if flag_openmp.
	(c_parser_postfix_expression): Diagnose uses of omp_all_memory
	in postfix expressions.
	(c_parser_omp_variable_list): Handle omp_all_memory in depend
	clause.
	* c-typeck.cc (c_finish_omp_clauses): Handle omp_all_memory
	keyword in depend clause as null_pointer_node, diagnose invalid
	uses.
gcc/cp/
	* lex.cc (init_reswords): Register omp_all_memory as keyword
	if flag_openmp.
	* parser.cc (cp_parser_primary_expression): Diagnose uses of
	omp_all_memory in postfix expressions.
	(cp_parser_omp_var_list_no_open): Handle omp_all_memory in depend
	clause.
	* semantics.cc (finish_omp_clauses): Handle omp_all_memory
	keyword in depend clause as null_pointer_node, diagnose invalid
	uses.
	* pt.cc (tsubst_omp_clause_decl): Pass through omp_all_memory.
gcc/testsuite/
	* c-c++-common/gomp/all-memory-1.c: New test.
	* c-c++-common/gomp/all-memory-2.c: New test.
	* c-c++-common/gomp/all-memory-3.c: New test.
	* g++.dg/gomp/all-memory-1.C: New test.
	* g++.dg/gomp/all-memory-2.C: New test.
libgomp/
	* libgomp.h (struct gomp_task): Add depend_all_memory member.
	* task.c (gomp_init_task): Initialize depend_all_memory.
	(gomp_task_handle_depend): Handle omp_all_memory.
	(gomp_task_run_post_handle_depend_hash): Clear
	parent->depend_all_memory if equal to current task.
	(gomp_task_maybe_wait_for_dependencies): Handle omp_all_memory.
	* testsuite/libgomp.c-c++-common/depend-1.c: New test.
	* testsuite/libgomp.c-c++-common/depend-2.c: New test.
	* testsuite/libgomp.c-c++-common/depend-3.c: New test.

20 files changed, 817 insertions, 36 deletions

diff --git a/gcc/c-family/c-common.h b/gcc/c-family/c-common.h
index aa043de..47442c9 100644
--- a/gcc/c-family/c-common.h
+++ b/gcc/c-family/c-common.h
@@ -219,6 +219,9 @@ enum rid
   RID_AT_INTERFACE,
   RID_AT_IMPLEMENTATION,
 
+  /* OpenMP */
+  RID_OMP_ALL_MEMORY,
+
   /* Named address support, mapping the keyword to a particular named address
      number.  Named address space 0 is reserved for the generic address.  If
      there are more than 254 named addresses, the addr_space_t type will need
diff --git a/gcc/c-family/c-omp.cc b/gcc/c-family/c-omp.cc
index 777cdc6..987ba7d 100644
--- a/gcc/c-family/c-omp.cc
+++ b/gcc/c-family/c-omp.cc
@@ -757,7 +757,7 @@ c_finish_omp_depobj (location_t loc, tree depobj,
 	      t = build2 (COMPOUND_EXPR, TREE_TYPE (t1), TREE_OPERAND (t, 0),
 			  t1);
 	    }
-	  else
+	  else if (t != null_pointer_node)
 	    t = build_fold_addr_expr (t);
 	  break;
 	default:
diff --git a/gcc/c/c-parser.cc b/gcc/c/c-parser.cc
index d431d5f..51a0725 100644
--- a/gcc/c/c-parser.cc
+++ b/gcc/c/c-parser.cc
@@ -165,6 +165,14 @@ c_parse_init (void)
       C_SET_RID_CODE (id, RID_FIRST_INT_N + i);
       C_IS_RESERVED_WORD (id) = 1;
     }
+
+  if (flag_openmp)
+    {
+      id = get_identifier ("omp_all_memory");
+      C_SET_RID_CODE (id, RID_OMP_ALL_MEMORY);
+      C_IS_RESERVED_WORD (id) = 1;
+      ridpointers [RID_OMP_ALL_MEMORY] = id;
+    }
 }
 
 /* A parser structure recording information about the state and
@@ -10202,6 +10210,13 @@ c_parser_postfix_expression (c_parser *parser)
 	case RID_GENERIC:
 	  expr = c_parser_generic_selection (parser);
 	  break;
+	case RID_OMP_ALL_MEMORY:
+	  gcc_assert (flag_openmp);
+	  c_parser_consume_token (parser);
+	  error_at (loc, "%<omp_all_memory%> may only be used in OpenMP "
+			 "%<depend%> clause");
+	  expr.set_error ();
+	  break;
 	default:
 	  c_parser_error (parser, "expected expression");
 	  expr.set_error ();
@@ -13025,7 +13040,19 @@ c_parser_omp_variable_list (c_parser *parser,
 	  if (c_parser_next_token_is_not (parser, CPP_NAME)
 	      || c_parser_peek_token (parser)->id_kind != C_ID_ID)
 	    {
-	      struct c_expr expr = c_parser_expr_no_commas (parser, NULL);
+	      struct c_expr expr;
+	      if (kind == OMP_CLAUSE_DEPEND
+		  && c_parser_next_token_is_keyword (parser,
+						     RID_OMP_ALL_MEMORY)
+		  && (c_parser_peek_2nd_token (parser)->type == CPP_COMMA
+		      || (c_parser_peek_2nd_token (parser)->type
+			  == CPP_CLOSE_PAREN)))
+		{
+		  expr.value = ridpointers[RID_OMP_ALL_MEMORY];
+		  c_parser_consume_token (parser);
+		}
+	      else
+		expr = c_parser_expr_no_commas (parser, NULL);
 	      if (expr.value != error_mark_node)
 		{
 		  tree u = build_omp_clause (clause_loc, kind);
diff --git a/gcc/c/c-typeck.cc b/gcc/c/c-typeck.cc
index e130196..bcfe08b 100644
--- a/gcc/c/c-typeck.cc
+++ b/gcc/c/c-typeck.cc
@@ -14832,6 +14832,18 @@ c_finish_omp_clauses (tree clauses, enum c_omp_region_type ort)
 	    }
 	  if (t == error_mark_node)
 	    remove = true;
+	  else if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_DEPEND
+		   && t == ridpointers[RID_OMP_ALL_MEMORY])
+	    {
+	      if (OMP_CLAUSE_DEPEND_KIND (c) != OMP_CLAUSE_DEPEND_OUT
+		  && OMP_CLAUSE_DEPEND_KIND (c) != OMP_CLAUSE_DEPEND_INOUT)
+		{
+		  error_at (OMP_CLAUSE_LOCATION (c),
+			    "%<omp_all_memory%> used with %<depend%> kind "
+			    "other than %<out%> or %<inout%>");
+		  remove = true;
+		}
+	    }
 	  else if (!lvalue_p (t))
 	    {
 	      error_at (OMP_CLAUSE_LOCATION (c),
@@ -14873,24 +14885,32 @@ c_finish_omp_clauses (tree clauses, enum c_omp_region_type ort)
 	    }
 	  if (!remove)
 	    {
-	      tree addr = build_unary_op (OMP_CLAUSE_LOCATION (c), ADDR_EXPR,
-					  t, false);
-	      if (addr == error_mark_node)
-		remove = true;
+	      if (t == ridpointers[RID_OMP_ALL_MEMORY])
+		t = null_pointer_node;
 	      else
 		{
+		  tree addr = build_unary_op (OMP_CLAUSE_LOCATION (c),
+					      ADDR_EXPR, t, false);
+		  if (addr == error_mark_node)
+		    {
+		      remove = true;
+		      break;
+		    }
 		  t = build_indirect_ref (OMP_CLAUSE_LOCATION (c), addr,
 					  RO_UNARY_STAR);
 		  if (t == error_mark_node)
-		    remove = true;
-		  else if (TREE_CODE (OMP_CLAUSE_DECL (c)) == TREE_LIST
-			   && TREE_PURPOSE (OMP_CLAUSE_DECL (c))
-			   && (TREE_CODE (TREE_PURPOSE (OMP_CLAUSE_DECL (c)))
-			       == TREE_VEC))
-		    TREE_VALUE (OMP_CLAUSE_DECL (c)) = t;
-		  else
-		    OMP_CLAUSE_DECL (c) = t;
+		    {
+		      remove = true;
+		      break;
+		    }
 		}
+	      if (TREE_CODE (OMP_CLAUSE_DECL (c)) == TREE_LIST
+		  && TREE_PURPOSE (OMP_CLAUSE_DECL (c))
+		  && (TREE_CODE (TREE_PURPOSE (OMP_CLAUSE_DECL (c)))
+		      == TREE_VEC))
+		TREE_VALUE (OMP_CLAUSE_DECL (c)) = t;
+	      else
+		OMP_CLAUSE_DECL (c) = t;
 	    }
 	  break;
 
diff --git a/gcc/cp/lex.cc b/gcc/cp/lex.cc
index 784debc..0b121a9 100644
--- a/gcc/cp/lex.cc
+++ b/gcc/cp/lex.cc
@@ -273,6 +273,14 @@ init_reswords (void)
       C_SET_RID_CODE (id, RID_FIRST_INT_N + i);
       set_identifier_kind (id, cik_keyword);
     }
+
+  if (flag_openmp)
+    {
+      id = get_identifier ("omp_all_memory");
+      C_SET_RID_CODE (id, RID_OMP_ALL_MEMORY);
+      set_identifier_kind (id, cik_keyword);
+      ridpointers [RID_OMP_ALL_MEMORY] = id;
+    }
 }
 
 static void
diff --git a/gcc/cp/parser.cc b/gcc/cp/parser.cc
index d09d5b7..84f379c 100644
--- a/gcc/cp/parser.cc
+++ b/gcc/cp/parser.cc
@@ -5876,6 +5876,14 @@ cp_parser_primary_expression (cp_parser *parser,
 	case RID_AT_SELECTOR:
 	  return cp_parser_objc_expression (parser);
 
+	case RID_OMP_ALL_MEMORY:
+	  gcc_assert (flag_openmp);
+	  cp_lexer_consume_token (parser->lexer);
+	  error_at (token->location,
+		    "%<omp_all_memory%> may only be used in OpenMP "
+		    "%<depend%> clause");
+	  return error_mark_node;
+
 	case RID_TEMPLATE:
 	  if (parser->in_function_body
 	      && (cp_lexer_peek_nth_token (parser->lexer, 2)->type
@@ -36735,6 +36743,15 @@ cp_parser_omp_var_list_no_open (cp_parser *parser, enum omp_clause_code kind,
 	  decl = cp_parser_primary_expression (parser, false, false, false,
 					       &idk);
 	}
+      else if (kind == OMP_CLAUSE_DEPEND
+	       && cp_parser_is_keyword (token, RID_OMP_ALL_MEMORY)
+	       && (cp_lexer_nth_token_is (parser->lexer, 2, CPP_COMMA)
+		   || cp_lexer_nth_token_is (parser->lexer, 2,
+					     CPP_CLOSE_PAREN)))
+	{
+	  decl = ridpointers[RID_OMP_ALL_MEMORY];
+	  cp_lexer_consume_token (parser->lexer);
+	}
       else
 	{
 	  name = cp_parser_id_expression (parser, /*template_p=*/false,
diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
index 2c7c5f8..06b4a7d 100644
--- a/gcc/cp/pt.cc
+++ b/gcc/cp/pt.cc
@@ -17601,8 +17601,8 @@ static tree
 tsubst_omp_clause_decl (tree decl, tree args, tsubst_flags_t complain,
 			tree in_decl, tree *iterator_cache)
 {
-  if (decl == NULL_TREE)
-    return NULL_TREE;
+  if (decl == NULL_TREE || decl == ridpointers[RID_OMP_ALL_MEMORY])
+    return decl;
 
   /* Handle OpenMP iterators.  */
   if (TREE_CODE (decl) == TREE_LIST
diff --git a/gcc/cp/semantics.cc b/gcc/cp/semantics.cc
index 10478d1..61f49be 100644
--- a/gcc/cp/semantics.cc
+++ b/gcc/cp/semantics.cc
@@ -7815,6 +7815,20 @@ finish_omp_clauses (tree clauses, enum c_omp_region_type ort)
 	    }
 	  if (t == error_mark_node)
 	    remove = true;
+	  else if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_DEPEND
+		   && t == ridpointers[RID_OMP_ALL_MEMORY])
+	    {
+	      if (OMP_CLAUSE_DEPEND_KIND (c) != OMP_CLAUSE_DEPEND_OUT
+		  && OMP_CLAUSE_DEPEND_KIND (c) != OMP_CLAUSE_DEPEND_INOUT)
+		{
+		  error_at (OMP_CLAUSE_LOCATION (c),
+			    "%<omp_all_memory%> used with %<depend%> kind "
+			    "other than %<out%> or %<inout%>");
+		  remove = true;
+		}
+	      if (processing_template_decl)
+		break;
+	    }
 	  else if (processing_template_decl && TREE_CODE (t) != OVERLOAD)
 	    break;
 	  else if (!lvalue_p (t))
@@ -7867,24 +7881,32 @@ finish_omp_clauses (tree clauses, enum c_omp_region_type ort)
 	    }
 	  if (!remove)
 	    {
-	      tree addr = cp_build_addr_expr (t, tf_warning_or_error);
-	      if (addr == error_mark_node)
-		remove = true;
+	      if (t == ridpointers[RID_OMP_ALL_MEMORY])
+		t = null_pointer_node;
 	      else
 		{
+		  tree addr = cp_build_addr_expr (t, tf_warning_or_error);
+		  if (addr == error_mark_node)
+		    {
+		      remove = true;
+		      break;
+		    }
 		  t = cp_build_indirect_ref (OMP_CLAUSE_LOCATION (c),
 					     addr, RO_UNARY_STAR,
 					     tf_warning_or_error);
 		  if (t == error_mark_node)
-		    remove = true;
-		  else if (TREE_CODE (OMP_CLAUSE_DECL (c)) == TREE_LIST
-			   && TREE_PURPOSE (OMP_CLAUSE_DECL (c))
-			   && (TREE_CODE (TREE_PURPOSE (OMP_CLAUSE_DECL (c)))
-			       == TREE_VEC))
-		    TREE_VALUE (OMP_CLAUSE_DECL (c)) = t;
-		  else
-		    OMP_CLAUSE_DECL (c) = t;
+		    {
+		      remove = true;
+		      break;
+		    }
 		}
+	      if (TREE_CODE (OMP_CLAUSE_DECL (c)) == TREE_LIST
+		  && TREE_PURPOSE (OMP_CLAUSE_DECL (c))
+		  && (TREE_CODE (TREE_PURPOSE (OMP_CLAUSE_DECL (c)))
+		      == TREE_VEC))
+		TREE_VALUE (OMP_CLAUSE_DECL (c)) = t;
+	      else
+		OMP_CLAUSE_DECL (c) = t;
 	    }
 	  break;
 	case OMP_CLAUSE_DETACH:
diff --git a/gcc/gimplify.cc b/gcc/gimplify.cc
index 822e0cf..13413d0 100644
--- a/gcc/gimplify.cc
+++ b/gcc/gimplify.cc
@@ -8623,7 +8623,8 @@ gimplify_omp_depend (tree *list_p, gimple_seq *pre_p)
 	      }
 	    if (error_operand_p (TREE_VALUE (t)))
 	      return 2;
-	    TREE_VALUE (t) = build_fold_addr_expr (TREE_VALUE (t));
+	    if (TREE_VALUE (t) != null_pointer_node)
+	      TREE_VALUE (t) = build_fold_addr_expr (TREE_VALUE (t));
 	    r = build4 (ARRAY_REF, ptr_type_node, array, cnts[i],
 			NULL_TREE, NULL_TREE);
 	    tem = build2_loc (OMP_CLAUSE_LOCATION (c), MODIFY_EXPR,
@@ -8650,7 +8651,8 @@ gimplify_omp_depend (tree *list_p, gimple_seq *pre_p)
 	      }
 	    if (error_operand_p (OMP_CLAUSE_DECL (c)))
 	      return 2;
-	    OMP_CLAUSE_DECL (c) = build_fold_addr_expr (OMP_CLAUSE_DECL (c));
+	    if (OMP_CLAUSE_DECL (c) != null_pointer_node)
+	      OMP_CLAUSE_DECL (c) = build_fold_addr_expr (OMP_CLAUSE_DECL (c));
 	    if (gimplify_expr (&OMP_CLAUSE_DECL (c), pre_p, NULL,
 			       is_gimple_val, fb_rvalue) == GS_ERROR)
 	      return 2;
@@ -10346,12 +10348,15 @@ gimplify_scan_omp_clauses (tree *list_p, gimple_seq *pre_p,
 	      remove = true;
 	      break;
 	    }
-	  OMP_CLAUSE_DECL (c) = build_fold_addr_expr (OMP_CLAUSE_DECL (c));
-	  if (gimplify_expr (&OMP_CLAUSE_DECL (c), pre_p, NULL,
-			     is_gimple_val, fb_rvalue) == GS_ERROR)
+	  if (OMP_CLAUSE_DECL (c) != null_pointer_node)
 	    {
-	      remove = true;
-	      break;
+	      OMP_CLAUSE_DECL (c) = build_fold_addr_expr (OMP_CLAUSE_DECL (c));
+	      if (gimplify_expr (&OMP_CLAUSE_DECL (c), pre_p, NULL,
+				 is_gimple_val, fb_rvalue) == GS_ERROR)
+		{
+		  remove = true;
+		  break;
+		}
 	    }
 	  if (code == OMP_TASK)
 	    ctx->has_depend = true;
diff --git a/gcc/testsuite/c-c++-common/gomp/all-memory-1.c b/gcc/testsuite/c-c++-common/gomp/all-memory-1.c
new file mode 100644
index 0000000..5d63e0d
--- /dev/null
+++ b/gcc/testsuite/c-c++-common/gomp/all-memory-1.c
@@ -0,0 +1,52 @@
+int omp_all_memory;			/* { dg-error "expected" } */
+
+void
+foo (void)
+{
+  int p = (&omp_all_memory)[0];		/* { dg-error "'omp_all_memory' may only be used in OpenMP 'depend' clause" } */
+}
+
+void
+bar (void)
+{
+  int *omp_all_memory;			/* { dg-error "expected" } */
+}
+
+void
+baz (void)
+{
+  struct omp_all_memory { int a; };	/* { dg-error "expected" } */
+}
+
+void
+qux (void)
+{
+  union omp_all_memory { int a; };	/* { dg-error "expected" } */
+}
+
+void
+corge (void)
+{
+  enum omp_all_memory { OAM; };		/* { dg-error "expected" } */
+}
+
+void
+garply (void)
+{
+  enum E { omp_all_memory }; }		/* { dg-error "expected" } */
+
+void
+boo (void)
+{
+  int x, y;
+  #pragma omp task private (omp_all_memory)			/* { dg-error "expected" } */
+  ;
+  #pragma omp task depend(inout: *&omp_all_memory)		/* { dg-error "'omp_all_memory' may only be used in OpenMP 'depend' clause" } */
+  ;
+  #pragma omp task depend(inout: omp_all_memory[0])		/* { dg-error "'omp_all_memory' may only be used in OpenMP 'depend' clause" } */
+  ;
+  #pragma omp task depend(in: omp_all_memory)			/* { dg-error "'omp_all_memory' used with 'depend' kind other than 'out' or 'inout'" } */
+  ;
+  #pragma omp task depend(mutexinoutset: omp_all_memory)	/* { dg-error "'omp_all_memory' used with 'depend' kind other than 'out' or 'inout'" } */
+  ;
+}
diff --git a/gcc/testsuite/c-c++-common/gomp/all-memory-2.c b/gcc/testsuite/c-c++-common/gomp/all-memory-2.c
new file mode 100644
index 0000000..6f5d31b
--- /dev/null
+++ b/gcc/testsuite/c-c++-common/gomp/all-memory-2.c
@@ -0,0 +1,55 @@
+/* { dg-options "-fno-openmp" } */
+
+int omp_all_memory;			/* { dg-bogus "expected" } */
+
+void
+foo (void)
+{
+  int p = (&omp_all_memory)[0];		/* { dg-bogus "'omp_all_memory' may only be used in OpenMP 'depend' clause" } */
+}
+
+void
+bar (void)
+{
+  int *omp_all_memory;			/* { dg-bogus "expected" } */
+}
+
+void
+baz (void)
+{
+  struct omp_all_memory { int a; };	/* { dg-bogus "expected" } */
+}
+
+void
+qux (void)
+{
+  union omp_all_memory { int a; };	/* { dg-bogus "expected" } */
+}
+
+void
+corge (void)
+{
+  enum omp_all_memory { OAM };		/* { dg-bogus "expected" } */
+}
+
+void
+garply (void)
+{
+  enum E { omp_all_memory };		/* { dg-bogus "expected" } */
+}
+
+void
+boo (void)
+{
+  int x, y;
+  #pragma omp task private (omp_all_memory)
+  ;
+  #pragma omp task depend(inout: *&omp_all_memory)
+  ;
+  #pragma omp task depend(inout: omp_all_memory[0])
+  ;
+  #pragma omp task depend(in: omp_all_memory)
+  ;
+  #pragma omp task depend(mutexinoutset: omp_all_memory)
+  ;
+}
diff --git a/gcc/testsuite/c-c++-common/gomp/all-memory-3.c b/gcc/testsuite/c-c++-common/gomp/all-memory-3.c
new file mode 100644
index 0000000..f178b8d
--- /dev/null
+++ b/gcc/testsuite/c-c++-common/gomp/all-memory-3.c
@@ -0,0 +1,22 @@
+typedef struct __attribute__((__aligned__ (sizeof (void *)))) omp_depend_t {
+  char __omp_depend_t__[2 * sizeof (void *)];
+} omp_depend_t;
+
+omp_depend_t z;
+
+void
+foo (void)
+{
+  int x = 0, y = 0;
+  #pragma omp task depend(out: omp_all_memory)
+  ;
+  #pragma omp task depend(inout: omp_all_memory)
+  ;
+  #pragma omp task depend(out: x, omp_all_memory, y)
+  ;
+  #pragma omp task depend(inout: omp_all_memory, y)
+  ;
+  #pragma omp task depend(out: x, omp_all_memory)
+  ;
+  #pragma omp depobj (z) depend (inout: omp_all_memory)
+}
diff --git a/gcc/testsuite/g++.dg/gomp/all-memory-1.C b/gcc/testsuite/g++.dg/gomp/all-memory-1.C
new file mode 100644
index 0000000..8f3358d
--- /dev/null
+++ b/gcc/testsuite/g++.dg/gomp/all-memory-1.C
@@ -0,0 +1,24 @@
+namespace A
+{
+  namespace omp_all_memory		// { dg-error "expected" }
+  {
+  }
+}
+
+namespace B
+{
+  template <int N>
+  void omp_all_memory () {}		// { dg-error "expected" }
+}
+
+namespace C
+{
+  template <int N>
+  struct omp_all_memory {};		// { dg-error "expected" }
+}
+
+namespace D
+{
+  template <int omp_all_memory>		// { dg-error "expected" }
+  struct S {};
+}
diff --git a/gcc/testsuite/g++.dg/gomp/all-memory-2.C b/gcc/testsuite/g++.dg/gomp/all-memory-2.C
new file mode 100644
index 0000000..1acf391
--- /dev/null
+++ b/gcc/testsuite/g++.dg/gomp/all-memory-2.C
@@ -0,0 +1,26 @@
+// { dg-options "-fno-openmp" }
+
+namespace A
+{
+  namespace omp_all_memory		// { dg-bogus "expected" }
+  {
+  }
+}
+
+namespace B
+{
+  template <int N>
+  void omp_all_memory () {}		// { dg-bogus "expected" }
+}
+
+namespace C
+{
+  template <int N>
+  struct omp_all_memory {};		// { dg-bogus "expected" }
+}
+
+namespace D
+{
+  template <int omp_all_memory>		// { dg-bogus "expected" }
+  struct S {};
+}
diff --git a/gcc/tree-pretty-print.cc b/gcc/tree-pretty-print.cc
index 99af977..d7615aa 100644
--- a/gcc/tree-pretty-print.cc
+++ b/gcc/tree-pretty-print.cc
@@ -850,7 +850,10 @@ dump_omp_clause (pretty_printer *pp, tree clause, int spc, dump_flags_t flags)
 	    pp_string (pp, name);
 	    pp_colon (pp);
 	  }
-	dump_generic_node (pp, t, spc, flags, false);
+	if (t == null_pointer_node)
+	  pp_string (pp, "omp_all_memory");
+	else
+	  dump_generic_node (pp, t, spc, flags, false);
 	pp_right_paren (pp);
       }
       break;
diff --git a/libgomp/libgomp.h b/libgomp/libgomp.h
index b9e0391..295d10f 100644
--- a/libgomp/libgomp.h
+++ b/libgomp/libgomp.h
@@ -574,6 +574,8 @@ struct gomp_task
   struct gomp_dependers_vec *dependers;
   struct htab *depend_hash;
   struct gomp_taskwait *taskwait;
+  /* Last depend({,in}out:omp_all_memory) child if any.  */
+  struct gomp_task *depend_all_memory;
   /* Number of items in DEPEND.  */
   size_t depend_count;
   /* Number of tasks this task depends on.  Once this counter reaches
diff --git a/libgomp/task.c b/libgomp/task.c
index 828348c..db4a6f7 100644
--- a/libgomp/task.c
+++ b/libgomp/task.c
@@ -80,6 +80,7 @@ gomp_init_task (struct gomp_task *task, struct gomp_task *parent_task,
   task->dependers = NULL;
   task->depend_hash = NULL;
   task->taskwait = NULL;
+  task->depend_all_memory = NULL;
   task->depend_count = 0;
   task->completion_sem = NULL;
   task->deferred_p = false;
@@ -171,6 +172,7 @@ gomp_task_handle_depend (struct gomp_task *task, struct gomp_task *parent,
   size_t ndepend = (uintptr_t) depend[0];
   size_t i;
   hash_entry_type ent;
+  bool all_memory = false;
 
   if (ndepend)
     {
@@ -181,6 +183,7 @@ gomp_task_handle_depend (struct gomp_task *task, struct gomp_task *parent,
 	{
 	  task->depend[i].addr = depend[2 + i];
 	  task->depend[i].is_in = i >= nout;
+	  all_memory |= i < nout && depend[2 + i] == NULL;
 	}
     }
   else
@@ -201,6 +204,8 @@ gomp_task_handle_depend (struct gomp_task *task, struct gomp_task *parent,
 	    {
 	    case GOMP_DEPEND_OUT:
 	    case GOMP_DEPEND_INOUT:
+	      all_memory |= d[0] == NULL;
+	      break;
 	    case GOMP_DEPEND_MUTEXINOUTSET:
 	      break;
 	    case GOMP_DEPEND_IN:
@@ -226,8 +231,126 @@ gomp_task_handle_depend (struct gomp_task *task, struct gomp_task *parent,
 	  task->depend[n++].is_in = 1;
 	}
     }
-  task->depend_count = ndepend;
   task->num_dependees = 0;
+  if (__builtin_expect (parent->depend_all_memory && ndepend, false))
+    {
+      struct gomp_task *tsk = parent->depend_all_memory;
+      if (tsk->dependers == NULL)
+	{
+	  tsk->dependers
+	    = gomp_malloc (sizeof (struct gomp_dependers_vec)
+			   + 6 * sizeof (struct gomp_task *));
+	  tsk->dependers->n_elem = 1;
+	  tsk->dependers->allocated = 6;
+	  tsk->dependers->elem[0] = task;
+	}
+      else
+	{
+	  if (tsk->dependers->n_elem == tsk->dependers->allocated)
+	    {
+	      tsk->dependers->allocated
+		= tsk->dependers->allocated * 2 + 2;
+	      tsk->dependers
+		= gomp_realloc (tsk->dependers,
+				sizeof (struct gomp_dependers_vec)
+				+ (tsk->dependers->allocated
+				   * sizeof (struct gomp_task *)));
+	    }
+	  tsk->dependers->elem[tsk->dependers->n_elem++] = task;
+	}
+      task->num_dependees++;
+    }
+  if (__builtin_expect (all_memory, false))
+    {
+      /* A task with depend(inout: omp_all_memory) depends on all previous
+	 sibling tasks which have any dependencies and all later sibling
+	 tasks which have any dependencies depend on it.  */
+      task->depend_count = 1;
+      task->depend[0].addr = NULL;
+      task->depend[0].next = NULL;
+      task->depend[0].prev = NULL;
+      task->depend[0].task = task;
+      task->depend[0].redundant = true;
+      task->depend[0].redundant_out = false;
+      if (parent->depend_hash)
+	{
+	  /* Inlined htab_traverse + htab_clear.  All newer siblings can
+	     just depend on this task.  Add dependencies on all previous
+	     sibling tasks with dependencies and make them redundant and
+	     clear the hash table.  */
+	  hash_entry_type *slot = &parent->depend_hash->entries[0];
+	  hash_entry_type *end = slot + htab_size (parent->depend_hash);
+	  for (; slot != end; ++slot)
+	    {
+	      if (*slot == HTAB_EMPTY_ENTRY)
+		continue;
+	      if (*slot != HTAB_DELETED_ENTRY)
+		{
+		  for (ent = *slot; ent; ent = ent->next)
+		    {
+		      struct gomp_task *tsk = ent->task;
+
+		      if (ent->redundant_out)
+			break;
+
+		      ent->redundant = true;
+		      if (tsk->dependers == NULL)
+			{
+			  tsk->dependers
+			    = gomp_malloc (sizeof (struct gomp_dependers_vec)
+					   + 6 * sizeof (struct gomp_task *));
+			  tsk->dependers->n_elem = 1;
+			  tsk->dependers->allocated = 6;
+			  tsk->dependers->elem[0] = task;
+			  task->num_dependees++;
+			  continue;
+			}
+		      /* We already have some other dependency on tsk from
+			 earlier depend clause.  */
+		      else if (tsk->dependers->n_elem
+			       && (tsk->dependers->elem[tsk->dependers->n_elem
+							- 1] == task))
+			continue;
+		      else if (tsk->dependers->n_elem
+			       == tsk->dependers->allocated)
+			{
+			  tsk->dependers->allocated
+			    = tsk->dependers->allocated * 2 + 2;
+			  tsk->dependers
+			    = gomp_realloc (tsk->dependers,
+					    sizeof (struct gomp_dependers_vec)
+					    + (tsk->dependers->allocated
+					       * sizeof (struct gomp_task *)));
+			}
+		      tsk->dependers->elem[tsk->dependers->n_elem++] = task;
+		      task->num_dependees++;
+		    }
+		  while (ent)
+		    {
+		      ent->redundant = true;
+		      ent = ent->next;
+		    }
+		}
+	      *slot = HTAB_EMPTY_ENTRY;
+	    }
+	  if (htab_size (parent->depend_hash) <= 32)
+	    {
+	      parent->depend_hash->n_elements = 0;
+	      parent->depend_hash->n_deleted = 0;
+	    }
+	  else
+	    {
+	      /* Shrink the hash table if it would be too large.
+		 We don't want to walk e.g. megabytes of empty hash
+		 table for every depend(inout: omp_all_memory).  */
+	      free (parent->depend_hash);
+	      parent->depend_hash = htab_create (12);
+	    }
+	}
+      parent->depend_all_memory = task;
+      return;
+    }
+  task->depend_count = ndepend;
   if (parent->depend_hash == NULL)
     parent->depend_hash = htab_create (2 * ndepend > 12 ? 2 * ndepend : 12);
   for (i = 0; i < ndepend; i++)
@@ -1175,6 +1298,8 @@ gomp_task_run_post_handle_depend_hash (struct gomp_task *child_task)
   struct gomp_task *parent = child_task->parent;
   size_t i;
 
+  if (parent->depend_all_memory == child_task)
+    parent->depend_all_memory = NULL;
   for (i = 0; i < child_task->depend_count; i++)
     if (!child_task->depend[i].redundant)
       {
@@ -1738,6 +1863,17 @@ gomp_task_maybe_wait_for_dependencies (void **depend)
       n = 5;
     }
   gomp_mutex_lock (&team->task_lock);
+  if (__builtin_expect (task->depend_all_memory && ndepend, false))
+    {
+      struct gomp_task *tsk = task->depend_all_memory;
+      if (!tsk->parent_depends_on)
+	{
+	  tsk->parent_depends_on = true;
+	  ++num_awaited;
+	  if (tsk->num_dependees == 0 && tsk->kind == GOMP_TASK_WAITING)
+	    priority_queue_upgrade_task (tsk, task);
+	}
+    }
   for (i = 0; i < ndepend; i++)
     {
       elem.addr = depend[i + n];
@@ -1760,6 +1896,36 @@ gomp_task_maybe_wait_for_dependencies (void **depend)
 	    }
 	  elem.addr = d[0];
 	}
+      if (__builtin_expect (elem.addr == NULL && !elem.is_in, false))
+	{
+	  size_t size = htab_size (task->depend_hash);
+	  if (htab_elements (task->depend_hash) * 8 < size && size > 32)
+	    htab_expand (task->depend_hash);
+
+	  /* depend(inout: omp_all_memory) - depend on all previous
+	     sibling tasks that do have dependencies.  Inlined
+	     htab_traverse.  */
+	  hash_entry_type *slot = &task->depend_hash->entries[0];
+	  hash_entry_type *end = slot + htab_size (task->depend_hash);
+	  for (; slot != end; ++slot)
+	    {
+	      if (*slot == HTAB_EMPTY_ENTRY || *slot == HTAB_DELETED_ENTRY)
+		continue;
+	      for (ent = *slot; ent; ent = ent->next)
+		{
+		  struct gomp_task *tsk = ent->task;
+		  if (!tsk->parent_depends_on)
+		    {
+		      tsk->parent_depends_on = true;
+		      ++num_awaited;
+		      if (tsk->num_dependees == 0
+			  && tsk->kind == GOMP_TASK_WAITING)
+			priority_queue_upgrade_task (tsk, task);
+		    }
+		}
+	    }
+	  break;
+	}
       ent = htab_find (task->depend_hash, &elem);
       for (; ent; ent = ent->next)
 	if (elem.is_in && ent->is_in)
diff --git a/libgomp/testsuite/libgomp.c-c++-common/depend-1.c b/libgomp/testsuite/libgomp.c-c++-common/depend-1.c
new file mode 100644
index 0000000..3376b99
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c-c++-common/depend-1.c
@@ -0,0 +1,110 @@
+#include <stdlib.h>
+#include <unistd.h>
+
+void
+test (int ifval)
+{
+  int a[8], b[8], i;
+  for (i = 0; i < 8; i++)
+    {
+      a[i] = i;
+      b[i] = 2 * i;
+    }
+  #pragma omp parallel
+  #pragma omp single
+  {
+    #pragma omp task shared(a) depend(in: a[0])
+    {
+      usleep (5000);
+      a[0] = 42;
+    }
+    #pragma omp task shared(a) depend(out: a[1])
+    {
+      usleep (5000);
+      a[1] = 43;
+    }
+    #pragma omp task shared(a) depend(inout: a[2])
+    {
+      usleep (5000);
+      a[2] = 44;
+    }
+    #pragma omp task shared(a) depend(mutexinoutset: a[3])
+    {
+      usleep (5000);
+      a[3] = 45;
+    }
+    #pragma omp task shared(a)
+    {
+      usleep (15000);
+      a[4] = 46;
+    }
+    #pragma omp task shared(b) depend(in: b[0])
+    {
+      usleep (5000);
+      b[0] = 47;
+    }
+    #pragma omp task shared(b) depend(in: b[4])
+    {
+      usleep (5000);
+      b[4] = 48;
+    }
+    /* None of the above tasks depend on each other.
+       The following task depends on all but the a[4] = 46; one.  */
+    #pragma omp task shared(a, b) depend(out: omp_all_memory) private(i) if(ifval)
+    {
+      if (a[0] != 42 || a[1] != 43 || a[2] != 44 || a[3] != 45
+	  || a[5] != 5 || a[6] != 6 || a[7] != 7
+	  || b[0] != 47 || b[1] != 2 || b[2] != 4 || b[3] != 6
+	  || b[4] != 48 || b[5] != 10 || b[6] != 12 || b[7] != 14)
+	abort ();
+      for (i = 0; i < 8; ++i)
+	if (i != 4)
+	  a[i] = 3 * i + 7;
+      for (i = 0; i < 8; ++i)
+	b[i] = 4 * i - 7;
+    }
+    /* The following task depends on both b[0] = 47; and
+       above omp_all_memory tasks, but as the latter depends on
+       the former, effectively it is dependent just on the omp_all_memory
+       task.  */
+    #pragma omp task shared(b) depend(inout: b[0])
+    {
+      usleep (5000);
+      b[0] = 49;
+    }
+    /* The following task depends on all the above except a[4] = 46; one,
+       but it can be reduced to dependency on the above omp_all_memory
+       one and b[0] = 49; one.  */
+    #pragma omp task shared(a, b) depend(inout: b[7], omp_all_memory, b[6]) \
+		     private(i) if(ifval)
+    {
+      for (i = 0; i < 8; ++i)
+	if (i != 4)
+	  {
+	    if (a[i] != 3 * i + 7)
+	      abort ();
+	    a[i] = 5 * i + 50;
+	  }
+      if (b[0] != 49)
+	abort ();
+      b[0] = 6 * i + 57;
+      for (i = 1; i < 8; ++i)
+	{
+	  if (b[i] != 4 * i - 7) 
+	    abort ();
+	  b[i] = 6 * i + 57;
+	}
+    }
+    #pragma omp taskwait
+    if (a[4] != 46)
+      abort ();
+  }
+}
+
+int
+main ()
+{
+  test (1);
+  test (0);
+  return 0;
+}
diff --git a/libgomp/testsuite/libgomp.c-c++-common/depend-2.c b/libgomp/testsuite/libgomp.c-c++-common/depend-2.c
new file mode 100644
index 0000000..d7b5335
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c-c++-common/depend-2.c
@@ -0,0 +1,116 @@
+#include <omp.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+void
+test (int ifval)
+{
+  int a[8], b[8], i;
+  omp_depend_t d1, d2;
+  #pragma omp depobj (d1) depend(inout: omp_all_memory) 
+  #pragma omp depobj (d2) depend(out: omp_all_memory)
+  for (i = 0; i < 8; i++)
+    {
+      a[i] = i;
+      b[i] = 2 * i;
+    }
+  #pragma omp parallel
+  #pragma omp single
+  {
+    #pragma omp task shared(a) depend(in: a[0])
+    {
+      usleep (5000);
+      a[0] = 42;
+    }
+    #pragma omp task shared(a) depend(out: a[1])
+    {
+      usleep (5000);
+      a[1] = 43;
+    }
+    #pragma omp task shared(a) depend(inout: a[2])
+    {
+      usleep (5000);
+      a[2] = 44;
+    }
+    #pragma omp task shared(a) depend(mutexinoutset: a[3])
+    {
+      usleep (5000);
+      a[3] = 45;
+    }
+    #pragma omp task shared(a)
+    {
+      usleep (15000);
+      a[4] = 46;
+    }
+    #pragma omp task shared(b) depend(in: b[0])
+    {
+      usleep (5000);
+      b[0] = 47;
+    }
+    #pragma omp task shared(b) depend(in: b[4])
+    {
+      usleep (5000);
+      b[4] = 48;
+    }
+    /* None of the above tasks depend on each other.
+       The following task depends on all but the a[4] = 46; one.  */
+    #pragma omp task shared(a, b) depend(depobj: d1) private(i) if(ifval)
+    {
+      if (a[0] != 42 || a[1] != 43 || a[2] != 44 || a[3] != 45
+	  || a[5] != 5 || a[6] != 6 || a[7] != 7
+	  || b[0] != 47 || b[1] != 2 || b[2] != 4 || b[3] != 6
+	  || b[4] != 48 || b[5] != 10 || b[6] != 12 || b[7] != 14)
+	abort ();
+      for (i = 0; i < 8; ++i)
+	if (i != 4)
+	  a[i] = 3 * i + 7;
+      for (i = 0; i < 8; ++i)
+	b[i] = 4 * i - 7;
+    }
+    /* The following task depends on both b[0] = 47; and
+       above omp_all_memory tasks, but as the latter depends on
+       the former, effectively it is dependent just on the omp_all_memory
+       task.  */
+    #pragma omp task shared(b) depend(inout: b[0])
+    {
+      usleep (5000);
+      b[0] = 49;
+    }
+    /* The following task depends on all the above except a[4] = 46; one,
+       but it can be reduced to dependency on the above omp_all_memory
+       one and b[0] = 49; one.  */
+    #pragma omp task shared(a, b) depend(inout: b[6]) depend(depobj: d2) \
+		     depend(out: b[7]) private(i) if(ifval)
+    {
+      for (i = 0; i < 8; ++i)
+	if (i != 4)
+	  {
+	    if (a[i] != 3 * i + 7)
+	      abort ();
+	    a[i] = 5 * i + 50;
+	  }
+      if (b[0] != 49)
+	abort ();
+      b[0] = 6 * i + 57;
+      for (i = 1; i < 8; ++i)
+	{
+	  if (b[i] != 4 * i - 7) 
+	    abort ();
+	  b[i] = 6 * i + 57;
+	}
+    }
+    #pragma omp taskwait
+    if (a[4] != 46)
+      abort ();
+  }
+  #pragma omp depobj (d2) destroy
+  #pragma omp depobj (d1) destroy
+}
+
+int
+main ()
+{
+  test (1);
+  test (0);
+  return 0;
+}
diff --git a/libgomp/testsuite/libgomp.c-c++-common/depend-3.c b/libgomp/testsuite/libgomp.c-c++-common/depend-3.c
new file mode 100644
index 0000000..052e77c
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c-c++-common/depend-3.c
@@ -0,0 +1,103 @@
+#include <stdlib.h>
+#include <unistd.h>
+
+int
+main ()
+{
+  int a[8], b[8], i;
+  for (i = 0; i < 8; i++)
+    {
+      a[i] = i;
+      b[i] = 2 * i;
+    }
+  #pragma omp parallel
+  #pragma omp single
+  {
+    #pragma omp task shared(a) depend(in: a[0])
+    {
+      usleep (5000);
+      a[0] = 42;
+    }
+    #pragma omp task shared(a) depend(out: a[1])
+    {
+      usleep (5000);
+      a[1] = 43;
+    }
+    #pragma omp task shared(a) depend(inout: a[2])
+    {
+      usleep (5000);
+      a[2] = 44;
+    }
+    #pragma omp task shared(a) depend(mutexinoutset: a[3])
+    {
+      usleep (5000);
+      a[3] = 45;
+    }
+    #pragma omp task shared(a)
+    {
+      usleep (15000);
+      a[4] = 46;
+    }
+    #pragma omp task shared(b) depend(in: b[0])
+    {
+      usleep (5000);
+      b[0] = 47;
+    }
+    #pragma omp task shared(b) depend(in: b[4])
+    {
+      usleep (5000);
+      b[4] = 48;
+    }
+    /* None of the above tasks depend on each other.
+       The following task depends on all but the a[4] = 46; one.  */
+    #pragma omp task shared(a, b) depend(iterator (j=0:7), inout: omp_all_memory) private(i)
+    {
+      if (a[0] != 42 || a[1] != 43 || a[2] != 44 || a[3] != 45
+	  || a[5] != 5 || a[6] != 6 || a[7] != 7
+	  || b[0] != 47 || b[1] != 2 || b[2] != 4 || b[3] != 6
+	  || b[4] != 48 || b[5] != 10 || b[6] != 12 || b[7] != 14)
+	abort ();
+      for (i = 0; i < 8; ++i)
+	if (i != 4)
+	  a[i] = 3 * i + 7;
+      for (i = 0; i < 8; ++i)
+	b[i] = 4 * i - 7;
+    }
+    /* The following task depends on both b[0] = 47; and
+       above omp_all_memory tasks, but as the latter depends on
+       the former, effectively it is dependent just on the omp_all_memory
+       task.  */
+    #pragma omp task shared(b) depend(inout: b[0])
+    {
+      usleep (5000);
+      b[0] = 49;
+    }
+    /* The following task depends on all the above except a[4] = 46; one,
+       but it can be reduced to dependency on the above omp_all_memory
+       one and b[0] = 49; one.  */
+    #pragma omp task shared(a, b) depend(inout: b[7]) depend(iterator(j=4:5), out: omp_all_memory) \
+		     depend(inout: b[6]) private(i)
+    {
+      for (i = 0; i < 8; ++i)
+	if (i != 4)
+	  {
+	    if (a[i] != 3 * i + 7)
+	      abort ();
+	    a[i] = 5 * i + 50;
+	  }
+      if (b[0] != 49)
+	abort ();
+      b[0] = 6 * i + 57;
+      for (i = 1; i < 8; ++i)
+	{
+	  if (b[i] != 4 * i - 7) 
+	    abort ();
+	  b[i] = 6 * i + 57;
+	}
+    }
+    #pragma omp taskwait
+    if (a[4] != 46)
+      abort ();
+  }
+  return 0;
+}