Hi!

As the following testcase shows, we need to consider even target to be a 
construct
that forces not to use copy in/out for shared on parallel inside of the target.
E.g. for parallel nested inside another parallel or host teams, we already avoid
copy in/out and we need to treat target the same.

Bootstrapped/regtested on x86_64-linux and i686-linux, committed to trunk.

2020-02-06  Jakub Jelinek  <ja...@redhat.com>

        PR libgomp/93515
        * omp-low.c (use_pointer_for_field): For nested constructs, also
        look for map clauses on target construct.
        (scan_omp_1_stmt) <case GIMPLE_OMP_TARGET>: Bump temporarily
        taskreg_nesting_level.

        * testsuite/libgomp.c-c++-common/pr93515.c: New test.

--- gcc/omp-low.c.jj    2020-01-12 11:54:36.688409260 +0100
+++ gcc/omp-low.c       2020-01-31 15:00:46.852168424 +0100
@@ -477,18 +477,30 @@ use_pointer_for_field (tree decl, omp_co
          omp_context *up;
 
          for (up = shared_ctx->outer; up; up = up->outer)
-           if (is_taskreg_ctx (up) && maybe_lookup_decl (decl, up))
+           if ((is_taskreg_ctx (up)
+                || (gimple_code (up->stmt) == GIMPLE_OMP_TARGET
+                    && is_gimple_omp_offloaded (up->stmt)))
+               && maybe_lookup_decl (decl, up))
              break;
 
          if (up)
            {
              tree c;
 
-             for (c = gimple_omp_taskreg_clauses (up->stmt);
-                  c; c = OMP_CLAUSE_CHAIN (c))
-               if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_SHARED
-                   && OMP_CLAUSE_DECL (c) == decl)
-                 break;
+             if (gimple_code (up->stmt) == GIMPLE_OMP_TARGET)
+               {
+                 for (c = gimple_omp_target_clauses (up->stmt);
+                      c; c = OMP_CLAUSE_CHAIN (c))
+                   if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_MAP
+                       && OMP_CLAUSE_DECL (c) == decl)
+                     break;
+               }
+             else
+               for (c = gimple_omp_taskreg_clauses (up->stmt);
+                    c; c = OMP_CLAUSE_CHAIN (c))
+                 if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_SHARED
+                     && OMP_CLAUSE_DECL (c) == decl)
+                   break;
 
              if (c)
                goto maybe_mark_addressable_and_ret;
@@ -3781,7 +3793,14 @@ scan_omp_1_stmt (gimple_stmt_iterator *g
       break;
 
     case GIMPLE_OMP_TARGET:
-      scan_omp_target (as_a <gomp_target *> (stmt), ctx);
+      if (is_gimple_omp_offloaded (stmt))
+       {
+         taskreg_nesting_level++;
+         scan_omp_target (as_a <gomp_target *> (stmt), ctx);
+         taskreg_nesting_level--;
+       }
+      else
+       scan_omp_target (as_a <gomp_target *> (stmt), ctx);
       break;
 
     case GIMPLE_OMP_TEAMS:
--- libgomp/testsuite/libgomp.c-c++-common/pr93515.c.jj 2020-01-31 
14:53:01.163112148 +0100
+++ libgomp/testsuite/libgomp.c-c++-common/pr93515.c    2020-01-31 
14:52:38.627448474 +0100
@@ -0,0 +1,36 @@
+/* PR libgomp/93515 */
+
+#include <omp.h>
+#include <stdlib.h>
+
+int
+main ()
+{
+  int i;
+  int a = 42;
+#pragma omp target teams distribute parallel for defaultmap(tofrom: scalar)
+  for (i = 0; i < 64; ++i)
+    if (omp_get_team_num () == 0)
+      if (omp_get_thread_num () == 0)
+       a = 142;
+  if (a != 142)
+    __builtin_abort ();
+  a = 42;
+#pragma omp target parallel for defaultmap(tofrom: scalar)
+  for (i = 0; i < 64; ++i)
+    if (omp_get_thread_num () == 0)
+      a = 143;
+  if (a != 143)
+    __builtin_abort ();
+  a = 42;
+#pragma omp target firstprivate(a)
+  {
+    #pragma omp parallel for
+    for (i = 0; i < 64; ++i)
+      if (omp_get_thread_num () == 0)
+       a = 144;
+    if (a != 144)
+      abort ();
+  }
+  return 0;
+}


        Jakub

Reply via email to