Further simplify 'gcc/omp-oacc-neuter-broadcast.cc:record_field_map_t'
authorThomas Schwinge <thomas@codesourcery.com>
Fri, 13 Aug 2021 19:17:55 +0000 (21:17 +0200)
committerThomas Schwinge <thomas@codesourcery.com>
Tue, 22 Feb 2022 16:43:39 +0000 (17:43 +0100)
Now that I've resolved GCC 'hash_map' issues (a while ago already), we may
further simplify this after commit 049eda8274b7394523238b17ab12c3e2889f253e
"Avoid 'GTY' use for 'gcc/omp-oacc-neuter-broadcast.cc:field_map'": as
'hash_map' Value, directly store 'field_map_t' objects, not pointers to
manually allocated 'field_map_t' objects.

gcc/
* omp-oacc-neuter-broadcast.cc (record_field_map_t): Further
simplify.  Adjust all users.

gcc/omp-oacc-neuter-broadcast.cc

index 7fb691d..314161e 100644 (file)
@@ -538,7 +538,7 @@ typedef hash_map<tree, tree> field_map_t;
    to propagate, to the field in the record type that should be used for
    transmission and reception.  */
 
-typedef hash_map<tree, field_map_t *> record_field_map_t;
+typedef hash_map<tree, field_map_t> record_field_map_t;
 
 static void
 install_var_field (tree var, tree record_type, field_map_t *fields)
@@ -1168,8 +1168,7 @@ worker_single_copy (basic_block from, basic_block to,
        gcc_assert (TREE_CODE (var) == VAR_DECL);
 
       /* If we had no record type, we will have no fields map.  */
-      field_map_t **fields_p = record_field_map->get (record_type);
-      field_map_t *fields = fields_p ? *fields_p : NULL;
+      field_map_t *fields = record_field_map->get (record_type);
 
       if (worker_partitioned_uses->contains (var)
          && fields
@@ -1684,10 +1683,9 @@ oacc_do_neutering (unsigned HOST_WIDE_INT bounds_lo,
 
          field_vec.qsort (sort_by_size_then_ssa_version_or_uid);
 
-         field_map_t *fields = new field_map_t;
-
          bool existed;
-         existed = record_field_map.put (record_type, fields);
+         field_map_t *fields
+           = &record_field_map.get_or_insert (record_type, &existed);
          gcc_checking_assert (!existed);
 
          /* Insert var fields in reverse order, so the last inserted element
@@ -1818,8 +1816,6 @@ oacc_do_neutering (unsigned HOST_WIDE_INT bounds_lo,
                        &partitioned_var_uses, &record_field_map,
                        &blk_offset_map, writes_gang_private);
 
-  for (auto it : record_field_map)
-    delete it.second;
   record_field_map.empty ();
 
   /* These are supposed to have been 'delete'd by 'neuter_worker_single'.  */