OpenE2K/gcc
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Make the PR99220 fix more robust

Browse Source 
This avoids interleaving pattern recognition and load redistribution
optimization since the load_map used in the latter is fragile with
respect to release and reuse of SLP nodes, something which can also
occur within the pattern recognition machinery.

2021-02-25  Richard Biener  <rguenther@suse.de>

	* tree-vect-slp.c (optimize_load_redistribution_1): Delay
	load_map population.
	(vect_match_slp_patterns_2): Revert part of last change.
	(vect_analyze_slp): Do not interleave optimize_load_redistribution
	with pattern detection but do it afterwards.  Dump the
	whole SLP graph after pattern recognition and load
	redistribution optimization finished.
This commit is contained in:
Richard Biener 2021-02-25 08:37:36 +01:00
parent a6baafcac5
commit 5e7c0927cb
1 changed files with 30 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

53
gcc/tree-vect-slp.c
View File

@ -2294,8 +2294,6 @@ optimize_load_redistribution_1 (scalar_stmts_to_slp_tree_map_t *bst_map,
if (slp_tree *leader = load_map->get (root))
return *leader;
load_map->put (root, NULL);
slp_tree node;
unsigned i;
@ -2342,6 +2340,8 @@ optimize_load_redistribution_1 (scalar_stmts_to_slp_tree_map_t *bst_map,
}
next:
load_map->put (root, NULL);
FOR_EACH_VEC_ELT (SLP_TREE_CHILDREN (root), i , node)
{
slp_tree value
@ -2352,11 +2352,10 @@ next:
SLP_TREE_REF_COUNT (value)++;
SLP_TREE_CHILDREN (root)[i] = value;
/* ??? We know the original leafs of the replaced nodes will
be referenced by bst_map, only the permutes created by
pattern matching are not. */
be referenced by bst_map, only the permutes created by
pattern matching are not. */
if (SLP_TREE_REF_COUNT (node) == 1)
load_map->remove (node);
vect_free_slp_tree (node);
}
}
@ -2394,7 +2393,6 @@ optimize_load_redistribution (scalar_stmts_to_slp_tree_map_t *bst_map,
pattern matching are not. */
if (SLP_TREE_REF_COUNT (node) == 1)
load_map->remove (node);
vect_free_slp_tree (node);
}
}
@ -2417,9 +2415,7 @@ vect_match_slp_patterns_2 (slp_tree *ref_node, vec_info *vinfo,
slp_tree node = *ref_node;
bool found_p = false;
if (!node || visited->add (node))
return node
&& SLP_TREE_REPRESENTATIVE (node)
&& STMT_VINFO_SLP_VECT_ONLY_PATTERN (SLP_TREE_REPRESENTATIVE (node));
return false;
slp_tree child;
FOR_EACH_VEC_ELT (SLP_TREE_CHILDREN (node), i, child)
@ -2898,23 +2894,24 @@ vect_analyze_slp (vec_info *vinfo, unsigned max_tree_size)
hash_set<slp_tree> visited_patterns;
slp_tree_to_load_perm_map_t perm_cache;
hash_map<slp_tree, slp_tree> load_map;
/* See if any patterns can be found in the SLP tree. */
bool pattern_found = false;
FOR_EACH_VEC_ELT (LOOP_VINFO_SLP_INSTANCES (vinfo), i, instance)
if (vect_match_slp_patterns (instance, vinfo, &visited_patterns,
&perm_cache))
{
slp_tree root = SLP_INSTANCE_TREE (instance);
optimize_load_redistribution (bst_map, vinfo, SLP_TREE_LANES (root),
&load_map, root);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"Pattern matched SLP tree\n");
vect_print_slp_graph (MSG_NOTE, vect_location, root);
}
}
pattern_found |= vect_match_slp_patterns (instance, vinfo,
&visited_patterns, &perm_cache);
/* If any were found optimize permutations of loads. */
if (pattern_found)
{
hash_map<slp_tree, slp_tree> load_map;
FOR_EACH_VEC_ELT (LOOP_VINFO_SLP_INSTANCES (vinfo), i, instance)
{
slp_tree root = SLP_INSTANCE_TREE (instance);
optimize_load_redistribution (bst_map, vinfo, SLP_TREE_LANES (root),
&load_map, root);
}
}
@ -2925,6 +2922,16 @@ vect_analyze_slp (vec_info *vinfo, unsigned max_tree_size)
vect_free_slp_tree ((*it).second);
delete bst_map;
if (pattern_found && dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"Pattern matched SLP tree\n");
hash_set<slp_tree> visited;
FOR_EACH_VEC_ELT (LOOP_VINFO_SLP_INSTANCES (vinfo), i, instance)
vect_print_slp_graph (MSG_NOTE, vect_location,
SLP_INSTANCE_TREE (instance), visited);
}
return opt_result::success ();
}