public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r11-3820] fix SLP subgraph detection wrt fully shared lanes
@ 2020-10-12 12:31 Richard Biener
0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2020-10-12 12:31 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:a7825bb6556da6fa991b7ea1b72e3851026dc2cd
commit r11-3820-ga7825bb6556da6fa991b7ea1b72e3851026dc2cd
Author: Richard Biener <rguenther@suse.de>
Date: Mon Oct 12 14:29:35 2020 +0200
fix SLP subgraph detection wrt fully shared lanes
When a VEC_PERM SLP node just permutes existing lanes this confuses
the SLP subgraph detection where I tried to elide a node-based
visited hash-map in a way that doesn't work. Fixed by adding such.
2020-10-12 Richard Biener <rguenther@suse.de>
* tree-vect-slp.c (vect_bb_partition_graph_r): Use visited
hash-map.
(vect_bb_partition_graph): Likewise.
Diff:
---
gcc/tree-vect-slp.c | 17 ++++++++++-------
1 file changed, 10 insertions(+), 7 deletions(-)
diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c
index dd2042a4db5..8acef6f3cef 100644
--- a/gcc/tree-vect-slp.c
+++ b/gcc/tree-vect-slp.c
@@ -3254,18 +3254,19 @@ static void
vect_bb_partition_graph_r (bb_vec_info bb_vinfo,
slp_instance instance, slp_tree node,
hash_map<stmt_vec_info, slp_instance> &stmt_to_instance,
- hash_map<slp_instance, slp_instance> &instance_leader)
+ hash_map<slp_instance, slp_instance> &instance_leader,
+ hash_set<slp_tree> &visited)
{
stmt_vec_info stmt_info;
unsigned i;
- bool all = true;
+
FOR_EACH_VEC_ELT (SLP_TREE_SCALAR_STMTS (node), i, stmt_info)
{
bool existed_p;
slp_instance &stmt_instance
= stmt_to_instance.get_or_insert (stmt_info, &existed_p);
if (!existed_p)
- all = false;
+ ;
else if (stmt_instance != instance)
{
/* If we're running into a previously marked stmt make us the
@@ -3279,15 +3280,15 @@ vect_bb_partition_graph_r (bb_vec_info bb_vinfo,
}
stmt_instance = instance;
}
- /* If not all stmts had been visited we have to recurse on children. */
- if (all)
+
+ if (visited.add (node))
return;
slp_tree child;
FOR_EACH_VEC_ELT (SLP_TREE_CHILDREN (node), i, child)
if (SLP_TREE_DEF_TYPE (child) == vect_internal_def)
vect_bb_partition_graph_r (bb_vinfo, instance, child, stmt_to_instance,
- instance_leader);
+ instance_leader, visited);
}
/* Partition the SLP graph into pieces that can be costed independently. */
@@ -3302,13 +3303,15 @@ vect_bb_partition_graph (bb_vec_info bb_vinfo)
marked stmt, make the stmts leader the current SLP graph entry. */
hash_map<stmt_vec_info, slp_instance> stmt_to_instance;
hash_map<slp_instance, slp_instance> instance_leader;
+ hash_set<slp_tree> visited;
slp_instance instance;
for (unsigned i = 0; bb_vinfo->slp_instances.iterate (i, &instance); ++i)
{
instance_leader.put (instance, instance);
vect_bb_partition_graph_r (bb_vinfo,
instance, SLP_INSTANCE_TREE (instance),
- stmt_to_instance, instance_leader);
+ stmt_to_instance, instance_leader,
+ visited);
}
/* Then collect entries to each independent subgraph. */
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2020-10-12 12:31 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-10-12 12:31 [gcc r11-3820] fix SLP subgraph detection wrt fully shared lanes Richard Biener
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).