public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r11-8696] tree-optimization/101088 - fix SM invalidation issue
@ 2021-07-07 14:06 Richard Biener
0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2021-07-07 14:06 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:91fc9610464281bb28e3c6529399fdcfb57fa812
commit r11-8696-g91fc9610464281bb28e3c6529399fdcfb57fa812
Author: Richard Biener <rguenther@suse.de>
Date: Wed Jun 16 09:49:18 2021 +0200
tree-optimization/101088 - fix SM invalidation issue
When we face a sm_ord vs sm_unord for the same ref during
store sequence merging we assert that the ref is already marked
unsupported. But it can be that it will only be marked so
during the ongoing merging so instead of asserting mark it here.
Also apply some optimization to not waste resources to search
for already unsupported refs.
2021-06-16 Richard Biener <rguenther@suse.de>
PR tree-optimization/101088
* tree-ssa-loop-im.c (sm_seq_valid_bb): Only look for
supported refs on edges. Do not assert same ref but
different kind stores are unsuported but mark them so.
(hoist_memory_references): Only look for supported refs
on exits.
* gcc.dg/torture/pr101088.c: New testcase.
(cherry picked from commit 43fc4234ad3d9302d3460385b6fdb5e3f59b6986)
Diff:
---
gcc/testsuite/gcc.dg/torture/pr101088.c | 45 +++++++++++++++++++++++++++++++++
gcc/tree-ssa-loop-im.c | 21 +++++++++++----
2 files changed, 61 insertions(+), 5 deletions(-)
diff --git a/gcc/testsuite/gcc.dg/torture/pr101088.c b/gcc/testsuite/gcc.dg/torture/pr101088.c
new file mode 100644
index 00000000000..00fce39d2f4
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/pr101088.c
@@ -0,0 +1,45 @@
+/* { dg-do compile } */
+
+int bh, on, h0;
+
+void
+qw (int n2)
+{
+ int *e5;
+
+ if (n2 == 0)
+ {
+ n2 = 1;
+ while (n2 != 0)
+ for (n2 = 0; n2 < 1; ++n2)
+ {
+ }
+
+ e5 = &n2;
+ }
+ else
+ e5 = &on;
+
+ while (h0 < 1)
+ {
+ if (on == 0)
+ {
+ ++*e5;
+ bh = 0;
+ }
+ else
+ {
+ bh = 0;
+ ++on;
+ *e5 = on;
+ h0 = *e5;
+ if (h0 == 0)
+ {
+ *e5 = 0;
+ ++h0;
+ }
+ }
+
+ ++h0;
+ }
+}
diff --git a/gcc/tree-ssa-loop-im.c b/gcc/tree-ssa-loop-im.c
index 1c865b28fd6..7de47edbcb3 100644
--- a/gcc/tree-ssa-loop-im.c
+++ b/gcc/tree-ssa-loop-im.c
@@ -2340,7 +2340,13 @@ sm_seq_valid_bb (class loop *loop, basic_block bb, tree vdef,
tree vuse = gimple_phi_arg_def (phi, i);
edge e = gimple_phi_arg_edge (phi, i);
auto_vec<seq_entry> edge_seq;
- bitmap_copy (tem_refs_not_in_seq, refs_not_in_seq);
+ bitmap_and_compl (tem_refs_not_in_seq,
+ refs_not_in_seq, refs_not_supported);
+ /* If we've marked all refs we search for as unsupported
+ we can stop processing and use the sequence as before
+ the PHI. */
+ if (bitmap_empty_p (tem_refs_not_in_seq))
+ return 1;
eret = sm_seq_valid_bb (loop, e->src, vuse, edge_seq,
tem_refs_not_in_seq, refs_not_supported,
true, fully_visited);
@@ -2379,9 +2385,9 @@ sm_seq_valid_bb (class loop *loop, basic_block bb, tree vdef,
/* sm_other prevails. */
else if (first_edge_seq[i].second != edge_seq[i].second)
{
- /* This is just an optimization. */
- gcc_assert (bitmap_bit_p (refs_not_supported,
- first_edge_seq[i].first));
+ /* Make sure the ref is marked as not supported. */
+ bitmap_set_bit (refs_not_supported,
+ first_edge_seq[i].first);
first_edge_seq[i].second = sm_other;
first_edge_seq[i].from = NULL_TREE;
}
@@ -2533,7 +2539,12 @@ hoist_memory_references (class loop *loop, bitmap mem_refs,
vec<seq_entry> seq;
seq.create (4);
auto_bitmap refs_not_in_seq (&lim_bitmap_obstack);
- bitmap_copy (refs_not_in_seq, mem_refs);
+ bitmap_and_compl (refs_not_in_seq, mem_refs, refs_not_supported);
+ if (bitmap_empty_p (refs_not_in_seq))
+ {
+ seq.release ();
+ break;
+ }
auto_bitmap fully_visited;
int res = sm_seq_valid_bb (loop, e->src, NULL_TREE,
seq, refs_not_in_seq,
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2021-07-07 14:06 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-07-07 14:06 [gcc r11-8696] tree-optimization/101088 - fix SM invalidation issue 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).