* [COMMITTED] Correctly unify recomputation with existing range.
@ 2021-06-24 17:34 Andrew MacLeod
0 siblings, 0 replies; only message in thread
From: Andrew MacLeod @ 2021-06-24 17:34 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 496 bytes --]
When propagating the on-entry cache, new block ranges are calculated by
combining all the incoming edges and comparing to the old value.
When a recomputation was performed on an edge, it didn't take into
account that the value in the block may already be better than a
potential recompuation... Thus a worse values was sometimes propagated.
Fixed by simply calling the now correct range_on_edge the cache provides.
bootstraps on x86_64-pc-linux-gnu with no regressions. pushed.
Andrew
[-- Attachment #2: 0002-Correctly-unify-recomputation-with-existing-range.patch --]
[-- Type: text/x-patch, Size: 2468 bytes --]
From 5bdcfb74ff97b42f08993af8614c35685fdd8689 Mon Sep 17 00:00:00 2001
From: Andrew MacLeod <amacleod@redhat.com>
Date: Wed, 23 Jun 2021 15:25:45 -0400
Subject: [PATCH 2/3] Correctly unify recomputation with existing range.
When propagating the on-entry cache, new block ranges are calculated
by combining all the incoming edges and comparing to the old value.
When a recomputation was performed on an edge, it didn't take into account
that the value in the block may already be better than a potential
recompuation... Thus a worse values was sometimes propagated.
Fixed by simply calling the now correct range_on_edge the cache provides.
* gimple-range-cache.cc (ranger_cache::propagate_cache): Call
range_on_edge instead of manually calculating.
---
gcc/gimple-range-cache.cc | 29 +++++++++--------------------
1 file changed, 9 insertions(+), 20 deletions(-)
diff --git a/gcc/gimple-range-cache.cc b/gcc/gimple-range-cache.cc
index a377261c5c7..98ecdbbd68e 100644
--- a/gcc/gimple-range-cache.cc
+++ b/gcc/gimple-range-cache.cc
@@ -1037,27 +1037,12 @@ ranger_cache::propagate_cache (tree name)
new_range.set_undefined ();
FOR_EACH_EDGE (e, ei, bb->preds)
{
+ range_on_edge (e_range, e, name);
if (DEBUG_RANGE_CACHE)
- fprintf (dump_file, " edge %d->%d :", e->src->index, bb->index);
- // Get whatever range we can for this edge.
- if (!m_gori.outgoing_edge_range_p (e_range, e, name, *this))
{
- exit_range (e_range, name, e->src);
- if (DEBUG_RANGE_CACHE)
- {
- fprintf (dump_file, "No outgoing edge range, picked up ");
- e_range.dump (dump_file);
- fprintf (dump_file, "\n");
- }
- }
- else
- {
- if (DEBUG_RANGE_CACHE)
- {
- fprintf (dump_file, "outgoing range :");
- e_range.dump (dump_file);
- fprintf (dump_file, "\n");
- }
+ fprintf (dump_file, " edge %d->%d :", e->src->index, bb->index);
+ e_range.dump (dump_file);
+ fprintf (dump_file, "\n");
}
new_range.union_ (e_range);
if (new_range.varying_p ())
@@ -1074,7 +1059,11 @@ ranger_cache::propagate_cache (tree name)
if (DEBUG_RANGE_CACHE)
{
if (!ok_p)
- fprintf (dump_file, " Cache failure to store value.");
+ {
+ fprintf (dump_file, " Cache failure to store value:");
+ print_generic_expr (dump_file, name, TDF_SLIM);
+ fprintf (dump_file, " ");
+ }
else
{
fprintf (dump_file, " Updating range to ");
--
2.17.2
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2021-06-24 17:34 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-06-24 17:34 [COMMITTED] Correctly unify recomputation with existing range Andrew MacLeod
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).