diff mbox series

[COMMITTED] Fix ifcvt-4.c to not depend on VRP2 asserts.

Message ID a4c304da-7898-34ba-a8af-175cc1f0f0ba@redhat.com
State New
Headers show
Series [COMMITTED] Fix ifcvt-4.c to not depend on VRP2 asserts. | expand

Commit Message

Andrew MacLeod Oct. 28, 2021, 2:51 p.m. UTC 
as discussed elsewhere, gcc.dg/ifcvt-4.c seems to be a flawed testcase 
as it is.

it does not use __builtin_expect properly, so edges are predicted 
50/50.  It also turns out that it only works due to an oddity that 
causes the basic blocks to be restructured by the VRP2 pass when it 
removes all the ASSERT_EXPRs that it uses. If we run ranger for VRP2, 
the testcase fails because the blocks do not get rearranged.

This patch tweaks the testcase so that it uses expect properly and 
passes with both classic vrp2, and ranger vrp2.   I have tested it on 
both powerpc and x86 which seem the be the primaries.  Pushed.

Andrew
diff mbox series

Patch

From d123daec0c237533cf974334d98bc6d357d4273e Mon Sep 17 00:00:00 2001
From: Andrew MacLeod <amacleod@redhat.com>
Date: Mon, 25 Oct 2021 13:34:36 -0400
Subject: [PATCH 3/3] Fix ifcvt-4.c to not depend on VRP2 asserts.

The testcase fails if VRP2 is replaced with a non-assert based VRP because it
accidentally depends on specific IL changes when the asserts are removed.  This
removes that dependency.

	gcc/testsuite/
	* gcc.dg/ifcvt-4.c: Adjust.
---
 gcc/testsuite/gcc.dg/ifcvt-4.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/gcc/testsuite/gcc.dg/ifcvt-4.c b/gcc/testsuite/gcc.dg/ifcvt-4.c
index ec142cfd943..e74e449b402 100644
--- a/gcc/testsuite/gcc.dg/ifcvt-4.c
+++ b/gcc/testsuite/gcc.dg/ifcvt-4.c
@@ -13,8 +13,7 @@  foo (word x, word y, word a)
   word i = x;
   word j = y;
   /* Try to make taking the branch likely.  */
-  __builtin_expect (x > y, 1);
-  if (x > y)
+  if (__builtin_expect (x > y, 0))
     {
       i = a;
       j = i;
-- 
2.17.2