]> git.ipfire.org Git - thirdparty/gcc.git/commit
For the `-A CMP -B -> B CMP A` pattern allow EQ/NE for all integer types
authorAndrew Pinski <apinski@marvell.com>
Tue, 6 Jun 2023 02:12:43 +0000 (19:12 -0700)
committerAndrew Pinski <apinski@marvell.com>
Wed, 7 Jun 2023 02:57:18 +0000 (19:57 -0700)
commitcc155ff9c38848a8e6a7125dd0b66ac0aef47880
treea9937355faa4b4d301d58e2c7589f1033d622643
parent3fe017ee79211820ca774f4c87ef86753eeef567
For the `-A CMP -B -> B CMP A` pattern allow EQ/NE for all integer types

I noticed while looking at some code generation issue, that forwprop
was not handling `-a == 0` for unsigned types and I was confused why
it was not.
r6-1814-g66e1cacf608045 removed these from fold because they
were supposed to be already handled by the match.pd patterns
but it was missed that the match.pd patterns checked
TYPE_OVERFLOW_UNDEFINED while fold didn't do that for NE/EQ.
This patch removes the restriction on NE/EQ on TYPE_OVERFLOW_UNDEFINED.

OK? Bootstrapped and tested on x86_64-linux-gnu.

gcc/ChangeLog:

PR tree-optimization/110134
* match.pd (-A CMP -B -> B CMP A): Allow EQ/NE for all integer
types.
(-A CMP CST -> B CMP (-CST)): Likewise.

gcc/testsuite/ChangeLog:

PR tree-optimization/110134
* gcc.dg/tree-ssa/negneq-1.c: New test.
* gcc.dg/tree-ssa/negneq-2.c: New test.
* gcc.dg/tree-ssa/negneq-3.c: New test.
* gcc.dg/tree-ssa/negneq-4.c: New test.
gcc/match.pd
gcc/testsuite/gcc.dg/tree-ssa/negneq-1.c [new file with mode: 0644]
gcc/testsuite/gcc.dg/tree-ssa/negneq-2.c [new file with mode: 0644]
gcc/testsuite/gcc.dg/tree-ssa/negneq-3.c [new file with mode: 0644]
gcc/testsuite/gcc.dg/tree-ssa/negneq-4.c [new file with mode: 0644]