This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: RFA: PR62208: const0_rtx used for vectors in simplify-rtx.c
- From: Richard Biener <richard dot guenther at gmail dot com>
- To: Richard Sandiford <rdsandiford at googlemail dot com>,gcc-patches at gcc dot gnu dot org
- Date: Sat, 06 Sep 2014 16:06:43 +0200
- Subject: Re: RFA: PR62208: const0_rtx used for vectors in simplify-rtx.c
- Authentication-results: sourceware.org; auth=none
- References: <87tx4lxctu dot fsf at googlemail dot com>
On September 6, 2014 12:28:29 PM CEST, Richard Sandiford <rdsandiford@googlemail.com> wrote:
>This PR is about a case where an XOR+EQ sequence can legitimately
>be simplified. The problem was that the sequence was operating on
>vectors while the simplified version used a 0 CONST_INT (const0_rtx).
>
>The code seems to have been like that for a long time but presumably
>the bug has been latent until now. I suppose normally we'd want
>earlier optimisations to catch this.
>
>Tested on x86_64-linux-gnu. OK to install?
OK.
Thanks,
Richard.
>Thanks,
>Richard
>
>
>gcc/
> PR rtl-optimization/62208
> * simplify-rtx.c (simplify_relational_operation_1): Use CONST0_RTX
> rather than const0_rtx in eq/ne-xor simplifications.
>
>gcc/testsuite/
> * gcc.target/i386/pr62208.c: New test.
>
>Index: gcc/simplify-rtx.c
>===================================================================
>--- gcc/simplify-rtx.c 2014-09-06 11:05:20.859563557 +0100
>+++ gcc/simplify-rtx.c 2014-09-06 11:05:20.935564244 +0100
>@@ -4480,16 +4480,16 @@ simplify_relational_operation_1 (enum rt
> && op0code == XOR
> && rtx_equal_p (XEXP (op0, 0), op1)
> && !side_effects_p (XEXP (op0, 0)))
>- return simplify_gen_relational (code, mode, cmp_mode,
>- XEXP (op0, 1), const0_rtx);
>+ return simplify_gen_relational (code, mode, cmp_mode, XEXP (op0,
>1),
>+ CONST0_RTX (mode));
>
> /* Likewise (eq/ne (xor x y) y) simplifies to (eq/ne x 0). */
> if ((code == EQ || code == NE)
> && op0code == XOR
> && rtx_equal_p (XEXP (op0, 1), op1)
> && !side_effects_p (XEXP (op0, 1)))
>- return simplify_gen_relational (code, mode, cmp_mode,
>- XEXP (op0, 0), const0_rtx);
>+ return simplify_gen_relational (code, mode, cmp_mode, XEXP (op0,
>0),
>+ CONST0_RTX (mode));
>
> /* (eq/ne (xor x C1) C2) simplifies to (eq/ne x (C1^C2)). */
> if ((code == EQ || code == NE)
>Index: gcc/testsuite/gcc.target/i386/pr62208.c
>===================================================================
>--- /dev/null 2014-09-03 21:33:59.721983712 +0100
>+++ gcc/testsuite/gcc.target/i386/pr62208.c 2014-09-06
>11:09:18.220708443 +0100
>@@ -0,0 +1,23 @@
>+/* { dg-options "-O3 -fwhole-program -march=x86-64" } */
>+
>+int *a;
>+unsigned int b;
>+
>+void fn2 ()
>+{
>+ int t[9];
>+ for (; b; b++)
>+ *a ^= (~t[b] != t[b]);
>+}
>+
>+int fn1 ()
>+{
>+ fn2 ();
>+ return 0;
>+}
>+
>+int main ()
>+{
>+ fn1 ();
>+ return 0;
>+}