On August 29, 2016 9:43:50 PM GMT+02:00, Jakub Jelinek <ja...@redhat.com> wrote:
>Hi!
>
>Returning c from avoid_constant_pool_reference for constant pool
>references
>is only fine if the modes match and offset is zero, otherwise if the
>offset
>is sane, we can try to simplify it as subreg. But, returning c for out
>of
>bounds offsets is always wrong, and especially if the mode doesn't
>match, it
>can results in all kinds of ICEs.
>
>Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
OK.
Thanks,
Richard.
>2016-08-29 Jakub Jelinek <ja...@redhat.com>
>
> PR middle-end/77377
> * simplify-rtx.c (avoid_constant_pool_reference): For out of bounds
> constant pool reference return x instead of c.
>
> * gcc.target/i386/pr77377.c: New test.
>
>--- gcc/simplify-rtx.c.jj 2016-08-12 17:33:46.000000000 +0200
>+++ gcc/simplify-rtx.c 2016-08-29 13:32:38.126859949 +0200
>@@ -251,15 +251,14 @@ avoid_constant_pool_reference (rtx x)
> /* If we're accessing the constant in a different mode than it was
> originally stored, attempt to fix that up via subreg simplifications.
> If that fails we have no choice but to return the original memory. */
>- if ((offset != 0 || cmode != GET_MODE (x))
>- && offset >= 0 && offset < GET_MODE_SIZE (cmode))
>+ if (offset == 0 && cmode == GET_MODE (x))
>+ return c;
>+ else if (offset >= 0 && offset < GET_MODE_SIZE (cmode))
> {
> rtx tem = simplify_subreg (GET_MODE (x), c, cmode, offset);
> if (tem && CONSTANT_P (tem))
> return tem;
> }
>- else
>- return c;
> }
>
> return x;
>--- gcc/testsuite/gcc.target/i386/pr77377.c.jj 2016-08-29
>13:39:01.707328239 +0200
>+++ gcc/testsuite/gcc.target/i386/pr77377.c 2016-08-29
>13:38:55.901396832 +0200
>@@ -0,0 +1,6 @@
>+/* PR middle-end/77377 */
>+/* { dg-do compile } */
>+/* { dg-options "-O3 -msse" } */
>+/* { dg-additional-options "-fpic" { target fpic } } */
>+
>+#include "../../c-c++-common/pr59037.c"
>
> Jakub
