On Tue, 21 Feb 2012, Jakub Jelinek wrote:
> Hi!
>
> This function spends significant amount of code to update the virtual
> uses/defs in the new sequence, but only handles that way stores, not
> non-pure/const calls, so we ICE during tree DSE on this testcase, because
> vop has been marked for renaming.
>
> Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok
> for trunk?
Ok.
Thanks,
Richard.
> 2012-02-20 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/52318
> * gimple-fold.c (gimplify_and_update_call_from_tree): Add
> vdef also to non-pure/const call stmts in the sequence.
>
> * gcc.dg/pr52318.c: New test.
>
> --- gcc/gimple-fold.c.jj 2011-12-16 08:37:45.000000000 +0100
> +++ gcc/gimple-fold.c 2012-02-20 21:51:20.853133362 +0100
> @@ -591,8 +591,11 @@ gimplify_and_update_call_from_tree (gimp
> for (i = gsi_last (stmts); !gsi_end_p (i); gsi_prev (&i))
> {
> new_stmt = gsi_stmt (i);
> - if (gimple_assign_single_p (new_stmt)
> - && !is_gimple_reg (gimple_assign_lhs (new_stmt)))
> + if ((gimple_assign_single_p (new_stmt)
> + && !is_gimple_reg (gimple_assign_lhs (new_stmt)))
> + || (is_gimple_call (new_stmt)
> + && (gimple_call_flags (new_stmt)
> + & (ECF_NOVOPS | ECF_PURE | ECF_CONST | ECF_NORETURN)) == 0))
> {
> tree vdef;
> if (!laststore)
> --- gcc/testsuite/gcc.dg/pr52318.c.jj 2012-02-20 22:10:09.211778710 +0100
> +++ gcc/testsuite/gcc.dg/pr52318.c 2012-02-20 22:09:10.000000000 +0100
> @@ -0,0 +1,17 @@
> +/* PR tree-optimization/52318 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -ftracer -fno-tree-ccp -fno-tree-copy-prop
> -fno-tree-dce" } */
> +
> +int c;
> +char *p;
> +
> +void
> +foo (int i)
> +{
> + char a[2];
> + char b[20];
> + p = __builtin___stpcpy_chk (a, "", 2);
> + p = __builtin___stpcpy_chk (&b[16], i ? "e" : "jkl", 4);
> + if (c)
> + foo (i);
> +}
>
> Jakub
>
>
--
Richard Guenther <rguent...@suse.de>
SUSE / SUSE Labs
SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer
