On Mon, Mar 04, 2013 at 11:17:41PM +0100, Steven Bosscher wrote:
> On Mon, Mar 4, 2013 at 10:17 PM, Jakub Jelinek wrote:
> > Something that again hits lots of testcases during valgrind checking
> > bootstrap. init_alias_analysis apparently does
> > vec_safe_grow_cleared (reg_known_value, maxreg - FIRST_PSEUDO_REGISTER);
> > reg_known_equiv_p = sbitmap_alloc (maxreg - FIRST_PSEUDO_REGISTER);
> > but doesn't bitmap_clear (reg_known_equiv_p), perhaps as an optimization?
>
> No, an incorrect replacement with sbitmap. We used to have:
>
> reg_known_equiv_p = XCNEWVEC (bool, reg_known_value_size);
>
> I'm probably to blame for this one, actually :-)
This works too on the testcase (e.g. gcc.dg/54455.c).
Will bootstrap/regtest it overnight.
2013-03-04 Jakub Jelinek <ja...@redhat.com>
* alias.c (init_alias_analysis): Clear reg_known_equiv_p bitmap.
--- gcc/alias.c.jj 2013-01-18 12:49:31.000000000 +0100
+++ gcc/alias.c 2013-03-04 23:22:38.865751120 +0100
@@ -2812,6 +2812,7 @@ init_alias_analysis (void)
vec_safe_grow_cleared (reg_known_value, maxreg - FIRST_PSEUDO_REGISTER);
reg_known_equiv_p = sbitmap_alloc (maxreg - FIRST_PSEUDO_REGISTER);
+ bitmap_clear (reg_known_equiv_p);
/* If we have memory allocated from the previous run, use it. */
if (old_reg_base_value)
Jakub
