This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PATCH: calculate CDI_DOMINATORS in tree-ssa-sink
Steven Bosscher wrote:
> On Wednesday 03 January 2007 01:29, Josh Conner wrote:
>> All -
>>
>> There's a relatively obvious fix that should be made in
>> execute_sink_code. This statement:
>>
>> calculate_dominance_info (CDI_DOMINATORS | CDI_POST_DOMINATORS);
>>
>> assuming it is intended to calculate both dominators and
>> post-dominators, is incorrect. Since CDI_*DOMINATORS are enums and not
>> bitfields (CDI_DOMINATORS==0 and CDI_POST_DOMINATORS==1), we're only
>> calculating post-dom info.
>
> This is not the first time that this bug had to be fixed in some
> pass. You get to score Mega Bonus Points if you can fix the
> problem at the root:
>
> Index: basic-block.h
> ===================================================================
> --- basic-block.h (revision 120352)
> +++ basic-block.h (working copy)
> @@ -935,8 +935,8 @@ extern void reorder_basic_blocks (unsign
>
> enum cdi_direction
> {
> - CDI_DOMINATORS,
> - CDI_POST_DOMINATORS
> + CDI_DOMINATORS = 1,
> + CDI_POST_DOMINATORS = 2
> };
>
> enum dom_state
>
> ...and a larger number of fixes in dominance.c everywhere where
> we look at dom_computed[].
OK, I could use some Mega Bonus Points -- hopefully they can be
converted into the Supreme Ultimate iPoints that my manager accepts... :)
I'll take a look into this.
- Josh