This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch RFC [tree-ssa]: ICE in DSE
- From: law at redhat dot com
- To: Dale Johannesen <dalej at apple dot com>
- Cc: gcc-patches Patches <gcc-patches at gcc dot gnu dot org>
- Date: Thu, 18 Mar 2004 14:09:15 -0700
- Subject: Re: Patch RFC [tree-ssa]: ICE in DSE
- Reply-to: law at redhat dot com
In message <1F2CABE7-791D-11D8-A952-000A95D7CD40@apple.com>, Dale Johannesen wr
ites:
>I'm getting an ICE in DSE in sixtrack, which I can't post, and haven't
>been able to
>duplicate it in a small example. However the situation is
>straightforward:
>
># i__9878 = VDEF <i__9879>; [A]
>i_ = 6;
># i__9164 = VDEF <i__9878>; [B]
>i_ = 1;
>
>#i__10497 = PHI <i__9164(4), i__2915(7)>; [C]
>
># i__2915 = VDEF <i__10497>; [D]
>i_ = T.17897_9327;
>
>When deleting [B], the use from [A] is updated to point to [D] rather
>than [C]
>(which is correctly updated to use i_9878). This leads to a crash
>later when deleting [A].
>The fix is also easy. OK, pending tesitng and ChangeLog?
This is fine. What's odd is I thought the code already did precisely
what your patch does. Presumably that was in one of my development
versions and got lost.
You might want to twiddle the comment to make it match reality.
jeff