This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: RFA: Don't destroy SCEV cache in copy-prop
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: Michael Matz <matz at suse dot de>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Tue, 7 Dec 2010 05:34:44 -0800
- Subject: Re: RFA: Don't destroy SCEV cache in copy-prop
- References: <Pine.LNX.4.64.1011261558470.29956@wotan.suse.de> <AANLkTim9w1N8LT1-UZMkjbUA0NFMpK+wdjkJaNseLPWS@mail.gmail.com>
On Sat, Dec 4, 2010 at 9:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Nov 26, 2010 at 7:03 AM, Michael Matz <matz@suse.de> wrote:
>> Hello,
>>
>> while working on something else I got a segfault in SCEV analysis because
>> it contained stale SSA names. ?The reason is that copy-prop always DCEs
>> the propagated copies, invalidating SCEV entries referring to the LHS of
>> those. ?We could either reset the SCEV cache (at the expense of having to
>> recompute it again), or simply not do DCE when we have loops. ?I chose the
>> latter.
>>
>> I've also removed the superflous copy-prop in the graphite pipeline, it's
>> no use to run two copy-props next to each other.
>>
>> Unfortunately I don't have a testcase that fails with an unmodified GCC.
>>
>> Regstrapping on x86_64-linux (all langs + Ada) in progress. ?Okay if that
>> passes?
>>
>>
>
> This caused:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46799
>
This also caused:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46834
--
H.J.