--- Comment #20 from rguenth at gcc dot gnu dot org 2007-11-23 14:29
---
Fixed.
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
Status|NEW
--- Comment #19 from rguenth at gcc dot gnu dot org 2007-11-23 14:29
---
Subject: Bug 34176
Author: rguenth
Date: Fri Nov 23 14:28:59 2007
New Revision: 130379
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=130379
Log:
2007-11-23 Richard Guenther <[EMAIL PROTECTED]>
M
--- Comment #18 from dberlin at gcc dot gnu dot org 2007-11-22 23:02
---
Subject: Re: [4.3 Regression] SCCVN breaks gettext
On 22 Nov 2007 22:51:12 -, matz at gcc dot gnu dot org
<[EMAIL PROTECTED]> wrote:
>
>
> --- Comment #17 from matz at gcc dot gnu dot org 2007-11-22 22:5
--- Comment #17 from matz at gcc dot gnu dot org 2007-11-22 22:51 ---
We could also save the deletes by using a tick counter, but with that patch
the hash tables will be mostly small anyway, so emptying them should be fast
enough I hope. And of course we won't get as optimal results as
--- Comment #16 from rguenth at gcc dot gnu dot org 2007-11-22 22:24
---
... Indeed - wrong types in the testcase :)
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34176
--- Comment #15 from rguenth at gcc dot gnu dot org 2007-11-22 22:19
---
Bootstrapped / tested on x86_64, interestingly we manage to miscompile the
testcase at -O0 now :( (but this _must_ be unrelated to this fix!)
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34176
--- Comment #14 from rguenth at gcc dot gnu dot org 2007-11-22 20:54
---
Created an attachment (id=14615)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14615&action=view)
prototype patch
Patch clearing the optimistic tables before every iteration and doing a
fallback
lookup on th
--- Comment #13 from dberlin at gcc dot gnu dot org 2007-11-22 15:35
---
Subject: Re: [4.3 Regression] SCCVN breaks gettext
On 22 Nov 2007 14:03:35 -, matz at suse dot de
<[EMAIL PROTECTED]> wrote:
>
>
> --- Comment #9 from matz at suse dot de 2007-11-22 14:03 ---
> Subje
--- Comment #12 from rguenther at suse dot de 2007-11-22 15:08 ---
Subject: Re: [4.3 Regression] SCCVN breaks
gettext
On Thu, 22 Nov 2007, matz at gcc dot gnu dot org wrote:
> --- Comment #11 from matz at gcc dot gnu dot org 2007-11-22 14:13 ---
> Btw. canonicalization/rewri
--- Comment #11 from matz at gcc dot gnu dot org 2007-11-22 14:13 ---
Btw. canonicalization/rewriting on INSERT is indeed okay. As long as the hash
tables are deleted after every iteration (or otherwise invalidated). At least
the now provably false information needs to be removed, whic
--- Comment #10 from matz at gcc dot gnu dot org 2007-11-22 14:07 ---
Created an attachment (id=14608)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14608&action=view)
my dump (without patch)
This is the dump I'm talking about. Richard uses a different compiler, so
our SSA versio
--- Comment #9 from matz at suse dot de 2007-11-22 14:03 ---
Subject: Re: [4.3 Regression] SCCVN breaks
gettext
[sorry for the breakage in last response]
It does not. The RPO algorithm (the one proven) uses hash table deletes
per iteration. About the SCC algorithm they have to say
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
Target Milestone|--- |4.3.0
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34176
--- Comment #8 from rguenth at gcc dot gnu dot org 2007-11-22 14:00 ---
If you go through the SCCVN dump you realize that Michas analysis of the
problem
is correct btw:
SCC consists of: nitems_21 nitems_1 new_max_9 dest_10 destptr_3 destptr.2_16
dest.3_17 D.1581_18 D.1582_19 nitems_20 d
--- Comment #7 from matz at suse dot de 2007-11-22 13:58 ---
Subject: Re: [4.3 Regression] SCCVN breaks
gettext
Hi,
On Thu, 22 Nov 2007, dberlin at dberlin dot org wrote:
> Right, but this is the optimistic set of hash tables, so that is okay.
I initially thought so too, but it rea
--- Comment #6 from rguenther at suse dot de 2007-11-22 09:35 ---
Subject: Re: [4.3 Regression] SCCVN breaks
gettext
On Thu, 22 Nov 2007, dberlin at dberlin dot org wrote:
> --- Comment #4 from dberlin at gcc dot gnu dot org 2007-11-22 04:48
> ---
> Subject: Re: [4.3 Regre
--- Comment #5 from steven at gcc dot gnu dot org 2007-11-22 07:10 ---
> The one in a real compiler I have of SCCVN both do canonicalization on
> insert, as does the original code from Rice's massively scalar
> compiler (which is where the algorithm comes from).
"The one"? "real compile
--- Comment #4 from dberlin at gcc dot gnu dot org 2007-11-22 04:48 ---
Subject: Re: [4.3 Regression] SCCVN breaks gettext
On 22 Nov 2007 04:26:57 -, matz at gcc dot gnu dot org
<[EMAIL PROTECTED]> wrote:
>
>
> --- Comment #2 from matz at gcc dot gnu dot org 2007-11-22 04:26 -
--- Comment #3 from matz at gcc dot gnu dot org 2007-11-22 04:31 ---
Created an attachment (id=14600)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14600&action=view)
incomplete fix
Concept patch touching only the unary case. binary, phi nodes and maybe
references would need some
--- Comment #2 from matz at gcc dot gnu dot org 2007-11-22 04:26 ---
The problem starts already in the first iteration:
Value numbering destptr_3 stmt = destptr_3 = PHI
Setting value number of destptr_3 to dest_9
So, for now we assume dest_9 == destptr_3, quite okay, lets assume so. N
--- Comment #1 from rguenth at gcc dot gnu dot org 2007-11-21 23:02 ---
If you look at what SCCVN does for the affected SCC then you see that iteration
converges to
Setting value number of dest_7 to dest_7
Setting value number of list_23 to list_23
Value numbering destptr_3 stmt = destp
21 matches
Mail list logo
