--- Comment #24 from rakdver at gcc dot gnu dot org 2005-12-08 09:34
---
Subject: Bug 25248
Author: rakdver
Date: Thu Dec 8 09:34:26 2005
New Revision: 108225
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=108225
Log:
PR tree-optimization/25248
* tree-scalar-ev
--- Comment #23 from rakdver at gcc dot gnu dot org 2005-12-06 20:33
---
Patch: http://gcc.gnu.org/ml/gcc-patches/2005-12/msg00454.html
--
rakdver at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #22 from bergner at vnet dot ibm dot com 2005-12-06 18:47
---
...and I can verify that the patch fixes the reduced test case. Thanks!
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25248
--- Comment #21 from olh at suse dot de 2005-12-06 18:11 ---
Created an attachment (id=10420)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=10420&action=view)
pr25248.patch
This patch fixes it for me.
Let me attach it in a readable form.
--
http://gcc.gnu.org/bugzilla/show_bu
--- Comment #20 from rakdver at gcc dot gnu dot org 2005-12-06 15:43
---
This patch should fix the problem:
Index: tree-scalar-evolution.c
===
*** tree-scalar-evolution.c (revision 108078)
--- tree-scalar-evolution.c
--- Comment #19 from rakdver at gcc dot gnu dot org 2005-12-06 15:28
---
Seems like some problem in iv analysis:
ssa name pstart_17
type long unsigned int
base (long unsigned int) (unsigned int) D.1301_7 + (long unsigned int)
(unsigned int) pstart_10
step (long unsigned int) step
--- Comment #18 from olh at suse dot de 2005-12-06 14:26 ---
Adding -fno-ivopts to CFLAGS for hash_utils_64.o fixes it for me.
Tested with gcc-4_1-branch r108104 and current binutils.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25248
--- Comment #17 from rguenth at gcc dot gnu dot org 2005-12-06 13:59
---
IVOPTs truncates pstartD.1866_10 down to int, which is wrong:
htab_bolt_mapping (pstartD.1866, cntD.1867)
{
unsigned intD.3 D.1959;
long unsigned intD.4 D.1960;
long unsigned intD.4 ivtmp.50D.1954;
unsigne
--- Comment #16 from rguenth at gcc dot gnu dot org 2005-12-06 13:48
---
Confirmed. -fno-ivopts fixes it.
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #15 from rguenth at gcc dot gnu dot org 2005-12-06 13:43
---
So, CCing zdenek. Btw., checking -floop-optimize2 may also be worth trying
(setting up a rpoper chroot now).
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
-
--- Comment #14 from bergner at vnet dot ibm dot com 2005-12-06 13:40
---
sizeof(unsigned long) and sizeof(unsigned long long) are both 8 bytes on ppc64.
Olaf, -O1 isn't a workaround, it's the minimum optimization level that still
shows the bug. Trying some -fno-* options, using -fno-
--- Comment #13 from rguenth at gcc dot gnu dot org 2005-12-06 13:39
---
hmhm, maybe I'm in a ppc32 chroot.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25248
--- Comment #12 from rguenth at gcc dot gnu dot org 2005-12-06 13:37
---
Note that in the reduced testcase we have
pr25248.c: In function ?htab_bolt_mapping?:
pr25248.c:15: warning: integer constant is too large for ?unsigned long? type
pr25248.c: In function ?main?:
pr25248.c:24: warn
--- Comment #11 from olh at suse dot de 2005-12-06 08:44 ---
same is true for a r99558 compiled object file, in a gcc40 kernel.
-O1 doesnt help there.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25248
--- Comment #10 from olh at suse dot de 2005-12-06 08:38 ---
Odd, -O1 doesnt fix it for me:
+CFLAGS_hash_utils_64.o += -O1
gcc version 4.2.0 20051206 (experimental)
GNU ld version 2.16.91 20051206
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25248
--- Comment #9 from bergner at vnet dot ibm dot com 2005-12-06 06:26
---
Created an attachment (id=10414)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=10414&action=view)
Minimal test case.
Here's a minimal test case. This works with older gcc's and without
optimization on newer
--- Comment #8 from bergner at vnet dot ibm dot com 2005-12-06 01:57
---
We're miscompiling htab_bolt_mapping(). What do I win? ;-)
I have a userland test case which I'll attach once I've shunk it down a little.
--
bergner at vnet dot ibm dot com changed:
What|Remov
--- Comment #7 from giovannibajo at libero dot it 2005-12-04 23:17 ---
Further bonus points if you can spot which function is miscompiled.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25248
--- Comment #6 from olh at suse dot de 2005-12-04 20:58 ---
Created an attachment (id=10403)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=10403&action=view)
PR25248-3.tar.bz2
If someone can spot the bug, I cant.
Unified all asm labels to reduce diff noise.
The object file preven
--- Comment #5 from pinskia at gcc dot gnu dot org 2005-12-04 19:24 ---
(In reply to comment #4)
> Created an attachment (id=10402)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=10402&action=view) [edit]
> PR25248-2.tar.bz2
> this change breaks it. I replaced only arch/powerpc/mm/h
20 matches
Mail list logo
