--- Comment #48 from hjl at gcc dot gnu dot org 2010-09-07 21:19 ---
Subject: Bug 36502
Author: hjl
Date: Tue Sep 7 21:18:55 2010
New Revision: 163971
URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=163971
Log:
Redefine STACK_BOUNDARY/PREFERRED_STACK_BOUNDARY for Darwin/x86.
--- Comment #49 from hjl dot tools at gmail dot com 2010-09-08 02:16
---
Fixed.
--
hjl dot tools at gmail dot com changed:
What|Removed |Added
Status|NEW
--- Comment #44 from howarth at nitro dot med dot uc dot edu 2010-09-06
13:17 ---
Created an attachment (id=21709)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21709action=view)
retain redefinition of MAIN_STACK_BOUNDARY as required
--
howarth at nitro dot med dot uc dot edu
--- Comment #45 from howarth at nitro dot med dot uc dot edu 2010-09-06
13:57 ---
(In reply to comment #44)
Created an attachment (id=21709)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21709action=view) [edit]
retain redefinition of MAIN_STACK_BOUNDARY as required
Testsuite
--- Comment #46 from dominiq at lps dot ens dot fr 2010-09-06 14:04 ---
gfortran.dg/backspace_1.f, gfortran.dg/record_marker_2.f, ...
They are pr45534 and probably fixed at revision 163913 (testing).
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36502
--- Comment #47 from howarth at nitro dot med dot uc dot edu 2010-09-07
03:27 ---
Testsuite results for gcc-pr45234-2.patch and PR36502v9.patch on
i386-apple-darwin10 are at...
http://gcc.gnu.org/ml/gcc-testresults/2010-09/msg00603.html
PR36502, PR42313 and PR44651 fixed as well as
--- Comment #42 from howarth at nitro dot med dot uc dot edu 2010-09-04
20:39 ---
Posted final version of proposed patch to...
http://gcc.gnu.org/ml/gcc-patches/2010-09/msg00237.html
This patch also fixes PR42313 and PR44651 as yet another added bonus.
--
--- Comment #43 from howarth at nitro dot med dot uc dot edu 2010-09-04
21:21 ---
Updated patch to reflect the wider coverage of PRs fixed...
http://gcc.gnu.org/ml/gcc-patches/2010-09/msg00365.html
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36502
--- Comment #41 from howarth at nitro dot med dot uc dot edu 2010-09-03
01:58 ---
Created an attachment (id=21681)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21681action=view)
don't redefine MAIN_STACK_BOUNDARY so -fstack-usage can work
--
--- Comment #31 from hjl dot tools at gmail dot com 2010-09-01 14:51
---
(In reply to comment #30)
With only gcc-pr45234-2.patch at r163712 , I am seeing the following
regressions...
FAIL: gcc.c-torture/execute/builtins/sprintf-chk.c compilation, -Os
(internal
compiler error)
--- Comment #32 from howarth at nitro dot med dot uc dot edu 2010-09-01
15:18 ---
(In reply to comment #31)
I compared revision 163733 against revision 163733 + gcc-pr45234-2.patch
with a cross compiler to x86_64-apple-darwin10.4.0. I saw no
differences in .su files for
--- Comment #33 from howarth at nitro dot med dot uc dot edu 2010-09-01
15:30 ---
Created an attachment (id=21647)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21647action=view)
Don't redefine STACK_BOUNDARY and replace STACK_BOUNDARY with 128.
--
howarth at nitro dot med dot
--- Comment #34 from hjl dot tools at gmail dot com 2010-09-01 15:31
---
(In reply to comment #33)
Created an attachment (id=21647)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21647action=view) [edit]
Don't redefine STACK_BOUNDARY and replace STACK_BOUNDARY with 128.
Please
--- Comment #35 from howarth at nitro dot med dot uc dot edu 2010-09-01
15:37 ---
(In reply to comment #34)
Please do a proper regression test and report REAL regressions.
First we need to do a regression hunt in gcc trunk for the new
stack test case failures. It is impossible to
--- Comment #36 from howarth at nitro dot med dot uc dot edu 2010-09-01
17:45 ---
FAIL: gcc.c-torture/execute/builtins/sprintf-chk.c compilation, -Os (internal
compiler error)
UNRESOLVED: gcc.c-torture/execute/builtins/sprintf-chk.c execution, -Os
FAIL: gcc.dg/stack-usage-1.c
--- Comment #37 from howarth at nitro dot med dot uc dot edu 2010-09-01
18:01 ---
The nest.c failure doesn't occur with r163659 and gcc-pr45234-2.patch. So
PR36502v4.patch solves PR36502 but introduces a single regression in the gcc
testsuite.
--
--- Comment #38 from hjl dot tools at gmail dot com 2010-09-01 19:49
---
Created an attachment (id=21649)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21649action=view)
A patrch
Darwin needs 128bit stack boundary for all function calls. This
patch should fix nest.c.
--
--- Comment #39 from hjl dot tools at gmail dot com 2010-09-01 20:03
---
Created an attachment (id=21650)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21650action=view)
Another patch
Darwin needs 128bit stack alignment for -pg.
--
hjl dot tools at gmail dot com changed:
--- Comment #40 from howarth at nitro dot med dot uc dot edu 2010-09-01
21:28 ---
(In reply to comment #39)
Created an attachment (id=21650)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21650action=view) [edit]
Another patch
Darwin needs 128bit stack alignment for -pg.
--- Comment #11 from howarth at nitro dot med dot uc dot edu 2010-08-31
07:04 ---
Using the proposed PR36502v2.patch, which eliminates any attempts to change the
default stack boundary handling in gcc trunk, I find the following regressions
at -m32 on x86_64-apple-darwin10...
FAIL:
--- Comment #12 from howarth at nitro dot med dot uc dot edu 2010-08-31
12:57 ---
Testresults for PR36502v2.patch at
http://gcc.gnu.org/ml/gcc-testresults/2010-08/msg03098.html. It appears that
the only additional failures triggered by this patch are listed in comment 11.
--
--- Comment #13 from hjl dot tools at gmail dot com 2010-08-31 13:48
---
(In reply to comment #9)
Created an attachment (id=21599)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21599action=view) [edit]
rely on PREFERRED_STACK_BOUNDARY_DEFAULT for intel darwin
You should keep
--- Comment #14 from hjl dot tools at gmail dot com 2010-08-31 13:54
---
(In reply to comment #12)
Testresults for PR36502v2.patch at
http://gcc.gnu.org/ml/gcc-testresults/2010-08/msg03098.html. It appears that
the only additional failures triggered by this patch are listed in
--- Comment #15 from howarth at nitro dot med dot uc dot edu 2010-08-31
15:39 ---
Created an attachment (id=21607)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21607action=view)
rely on PREFERRED_STACK_BOUNDARY_DEFAULT and MAIN_STACK_BOUNDARY
--
--- Comment #16 from howarth at nitro dot med dot uc dot edu 2010-08-31
15:48 ---
(In reply to comment #14)
Please try this patch:
http://gcc.gnu.org/ml/gcc-patches/2010-08/msg01916.html
With the patch above and PR36502v3.patch, I still get...
FAIL:
--- Comment #17 from howarth at nitro dot med dot uc dot edu 2010-08-31
16:30 ---
With http://gcc.gnu.org/ml/gcc-patches/2010-08/msg01916.html and the patch in
comment 15, I am also still failing at -m32 on x86_64-apple-darwin10...
FAIL: gcc.dg/nest.c execution test
FAIL:
--- Comment #18 from howarth at nitro dot med dot uc dot edu 2010-08-31
16:36 ---
Created an attachment (id=21608)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21608action=view)
assembly from failing gcc.dg/nest.c execution test at -m32 on
x86_64-apple-darwin10
Generated with...
--- Comment #19 from howarth at nitro dot med dot uc dot edu 2010-08-31
16:41 ---
(In reply to comment #18)
In gdb, the failing nest.exe exection shows...
Starting program: /Users/howarth/stack_align_bug2/nest.exe
Reading symbols for shared libraries ++. done
Program received
--- Comment #20 from howarth at nitro dot med dot uc dot edu 2010-08-31
16:42 ---
Created an attachment (id=21609)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21609action=view)
assembly from failing gcc.dg/torture/stackalign/alloca-4.c -O1 execution test
at -m32 on
--- Comment #21 from howarth at nitro dot med dot uc dot edu 2010-08-31
16:46 ---
(In reply to comment #20)
Created alloca-4.s with...
/sw/src/fink.build/gcc46-4.6.0-1000/darwin_objdir/gcc/xgcc
-B/sw/src/fink.build/gcc46-4.6.0-1000/darwin_objdir/gcc/
--- Comment #22 from howarth at nitro dot med dot uc dot edu 2010-08-31
19:50 ---
Created an attachment (id=21615)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21615action=view)
assembly from stock build of gcc.dg/nest.c execution test at -m32 on
x86_64-apple-darwin10
--
--- Comment #23 from howarth at nitro dot med dot uc dot edu 2010-08-31
19:55 ---
Created an attachment (id=21616)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21616action=view)
assembly from stock build of gcc.dg/torture/stackalign/alloca-4.c -O1
execution test at -m32 on
--- Comment #24 from howarth at nitro dot med dot uc dot edu 2010-08-31
19:57 ---
Created an attachment (id=21617)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21617action=view)
diff between nest.s.stock and nest.s show alignment changes in failing test
case
--
--- Comment #25 from howarth at nitro dot med dot uc dot edu 2010-08-31
19:59 ---
Created an attachment (id=21618)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21618action=view)
diff between alloca-4.s.stock and alloca-4.s show alignment changes in failing
test case
--
--- Comment #26 from hjl dot tools at gmail dot com 2010-08-31 20:30
---
(In reply to comment #15)
Created an attachment (id=21607)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21607action=view) [edit]
rely on PREFERRED_STACK_BOUNDARY_DEFAULT and MAIN_STACK_BOUNDARY
This is
--- Comment #27 from mrs at gcc dot gnu dot org 2010-09-01 00:25 ---
I don't think MAIN_STACK_BOUNDARY needs to be set, nor will it help to set it.
The alignment is set up by the crt runtime, and just to call to main, the
alignment of the stack must be 128, before we issue the call.
--- Comment #28 from hjl dot tools at gmail dot com 2010-09-01 01:08
---
(In reply to comment #27)
I don't think MAIN_STACK_BOUNDARY needs to be set, nor will it help to set
it.
The alignment is set up by the crt runtime, and just to call to main, the
alignment of the stack must
--- Comment #29 from howarth at nitro dot med dot uc dot edu 2010-09-01
04:02 ---
(In reply to comment #28)
If you don't set MAIN_STACK_BOUNDARY to 128, gcc may
align stack in main.
I am seeing some instability in the testsuite results when I don't redefine
MAIN_STACK_BOUNDARY to
--- Comment #30 from howarth at nitro dot med dot uc dot edu 2010-09-01
05:34 ---
With only gcc-pr45234-2.patch at r163712 , I am seeing the following
regressions...
FAIL: gcc.c-torture/execute/builtins/sprintf-chk.c compilation, -Os (internal
compiler error)
FAIL:
--- Comment #3 from rth at gcc dot gnu dot org 2010-08-30 18:20 ---
It's caused by this:
config/i386/darwin.h:#define STACK_BOUNDARY 128
I think you want to delete that and allow STACK_BOUNDARY to be defined
by i386.h to UNITS_PER_WORD. Non-leaf functions should be handled by
--- Comment #4 from howarth at nitro dot med dot uc dot edu 2010-08-30
22:44 ---
(In reply to comment #3)
Richard,
The following patch fails to bootstrap...
Index: i386/darwin.h
===
--- i386/darwin.h
--- Comment #5 from howarth at nitro dot med dot uc dot edu 2010-08-31
01:23 ---
Created an attachment (id=21598)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21598action=view)
patch to properly use MIN_STACK_BOUNDARY on intel darwin.
--
--- Comment #6 from howarth at nitro dot med dot uc dot edu 2010-08-31
01:25 ---
For the testcase compiled with...
gcc-4 -m32 -O -fomit-frame-pointer -fno-pic -S add.c
...current gcc trunk gives...
_f:
LFB0:
subl$12, %esp
LCFI0:
addl$1, _a
addl
--- Comment #7 from hjl dot tools at gmail dot com 2010-08-31 02:18 ---
Please check how MIN_STACK_BOUNDARY is used in i386.c. If you
define it to MIN_STACK_BOUNDARY, -mstackrealign won't work
correctly.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36502
--- Comment #8 from hjl dot tools at gmail dot com 2010-08-31 02:21 ---
(In reply to comment #7)
Please check how MIN_STACK_BOUNDARY is used in i386.c. If you
define it to MIN_STACK_BOUNDARY, -mstackrealign won't work
correctly.
I meant -mstackrealign won't work correctly if
--- Comment #9 from howarth at nitro dot med dot uc dot edu 2010-08-31
05:13 ---
Created an attachment (id=21599)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21599action=view)
rely on PREFERRED_STACK_BOUNDARY_DEFAULT for intel darwin
--
howarth at nitro dot med dot uc dot
--- Comment #10 from howarth at nitro dot med dot uc dot edu 2010-08-31
05:15 ---
On reflection, we should be able to dump all the stack related setting in
darwin.h and use the current setting of PREFERRED_STACK_BOUNDARY_DEFAULT 128
since current gcc now defaults on SSE2.
--
--- Comment #2 from fxcoudert at gcc dot gnu dot org 2010-01-02 20:52
---
Still present as of rev. 155544.
--
fxcoudert at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #1 from pinskia at gcc dot gnu dot org 2008-06-11 23:43 ---
This is due to alignment requirement interacting weirdly with the back-end.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36502
49 matches
Mail list logo
