I see the following ICE with current trunk and with trunk from
2007-09-16. It doesn't happen with 2007-09-02.
(sid)29744:[EMAIL PROTECTED]: ~] /usr/lib/gcc-snapshot/bin/gcc -Wall -O2
-ftree-vectorize -c allegro-mixer.c
allegro-mixer.c: In function '_mix_some_samples':
allegro-mixer.c:19:
--- Comment #1 from tbm at cyrius dot com 2007-10-21 06:03 ---
Created an attachment (id=14379)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14379action=view)
preprocessed source
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33846
--- Comment #2 from tbm at cyrius dot com 2007-10-21 06:05 ---
/* Testcase by Martin Michlmayr [EMAIL PROTECTED] */
/* { dg-do compile } */
/* { dg-options -O2 -ftree-vectorize } */
int clamp_val (int i)
{
return ~i 31;
}
void _mix_some_samples (long buf, int *mix_buffer, int
--- Comment #3 from tbm at cyrius dot com 2007-10-21 06:05 ---
Starting program: /home/tbm/tmp/gcc/gcc-4.3-20071020-r129504/gcc/cc1 -O2
-ftree-vectorize ~/allegro-mixer.c
clamp_val _mix_some_samples
Analyzing compilation unit
Performing interprocedural optimizations
visibility
--- Comment #4 from tbm at cyrius dot com 2007-10-21 06:25 ---
Here's another one (requires -O3):
/* Testcase by Martin Michlmayr [EMAIL PROTECTED] */
extern int sscanf (__const char *__restrict __s,
__const char *__restrict __format, ...);
unsigned char
--- Comment #4 from dorit at gcc dot gnu dot org 2007-10-21 06:39 ---
I was able to reproduce this on i386-linux.
Looks like it's related to PLUS_EXPR vs. POINTER_PLUS_EXPR. The folowing patch
fixes this testcase:
Index: tree-vect-analyze.c
The following code gives:
t.f90:41.12:
call sub ( shape_line_P1 )
1
Error: Type/rank mismatch in argument 'sub1' at (1)
I tested ifort, lahey, g95 without any problem.
module int_m
interface shape_line_P1
module procedure shape_line_P1, shape_line_P1_2
end interface
--- Comment #5 from dorit at gcc dot gnu dot org 2007-10-21 07:14 ---
This patch fixes it:
Index: tree-vect-transform.c
===
*** tree-vect-transform.c (revision 129521)
--- tree-vect-transform.c (working copy)
--- Comment #2 from pault at gcc dot gnu dot org 2007-10-21 07:30 ---
Given Steve's comment, this clearly should be confirmed!
Paul
--
pault at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #3 from dorit at gcc dot gnu dot org 2007-10-21 08:07 ---
The proposed fix/work-around for PR33834 also happens to fix this PR. But the
real problem is that we try to access a NULL argument (operand 2 of a CALL_EXPR
may be NULL). So we should probably at least add something
--- Comment #5 from irar at il dot ibm dot com 2007-10-21 08:45 ---
(In reply to comment #4)
Created an attachment (id=14370)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14370action=view) [edit]
Vectorization dump file
Thanks!
The vectorizer fails in transformation phase in
--- Comment #4 from burnus at gcc dot gnu dot org 2007-10-21 09:30 ---
Several intrinsic functions come as specific and as generic functions. For
instance
COS as generic function takes as argument REAL(k) and COMPLEX(k), k=4,8,10,16
and returns the same type and kind as it got
[ Forwarded from http://bugs.debian.org/441633 ]
From: Sergei Golovan [EMAIL PROTECTED]
Current gcc 4.2 fails to build erlang package on mips and mipsel
architectures. It works on all other architectures.
Linker error messages are the following (the same for both mips and mipsel):
--- Comment #1 from tbm at cyrius dot com 2007-10-21 09:46 ---
The problem is that gcc-4.2 generates a bne instruction that references a
label that doesn't exist. This happens with -O1 or higher. It doesn't
happen with gcc 4.1 or current trunk.
If you compile the attached
--- Comment #2 from tbm at cyrius dot com 2007-10-21 09:46 ---
Created an attachment (id=14380)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14380action=view)
preprocessed source
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33848
--- Comment #13 from jkherciueh at gmx dot net 2007-10-21 09:49 ---
(In reply to comment #12)
In order to have a fully conforming implementation, we need something which
works for *any* source range and *any* target range. Are you willing to
work on that? Would be mainline-only of
--- Comment #3 from steven at gcc dot gnu dot org 2007-10-21 10:00 ---
The way for you to narrow down the bug is this:
1. Compile with -daAP
2. Look in the .s file which instruction references the missing label. There
should be a LABEL_REF with a number.
3. Grep for code_label.*number
--- Comment #1 from burnus at gcc dot gnu dot org 2007-10-21 10:07 ---
(And the program is accepted using the patch from bug 33162 comment 2, however,
I think the patch is the wrong to solution for this PR and bug 33162.)
NAG f95 also gives the following error:
Error: line 41: Actual
--- Comment #4 from tbm at cyrius dot com 2007-10-21 10:15 ---
(In reply to comment #3)
This will help you find the pass during/after which the label disappears.
It's there from 104r (expand) to 139r (postreload) but not anymore in 141r
(flow2).
--
--- Comment #2 from burnus at gcc dot gnu dot org 2007-10-21 10:33 ---
I cannot read. You used:
interface shape_line_P1
module procedure shape_line_P1, shape_line_P1_2
end interface
Thus shape_line_P1 is both a specific subroutine and a generic subroutine,
however, the
The following error message is misleading:
Error: GENERIC non-INTRINSIC procedure '%s' is not allowed as an actual
argument
The problem is that any GENERIC procedure - intrinsic or not - is not allowed
as actual argument. (For a dummy procedure[...], the associated actual
argument shall be the
--- Comment #5 from tbm at cyrius dot com 2007-10-21 10:44 ---
Trying to reduce...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33848
--- Comment #5 from burnus at gcc dot gnu dot org 2007-10-21 10:44 ---
This patch: Allows the test case in comment #1 to compile and run correctly
with
no regressions in the test suite.
It also wrongly allows the test case in the bugreport PR 33847 to compile. Thus
we should add a
--- Comment #4 from irar at il dot ibm dot com 2007-10-21 11:02 ---
The problem is with vector shift with scalar shift argument.
For the code created by the vectorizer:
vect_var_.49_103 = ~vect_var_.47_101;
vect_var_.50_105 = vect_var_.49_103 31;
(ashiftrt:V4SI (not:V4SI
--- Comment #14 from pcarlini at suse dot de 2007-10-21 11:33 ---
(In reply to comment #13)
Sure thing.
Excellent.
I will need a little bit of time to familiarize myself with the current code.
Also, I will need a little bit of help in finding my way around. E.g., where
do
the
For
integer(4) :: p(4) = (/2,4,1,3/)
integer(4) :: q(4) = (/2,4,1,3/)
p(q) = (/(i, i = 1, 4)/)
gfortran generates an unneeded temporary, which is only needed for
integer(4) :: p(4) = (/2,4,1,3/)
p(p) = (/(i, i = 1, 4)/)
Cf. PR 33749
--
Summary: Unneeded temporary
--- Comment #1 from burnus at gcc dot gnu dot org 2007-10-21 12:40 ---
See also http://gcc.gnu.org/ml/fortran/2007-10/msg00298.html
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33850
--- Comment #6 from irar at il dot ibm dot com 2007-10-21 12:52 ---
The solution can be just not check if the vectorization is worthwhile during
the transformation. The decision whether to vectorize or not should be made
during the analysis anyway.
The vectorization factor can get
libtool adds all sorts of wonderful flags to gcc, whose purpose is unclear.
Nonetheless, the following works on gcc-4.2.1 but fails on gcc version 4.3.0
20071012 (experimental) (GCC) on the same box, a Mac Pro running OS X 10.4:
g++ -m64 -dynamiclib -single_module ${wl}-flat_namespace
--- Comment #1 from benoit dot hudson at gmail dot com 2007-10-21 15:15
---
I should specify: I can *compile* but I can't *link*.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33851
--- Comment #2 from oliver dot kellogg at eads dot com 2007-10-21 15:44
---
Also happens with 4.3.0-20071021 r129524.
Does not happen with -O (optimize).
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33788
--- Comment #6 from jvdelisle at gcc dot gnu dot org 2007-10-21 16:39
---
Created an attachment (id=14381)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14381action=view)
Patch to add checks for double specifics
This patch adds checks for double specific intrinsics. I added
--- Comment #2 from pinskia at gcc dot gnu dot org 2007-10-21 16:45 ---
Well, libtool needs to check the gcc version. The issue is that in 4.3, we
call ld directly instead of libtool (Apple's libtool and not GNU libtool). So
libtool needs to use
--
pinskia at gcc dot gnu dot org changed:
What|Removed |Added
CC||pinskia at gcc dot gnu dot
|
--- Comment #6 from pault at gcc dot gnu dot org 2007-10-21 18:10 ---
Subject: Bug 33749
Author: pault
Date: Sun Oct 21 18:10:00 2007
New Revision: 129539
URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=129539
Log:
2007-10-21 Paul Thomas [EMAIL PROTECTED]
PR
--- Comment #3 from benoit dot hudson at gmail dot com 2007-10-21 18:23
---
(In reply to comment #2)
Well, libtool needs to check the gcc version.
I fully understand that morally, it's libtool's fault for releasing a program
that didn't correctly divine that gcc 4.3 would change the
--- Comment #15 from Tobias dot Schlueter at physik dot uni-muenchen dot de
2007-10-21 20:13 ---
Subject: Re: Diagnose different string lengths in array
constructors at run time
dominiq at lps dot ens dot fr wrote:
--- Comment #14 from dominiq at lps dot ens dot fr 2007-10-21
--- Comment #16 from tobi at gcc dot gnu dot org 2007-10-21 21:18 ---
Re this testcase:
(In reply to comment #5)
program array_char
implicit none
character (len=2) :: x, y
character (len=2) :: z(2)
x = a
y = cd
z = (/y(1:len(trim(y))), x(1:len(trim(x)))/) ! causes segfault
--- Comment #7 from jvdelisle at gcc dot gnu dot org 2007-10-21 21:32
---
Created an attachment (id=14382)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14382action=view)
Preliminary argument checking with intrinsics
This draft patch catches the latest invalid case provided in
--- Comment #3 from hjl at lucon dot org 2007-10-21 21:41 ---
I believe all those header files should be restored unless we have a very
good reason to break libstdc++ API.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33831
--- Comment #1 from hjl at lucon dot org 2007-10-21 21:45 ---
I believe we should add those deleted ext/xxx header files with
#warning __FILE__ is deprecated
#include backward/xxx
if we want to move ext/xxx to backward/xxx. It is never a good idea to break
the existing C++
Here is incorrect text formating with using cout() with setw() in UNICODE
(UTF-8)
Two text: first in English - correct formating, second - russian in UTF-8 -
incorrect.
#include iostream
#include iomanip
int main() {
std::cout.setf(std::ios::left);
std::cout.fill('_');
--- Comment #1 from jvdelisle at gcc dot gnu dot org 2007-10-21 22:56
---
Is the issue here only the wording of the error message or is the checking
incorrect as well?
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33849
--- Comment #4 from danglin at gcc dot gnu dot org 2007-10-22 00:43 ---
The actual code that is wrong is the following:
0x00010670 main+176: ldw 10(r20),r19
0x00010674 main+180: uaddcm r0,r23,r21
0x00010678 main+184: uaddcm r0,r24,r22
0x0001067c main+188: depw,z r19,30,31,r19
--- Comment #5 from dave at hiauly1 dot hia dot nrc dot ca 2007-10-22
01:30 ---
Subject: Re: gcc.c-torture/execute/longlong.c
execution at -O3
Things appear to go wrong in the greg pass:
(insn 73 133 70 4
/home/dave/gnu/gcc-4.3/gcc/gcc/testsuite/gcc.c-torture/execute
--
pinskia at gcc dot gnu dot org changed:
What|Removed |Added
CC||pinskia at gcc dot gnu dot
|
--
jakub at gcc dot gnu dot org changed:
What|Removed |Added
AssignedTo|unassigned at gcc dot gnu |jakub at gcc dot gnu dot org
|dot org
--- Comment #6 from dorit at gcc dot gnu dot org 2007-10-22 04:28 ---
I'm testing this patch. It fixes the two testcases, while allowing the first
testcase to get vectorized. (the last bit in the patch is the fix for PR33835):
Index: tree-vect-analyze.c
--- Comment #4 from dorit at gcc dot gnu dot org 2007-10-22 04:37 ---
I'm testing a patch that would fix both this PR and PR33834 (posted it under
the PR33834 entry). By the way, this testcase does not get vectorized with
current mainline (an Oct21 snapshot) because the call to cos is
49 matches
Mail list logo
