--
What|Removed |Added
Target Milestone|--- |4.0.0
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19609
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
08:40 ---
I bootstraped a 20050123 shapshot with the d-19609
patch attached and lags_complex_divide_method=1 on
ia64-unknown-linux-gnu, and I got this bootstrap failure
in libgfortran:
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
08:48 ---
Same thing:
$ cat cdivide.c
#include stdio.h
#include math.h
#include complex.h
int main()
{
float complex a,b,c;
c = a/b;
}
$ ~/gcc-bin/gcc/xgcc -fdump-rtl-all-all -fdump-tree-all -B
--- Additional Comments From pcarlini at suse dot de 2005-01-25 10:49
---
Hi, everything looks fine here (20050124 on x86-linux), double checked the
formulas, bootstrapped c, c++, tested a few other divisions besides the
testcase (now ok), and cannot reproduce the last dump of Thomas.
--- Additional Comments From pcarlini at suse dot de 2005-01-25 12:39
---
Thomas, before attaching stuff to the PR, please double check your setup: on
x86, at least, the problem is definitely fixed and I cannot reproduce your
dumps. I'm currently testing on x86_64.
--
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
14:50 ---
(In reply to comment #13)
Thomas, before attaching stuff to the PR, please double check your setup: on
x86, at least, the problem is definitely fixed and I cannot reproduce your
dumps. I'm currently
--- Additional Comments From pcarlini at suse dot de 2005-01-25 15:01
---
I just redid the following commands, in order, on an
ia64-unknown-linux-gnu machine:
With the most recent snapshot:
$ bzcat gcc-4.0-20050123.tar.bz2 | tar -xvf -
... with the same bootstrap failure.
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
15:12 ---
(In reply to comment #15)
This is an ia64 specific issue, therefore: testing on x86 and
x86-64 is ok, and definitely no bootstrap failures.
Interesting. How did you do the testing? What version
--- Additional Comments From pcarlini at suse dot de 2005-01-25 15:20
---
Interesting. How did you do the testing? What version did you test
against? How did you test? Maybe something is broken in 20050123
that got fixed later, which caused this.
Nothing special, just current
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
15:40 ---
(In reply to comment #17)
Therefore, how did you get
the last dumps?
I used the xgcc from the build subdirectoy:
$ ~/gcc-bin/gcc/xgcc -fdump-rtl-all-all -fdump-tree-all -B ~/gcc-bin/gcc/
cdivide.c
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
15:57 ---
Paolo,
could you upload the .t14.oplower dump that you got from your working
version with the d-19609 patch?
Thomas
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19609
--- Additional Comments From pcarlini at suse dot de 2005-01-25 16:01
---
Is in Comment #11, and looks fine.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19609
--- Additional Comments From pcarlini at suse dot de 2005-01-25 15:51
---
Ok, sorry, didnt' pay attention to that detail (while working on the library,
usually we install the compiler... ;)
Anyway, I'm going to build and test on ia64 too.
--
--- Additional Comments From pcarlini at suse dot de 2005-01-25 16:13
---
... indeed, IMH-not-compiler-hacker, opinion, t14 should be the same on x86
and ia64...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19609
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
16:43 ---
(In reply to comment #22)
... indeed, IMH-not-compiler-hacker, opinion, t14 should be the same on x86
and ia64...
Quite.
Could you maybe run a C-only bootstrap with the 20050123 snapshot,
modified
--- Additional Comments From Thomas dot Koenig at online dot de 2005-01-25
17:00 ---
#ifdef EGG_ON_FACE_MODE
Sorry to have kept everybody from doing something more worthwile.
The problem was that I hadn't noticed that the patch for
PR 19468 wasn't yet in the 20050123 snapshot which I
--- Additional Comments From pcarlini at suse dot de 2005-01-25 17:06
---
OK ;) In Italy we say something like anything is ok, if ends ok ;)
Anyway, in the meanwhile an ia64-linux bootstrap completed fine here. I'm
running
the testsuite now.
--
--- Additional Comments From cvs-commit at gcc dot gnu dot org 2005-01-25
18:19 ---
Subject: Bug 19609
CVSROOT:/cvs/gcc
Module name:gcc
Changes by: [EMAIL PROTECTED] 2005-01-25 18:19:10
Modified files:
gcc: ChangeLog tree-complex.c
Log
--- Additional Comments From rth at gcc dot gnu dot org 2005-01-25 18:26
---
Fixed.
--
What|Removed |Added
Status|ASSIGNED|RESOLVED
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-01-24
21:49 ---
Confirmed.
--
What|Removed |Added
Status|UNCONFIRMED |NEW
--
What|Removed |Added
AssignedTo|unassigned at gcc dot gnu |rth at gcc dot gnu dot org
|dot org |
Status|NEW
--- Additional Comments From pcarlini at suse dot de 2005-01-24 23:01
---
Hi Richard. It seems to me that the problem is in the implemented formula.
Looking at Goldberg, when |d| = |c| in its paper,that is |bi| |br| in the
Gcc code, we should have, at the end of the calculation:
--- Additional Comments From rth at gcc dot gnu dot org 2005-01-24 23:23
---
Yes, I see that. My eyes crossed while transcribing the algorithm to trees,
and I failed to notice that a computation is (b - (a * ratio)) in one branch
and ((a * ratio) - b) in the other. And so incorrectly
--- Additional Comments From pcarlini at suse dot de 2005-01-24 23:33
---
Yes, I see that. My eyes crossed while transcribing the algorithm to trees,
and I failed to notice that a computation is (b - (a * ratio)) in one branch
and ((a * ratio) - b) in the other. And so incorrectly
--- Additional Comments From pcarlini at suse dot de 2005-01-25 01:53
---
Great! In case you like me to double check a bit, only tomorrow, given my
timezone... Thanks again!
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19609
--- Additional Comments From rth at gcc dot gnu dot org 2005-01-25 06:28
---
Please do. I seem to have screwed up the ia64 build as well (user error
on my side), and won't see results until tommorow gmt-8.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19609
26 matches
Mail list logo
