--- Comment #16 from mmitchel at gcc dot gnu dot org 2007-06-29 18:00
---
This problem is reported fixed; may we close this PR?
--
mmitchel at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #17 from pcarlini at suse dot de 2007-06-29 18:09 ---
Great, let's close it, then.
--
pcarlini at suse dot de changed:
What|Removed |Added
--- Comment #15 from dominiq at lps dot ens dot fr 2007-03-22 16:22 ---
Note that the drawback of the optimization replacing sin+cos by cexpi on PPC
Darwin has been dissected in PR31249.
In comment #16, I proposed a patch. Before applying it, it would be nice to
test if the other
--- Comment #14 from dominiq at lps dot ens dot fr 2007-03-17 00:26 ---
Thanks to Richard Guenther, the bug seems to be fixed (hopefully in the
20070316 snapshot):
see PR31161 for details.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30980
--- Comment #10 from dominiq at lps dot ens dot fr 2007-03-13 12:56 ---
The problem seems to come from a broken/unavailable __builtin_cexpi,
see PR31161.
My understanding is that __builtin_cexpi and __builtin_sincos are twin
objects(?). Now I see in gcc/tree-ssa-math-opts.c:
...
--- Comment #11 from rguenth at gcc dot gnu dot org 2007-03-13 13:22
---
Because if TARGET_C99_FUNCTIONS is set we can expand __builtin_cexpi by using
cexp(). The ICE you get in PR30969 shows that TARGET_C99_FUNCTIONS is _not_
set:
/* We can expand via the C99 cexp function.
--- Comment #12 from rguenth at gcc dot gnu dot org 2007-03-13 13:26
---
Now,
2007-02-05 Roger Sayle [EMAIL PROTECTED]
* fold-const.c (fold_unary) REAL_PART: Test for availability of
BUILT_IN_COS before simplifying REAL_PART(CEXPI)) to COS.
IMAG_PART:
--- Comment #13 from dominiq at lps dot ens dot fr 2007-03-13 14:48 ---
The ICE you get in PR30969 shows that TARGET_C99_FUNCTIONS is _not_ set:
Is there a way to test it once the building directory is gone? In
gcc/config/rs6000/darwin.h I read:
...
/* Old versions of Mac OS/Darwin
--- Comment #9 from dominiq at lps dot ens dot fr 2007-03-11 12:19 ---
I confirm that the bug does not show with gcc. Is theis related to the
following failures in the regtest?
FAIL: gcc.dg/builtins-59.c scan-tree-dump __builtin_cexpi
FAIL: gcc.dg/builtins-59.c scan-tree-dump-not
--- Comment #2 from rguenth at gcc dot gnu dot org 2007-02-27 13:27 ---
As I cannot reproduce this on any linux target I guess the following change did
it:
+2007-02-16 Geoffrey Keating [EMAIL PROTECTED]
+
+ * config/darwin.h (LINK_SPEC): Always pass -macosx_version_min
+
--- Comment #3 from pcarlini at suse dot de 2007-02-27 13:30 ---
Yes, I can confirm it cannot be reproduced on linux targets. Anyway, we can
easily nail down between 122051 and 122067 (from testresults).
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30980
--- Comment #4 from pcarlini at suse dot de 2007-02-27 13:39 ---
Note, however, that per Kaveh' reports, around the same dates (between
2007-02-14 and 2007-02-19) it started failing also on sparc-sun-solaris2.10...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30980
--- Comment #5 from pcarlini at suse dot de 2007-02-27 13:41 ---
I'm adding Eric too, maybe he wants to investigate the sparc version of the
issue.
--
pcarlini at suse dot de changed:
What|Removed |Added
--- Comment #6 from pinskia at gcc dot gnu dot org 2007-02-27 15:42 ---
PR 30969 is related. The reduced fortran looks like the reduced C++.
--
pinskia at gcc dot gnu dot org changed:
What|Removed |Added
--- Comment #7 from pinskia at gcc dot gnu dot org 2007-02-27 15:47 ---
I'm adding Eric too, maybe he wants to investigate the sparc version of the
issue.
The patch which I was worried about causing a regression with respect of
calling complex functions is:
2007-02-18 Sandra
--- Comment #8 from geoffk at gcc dot gnu dot org 2007-02-27 17:25 ---
I'm confident that my patch could not possibly have affected targets other than
Darwin, and should not have significantly affected code generation even there.
Complex arithmetic testcases will fail for other reasons
16 matches
Mail list logo
