[Bug target/47522] [4.4/4.5/4.6 Regression] wrong code at -O3 -ffast-math

2011-01-31 Thread rguenth at gcc dot gnu.org 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47522

Richard Guenther rguenth at gcc dot gnu.org changed:

   What|Removed |Added

   Keywords||wrong-code
  Component|middle-end  |target
 Blocks||44183
   Target Milestone|--- |4.4.6

--- Comment #2 from Richard Guenther rguenth at gcc dot gnu.org 2011-01-31 
10:03:25 UTC ---
Target piece, if there is one.  PR44183 for the vectorizer piece, if there is
one.

[Bug target/47522] [4.4/4.5/4.6 Regression] wrong code at -O3 -ffast-math

2011-01-31 Thread Joost.VandeVondele at pci dot uzh.ch 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47522

--- Comment #3 from Joost VandeVondele Joost.VandeVondele at pci dot uzh.ch 
2011-01-31 10:16:24 UTC ---
(In reply to comment #2)
 Target piece, if there is one.  PR44183 for the vectorizer piece, if there is
 one.

I'm almost certain that this is a dup of PR44183. As far as I can judge, the
result of the calculation is indeed correct.

The annoying thing is that this makes valgrind's 'out of bounds checking' kind
of useless.

[Bug target/47522] [4.4/4.5/4.6 Regression] wrong code at -O3 -ffast-math

2011-01-31 Thread rguenther at suse dot de 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47522

--- Comment #4 from rguenther at suse dot de rguenther at suse dot de 
2011-01-31 10:47:53 UTC ---
On Mon, 31 Jan 2011, Joost.VandeVondele at pci dot uzh.ch wrote:

 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47522
 
 --- Comment #3 from Joost VandeVondele Joost.VandeVondele at pci dot uzh.ch 
 2011-01-31 10:16:24 UTC ---
 (In reply to comment #2)
  Target piece, if there is one.  PR44183 for the vectorizer piece, if there 
  is
  one.
 
 I'm almost certain that this is a dup of PR44183. As far as I can judge, the
 result of the calculation is indeed correct.
 
 The annoying thing is that this makes valgrind's 'out of bounds checking' kind
 of useless.

I think valgrind should simply special-case these kind of out of bounds
checks based on the instruction that was used.  This is an optimization
that is useful (even glibc memcpy routines do that, but they have
exception rules in valgrind ...)

Richard.

[Bug target/47522] [4.4/4.5/4.6 Regression] wrong code at -O3 -ffast-math

2011-01-31 Thread rguenth at gcc dot gnu.org 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47522

Richard Guenther rguenth at gcc dot gnu.org changed:

   What|Removed |Added

 Status|NEW |RESOLVED
 Resolution||WONTFIX

--- Comment #5 from Richard Guenther rguenth at gcc dot gnu.org 2011-01-31 
10:49:53 UTC ---
WONTFIX.  The transformation is legal.  valgrind sucks (maybe report this
to them).

[Bug target/47522] [4.4/4.5/4.6 Regression] wrong code at -O3 -ffast-math

2011-01-31 Thread Joost.VandeVondele at pci dot uzh.ch 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47522

--- Comment #6 from Joost VandeVondele Joost.VandeVondele at pci dot uzh.ch 
2011-01-31 11:17:46 UTC ---
(In reply to comment #5)
 valgrind sucks (maybe report this to them).

I think this is an unnecessary comment (all useful tools have bugs, or features
we don't like). The issue has been reported as

https://bugs.kde.org/show_bug.cgi?id=264936