[Bug target/34709] New: [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[Bug target/34709] New: [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[hjl at lucon dot org]

This patch

http://gcc.gnu.org/ml/gcc-patches/2008-01/msg00191.html

miscompiled 481.wrf in SPEC CPU 2006 on Linux/Intel64 with -O2 -ffast-math.
The resulting 481.wrf segfaulted.


-- 
           Summary: [4.3 regression]: revision 131342 miscompiled 481.wrf on
                    Linux/Intel64
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: hjl at lucon dot org
GCC target triplet: x86_64-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2008-01-08 10:21 -------
No idea if Uros has access to spec, so maybe you can quote the snippet where
rsqrtss is used from the assembly of 481.wrf?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[ubizjak at gmail dot com]

------- Comment #2 from ubizjak at gmail dot com  2008-01-08 11:23 -------
(In reply to comment #1)
> No idea if Uros has access to spec, so maybe you can quote the snippet where
> rsqrtss is used from the assembly of 481.wrf?

Unfortunatelly no... does this patch help:

Index: i386.c
===================================================================
--- i386.c      (revision 131392)
+++ i386.c      (working copy)
@@ -21449,7 +21449,7 @@ static tree
 ix86_builtin_reciprocal (unsigned int fn, bool md_fn,
                         bool sqrt ATTRIBUTE_UNUSED)
 {
-  if (! (TARGET_SSE_MATH && !optimize_size
+  if (! (TARGET_SSE_MATH && TARGET_RECIP && !optimize_size
         && flag_finite_math_only && !flag_trapping_math
         && flag_unsafe_math_optimizations))
     return NULL_TREE;

BTW: What happens if -mrecip is used to compile 481.wrf?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[hjl at lucon dot org]

------- Comment #3 from hjl at lucon dot org  2008-01-08 18:01 -------
(In reply to comment #2)
> (In reply to comment #1)
> > No idea if Uros has access to spec, so maybe you can quote the snippet where
> > rsqrtss is used from the assembly of 481.wrf?
> 
> Unfortunatelly no... does this patch help:
> 
> Index: i386.c
> ===================================================================
> --- i386.c      (revision 131392)
> +++ i386.c      (working copy)
> @@ -21449,7 +21449,7 @@ static tree
>  ix86_builtin_reciprocal (unsigned int fn, bool md_fn,
>                          bool sqrt ATTRIBUTE_UNUSED)
>  {
> -  if (! (TARGET_SSE_MATH && !optimize_size
> +  if (! (TARGET_SSE_MATH && TARGET_RECIP && !optimize_size
>          && flag_finite_math_only && !flag_trapping_math
>          && flag_unsafe_math_optimizations))
>      return NULL_TREE;

It works.

> BTW: What happens if -mrecip is used to compile 481.wrf?

I will give it a try.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[hjl at lucon dot org]

------- Comment #4 from hjl at lucon dot org  2008-01-08 18:24 -------
(In reply to comment #2)
> BTW: What happens if -mrecip is used to compile 481.wrf?
> 

"-O2 -mrecip -ffast-math" miscompiles 481.wrf on Linux/Intel64.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[ubizjak at gmail dot com]

------- Comment #5 from ubizjak at gmail dot com  2008-01-08 18:36 -------
(In reply to comment #3)

> It works.

Ok, then we have the same issue with rsqrtss as with rcpss in PR target/34702.
I'll disable -mrecip as default for -ffast-math.


-- 

ubizjak at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |ubizjak at gmail dot com
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-01-08 18:36:06
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[uros at gcc dot gnu dot org]

------- Comment #6 from uros at gcc dot gnu dot org  2008-01-08 19:23 -------
Subject: Bug 34709

Author: uros
Date: Tue Jan  8 19:22:46 2008
New Revision: 131403

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=131403
Log:
        PR target/34709
        Revert:

        2008-01-05  Uros Bizjak  <ubizjak@gmail.com>
        * config/i386/i386.c (ix86_builtin_reciprocal): Remove check
        for TARGET_RECIP.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[ubizjak at gmail dot com]

------- Comment #7 from ubizjak at gmail dot com  2008-01-08 19:27 -------
Fixed.


-- 

ubizjak at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709

[Bug target/34709] [4.3 regression]: revision 131342 miscompiled 481.wrf on Linux/Intel64

[Joey dot ye at intel dot com]

------- Comment #8 from Joey dot ye at intel dot com  2008-01-17 10:11 -------
A small case and patch are available at
http://gcc.gnu.org/ml/gcc-patches/2008-01/msg00747.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34709