[Bug middle-end/50083] New: [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[Bug middle-end/50083] New: [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

             Bug #: 50083
           Summary: [4.7 regression] All 32-bit fortran tests fail on
                    32-bit Solaris
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ro@gcc.gnu.org
                CC: ubizjak@gmail.com
              Host: i386-pc-solaris2.[89], sparc-sun-solaris2.[89]
            Target: i386-pc-solaris2.[89], sparc-sun-solaris2.[89]
             Build: i386-pc-solaris2.[89], sparc-sun-solaris2.[89]


Uros,

your patch

2011-08-12  Uros Bizjak  <ubizjak@gmail.com>

        * builtins.def (BUILT_IN_ICEIL{,F,L}, BUILT_IN_IFLOOR{,F,L}
        BUILT_IN_IRINT{,F,L}, BUILT_IN_IROUND{,F,L}: New builtin definitions.
[...]

completely broke 32-bit Fortran testing on Solaris 8 and 9, both SPARC and x86:

FAIL: gfortran.dg/coarray/atomic_1.f90 -fcoarray=single  -O2  (test for excess
errors)
Excess errors:
Undefined            first referenced
 symbol                  in file
__builtin_iroundf                  
/var/gcc/regression/trunk/8-gcc/build/sparc-sun-solaris2.8/./libgfortran/.libs/libgfortran.so
__builtin_iround                   
/var/gcc/regression/trunk/8-gcc/build/sparc-sun-solaris2.8/./libgfortran/.libs/libgfortran.so
ld: fatal: Symbol referencing errors. No output written to ./atomic_1.exe

Please fix.

  Rainer

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.7.0

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2011-08-19
         AssignedTo|unassigned at gcc dot       |ubizjak at gmail dot com
                   |gnu.org                     |
     Ever Confirmed|0                           |1

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2011-08-19 19:28:11 UTC ---
Eh, usual problems with this kind of optimizations ;)

I will look into it tomorrow...

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

--- Comment #2 from Uros Bizjak <ubizjak at gmail dot com> 2011-08-19 22:21:53 UTC ---
Created attachment 25058
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25058
Proposed patch to fallback to lround call

I can't trigger the problem on Linux, can you please test attached patch - or
eventually provide a testcase with all necessary compile flags.

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |WAITING

--- Comment #3 from Uros Bizjak <ubizjak at gmail dot com> 2011-08-23 18:16:49 UTC ---
(In reply to comment #2)
> Created attachment 25058 [details]
> Proposed patch to fallback to lround call
> 
> I can't trigger the problem on Linux, can you please test attached patch - or
> eventually provide a testcase with all necessary compile flags.

Clock starts ticking ...

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

--- Comment #4 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-08-24 10:43:58 UTC ---
> --- Comment #3 from Uros Bizjak <ubizjak at gmail dot com> 2011-08-23 18:16:49 UTC ---
> (In reply to comment #2)
>> Created attachment 25058 [details]
>> Proposed patch to fallback to lround call
>> 
>> I can't trigger the problem on Linux, can you please test attached patch - or
>> eventually provide a testcase with all necessary compile flags.
>
> Clock starts ticking ...

Sorry, I missed this.  The following test (reduced from
libgfortran/intrinsics/c99_functions.c) fails for me:

$ cat iround.i
float roundf (float x);

long int
lroundf (float x)
{
  return (long int) roundf (x);
}
$ ./cc1 -quiet -O2 -march=pentiumpro iround.i
$ cat iround.s
        .file   "iround.i"
        .text
        .align 16
        .globl  lroundf
        .type   lroundf, @function
lroundf:
.LFB0:
        .cfi_startproc
        jmp     __builtin_iroundf
        .cfi_endproc
.LFE0:
        .size   lroundf, .-lroundf
        .ident  "GCC: (GNU) 4.7.0 20110819 (experimental) [trunk revision
177911]"

I'll try the patch next thing.

    Rainer

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

--- Comment #5 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-08-24 10:47:56 UTC ---
At least rebuilding cc1 with the patched builtins.c doesn't make a
difference.

    Rainer

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

--- Comment #6 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-08-25 15:29:38 UTC ---
I can now also reproduce the failure on x86_64-unknown-linux-gnu:

cc1 -quiet -O2 -m32 iround.i -muclibc

With the default (-mglibc), it doesn't occur.

    Rainer

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |ASSIGNED

--- Comment #7 from Uros Bizjak <ubizjak at gmail dot com> 2011-08-25 16:48:38 UTC ---
Thanks, confirmed.

Looking into it.

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

--- Comment #8 from Uros Bizjak <ubizjak at gmail dot com> 2011-08-25 18:30:54 UTC ---
Too many trees to see the forest case ;)

We also have to protect conversion of round, rint and nearbyint with
TARGET_C99.

Obvious patch (also includes non-harmful typo in existing code):

--cut here--
Index: convert.c
===================================================================
--- convert.c    (revision 178071)
+++ convert.c    (working copy)
@@ -469,6 +469,9 @@ convert_to_integer (tree type, tree expr)
       break;

     CASE_FLT_FN (BUILT_IN_ROUND):
+      /* Only convert in ISO C99 mode.  */
+      if (!TARGET_C99_FUNCTIONS)
+        break;
       if (outprec < TYPE_PRECISION (integer_type_node)
           || (outprec == TYPE_PRECISION (integer_type_node)
           && !TYPE_UNSIGNED (type)))
@@ -487,11 +490,14 @@ convert_to_integer (tree type, tree expr)
         break;
       /* ... Fall through ...  */
     CASE_FLT_FN (BUILT_IN_RINT):
+      /* Only convert in ISO C99 mode.  */
+      if (!TARGET_C99_FUNCTIONS)
+        break;
       if (outprec < TYPE_PRECISION (integer_type_node)
           || (outprec == TYPE_PRECISION (integer_type_node)
           && !TYPE_UNSIGNED (type)))
         fn = mathfn_built_in (s_intype, BUILT_IN_IRINT);
-      else if (outprec < TYPE_PRECISION (long_integer_type_node)
+      else if (outprec == TYPE_PRECISION (long_integer_type_node)
            && !TYPE_UNSIGNED (type))
         fn = mathfn_built_in (s_intype, BUILT_IN_LRINT);
       else if (outprec == TYPE_PRECISION (long_long_integer_type_node)
--cut here--

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[uros at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

--- Comment #9 from uros at gcc dot gnu.org 2011-08-26 16:10:50 UTC ---
Author: uros
Date: Fri Aug 26 16:10:45 2011
New Revision: 178119

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=178119
Log:
    PR middle-end/50083
    * convert.c (convert_to_integer) <BUIT_IN_ROUND{,F,L}>: Convert
    only when TARGET_C99_FUNCTIONS.
    <BUILT_IN_NEARBYINT{,F,L}>: Ditto.
    <BUILT_IN_RINT{,F,L}>: Ditto.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/convert.c

[Bug middle-end/50083] [4.7 regression] All 32-bit fortran tests fail on 32-bit Solaris

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50083

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
                URL|                            |http://gcc.gnu.org/ml/gcc-p
                   |                            |atches/2011-08/msg02097.htm
                   |                            |l
         Resolution|                            |FIXED

--- Comment #10 from Uros Bizjak <ubizjak at gmail dot com> 2011-08-26 16:13:18 UTC ---
Fixed.