[Bug libgomp/49965] New: libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[Bug libgomp/49965] New: libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

           Summary: libgomp.c++/reduction-4.C and libgomp.c++/task-8.C
                    FAIL on Solaris 11/SPARC
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libgomp
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ro@gcc.gnu.org
                CC: jakub@gcc.gnu.org
              Host: sparc-sun-solaris2.11
            Target: sparc-sun-solaris2.11
             Build: sparc-sun-solaris2.11


Two new libgomp testcases FAIL on Solaris 11/SPARC:

FAIL: libgomp.c++/reduction-4.C  -O0  execution test

  both 32 and 64-bit

  abort () here:

#3  0x000183b8 in foo<long long, long double> () at
/vol/gcc/src/hg/trunk/solaris/libgomp/testsuite/libgomp.c++/reduction-4.C:26

  j = 8, p = inf

  This doesn't happen on Solaris 10/x86.

FAIL: libgomp.c++/task-8.C  -O0  (test for excess errors)
Excess errors:
ld: warning: symbol 'err' has differing types:
    (file /var/tmp//ccD0aiQD.o type=OBJT; file /lib/libc.so type=FUNC);
    /var/tmp//ccD0aiQD.o definition taken
ld: warning: symbol 'err' has differing types:
    (file /var/tmp//ccD0aiQD.o type=OBJT; file /lib/libc.so type=FUNC);

  err() was only introduced in Solaris 11.  Replacing err by error fixes this.

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

Rainer Orth <ro at gcc dot gnu.org> changed:

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

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-03 15:10:13 UTC ---
for task-8.C, error is a function on linux, so please replace it by errval
or err_atomic or similar instead if err doesn't work on Solaris.

Fur reduction-4.C, perhaps Solaris long double handling is broken?  The
testcase doesn't use anything complicated, in *.omp_fn* of the long long/long
double version the private f is initialized to +Inf, then in some iterations
set to 9.0 or -2.0 and finally everything is MIN_EXPRed together in a critical
section.

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #2 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-08-03 16:26:45 UTC ---
> --- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-03 15:10:13 UTC ---
> for task-8.C, error is a function on linux, so please replace it by errval
> or err_atomic or similar instead if err doesn't work on Solaris.

Sure: I'd have tested that on Linux/x86_64 before submitting anyway.

> Fur reduction-4.C, perhaps Solaris long double handling is broken?  The

Not in general :-)

> testcase doesn't use anything complicated, in *.omp_fn* of the long long/long
> double version the private f is initialized to +Inf, then in some iterations
> set to 9.0 or -2.0 and finally everything is MIN_EXPRed together in a critical
> section.

What I've done is adding a couple of printf's all over
template <typename I, typename F> void foo (): before the #pragma omp
parallel for, before and after each assignement in the for loop, and
before and after the abort, every time casting the args to long long
resp. long double with matching format strings.  It I initialize i to 0
before the #pragma so I get comparable output, there's no difference in
output between i386-pc-solaris2.10 and sparc-sun-solaris2.11 except for
the abort.

    Rainer

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-03 16:37:36 UTC ---
So what values it printed?  Did it print -2.0 and 9.0 in some iterations?
The final merging is done in a critical section between GOMP_atomic_start and
GOMP_atomic_end, perhaps you can put a breakpoint in there and watch how the
values are merged using the MIN_EXPR.
Can you reproduce the failure with OMP_NUM_THREADS=1 BTW?

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #4 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-08-03 16:50:27 UTC ---
> --- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-03 16:37:36 UTC ---
> So what values it printed?  Did it print -2.0 and 9.0 in some iterations?

Here's the complete output:

i = 0: j = -10000, f = 1024
i = 1: f = Inf
i = 2: j = -2147483648, f = Inf
i = 1: f = -2
i = 2: j = 8, f = Inf
i = 3
i = 2: j = 8, f = 9
i = 0: j = -2147483648
i = 0: j = -16
i = 0: j = 8, f = -2

i = 0: j = -10000, f = 1024
i = 1: f = Inf
i = 1: f = -2
i = 3
i = 2: j = -2147483648, f = Inf
i = 2: j = 8, f = Inf
i = 2: j = 8, f = 9
i = 0: j = -2147483648
i = 0: j = -16
i = 0: j = 8, f = -2

i = 0: j = -10000, f = 1024
i = 1: f = Inf
i = 1: f = -2
i = 3
i = 2: j = -9223372036854775808, f = Inf
i = 2: j = 8, f = Inf
i = 2: j = 8, f = 9
i = 0: j = -9223372036854775808
i = 0: j = -16
i = 0: j = 8, f = Inf

> The final merging is done in a critical section between GOMP_atomic_start and
> GOMP_atomic_end, perhaps you can put a breakpoint in there and watch how the
> values are merged using the MIN_EXPR.

I've missed the right call so far.  Will give it another try tomorrow.

> Can you reproduce the failure with OMP_NUM_THREADS=1 BTW?

Yes, the failure is the same, just with

i = 0: j = -10000, f = 1024
i = 0: j = -9223372036854775808
i = 0: j = -16
i = 1: f = Inf
i = 1: f = -2
i = 2: j = -16, f = -2
i = 2: j = 8, f = -2
i = 2: j = 8, f = -2
i = 3
i = 0: j = 8, f = 1024
Abort

    Rainer

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #5 from Rainer Orth <ro at gcc dot gnu.org> 2011-08-05 12:22:16 UTC ---
Author: ro
Date: Fri Aug  5 12:22:10 2011
New Revision: 177428

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=177428
Log:
    PR libgomp/49965
    * testsuite/libgomp.c++/task-8.C: Replaced err by errval.

Modified:
    trunk/libgomp/ChangeLog
    trunk/libgomp/testsuite/libgomp.c++/task-8.C

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #6 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-08-09 15:10:25 UTC ---
> --- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-03 16:37:36 UTC ---
> So what values it printed?  Did it print -2.0 and 9.0 in some iterations?
> The final merging is done in a critical section between GOMP_atomic_start and
> GOMP_atomic_end, perhaps you can put a breakpoint in there and watch how the
> values are merged using the MIN_EXPR.
> Can you reproduce the failure with OMP_NUM_THREADS=1 BTW?

I think there's a code generation bug here: I could reduce the example
to a considerably smaller testcase (red-4.C, attached), which still
fails with OMP_NUM_THREADS=1.

Between GOMP_atomic_start and GOMP_atomic_end, there's a single call to
_Q_fle (-2, 1024), which correctly returns 1 (i.e. first arg <= second).

With g++ -mcpu=v9 -fopenmp -O3 red-4.c -lm, I get this code sequence:

    ldd    [%i0], %f8    %i1
    ldd    [%i0+8], %f10
    ldd    [%fp-48], %f12    %i0
    ldd    [%fp-40], %f14
    std    %f12, [%fp-16]    %i0
    std    %f14, [%fp-8]
    std    %f8, [%fp-32]    %i1
    std    %f10, [%fp-24]
    std    %f8, [%fp-56]    %i1
    std    %f10, [%fp-64]
    add    %fp, -16, %o0    %i0 <= %i1
    call    _Q_fle, 0
     add    %fp, -32, %o1
    cmp    %o0, 0        1 on true, 0 on false
    ldd    [%fp-56], %f8    %i1
    ldd    [%fp-64], %f10
    ldd    [%fp-48], %f12    %i0
    ldd    [%fp-40], %f14
    fmovdle    %icc, %f12, %f8        on <= 0 (false): %i0 -> %i1
    fmovdle    %icc, %f14, %f10
    std    %f8, [%i0]    return %i1
    std    %f10, [%i0+8]

I think the fmovdle is wrong: if _Q_fle returns 1 (<=), the first arg
should be returned, but it's the other way round.

    Rainer

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #7 from Rainer Orth <ro at gcc dot gnu.org> 2011-08-09 15:11:57 UTC ---
Created attachment 24962
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24962
reduced testcase

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #8 from Rainer Orth <ro at gcc dot gnu.org> 2011-08-09 15:12:53 UTC ---
Created attachment 24963
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24963
corresponding assembler output

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ebotcazou at gcc dot
                   |                            |gnu.org

--- Comment #9 from Rainer Orth <ro at gcc dot gnu.org> 2011-08-09 15:18:45 UTC ---
Eric, any idea?

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-09-30
     Ever Confirmed|0                           |1

--- Comment #10 from Rainer Orth <ro at gcc dot gnu.org> 2011-09-30 09:44:03 UTC ---
Eric, could you please have a look to see if this is a SPARC backend issue or
the
problem lies elsewhere?

Thanks.
   Rainer

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |ebotcazou at gcc dot
                   |gnu.org                     |gnu.org

--- Comment #11 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-09-30 10:14:23 UTC ---
Investigating.

[Bug libgomp/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #12 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-11 15:37:29 UTC ---
This is a fallout of the merge of the cond-optab branch in the 4.5 series.

[Bug target/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #14 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-11 21:34:01 UTC ---
Author: ebotcazou
Date: Tue Oct 11 21:33:57 2011
New Revision: 179828

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=179828
Log:
    PR target/49965
    * config/sparc/sparc.md (mov<I:mode>cc): Do not save comparison code.
    (mov<F:mode>cc): Likewise.

Modified:
    branches/gcc-4_6-branch/gcc/ChangeLog
    branches/gcc-4_6-branch/gcc/config/sparc/sparc.md

[Bug target/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #13 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-11 21:33:28 UTC ---
Author: ebotcazou
Date: Tue Oct 11 21:33:24 2011
New Revision: 179827

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=179827
Log:
    PR target/49965
    * config/sparc/sparc.md (mov<I:mode>cc): Do not save comparison code.
    (mov<F:mode>cc): Likewise.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/sparc/sparc.md

[Bug target/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #15 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-11 21:34:48 UTC ---
Author: ebotcazou
Date: Tue Oct 11 21:34:42 2011
New Revision: 179829

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=179829
Log:
    PR target/49965
    * config/sparc/sparc.md (mov<I:mode>cc): Do not save comparison code.
    (mov<F:mode>cc): Likewise.

Modified:
    branches/gcc-4_5-branch/gcc/ChangeLog
    branches/gcc-4_5-branch/gcc/config/sparc/sparc.md

[Bug target/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

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

--- Comment #16 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-11 21:39:14 UTC ---
The testsuite should be clean again.

[Bug target/49965] libgomp.c++/reduction-4.C and libgomp.c++/task-8.C FAIL on Solaris 11/SPARC

[davem at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49965

--- Comment #17 from davem at gcc dot gnu.org 2011-11-04 20:26:07 UTC ---
Author: davem
Date: Fri Nov  4 20:25:59 2011
New Revision: 180982

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=180982
Log:
Fix sparc regression due to recent movcc pattern changes.

    PR target/49965
    * config/sparc/sparc.c (sparc_expand_conditional_move): Handle the
    fact that sparc_emit_float_lib_cmp modifies the comparison in
    operands[1].

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