[Bug target/51134] New: [4.7 Regression] FAIL: gfortran.fortran-torture/execute/arrayarg.f90

[Bug target/51134] New: [4.7 Regression] FAIL: gfortran.fortran-torture/execute/arrayarg.f90

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

             Bug #: 51134
           Summary: [4.7 Regression] FAIL:
                    gfortran.fortran-torture/execute/arrayarg.f90
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com
                CC: hubicka@gcc.gnu.org


On Linux/x86-64, revision 181359 gave

FAIL: gfortran.fortran-torture/execute/arrayarg.f90 execution,  -O2
-fomit-frame-pointer -finline-functions -funroll-loops 

Revision 181355 is OK. It may be caused by revision 181357:

http://gcc.gnu.org/ml/gcc-cvs/2011-11/msg00649.html

[Bug target/51134] [4.7 Regression] FAIL: gfortran.fortran-torture/execute/arrayarg.f90

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-15 04:03:41 UTC ---
It is caused by revision 181357.

[Bug target/51134] [4.7 Regression] FAIL: gfortran.fortran-torture/execute/arrayarg.f90

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-15 04:05:06 UTC ---
gfortran.fortran-torture/execute/arrayarg.f90 is miscompiled.

[Bug target/51134] [4.7 Regression] FAIL: gfortran.fortran-torture/execute/arrayarg.f90

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #3 from Jan Hubicka <hubicka at ucw dot cz> 2011-11-15 12:37:07 UTC ---
> gfortran.fortran-torture/execute/arrayarg.f90 is miscompiled.
There are 4 expanded memset calls in the testcase, all ends up being unrolled
loop.
Looks like we misoptmize one of them.  I am looking into that.

Honza

[Bug target/51134] [4.7 Regression] FAIL: gfortran.fortran-torture/execute/arrayarg.f90

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #4 from Jan Hubicka <hubicka at ucw dot cz> 2011-11-15 13:28:42 UTC ---
Hi,
the bug is loopy epilogue not being used because promoted value is not
availbale. The non-loopy epilogue does not expect alignments
greater than 16.

I am testing the following patch.
I think we might still have bugs for blocks >8 && <16. Will try to produce
testcase.

Honza

Index: config/i386/i386.c
===================================================================
--- config/i386/i386.c    (revision 181357)
+++ config/i386/i386.c    (working copy)
@@ -23073,13 +23073,13 @@ ix86_expand_setmem (rtx dst, rtx count_e
       rtx tmp;
       if (align_unknown && unroll_factor > 1
       && epilogue_size_needed >= GET_MODE_SIZE (move_mode)
-      && vec_promoted_val)
+      && (vec_promoted_val || gpr_promoted_val))
     {
       /* Reduce epilogue's size by creating not-unrolled loop.  If we won't
          do this, we can have very big epilogue - when alignment is statically
          unknown we'll have the epilogue byte by byte which may be very slow. 
*/
       loop_iter = expand_set_or_movmem_via_loop_with_iter (dst, NULL, destreg,
-          NULL, vec_promoted_val, count_exp,
+          NULL, vec_promoted_val ? vec_promoted_val : gpr_promoted_val,
count_exp,
           loop_iter, move_mode, 1,
           expected_size, false);
       dst = change_address (dst, BLKmode, destreg);

[Bug target/51134] [4.7 Regression] FAIL: gfortran.fortran-torture/execute/arrayarg.f90

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #5 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-15 16:34:22 UTC ---
(In reply to comment #4)
> Hi,
> the bug is loopy epilogue not being used because promoted value is not
> availbale. The non-loopy epilogue does not expect alignments
> greater than 16.
> 
> I am testing the following patch.
> I think we might still have bugs for blocks >8 && <16. Will try to produce
> testcase.

I believe so.  If you configure GCC for Linux/x86-64 with

--with-arch=core2 --with-cpu=atom

you will get

gengtype: malloc.c:3840: _int_malloc: Assertion `(unsigned long)(size) >=
(unsigned long)(nb)' failed.
make[6]: *** [s-gtype] Aborted

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.7.0
            Summary|[4.7 Regression] FAIL:      |[4.7 Regression] x86
                   |gfortran.fortran-torture/ex |memset/memcpy expansion is
                   |ecute/arrayarg.f90          |broken

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #6 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-16 01:56:08 UTC ---
On Linux/ia32, If you configure GCC with

--with-arch=core2 --with-cpu=atom

you will get

bin/sh ../../src-trunk/gcc/../move-if-change tmp-check.h tree-check.h
*** glibc detected *** build/gengtype: malloc(): memory corruption: 0x09561138
***

and it hangs forever.

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2011-11-18
     Ever Confirmed|0                           |1

--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-11-18 01:03:17 UTC ---
I am testing patch fixing the Atom issues and other problems.

The gfortran.fortran-torture/execute/arrayarg.f90 testcase is however
different: the new inline code enables loop unrolling that in turn makes sched2
to produce wrong code.  The testcase passes with -fno-unroll-loop (or a hack
disabling only RTL unroller) as well as with -fno-schedule-insns2 added.

After hours looking into the code produced I am quite convinced that the memset
expansion involved is correct.  Can someone with better Fortran-fu than me
verify that the testcase is well defined WRT aliasing?

Honza

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #8 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-11-18 02:18:32 UTC ---
Author: hubicka
Date: Fri Nov 18 02:18:28 2011
New Revision: 181466

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=181466
Log:

    PR bootstrap/51134
    * i386.c (atom_cost): Fix 32bit memset description.
    (expand_set_or_movmem_via_loop_with_iter): Output proper bounds check for
epilogue loops.
    (expand_movmem_epilogue): Handle epilogues up to size 15 w/o producing byte
loop.
    (decide_alg): sse_loop is not useable wthen SSE2 is disabled; when not
optimizing always
    use rep movsb or lincall; do not produce word sized loops when optimizing
memset for
    size (to avoid need for large constants).
    (ix86_expand_movmem): Get into sync with ix86_expand_setmem; choose unroll
factors
    better; always do 128bit moves when producing SSE loops; do not produce
loopy epilogue
    when size is too small.
    (promote_duplicated_reg_to_size): Do not look into desired alignments when
    doing vector expansion.
    (ix86_expand_setmem): Track better when promoted value is available; choose
unroll factors
    more sanely.; output loopy epilogue only when needed.

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

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #9 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-22 03:30:54 UTC ---
On Linux/x86-64, configured with

--enable-clocale=gnu --with-system-zlib --with-demangler-in-ld --enable-shared
--prefix=/usr/gcc-4.7.0 --with-local-prefix=/usr/local -with-arch=core2
--with-cpu=atom --with-fpmath=sse

revision 181584 result is at

http://gcc.gnu.org/ml/gcc-testresults/2011-11/msg02226.html

revision 181584 + the old memset/memcpy expansion result us at

http://gcc.gnu.org/ml/gcc-testresults/2011-11/msg02225.html

Regressions caused by the new memset/memcpy expansion are

FAIL: 27_io/basic_filebuf/seekoff/wchar_t/1.cc execution test
FAIL: 27_io/basic_filebuf/seekpos/wchar_t/9874.cc execution test
FAIL: 27_io/manipulators/basefield/char/1.cc execution test
FAIL: 27_io/manipulators/basefield/wchar_t/1.cc execution test
FAIL: 27_io/objects/wchar_t/12.cc execution test
FAIL: 27_io/objects/wchar_t/13.cc execution test
FAIL: events run
FAIL: gcc.target/i386/cleanup-1.c execution test
FAIL: gcc.target/i386/cleanup-2.c execution test
FAIL: gcc.target/i386/pr34077.c (internal compiler error)
FAIL: gcc.target/i386/pr34077.c (internal compiler error)
FAIL: gcc.target/i386/pr34077.c (test for excess errors)
FAIL: gcc.target/i386/pr34077.c (test for excess errors)
FAIL: gfortran.fortran-torture/execute/arrayarg.f90 execution,  -O2
-fomit-frame-pointer -finline-functions -funroll-loops

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[michael.v.zolotukhin at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #10 from Michael Zolotukhin <michael.v.zolotukhin at gmail dot com> 2011-11-22 13:17:38 UTC ---
Created attachment 25882
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25882
Patch for new memset/memcpy implementation

(In reply to comment #9)
> Regressions caused by the new memset/memcpy expansion are
> 
> FAIL: 27_io/basic_filebuf/seekoff/wchar_t/1.cc execution test
> FAIL: 27_io/basic_filebuf/seekpos/wchar_t/9874.cc execution test
> FAIL: 27_io/manipulators/basefield/char/1.cc execution test
> FAIL: 27_io/manipulators/basefield/wchar_t/1.cc execution test
> FAIL: 27_io/objects/wchar_t/12.cc execution test
> FAIL: 27_io/objects/wchar_t/13.cc execution test
> FAIL: events run
> FAIL: gcc.target/i386/cleanup-1.c execution test
> FAIL: gcc.target/i386/cleanup-2.c execution test
> FAIL: gcc.target/i386/pr34077.c (internal compiler error)
> FAIL: gcc.target/i386/pr34077.c (internal compiler error)
> FAIL: gcc.target/i386/pr34077.c (test for excess errors)
> FAIL: gcc.target/i386/pr34077.c (test for excess errors)
> FAIL: gfortran.fortran-torture/execute/arrayarg.f90 execution,  -O2
> -fomit-frame-pointer -finline-functions -funroll-loops

With the attached patch fails on PR34077 and arrayarg.f90 are fixed, fails on
cleanup-tests are probably caused by incorrect testcase (see
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49503).

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-22 16:34:54 UTC ---
(In reply to comment #10)
> Created attachment 25882 [details]
> Patch for new memset/memcpy implementation
> 
> (In reply to comment #9)
> > Regressions caused by the new memset/memcpy expansion are
> > 
> > FAIL: 27_io/basic_filebuf/seekoff/wchar_t/1.cc execution test
> > FAIL: 27_io/basic_filebuf/seekpos/wchar_t/9874.cc execution test
> > FAIL: 27_io/manipulators/basefield/char/1.cc execution test
> > FAIL: 27_io/manipulators/basefield/wchar_t/1.cc execution test
> > FAIL: 27_io/objects/wchar_t/12.cc execution test
> > FAIL: 27_io/objects/wchar_t/13.cc execution test
> > FAIL: events run
> > FAIL: gcc.target/i386/cleanup-1.c execution test
> > FAIL: gcc.target/i386/cleanup-2.c execution test
> > FAIL: gcc.target/i386/pr34077.c (internal compiler error)
> > FAIL: gcc.target/i386/pr34077.c (internal compiler error)
> > FAIL: gcc.target/i386/pr34077.c (test for excess errors)
> > FAIL: gcc.target/i386/pr34077.c (test for excess errors)
> > FAIL: gfortran.fortran-torture/execute/arrayarg.f90 execution,  -O2
> > -fomit-frame-pointer -finline-functions -funroll-loops
> 
> With the attached patch fails on PR34077 and arrayarg.f90 are fixed, fails on
> cleanup-tests are probably caused by incorrect testcase (see
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49503).

Do you have a patch for those C++ and Java regressions?

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[michael.v.zolotukhin at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #12 from Michael Zolotukhin <michael.v.zolotukhin at gmail dot com> 2011-11-22 16:54:35 UTC ---
> Do you have a patch for those C++ and Java regressions?

What regressions do you mean exactly? I managed to fix the bootstraps
(with sse_loop enabled again), but there are still some fails, so I
don't send the patch.

Currently I don't have a fix that solves all the problems - the
attached to previous letter patch fixes some of them, but there are
other fails (in 27_io and in some specs2k). I'm continuing debugging
and hope to finish fixes soon.

On 22 November 2011 20:34, hjl.tools at gmail dot com
<gcc-bugzilla@gcc.gnu.org> wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134
>
> --- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-22 16:34:54 UTC ---
> (In reply to comment #10)
>> Created attachment 25882 [details]
>> Patch for new memset/memcpy implementation
>>
>> (In reply to comment #9)
>> > Regressions caused by the new memset/memcpy expansion are
>> >
>> > FAIL: 27_io/basic_filebuf/seekoff/wchar_t/1.cc execution test
>> > FAIL: 27_io/basic_filebuf/seekpos/wchar_t/9874.cc execution test
>> > FAIL: 27_io/manipulators/basefield/char/1.cc execution test
>> > FAIL: 27_io/manipulators/basefield/wchar_t/1.cc execution test
>> > FAIL: 27_io/objects/wchar_t/12.cc execution test
>> > FAIL: 27_io/objects/wchar_t/13.cc execution test
>> > FAIL: events run
>> > FAIL: gcc.target/i386/cleanup-1.c execution test
>> > FAIL: gcc.target/i386/cleanup-2.c execution test
>> > FAIL: gcc.target/i386/pr34077.c (internal compiler error)
>> > FAIL: gcc.target/i386/pr34077.c (internal compiler error)
>> > FAIL: gcc.target/i386/pr34077.c (test for excess errors)
>> > FAIL: gcc.target/i386/pr34077.c (test for excess errors)
>> > FAIL: gfortran.fortran-torture/execute/arrayarg.f90 execution,  -O2
>> > -fomit-frame-pointer -finline-functions -funroll-loops
>>
>> With the attached patch fails on PR34077 and arrayarg.f90 are fixed, fails on
>> cleanup-tests are probably caused by incorrect testcase (see
>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49503).
>
> Do you have a patch for those C++ and Java regressions?

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

--- Comment #13 from hjl at gcc dot gnu.org <hjl at gcc dot gnu.org> 2011-11-24 22:11:16 UTC ---
Author: hjl
Date: Thu Nov 24 22:11:12 2011
New Revision: 181701

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=181701
Log:
Revert revision 181357.

gcc/

2011-11-24  H.J. Lu  <hongjiu.lu@intel.com>

    PR target/51134
    * config/i386/i386.h (processor_costs): Revert revision 181357.
    * config/i386/i386.c (cost models): Likewise.
    (core_cost): Likewise.
    (promote_duplicated_reg): Likewise.
    (promote_duplicated_reg_to_size): Likewise.
    (processor_target): Likewise.
    (expand_set_or_movmem_via_loop_with_iter): Likewise.
    (expand_set_or_movmem_via_loop): Likewise.
    (emit_strset): Likewise.
    (expand_movmem_epilogue): Likewise.
    (expand_setmem_epilogue): Likewise.
    (expand_movmem_prologue): Likewise.
    (expand_setmem_prologue): Likewise.
    (expand_constant_movmem_prologue): Likewise.
    (expand_constant_setmem_prologue): Likewise.
    (decide_alg): Likewise.
    (decide_alignment): Likewise.
    (ix86_expand_movmem): Likewise.
    (ix86_expand_setmem): Likewise.
    (ix86_slow_unaligned_access): Likewise.
    * config/i386/i386.md (strset): Likewise.
    * config/i386/sse.md (vec_dupv4si): Likewise.
    (vec_dupv2di): Likewise.

gcc/testsuite/

2011-11-24  H.J. Lu  <hongjiu.lu@intel.com>

    PR target/51134
    * gcc.target/i386/sw-1.c: Revert revision 181357.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386-opts.h
    trunk/gcc/config/i386/i386.c
    trunk/gcc/config/i386/i386.h
    trunk/gcc/config/i386/i386.opt
    trunk/gcc/config/i386/sse.md
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.target/i386/sw-1.c

[Bug target/51134] [4.7 Regression] x86 memset/memcpy expansion is broken

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51134

H.J. Lu <hjl.tools at gmail dot com> changed:

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

--- Comment #14 from H.J. Lu <hjl.tools at gmail dot com> 2011-11-25 01:04:37 UTC ---
Fixed.