[Bug tree-optimization/60236] New: gfortran.dg/vect/pr32380.f fails on ARM

[Bug tree-optimization/60236] New: gfortran.dg/vect/pr32380.f fails on ARM

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

            Bug ID: 60236
           Summary: gfortran.dg/vect/pr32380.f fails on ARM
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: bernd.edlinger at hotmail dot de

Hi,

this test case fails, because the only 5 of 6 loops get optimized:

pr32380.f:162:0: note: function is not vectorizable.                            
pr32380.f:162:0: note: not vectorized: relevant stmt not supported: _113 =
__builtin_sqrtf (_112);

pr32380.f:5:0: note: vectorized 5 loops in function.

but the test case expects 6 loops here.


gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/home/ed/gnu/arm-linux-gnueabihf/libexec/gcc/armv7l-unknown-linux-gnueabihf/4.9.0/lto-wrapper
Target: armv7l-unknown-linux-gnueabihf
Configured with: ../gcc-4.9-20140209/configure
--prefix=/home/ed/gnu/arm-linux-gnueabihf
--enable-languages=c,c++,objc,obj-c++,fortran,ada,go --with-arch=armv7-a
--with-tune=cortex-a9 --with-fpu=vfpv3-d16 --with-float=hard
Thread model: posix
gcc version 4.9.0 20140209 (experimental) (GCC)

[Bug tree-optimization/60236] gfortran.dg/vect/pr32380.f fails on ARM

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Does ARM have a builtin-vectorized-function hook that handles sqrtf?  If not,
then that's expected.

Please adjust the expected number of vectorizations with vect_call_sqrt like

! { dg-final { scan-tree-dump-times "vectorized 5 loops" 1 "vect" { ! target
vect_call_sqrt } } }
! { dg-final { scan-tree-dump-times "vectorized 6 loops" 1 "vect" { target
vect_call_sqrt  } } }

(and hope that powerpc and ia64 hit exactly the same issue ...)

[Bug tree-optimization/60236] gfortran.dg/vect/pr32380.f fails on ARM

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

--- Comment #2 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
(In reply to Richard Biener from comment #1)
> Does ARM have a builtin-vectorized-function hook that handles sqrtf?  If not,
> then that's expected.
> 

Yes, it does have a builtin-vectorized-function which does _not_ handle sqrtf.

> Please adjust the expected number of vectorizations with vect_call_sqrt like
> 
> ! { dg-final { scan-tree-dump-times "vectorized 5 loops" 1 "vect" { ! target
> vect_call_sqrt } } }
> ! { dg-final { scan-tree-dump-times "vectorized 6 loops" 1 "vect" { target
> vect_call_sqrt  } } }
> 
> (and hope that powerpc and ia64 hit exactly the same issue ...)

you mean vect_call_sqrtf ?

[Bug tree-optimization/60236] gfortran.dg/vect/pr32380.f fails on ARM

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

--- Comment #3 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
ERROR: gfortran.dg/vect/pr32380.f  -O : error executing dg-final: syntax error
in target selector "! target vect_call_sqrtf"

what is the right syntax here??

[Bug tree-optimization/60236] gfortran.dg/vect/pr32380.f fails on ARM

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

--- Comment #4 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
hmm...

I just looked at PR#52229.
At least on PowerPC many more loops are not vectorizable.
because of unaligned access.

Jakub's last comment there is that the xfail
should be based on the vect_hw_misaling attribute.

Well, but the problem on ARMv7 was not about misalignment
instead it was about we can not vectorize the SQRT on Arm,
and certainly not on IA64, because they have no
builtin-vectorized-function hook at all:

       set et_vect_hw_misalign_saved 0
       if { ([istarget x86_64-*-*]
            || [istarget aarch64*-*-*]
            || [istarget i?86-*-*]) } {
          set et_vect_hw_misalign_saved 1
       }

is this correct for ARM?
Or can the ARM CPU vectorize unaligned data,
I think they just have to know if the data is aligned or not,
and choose different opcodes?

[Bug tree-optimization/60236] gfortran.dg/vect/pr32380.f fails on ARM

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

--- Comment #5 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
Created attachment 32166
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32166&action=edit
proposed patch

Ok, now I have fixed the syntax.

This patch works for arm and x86_64.
And takes into account the Problem that was reported
for PowerPC at PR52229.

But I cannot run the test on PowerPC.

[Bug tree-optimization/60236] gfortran.dg/vect/pr32380.f fails on ARM

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

Bernd Edlinger <bernd.edlinger at hotmail dot de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dominiq at lps dot ens.fr,
                   |                            |tkoenig at gcc dot gnu.org

--- Comment #6 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
Hi,

This patch tries to solve the problem in a more general way.
I wonder if one of you could try this patch on PowerPC?

Thanks
Bernd.

[Bug fortran/60236] gfortran.dg/vect/pr32380.f fails on ARM

[edlinger at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

--- Comment #8 from edlinger at gcc dot gnu.org ---
Author: edlinger
Date: Sun Mar  2 18:06:49 2014
New Revision: 208257

URL: http://gcc.gnu.org/viewcvs?rev=208257&root=gcc&view=rev
Log:
2014-03-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR fortran/60236
        * gfortran.dg/vect/pr32380.f: Fix expected test results.

Modified:
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gfortran.dg/vect/pr32380.f

[Bug fortran/60236] gfortran.dg/vect/pr32380.f fails on ARM

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60236

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |Greta.Yorsh at arm dot com

--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 52249 has been marked as a duplicate of this bug. ***