[Bug math/18731] New: math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[Bug math/18731] New: math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

            Bug ID: 18731
           Summary: math/test-double-vlen4 and math/test-float-vlen8 fail
                    on non-AVX machines
           Product: glibc
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
          Assignee: unassigned at sourceware dot org
          Reporter: hjl.tools at gmail dot com
  Target Milestone: ---

On non-AVX x86-64 machine, I got

FAIL: math/test-double-vlen4
FAIL: math/test-float-vlen8

Program received signal SIGILL, Illegal instruction.
0x0000000000403a45 in cosf_vlen8 (x=inf)
    at ../sysdeps/x86_64/fpu/test-float-vlen8-wrappers.c:24
24      VECTOR_WRAPPER (WRAPPER_NAME (cosf), _ZGVcN8v_cosf)
(gdb) disass
Dump of assembler code for function cosf_vlen8:
   0x0000000000403a40 <+0>:     lea    0x8(%rsp),%r10

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |andrew.n.senkevich at gmail dot co
                   |                            |m

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> ---
These are wrong:

ifeq ($(build-mathvec),yes)
libmvec-tests += double-vlen2 double-vlen4 double-vlen4-avx2 \
                 float-vlen4 float-vlen8 float-vlen8-avx2

ifeq (yes,$(config-cflags-avx512))
libmvec-tests += double-vlen8 float-vlen16
endif

You can't run those tests just because compiler works.  They will
fail if hardware doesn't support AVX or AVX512.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[andrew.n.senkevich at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

Andrew Senkevich <andrew.n.senkevich at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at sourceware dot org   |andrew.n.senkevich at gmail dot co
                   |                            |m

--- Comment #2 from Andrew Senkevich <andrew.n.senkevich at gmail dot com> ---
(In reply to H.J. Lu from comment #1)
> These are wrong:
> 
> ifeq ($(build-mathvec),yes)
> libmvec-tests += double-vlen2 double-vlen4 double-vlen4-avx2 \
>                  float-vlen4 float-vlen8 float-vlen8-avx2
> 
> ifeq (yes,$(config-cflags-avx512))
> libmvec-tests += double-vlen8 float-vlen16
> endif
> 
> You can't run those tests just because compiler works.  They will
> fail if hardware doesn't support AVX or AVX512.

double-vlen8, float-vlen16 are AVX512 tests and they have runtime checks.

double-vlen4-avx2, float-vlen8-avx2 are AVX2 and they also have runtime checks.

test-double-vlen4, test-float-vlen8 are AVX tests, and yes, for these I didn't
add runtime checks.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[andrew.n.senkevich at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

--- Comment #3 from Andrew Senkevich <andrew.n.senkevich at gmail dot com> ---
Patch sent in https://sourceware.org/ml/libc-alpha/2015-07/msg00957.html

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[cvs-commit at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

--- Comment #4 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "GNU C Library master sources".

The branch, master has been updated
       via  febce2ac5f46a0d5c67ca8b535a028425d421be4 (commit)
      from  95d8f022bf0795cbac473c4f6182595e75cafe5f (commit)

Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.

- Log -----------------------------------------------------------------
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=febce2ac5f46a0d5c67ca8b535a028425d421be4

commit febce2ac5f46a0d5c67ca8b535a028425d421be4
Author: Andrew Senkevich <andrew.senkevich@intel.com>
Date:   Wed Jul 29 19:47:29 2015 +0300

    Added runtime check for AVX vector math tests.

        [BZ #18731]
        * sysdeps/x86_64/fpu/math-tests-arch.h: Added AVX runtime check.
        * sysdeps/x86_64/fpu/test-double-vlen4.c: Likewise.
        * sysdeps/x86_64/fpu/test-float-vlen8.c: Likewise.

-----------------------------------------------------------------------

Summary of changes:
 ChangeLog                              |    7 +++++++
 sysdeps/x86_64/fpu/math-tests-arch.h   |   24 +++++++++++++++++++++++-
 sysdeps/x86_64/fpu/test-double-vlen4.c |    2 ++
 sysdeps/x86_64/fpu/test-float-vlen8.c  |    2 ++
 4 files changed, 34 insertions(+), 1 deletions(-)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[andrew.n.senkevich at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

Andrew Senkevich <andrew.n.senkevich at gmail dot com> changed:

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

--- Comment #5 from Andrew Senkevich <andrew.n.senkevich at gmail dot com> ---
Fixed.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[markus at trippelsdorf dot de]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

Markus Trippelsdorf <markus at trippelsdorf dot de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |markus at trippelsdorf dot de

--- Comment #6 from Markus Trippelsdorf <markus at trippelsdorf dot de> ---
I still get:
 FAIL: math/test-float-vlen4

traps: ld-linux-x86-64[24556] trap invalid opcode ip:403543 sp:7ffc7c36c7c8
error:0 in test-float-vlen4[400000+13000]

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

--- Comment #7 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Markus Trippelsdorf from comment #6)
> I still get:
>  FAIL: math/test-float-vlen4
> 
> traps: ld-linux-x86-64[24556] trap invalid opcode ip:403543 sp:7ffc7c36c7c8
> error:0 in test-float-vlen4[400000+13000]

Please configure glibc with --enable-hardcoded-path-in-tests and
run it under gdb to see why AVX check fails to work.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[markus at trippelsdorf dot de]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

--- Comment #8 from Markus Trippelsdorf <markus at trippelsdorf dot de> ---
Program received signal SIGILL, Illegal instruction.
0x0000000000403663 in cosf_vlen4 (x=inf) at
../sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c:24
24      VECTOR_WRAPPER (WRAPPER_NAME (cosf), _ZGVbN4v_cosf)
(gdb) bt
#0  0x0000000000403663 in cosf_vlen4 (x=inf) at
../sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c:24
#1  0x0000000000402e61 in cos_test () at
/var/tmp/glibc-build/math/libm-test.c:65576
#2  0x0000000000401908 in main (argc=<optimized out>, argv=<optimized out>) at
/var/tmp/glibc-build/math/libm-test.c:150244
(gdb) disass
Dump of assembler code for function cosf_vlen4:
   0x0000000000403660 <+0>:     movaps %xmm0,%xmm1
=> 0x0000000000403663 <+3>:     insertps $0xe,%xmm0,%xmm0
   0x0000000000403669 <+9>:     sub    $0x8,%rsp

The CPU on this test machine doesn't support SSE4_1.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/18731] math/test-double-vlen4 and math/test-float-vlen8 fail on non-AVX machines

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18731

--- Comment #9 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Markus Trippelsdorf from comment #8)
> Program received signal SIGILL, Illegal instruction.
> 0x0000000000403663 in cosf_vlen4 (x=inf) at
> ../sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c:24
> 24      VECTOR_WRAPPER (WRAPPER_NAME (cosf), _ZGVbN4v_cosf)
> (gdb) bt
> #0  0x0000000000403663 in cosf_vlen4 (x=inf) at
> ../sysdeps/x86_64/fpu/test-float-vlen4-wrappers.c:24
> #1  0x0000000000402e61 in cos_test () at
> /var/tmp/glibc-build/math/libm-test.c:65576
> #2  0x0000000000401908 in main (argc=<optimized out>, argv=<optimized out>)
> at /var/tmp/glibc-build/math/libm-test.c:150244
> (gdb) disass
> Dump of assembler code for function cosf_vlen4:
>    0x0000000000403660 <+0>:     movaps %xmm0,%xmm1
> => 0x0000000000403663 <+3>:     insertps $0xe,%xmm0,%xmm0
>    0x0000000000403669 <+9>:     sub    $0x8,%rsp
> 
> The CPU on this test machine doesn't support SSE4_1.

Please open a new bug against SSE4.1.

-- 
You are receiving this mail because:
You are on the CC list for the bug.