Strange error

[Salvatore Filippone]

Hi,
This is probably for the main GCC list, but anyway: I have a new machine
on which I installed multiple versions of GCC/GNU Fortran. To be
precise, I have bootstrapped and installted versions 4.2.4, 4.3.5,
4.4.0, 4.5.3 and 4.6.0.
Now, from 4.2 to 4.4 things work as expected. With 4.5.3 and 4.6.0 I get
the following:
[sfilippo@node64-1 modules]$ gfortran -O3  -march=native  -I.. -I. -I. -DHAVE_ISO_FORTRAN_ENV -DHAVE_FLUSH_STMT -DHAVE_VOLATILE -DHAVE_MOVE_ALLOC -DMPI_H  -c psb_realloc_mod.F90
/tmp/cceVTH1u.s: Assembler messages:
/tmp/cceVTH1u.s:158: Error: no such instruction: `vmovdqa 368(%rsp),%xmm0'
/tmp/cceVTH1u.s:159: Error: no such instruction: `vmovdqa %xmm0,336(%rsp)'
/tmp/cceVTH1u.s:335: Error: no such instruction: `vmovups (%r9,%rdx),%xmm0'
/tmp/cceVTH1u.s:337: Error: no such instruction: `vmovaps %xmm0,(%rdi,%rdx)'
....................................
/tmp/cceVTH1u.s:12566: Error: no such instruction: `vmovdqu (%r8,%rax),%xmm0'
/tmp/cceVTH1u.s:12568: Error: no such instruction: `vmovdqa %xmm0,(%rdi,%rax)'
/tmp/cceVTH1u.s:12783: Error: no such instruction: `vmovdqu (%r14,%rax),%xmm0'
/tmp/cceVTH1u.s:12785: Error: no such instruction: `vmovdqa %xmm0,(%rbx,%rax)'

this is CentOS 5.6, with CPU
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 6
model           : 42
model name      :        Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz
stepping        : 7
cpu MHz         : 3401.000
cache size      : 8192 KB
physical id     : 0
siblings        : 8
core id         : 0
cpu cores       : 4
apicid          : 0
fpu             : yes
fpu_exception   : yes
cpuid level     : 13
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx rdtscp lm constant_tsc ida nonstop_tsc arat pni monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr sse4_1 sse4_2 popcnt lahf_lm
bogomips        : 6784.94
clflush size    : 64
cache_alignment : 64
address sizes   : 36 bits physical, 48 bits virtual
power management: [8]


Question is obvious: what the *@!++* is going on? What would be needed
to further debug this issue?
 
Thanks
Salvatore

[Richard Guenther-2]

On Wed, Jun 1, 2011 at 3:34 PM, Salvatore Filippone
<salvatore.filippone@...> wrote:

> Hi,
> This is probably for the main GCC list, but anyway: I have a new machine
> on which I installed multiple versions of GCC/GNU Fortran. To be
> precise, I have bootstrapped and installted versions 4.2.4, 4.3.5,
> 4.4.0, 4.5.3 and 4.6.0.
> Now, from 4.2 to 4.4 things work as expected. With 4.5.3 and 4.6.0 I get
> the following:
> [sfilippo@node64-1 modules]$ gfortran -O3  -march=native  -I.. -I. -I. -DHAVE_ISO_FORTRAN_ENV -DHAVE_FLUSH_STMT -DHAVE_VOLATILE -DHAVE_MOVE_ALLOC -DMPI_H  -c psb_realloc_mod.F90
> /tmp/cceVTH1u.s: Assembler messages:
> /tmp/cceVTH1u.s:158: Error: no such instruction: `vmovdqa 368(%rsp),%xmm0'
> /tmp/cceVTH1u.s:159: Error: no such instruction: `vmovdqa %xmm0,336(%rsp)'
> /tmp/cceVTH1u.s:335: Error: no such instruction: `vmovups (%r9,%rdx),%xmm0'
> /tmp/cceVTH1u.s:337: Error: no such instruction: `vmovaps %xmm0,(%rdi,%rdx)'
> ....................................
> /tmp/cceVTH1u.s:12566: Error: no such instruction: `vmovdqu (%r8,%rax),%xmm0'
> /tmp/cceVTH1u.s:12568: Error: no such instruction: `vmovdqa %xmm0,(%rdi,%rax)'
> /tmp/cceVTH1u.s:12783: Error: no such instruction: `vmovdqu (%r14,%rax),%xmm0'
> /tmp/cceVTH1u.s:12785: Error: no such instruction: `vmovdqa %xmm0,(%rbx,%rax)'
>
> this is CentOS 5.6, with CPU
> processor       : 0
> vendor_id       : GenuineIntel
> cpu family      : 6
> model           : 42
> model name      :        Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz
> stepping        : 7
> cpu MHz         : 3401.000
> cache size      : 8192 KB
> physical id     : 0
> siblings        : 8
> core id         : 0
> cpu cores       : 4
> apicid          : 0
> fpu             : yes
> fpu_exception   : yes
> cpuid level     : 13
> wp              : yes
> flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx rdtscp lm constant_tsc ida nonstop_tsc arat pni monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr sse4_1 sse4_2 popcnt lahf_lm
> bogomips        : 6784.94
> clflush size    : 64
> cache_alignment : 64
> address sizes   : 36 bits physical, 48 bits virtual
> power management: [8]
>
>
> Question is obvious: what the *@!++* is going on? What would be needed
> to further debug this issue?

Update your assember - it doesn't seem to understand AVX instructions.

Richard.

> Thanks
> Salvatore
>
>
>

[Salvatore Filippone]

Il giorno mer, 01/06/2011 alle 15.38 +0200, Richard Guenther ha scritto:
>
> Update your assember - it doesn't seem to understand AVX instructions.
>
> Richard.

Which means that on this particular opsys (CentOS 5.6) I have to build
and install my own binutils, at least for recent GCC versions, since the
system binutils are at 2.17 and no newer precompiled version is
available.

Thanks a lot
Salvatore

[Salvatore Filippone]

On further reflection, isn't this something that configure should warn
me about when I do the bootstrap? Or am I asking for too much ?

Il giorno mer, 01/06/2011 alle 15.38 +0200, Richard Guenther ha scritto:
> Update your assember - it doesn't seem to understand AVX instructions.
>
> Richard.

[Tobias Burnus]

On 06/01/2011 03:46 PM, Salvatore Filippone wrote:
> Il giorno mer, 01/06/2011 alle 15.38 +0200, Richard Guenther ha scritto:
>> Update your assember - it doesn't seem to understand AVX instructions.
> Which means that on this particular opsys (CentOS 5.6) I have to build
> and install my own binutils

Well, the -mavx is added by -march=native. Thus, using -mno-avx might be
enough - or you replace -march=native by something else. (However, in
terms of performance, using AVX can make a diffence; thus, updating
binutils cannot harm.)

Tobias

[Toon Moene-3]

On 06/01/2011 03:34 PM, Salvatore Filippone wrote:

> this is CentOS 5.6, with CPU

> flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36
> clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx rdtscp lm constant_tsc ida
> nonstop_tsc arat pni monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr sse4_1 sse4_2 popcnt lahf_lm

Are you sure your kernel supports AVX (i.e., the larger
"save-the-vector-registers" area) ?

I don't see 'avx' mentioned in the flags (nor do I know for certain if
they would, but the kernel does have to find out the processor *has* AVX
in order to support its 256-bit vectors).

--
Toon Moene - e-mail: toon@... - phone: [phone]
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
At home: http://moene.org/~toon/; weather: http://moene.org/~hirlam/
Progress of GNU Fortran: http://gcc.gnu.org/wiki/GFortran#news

[Tim Prince-4]

On 6/1/2011 1:43 PM, Toon Moene wrote:

> On 06/01/2011 03:34 PM, Salvatore Filippone wrote:
>
>> this is CentOS 5.6, with CPU
>
>> flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov
>> pat pse36
>> clflush dts acpi mmx fxsr sse sse2 ss ht tm syscall nx rdtscp lm
>> constant_tsc ida
>> nonstop_tsc arat pni monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr sse4_1
>> sse4_2 popcnt lahf_lm
>
> Are you sure your kernel supports AVX (i.e., the larger
> "save-the-vector-registers" area) ?
>
> I don't see 'avx' mentioned in the flags (nor do I know for certain if
> they would, but the kernel does have to find out the processor *has* AVX
> in order to support its 256-bit vectors).
>

We have confirmed that AVX code won't run on CentOS 5.6.  As Red Hat 6.1
(recently finalized) is meant to have full support for AVX, we
anticipate availability of CentOS 6.1.  I guess 5.8 may have a degree of
AVX support.

--
Tim Prince

[Salvatore Filippone]

Il giorno mer, 01/06/2011 alle 19.43 +0200, Toon Moene ha scritto:
> Are you sure your kernel supports AVX (i.e., the larger
> "save-the-vector-registers" area) ?
>
After installing a new binutils, it worked, therefore I suspect it
does.
My complain was really that the error came  out of the blue, with no
warning at compiler configure time.
Enough time spent on this, back to rela work :)
Salvatore