Failing benchmark / Compile farm

[Tobias Grosser]

Hi Konrad,

I attached you the interesting part of the run log. However the
config.log is already deleted. However I do not add anything than the
flags necessary to find ppl/cloog/mpfr/gmp.

To get an compile farm account, look at this website.

http://gcc.gnu.org/wiki/CompileFarm#head-0db7ff08ecf4b66b4d73df10049ae56d8203b97c

Tobi

[David Edelsohn]

On Wed, Jul 1, 2009 at 11:07 AM, Tobias
Grosser<gro...@fim.uni-passau.de> wrote:
> Hi Konrad,
>
> I attached you the interesting part of the run log. However the
> config.log is already deleted. However I do not add anything than the
> flags necessary to find ppl/cloog/mpfr/gmp.

Are both Konrad and you testing Intel or AMD systems? If GCC
config.guess discovers slightly different systems, it could cause
subtly different code generation.

David

[Konrad Trifunovic]

2009/7/1 David Edelsohn <dje...@gmail.com>:

I'm testing on Intel Core 2.
But this is immaterial. Enabling the legality check for
identity transformation (no-transformation) does not affect the code
generation. When enabled, it causes ICE on this benchmark (according
to tests ran by Tobias).

I would rather look whether this function call is activated on my
machine at all.
What I suspect is the set of the flags that is used.
The set of the flags can determine what is happening in the passes
before Graphite Scop detection. It probably seems that I use flags
that do not trigger any Scop detection, thus my code would not be called at all.

I will figure that out.

Anyway, I asked for an account at compilerfarm.

Konrad

> >
>

[Tobias Grosser]

On Wed, 2009-07-01 at 11:38 -0400, David Edelsohn wrote:
> On Wed, Jul 1, 2009 at 11:07 AM, Tobias
> Grosser<gro...@fim.uni-passau.de> wrote:

> > Hi Konrad,
> >
> > I attached you the interesting part of the run log. However the
> > config.log is already deleted. However I do not add anything than the
> > flags necessary to find ppl/cloog/mpfr/gmp.
>

> Are both Konrad and you testing Intel or AMD systems? If GCC
> config.guess discovers slightly different systems, it could cause
> subtly different code generation.

I test on AMD systems. Namely gcc17 on the compile farm.

The CPU is:

processor : 7
vendor_id : AuthenticAMD
cpu family : 16
model : 2
model name : Quad-Core AMD Opteron(tm) Processor 8354
stepping : 3
cpu MHz : 2194.594
cache size : 512 KB
physical id : 1
siblings : 4
core id : 3
cpu cores : 4
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov
pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt rdtscp
lm 3dnowext 3dnow constant_tsc pni monitor cx16 lahf_lm cmp_legacy svm
cr8_legacy
bogomips : 4389.30
TLB size : 1024 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 48 bits physical, 48 bits virtual
power management: ts ttp tm stc [6] [7] [8]

Tobias

[Konrad Trifunovic]

I finally found it out.
Sorry for disturbing so much.
I made a very stupid mistake: I was changing the sources in one directory,
and testing the other one....

I can now reproduce the bug on gfortran.dg/cray_pointers_2.f90.

thanks,
and sorry again,

Konrad

2009/7/1 Tobias Grosser <tobi-g...@web.de>: