This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: Some SPECfp results for the simplify-rtx problem


> 
> Fortran 77 tests: 168.wupwise, 171.swim, 172.mgrid, 173.applu,
> 200.sixtrack, 301.apsi.  
> C tests (for reference): 177.mesa, 179.art, 183.equake, 188.ammp
> 
> Unfortunatly both compilers have errors and 177.mesa, 188.ammp and
> 301.apsi fail with an "Output miscompare".
> 
> I now run on the same machine tests with GCC with RTH's "address_cost
> tweak" installed:
Please can you run it with the strength-reduction tweek I pointed out
in the reply to Richard's email?  I plan to install it today, but I want
automated checker to come with results before I do so. (and the patch
needs re-bootstrapping as it aged somewhat)

It should track the same problem from other direction and thus I would
like to see effect of Richard's change on the updated code.

Also please run the integer tests too.  The effect of Richard's patch can
be increase of register pressure in the integer code that can negatively affect
the performance.
> 
>      Base flags: -O2 -march=athlon -malign-double
>      Peak flags: -O2 -march=athlon -malign-double -fno-reduce-all-givs
>                                      Estimated                     Estimated
>                    Base      Base      Base      Peak      Peak      Peak
>    Benchmarks    Ref Time  Run Time   Ratio    Ref Time  Run Time   Ratio
>    ------------  --------  --------  --------  --------  --------  --------
>    168.wupwise       1600     336         477*     1600     339         471*
>    171.swim          3100     980         316*     3100     783         396*
>    172.mgrid         1800     518         348*     1800     533         338*
>    173.applu         2100     694         303*     2100     654         321*
>    179.art           2600    1730         150*     2600    1709         152*
>    183.equake        1300     339         384*     1300     340         383*
>    200.sixtrack      1100     520         211*     1100     491         224*
Hmm, looks like we are at something...
> 
> Btw. Honza just committed a fix for "fcmov" which might fix the
> failures I see.
Hope so.  The specint misscomparisons are gone.

Honza
> 
> Andreas
> -- 
>  Andreas Jaeger
>   SuSE Labs aj@suse.de
>    private aj@arthur.inka.de
>     http://www.suse.de/~aj


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]