[patch] Merge to trunk from Graphite branch

[Richard Guenther]

On Mon, Mar 15, 2010 at 4:12 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> On Mon, Mar 15, 2010 at 03:11:21PM +0100, Richard Guenther wrote:
>>
>> Sorry, but the biggest win for 4.5 is the various improvements
>> SUSE / AMD has contributed to improve SPEC CPU 2006 scores
>> (7% overall SPEC FP improvement).
>>
>> Updating the default x86 arch does bring you nothing (well, if
>> you're not clueless in case you shouldn't build GCC yourself).
>
>   I mentioned that because of the 2% improvement in the Polyhedron
> 2005 benchmarks which coincided with the x86 arch changes which
> enabled -msse2 across the board...
>
> http://gcc.gnu.org/ml/gcc/2010-03/msg00075.html
>
>>
>> As far as I know Graphite does not bring any performance advantages
>> yet, so I see no reason to enable it by default.
>>
>> If you want a specific benchmark to run faster you have to start
>> analyzing why it is slow (it helps to have a competing compiler
>> generate faster code), produce a testcase and at least file
>> a missed-optimization bugreport.  Of course even better if you
>> can think of a solution and implement it.
>>
>> I am not aware of any major vectorization missed-optimizations
>> vs for example the Intel compiler for SPEC CPU 2006.
>
>    Don't we still lack the ability to vectorize the cases described
> here?
>
> http://gcc.gnu.org/ml/gcc/2009-08/msg00057.html
>
> Perhaps current graphite already would handle some of those if the
> vectorization wasn't getting accidentally lost.

This case doesn't happen in hot areas of SPEC CPU 2006.

Richard.