[PATCH][4.3][RFT] Fix PR36519 - deal with unpartitionable but call-clobbered SFTs

[Richard Guenther]

On Wed, Jun 18, 2008 at 7:33 AM, Paolo Bonzini <bonzini@gnu.org> wrote:
> Paolo Bonzini wrote:
>>
>> Paolo Bonzini wrote:
>>>
>>> Paolo Bonzini wrote:
>>>>
>>>>> Any better ideas?  Any volunteers for SPEC-testing this patch vs.
>>>>> the 4.3.0 release?
>>>>
>>>> I can try.
>>>
>>> Uhm, it got stuck on galgel.  In other words, 4.3.0 gives an infinite
>>> loop on galgel for i686-pc-linux-gnu. :-(
>>>
>>> Sorry for not noticing before -- I was travelling.  I restarted it now,
>>> I'll get SPECint2000 later today and SPECfp2000 tomorrow morning.
>>
>> Time flies... in the meanwhile I got SPECint results, they are mostly
>> neutral with slight loss on eon/perlbmk and slight gains on gcc/gap.
>
> SPECfp (except galgel) is -1% with small losses across almost all benchmarks
> (-5% on equake, -3% on lucas).

So, what do people think?  Is it worth to enable building those testcases at -O2
for this cost?  I would obviously say yes.

Richard.