Jakub's Recommendations for ia32 Support

[Josh Boyer]

On Wed, Feb 04, 2009 at 01:40:10AM +0100, Dominik 'Rathann' Mierzejewski wrote:
>On Tuesday, 03 February 2009 at 22:08, Josh Boyer wrote:
>> On Tue, Feb 03, 2009 at 09:57:40PM +0100, Dominik 'Rathann' Mierzejewski wrote:
>> >On Tuesday, 03 February 2009 at 21:52, Josh Boyer wrote:
>> >> On Tue, Feb 03, 2009 at 09:45:46PM +0100, Dominik 'Rathann' Mierzejewski wrote:
>> >[...]
>> >> >Well, I talk to people who write hand-optimized assembly and care to
>> >> >squeeze every cycle out of various CPUs and they say it's definitely
>> >> >a win. So please, show me some code instead of hand-waving.
>> >> 
>> >> If they can do that, then why can't they rebuild things themselves?
>> >
>> >Oh, right. Let's all use LFS or Gentoo.
>> >Why can't everyone rebuild things themselves?
>> 
>> They can.
>> 
>> Look, it's very simple.  Fedora _cannot_ be everything to everyone.  There
>> are too many variations, optimizations, etc.  So, in light of that, it
>> tries to fit the best usecase.  The common ground.
>> 
>> People that hand code assembly files are _not_ the common users of Fedora.
>
>I know. I'm just saying Ulrich's statement about "cmov not being a win"
>is not convincing without some code samples. I, on the other hand, can
>point you to x264 cabac asm code, which uses cmov to attain 8%(p4)
>to 25%(core2) speedups.

I can believe that.  So why can't you build that codec code with the proper
flags to include the support you need?

Ulrich's statement was about the general case again.  Throwing around
specific examples as counter points isn't going to be very productive unless
those examples _are_ the general cases.

josh