Re: spring and -march=i686

[Marco Amadori <marco.amadori@gmail.com>]

On Monday 25 January 2010 23:55:43 Paul Wise wrote:
> Has anyone done any benchmarking/testing to see if -march=686 makes
> any difference?


If the difference between i486 and i686 is just cmov, someone has a bad opinion 
on it:

http://ondioline.org/mail/cmov-a-bad-idea-on-out-of-order-cpus

> I suspect the GPU probably matters more than the CPU?

OpenGL 1.4 is needed, and probably "SSE is needed to have not online desyncs" 
upstream said.
 
> Usually binaries have to work on the base CPU instruction set for an
> architecture, IIRC that is 486-level instructions for i386. I imagine
> detecting CPU support for SSE at runtime is an acceptable way to make
> one set of binaries.

This already happens in latest binaries, it still lacks i486/i686 checks. 
(although is really unprobable that one having an OpenGL card is using an i486 
CPU).

-- 
ESC:wq