Re: [Bug-gnubg] Benchmarks, experiments, speedups

[Ingo Macherius]

I got figures on the impact of SSE3. All data from the same machine, which runs 
a P4-F ("Prescott") using my beta-2 binary.

 386: 21800 eval/s  (100%)
sse : 28600 eval/s  (131%)
sse2: 29800 eval/s (137%)
sse3: 31400 eval/s (144%)

So not unexpectedly the gain from 386 to SSE is largest (about 1/3), and SSE2 
and SSE3 each add about 6-7% to that.

For an intrinsic implementation this probably means that restricting oneself to 
SSE (rather than producing SSE2 and SSE3 versions as well) is good enough. See 
the 80:20 rule :)

Ingo






______________________________________________________________
Verschicken Sie romantische, coole und witzige Bilder per SMS!
Jetzt bei WEB.DE FreeMail: http://f.web.de/?mc=021193