Subject: Re: [linux-audio-dev] Athlon/3dnow! dsp benchmark results
From: est_AT_hyperreal.org
Date: ma syys 06 1999 - 11:13:15 EDT
Benno Senoner discourseth:
>
> Eric: are you sure that the biquad filter is coded in an optimal way.
> Maybe you have to reorder things ?
Well, it's pretty optimal for the k6-2. I sweated over it a *lot* but
couldn't get it above that 3.9-fold speedup even though I had a spare
register to play with. I tried various unrollings and interleavings
and many, many reorderings (even some that were incorrect).
> I can't believe that 3Dnow is slower than optimized C on the Athlon.
..or that biquad/3dnow is slower on an Athlon/500 than a K6-2/350? I
found it shocking but easier to believe than that Dieter is hoaxing
me. Again, the Athlon has greater latency for 3dnow instructions.
Eric
This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:27:11 EST