On 11-Jun-99 Frank Baumgart wrote:
> About the most expensive in quantize() seem to be the sqrt()
> calculations.
Yes. been thinking about finding a faster approximate sqrt for a while.
> Andrew Glassner
Glassner Rocks! Check out his column in IEEE computer graphics and
applications. Pure genius.
> Sorry, my time is extremely limited right now and I do not think I can
> follow this or make a substitute for lame in the next week. But I will
> look again for the author of the algo next week, maybe there is
> something on the web.
I'm still marking all this weekend, then camp next week. But then school
holidays for two weeks :) One week will be programming :)
> The efficiency of the number of comparisons with constants (temp <
> 5.5xxx) is directly related to the number of available FP registers,
Yeah. I noted this in my changes to the 8hz-mp3 code, but my comments have been
ditched in LAME. I tuned it (sort of) for performance on my p166mmx. YMMV.
Maybe a #def in the makefile for particular architectures. Which means
everyone has to tune it on their system. Or maybe I could write a quick test
file that determines the optiminum by trying all combinations.
"more speed!"
later
mike
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
