Michael Cheng writes:
> > Do you have any idea why would LAMEv2.1 segfault when compiled with
> > the pgcc compiler, but work fine when compiled with the good old gcc
> > 2.7? This is sad, because for I got a 15% improvement in speed for
> > 8hz-mp3 by just recompiling the thing with pgcc and -mpentium.
>
> Cool. I wish the freebsd port of pgcc was still happening :(
> Anyone got a working pgcc on there freebsd system? (maybe I should just
> use the linux one under linux emu...)
> I just remembered this morning on the way to uni that there is a memory
> allocation error in the code. I'll look it up... There's not enough mem
> alloced for stereo buffers. It'll work fine in mono though. Give it a
> try.
I just tried both downmixing a stereo sample to mono ("-m m"), and
encoding a mono sample, always with the same tragic results. :-(
> > On the same subject of optimizations - for those guys who have AMD K6,
> > remove the -malign-* options from the Makefile before building; they
> > just slow the program on the K6.
>
> Handy tip. thanks. the maligns improve performance on freebsd which is
> why they're in there :)
Well, it doesn't improve by a lot - encoding a 14-sec sample went down
from ~20.05 to ~19.996 seconds after this change. In any case, the K6
sucks for MP3 encoding. These results were achieved with a 300-mHz
processor... Probably it's the FPU to blame.
--
Alex Shnitman
[EMAIL PROTECTED], [EMAIL PROTECTED]
http://alexsh.home.ml.org
