Pick up sfront version 0.51 01/01/00 at:
http://www.cs.berkeley.edu/~lazzaro/sa/
From the change log:
Version 0.51 -- 01/01/00
[1] Extensive optimizations
and bug fixes for the
filter, spectral analysis,
gain control, sample
conversion, delay, and
effects core opcodes, the
pluck and grain signal
generator core opcodes,
and the spatialize statement.
[2] All a-rate core opcodes
now function correctly when
enclosed in if and while
statements with a-rate guards
(thanks to Ross Bencina).
-----
Highlights of the new release:
-*- Core opcodes fft() and ifft()
were rewritten, implementing a
reasonably-fast FFT algorithm.
An fft/ifft pair doing a 512
point FFT with a 256 point
sliding window runs at about
2 microseconds per audio sample
processed on a 450 Mhz PIII,
supporting about 10 concurrent
pairs at 44.1 kHz in batch
rendering.
-*- The fracdelay opcode is
significantly faster as well,
resulting in this speedup for
the elpelele.saol example:
sfront 0.50: user 272.20sec PIII 450Mhz
sfront 0.50: user 226.99sec PIII 450Mhz
-*- The spatialize statement
runs much faster now, as shown
by this performance comparison
of the spatialize-dependent
example program perc.saol in
the sfront distribution
sfront 0.50: user 37.88sec, PIII 450Mhz
sfront 0.51: user 19.17sec, PIII 450Mhz
-*- Other opcodes with speedups in
the 10-20% or greater range include
pluck, grain, port, firt, iirt, rms,
downsamp.
This release concludes optimization
work on the heavyweight core opcodes.
The focus is now returning to optimizing
the code generated by core language elements
in SAOL.
Oh, and Happy Y2K! Some of our servers
actually hung here during the date rollover,
I had to reboot my Linux machine this morning
to get it to see the disks properly ...
--jl
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 11:46:36 EST