5 Aug
2007
5 Aug
'07
5:33 p.m.
Mike Taht wrote:
I completely rewrote tap eq while on vacation last
month to be pure
sse2, using doubles for extra resolution, with no denormal problems,
and, at least on x86_64, far greater speed, especially for the stereo
version.
Excellent work, I was hoping someone would write an EQ plugin using SSE2
(I don't know of any existing ones).
I have tested extensively on my own machine but have
not cleaned up
the code enough to justify a formal release, perhaps our efforts can
be synced....
My own 'efforts' don't amount to much so far, but I'm working on a
test-bed to analyse LADSPA plugins, and I can continue to monitor your
snapshots for any issues.
One bug I found so far, is that if I call run() with a block size of 1,
do_eq() is run with sample_count of 2. This causes the 'output' buffer
to be overrun inside of the switch(numfilt) block.
Also I have to ask whether both the original TAP plugins and your own,
should set HINT_LOGARITHMIC for the frequency and/or gain parameters?
And I'm not sure what msp_eq.c.good is for, maybe that wants deleting?
Anyway, please keep up the good work.
Paul Davis wrote:
you'd better make sure you turn off Ardour's
denormal protection if
you want to duplicate it. This was added to 2.0.3 IIRC.
That's a good point, but I had that turned off.
Regards,
--
Steven Chamberlain
steven(a)pyro.eu.org