It might be that some hosts access the raw Csound buffers, i.e. the
spin and spout buffers, directly. They contain one kperiod worth of
audio. It might be that other hosts do not do this, but rely on the
intermediate buffers, whose size is controlled by -b. These two
methods might possibly behave differently.

I recommend that programmers putting Csound into hosts use spin and
spout directly. This affords the lowest latency. However this is the
"lower half" of the audio driver. The host's own send and receive
buffers would be the "upper half" of the audio driver and might have a
very different size/latency. This is not a problem for feeding audio,
the receive and send loops for copying data into and out of spin and
spout would have one set of indices for spin and spout, and another
set of indices for the host output and input buffers (see
CsoundVST.cpp for an example).

If the plugin uses -b then it is possible that Csound's buffers and
the host buffers need to be aligned in size. Or not. It depends on how
the plugin is programmed.

Hope this helps,
Mike

On Mon, Feb 20, 2012 at 10:53 AM, Bernt Isak Wærstad
 wrote:
> This mail follows up the discussion I had with jpff and Victor in Hannover
> at the Csound conference concerning using different buffer sizes at the same
> time. Since this goes out to all I should briefly explain the backdrop.
>
> In my live setup I use Ableton Live as a host combining different processing
> from VSTs, Max, Lives internal effects and Csound where Hadron is a key
> piece of the setup. The problem is that with Hadron I really can't have a
> low buffer size without the sound starting to crack up. At first I thought
> this was natural since it's quite CPU intensive indeed, but then
> I realized that it only uses about 14-15% for one instance and with VSTs,
> internal effects and some of the other Csound effects I could go as high as
> 70% before the audio started to break. Since Hadron sort of introduces a
> minimum latency because of the granular synthesis and the fact that I'm
> mostly using it as a  time based "send effect", I didn't mind having high
> buffer size with lots of latency, but the problem is combining Hadron with
> serial/insert effects (like a phaser for instance) in the same setup.
>
> So that's why I asked John and Victor if it was possible to have different
> buffer sizes at the same time (high latency and low latency tracks). Working
> with Csound through a host disables -b and -B if I understood correctly, so
> setting a ksmps higher than host buffer size is the only way to force Csound
> to have higher buffer size than the host. That is ksmps=Csound buffer size
> if ksmps>host buffer size. I've tested this and it seems correct (with host
> buffer size=64 and ksmps=256, more latency was introduced on the track with
> audio straight through Csound than on a clean track with audio straight
> through), but the problem is that it doesn't help. While investigating this,
> I discovered that routing the audio through Jack or Max for Live in edit
> mode (with preview on of course) does help!
>
> I haven't tested this on Hadron, but on a simple pvsblur example (attached)
> which produced the same errors. Here's a list of my test results:
>
> Csound in Max for Live or CsoundPlugin in Live: Live buffer size=64,
> ksmps=32 : NOT OK
> Csound in Max for Live or CsoundPlugin in Live: Live buffer size=64,
> ksmps=64 : NOT OK
> Csound in Max for Live or CsoundPlugin in Live: Live buffer size=64,
> ksmps=128 : NOT OK
> Csound in Max for Live or CsoundPlugin in Live: Live buffer size=128,
> ksmps=32 : OK with blur time up to ~0.6
> Csound in Max for Live or CsoundPlugin in Live: Live buffer size=256,
> ksmps=32 : OK
>
> Csound in Max for Live in Live - in preview mode (patch open for editing):
> buffer size=64, ksmps=32 : OK
>
> Reaper with Sigurds Csound VST Plugin:
> On a playback track in Reaper: Reaper buffer size=64, ksmps=32 : OK
> On a input track (record enable and monitoring) in Reaper: Reaper buffer
> size=64, ksmps=32 : NOT OK
> On a input track (record enable and monitoring) in Reaper: Reaper buffer
> size=64, ksmps=64 : NOT OK
> On a input track (record enable and monitoring) in Reaper: Reaper buffer
> size=64, ksmps=128 : NOT OK
> On a input track (record enable and monitoring) in Reaper: Reaper buffer
> size=128, ksmps=32 : OK with blur time up to ~0.6
> On a input track (record enable and monitoring) in Reaper: Reaper buffer
> size=256, ksmps=32 : OK
>
> Connecting Csound and Live through Jack: Jack buffer size=64, ksmps=32 : OK
>
> Csound standalone: -b32, -B64, ksmps=16 : NOT OK, but fewer crackle
> artifacts than with Reaper or Live
> Csound standalone: -b32, -B64, ksmps=32 : NOT OK, but fewer crackle
> artifacts than with Reaper or Live
> Csound standalone: -b64, -B128, ksmps=32 : NOT OK, but almost
> Csound standalone: -b128, -B256, ksmps=64 : OK
>
> Csound standalone through Jack: Jack buffer size=64, ksmps=32 : OK
>
> --
> Mvh.
>
> Bernt Isak Wærstad
>
>
>



-- 
Michael Gogins
Irreducible Productions
http://www.michael-gogins.com
Michael dot Gogins at gmail dot com