6 Feb
2003
6 Feb
'03
5:37 p.m.
Steve Harris wrote:
seems to be a siuggestion in the PTAF document that host may have to be
responsibe for killing denormal numbers anyway, so they will have to snoop
controls in that case.
More exactly, there are properties telling the host if :
- Plug-in accepts denormalized incoming audio stream
without cpu load penalty.
- Plug-in doesn't output denormal numbers in audio stream,
or just a few.
This gives the host hints to properly place denormal killers,
if needed.
About silent buffers, I will probably add to PTAF a third,
intermediate buffer state, marking it as "virtually silent
but actually full of garbage". Encapsulated at a higher
level, it becomes the "defered zeroing" concept, seen in
the DXi2 API. It can improve a bit the CPU load dedicated
to audio routing.
-- Laurent
==================================+========================
Laurent de Soras | Ohm Force
DSP developer & Software designer | Digital Audio Software
mailto:laurent@ohmforce.com | http://www.ohmforce.com
==================================+========================