Paul Winkler wrote:

>Why isn't [32 bit float] sufficient? Is it because you want to connect 
>to a 16-bit soundcard? 

No.  It is because some of my planned functions will utilize interprocess
communication of information derived from samples (one example would be
spectral data; there are others), in which the 23 or 24 bit mantissa of the
32 bit float is not sufficient.  Thirty-two bit integer -- perhaps.  But it
is so much easier just to use doubles.

>You can still write double-precision code in LADSPA plugins as long as 
>the inputs and outputs are 32-bit floats. 

Are they necessarily floats, or can they be viewed as generic 32 bit
entities?  That is to say, could I create a C-language union in my code
that takes 32 bit floats, as far as the I/O is concerned, but regards them
internally as 32 bit integers or concatenates them into doubles?

Thanks again,
Greg Berchin

Reply via email to