Subject: Re: [linux-audio-dev] Simple Plugin API: In/Out Ports
From: Eli Brandt (eli_AT_t.gp.cs.cmu.edu)
Date: ti maalis 07 2000 - 11:35:48 EST
Paul Winkler wrote:
> And I think this has been mentioned before, but we should DEFINITELY
> try to keep it simple to compile 64-bit versions of hosts and
> plugins. I don't know if this takes more than a typedef, but it
> needs to be considered now.
I like to have a fastfloat_t and a smallfloat_t, since even when
double-precision is faster (IBM PPC) I may not want to store my tracks
that way in memory.
> During discussions of double vs.
> single-precision floats on the csound mailing list, several more
> knowledgeable persons than I asserted that even on
> current-generation intel hardware, double-precision arithmetic isn't
> really slower than single-precision,
On an x86, everything becomes a 10-byte long double when it's loaded
into the FPU. In fact, long doubles are slightly faster to load.
Shorter data is still faster overall because PCs persist in having lousy
bandwidth to memory. On other processors, I'm not keeping track -- it
used to be that single-precision was often faster than double, but the
trend is away from this.
As long as we're talking data types, have I mentioned these recently?
http://reality.sgi.com/dscott/audio.apps/dev/sp.html
http://www.cnmat.berkeley.edu/~adrian/Csigproc.html
They're both MIPS-centric, but generalizable.
--
Eli Brandt | eli+@cs.cmu.edu | http://www.cs.cmu.edu/~eli/
This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:23:28 EST