[gmpi] Re: Where are we?: 7.1.2

On Tue, Jun 17, 2003 at 06:55:20AM -0400, Angus F. Hewlett wrote:
> On Tue, 17 Jun 2003, Steve Harris wrote:
> 
> > I think youre missing the point slightly, currently the only thing that
> > profiles wrap is the datatype because thats the only area we've discovered
> > where paltops and desktops diverge.
> > The idea is that there would be one palmtop profile and one desktop
> > profile (and maybe others), but these could de independently versioned so
> > that eg. the paltop profile may move to float and the desktop may move to
> > doubles, at some point in the future.
> 
> Profile can also potentially encapsulate platform-binary and CPU-binary
> issues (x86 vs PPC, Linux vs Mac OS X vs WinNT/2K/XP).

Yes, I suggested this too, but its a slightly thornier issue.
 
> If we can agree to support one datatype per graph - and ultimately it
> looks like this will be worth doing for simplicity's sake, although I
> would still like to see test results on:-
> 
> - interleaved vs non-interleaved performance

Yes, this one could be important.
 
> - a long chain of plugs with high precision internal processing wired with
> normal-precision vs long-precision wires, and the resulting effect on
> signal quality and noise floor.

I have done some tests on this: http://plugin.org.uk/32vs64/
feel free to crique my test critera and conclusions, but it looks to me
that there is no useful signal quality gain from using 64 bit i/o in plugins.

The summary is: after a chain of 1000 plugins the float noise floor is
-300dB and the accumulated error is < 10^-4 dB.

The tests are of simulated chains of plugins with large gain changes at
input and output, errors are compared to the 387's 96 bit "long double"
format.

- Steve

----------------------------------------------------------------------
Generalized Music Plugin Interface (GMPI) public discussion list
Participation in this list is contingent upon your abiding by the
following rules:  Please stay on topic.  You are responsible for your own
words.  Please respect your fellow subscribers.  Please do not
redistribute anyone else's words without their permission.

Archive: http://www.freelists.org/archives/gmpi
Email gmpi-request@xxxxxxxxxxxxx w/ subject "unsubscribe" to unsubscribe

Other related posts: