[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
On Thu, 2004-04-22 at 14:57, Ben Davis wrote:
> On Thursday 22 April 2004 11:21, Elias Pschernig wrote:
> > OLD NEW
> > 0 3330 3340 3240 3240
> > 1 4510 4480 3780 3800
> > 2 4550 4580 4390 4390
>
> There still seems to be a difference between quality 0 and quality 1.
>
> In the later results, the difference is probably offset by the conversion that
> has to happen at the end with quality 1. However, that is a constant-time
> operation (per block of samples), and the actual mixing scales with the
> number of voices.
>
> It would be instructive to do some measurements with fewer and more voices
> being mixed. Go all the way up to Allegro's maximum of 64.
>
I uploaded my test to: http://allefant.sf.net/allegro/mixer.zip. The
problem with voices is, if I use fewer, it gets too fast. Maybe I should
let it run longer. The above numbers all are with 1024 voices mixed
together at a time (My aintern.h and digi.h are modified accordingly.).
I just create as many samples, and let them loop for some seconds. With
64 voices CPU stays at about 1%, which is below the normal fluctuation
of all the running background processes. And the user-time bash reports
doesn't say anything meaningful in that case.
I guess, someone with a very slow CPU should test it.
--
Elias Pschernig <elias@xxxxxxxxxx>