[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
On Tue, 2004-04-20 at 08:00, Chris wrote:
> There's about two things left to do. The first.. given being able to see
> the code between the low quality and hq1 mixers, as well as the
> profiling provided previously, would anyone object to removing the low
> quality mixer? The information I have strongly suggests there is no
> longer a need for it, as well it would haelp to trim down the code some
> more. If I don't hear any sustainable arguments to keep it, I'll remove
> it for the next version.
Just did another short speed comparison. I hacked jack.c to not transfer
any sound, so it should get a NULL driver like Bob suggested. Didn't
seem to work though.
This is the reported user CPU time with the hacked jack.c, which just
calls mix_some_samples (always result of 2 runs, with 16it, stereo,
44100 Hz):
OLD NEW
0 3330 3340 3240 3240
1 4510 4480 3780 3800
2 4550 4580 4390 4390
This is the normal ALSA driver - which is much faster although it
transfers and plays the sound :)
OLD NEW
0 2200 2210 2210 2220
1 2570 2570 2240 2240
2 3060 3060 3060 3060
So I also modified the ALSA driver to just have mix_some_samples in its
callback, not playing anything:
OLD NEW
0 2530 2530 2540 2540
1 3370 3370 2530 2530
2 3810 3810 3810 3810
So, although I again have no idea *what* I was measuring, at least in
the JACK case something was measured.. and your new mixer was faster
with all 3 qualities.
With the ALSA driver, it looks just like last time. Everything is faster
than the jack driver (although I removed the actual transfer), but the
new mixer only was faster than the old with quality 1. Surprisingly, not
transfering data to ALSA needed more CPU than transfering.
In any case, just looking at the ALSA results: quality 0 and quality 1
take about the same CPU with your new mixer, and not really more than
the original quality 0. Quality 2 had exactly the same speed in all 4
runs, no matter if I used the patched mixer or not. So I think, from the
CPU usage point of view, it can go in, and quality 0 can be dropped.
(But this should probably get tested better.)
--
Elias Pschernig <elias@xxxxxxxxxx>