Subject: Re: [linux-audio-dev] Re: more bad low latency results
From: Andre Pang (ozone_AT_algorithm.com.au)
Date: Sat Nov 17 2001 - 05:32:51 EET
On Fri, Nov 16, 2001 at 10:14:28AM +0100, Maarten de Boer wrote:
> > CONFIG_MK7=y
> >
> > Did I read correctly that some 3DNow instructions cause latency issues?
>
> That's what Roger Larsson suggested when I showed bad latency results.
> Compiling for 586 solved that problem for me, and I find it very strange
> you got good results compiling for CONFIG_MK7=y
Ah, one other thing you may want to try, and which probably won't
help but is worth a shot anyway, is to adjust the latency timer
for your PCI devices.
You want to adjust the PCI latency timer for your soundcard to a
_high_ value (counter-intuitive, I know), and everything else low.
1. cat /proc/pci, find the bus, device and function numbers for
your soundcard. e.g. mine looks like:
Bus 0, device 12, function 0:
Multimedia audio controller: Yamaha Corporation YMF-744B [DS-1S Audio Controller] (rev 2).
IRQ 11.
Master Capable. Latency=248. Min Gnt=5.Max Lat=25.
Non-prefetchable 32 bit memory at 0xefff8000 [0xefffffff].
I/O at 0xff00 [0xff3f].
I/O at 0xfefc [0xfeff].
2. Take note of the latency setting; in my example above, the
latency figure is 248, which is the maximum for the PCI bus.
You may find that yours is much lower, such as 32 or 64.
Change the latency setting for the soundcard:
setpci -s 00:0c.0 latency_timer=f8
| | +- function number
| +---- device number
+------- bus number
The function/device/bus numbers you specify to setpci are in
hex; e.g. my bus/device/function is 0, 12, and 0, so it looks
like 00:0c.0 as the setpci parameter.
3. Adjust the PCI latency timer for other cards in your system
to something much lower if they're currently very high. In
particular, your video card will probably have a latency timer
of 248. I reset mine to 128.
As I said, it probably won't help, but it's definitely worth a
shot.
-- #ozone/algorithm <ozone_AT_algorithm.com.au> - trust.in.love.to.save
This archive was generated by hypermail 2b28 : Sat Nov 17 2001 - 05:36:24 EET