[wdmaudiodev] Re: WavePci vs. WaveCyclic (was: Already got cool latency figures - how about everyone else)
- From: "Frank Berreth" <fberreth@xxxxxxxxxxxxx>
- To: <wdmaudiodev@xxxxxxxxxxxxx>
- Date: Fri, 31 Jan 2003 10:22:39 -0800
If I'm not totally mistaken, then you are required to handle
byte-alignment if you can play 8bit data, WORD-alignment if you can play
16bit data etc.
Do NEVER use option 3. This will be just as "bad" as wave cyclic with
the additional complexity of wave pci. The real question if you can use
wave pci or not is this:
Does HW support 64bit addressing?
Does HW support sample-split over page boundaries (6ch playback)?
Can you handle the alignment issues.
I'm guessing your HW does not support 64bit addressing (like most do),
therefore go to wave cyclic. Please do honor SetNotificationFreq on
WaveCyc. Portcls might want to change this from 10ms down to some other
frequency.
This posting is provided "AS IS" with no warranties, and confers no
rights.
-----Original Message-----
From: Daniel E. Germann [mailto:deg@xxxxxxxx]=20
Sent: Friday, January 31, 2003 7:39 AM
To: WDM Audio Developers List
Subject: [wdmaudiodev] WavePci vs. WaveCyclic (was: Already got cool
latency figures - how about everyone else)
One of our clients wants us to look into the complexity of writing a WDM
Audio driver for a PCI card that has a scatter/gather bus-master PCI DMA
controller. Unfortunately, the DMA controller has a restriction that
the
buffer starting address must be DWORD-aligned. The end of the buffer
can be
in the middle of a DWORD, but the start of the buffer must be
DWORD-aligned.
I read in the Audio Device Performance white paper
(http://www.microsoft.com/hwdev/tech/audio/highperf-drv.asp) that, "A
WavePci device is able to perform DMA transfers to or from buffers that
can
be located at arbitrary memory addresses, and that begin and end with
arbitrary byte alignments."
I can understand why the spec would be written that way. But does that
ever
really happen? We've looked at many, many applications, and all of them
use
buffers that are DWORD-aligned. Granted, the addresses we were looking
at
came right from the app, so there was no WDM/KMixer involvement. It may
be
that WDM/KMixer does use non-aligned buffers. Is this a case of the
specification being written to be "powerful and general", or is there a
real-world reason for this requirement?
Given that WavePci requires the hardware to handle arbitrary alignment,
it
looks like there are at least three options. (1) Redesign the hardware
so
it can handle buffers with arbitrary alignment. The hardware engineer
tells
me that will greatly increase the complexity of his DMA controller
design
(and take a lot of debugging time). (2) Use WaveCyclic instead of
WavePci,
which would result in a lot of needless memcpy()s, and incur the 10 ms
WaveCyclic latency Dirk mentioned. (3) Set up WavePci to run with a
driver-allocated DWORD aligned buffer and memcpy() to/from it, which
would
get rid of the 10 ms latency but still do the needless memcpy()s (not
100%
sure this will work).
What are other people doing? Does everyone else using WavePci have
"smarter" DMA controllers? ;-)
Best regards,
-Dan
> From: "Van Mieghem, Dirk" <dvm@xxxxxxxx>
> Subject: [wdmaudiodev] Re: Already got cool latency figures - how abou
> Date: Thu, 30 Jan 2003 19:18:26 +0100
>
> Hi Tom,
>
> > I've been using smaller buffers with WaveCyclic with no unusual
> > problems. Granted, they don't work reliably at very low latencies,
but I
> > thought that was my old, slow system.
> >
> > Can this have dire consequences I'm not aware of?
>
> It's been a while since I used WaveCyclic. My first driver used it
because
I
> liked the straightforward approach of WaveCyclic compared to the
comlicated
> way WavePCI deals with buffers. IMO, scatter/gather techniques are too
much
> overhead while using very small buffers (low latency). I remember that
the
> WaveCyclic DMA buffer couldn't go below 10ms because PortCls copies
the
data
> in and out of this buffer. PortCls is handling the Allocator Framing
to
the
> application and the miniport can't change this. However, it's possible
that
> this behavior changed with a Service Pack update.
>
> Dirk
******************
WDMAUDIODEV addresses:
Post message: mailto:wdmaudiodev@xxxxxxxxxxxxx
Subscribe: =
mailto:wdmaudiodev-request@xxxxxxxxxxxxx?subject=3Dsubscribe
Unsubscribe:
mailto:wdmaudiodev-request@xxxxxxxxxxxxx?subject=3Dunsubscribe
Moderator: mailto:wdmaudiodev-moderators@xxxxxxxxxxxxx
URL to WDMAUDIODEV page:
http://www.wdmaudiodev.de/
******************
WDMAUDIODEV addresses:
Post message: mailto:wdmaudiodev@xxxxxxxxxxxxx
Subscribe: mailto:wdmaudiodev-request@xxxxxxxxxxxxx?subject=subscribe
Unsubscribe: mailto:wdmaudiodev-request@xxxxxxxxxxxxx?subject=unsubscribe
Moderator: mailto:wdmaudiodev-moderators@xxxxxxxxxxxxx
URL to WDMAUDIODEV page:
http://www.wdmaudiodev.de/
Other related posts:
