From: Andre Schwarz <hidden> Date: 2008-07-07 09:40:25
Grant,
do you know if someone's working on a more generic DMA solution using
BestComm engine on 5200B ?
Maybe somthing that accepts a sg-list with callback ops or completion ?
Is it possible right now to accelerate simple memcpy ops ?
regards,
Andre
MATRIX VISION GmbH, Talstraße 16, DE-71570 Oppenweiler - Registergericht: Amtsgericht Stuttgart, HRB 271090
Geschäftsführer: Gerhard Thullner, Werner Armingeon, Uwe Furtner
From: Matt Sealey <hidden> Date: 2008-07-07 12:47:10
Andre Schwarz wrote:
Grant,
I know I'm not Grant, but..
do you know if someone's working on a more generic DMA solution using
BestComm engine on 5200B ?
Maybe somthing that accepts a sg-list with callback ops or completion ?
It was suggested once or twice, not least by me.
Is it possible right now to accelerate simple memcpy ops ?
From my discussions on the subject with Sylvain, it's possible, you just
need to use the GenBD and initiate it manually, however it's probably more
overhead than work if the data size is small, and BestComm would do better
to copy full 32-bit words at a time, and stay aligned, if you have an odd-sized
transfer from a non-32-bit aligned address, then you may have to do some
incredible amount of work which makes the actual transfer not worth doing
(by the time you set it up, the CPU could have copied it on it's own
already, I guess what you DO gain is a kernel preemption point.. the CPU
can do other things that are important)
I for one, though, whether it speeds stuff up or not, love to see this in
action and am very willing to test and benchmark it. I would love to see
more users, too, as the network stack is not the only system that can
benefit..
On a related note does anyone know of the status or what is going on with
Clifford Wolf's dmatransfer API?
--
Matt Sealey [off-list ref]
Genesi, Manager, Developer Relations
From: Jon Smirl <hidden> Date: 2008-07-07 13:39:45
On 7/7/08, Matt Sealey [off-list ref] wrote:
Andre Schwarz wrote:
quoted
Grant,
I know I'm not Grant, but..
quoted
do you know if someone's working on a more generic DMA solution using
BestComm engine on 5200B ?
Maybe somthing that accepts a sg-list with callback ops or completion ?
It was suggested once or twice, not least by me.
quoted
Is it possible right now to accelerate simple memcpy ops ?
From my discussions on the subject with Sylvain, it's possible, you just
need to use the GenBD and initiate it manually, however it's probably more
overhead than work if the data size is small, and BestComm would do better
to copy full 32-bit words at a time, and stay aligned, if you have an
odd-sized
transfer from a non-32-bit aligned address, then you may have to do some
incredible amount of work which makes the actual transfer not worth doing
(by the time you set it up, the CPU could have copied it on it's own
already, I guess what you DO gain is a kernel preemption point.. the CPU
can do other things that are important)
If you want an Efika specific problem, the audio hardware is capable
of simultaneously playing music on the S/PDIF and analog outputs. But
to do that the samples have to be alternated as they are fed into the
AC97 stream. I think the codec can capture that way too but you
didn't put a transceiver on the S/PDIF line.
In my test driver you only get AC97 or S/PDIF. Grant and I are both
working on i2s drivers, when we get those sorted out it shouldn't be
too hard to add ac97 back in. That codec driver I sent you was about
95% complete.
I for one, though, whether it speeds stuff up or not, love to see this in
action and am very willing to test and benchmark it. I would love to see
more users, too, as the network stack is not the only system that can
benefit..
On a related note does anyone know of the status or what is going on with
Clifford Wolf's dmatransfer API?
--
Matt Sealey [off-list ref]
Genesi, Manager, Developer Relations
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev
From: Matt Sealey <hidden> Date: 2008-07-07 14:18:45
Jon Smirl wrote:
On 7/7/08, Matt Sealey [off-list ref] wrote:
quoted
to do that the samples have to be alternated as they are fed into the
AC97 stream. I think the codec can capture that way too but you
didn't put a transceiver on the S/PDIF line.
OT, but the IDT STAC9766 doesn't support S/PDIF input, only output.
In my test driver you only get AC97 or S/PDIF. Grant and I are both
working on i2s drivers, when we get those sorted out it shouldn't be
too hard to add ac97 back in. That codec driver I sent you was about
95% complete.
Indeed and to be honest I only gave it a short glance before having to
do about a thousand other things, but I will take another peek some
time this week :)
--
Matt Sealey [off-list ref]
Genesi, Manager, Developer Relations
From: Jon Smirl <hidden> Date: 2008-07-07 14:50:48
On 7/7/08, Matt Sealey [off-list ref] wrote:
Jon Smirl wrote:
quoted
On 7/7/08, Matt Sealey [off-list ref] wrote:
quoted
to do that the samples have to be alternated as they are fed into the
AC97 stream. I think the codec can capture that way too but you
didn't put a transceiver on the S/PDIF line.
OT, but the IDT STAC9766 doesn't support S/PDIF input, only output.
quoted
In my test driver you only get AC97 or S/PDIF. Grant and I are both
working on i2s drivers, when we get those sorted out it shouldn't be
too hard to add ac97 back in. That codec driver I sent you was about
95% complete.
Indeed and to be honest I only gave it a short glance before having to
do about a thousand other things, but I will take another peek some
time this week :)
You should junk my PSC and DMA code and switch to the code Grant
recently posted. Codec should work as is. My PSC code is for a six
month old version of ASOC v2.
Grant's PSC driver needs to be split into a DMA library and i2s
driver. Then make an AC97 version based on his i2s driver that uses
the DMA library. Not hard once he splits the i2s driver.
We are getting closer to having full featured audio drivers for the
mpc5200 platform.
--
Matt Sealey [off-list ref]
Genesi, Manager, Developer Relations