8260 IDMA experience?

8260 IDMA experience?

[Allen Curtis]

Does anyone have experience with IDMA on the 8260? Are there any special
considerations if you wanted to use the memory-to-memory copy function in
place of a memcpy()? (kernel memory only at this time)


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: 8260 IDMA experience?

[Arto Vuori]

Allen Curtis wrote:
>
> Does anyone have experience with IDMA on the 8260? Are there any special
> considerations if you wanted to use the memory-to-memory copy function in
> place of a memcpy()? (kernel memory only at this time)

I have been using IDMA to transfer data between kernel memory and a
FPGA. I also did some testing with memory-to-memory transfers and i
didn't find them to be worth the trouble. Performance ain't that great.
You probably get more speed by copying with your CPU, though you might
avoid some cache trashing.

I did also notice that when i am doing memory-to-peripheral (or
vice-versa) transfers, IDMA corrupts simultaneous Ethernet traffic
unless IDMA is set to low-priority mode. Same might apply for
memory-to-memory transfers.

Arto

--
Arto Vuori
email: avuori@ssh.com
mobile:	+358 40 754 5223

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: 8260 IDMA experience?

[Allen Curtis]

> I have been using IDMA to transfer data between kernel memory and a
> FPGA. I also did some testing with memory-to-memory transfers and i
> didn't find them to be worth the trouble. Performance ain't that great.
> You probably get more speed by copying with your CPU, though you might
> avoid some cache trashing.

So it works but it may not be as fast as a memcpy(). Sounds like more
testing is required to determine the overall benefit of freeing the
processor.

> I did also notice that when i am doing memory-to-peripheral (or
> vice-versa) transfers, IDMA corrupts simultaneous Ethernet traffic
> unless IDMA is set to low-priority mode. Same might apply for
> memory-to-memory transfers.

What 8260 chip revision are you using? I have found lots of problems when
rev A processors but rev C has been much better.


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: 8260 IDMA experience?

[Arto Vuori]

Allen Curtis wrote:
> So it works but it may not be as fast as a memcpy(). Sounds like more
> testing is required to determine the overall benefit of freeing the
> processor.

Right. We are using IDMA only for accessing slow devices, so that CPU
don't need to wait for the data and can do some usefull work
simultaneously.

>
> > I did also notice that when i am doing memory-to-peripheral (or
> > vice-versa) transfers, IDMA corrupts simultaneous Ethernet traffic
> > unless IDMA is set to low-priority mode. Same might apply for
> > memory-to-memory transfers.
>
> What 8260 chip revision are you using? I have found lots of problems when
> rev A processors but rev C has been much better.

I have done most of my testing with revision A chips. I've had many
problems with those chips and it took some time with logic analyzer to
debug those IDMA transfers before i got them working.

Arto

--
Arto Vuori
email: avuori@ssh.com
mobile:	+358 40 754 5223

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/