Hi,
2013/8/12 Karl Rupp <r...@iue.tuwien.ac.at>
> Hi,
>
> > Good news : the GEMMs calls for OpenCL on dense non-proxy matrix now
> > call the generator ! It's a good step towards performance portability.
>
> Hurray, indeed it is! Well done! :-)
> Now as you fixed some things in the autotuner, I could also give it
> another shot on the MIC. Does the autotuner print all the interesting
> device information already?
>
> Oh, and I could also go for a K20X. Ready? ;-)
>
>
I still have some polishing to do for the autotuner, so that it indeed
print the same thing as viennacl-info. The problem of the MIC is that I
have aaaaabsolutely ZERO idea of how to prune profiles. We can try, with
the GPU configuration so that it remains tractable, but I don't guarantee
any good result :P
Nice for the K20X :) I still have to work out things though, to make sure
#pragma unroll is always called on NVidia Hardware, and never called on AMD
hardware... Performing an extensive search just creates so many constraints.
I'll send an update when I consider the autotuner is ready for the K20X and
the MIC...
>
> > For now, single precision : (...)
> >
> > All row-major.
>
> This does not yet include the 'trick' with using the transposed case for
> column-major, does it?
>
>
> > Bad news : peaky performance :
> > There's no missing digit for the "2048" case. Dealing with it is fairly
> > complicated, since it involves having different profiles for different
> > sizes. Since it seems to only affect AMD Hardware, I think we just warn
> > about the issue...
>
> We should document this, yes. However, I don't know whether we can
> reliably warn about this. It may even be driver or OS dependent.
>
Hmm, true that I should pass this specific benchmark to codeXL, to know
where the bad performance comes from. My guess for now is that it comes
from channel conflict...
>
> > However, this is the size used by default in the blas3_bench, which made
> > me freak out.
>
> Ouch... :-/
>
>
> > What do you guys would think about a more "graphical" (ie either plot or
> > a list, like above) benchmark, so that people who really care a lot
> > about performance can have an idea of when to use what.
>
> Printing a list with performances for different sizes would be nice,
> yes. Feel free to just print the output above, but better truncate near
> 1500 so that it does not crash (or take too long) on weaker hardware.
>
>
> > *Examples:*
> > -> I'm having a matrix of internal size 384*384 => In C=A*B, I'd prefer
> > to either order A as column-major (or to use C=trans(A)*B where A is
> > row-major instead, on my specific hardware)
> > -> If I want some performance increase for big problems, I'd rather go
> > for C=A*B in all row-major.
> >
> > What do you think about that?
>
> We should also have a section in the manual about this including some of
> our (i.e. particularly your) experiences.
>
Alright :)
Do you still plan to make a GUI for the Autotuner. I'm realizing that this
could be a really nice feature, since it allows one to freely choose the
parameter space ...
Best regards,
Philippe
> Best regards,
> Karli
>
>
>
> ------------------------------------------------------------------------------
> Get 100% visibility into Java/.NET code with AppDynamics Lite!
> It's a free troubleshooting tool designed for production.
> Get down to code-level detail for bottlenecks, with <2% overhead.
> Download for free and get started troubleshooting in minutes.
> http://pubads.g.doubleclick.net/gampad/clk?id=48897031&iu=/4140/ostg.clktrk
> _______________________________________________
> ViennaCL-devel mailing list
> ViennaCL-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/viennacl-devel
>
------------------------------------------------------------------------------
Get 100% visibility into Java/.NET code with AppDynamics Lite!
It's a free troubleshooting tool designed for production.
Get down to code-level detail for bottlenecks, with <2% overhead.
Download for free and get started troubleshooting in minutes.
http://pubads.g.doubleclick.net/gampad/clk?id=48897031&iu=/4140/ostg.clktrk
_______________________________________________
ViennaCL-devel mailing list
ViennaCL-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/viennacl-devel
