Hi again,
Oops, sorry. I had read : C = alpha * A*B + C, quite oddly !
Another possibility I can see is that the "mixed layout" kernels are pretty
slow (unoptimized memory accesses) and I think involve temporaries. There
could be a problem at this level, like a missing finish(); for example... I
guess.
Good night. I'm off to bed too !
Philippe
2013/8/26 Toby St Clere Smithe <m...@tsmithe.net>
> Hey Philippe,
>
> Philippe Tillet <phil.til...@gmail.com>
> writes:
> > I've double-checked the generator for the test-suite of the generated
> > matrices, and all layout/transpose combination work on my hardware on
> > rectangular matrices. To make sure the generated kernels are correct on
> > your hardware, you can run ./tests/generator_blas3-test-opencl...
> > Since the diagonal elements are correct, a synchronization problem is
> > unlikely, and it seems like the row-majors are mapped to a col-major, or
> > conversely. Does the problem appear when you mix layout with non-prod
> > operations?
>
> Thanks for your help. I'm about to head off to bed, but I thought I
> might as well answer your question: I've tested all combinations of
> layout/tranposition/range/slice for product operations, and I have no
> problems at all; for this reason, I also don't think that the problem
> comes from copying and accessing the matrix buffer incorrectly for the
> column-major layout. So, yeah, the problem only seems to appear when I
> mix layouts in a non-prod operation... Odd.
>
>
> Cheers,
>
> Toby
>
>
>
> ------------------------------------------------------------------------------
> Introducing Performance Central, a new site from SourceForge and
> AppDynamics. Performance Central is your source for news, insights,
> analysis and resources for efficient Application Performance Management.
> Visit us today!
> http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk
> _______________________________________________
> ViennaCL-devel mailing list
> ViennaCL-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/viennacl-devel
>
------------------------------------------------------------------------------
Introducing Performance Central, a new site from SourceForge and
AppDynamics. Performance Central is your source for news, insights,
analysis and resources for efficient Application Performance Management.
Visit us today!
http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk
_______________________________________________
ViennaCL-devel mailing list
ViennaCL-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/viennacl-devel
