You raise an interesting point about automatically uploading the results.  I noticed the option was checked by default and it didn't bother me in the least.  I definitely want to publish my results but could see how others may not want to. 

We should probably list the website where results could be found right next to the flag and a way to tag the results with a username or nickname and supply the person with a direct URL to their results so they can send to their friends and brag :)


On Fri, Aug 22, 2014 at 7:13 AM, Karl Rupp <rupp@iue.tuwien.ac.at> wrote:
Hi Namik,

 > I guess I forgot to work on Qt4 compatibility. It's fixed now:
> https://github.com/viennacl/viennacl-benchmark-gui/commit/c213ef1058cd84841c0ca4c45de8c79b7be833bf
> It seems the MOC does not support declaring signal/slot functions within
> preprocessor if statements... That's why
> #if (QT_VERSION > QT_VERSION_CHECK(5, 0, 0))
> someSlotFunction();
> #endif
> didn't work.

Thanks, it works like a charm again :-)

>     As for the sparse matrix reading: The code looks godd, just the way I'd
>     expect it to work. Maybe the file format isn't recognized properly? Can
>     you run in a debugger to see where the error is caused?
> I assumed my code wasn't good, so I didn't do much testing. I'll take a
> better look at it now.


Now as I could run the GUI again, I must say that it is getting into
better and better shape, so we should be good with releasing it together
with ViennaCL 1.6.0 and get a bunch of attention :-) Some more minor
suggestions to make the GUI even better:

* I notice that the splash screen has some compression artifacts. Is it
enough to just replace the respective graphics file with a higher
resolution image?

* The benchmark-logo ViennaCL_Benchmark_experimentLogo.png is neat. If
the 'V' gets tweaked a little in style so that it matches the font, it
would awesome! Also, the red in 'CL' is on the other end of the color
spectrum as blue, so it's worth to try colors 'closer' to blue, e.g.
orange. Maybe just use black for 'CL', like in 'Benchmark'?

* The automatic benchmark result upload functionality is perhaps a bit
rude, particularly now as all the NSA stuff has happened. What about
asking the user at the end of the benchmark to upload (yes/no)? Also, as
there is no autotuning available right now, we could just put the upload
option on standby and wait for the autotuning to be added.

* The values for "Vector size" take quite a lot of digits. What if you
plot "Vector size (in thousands)" instead to crop the unnecessary zeros?

I'll go over the routines for "Copy" and "Vector" to really benchmark
device-specifics rather than ViennaCL-specifics. This should be
straightforward - I'll ping you if I encounter any unexpected problems.
Maybe I find some time tonight, otherwise on Sunday.

Best regards,

Slashdot TV.
Video for Nerds.  Stuff that matters.
ViennaCL-devel mailing list

Matthew Musto