On Tue, 14 Oct 2014 11:49:33 +0800
"Zhigang Gong" <[email protected]> wrote:

> Jerome,
> 
> I'm interested in which benchmarks you were using, and the details of the
> comparison of beignet and macos's.

Dear Zhigang,

I didn't use "specific" benchmarks but rather used the application I am 
developping as benchmark.
https://github.com/kif/pyFAI

The benchmark measures the speed of images of various size, some
implementation are full OpenCL (i.e. no processing done outside OpenCL)
The methodology of the benchmark is explained in the EuroScipy proceedings:
http://pdebuyl.be/tmp/esp2014_draft.pdf

> If you could share it directly on the mail list, I will be better. Thanks!

If you want to reproduce the benchmark:

* clone the project from GitHub
* the dependencies list is rather long: numpy, scipy, matplotlib, pyopencl and 
fabio.
* in the directory, run "python setup.py build test"
The tests are making a lot of warning. Be sure to be connected on internet as 
it downloads test images (~100MB).

Then in the benchmark directory run "python benchmark.py -c -g -r 3 -n 10"
-r3: take the best of 3 runs
-n10: average over 10 measurements
-c and -g to test OpenCL both on CPU & GPU

Cheers,

-- 
Jérôme Kieffer <[email protected]>
_______________________________________________
Beignet mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/beignet

Reply via email to