[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Public WebGL] WebGL performance regression tests!

On Thu, Mar 29, 2012 at 2:54 PM, Kenneth Russell <kbr@google.com> wrote:

Nice work! Performance regressions are difficult to detect, so any
efforts to do so in an automated fashion are great.

Embarrassingly, Chrome Canary, at least, sad-tabs running one of the
tests; I've filed http://crbug.com/120977 to track this.

I hope that someone on the MapsGL team at Google might see this post
and comment. In general the tests you've got look pretty good; they
seem to exercise aspects of the WebGL implementation such as reading
back pixels outside the viewport and texture format conversions. I
think the MapsGL team has more tests in their built-in benchmark which
measured fill rate, certain aspects of shader performance, etc. If we
were trying to write a cross-platform Windows Experience Index
benchmark we'd probably want more of the latter kind of tests.

MapsGL guy here.  This is great that you're putting together this suite, Benoit.
Our benchmark is based mostly on doing lots of texture lookups and dependent texture lookups because that was what we found to be most correlated with MapsGL performance.
We also have a compute-heavy fragment shader that does Mandelbrot calcs.

We also run continuous perf tests of MapsGL itself, and this shows things that may not show up in the synthetic 1-trick benchmarks.   I think following the lead of the other 3D benchmarks and including some packaged up bits of real apps or demos in addition to the synthetic tests would be a good thing for the suite.