[Piglit] [PATCH 00/10] shader_runner support for micro benchmarks

[Paul Berry]

On 16 October 2013 13:33, Jordan Justen <jljusten at gmail.com> wrote:

> On Wed, Oct 16, 2013 at 10:58 AM, Eric Anholt <eric at anholt.net> wrote:
> > Jordan Justen <jordan.l.justen at intel.com> writes:
> >
> >> git://people.freedesktop.org/~jljusten/piglit shader_runner-time-v1
> >>
> >> I think shader_runner could be an easy way to develop
> >> quick micro-benchmarks when working on performance.
> >>
> >> I found shader_runner only required a few tweaks to
> >> be usable for this with depth clears.
> >>
> >> I'm not suggesting (at least in this series), that
> >> we add any micro benchmark scripts to the tree. Rather
> >> I would just like to make it possible to write such
> >> scripts for shader_runner.
> >>
> >> The last patch in this series provides an example
> >> usage, but I don't want that patch to be added to piglit.
> >
> > I don't think we should add this to shader_runner.
>
> So, none of the patches?
>
> For example, are 1 & 2 valuable? My thought is, aren't many/most
> shader_test's indifferent to the window size? So, perhaps we could
> shrink the default size down smaller for Linux runs? (I know windows
> has some lower bound for size.)
>

I think patches 1 and 2 are valuable and should be kept.


>
> > You spent more code
> > putting this in shader_runner than it would have taken to just hack
> > something up standalone,
>
> Possibly. The shader_runner changes aren't that fancy though.
>
> But, I find tweaking and re-running a shader_test is faster/easier.
>
> Regarding the 'time' commands, I thought it might be an convenient way
> to micro benchmark shader code issue, although my series doesn't do
> this. But, if you don't agree that this is valuable, well, then it
> probably isn't.
>
> > and shader_runner is already a frankenstein.
>
> Without a doubt. Have we officially drawn a line that shader_runner is
> too much of a monster, and we should avoid adding new features to it?
>

I don't think we've drawn that line.  Yes, shader_runner is ugly and hacky,
but there's a large class of tests where it's way easier to write
shader_runner tests than to write c tests.  If we can broaden this class by
small, incremental improvements to shader_runner, I'm all for it.

If someone wants to submit some patches that make shader_runner less hacky,
I'm in favor of that too.


>
> > I do most of my throwaway microbenchmarks in the mesa-demos repo.
>
> Would you be willing to consider ways to make this convenient in
> piglit, such as patch 3/piglit_get_microseconds?
>
> -Jordan
> _______________________________________________
> Piglit mailing list
> Piglit at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/piglit
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/piglit/attachments/20131018/4e264c73/attachment-0001.html>