[Intel-gfx] [PATCH 07/18] tests/gem_ctx_exec: mark lrc lite restore as basic
Daniel Vetter
daniel at ffwll.ch
Fri Aug 14 09:03:00 PDT 2015
On Fri, Aug 14, 2015 at 08:31:01AM -0700, Jesse Barnes wrote:
> On 08/14/2015 05:32 AM, Daniel Vetter wrote:
> > On Thu, Aug 13, 2015 at 01:31:30PM -0700, Jesse Barnes wrote:
> >> Need some LRC tests in the 'basic' subset, and this is a good simple
> >> one.
> >>
> >> Signed-off-by: Jesse Barnes <jbarnes at virtuousgeek.org>
> >
> > This is just a testcase for a very specific lrc corner case. We do already
> > exercise lrc with all the other execbuf testcases. Imo we're covered
> > enough already with what we have in the basic testset - testing for all 3
> > billion cornercases will make it grow out of scope I fear. I'd just drop
> > this one here as not needed for BAT.
> >
> > If you want to extend execbuffer scope a bit then we should add a
> > concurrency test, i.e. one of the gem_concurrent_blt testcases as basic
> > ones. Unfortunately to be able to reliable trigger race conditions those
> > all take a few seconds. But inter-batch sync is a _big_ gap across all
> > archs, and something which is even more tricky with lrc (and scheduler).
> > Imo that would be a lot more useful than this test here.
>
> Yeah that's a good point; I just saw 'lrc' and though "I want that", but
> you're right we should already be covered. Definitely open to adding
> some concurrency stuff (maybe just a few seconds worth) as we get things
> in place.
For concurrency mabye we can do that in a 2nd round - we need to go over
gem_concurrent_* anyway and reshuffle that one into basic, full and
subtests that we're only going to expose through cmdline options for
manual testing. Atm there's way too many of those tests.
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
More information about the Intel-gfx
mailing list