Re: [PATCH i-g-t 3/4] igt/gem_exec_schedule: Trim deep runtime

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

 



Quoting Katarzyna Dec (2018-07-24 13:08:25)
> On Mon, Jul 23, 2018 at 09:07:35PM +0100, Chris Wilson wrote:
> > Time the runtime for emitting deep dependency tree, while keeping it
> > full of umpteen thousand requests.
> > 
> > Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> 
> After conversation on IRC with dispelling doubts:
> Reviewed-by: Katarzyna Dec <katarzyna.dec@xxxxxxxxx>

The test itself is just a nonsense stress test, trying to push the
system to the breaking point. Unfortunately the setup also demands that
it complete with 10s due to the use of an external fence, and CI runs
slowly (lockdep + allocation debugging is not fun). There are a myriad
possible dependency webs that are all interesting, so the only thing we
want to avoid are degenerate patterns that are resolved without looking
in the dfs (although analysing such degenerate patterns to make sure we
can take advantage of the degeneracy, because frankly most users fall
into simple patterns is also interesting, but not the purpose of this
investigation). As it is just a smoketest trying to exercise the dfs,
any such pattern is as good as any other. Again, this would be ideal to
script such that we could construct some arbitrary tree and evaluate the
execution order (going much further than the simple depth 1 trees we use
to prove the basics).

Hmm, really should investigate a few graph expression languages to see
if we can truly script this. Still have that 10s to worry about though.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx




[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux