On 06/07/2020 13:53, Maarten Lankhorst wrote:
Op 03-07-2020 om 15:49 schreef Tvrtko Ursulin:
On 03/07/2020 13:22, Maarten Lankhorst wrote:
We want to introduce backoff logic, but we need to lock the
pool object as well for command parsing. Because of this, we
will need backoff logic for the engine pool obj, move the batch
validation up slightly to eb_lookup_vmas, and the actual command
parsing in a separate function which can get called from execbuf
relocation fast and slowpath.
On this one I also had some feedback in the previous round which you maybe missed.
Regards,
Compile tested this one now, so should work. :)
Cool.
And what about the other comments I had? More splitting of logical steps
(lookup/validate/parse/whatever)? Working on removing the reverts and
making eb_relocate_parse_slow have less than 11 gotos and 4-5 labels (by
the end of the series) is out the question? Removing the "dirty hack" as
well later in the series?
It would be really nice if you could also scan through Chris' idea. I
particularly like how he manages to do it without needing to touch
execlists and context handling (less invasiness should help us). But I
haven't studied his series fully yet, to understand all the details. So
it would be good to join forces there as well.
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
