Re: [PATCH 00/10] Venus stateful Codec API

[Alexandre Courbot]

On Fri, Jan 25, 2019 at 7:35 PM Stanimir Varbanov
<stanimir.varbanov@xxxxxxxxxx> wrote:
>
> Hi Alex,
>
> On 1/25/19 7:34 AM, Alexandre Courbot wrote:
> > On Thu, Jan 24, 2019 at 7:13 PM Stanimir Varbanov
> > <stanimir.varbanov@xxxxxxxxxx> wrote:
> >>
> >> Hi Alex,
> >>
> >> Thank you for review and valuable comments!
> >>
> >> On 1/24/19 10:43 AM, Alexandre Courbot wrote:
> >>> Hi Stanimir,
> >>>
> >>> On Fri, Jan 18, 2019 at 1:20 AM Stanimir Varbanov
> >>> <stanimir.varbanov@xxxxxxxxxx> wrote:
> >>>>
> >>>> Hello,
> >>>>
> >>>> This aims to make Venus decoder compliant with stateful Codec API [1].
> >>>> The patches 1-9 are preparation for the cherry on the cake patch 10
> >>>> which implements the decoder state machine similar to the one in the
> >>>> stateful codec API documentation.
> >>>
> >>> Thanks *a lot* for this series! I am still stress-testing it against
> >>> the Chromium decoder tests, but so far it has been rock-solid. I have
> >>> a few inline comments on some patches ; I will also want to review the
> >>> state machine more thoroughly after refreshing my mind on Tomasz doc,
> >>> but this looks pretty promising already.
> >>
> >> I'm expecting problems with ResetAfterFirstConfigInfo. I don't know why
> >> but this test case is very dirty. I'd appreciate any help to decipher
> >> what is the sequence of v4l2 calls made by this unittest case.
> >
> > I did not see any issue with ResetAfterFirstConfigInfo, however
> > ResourceExhaustion seems to hang once in a while. But I could already
> > see this behavior with the older patchset.
>
> Is it hangs badly?

It just stops processing, no crash. I need to investigate more closely.

> >
> > In any case I plan to thoroughly review the state machine. I agree it
> > is a bit complex to grasp.
>
> yes the state machine isn't simple and I blamed Tomasz many times for
> that :)

If I feel inspired I may try to extract the useful part of your patch
into a framework that we can use everywhere. :) We don't want every
driver to reimplement this complex state machine and introduce new
bugs every time.