Re: [EXT] Re: [PATCH 0/5] Add V4L2 driver for i.MX8 JPEG Encoder/Decoder

[Nicolas Dufresne]

Le vendredi 08 novembre 2019 Ã 15:06 +0100, Hans Verkuil a Ãcrit :
> On 11/8/19 2:36 PM, Mirela Rabulea wrote:
> > Hi Hans,
> > 
> > On Mi, 2019-11-06 at 17:52 +0100, Hans Verkuil wrote:
> > >       test Scaling: OK
> > > The presence of a scaler is suspicious: is the encoder indeed
> > > capable of scaling? I suspect this is a bug.
> > 
> > No, it's not capable of scaling. You suspect a bug in the driver or the
> > tests?
> 
> Actually, I think that's an outstanding bug in v4l-utils. It doesn't
> correctly handle the m2m case with respect to scaling. I think. I'll
> look into this a bit more.
> 
> > > Codec ioctls:
> > > >       test VIDIOC_(TRY_)ENCODER_CMD: OK
> > > The presence of this...
> > > 
> > > >       test VIDIOC_G_ENC_INDEX: OK (Not Supported)
> > > >       test VIDIOC_(TRY_)DECODER_CMD: OK
> > > ...and this is also strange for a JPEG codec. These ioctls are
> > > typically only
> > > needed for MPEG/H264/etc. codecs, and not for a simple JPEG codec.
> > > 
> > > The same issues are found for the JPEG decoder.
> > 
> > I implemented the CMD_STOP for both encoder & decoder, because it was
> > requested by our developer for gstreamer plugin for this codec.
> > The context in which this was requested was for playing MJPEG videos (a
> > concatenation of JPEG frames).
> 
> This ioctl makes no sense for JPEG codecs, and in fact jpeg drivers like
> s5p-jpeg or mtk-jpeg do not implement this. This sounds like a gstreamer bug.
> Nicolas, do you know anything about this?

This code was added in generic bits of GStreamer. So we just always do
CMD_STOP to drain the remaining buffers and expect EPIPE later on when
done. We can special case this, but at the same time why ? This could
be buried into the m2m helpers with a default implementation.

If you feel like this is completely wrong and userspace should do more
case-by-case code path for different CODEC handling, then someone will
have to update GStreamer accordingly. This is all moving target at the
spec just got merged.

> 
> > > Streaming ioctls:
> > > >       test read/write: OK (Not Supported)
> > > >       test blocking wait: OK
> > > >               fail: v4l2-test-buffers.cpp(254): g_field() ==
> > > > V4L2_FIELD_ANY
> > > The driver shall never return FIELD_ANY. This needs to be FIELD_NONE.
> > 
> > Is there a "good example" of a v4l m2m driver that passes these
> 
> vim2m. Also drivers/media/platform/mtk-jpeg/ (although I'm not sure when
> it was last tested with v4l2-compliance, so it might be a bit out of date).
> 
> > streaming tests? That would save some time on my side.
> > For the FIELD_ANY issue, I got inspired from your commit:
> > ab7afaf3 media: vim2m: add buf_out_validate callback
> > But there's a lot more to go...
> > 
> > Thanks,
> > Mirela
> > 
> 
> Regards,
> 
> 	Hans

Attachment: signature.asc
Description: This is a digitally signed message part