Re: [PATCH] V4L: add Y12I, Y8I and Z16 pixel format documentation

[Sakari Ailus <sakari.ailus@iki.fi>]

Hi Guennadi,

On Thu, Jan 14, 2016 at 12:12:08PM +0100, Guennadi Liakhovetski wrote:
> Hi Sakari,
> 
> Thanks for a review! I'll fix all the cosmetic issues, thanks. As for 
> other comments:
> 
> On Wed, 13 Jan 2016, Sakari Ailus wrote:
> 
> [snip]
> 
> > > --- /dev/null
> > > +++ b/Documentation/DocBook/media/v4l/pixfmt-y12i.xml
> > > @@ -0,0 +1,49 @@
> > > +<refentry id="V4L2-PIX-FMT-Y12I">
> > > +  <refmeta>
> > > +    <refentrytitle>V4L2_PIX_FMT_Y12I ('Y12I ')</refentrytitle>
> > 
> > Extra space after 4cc.                        ^
> > 
> > > +    &manvol;
> > > +  </refmeta>
> > > +  <refnamediv>
> > > +    <refname><constant>V4L2_PIX_FMT_Y12I</constant></refname>
> > > +    <refpurpose>Interleaved grey-scale image, e.g. from a stereo-pair</refpurpose>
> > > +  </refnamediv>
> > > +  <refsect1>
> > > +    <title>Description</title>
> > > +
> > > +    <para>This is a grey-scale image with a depth of 12 bits per pixel, but with
> > > +pixels from 2 sources interleaved and bit-packed. Each pixel is stored in a
> > > +24-bit word. E.g. data, stored by a R200 RealSense camera on a little-endian
> > > +machine can be deinterlaced using</para>
> > 
> > I think we should precisely define the format, either big or little. Is the
> > endianness of the format affected by the machine endianness? (I'd guess no,
> > but that's just a guess.)
> 
> Ok, since this works on a LE machine:
> 
> left0 = 0xfff & *(__u16 *)buf;
> 
> I think we can call data LE in the buffer. But specifying left-right order 
> cannot be done in terms of endianness, so, I provided that code snippet.

I meant that the the format definition should clearly say which one is the
order.

> 
> > I wonder if the format should convey the information which one is right and
> > which one is left, e.g. by adding "LR" to the name.
> 
> You mean to distinguish between LR and RL? Can do in principle, yes.

If we want the format to have an exact definition, we should have this as
well.

I think the formats increasingly have little details such as this one which
require adding many format variants but I'm not sure if it's even a problem.

I'd postfix the name with "LR" or at least document that this is the pixel
order.

>
> > No need to mention RealSense specifically IMO.
> 
> Ok.
> 
> > > +
> > > +<para>
> > > +<programlisting>
> > > +__u8 *buf;
> > > +left0 = 0xfff &amp; *(__u16 *)buf;
> > > +rirhgt0 = *(__u16 *)(buf + 1) >> 4;
> > 
> > "right"
> 
> [snip]
> 
> > > diff --git a/Documentation/DocBook/media/v4l/pixfmt-z16.xml b/Documentation/DocBook/media/v4l/pixfmt-z16.xml
> > > new file mode 100644
> > > index 0000000..fac3c68
> > > --- /dev/null
> > > +++ b/Documentation/DocBook/media/v4l/pixfmt-z16.xml
> > > @@ -0,0 +1,79 @@
> > > +<refentry id="V4L2-PIX-FMT-Z16">
> > > +  <refmeta>
> > > +    <refentrytitle>V4L2_PIX_FMT_Z16 ('Z16 ')</refentrytitle>
> > > +    &manvol;
> > > +  </refmeta>
> > > +  <refnamediv>
> > > +    <refname><constant>V4L2_PIX_FMT_Z16</constant></refname>
> > > +    <refpurpose>Interleaved grey-scale image, e.g. from a stereo-pair</refpurpose>
> > > +  </refnamediv>
> > > +  <refsect1>
> > > +    <title>Description</title>
> > > +
> > > +    <para>This is a 16-bit format, representing depth data. Each pixel is a
> > > +distance in mm to the respective point in the image coordinates. Each pixel is
> > > +stored in a 16-bit word in the little endian byte order.</para>
> > 
> > The format itself looks quite generic but the unit might be specific to the
> > device. It'd sound silly to add a new format if just the unit is different.
> 
> My understanding is, that each format must have a fixed meaning, i.e. a 
> fixed depth unit too, although it would definitely help to be able to 
> relax that requirement in this case.

Agreed.

> > How about re-purpose the colourspace field for depth formats and
> > add a flag telling the colour space field contains the unit and the unit
> > prefix.
> 
> Hmmm... Not sure I find this a proper use of the .colorspace field...

I think colour space doesn't make much sense in context of depth.

> 
> > Not something to have in this patch nor patchset though: controls
> > should gain that as well.
> 
> Sorry, didn't get this - how can a control tell you what units a specific 
> format uses? What if your camera can output depth in multiple units?

Controls do not have units at the moment. The specification often suggests a
unit but using the unit suggested by the spec isn't always possible, thus
it'd be useful to have this available through VIDIOC_QUERYCTRL.

-- 
Kind regards,

Sakari Ailus
e-mail: sakari.ailus@iki.fi	XMPP: sailus@retiisi.org.uk