[libcamera-devel] [PATCH v4 5/7] libcamera: v4l2_subdevice: Add format information

Tue Apr 28 09:18:48 CEST 2020

Hi Laurent, Niklas,

On Tue, Apr 28, 2020 at 05:01:23AM +0300, Laurent Pinchart wrote:
> Hi Jacopo and Niklas,
>
> On Tue, Apr 28, 2020 at 02:06:34AM +0200, Niklas Söderlund wrote:
> > On 2020-04-27 23:32:34 +0200, Jacopo Mondi wrote:
> > > Define a map of static information associated with a media bus code.
> > > Start by reporting the bits-per-pixel for each media bus code defined by
> > > the V4L2 kernel API, to later expand it when necessary.
> > >
> > > Add to the V4L2SubdeviceFormat class a method to return the bits per
> > > pixel, retrieved by inspecting the static map of format information.
> > >
> > > While at there, remove a rouge map inclusion from header file.
>
> s/there/it/ (or "while there")
> s/rouge/rogue/ (unless you really mean red :-))
>
> > >
> > > Signed-off-by: Jacopo Mondi <jacopo at jmondi.org>
> > >
> > > ---
> > >  src/libcamera/include/v4l2_subdevice.h |   2 +-
> > >  src/libcamera/v4l2_subdevice.cpp       | 104 +++++++++++++++++++++++++
> > >  2 files changed, 105 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/src/libcamera/include/v4l2_subdevice.h b/src/libcamera/include/v4l2_subdevice.h
> > > index 27ba5b17f61e..d0e565dbdaab 100644
> > > --- a/src/libcamera/include/v4l2_subdevice.h
> > > +++ b/src/libcamera/include/v4l2_subdevice.h
> > > @@ -7,7 +7,6 @@
> > >  #ifndef __LIBCAMERA_V4L2_SUBDEVICE_H__
> > >  #define __LIBCAMERA_V4L2_SUBDEVICE_H__
> > >
> > > -#include <map>
> > >  #include <string>
> > >  #include <vector>
> > >
> > > @@ -27,6 +26,7 @@ struct V4L2SubdeviceFormat {
> > >  	Size size;
> > >
> > >  	const std::string toString() const;
> > > +	uint8_t bitsPerPixel() const;
>
> I think unsigned int would be more efficient (or at least not less). It
> would also avoid interesting issues when writing
>
> 	LOG(Info) << "bits per pixel: " << format.bitsPerPixel();
>
> as with uint8_t it will be considered as a char and not a number (I
> know...).

I stumbled into this a few days ago, yes, I know :(

I can change this to unsigned int, sure

>
> > >  };
> > >
> > >  class V4L2Subdevice : public V4L2Device
> > > diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
> > > index 74788ce7cf4f..93fe4b8c3d32 100644
> > > --- a/src/libcamera/v4l2_subdevice.cpp
> > > +++ b/src/libcamera/v4l2_subdevice.cpp
> > > @@ -14,6 +14,7 @@
> > >  #include <sys/ioctl.h>
> > >  #include <unistd.h>
> > >
> > > +#include <linux/media-bus-format.h>
> > >  #include <linux/v4l2-subdev.h>
> > >
> > >  #include <libcamera/geometry.h>
> > > @@ -32,6 +33,92 @@ namespace libcamera {
> > >
> > >  LOG_DECLARE_CATEGORY(V4L2)
> > >
> > > +namespace {
> > > +
> > > +/*
> > > + * \var formatInfoMap
> > > + * \brief A map that associates bits per pixel to V4L2 media bus codes
> > > + */
> > > +const std::map<uint32_t, uint8_t> formatInfoMap = {
> > > +	{ V4L2_MBUS_FMT_RGB444_2X8_PADHI_BE, 16 },
> > > +	{ V4L2_MBUS_FMT_RGB444_2X8_PADHI_LE, 16 },
> > > +	{ V4L2_MBUS_FMT_RGB555_2X8_PADHI_BE, 16 },
> > > +	{ V4L2_MBUS_FMT_RGB555_2X8_PADHI_LE, 16 },
> > > +	{ V4L2_MBUS_FMT_BGR565_2X8_BE, 16 },
> > > +	{ V4L2_MBUS_FMT_BGR565_2X8_LE, 16 },
> > > +	{ V4L2_MBUS_FMT_RGB565_2X8_BE, 16 },
> > > +	{ V4L2_MBUS_FMT_RGB565_2X8_LE, 16 },
> > > +	{ V4L2_MBUS_FMT_RGB666_1X18, 18 },
> > > +	{ V4L2_MBUS_FMT_RGB888_1X24, 24 },
> > > +	{ V4L2_MBUS_FMT_RGB888_2X12_BE, 24 },
> > > +	{ V4L2_MBUS_FMT_RGB888_2X12_LE, 24 },
> > > +	{ V4L2_MBUS_FMT_ARGB8888_1X32, 32 },
> > > +	{ V4L2_MBUS_FMT_Y8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_UV8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_UYVY8_1_5X8, 40 },
>
> 1_5X8 means 1.5x8 = 12 :-)
>

Ah!

> > > +	{ V4L2_MBUS_FMT_VYUY8_1_5X8, 40 },
> > > +	{ V4L2_MBUS_FMT_YUYV8_1_5X8, 40 },
> > > +	{ V4L2_MBUS_FMT_YVYU8_1_5X8, 40 },
> > > +	{ V4L2_MBUS_FMT_UYVY8_2X8, 16 },
> > > +	{ V4L2_MBUS_FMT_VYUY8_2X8, 16 },
> > > +	{ V4L2_MBUS_FMT_YUYV8_2X8, 16 },
> > > +	{ V4L2_MBUS_FMT_YVYU8_2X8, 16 },
> > > +	{ V4L2_MBUS_FMT_Y10_1X10, 10 },
> > > +	{ V4L2_MBUS_FMT_UYVY10_2X10, 20 },
> > > +	{ V4L2_MBUS_FMT_VYUY10_2X10, 20 },
> > > +	{ V4L2_MBUS_FMT_YUYV10_2X10, 20 },
> > > +	{ V4L2_MBUS_FMT_YVYU10_2X10, 20 },
> > > +	{ V4L2_MBUS_FMT_Y12_1X12, 12 },
> > > +	{ V4L2_MBUS_FMT_UYVY8_1X16, 16 },
> > > +	{ V4L2_MBUS_FMT_VYUY8_1X16, 16 },
> > > +	{ V4L2_MBUS_FMT_YUYV8_1X16, 16 },
> > > +	{ V4L2_MBUS_FMT_YVYU8_1X16, 16 },
> > > +	{ V4L2_MBUS_FMT_YDYUYDYV8_1X16, 16 },
> > > +	{ V4L2_MBUS_FMT_UYVY10_1X20, 20 },
> > > +	{ V4L2_MBUS_FMT_VYUY10_1X20, 20 },
> > > +	{ V4L2_MBUS_FMT_YUYV10_1X20, 20 },
> > > +	{ V4L2_MBUS_FMT_YVYU10_1X20, 20 },
> > > +	{ V4L2_MBUS_FMT_YUV10_1X30, 30 },
> > > +	{ V4L2_MBUS_FMT_AYUV8_1X32, 32 },
> > > +	{ V4L2_MBUS_FMT_UYVY12_2X12, 24 },
> > > +	{ V4L2_MBUS_FMT_VYUY12_2X12, 24 },
> > > +	{ V4L2_MBUS_FMT_YUYV12_2X12, 24 },
> > > +	{ V4L2_MBUS_FMT_YVYU12_2X12, 24 },
> > > +	{ V4L2_MBUS_FMT_UYVY12_1X24, 24 },
> > > +	{ V4L2_MBUS_FMT_VYUY12_1X24, 24 },
> > > +	{ V4L2_MBUS_FMT_YUYV12_1X24, 24 },
> > > +	{ V4L2_MBUS_FMT_YVYU12_1X24, 24 },
> > > +	{ V4L2_MBUS_FMT_SBGGR8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SGBRG8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SGRBG8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SRGGB8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SBGGR10_ALAW8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SGBRG10_ALAW8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SGRBG10_ALAW8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SRGGB10_ALAW8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SBGGR10_DPCM8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SGBRG10_DPCM8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SGRBG10_DPCM8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SRGGB10_DPCM8_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_SBGGR10_2X8_PADHI_BE, 16 },
> > > +	{ V4L2_MBUS_FMT_SBGGR10_2X8_PADHI_LE, 16 },
> > > +	{ V4L2_MBUS_FMT_SBGGR10_2X8_PADLO_BE, 16 },
> > > +	{ V4L2_MBUS_FMT_SBGGR10_2X8_PADLO_LE, 16 },
> > > +	{ V4L2_MBUS_FMT_SBGGR10_1X10, 10 },
> > > +	{ V4L2_MBUS_FMT_SGBRG10_1X10, 10 },
> > > +	{ V4L2_MBUS_FMT_SGRBG10_1X10, 10 },
> > > +	{ V4L2_MBUS_FMT_SRGGB10_1X10, 10 },
> > > +	{ V4L2_MBUS_FMT_SBGGR12_1X12, 24 },
> > > +	{ V4L2_MBUS_FMT_SGBRG12_1X12, 24 },
> > > +	{ V4L2_MBUS_FMT_SGRBG12_1X12, 24 },
> > > +	{ V4L2_MBUS_FMT_SRGGB12_1X12, 24 },
>
> These last four tshould be 12 bits per pixel.
>

Ouch, sorry, I re-checked this twice but I've missed this it seems

> > > +	{ V4L2_MBUS_FMT_JPEG_1X8, 8 },
> > > +	{ V4L2_MBUS_FMT_S5C_UYVY_JPEG_1X8, 8 },
>
> I'd drop these two as bits per pixel is really ill-defined for JPEG.
>

Ack

> I think the bits per pixel concept is ill-defined in general. Our
> current use case for this is limited to Bayer formats, to know the
> dynamic range of the data. When transmitting, let's say,
> V4L2_MBUS_FMT_SBGGR10_2X8_PADHI_BE, I can imagine the pipeline handler
> possibly being interested in knowing that the bus width is 16 bits in
> order to configure the pipeline, and the IPA being interested in knowing
> the dynamic range is 10 bits. We'll have to revisiti all this later,
> there's no reason to block this patch for this, but please remember this
> issue in the future as I'll want to refactor this before getting more
> users of this API.

I agree... should we start thinking of a 'sample width' vs 'bus width'
?

>
> > > +	{ V4L2_MBUS_FMT_AHSV8888_1X32, 32 },
> > > +};
> > > +
> > > +}; /* namespace */
>
> s/};/}/
>
> > > +
> > >  /**
> > >   * \struct V4L2SubdeviceFormat
> > >   * \brief The V4L2 sub-device image format and sizes
> > > @@ -81,6 +168,23 @@ const std::string V4L2SubdeviceFormat::toString() const
> > >  	return ss.str();
> > >  }
> > >
> > > +/**
> > > + * \brief Retrieve the number of bits per pixel for the V4L2 subdevice format
> > > + * \return The number of bits per pixel for the format, or default it to 8 if
> > > + * the format's media bus code is not supported
> > > + */
> > > +uint8_t V4L2SubdeviceFormat::bitsPerPixel() const
> > > +{
> > > +	const auto it = formatInfoMap.find(mbus_code);
> > > +	if (it == formatInfoMap.end()) {
> > > +		LOG(V4L2, Error) << "No information available for format '"
> > > +				 << toString() << "'";
> > > +		return 0;
> >
> > This does not match the documentation of defaulting to 8 if the format
> > is not found. I'm wondering however if we shall not make the error fatal
> > as a default value that is wrong could lead to really odd behaviors.
>
> The code was updated but the documentation stayed as-is. s/or default it
> to 8/or 0/.

Yeah, leftover

>
> Reviewed-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
>

Thanks
  j

> > > +	}
> > > +
> > > +	return it->second;
> > > +}
> > > +
> > >  /**
> > >   * \class V4L2Subdevice
> > >   * \brief A V4L2 subdevice as exposed by the Linux kernel
>
> --
> Regards,
>
> Laurent Pinchart