[libcamera-devel] [PATCH 11/17] libcamera: stream: Add StreamFormats

[Kieran Bingham]

Hi Niklas,

On 27/05/2019 01:15, Niklas Söderlund wrote:
> Add a StreamFormats which describes all the formats a stream can
> support. The object does not collect any formation itself but can
> simplify users interaction with formats as it's able to translate a
> stream format range into discrete list and a discrete list to a range.
> 
> Signed-off-by: Niklas Söderlund <niklas.soderlund at ragnatech.se>
> ---
>  include/libcamera/stream.h |  16 +++
>  src/libcamera/stream.cpp   | 216 +++++++++++++++++++++++++++++++++++++
>  2 files changed, 232 insertions(+)
> 
> diff --git a/include/libcamera/stream.h b/include/libcamera/stream.h
> index e38c0e7e827d5888..48daf5ac23f55d85 100644
> --- a/include/libcamera/stream.h
> +++ b/include/libcamera/stream.h
> @@ -7,6 +7,7 @@
>  #ifndef __LIBCAMERA_STREAM_H__
>  #define __LIBCAMERA_STREAM_H__
>  
> +#include <map>
>  #include <string>
>  #include <vector>
>  
> @@ -18,6 +19,21 @@ namespace libcamera {
>  class Camera;
>  class Stream;
>  
> +class StreamFormats
> +{
> +public:
> +	StreamFormats();
> +	StreamFormats(const std::map<unsigned int, std::vector<SizeRange>> &formats);
> +
> +	std::vector<unsigned int> pixelformats() const;
> +	std::vector<Size> sizes(unsigned int pixelformat) const;
> +
> +	SizeRange range(unsigned int pixelformat) const;
> +
> +private:
> +	std::map<unsigned int, std::vector<SizeRange>> formats_;
> +};
> +
>  struct StreamConfiguration {
>  	StreamConfiguration()
>  		: stream_(nullptr)
> diff --git a/src/libcamera/stream.cpp b/src/libcamera/stream.cpp
> index eecd37160150d55c..a2931902fda2baa5 100644
> --- a/src/libcamera/stream.cpp
> +++ b/src/libcamera/stream.cpp
> @@ -7,6 +7,7 @@
>  
>  #include <libcamera/stream.h>
>  
> +#include <algorithm>
>  #include <iomanip>
>  #include <sstream>
>  
> @@ -36,6 +37,221 @@ namespace libcamera {
>  
>  LOG_DEFINE_CATEGORY(Stream)
>  
> +/**
> + * \class StreamFormats
> + * \brief Hold information about supported stream formats
> + *
> + * The StreamFormats class holds information about pixel formats and frame

/about pixel formats/about the pixel formats/

> + * sizes a stream supports. The class groups size information by the pixel
> + * format which can produce it. There are two ways to examine the size
> + * information, as a range or as a list of discrete sizes.
> + *
> + * When sizes are viewed as a range it describes the minimum and maximum width
> + * and height values. There is a possibility to supplement the range description
> + * with a horizontal och vertical stepping size. The stepping size describes the

/och/and/ ?


> + * step size in pixel from the minimum with/height.

/in pixel/in pixels/

> + *
> + * When sizes is viewed as a list of discrete sizes it describes exact dimensions

When sizes are viewed...  ... they describe exact ...

> + * which can be selected and used.
> + *
> + * Pipeline handlers can create StreamFormats describing each pixel format using
> + * either a range or a list of  discrete sizes. The StreamFormats class attempts

/of  discrete/of discrete/ <remove extra space>

> + * to translates between the two different ways to view them. The translations

/translates/translate/

> + * are performed as:
> + *
> + *  - If a pixel format is described as a list of discrete sizes a range is

... of discrete sizes then a range ....

> + *    created by taking the minim and maximum width/height in the list.

/minim/minimum/

> + *    The stepping information is not recreated and set to 1.

... and is set to 1.


> + *
> + *  - If a pixel format is described as a range a list of discrete sizes which

... as a range then a list of ...

> + *    fits inside that range are selected from a list of common sizes. The
> + *    stepping information is taken into consideration when generating the
> + *    sizes.
> + *
> + * Applications examining sizes as a range with stepping values of 1 should be
> + * aware that the range could be generated form a list of discrete sizes and

/form/from/


> + * there could be big gaps in the range to what the stream can support.

... there could be a large quantity of possible Size combinations which
may not be supported by the Stream.


Could we set the stepping size to '-1' or '0' or something here to
identify that the range is really only specifying the maximum, and
minimum width/height ?

> + *
> + * All sizes retrieved from StreamFormats should be treated as advisory and no
> + * size should be considered to be supported until its been verified using
> + * CameraConfiguration::validate().
> + */
> +
> +/**
> + * \brief Construct a empty StreamFormats object
> + */
> +StreamFormats::StreamFormats()
> +{
> +}
> +
> +/**
> + * \brief Construct a StreamFormats object
> + * \param[in] formats A map of pixel formats to a sizes description
> + */
> +StreamFormats::StreamFormats(const std::map<unsigned int, std::vector<SizeRange>> &formats)
> +	: formats_(formats)
> +{
> +}
> +
> +/**
> + * \brief Retrieve the list of pixel formats supported
> + * \returns List of pixel formats
> + */
> +std::vector<unsigned int> StreamFormats::pixelformats() const
> +{
> +	std::vector<unsigned int> formats;
> +
> +	for (auto const &it : formats_)
> +		formats.push_back(it.first);
> +
> +	return formats;
> +}
> +
> +/**
> + * \brief Retrieve the list of frame sizes
> + * \param[in] pixelformat Pixel format to retrieve sizes for
> + * \returns List description of frame sizes
> + */
> +std::vector<Size> StreamFormats::sizes(unsigned int pixelformat) const
> +{
> +	/*
> +	 * Sizes to try and extract from ranges.
> +	 * \todo Verify list of resolutions are good, current list compiled
> +	 * from v4l2 documentation and source code as well as lists of
> +	 * common frame sizes.
> +	 */
> +	static const std::vector<Size> rangeDiscreteSizes = {
> +		Size(160, 120),
> +		Size(240, 160),
> +		Size(320, 240),
> +		Size(400, 240),
> +		Size(480, 320),
> +		Size(640, 360),
> +		Size(640, 480),
> +		Size(720, 480),
> +		Size(720, 576),
> +		Size(768, 480),
> +		Size(800, 600),
> +		Size(854, 480),
> +		Size(960, 540),
> +		Size(960, 640),
> +		Size(1024, 576),
> +		Size(1024, 600),
> +		Size(1024, 768),
> +		Size(1152, 864),
> +		Size(1280, 1024),
> +		Size(1280, 1080),
> +		Size(1280, 720),
> +		Size(1280, 800),
> +		Size(1360, 768),
> +		Size(1366, 768),
> +		Size(1400, 1050),
> +		Size(1440, 900),
> +		Size(1536, 864),
> +		Size(1600, 1200),
> +		Size(1600, 900),
> +		Size(1680, 1050),
> +		Size(1920, 1080),
> +		Size(1920, 1200),
> +		Size(2048, 1080),
> +		Size(2048, 1152),
> +		Size(2048, 1536),
> +		Size(2160, 1080),
> +		Size(2560, 1080),
> +		Size(2560, 1440),
> +		Size(2560, 1600),
> +		Size(2560, 2048),
> +		Size(2960, 1440),
> +		Size(3200, 1800),
> +		Size(3200, 2048),
> +		Size(3200, 2400),
> +		Size(3440, 1440),
> +		Size(3840, 1080),
> +		Size(3840, 1600),
> +		Size(3840, 2160),
> +		Size(3840, 2400),
> +		Size(4096, 2160),
> +		Size(5120, 2160),
> +		Size(5120, 2880),
> +		Size(7680, 4320),

I still wonder if we could generate this data table in a better form.

For a start, perhaps it might be worth grouping common Sizes to a single
line to form a table based on their ratios? That would show what ratios
are included maybe?

(perhaps we could generate each ratio from a single width or height
instead?)

/* 4:3                     16:9				16:10 */

Size(1600, 1200),	Size(1600, 900),	Size(1680, 1050),
			Size(1920, 1080),	Size(1920, 1200),

/* 2:1			 2:1 ish again?			4:3 */
Size(2048, 1080),	Size(2048, 1152),	Size(2048, 1536),


If you did look at doing this, I'd move the table outside of the
function so that it's indent is reduced.

> +	};
> +	std::vector<Size> sizes;
> +
> +	/* Make sure pixel format exists. */
> +	auto const &it = formats_.find(pixelformat);
> +	if (it == formats_.end())
> +		return {};
> +
> +	/* Try create list of discrete sizes. */
> +	const std::vector<SizeRange> &ranges = it->second;
> +	bool discrete = true;
> +	for (const SizeRange &range : ranges) {
> +		if (range.min != range.max) {
> +			discrete = false;
> +			break;
> +		}
> +		sizes.emplace_back(range.min.width, range.min.height);
> +	}
> +
> +	/* If discrete not possible generate from range. */
> +	if (!discrete) {
> +		if (ranges.size() != 1) {
> +			LOG(Stream, Error) << "Range format is ambiguous";
> +			return {};
> +		}
> +
> +		const SizeRange &limit = ranges.front();
> +		sizes.clear();
> +
> +		for (const Size &size : rangeDiscreteSizes) {
> +			if (size.width < limit.min.width ||
> +			    size.width > limit.max.width ||
> +			    size.height < limit.min.height ||
> +			    size.height > limit.max.height ||
> +			    (size.width - limit.min.width) % limit.vStep ||
> +			    (size.height - limit.min.height) % limit.hStep)
> +				continue;
> +
> +			sizes.push_back(size);
> +		}
> +	}
> +
> +	std::sort(sizes.begin(), sizes.end());
> +
> +	return sizes;
> +}
> +
> +/**
> + * \brief Retrieve a frame size range
> + * \param[in] pixelformat Pixel format to retrieve range for
> + * \returns Range description of frame size
> + */
> +SizeRange StreamFormats::range(unsigned int pixelformat) const
> +{
> +	auto const it = formats_.find(pixelformat);
> +	if (it == formats_.end())
> +		return {};
> +
> +	const SizeRange &first = it->second.front();
> +	if (it->second.size() == 1)
> +		return first;

There's a lot of mixing of 'first' and 'second' here which is a bit
awkward for comprehension....

> +
> +	LOG(Stream, Debug) << "Building range from discret";

/discret/discrete/

> +
> +	SizeRange range(first.min.width, first.min.height,
> +			first.max.width, first.max.height);
> +
> +	for (const SizeRange &limit : it->second) {
> +		if (limit.min < range.min)
> +			range.min = limit.min;
> +
> +		if (limit.max > range.max)
> +			range.max = limit.max;
> +	}
> +
> +	return range;
> +}
> +
>  /**
>   * \struct StreamConfiguration
>   * \brief Configuration parameters for a stream
> 

-- 
Regards
--
Kieran