[PATCH] drm/fourcc: Add bayer formats and modifiers

Tue Mar 3 08:26:27 UTC 2020

On Fri, 28 Feb 2020 17:31:35 +0100
Niklas Söderlund <niklas.soderlund at ragnatech.se> wrote:

> Bayer formats are used with cameras and contain green, red and blue
> components, with alternating lines of red and green, and blue and green
> pixels in different orders. For each block of 2x2 pixels there is one
> pixel with a red filter, two with a green filter, and one with a blue
> filter. The filters can be arranged in different patterns.
> 
> Add DRM fourcc formats to describe the most common Bayer formats. Also
> add a modifiers to describe the custom packing layouts used by the Intel
> IPU3 and in the MIPI (Mobile Industry Processor Interface) CSI-2
> specification.
> 
> Signed-off-by: Niklas Söderlund <niklas.soderlund at ragnatech.se>
> ---
>  include/uapi/drm/drm_fourcc.h | 95 +++++++++++++++++++++++++++++++++++
>  1 file changed, 95 insertions(+)

Hi,

here are some by-stander comments.

> 
> diff --git a/include/uapi/drm/drm_fourcc.h b/include/uapi/drm/drm_fourcc.h
> index 8bc0b31597d80737..561d5a08ffd16b69 100644
> --- a/include/uapi/drm/drm_fourcc.h
> +++ b/include/uapi/drm/drm_fourcc.h
> @@ -286,6 +286,62 @@ extern "C" {
>  #define DRM_FORMAT_YVU444	fourcc_code('Y', 'V', '2', '4') /* non-subsampled Cr (1) and Cb (2) planes */
>  
>  
> +/*
> + * Bayer formats
> + *
> + * Bayer formats contain green, red and blue components, with alternating lines
> + * of red and green, and blue and green pixels in different orders. For each
> + * block of 2x2 pixels there is one pixel with a red filter, two with a green
> + * filter, and one with a blue filter. The filters can be arranged in different
> + * patterns.
> + *
> + * For example, RGGB:
> + *	row0: RGRGRGRG...
> + *	row1: GBGBGBGB...
> + *	row3: RGRGRGRG...
> + *	row4: GBGBGBGB...
> + *	...
> + *
> + * Vendors have different methods to pack the sampling formats to increase data
> + * density. For this reason the fourcc only describes pixel sample size and the
> + * filter pattern for each block of 2x2 pixels. A modifier is needed to
> + * describe the memory layout.
> + *
> + * In addition to vendor modifiers for memory layout DRM_FORMAT_MOD_LINEAR may
> + * be used to describe a layout where all samples are placed consecutively in
> + * memory. If the sample does not fit inside a single byte, the sample storage
> + * is extended to the minimum number of (little endian) bytes that can hold the
> + * sample and any unused most-significant bits are defined as padding.

"Minimum number of (little endian) bytes" is probably not quite right,
because you could end up with a 3-byte word for e.g. 18-bit samples,
and for those I don't think endianess is even a defined concept. Yes,
you don't add any >16 bit formats here, but being careful here avoids
having to face the question and confusion when someone does add such
formats.

Alternatively, do not even pretend to define any layout for samples >
16 bits, and leave it for the future to be defined if/when the need
arises.

> + *
> + * For example, SRGGB10:
> + * Each 10-bit sample is contained in 2 consecutive little endian bytes, where
> + * the 6 most-significant bits are unused.

Nitpick: I think you mean "10-bit sample is contained in a uint16 word
(little endian), ..."

"little endian byte" sounds like a strange concept to me, as it seems
to delve into the order of bits in a byte, MSB or LSB first. I suspect
most people would not even think of this, but I've been scarred by
reading the Pixman pixel format definitions.

> + */
> +
> +/* 8-bit Bayer formats */
> +#define DRM_FORMAT_SRGGB8	fourcc_code('R', 'G', 'G', 'B')

The S in SRGGB is quite surprising to me. I saw it mentioned in IRC
that it is easy to read as sRGB and I agree. I would not know to
associate S with Bayer to begin with.

Why not e.g. DRM_FORMAT_BAYER_RGGB8?

> +#define DRM_FORMAT_SGRBG8	fourcc_code('G', 'R', 'B', 'G')
> +#define DRM_FORMAT_SGBRG8	fourcc_code('G', 'B', 'R', 'G')
> +#define DRM_FORMAT_SBGGR8	fourcc_code('B', 'A', '8', '1')
> +
> +/* 10-bit Bayer formats */
> +#define DRM_FORMAT_SRGGB10	fourcc_code('R', 'G', '1', '0')
> +#define DRM_FORMAT_SGRBG10	fourcc_code('B', 'A', '1', '0')
> +#define DRM_FORMAT_SGBRG10	fourcc_code('G', 'B', '1', '0')
> +#define DRM_FORMAT_SBGGR10	fourcc_code('B', 'G', '1', '0')
> +
> +/* 12-bit Bayer formats */
> +#define DRM_FORMAT_SRGGB12	fourcc_code('R', 'G', '1', '2')
> +#define DRM_FORMAT_SGRBG12	fourcc_code('B', 'A', '1', '2')

Conflict:

#define DRM_FORMAT_BGRA4444     fourcc_code('B', 'A', '1', '2') /* [15:0] B:G:R:A 4:4:4:4 little endian */

Does the kernel not have a "self-test" that ensures that all format
codes (and why not modifiers as well) are unique?

> +#define DRM_FORMAT_SGBRG12	fourcc_code('G', 'B', '1', '2')
> +#define DRM_FORMAT_SBGGR12	fourcc_code('B', 'G', '1', '2')
> +
> +/* 14-bit Bayer formats */
> +#define DRM_FORMAT_SRGGB14	fourcc_code('R', 'G', '1', '4')
> +#define DRM_FORMAT_SGRBG14	fourcc_code('B', 'A', '1', '4')
> +#define DRM_FORMAT_SGBRG14	fourcc_code('G', 'B', '1', '4')
> +#define DRM_FORMAT_SBGGR14	fourcc_code('B', 'G', '1', '4')
> +
>  /*
>   * Format Modifiers:
>   *
> @@ -309,6 +365,7 @@ extern "C" {
>  #define DRM_FORMAT_MOD_VENDOR_BROADCOM 0x07
>  #define DRM_FORMAT_MOD_VENDOR_ARM     0x08
>  #define DRM_FORMAT_MOD_VENDOR_ALLWINNER 0x09
> +#define DRM_FORMAT_MOD_VENDOR_MIPI 0x0a
>  
>  /* add more to the end as needed */
>  
> @@ -434,6 +491,17 @@ extern "C" {
>   */
>  #define I915_FORMAT_MOD_Y_TILED_GEN12_MC_CCS fourcc_mod_code(INTEL, 7)
>  
> +
> +/*
> + * IPU3 Bayer packing layout
> + *
> + * The IPU3 raw Bayer formats use a custom packing layout where there are no
> + * gaps between each 10-bit sample. It packs 25 pixels into 32 bytes leaving
> + * the 6 most significant bits in the last byte unused. The format is little
> + * endian.

Do I understand that right, that the collection of bytes (not words?)
represents a stream of bits? In which order do you read the bits of a byte
to produce the bits in a 10-bit unit?

I don't think "little endian" specifies that (even less for
non-2/4/8-byte units), and Pixman formats prove that the order could be
specified either way.

Does the "little endian" mean that in the 32 bytes long unit, one needs
to extract in chunks of uint16_t/uint32_t/uint64_t and inspect the bits
of those words, or is the 32-byte unit supposed to be read byte by byte
in which case endianess plays no role?

If possible, it would be good to reword the definition so that these
questions cannot arise.

> + */
> +#define IPU3_FORMAT_MOD_PACKED fourcc_mod_code(INTEL, 8)
> +
>  /*
>   * Tiled, NV12MT, grouped in 64 (pixels) x 32 (lines) -sized macroblocks
>   *
> @@ -804,6 +872,33 @@ extern "C" {
>   */
>  #define DRM_FORMAT_MOD_ALLWINNER_TILED fourcc_mod_code(ALLWINNER, 1)
>  
> +/* Mobile Industry Processor Interface (MIPI) modifiers */
> +
> +/*
> + * MIPI CSI-2 packing layout
> + *
> + * The CSI-2 RAW formats (for example Bayer) use a different packing layout
> + * depenindg on the sample size.
> + *
> + * - 10-bits per sample
> + *   Every four consecutive samples are packed into 5 bytes. Each of the first 4
> + *   bytes contain the 8 high order bits of the pixels, and the 5th byte
> + *   contains the 2 least-significant bits of each pixel, in the same order.

...in the same order? So bits 0-1 are the bits 0-1 of the 1st sample, bits
2-3 are the bits 0-1 of the 2nd sample, etc?

> + *
> + * - 12-bits per sample
> + *   Every two consecutive samples are packed into three bytes. Each of the
> + *   first two bytes contain the 8 high order bits of the pixels, and the third
> + *   byte contains the four least-significant bits of each pixel, in the same
> + *   order.
> + *
> + * - 14-bits per sample
> + *   Every four consecutive samples are packed into seven bytes. Each of the
> + *   first four bytes contain the eight high order bits of the pixels, and the
> + *   three following bytes contains the six least-significant bits of each
> + *   pixel, in the same order.

How do you count the bits when crossing the byte boundary on the last 3
bytes? Is it an imaginary 24-bit word, or do you take it byte by byte
like this:

byte0 bits 0-5: 1st sample bits 0-5
byte0 bits 6-7: 2nd sample bits 0-1
byte1 bits 0-3: 2nd sample bits 2-5
byte1 bits 4-7: 3rd sample bits 0-3
byte2 bits 0-1: 3rd sample bits 4-5
byte2 bits 2-7: 4th sample bits 0-5

> + */
> +#define MIPI_FORMAT_MOD_CSI2_PACKED fourcc_mod_code(MIPI, 1)
> +
>  #if defined(__cplusplus)
>  }
>  #endif

Thanks,
pq
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20200303/5fdb4386/attachment.sig>