Name

    OML_subsample

Name Strings

    GL_OML_subsample

Contact

    Jon Leech, Silicon Graphics (ljp 'at' sgi.com)

Status

    Complete. Approved by the Khronos SIG on July 19, 2001.

Version

    Last Modified Date: 07/23/2001
    Author Revision: $Header: //depot/main/doc/registry/extensions/OML/subsample.spec#10 $

Number

    240

Dependencies

    This extension is written against the OpenGL 1.2.1 Specification,

Overview

    Many video image formats and compression techniques utilize various
    component subsamplings, so it is necessary to provide a mechanism to
    specify the up- and down-sampling of components as pixel data is
    drawn from and read back to the client. Though subsampled components
    are normally associated with the video color space, YCrCb, use of
    subsampling in OpenGL does not imply a specific color space. Color
    space conversion may be performed using other extensions or core
    capabilities such as the color matrix.

    This extension defines two new pixel storage formats representing
    subsampled data on the client. It is loosely based on the
    SGIX_subsample extension, but specifies subsampling with the data
    format parameter rather than pixel packing parameters. It also
    adds support for CYA subsampled data.

    When pixel data is received from the client and an unpacking
    upsampling mode other than PIXEL_SUBSAMPLE_NONE_OML is specified,
    upsampling is performed via replication, unless otherwise specified
    by UNPACK_RESAMPLE_OML.

    Similarly, when pixel data is read back to the client and a packing
    downsampling mode other than PIXEL_SUBSAMPLE_NONE_OML is specified,
    downsampling is performed via simple component decimation (point
    sampling), unless otherwise specified by PACK_RESAMPLE_OML.

Issues

  * Which subsampled component orderings should be supported?

    Only CY and CYA component ordering, since this matches contemporary
    video hardware. YC and YCA ordering will require a separate
    extension defining new formats.

  * The new enumerant naming scheme gives the component frequencies in
    the same order as the components themselves; that is,
    FORMAT_SUBSAMPLE_24_24_OML corresponds to CY 4:2:2, and
    FORMAT_SUBSAMPLE_244_244_OML corresponds to CYA 4:2:2:4. This makes
    naming YC and YCA orderings easier.

  * Should subsampling be specified with new pixel storage parameters,
    like the SGIX_subsample extension, or with new formats, like the
    EXT_422 extension?

    With new formats. There are many invalid format/type combinations
    when specifying subsampling with a pixel storage parameter. Also,
    there's an ambiguity when doing this because the <format> parameter
    represents the after-upsampling data format, not the host format.

  * Because subsampled data is inherently pixel / texture oriented, this
    extension only supports the new formats for pixel and texture
    operations; it does not support them for convolution filters,
    histograms, minmax, or color tables.

  * The only packed pixel type supported is 10_10_10_2, since this is
    needed for video data interoperability. It would be possible to
    support many other packed pixel formats, but most are unused in
    practice.

    Is support for other packed pixels types, particularly
    2_10_10_10_REV, required?

  * Should readbacks of non-even widths be allowed when downsampling?

    No. This is not consistent with draw operations, where this
    constraint already exists. It also makes OML_resample more complex
    when using an AVERAGE filter, since the edge cases may also apply to
    even pixel coordinates. The spec may need to be more explicit about
    this restriction.

IP Status

    No known issues.

New Procedures and Functions

    None.

New Tokens

    Accepted by the <format> parameter of DrawPixels, ReadPixels,
    TexImage1D, TexImage2D, TexImage3D, TexSubImage1D, TexSubImage2D,
    TexSubImage3D, and GetTexImage

	FORMAT_SUBSAMPLE_24_24_OML	    0x8982
	FORMAT_SUBSAMPLE_244_244_OML	    0x8983

Additions to Chapter 2 of the OpenGL 1.2.1 Specification (OpenGL Operation)

    None.

Additions to Chapter 3 of the OpenGL 1.2.1 Specification (Rasterization)

  - (3.6.4, p. 88 "Rasterization of Pixel Rectangles")

    Add prior to the "Unpacking" subsection on page 90:

    If <format> is FORMAT_SUBSAMPLE_24_24_OML or
    FORMAT_SUBSAMPLE_244_244_OML, and <type> is one of the packed pixel
    formats in table 3.8 other than UNSIGNED_INT_10_10_10_2, then the
    error INVALID_OPERATION occurs; if <width> is not a multiple of 2
    pixels, or if the value of the UNPACK_SKIP_PIXELS or
    UNPACK_ROW_LENGTH parameters is not a multiple of 2 pixels, then the
    error INVALID_OPERATION occurs.

  - Add new entries to table 3.6:

    Format Name			    Element Meaning and Order	Target Buffer
    -----------			    -------------------------	-------------
    FORMAT_SUBSAMPLE_24_24_OML	    CbY / CrY			Color
    FORMAT_SUBSAMPLE_244_244_OML    CbYA / CrYA			Color

  - Append to the caption of table 3.6:

    Subsampled formats yield components that are further modified during
    conversion to uniform sampling. The subsampled components are
    denoted as Cb, Y, Cr, and A, although subsampled data is not defined
    to be in any specific color space.

  - Modify table 3.8:

    <type> Parameter	    GL Data Number of  Matching
    Token Name		    Type    Components Pixel Formats
    ----------------------- ------- ---------- -------------
    UNSIGNED_INT_10_10_10_2 uint    2, 3, 4    RGBA, BGRA,
					       FORMAT_SUBSAMPLE_24_24_OML,
					       FORMAT_SUBSAMPLE_244_244_OML

  - Append to the caption of table 3.8:

    Subsampled formats may pack components from multiple groups into a
    single uint.

  - Modify table 3.11's UNSIGNED_INT_10_10_10_2 entry:

    UNSIGNED_INT_10_10_10_2, <format>s RGBA and BGRA:

	(use existing 4-component diagram)

    UNSIGNED_INT_10_10_10_2, <format> SUBSAMPLE_24_24_OML:

	 31		22 21		  12 11		     2 1  0
	| pixel 0/comp. 0 | pixel 0/comp. 1 | pixel 1/comp. 0 | xx |
	| pixel 1/comp. 1 | pixel 2/comp. 0 | pixel 2/comp. 1 | xx |

    UNSIGNED_INT_10_10_10_2, <format> SUBSAMPLE_244_244_OML:

	 31		22 21		  12 11		     2 1  0
	| 1st comp.	  | 2nd comp.	    | 3rd comp.       | xx |

  - Change caption of table 3.11:

    Table 3.11: UNSIGNED_INT formats. Subsampled formats are packed into
    words, so components from a group may lie in different words. ``xx''
    fields are unused.

  - Add new subsection before "Conversion to RGB" on page 99:

    Conversion to Uniform Sampling

    This step is applied only to subsampled data. If <format> is
    FORMAT_SUBSAMPLE_24_24_OML, then the number of components per pixel
    is increased from two to three. If <format> is
    FORMAT_SUBSAMPLE_244_244_OML. then the number of components per
    pixel is increased from three to four.

    After conversion to uniform sampling (see figure 3.9). pixels are
    thereafter treated as though they were RGB (three component) or RGBA
    (four component) format.

    In the remainder of this section, the j'th component of the i'th
    pixel in a row is denoted by S_i,j (for source pixels in client
    memory) and D_i,j (for destination pixels in the color buffer).

    Destination component values are defined as:

    For even pixels ((i mod 2) == 0):

	D_i,0 = S_i,0
	D_i,1 = S_i,1
	D_i,2 = S_i+1,0
	D_i,3 = S_i,2

    For odd pixels ((i mod 2) == 1):

	D_i,0 = S_i-1,0
	D_i,1 = S_i,1
	D_i,2 = S_i,0
	D_i,3 = S_i,2

  - Add new figure 3.9 (renumber following figures):

    FORMAT_SUBSAMPLE_24_24_OML:
    <Cb0,Y0>	 <Cr0,Y1>     <Cb2,Y2>	  <Cr2, Y3>
      |  |	  |   |        |   |	    |	|
      |  |   ____/ \__|__      |   |   ____/ \__|__
      |  |  /	      |  \     |   |  /		|  \
      |\_|__|_____    |  |     |\__|__|_____	|  |
      |  |  |	  \   |  |     |   |  |     \	|  |
      V  V  V	  V   V  V     V   V  V     V	V  V
    <Cb0,Y0,Cr0> <Cb0,Y1,Cr0> <Cb2,Y2,Cr2> <Cb2,Y3,Cr2>
    < R0,G0,B1 > < R0,G1,B1 > < R2,G2,B3 > < R2,G3,B3 >

    FORMAT_SUBSAMPLE_244_244_OML:
    <Cb0,Y0,A0>     <Cr0,Y1,A1>     <Cb2,Y2,A2>     <Cr2,Y3,A3>
      |  |   \	     |	 |   \	     |	 |   \	     |	 |   \
      |  |   _\_____/ \__|__  \      |	 |   _\_____/ \__|__  \
      |  |  /  \	 |  \  \     |	 |  /  \	 |  \  \
      |\_|__|___\____	 |  |	\    |\__|__|___\____	 |  |	\
      |  |  |	|    \	 |  |	|    |	 |  |	|    \	 |  |	|
      V  V  V	V    V	 V  V	V    V	 V  V	V    V	 V  V	V
    <Cb0,Y0,Cr0,A0> <Cb0,Y1,Cr0,A1> <Cb2,Y2,Cr2,A2> <Cb2,Y3,Cr2,A3>
    < R0,G0,B1,A0 > < R0,G1,B1,A1 > < R2,G2,B3,A2 > < R2,G3,B3,A3 >

    Figure 3.9: Upsampling with component replication of subsampled data
    from client memory to form RGB or RGBA pixels.


Additions to Chapter 4 of the OpenGL 1.2.1 Specification (Per-Fragment
Operations and the Frame Buffer)

    4.3.2 Reading Pixels

  - Add new subsection before "Final Conversion" on page 160:

    Conversion to Subsampled Form

    This step is applied only if <format> is FORMAT_SUBSAMPLE_24_24_OML
    or FORMAT_SUBSAMPLE_244_244_OML.

    In the remainder of this section, the j'th component of the i'th
    pixel in a row is denoted by S_i,j (for source pixels in the color
    buffer) and D_i,j (for destination pixels in client memory).

    If <format> is FORMAT_SUBSAMPLE_24_24_OML, then the resulting pixels
    have 2 components; if <format> is FORMAT_SUBSAMPLE_244_244_OML, then
    the resulting pixels have 3 components (see figure 4.3). Destination
    component values are defined as:

    For even pixels ((i mod 2) == 0):

	D_i,0 = S_i,0
	D_i,1 = S_i,1
	D_i,2 = S_i,3	(only for FORMAT_SUBSAMPLE_244_244_OML)

    For odd pixels ((i mod 2) == 1):

	D_i,0 = S_i-1,2
	D_i,1 = S_i,1
	D_i,2 = S_i,3	(only for FORMAT_SUBSAMPLE_244_244_OML)

  - Add new figure 4.3 (renumber following figures):

    FORMAT_SUBSAMPLE_24_24_OML:
    <R0,G0,B0,A0> <R1,G1,B1,A1> <R2,G2,B2,A2> <R3,G3,B3,A3>
      |  | |   |   |  |  |  |	 |  |  |  |    |  |  |	|
      |  |  \  *   *  |  *  *	 |  |	\ *    *  |  *	*
      |  |   |	      |		 |  |	 |	  |
      V  V   V	      V		 V  V	 V	  V
    <Cb0,Y0><Cr0,     Y1>	<Cb2,Y2><Cr2,	  Y3>
    <--- pixel pair ---->	<--- pixel pair ---->

    FORMAT_SUBSAMPLE_244_244_OML:
    <R0,G0,B0,A0> <R1,G1,B1,A1> <R2,G2,B2,A2> <R3,G3,B3,A3>
      |  |  |  |   |   |  |  |	  |  |	|  |   |   |  |  |
      |  |  \__|__ *   |  *  |	  |  |	\__|__ *   |  *  |
      |  |     |  \    |     |	  |  |	   |  \    |	 |
      |  |    /    |   |    /	  |  |	  /    |   |	/
      V  V   V	   V   V   V	  V  V	 V     V   V   V
    <Cb0,Y0,A0>   <Cr0,Y1,A1>	<Cb2,Y2,A2>   <Cr2,Y3,A3>
    <--- pixel pair ---->	<--- pixel pair ---->

    Figure 4.3: Downsampling of RGB or RGBA pixels to form subsampled
    data in host memory.


  - Add prior to the last sentence of subsection "Placement in Client
    Memory" on page 162:

    If <format> is FORMAT_SUBSAMPLE_24_24_OML, then only the
    corresponding two elements (first two components of each group) are
    written. If <format> is FORMAT_SUBSAMPLE_244_244_OML, then only the
    corresponding three elements (first three components of each group)
    are written.

Additions to Chapter 5 of the OpenGL 1.2.1 Specification (Special Functions)

    None.

Additions to Chapter 6 of the OpenGL 1.2.1 Specification (State and State Requests)

    None.

Additions to the GLX 1.3 Specification

    TBD. Discussion of image formats in the GLX Protocol Specification
    may need to be expanded.

Errors

    See above.

New State

    None.

New Implementation Dependent State

    None.

Revision History

  * Revision 10, 07/24/2001 - Finalized Status for OpenML 1.0.
  * Revision 9, 07/16/2001 - Fix label in 24_24 packed pixel diagram.
  * Revisions 7-8, 07/11/2001 - Assign enum values and extension number
    for the registry.
  * Revision 6 - Correct errors in the equations describing subsampling.
  * Revision 5 - formatting changes for OpenML Specification
  * Revision 4 - Rewrite to use the <format> parameter, rather than a
    pixel storage mode, to specify subsampled data. Specify which
    format/type combinations are allowed for subsampled data, and define
    the representation of 10-bit component subsampled packed pixel data.
  * Revision 3 - Removed support for YC component orders. Renamed CY and
    CYA enumerants more sensibly. Changed text descriptions of sampling
    to equations. Made enum values undefined until we've determined if
    this extension is backwards compatible with SGIX_subsample.
  * Revision 2 - corrected 4224 upsampling and downsampling figures.
    Moved discussion of errors for non-even image widths from the
    OML_resample specification.
  * Revision 1 - derived from SGIX_subsample.