xref: /third_party/openGLES/extensions/SGIX/SGIX_pixel_texture_bits.txt

Name

    SGIX_pixel_texture_bits

Name Strings

    GL_SGIX_pixel_texture_bits

Version

    $Date: 1998/10/01 23:30:18 $ $Revision: 1.11 $

Number

    127

Dependencies

    GL_SGIS_pixel_texture
    can be implemented in parallel with SGIX_pixel_texture_lod

Overview

    The pixel_texture extension causes the "convert to fragment"
    stage of the pixel pipeline to derive the fragment's texture
    coordinates from the pixel group's color components.
    At this stage in the pixel pipeline, the color components have
    been converted into the framebuffer format.  These
    framebuffer format values are used to produce texture coordinates
    of exactly the same value.  The internal representation of
    texture coordinates is intentionally left vague in the spec
    so implementations can choose whatever they like.

    This works well when the frame buffer format for
    a color component is a fixed point value in the range [0,1].
    However, if SGIX_color_range is supported, the framebuffer
    format can be floating point, and the values are not
    a linear ramp.  This means that the texture must be
    large and sparsely populated in order to ensure that
    each texture coordinate value reaches a unique texel.

    This extension allows the application to specify that
    the color component be treated like an unsigned integer
    with the same number of bits as are in the framebuffer format,
    which makes each texel in the texture accessible through
    the texture lookup equations with a texture whose size
    is 2^n in each dimension, where n is the number of bits in
    the framebuffer format.

    The application must know the exact frambuffer format
    in order to define the texture correctly.

    For example, this approach allows the implementation of
    a bit-preserving 1D lookup table for the s10e5 bali pixel format.
    It specifies a mapping of the s10e5 pixel format to a texture
    coordinate index in a manner that allows one texel to be
    mapped to each of the 64K bit patterns possible in the
    s10e5 format.

    The motivation is to allow the shader toolkit to implement
    functions, e.g., cos(x), with a copy pixels
    and pixel texture enabled in a way that preserves as
    much accuracy as possible.

IP Status

    Silicon Graphics has filed for patent protection for some of the
    techniques described in this extension document.

Issues

    * Could we define a separate scale factor so that we don't
      necessarily have to specify that we convert to the framebuffer
      format?

      There isn't much need to mention the framebuffer format
      here; we just need to specify how we're going to interpret
      the bits.  However, the application must use a texture
      that corresponds to the framebuffer format.

    * To get to the exact center of a texel, we need to add
      0.5 to the calculation after the conversion to unsigned
      int and before the divide by 2^m-1.

      This is not necessary because the equations for calculating
      the 2x2 square of texels when TEXTURE_MIN_FILTER is LINEAR
      do it.  (Section 3.8 in the spec.)

New Procedures and Functions

    None

New Tokens

    Accepted by the <pname> parameter of PixelTexGenParameteriSGIX and
    PixelTexGenParameterivSGIX:

        COLOR_TO_TEXTURE_COORD_SGIX

    Accepted by the <param> parameter of PixelTexGenParameteriSGIX and
    PixelTexGenParameterivSGIX:

        COLOR_BIT_PATTERN_SGIX
        COLOR_VALUE_SGIX

Additions to Chapter 2 of the 1.0 Specification (OpenGL Operation)

    None

Additions to Chapter 3 of the 1.0 Specification (Rasterization)

    This extension, similar to the pixel_texture.spec modifies the
    "Conversion to Fragments" subsection of section 3.6.3 (Rasterization
    of Pixel Rectangles) of the GL Specification. Immediately following
    the text added by the pixel_texture.spec, insert the following:

    If the PixelTexGenParameterSGIX param COLOR_TO_TEXTURE_COORD_SGIX
    is set to COLOR_BIT_PATTERN_SGIX, the mapping from r,g,b,a
    to s,t,r,q is altered.

    At this stage, final color processing has been performed and the
    r,g,b,a values are in the form that matches the framebuffer.
    To map the red color value to the s texture coordinate the
    m bits of red, interpreted as an integer in the range [0,2^m-1], are
    divided by 2^m-1 to obtain an s coordinate in the range [0,1].
    Green and Blue are mapped similarly.  The q texture coordinate
    is set to one if a divide by q is in effect.  If TEXTURE_4D_SGIS
    is enabled, Alpha is mapped to q in the same way Red is
    mapped to s.

    If the PixelTexGenParameterSGIX param COLOR_TO_TEXTURE_COORD_SGIX
    is set to COLOR_VALUE_SGIX, the r,g,b,a values are used
    to produce floating point values for s,t,r,q.

Additions to Chapter 4 of the 1.0 Specification (Per-Fragment Operations
and the Frame Buffer)

    None


Additions to Chapter 5 of the 1.0 Specification (Special Functions)

    None

Additions to Chapter 6 of the 1.0 Specification (State and State Requests)

    None

Additions to the GLX Specification

    XXX

Errors

New State

    The following is added to Table 6.16. Pixels.

    Get Value                           Get Command                     Type    Initial Value                   Attrib
    ---------                           -----------                     ----    -------------                   ------
    COLOR_TO_TEXTURE_COORD_SGIX         GetPixelTexGenParameterivSGIS   Z2      COLOR_VALUE_SGIX                pixel

New Implementation Dependent State

    None