Name

    ARB_pixel_buffer_object

Name Strings

    GL_ARB_pixel_buffer_object

Notice

    Copyright (c) 2004-2013 The Khronos Group Inc. Copyright terms at
        http://www.khronos.org/registry/speccopyright.html

Status

    Complete. Approved by ARB on December 7, 2004.

Contributors

    Ralf Biermann
    Nick Carter
    Derek Cornish
    Matt Craighead
    Mark Kilgard
    Dale Kirkland
    Jon Leech
    Brian Paul
    Thomas Roell
    Ian Romanick
    Jeremy Sandmel
    Shazia Rahman

Contact

    Mark J. Kilgard, NVIDIA Corporation (mjk 'at' nvidia.com)
    Ralf Biermann, NVIDIA Corporation (rbiermann 'at' nvidia.com)
    Derek Cornish, NVIDIA Corporation (dcornish 'at' nvidia.com)

IP Status

    None.

Version

    Last Modified Date: October 8, 2013
    Revision: 1.2

Number

    ARB Extension #42

Dependencies

    Written based on the wording of the OpenGL 2.0 specification.

    Assumes support for (at least) OpenGL 1.5 or the
    ARB_vertex_buffer_object extension.

    NV_pixel_data_range affects the definition of this extension.

    EXT_pixel_buffer_object interacts with this extension.

Overview

    This extension expands on the interface provided by the
    ARB_vertex_buffer_object extension (and later integrated into OpenGL
    1.5) in order to permit buffer objects to be used not only with vertex
    array data, but also with pixel data.  The intent is to provide more
    acceleration opportunities for OpenGL pixel commands.

    While a single buffer object can be bound for both vertex arrays and
    pixel commands, we use the designations vertex buffer object (VBO)
    and pixel buffer object (PBO) to indicate their particular usage in
    a given situation.

    Recall that buffer objects conceptually are nothing more than arrays
    of bytes, just like any chunk of memory.  ARB_vertex_buffer_object
    allows GL commands to source data from a buffer object by binding the
    buffer object to a given target and then overloading a certain set of
    GL commands' pointer arguments to refer to offsets inside the buffer,
    rather than pointers to user memory.  An offset is encoded in a
    pointer by adding the offset to a null pointer.

    This extension does not add any new functionality to buffer objects
    themselves.  It simply adds two new targets to which buffer objects
    can be bound: GL_PIXEL_PACK_BUFFER and GL_PIXEL_UNPACK_BUFFER.  When a
    buffer object is bound to the GL_PIXEL_PACK_BUFFER target, commands
    such as glReadPixels pack (write) their data into a buffer object.
    When a buffer object is bound to the GL_PIXEL_UNPACK_BUFFER target,
    commands such as glDrawPixels and glTexImage2D unpack (read) their
    data from a buffer object.

    There are a several approaches to improve graphics performance
    with PBOs.  Some of the most interesting approaches are:

    - Streaming texture updates:  If the application uses
      glMapBuffer/glUnmapBuffer to write its data for glTexSubImage into
      a buffer object, at least one of the data copies usually required
      to download a texture can be eliminated, significantly increasing
      texture download performance.

    - Streaming draw pixels: When glDrawPixels sources client memory,
      OpenGL says the client memory can be modified immediately after the
      glDrawPixels command returns without disturbing the drawn image.
      This typically necessitates unpacking and copying the image prior
      to glDrawPixels returning.  However, when using glDrawPixels with
      a pixel pack buffer object, glDrawPixels may return prior to image
      unpacking because future modification of the buffer data requires
      explicit commands (glMapBuffer, glBufferData, or glBufferSubData).

    - Asynchronous glReadPixels:  If an application needs to read back a
      number of images and process them with the CPU, the existing GL
      interface makes it nearly impossible to pipeline this operation.
      The driver will typically send the hardware a readback command
      when glReadPixels is called, and then wait for all of the data to
      be available before returning control to the application.  Then,
      the application can either process the data immediately or call
      glReadPixels again; in neither case will the readback overlap with
      the processing.  If the application issues several readbacks
      into several buffer objects, however, and then maps each one to
      process its data, then the readbacks can proceed in parallel with
      the data processing.

    - Render to vertex array:  The application can use a fragment
      program to render some image into one of its buffers, then read
      this image out into a buffer object via glReadPixels.  Then, it can
      use this buffer object as a source of vertex data.

Issues

    1)  How does this extension relate to ARB_vertex_buffer_object?

        It builds on the ARB_vertex_buffer_object framework by adding
        two new targets that buffers can be bound to.

    2)  How does this extension relate to NV_pixel_data_range?

        This extension relates to NV_pixel_data_range in the same way
        that ARB_vertex_buffer_object relates to NV_vertex_array_range.
        To paraphrase the ARB_vertex_buffer_object spec, here are the
        main differences:

        - Applications are no longer responsible for memory management
          and synchronization.

        - Applications may still access high-performance memory directly,
          but this is optional, and such access is more restricted.

        - Buffer changes (glBindBuffer) are generally expected to be
          very lightweight, rather than extremely heavyweight
          (glPixelDataRangeNV).

        - A platform-specific allocator such as wgl/glXAllocateMemoryNV
          is no longer required.

    3)  Can a given buffer be used for both vertex and pixel data?

        RESOLVED: YES.  All buffers can be used with all buffer bindings,
        in whatever combinations the application finds useful.  Consider
        yourself warned, however, by the following issue.

    4)  May implementations make use of the target as a hint to select
        an appropriate memory space for the buffer?

        RESOLVED: YES, as long as such behavior is transparent to the
        application.  Some implementations may choose, for example, that
        they would rather stream vertex data from AGP memory, element
        (index) data from video memory, and pixel data from video memory.
        In fact, one can imagine arbitrarily complicated heuristics for
        selecting the memory space, based on factors such as the target,
        the "usage" argument, and the application's observed behavior.

        While it is entirely legal to create a buffer object by binding
        it to GL_ARRAY_BUFFER and loading it with data, then using it
        with the GL_PIXEL_UNPACK_BUFFER_ARB or GL_PIXEL_PACK_BUFFER_ARB
        binding, such behavior is liable to confuse the driver and may
        hurt performance.  If the driver implemented the hypothetical
        heuristic described earlier, such a buffer might have already
        been located in AGP memory, and so the driver would have to choose
        between two bad options: relocate the buffer into video memory, or
        accept lower performance caused by streaming pixel data from AGP.

    5)  Should all pixel path commands be supported, or just a subset
        of them?

        RESOLVED: ALL.  While there is little reason to believe that,
        say, glConvolutionFilter2D would benefit from this extension,
        there is no reason _not_ to support it.  The complete list of
        commands affected by this extension is listed in issues 17 and 18.

    6)  Should glPixelMap and glGetPixelMap be supported?

        RESOLVED: YES.  They're not really pixel path operations, but,
        again, there is no good reason to omit operations, and they _are_
        operations that pass around big chunks of pixel-related data.
        If we support glPolygonStipple, surely we should support this.

    7)  How does the buffer binding state push/pop?

        RESOLVED: As part of the pixel store client state.  This is
        analogous to how the ARB_vertex_buffer_object bindings
        pushed/popped as part of the vertex array client state.

    8)  Should NV_pixel_data_range (PDR) be used concurrently with pixel
        buffer objects ?

        RESOLVED: NO. While it would be possible to allocate a memory
        range for PDR, using a pointer into this memory range with one
        of the commands affected by PBOs will not work if a pixel buffer
        object other than zero is bound to the buffer binding point
        affecting the command.

        Pixel buffer objects always have higher precedence than PDR.

    9)  Should the INVALID_OPERATION error be generated if a pixel
        command would access data outside the range of the bound PBO?

        RESOLVED:  YES.  This requires considering the command parameters
        (such as width/height/depth/format/type/pointer), the current
        pixel store (pack/unpack) state, and the command operation itself
        to determine the maximum addressed byte for the pixel command.

        Brian Paul strongly recommends this behavior.

        This behavior should increase the reliability of using PBO and
        guard against programmer mistakes.

        This is particularly important for glReadPixels where returning
        data into a region outside the PBO could cause corruption of
        application memory.

        Such bounds checking is substantially more expensive for VBO
        accesses because bounds checking on a per-vertex element basis
        for each of multiple enabled vertex arrays prior to performing
        the command compromises the performance justification of VBO.

    10) If a pixel command with a bound PBO accesses data outside the
        range of the PBO, thereby generating a GL_INVALID_OPERATION error,
        can the pixel command end up being partially processed?

        RESOLVED:  NO.  As for all GL errors excepting GL_OUT_OF_MEMORY
        situations, "the command generating the error is ignored so that
        it has no effect on GL state or framebuffer contents."

        This means implementations must determine before the pixel command
        is performed whether the resulting read or write operations on
        the bound PBO will exceed the size of the PBO.

        This means an implementation is NOT allowed to detect out of
        bounds accesses in the middle of performing the command.

    11) How expensive is it to predetermine whether a pixel command
        accessing a PBO would have an out of bounds access?

        See the "Appendix on Pack/Unpack Range" to see the computations
        involved in computing the access limit.

        Implementations can further specialize and optimize the check
        to make this out of bounds checking negligible for any sizable
        pixel payload.

    12) Should feedback and select buffers output results into a
        buffer object?

        RESOLVED:  That might be useful for a future extension but is
        not appropriate for this extension.  New targets (other than
        PIXEL_PACK_BUFFER_ARB and PIXEL_UNPACK_BUFFER_ARB) make sense.

    13) Should NV_pixel_data_range interactions be documented in
        this specification?

        RESOLVED:  YES.  Interactions with NV_pixel_data_range are
        important to document to facilitate developers migrating to
        the multi-vendor ARB_pixel_buffer_object extension.  Discussion of
        interactions is limited to the issues and example usage sections.

        Other ARB specifications follow this policy, and Jon Leech agrees
        with this policy.

    14) Should an INVALID_OPERATION error be generated if the offset
        within a pixel buffer to a datum comprising of N basic machine
        units is not a multiple of N?

        RESOLVED:  YES.  This was stated for VBOs but no error was
        defined if the rule was violated.  Perhaps this needs to be
        better specified for VBO.

        For PBO, it is reasonable and cheap to enforce the alignment rule.
        For pixel commands it means making sure the offset is evenly
        divisible by the component or group size in basic machine units.

        This check is independent of the pixel store state because the
        pixel store state is specified in terms of pixels (not basic
        machine units) so pixel store addressing cannot create an
        unaligned access as long as the base offset is aligned.

        Certain commands (specifically glPolygonStipple,
        glGetPolygonStipple, glBitmap, glCompressedTexImage1D,
        glCompressedTexImage2D, glCompressedTexImage3D,
        glCompressedTexSubImage1D, glCompressedTexSubImage2D,
        glCompressedTexSubImage3D, and glGetCompressedTexImage) are not
        affected by this error because the data accessed is addressed
        at the granularity of basic machine units.

    15) Various commands do not make explicit reference to supporting
        packing or unpacking from a pixel buffer object but rather specify
        that parameters are handled in the same manner as glDrawPixels,
        glReadPixels, or the glCompressedTexImage commands.  So do such
        commands (example: glCompressedTexSubImage2D) use pixel buffers?

        RESOLVED:  YES.  Commands that have their behavior defined based
        on commands that read or write from pixel buffers will themselves
        read or write from pixel buffers.  Relying on this reduces the
        amount of specification language to be updated.

    16) What is the complete list of commands that can unpack (read)
        pixels from the current pixel unpack buffer object?

            glBitmap
            glColorSubTable
            glColorTable
            glCompressedTexImage1D
            glCompressedTexImage2D
            glCompressedTexImage3D
            glCompressedTexSubImage1D
            glCompressedTexSubImage2D
            glCompressedTexSubImage3D
            glConvolutionFilter1D
            glConvolutionFilter2D
            glDrawPixels
            glPixelMapfv
            glPixelMapuiv
            glPixelMapusv
            glPolygonStipple
            glSeparableFilter2D
            glTexImage1D
            glTexImage2D
            glTexImage3D
            glTexSubImage1D
            glTexSubImage2D
            glTexSubImage3D

    17) What is the complete list of commands that can pack (write)
        pixels into the current pixel pack buffer object?

            glGetCompressedTexImage
            glGetConvolutionFilter
            glGetHistogram
            glGetMinmax
            glGetPixelMapfv
            glGetPixelMapuiv
            glGetPixelMapusv
            glGetPolygonStipple
            glGetSeparableFilter,
            glGetTexImage
            glReadPixels

    18) How does support for pixel buffer objects affect the GLX protocol?

        UNRESOLVED:  See the "GLX Protocol" section.

    19) Prior to this extension, passing zero for the data argument of
        glTexImage1D, glTexImage2D, and glTexImage3D defined a texture
        image level without supplying an image.  How does this behavior
        change with this extension?

        RESOLVED:  The "unspecified image" behavior of the glTexImage
        calls only applies when bound to a zero pixel unpack buffer
        object.

        When bound to a non-zero pixel unpack buffer object, the data
        argument to these calls is treated as an offset rather than
        a pointer so zero is a reasonable and even likely value that
        corresponds to the very beginning of the buffer object's data.

        So to create a texture image level with unspecified image data,
        you MUST bind to the zero pixel unpack buffer object.

        See the ammended language at the end of section 3.8.1.

    20) How does this extension support video frame grabbers?

        RESOLVED:  This extension extends buffer objects so they can
        operate with pixel commands, rather than just vertex array
        commands.

        We anticipate that a future extension may provide a mechanism
        for transferring video frames from video frame grabber hardware
        or vertices from motion capture hardware (or any other source
        of aquired real-time data) directly into a buffer object to
        eliminate a copy.  Ideally, such transfers would be possible
        without requiring mapping of the buffer object.  But this
        extension does not provide such functionality.

        We anticipate such functionality to involve binding a buffer
        object to a new target type, configuring a source (or sink) for
        data (video frames, motion capture vertex sets, etc.), and then
        commands to initiate data transfers to the bound buffer object.

    21) Can this ARB extension share the same enumerants with the EXT
        version of this functionality?

        RESOLVED:  YES.  The ARB extension is functionally compatible
        with EXT_pixel_buffer_object except that the ARB version adds
        additional error checks for alignment and buffer bounds checking.

        The EXT behavior in the case of alignment violations and buffer
        bounds overflow are technically undefined.  The ARB extension
        simply defines the EXT extension's undefined behavior to be an
        OpenGL error.

        Using the same enumerants with firmed up error checking (that
        would otherwise indicate buggy usage) is preferable to two sets
        of enumerants where the older EXT set simply allows sloppy usage.

    22) The expected usage parameters (GL_STREAM_DRAW, etc.) for
        glBufferData are not clearly specified.  How can they be improved?

        RESOLVED:  To improve the clarity, replace the phrase "specified
        once" with "specified once per repetition of the usage pattern" so
        that it is clear for the STREAM_* usage modes (and the STATIC_*
        usage modes too, just much less frequently) that the repeated
        specification is part of a pattern and it is expected that the
        buffer can be, and will be for the STREAM_* usage patterns,
        specified again after being used and this is likely to repeat.

        Additionally, the *_COPY and *_DRAW usage patterns can source
        the data with "a GL drawing command" but also with image
        specification commands so change this phrase to "a GL drawing
        or image specification command."

    23) Is this the "right" way to expose render-to-vertex-array?

        DISCUSSION:  You can use this extension to render an image
        into a framebuffer, copy the pixels into a buffer object with
        glReadPixels, and then configure vertex arrays to source the pixel
        data as vertex attributes.  This necessarily involves a copy
        from the framebuffer to the buffer object.  Future extensions
        may provide mechanisms for copy-free render-to-vertex-array
        capabilities but that is not a design goal of this extension.

New Procedures and Functions

    None.


New Tokens

    Accepted by the <target> parameters of BindBuffer, BufferData,
    BufferSubData, MapBuffer, UnmapBuffer, GetBufferSubData,
    GetBufferParameteriv, and GetBufferPointerv:

        PIXEL_PACK_BUFFER_ARB                        0x88EB
        PIXEL_UNPACK_BUFFER_ARB                      0x88EC

    Accepted by the <pname> parameter of GetBooleanv, GetIntegerv,
    GetFloatv, and GetDoublev:

        PIXEL_PACK_BUFFER_BINDING_ARB                0x88ED
        PIXEL_UNPACK_BUFFER_BINDING_ARB              0x88EF


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

    None

 -- Section 2.9 "Buffer Objects"

    Replace the first two paragraphs with:

    "The vertex data arrays described in section 2.8 are stored in
    client memory.  It is sometimes desirable to store frequently accessed
    client data, such as vertex array and pixel data, in high-performance
    server memory.  GL buffer objects provide a mechanism for clients to
    use to allocate, initialize, and access such memory."

    The name space for buffer objects is the unsigned integer, with zero
    reserved for the GL.  A buffer object is created by binding an unused
    name to a buffer target.  A buffer object is bound by calling

       void BindBuffer(enum target, uint buffer);

    /target/ must be one of ARRAY_BUFFER, ELEMENT_ARRAY_BUFFER,
    PIXEL_UNPACK_BUFFER_ARB, or PIXEL_PACK_BUFFER_ARB.  The ARRAY_BUFFER
    target is discussed in section 2.9.1  The ELEMENT_ARRAY_BUFFER target
    is discussed in section 2.9.2.  The PIXEL_UNPACK_BUFFER_ARB and
    PIXEL_PACK_BUFFER_ARB targets are discussed later in sections 3.6,
    4.3.2, and 6.1.  If the buffer object named /buffer/ has not been
    previously bound or has been deleted since the last binding, the
    GL creates a new state vector, initialized with a zero-sized memory
    buffer and comprising the state values listed in table 2.6."

    Replace the 5th paragraph with:

    "Initially, each buffer object target is bound to zero.  There is
    no buffer object corresponding to the name zero so client attempts
    to modify or query buffer object state for a target bound to zero
    generate an INVALID_OPERATION error."

    Replace the phrase listing the valid targets for BufferData in the
    9th paragraph with:

    "with target set to one of ARRAY_BUFFER, ELEMENT_ARRAY_BUFFER,
    PIXEL_UNPACK_BUFFER_ARB, or PIXEL_PACK_BUFFER_ARB,"

    In the 10th paragraph describing buffer object usage modes, replace
    the phrase "specified once" with "specified once per repetition of
    the usage pattern" for the STREAM_* and STATIC_* usage values.

    Also in the 10th paragraph describing buffer object usage modes,
    replace the phrases "of a GL drawing command." and "for GL drawing
    commands." with "for GL drawing and image specification commands." for
    the *_DRAW and *_COPY usage values.

    Replace the phrase listing the valid targets for BufferSubData in
    the 15th paragraph with:

    "with target set to one of ARRAY_BUFFER, ELEMENT_ARRAY_BUFFER,
    PIXEL_UNPACK_BUFFER_ARB, or PIXEL_PACK_BUFFER_ARB."

    Replace the phrase listing the valid targets for MapBuffer in the
    16th paragraph with:

    "with target set to one of ARRAY_BUFFER, ELEMENT_ARRAY_BUFFER,
    PIXEL_UNPACK_BUFFER_ARB, or PIXEL_PACK_BUFFER_ARB."

    Replace the phrase listing the valid targets for UnmapBuffer in the
    21st paragraph with:

    "with target set to one of ARRAY_BUFFER, ELEMENT_ARRAY_BUFFER,
    PIXEL_UNPACK_BUFFER_ARB, or PIXEL_PACK_BUFFER_ARB."

 -- Section 2.9.2 "Array Indices in Buffer Objects"

    Delete the 3rd paragraph that explains how the ELEMENT_ARRAY_BUFFER
    target is acceptable for the commands specified in section 2.9.
    The updated section 2.9 language already says this.

 -- NEW Section 2.9.3 "Buffer Object Required State"

    "The state required to support buffer objects consists of binding
    names for the array buffer, element buffer, pixel unpack buffer, and
    pixel pack buffer.  Additionally, each vertex array has an associated
    binding so there is a buffer object binding for each of the vertex
    array, normal array, color array, index array, multiple texture
    coordinate arrays, edge flag array, secondary color array, fog
    coordinate array, and vertex attribute arrays.  The initial values for
    all buffer object bindings is zero.

    The state of each buffer object consists of a buffer size in basic
    machine units, a usage parameter, an access parameter, a mapped
    boolean, a pointer to the mapped buffer (NULL if unmapped), and the
    sized array of basic machine units for the buffer data."


Additions to Chapter 3 of the 1.2.1 Specification (Rasterization)

 -- Section 3.6 "Pixel Rectangles"

    Replace the 1st sentence in the 2nd paragraph:

    "A number of parameters control the encoding of pixels in buffer
    object or client memory (for reading and writing) and how pixels
    are processed before being placed in or after being read from the
    framebuffer (for reading, writing, and copying)."

 -- RENAME Section 3.6.1 "Pixel Storage Modes and Pixel Buffer Objects"

    Add to the end of the section:

    "In addition to storing pixel data in client memory, pixel data
    may also be stored in buffer objects (described in section 2.9).
    The current pixel unpack and pack buffer objects are designated
    by the PIXEL_UNPACK_BUFFER_ARB and PIXEL_PACK_BUFFER_ARB targets
    respectively.

    Initially, zero is bound for the PIXEL_UNPACK_BUFFER_ARB, indicating
    that image specification commands such as DrawPixels source their
    pixels from client memory pointer parameters.  However, if a non-zero
    buffer object is bound as the current pixel unpack buffer, then
    the pointer parameter is treated as an offset into the designated
    buffer object."

 -- Section 3.6.3 "Pixel Transfer Modes", page 116.

    Replace the last phrase in the 2nd paragraph with:

    "and /values/ refers to an array of size map values."

    [values is no longer necessarily a pointer.]

    Add the following paragraph after the third paragraph:

    "If a pixel unpack buffer is bound (as indicated by a non-zero
    value of PIXEL_UNPACK_BUFFER_BINDING_ARB), /values/ is an offset
    into the pixel unpack buffer; otherwise, /values/ is a pointer to a
    block client memory.  All pixel storage and pixel transfer modes are
    ignored when specifying a pixel map.  n machine units are read where
    n is the /size/ of the pixel map times the size of a float, uint,
    or ushort datum in basic machine units, depending on the respective
    PixelMap version.  If a pixel unpack buffer object is bound and data+n
    is greater than the size of the pixel buffer, INVALID_OPERATION
    results.  If a pixel unpack buffer object is bound and /values/ is
    not evenly divisible into the number of basic machine units needed
    to store in memory a float, uint, or ushort datum depending on their
    respective PixelMap version, INVALID_OPERATION results."

 -- Section 3.6.4 "Rasterization of Pixel Rectangles", page 126.

    Change the 1st sentence of the 1st paragraph to read:

    "The process of drawing pixels encoded in buffer objects or client
    memory is diagrammed in figure 3.7."

    Change the 4th sentence of the 2nd paragraph to read:

    "/data/ refers to the data to be drawn."

    [data is no longer necessarily a pointer.]

    Change the initial phrase in the 1st sentence of the 1st paragraph
    after "Unpacking" to read:

    "Data are taken from the currently bound pixel unpack buffer or
    client memory as a sequence of..."

    Insert this paragraph after the 1st paragraph after "Unpacking":

    "If a pixel unpack buffer is bound (as indicated by a non-zero
    value of PIXEL_UNPACK_BUFFER_BINDING_ARB), /data/ is an offset
    into the pixel unpack buffer and the pixels are unpacked from the
    buffer relative to this offset; otherwise, /data/ is a pointer to
    a block client memory and the pixels are unpacked from the client
    memory relative to the pointer.  If a pixel unpack buffer object
    is bound and unpacking the pixel data according to the process
    described below would access memory beyond the size of the pixel
    unpack buffer's memory size, INVALID_OPERATION results.  If a pixel
    unpack buffer object is bound and /data/ is not evenly divisible
    into the number of basic machine units needed to store in memory the
    corresponding GL data type from table 3.5 for the /type/ parameter,
    INVALID_OPERATION results."

 -- Section 3.8.1 "Texture Image Specification", page 150.

    Replace the last phrase in the 2nd to last sentence in the 1st
    paragraph with:

    "and a reference to the image data in the currently bound pixel unpack
    buffer or client memory."

    Replace the 1st sentence in the 13th paragraph with:

    "The image itself (referred to by /data/) is a sequence of groups
    of values."

    Replace the last paragraph with:

    "If the data argument of TexImage1D, TexImage2D, or TexImage3D
    is a null pointer (a zero-valued pointer in the C implementation)
    and the pixel unpack buffer object is zero, a one-, two-, or three-
    dimensional texture array is created with the specified target, level,
    internalformat, width, height, and depth border, but with unspecified
    image contents.  In this case no pixel values are access in client
    memory, and no pixel processing is performed.  Errors are generated,
    however, exactly as though the data pointer were valid.  Otherwise if
    the pixel unpack buffer object is non-zero, the data argument is
    treatedly normally to refer to the beginning of the pixel unpack
    buffer object's data."

 -- Section 3.8.3 "Compressed Texture Images", page 163.

    Replace the 3rd sentence of the 2nd paragraph with:

    "/data/ refers to compressed image data stored in the compressed
    image format corresponding to internalformat.  If a pixel
    unpack buffer is bound (as indicated by a non-zero value of
    PIXEL_UNPACK_BUFFER_BINDING_ARB), /data/ is an offset into the
    pixel unpack buffer and the compressed data is read from the buffer
    relative to this offset; otherwise, /data/ is a pointer to a block
    client memory and the compressed data is read from the client memory
    relative to the pointer."

    Replace the 2nd sentence in the 3rd paragraph with:

    "Compressed texture images are treated as an array of /imageSize/
    ubytes relative to /data/.  If a pixel unpack buffer object is bound
    and data+imageSize is greater than the size of the pixel buffer,
    INVALID_OPERATION results."

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

 -- Section 4.3.2 "Reading Pixels", page 219.

    Replace 1st sentence of the 1st paragraph with:

    "The method for reading pixels from the framebuffer and placing them in
    pixel pack buffer or client memory is diagrammed in figure 4.2."

    Add this paragraph after the 1st paragraph:

    "Initially, zero is bound for the PIXEL_PACK_BUFFER_ARB, indicating
    that image read and query commands such as ReadPixels return
    pixels results into client memory pointer parameters.  However, if
    a non-zero buffer object is bound as the current pixel pack buffer,
    then the pointer parameter is treated as an offset into the designated
    buffer object."

    Rename "Placement in Client Memory" to "Placement in Pixel Pack
    Buffer or Client Memory".

    Insert this paragraph after the newly renamed "Placement in Pixel
    Pack Buffer or Client Memory" heading:

    "If a pixel pack buffer is bound (as indicated by a non-zero value
    of PIXEL_PACK_BUFFER_BINDING_ARB), /data/ is an offset into the
    pixel pack buffer and the pixels are packed into the
    buffer relative to this offset; otherwise, /data/ is a pointer to a
    block client memory and the pixels are packed into the client memory
    relative to the pointer.  If a pixel pack buffer object is bound and
    packing the pixel data according to the pixel pack storage state
    would access memory beyond the size of the pixel pack buffer's
    memory size, INVALID_OPERATION results.  If a pixel pack buffer object
    is bound and /data/ is not evenly divisible into the number of basic
    machine units needed to store in memory the corresponding GL data type
    from table 3.5 for the /type/ parameter, INVALID_OPERATION results."


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

    None


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

 -- Section 6.1.3 "Enumerated Queries".

    After the sentence in the last paragraph describing GetPixelMap, add:

    "The GetPixelMapfv, GetPixelMapuiv, and GetPixelMapusv commands
    write all the values in the named pixel map to /data/.  If a
    pixel pack buffer is bound (as indicated by a non-zero value of
    PIXEL_PACK_BUFFER_BINDING_ARB), /data/ is an offset into the pixel
    pack buffer; otherwise, /data/ is a pointer to a block client memory.
    All pixel storage and pixel transfer modes are ignored when returning a
    pixel map.  n machine units are written where n is the size of the
    pixel map times the size of FLOAT, UNSIGNED_INT, or UNSIGNED_SHORT
    respectively in basic machine units.  If a pixel pack buffer object
    is bound and data+n is greater than the size of the pixel buffer,
    generate INVALID_OPERATION."

 -- Section 6.1.4 "Texture Queries".

    Remove the mention of img in the last phrase in the last sentence
    of the 1st paragraph so the sentence reads:

    "lod is a level-of-detail number, format is a pixel format from
    table 3.6, and type is a pixel type from table 3.5."

    Replace the 3rd sentence of the 2nd paragraph with:

    "These groups are then packed and placed in client or pixel buffer
    object memory.  If a pixel pack buffer is bound (as indicated by a
    non-zero value of PIXEL_PACK_BUFFER_BINDING_ARB), /img/ is an offset
    into the pixel pack buffer; otherwise, /img/ is a pointer to a block
    client memory."

    Add to the end of the 4th paragraph:

    "If a pixel pack buffer object is bound and packing the texture
    image into the buffer's memory would exceed the size of the buffer,
    generate INVALID_OPERATION."

    Replace the 2nd sentence of the 5th paragraph with:

    "When called, GetCompressedTexImage writes n ubytes of compressed
    image data to the pixel pack buffer or client memory pointed to by
    ptr, where n is the texture image's TEXTURE_COMPRESSED_IMAGE_SIZE
    value.

    Add to the end of the 6th paragraph:

    "If a pixel pack buffer object is bound and ptr+n is greater than
    the size of the buffer, generate INVALID_OPERATION."

 -- Section 6.1.5 "Stipple Query".

    "The pattern is packed into client or pixel pack buffer memory
    according to the procedures given in section 4.3.2 for ReadPixels;
    ..."

 -- Section 6.1.7 "Color Table Query".

    "The one-dimensional color table image is returned to client or
    pixel pack buffer memory starting at table."

 -- Section 6.1.8 "Convolution Query".

    "The one-dimensional or two-dimensional image is returned to client
    or pixel pack buffer memory starting at image."

    "The row and column images are returned to client or pixel pack
    buffer memory starting at row and column respectively."

 -- Section 6.1.9 "Histogram Query".

    "The one-dimensional histogram table image is returned to client or
    pixel pack buffer memory starting at values."

 -- Section 6.1.10 "Minmax Query".

    "A one-dimensional image of width 2 is returned to client or pixel
    pack buffer memory starting at values."

 -- Section 6.1.13 "Buffer Object Queries".

    Change the 2nd sentence of the 2nd paragraph to read:

    "target is ARRAY_BUFFER, ELEMENT_ARRAY_BUFFER, PIXEL_PACK_BUFFER_ARB,
    or PIXEL_UNPACK_BUFFER_ARB."

    Change the last phrase in the 1st sentence of the 4th paragraph to:

    "with target set to ARRAY_BUFFER, ELMENT_ARRAY_BUFFER,
    PIXEL_PACK_BUFFER_ARB, or PIXEL_UNPACK_BUFFER_ARB and pname set
    to BUFFER_MAP_POINTER."


GLX Protocol

    Pixel buffers, like texture objects and display lists, are server-side
    state.

    Prior to pixel buffer objects, pixel storage state for image packing
    and unpacking was considered client-side state. However, pixel
    buffers create the new situation where the server performs packing
    and unpacking into server-side pixel buffers.

    Fortunately, the GLX protocol, through foresight or oversight,
    has GLX protocol and non-rendering command opcodes (109 and 110)
    assigned for glPixelStoref and glPixelStorei respectively.

    It is better to use the existing protocol to send glPixelStorei and
    glPixelStoref GLX commands. This solves the problem of server-side
    pixel state the same way for both pack and unpack state. It may also
    allow implementations to minimize validation overhead for pixel
    commands because the pixel store modes are stateful rather than
    being parameters sent with every pixel command.

    When bound to a pixel unpack buffer object, the pixel payload for a 
    pixel command with no reply (for exmaple, glTexImage2D) is not sent.
    Instead, the pointer itself is sent as a 64-bit offset.
    Pixel storage modes are also not included in non-reply pixel commands.

    Therefore, the following new protocol for pixel commands that use the
    currently bound pixel unpack buffer object is specified:

        BitmapPBO
        ColorSubTablePBO
        ColorTablePBO
        CompressedTexImage1DPBO
        CompressedTexImage2DPBO
        CompressedTexImage3DPBO
        CompressedTexSubImage1DPBO
        CompressedTexSubImage2DPBO
        CompressedTexSubImage3DPBO
        ConvolutionFilter1DPBO
        ConvolutionFilter2DPBO
        DrawPixelsPBO
        PixelMapfvPBO
        PixelMapuivPBO
        PixelMapusvPBO
        PolygonStipplePBO
        SeparableFilter2DPBO
        TexImage1DPBO
        TexImage2DPBO
        TexImage3DPBO**
        TexSubImage1DPBO
        TexSubImage2DPBO
        TexSubImage3DPBO

    When bound to a pixel pack buffer object, the reply for pixel commands
    that return pixel data (for example, glReadPixels) is not required
    since the pixel data is actually transferred to the server-side pixel
    pack buffer object.  Indeed, forcing an unnecessary reply would hinder
    the performance advantages of using pixel buffer objects

    Therefore, protocol for "no reply" version of the following commands
    is specified:

        GetColorTablePBO
        GetCompressedTexImagePBO
        GetConvolutionFilterPBO
        GetHistogramPBO
        GetMinmaxPBO
        GetPixelMapfvPBO
        GetPixelMapuivPBO
        GetPixelMapusvPBO
        GetPolygonStipplePBO
        GetSeparableFilterPBO
        GetTexImagePBO
        ReadPixelsPBO

    If a "no reply" command is sent when the current pixel pack
    buffer object binding is zero, a GLXBadContextState error should
    be generated by the server.


    The following rendering commands are sent to the server as part of a
    glXRender request:

       BitmapPBO

            2           36              rendering command length
            2           311             rendering command opcode
            8           CARD64          bitmap
            4           CARD32          width
            4           CARD32          height
            4           FLOAT32         xorig
            4           FLOAT32         yorig
            4           FLOAT32         xmove
            4           FLOAT32         ymove

        ColorSubTablePBO

            2           32              rendering command length
            2           312             rendering command opcode
            8           CARD64          table
            4           ENUM            target
            4           CARD32          start
            4           CARD32          count
            4           ENUM            apiFormat  
            4           ENUM            apiType

        ColorTablePBO

            2           32              rendering command length
            2           313             rendering command opcode
            8           CARD64          table
            4           ENUM            target
            4           ENUM            internalformat
            4           CARD32          width
            4           ENUM            apiFormat
            4           ENUM            apiType

        CompressedTexImage1DPBO

            2           36              rendering command length
            2           314             rendering command opcode
            8           CARD64          data
            4           ENUM            target
            4           CARD32          level
            4           ENUM            internalFormat
            4           CARD32          width
            4           CARD32          border
            4           CARD32          imageSize

       CompressedTexImage2DPBO

            2           40              rendering command length
            2           315             rendering command opcode
            8           CARD64          data
            4           ENUM            target
            4           CARD32          level
            4           ENUM            internalFormat
            4           CARD32          width
            4           CARD32          height
            4           CARD32          border
            4           CARD32          imageSize

        CompressedTexImage3DPBO

            2           44              rendering command length
            2           316             rendering command opcode
            8           CARD64          data
            4           ENUM            target
            4           CARD32          level
            4           ENUM            internalFormat
            4           CARD32          width
            4           CARD32          height
            4           CARD32          depth
            4           CARD32          border
            4           CARD32          imageSize

        CompressedTexSubImage1DPBO

            2           36              rendering command length
            2           317             rendering command opcode
            8           CARD64          data
            4           ENUM            target
            4           CARD32          level
            4           CARD32          xoffset
            4           CARD32          width
            4           ENUM            apiFormat
            4           CARD32          imageSize

        CompressedTexSubImage2DPBO

            2           44              rendering command length
            2           318             rendering command opcode
            8           CARD64          data
            4           ENUM            target
            4           CARD32          level
            4           CARD32          xoffset
            4           CARD32          yoffset
            4           CARD32          width
            4           CARD32          height
            4           ENUM            apiFormat
            4           CARD32          imageSize

        CompressedTexSubImage3DPBO

            2           52              rendering command length
            2           319             rendering command opcode
            8           CARD64          data
            4           ENUM            target
            4           CARD32          level
            4           CARD32          xoffset
            4           CARD32          yoffset
            4           CARD32          zoffset
            4           CARD32          width
            4           CARD32          height
            4           CARD32          depth
            4           ENUM            apiFormat
            4           CARD32          imageSize

        ConvolutionFilter1DPBO

            2           32              rendering command length
            2           320             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           ENUM            internalformat
            4           CARD32          width
            4           ENUM            apiFormat
            4           ENUM            apiType

        ConvolutionFilter2DPBO

            2           36              rendering command length
            2           321             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           ENUM            internalformat
            4           CARD32          width
            4           CARD32          height
            4           ENUM            apiFormat
            4           ENUM            apiType

        DrawPixelsPBO

            2           28              rendering command length
            2           322             rendering command opcode
            8           CARD64          image
            4           CARD32          width
            4           CARD32          height
            4           ENUM            apiFormat
            4           ENUM            apiType
   
        PixelMapfvPBO

            2           20              rendering command length
            2           323             rendering command opcode
            8           CARD64          values
            4           ENUM            map
            4           CARD32          mapsize

        PixelMapuivPBO

            2           20              rendering command length
            2           324             rendering command opcode
            8           CARD64          values
            4           ENUM            map
            4           CARD32          mapsize

        PixelMapusvPBO

            2           20              rendering command length
            2           325             rendering command opcode
            8           CARD64          values
            4           ENUM            map
            4           CARD32          mapsize
   
        PolygonStipplePBO

            2           12              rendering command length
            2           326             rendering command opcode
            8           CARD64          mask

        SeparableFilter2DPBO

            2           44              rendering command length
            2           327             rendering command opcode
            8           CARD64          row
            8           CARD64          column
            4           ENUM            target
            4           ENUM            internalFormat
            4           CARD32          width
            4           CARD32          height
            4           ENUM            apiFormat
            4           ENUM            apiType

        TexImage1DPBO

            2           40              rendering command length
            2           328             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           CARD32          level
            4           CARD32          components
            4           CARD32          width
            4           CARD32          border
            4           ENUM            apiFormat
            4           ENUM            apiType

        TexImage2DPBO

            2           44              rendering command length
            2           329             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           CARD32          level
            4           CARD32          components
            4           CARD32          width
            4           CARD32          height
            4           CARD32          border
            4           ENUM            apiFormat
            4           ENUM            apiType

        TexImage3DPBO

            2           48              rendering command length
            2           330             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           CARD32          level
            4           CARD32          components
            4           CARD32          width
            4           CARD32          height
            4           CARD32          depth
            4           CARD32          border
            4           ENUM            apiFormat
            4           ENUM            apiType

        TexSubImage1DPBO

            2           36              rendering command length
            2           331             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           CARD32          level
            4           CARD32          xoffset
            4           CARD32          width
            4           ENUM            apiFormat
            4           ENUM            apiType

        TexSubImage2DPBO

            2           44              rendering command length
            2           332             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           CARD32          level
            4           CARD32          xoffset
            4           CARD32          yoffset
            4           CARD32          width
            4           CARD32          height
            4           ENUM            apiFormat
            4           ENUM            apiType

        TexSubImage3DPBO

            2           52              rendering command length
            2           333             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           CARD32          level
            4           CARD32          xoffset
            4           CARD32          yoffset
            4           CARD32          zoffset
            4           CARD32          width
            4           CARD32          height
            4           CARD32          depth
            4           ENUM            apiFormat
            4           ENUM            apiType

        GetColorTablePBO

            2           24              rendering command length
            2           334             rendering command opcode
            8           CARD64          table
            4           ENUM            target
            4           ENUM            apiFormat
            4           ENUM            apiType
 
        GetCompressedTexImagePBO

            2           20              rendering command length
            2           335             rendering command opcode
            8           CARD64          img
            4           ENUM            target
            4           CARD32          lod

        GetConvolutionFilterPBO

            2           24              rendering command length
            2           336             rendering command opcode
            8           CARD64          image
            4           ENUM            target
            4           ENUM            apiFormat
            4           ENUM            apiType

        GetHistogramPBO

            2           28              rendering command length
            2           337             rendering command opcode
            8           CARD64          values
            4           ENUM            target
            4           ENUM            apiFormat
            4           ENUM            apiType
            1           BOOL            reset
            3                           unused

        GetMinmaxPBO

            2           28              rendering command length
            2           338             rendering command opcode
            8           CARD64          values
            4           ENUM            target
            4           ENUM            apiFormat
            4           ENUM            apiType
            1           BOOL            reset
            3                           unused

        GetPixelMapfvPBO

            2           16              rendering command length
            2           339             rendering command opcode
            8           CARD64          values
            4           ENUM            map

        GetPixelMapuivPBO

            2           16              rendering command length
            2           340             rendering command opcode
            8           CARD64          values
            4           ENUM            map

        GetPixelMapusvPBO

            2           16              rendering command length
            2           341             rendering command opcode
            8           CARD64          values
            4           ENUM            map

        GetPolygonStipplePBO

            2           12              rendering command length
            2           342             rendering command opcode
            8           CARD64          mask

        GetSeparableFilterPBO

            2           40              rendering command length
            2           343             rendering command opcode
            8           CARD64          row
            8           CARD64          column
            8           CARD64          span
            4           ENUM            target
            4           ENUM            apiFormat
            4           ENUM            apiType

        GetTexImagePBO

            2           28              rendering command length
            2           344             rendering command opcode
            8           CARD64          texels
            4           ENUM            target
            4           CARD32          level
            4           ENUM            apiFormat
            4           ENUM            apiType

        ReadPixelsPBO

            2           36              rendering command length
            2           345             rendering command opcode
            8           CARD64          pixels
            4           CARD32          x
            4           CARD32          y
            4           CARD32          width
            4           CARD32          height
            4           ENUM            apiFormat
            4           ENUM            apiType


Errors

    INVALID_ENUM is generated if the <target> parameter of
    BindBuffer, BufferData, BufferSubData, MapBuffer, UnmapBuffer,
    GetBufferSubData, GetBufferParameteriv, or GetBufferPointerv is not
    one of ARRAY_BUFFER, ELEMENT_ARRAY_BUFFER, PIXEL_PACK_BUFFER_ARB,
    or PIXEL_UNPACK_BUFFER_ARB.

    INVALID_OPERATION is generated if Bitmap, ColorSubTable, ColorTable,
    CompressedTexImage1D, CompressedTexImage2D, CompressedTexImage3D,
    CompressedTexSubImage1D, CompressedTexSubImage2D,
    CompressedTexSubImage3D, ConvolutionFilter1D, ConvolutionFilter2D,
    DrawPixels, PixelMapfv, PixelMapuiv, PixelMapusv, PolygonStipple,
    SeparableFilter2D, TexImage1D, TexImage2D, TexImage3D, TexSubImage1D,
    TexSubImage2D, or TexSubImage3D would unpack (read) data from the
    currently bound PIXEL_UNPACK_BUFFER_ARB buffer object such that
    the memory reads required for the command would exceed the memory
    (data store) size of the buffer object.

    INVALID_OPERATION is generated if GetColorTable,
    GetCompressedTexImage, GetConvolutionFilter, GetHistogram, GetMinmax,
    GetPixelMapfv, GetPixelMapuiv, GetPixelMapusv, GetPolygonStipple,
    GetSeparableFilter, GetTexImage, or ReadPixels would pack (write) data
    to the currently bound PIXEL_PACK_BUFFER_ARB buffer object such that
    the memory writes required for the command would exceed the memory
    (data store) size of the buffer object.

    INVALID_OPERATION is generated by GetColorTable, GetConvolutionFilter,
    GetHistogram, GetMinmax, GetSeparableFilter, GetTexImage and ReadPixels
    if the current PIXEL_PACK_BUFFER_BINDING_ARB value is non-zero and the
    table/image/values/span/img/data parameter is not evenly divisible
    into the number of basic machine units needed to store in memory a
    datum indicated by the type parameter.

    INVALID_OPERATION is generated by ColorTable, ColorSubTable,
    ConvolutionFilter2D, ConvolutionFilter1D, SeparableFilter2D,
    TexImage1D, TexImage2D, TexImage3D, TexSubImage1D,
    TexSubImage2D, TexSubImage3D, and DrawPixels if the current
    PIXEL_UNPACK_BUFFER_BINDING_ARB value is non-zero and the data
    parameter is not evenly divisible into the number of basic machine
    units needed to store in memory a datum indicated by the type
    parameter.

    INVALID_OPERATION is generated by GetPixelMapfv if the current
    PIXEL_PACK_BUFFER_BINDING_ARB value is non-zero and the data parameter
    is not evenly divisible into the number of basic machine units needed
    to store in memory a float datum.

    INVALID_OPERATION is generated by GetPixelMapuiv if the current
    PIXEL_PACK_BUFFER_BINDING_ARB value is non-zero and the data parameter
    is not evenly divisible into the number of basic machine units needed
    to store in memory a uint datum.

    INVALID_OPERATION is generated by GetPixelMapusv if the current
    PIXEL_PACK_BUFFER_BINDING_ARB value is non-zero and the data parameter
    is not evenly divisible into the number of basic machine units needed
    to store in memory a ushort datum.

    INVALID_OPERATION is generated by PixelMapfv if the current
    PIXEL_UNPACK_BUFFER_BINDING_ARB value is non-zero and the data
    parameter is not evenly divisible into the number of basic machine
    units needed to store in memory a float datum.

    INVALID_OPERATION is generated by PixelMapuiv if the current
    PIXEL_UNPACK_BUFFER_BINDING_ARB value is non-zero and the data
    parameter is not evenly divisible into the number of basic machine
    units needed to store in memory a uint datum.

    INVALID_OPERATION is generated by PixelMapusv if the current
    PIXEL_UNPACK_BUFFER_BINDING_ARB value is non-zero and the data
    parameter is not evenly divisible into the number of basic machine
    units needed to store in memory a ushort datum.


Dependencies on EXT_pixel_buffer_object

    When this extension is supported, the EXT_pixel_buffer_object
    functionality adopts the tighter alignment and buffer bounds overflow
    error generation behavior of ARB_pixel_buffer_object (previously,
    EXT_pixel_buffer_object was not explicit about what happened in
    these situations).  This is because the two extensions share the
    same enumerants.

Dependencies on NV_pixel_data_range

    A non-zero pixel pack buffer binding takes priority over the
    READ_PIXEL_DATA_RANGE_NV  enable.

    A non-zero pixel unpack buffer binding takes priority over the
    WRITE_PIXEL_DATA_RANGE_NV enable.


New State

(table 6.20, Pixels, p. 235)

                                                         Initial
    Get Value                        Type   Get Command  Value    Sec     Attribute
    -------------------------------  ----   -----------  -------  ------  -----------
    PIXEL_PACK_BUFFER_BINDING_ARB    Z+     GetIntegerv  0        4.3.5   pixel-store
    PIXEL_UNPACK_BUFFER_BINDING_ARB  Z+     GetIntegerv  0        6.1.13  pixel-store


New Implementation Dependent State

    (none)


Usage Examples

    Convenient macro definition for specifying buffer offsets:

        #define BUFFER_OFFSET(i) ((char *)NULL + (i))

    Example 1: Render to vertex array:

        const int numberVertices = 100;

        // Create a buffer object for a number of vertices consisting of
        // 4 float values per vertex
        glGenBuffers(1, vertexBuffer);
        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, vertexBuffer);
        glBufferData(GL_PIXEL_PACK_BUFFER_ARB, numberVertices*4,
                     NULL, GL_DYNAMIC_DRAW);

        // Render vertex data into 100x1 strip of framebuffer using a
        // fragment program
        glBindProgram(FRAGMENT_PROGRAM_ARB, fragmentProgram);
        glDrawBuffer(GL_BACK);
        renderVertexData();
        glBindProgramARB(FRAGMENT_PROGRAM_ARB, 0);

        // Read the vertex data back from framebuffer
        glReadBuffer(GL_BACK);
        glReadPixels(0, 0, numberVertices, 1, GL_BGRA, GL_FLOAT,
                     BUFFER_OFFSET(0));

        // Change the binding point of the buffer object to
        // the vertex array binding point
        glBindBuffer(GL_ARRAY_BUFFER, vertexBuffer);

        glEnableClientState(VERTEX_ARRAY);
        glVertexPointer(4, GL_FLOAT, 0, BUFFER_OFFSET(0));
        glDrawArrays(TRIANGLE_STRIP, 0, numberVertices);


    Example 2: Streaming textures

    Streaming textures using NV_pixel_data_range:

        const int texWidth = 256;
        const int texHeight = 256;
        const int texsize = texWidth * texHeight * 4;
        void *pdrMemory, *texData;

        pdrMemory = glAllocateMemoryNV(texsize, 0.0, 1.0, 1.0);

        glPixelDataRangeNV(GL_WRITE_PIXEL_DATA_RANGE_NV, texsize,
                           pdrMemory);

        glEnableClientState(GL_WRITE_PIXEL_DATA_RANGE_NV);

        // Define texture level (without an image)
        glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, texWidth, texHeight, 0,
                     GL_BGRA, GL_UNSIGNED_BYTE, NULL);
        // Setup texture environment
        ...

        texData = getNextImage();

        while (texData) {

            memcpy(pdrMemory, texData, texsize);

            glFlushPixelDataRangeNV(GL_WRITE_PIXEL_DATA_RANGE_NV);

            glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, texWidth, texHeight,
                            GL_BGRA, GL_UNSIGNED_BYTE, pdrMemory);

            // Draw textured geometry
            glBegin(GL_QUADS);
            ...
            glEnd();

            texData = getNextImage();
        }

        glDisableClientState(GL_WRITE_PIXEL_DATA_RANGE_NV);

        glFreeMemoryNV(pdrMemory);


    Streaming textures using pixel buffer objects:

        const int texWidth = 256;
        const int texHeight = 256;
        const int texsize = texWidth * texHeight * 4;
        void *pboMemory, *texData;

        // Define texture level zero (without an image); notice the
        // explicit bind to the zero pixel unpack buffer object so that
        // pass NULL for the image data leaves the texture image
        // unspecified.
        glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
        glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, texWidth, texHeight, 0,
                     GL_BGRA, GL_UNSIGNED_BYTE, NULL);

        // Create and bind texture image buffer object
        glGenBuffers(1, &texBuffer);
        glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, texBuffer);

        // Setup texture environment
        ...

        texData = getNextImage();

        while (texData) {

            // Reset the contents of the texSize-sized buffer object
            glBufferData(GL_PIXEL_UNPACK_BUFFER_ARB, texSize, NULL,
                         GL_STREAM_DRAW);

            // Map the texture image buffer (the contents of which
            // are undefined due to the previous glBufferData)
            pboMemory = glMapBuffer(GL_PIXEL_UNPACK_BUFFER_ARB,
                                    GL_WRITE_ONLY);

            // Modify (sub-)buffer data
            memcpy(pboMemory, texData, texsize);

            // Unmap the texture image buffer
            glUnmapBuffer(GL_PIXEL_UNPACK_BUFFER_ARB);

            // Update (sub-)teximage from texture image buffer
            glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, texWidth, texHeight,
                            GL_BGRA, GL_UNSIGNED_BYTE, BUFFER_OFFSET(0));

            // Draw textured geometry
            glBegin(GL_QUADS);
            ...
            glEnd();

            texData = getNextImage();
        }

        glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);


    Example 3: Asynchronous glReadPixels

    Traditional glReadPixels:

        const int imagewidth = 640;
        const int imageheight = 480;
        GLubyte readBuffer[imagewidth*imageheight*4];

        // Render to framebuffer
        glDrawBuffer(GL_BACK);
        renderScene()

        // Read image from framebuffer
        glReadBuffer(GL_BACK);
        glReadPixels(0, 0, imagewidth, imageheight, GL_BGRA,
                     GL_UNSIGNED_BYTE, readBuffer);

        // Process image when glReadPixels returns after reading the
        // whole buffer
        processImage(readBuffer);


    Asynchronous glReadPixels:

        const int imagewidth = 640;
        const int imageheight = 480;
        const int imageSize = imagewidth*imageheight*4;

        glGenBuffers(2, imageBuffers);

        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[0]);
        glBufferData(GL_PIXEL_PACK_BUFFER_ARB, imageSize / 2, NULL,
                     GL_STREAM_READ);

        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[1]);
        glBufferData(GL_PIXEL_PACK_BUFFER_ARB, imageSize / 2, NULL,
                     GL_STREAM_READ);

        // Render to framebuffer
        glDrawBuffer(GL_BACK);
        renderScene();

        // Bind two different buffer objects and start the glReadPixels
        // asynchronously. Each call will return directly after
        // starting the DMA transfer.
        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[0]);
        glReadPixels(0, 0, imagewidth, imageheight/2, GL_BGRA,
                     GL_UNSIGNED_BYTE, BUFFER_OFFSET(0));

        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[1]);
        glReadPixels(0, imageheight/2, imagewidth, imageheight/2, GL_BGRA,
                     GL_UNSIGNED_BYTE, BUFFER_OFFSET(0));

        // Process partial images.  Mapping the buffer waits for
        // outstanding DMA transfers into the buffer to finish.
        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[0]);
        pboMemory1 = glMapBuffer(GL_PIXEL_PACK_BUFFER_ARB,
                                 GL_READ_ONLY);
        processImage(pboMemory1);
        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[1]);
        pboMemory2 = glMapBuffer(GL_PIXEL_PACK_BUFFER_ARB,
                                 GL_READ_ONLY);
        processImage(pboMemory2);

        // Unmap the image buffers
        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[0]);
        glUnmapBuffer(GL_PIXEL_PACK_BUFFER_ARB);
        glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, imageBuffers[1]);
        glUnmapBuffer(GL_PIXEL_PACK_BUFFER_ARB);


Appendix on Pack/Unpack Range

    The complexity of OpenGL's pixel pack/unpack state makes it difficult
    to express succinctly what range of a pixel buffer object will be
    accessed by a pixel command.

    The following code, following the conventions of the SGI OpenGL
    Sample Implementation, returns the limit (one byte more than the
    maximum allowed offset into the buffer object) for the memory a
    pixel command will read/write.

    /*
    ** Compute offset limit into user's data considering all pixel
    ** store modes.  This offset limit is ONE MORE than the largest byte
    ** offset for the image.
    */
    static GLsizeiptr OffsetLimitImage3D(__GLpixelStoreMode *pixelStoreMode,
                                         GLsizei width, GLsizei height,
                                         GLsizei depth,
                                         GLenum format, GLenum type,
                                         const void *userdata,
                                         GLint skip_images)
    {
        const GLint line_length = pixelStoreMode->lineLength;
        const GLint image_height = pixelStoreMode->imageHeight;
        const GLint alignment = pixelStoreMode->alignment;
        const GLint skip_pixels = pixelStoreMode->skipPixels;
        const GLint skip_lines = pixelStoreMode->skipLines;

        GLsizeiptr offsetLimit = (GLsizeiptr) userdata;

        GLint rowsize;
        GLint padding;
        GLint imagesize;

        assert(width > 0);
        assert(height > 0);
        assert(depth > 0);

        assert(line_length >= 0);
        assert(image_height >= 0);

        assert(skip_pixels >= 0);
        assert(skip_lines >= 0);
        assert(skip_images >= 0);

        assert((alignment == 1) ||
               (alignment == 2) ||
               (alignment == 4) ||
               (alignment == 8));

        /* All formats except GL_BITMAP fall out trivially */
        if (type == GL_BITMAP) {
            const GLint groups_per_line = (line_length > 0) ?
                                          line_length : width;
            const GLint rows_per_image = (image_height > 0) ?
                                         image_height : height;

            assert(1 == __glElementsPerGroup(format, type));

            rowsize = (groups_per_line + 7) / 8;
            padding = rowsize & (alignment-1);
            if (padding) {
                rowsize += alignment - padding;
            }
            imagesize = rows_per_image * rowsize;

            offsetLimit += imagesize    * (skip_images + depth-1);
            offsetLimit += rowsize      * (skip_lines  + height-1);
            offsetLimit += (skip_pixels + width+7)/8;
        } else {
            const GLint components = __glElementsPerGroup(format, type);
            const GLint element_size = __glBytesPerElement(type);
            const GLint group_size = element_size * components;

            if (0 == (line_length | image_height | skip_pixels |
                      skip_lines | skip_pixels)) {
                // Fast path: when above pixel store modes are all zero.
                rowsize = width * group_size;
                // Default alignment is 4 so allow arbitrary alignment
                // on fast path.
                padding = rowsize & (alignment-1);
                if (padding) {
                    rowsize += alignment - padding;
                }
                imagesize = depth * height * rowsize;
                offsetLimit += imagesize;
            } else {
                // General path: when one or more non-zero pixel store modes.
                const GLint groups_per_line = (line_length > 0) ?
                                              line_length : width;
                const GLint rows_per_image = (image_height > 0) ?
                                             image_height : height;

                rowsize = groups_per_line * group_size;
                padding = rowsize & (alignment-1);
                if (padding) {
                    rowsize += alignment - padding;
                }
                imagesize = rows_per_image * rowsize;

                offsetLimit += imagesize    * (skip_images  + depth-1);
                offsetLimit += rowsize      * (skip_lines   + height-1);
                offsetLimit += group_size   * (skip_pixels  + width);
            }
        }
        return offsetLimit;
    }

    GLsizeiptr __glOffsetLimitImage3D(__GLpixelStoreMode *pixelStoreMode,
                                      GLsizei width, GLsizei height,
                                      GLsizei depth,
                                      GLenum format, GLenum type,
                                      const void *userdata)
    {
        return OffsetLimitImage3D(pixelStoreMode,
                                  width, height, depth, format, type,
                                  userdata,
                                  pixelStoreMode->skipImages);
    }

    GLsizeiptr __glOffsetLimitImage(__GLpixelStoreMode *pixelStoreMode,
                                    GLsizei width, GLsizei height,
                                    GLenum format, GLenum type,
                                    const void *userdata)
    {
        /* NOTE: Non-3D image max offset computations ignore (treat as zero)
           the unpackModes.skipImages state! */
        return OffsetLimitImage3D(pixelStoreMode,
                                  width, height, 1, format, type,
                                  userdata,
                                  0);  // Treat skipImages as zero.
    }


Revision History

    revision 0.3: mjk

        Numbered issues.

        Add issues 14 through 18.

        Remove all gl/GL prefix/suffixing in specification sections.  Use
        gl/GL prefix/suffixing in sections other than the specification
        sections. Leaving off prefixes in non-specification sections is
        ambiguous, particularly within example source code.

        Base specification language updates on OpenGL 2.0 specification.

        Add buffer object required state section.

        Added GL_INVALID_OPERATION when an offset accessed (read or
        written) for a pixel command from/to a pixel buffer object would
        exceed the size of the buffer object.

        Added GL_INVALID_OPERATION when for misaligned offsets.

        Added "Appendix on Pack/Unpack Range".

        Add GLX protocol discussion.

    revision 0.4: mjk

        Fixed grammar issues from Brian Paul.

        Improved example code and fixed grammar from Nick Carter.

        Explain how a NULL data parameter to glTexImage commands works.

    revision 0.5: mjk

        Clarify that glBufferData usage modes apply to drawing _and_
        image specification commands.

    revision 0.6: mjk

        Add "streaming draw pixels" to the list of interesting approaches
        for this extension in the Overview.

        Add issue discussing the relationship of this extension to data
        aquisition hardware.

    revision 0.7: mjk

        Assign enumerant values to match the EXT_pixel_buffer_object values.

        Add issue explaining why the ARB extension shares enums with
        EXT_pixel_buffer_object.

        Apply Dale's suggestion to improve the clarity of the usage
        pattern parameters to glBufferData.

    revision 0.8 mjk

        Typo fixes from Ian Romanick and Nick Carter.

    revision 1.0 mjk

        Add issue 23 for Jeremy about render-to-vertex-array.  Move
        render-to-vertex-array justification in overview to bottom of
        the list.

   revision 1.1 srahman

       Add GLX protocol support. 

   revision 1.2 Jon Leech

       Change GLvoid -> void (Bug 10412).