NV_memory_object_sparse

GL_NV_memory_object_sparse

Carsten Rohde, NVIDIA
James Jones, NVIDIA

Carsten Rohde, NVIDIA Corporation (crohde 'at' nvidia.com)

Complete

Last Modified Date: August 14, 2020
Revision:           2

550
OpenGL ES Extension #329

Written against the OpenGL 4.6 and OpenGL ES 3.2 specifications
including ARB_sparse_texture and ARB_sparse_buffer.

GL_NV_memory_object_sparse requires GL_EXT_memory_object,
ARB_sparse_texture and/or ARB_sparse_buffer or a version of
OpenGL or OpenGL ES that incorporates it.

NV_memory_object_sparse interacts with ARB_direct_state_access (OpenGL)
when OpenGL < 4.6 is used.

ARB_sparse_texture (OpenGL) interacts with GL_EXT_memory_object_sparse.
ARB_sparse_buffer (OpenGL) interacts with GL_EXT_memory_object_sparse.
EXT_sparse_texture (OpenGL ES) interacts with GL_EXT_memory_object_sparse.

This extension adds sparse support to EXT_memory_object extension.

void BufferPageCommitmentMemNV(enum target,
                               intptr offset, sizeiptr size,
                               uint memory, uint64 memOffset,
                               boolean commit);

void NamedBufferPageCommitmentMemNV(uint buffer,
                                    intptr offset, sizeiptr size,
                                    uint memory, uint64 memOffset,
                                    boolean commit);

void TexPageCommitmentMemNV(enum target,
                            int layer, int level,
                            int xoffset, int yoffset, int zoffset,
                            sizei width, sizei height, sizei depth,
                            uint memory, uint64 offset,
                            boolean commit);

void TexturePageCommitmentMemNV(uint texture,
                                int layer, int level,
                                int xoffset, int yoffset, int zoffset,
                                sizei width, sizei height, sizei depth,
                                uint memory, uint64 offset,
                                boolean commit);

None

In Section 6.2, "Creating and Modifying Buffer Object Data Stores", add
the following add the following to the end of the description of
BufferSubData:

If NV_memory_object_sparse is supported, additionally, the commands

    void BufferPageCommitmentMemNV(enum target,
                                   intptr offset, sizeiptr size,
                                   uint memory, uint64 memOffset,
                                   boolean commit);

    void NamedBufferPageCommitmentMemNV(uint buffer,
                                        intptr offset,
                                        sizeiptr size,
                                        uint memory, uint64 memOffset,
                                        boolean commit);

behaves similarly to BufferPageCommitmentARB and
NamedBufferPageCommitmentMemARB except that the pages of the sparse buffer
are bound to the memory specified by <memory> and <memOffset>.

Errors (additionally to non-Mem variants)

  An INVALID_OPERATION error is generated if <memory> is not the name of
  an existing memory object.

  An INVALID_OPERATION error is generated if <offset> + <size> exceeds the
  size of the memory object.

Add the following to end of subsection 8.20.2. "Controlling Sparse Texture
Commitment":

If NV_memory_object_sparse is supported, additionally, the commands

    void TexPageCommitmentMemNV(enum target,
                                int layer, int level,
                                int xoffset, int yoffset, int zoffset,
                                sizei width, sizei height, sizei depth,
                                uint memory, uint64 offset,
                                boolean commit);

    void TexturePageCommitmentMemNV(uint texture,
                                    int layer, int level,
                                    int xoffset, int yoffset, int zoffset,
                                    sizei width, sizei height, sizei depth,
                                    uint memory, uint64 offset,
                                    boolean commit);

behaves similarly to TexPageCommitmentMemARB and
TexturePageCommitmentEXT except that the tiles of the sparse texture
are bound to the memory specified by <memory> and <offset>.

<layer> indicates the layer of a texture array or cube texture,
<zoffset> must be 0 and <depth> must 1 in this case. For other textures
<layer> must be 0.

Errors (additionally to non-Mem variants)

  An INVALID_OPERATION error is generated if <memory> is not the name of
  an existing memory object.

  An INVALID_OPERATION error is generated if <memory> is dedicated or
  imported from a non-opaque handle.

  An INVALID_OPERATION error is generated if <offset> plus the number of
  bytes required for the tiles to commit exceeds the size of the memory
  object.

  An INVALID_VALUE error is generated if <layer> is not 0 and the texture
  neither a texture array or a cube texture.

  An INVALID_VALUE error is generated if <zoffset> is not 0 or <depth> is
  not 1 if the texture is a texture array or a cube texture.

  An INVALID_VALUE error is generated if <layer> is greater or equal than
  the number of layers of the texture array.

  An INVALID_VALUE error is generated if <layer> is greater or equal than
  6 in case of a cube texture.

If EXT_direct_state_access is not supported, remove references to the
NamedBufferPageCommitmentMemNV and TexturePageCommitmentMemNV commands
added by this extension.

(1) Should we a 'aspect' parameter to the new gl.*CommitMemNV() functions?

RESOLVED: No. This can be deferred to a future EXT extension because there
          is currently no multi-planar texture support in GL and metadata
          isn't required for NVIDIA hardware.

Revision 2, 2020-08-14 (Piers Daniell)
    - Fix duplicate parameter names in BufferPageCommitmentMemNV and
      NamedBufferPageCommitmentMemNV.

Revision 1, 2020-08-04 (Carsten Rohde)
    - Initial draft.