Disclaimer

People reading the OpenGL ARB minutes are cautioned that statements made by attendees do not represent official company positions unless explicitly identified as such.

Action Items / Conclusions from the Meeting

• Reviewed latest Superbuffers specification from the working group and roadmap to final specification and vote.
• Reviewed new draft bylaws and membership agreements and proposed changes.
• Agreed to make minor editorial changes to OpenGL 1.5 and HLSL specifications. Adopted new procedural guidelines to speed adoption of future non-substantive, editorial changes.
• Discussed ARB_fragment_program_shadow extension proposal and agreed to call email vote after offline review.
• Will setup mailing lists for new technical infrastructure and marketing working groups and define specific goals.
• Discussed ARB_extensions_query extension proposal and will construct poll/discussion topic for opengl.org to get developer feedback.
• Reviewed OpenGL next release plans, potential features, and timeline for deciding on features to include. Marketing WG will discuss promotion of the HLSL to the core.
• Discussed behavior of vertex buffer objects shared in multiple contexts. Nick will suggest minor spec language changes to clarify matters.
• The March 2-3 meeting will be hosted by Apple in Cupertino, CA; the June meeting will probably be hosted by Intel in Sacramento.

Summary of Discussion Topics

• Superbuffers Review
• Bylaws WG Updates
• Specification Edits/Changes
• ARB_fragment_program_shadow proposal
• New Working Groups
• ARB_extensions_query proposal
• OpenGL Next Release Plans
• VBOs and Multiple Contexts
• Next Meeting

Tuesday, December 9

Superbuffers Review

Getting very close to the final version. Need clarifications and conversion to classic specification form. Rob has a spec based off the version 0.26 whitepaper but it needs updates. Expect to be able to hold a vote before, or by the March ARB meeting.

Reviewed changes from 0.26 discussed at the September meeting (see 0.29 document).

Reviewed examples (9.7, 9.8)

Q & A - John R. thinks number of entry points could be reduced by OO-encapsulation. Rob tried to follow existing GL practice. Primarily written as a white paper, rather than a classic extension spec, for development and discussion purposes,

Discussion of motivation - could expand the two paragraphs in section 1.3, which Rob wrote more than a year ago. General idea: moving from static to dynamic binding of resources, making them useful as both inputs and outputs to the GL state machine, and being able to bind them at many different points. This makes many new algorithms possible, so app writers can capture more of the power of modern GPUs. Demand is building quickly.

John R. suggests "views" may be a better way of thinking about "sub memories", and avoid conceptual collisions with SubImage calls. Rob will think about incorporating this change. Could incorporate additional views such as subimages; Rob thinks if we were to do this, we should add a new call to create a view, as contrasted to accessing a view of mipmaps or depth slices which were (at least conceptually) created at the same time as the base object.

NVIDIA issues with superbuffers:

• Breaks OpenGL paradigm of operating on objects that are in the state vector. Alternate proposal only operates on bound objects. [what about FB?]. Suggests altering to operate only on texture targets.
• Delayed validation is complex. Specifying attachments up front, as in the alternate, simplifies things.
• Poor symmetry of the object hierarchy.
• Submemories are overly complex. Instead, specify substructure when attaching memories to a container. Eliminates driver implementation overhead.
• General concern with high complexity of the specification. Solving the top priorities (rendering to vertex arrays / textures efficiently) can be solved with a much more straightforward interface.

Bylaws WG Updates

Timeline: need to get documents out of the WG in mid-January to allow for a vote at the March ARB meeting. Will have meetings to discuss transition plans.

Detailed review of the Membership Agreement, Bylaws, and Licensing documents with many small changes and additions proposed. Refer to Dave's WG notes for details.

Wednesday, December 10

Specification Edits/Changes

OpenGL 1.5 Spec Changes:

Reviewed a bunch of proposed changes for 1.5 spec. Most were non-controversial. Some need further wordsmithing. Some need more research. Will circulate proposed wording and new draft soon.

Shading Language Spec Changes:

Reviewed changes discussed in the arb-gl2 working group. Most were editorial and will be circulated to the mailing list. One functional change to be reviewed and email vote held.

Procedural update:

In the future, for editorial (non-substantive) changes to specs, we will simply put them out on the participants' mailing list and allow several weeks for feedback; failing pushback, they will be deemed adopted by consensus. For substantive changes recommended by the workgroup, the ARB should hold a formal vote.

ARB_fragment_program_shadow proposal

Bill and Pat reviewed a proposal to allows new shadow texture targets for ARB_fragment_program, and discussed different options for supporting it; they chose new targets over different instructions. This is a simple extension of low enough complexity not to demand its own working group.

New Working Groups

Marketing and Technical (infrastructure) Working Groups haven't got much traction in the way of volunteers since September, so we'll try to drive them from the top.

ARB_extensions_query proposal

Simple IsExtensionSupported query like the GLU 1.3 feature - except that GLU 1.3 isn't supported by the Microsoft GLU implementation. Possible augmentations: way to enumerate extension names + not adding future extension strings to the static string.

Since the main problem being addressed is old apps which don't allocate enough buffer space to copy the extensions string into, not adding them to the static string may be inevitable.

Problem: how do you know this extension is supported without querying the EXTENSIONS string? On Windows could assume non-NULL return from wglGetProcAddress would suffice.

To get developer feedback, we will come up with a poll or discussion topic for opengl.org by discussion on the mailing list.

OpenGL Next Release Plans

Should we stay on the June spec update cycle? Depends on what new functionality is available. Not much point, aside from marketing, in staying on a yearly release cycle if there isn't compelling functionality added.

Possible classes of things to add for the next version:

1) Promoting existing extensions to the core. Candidates:

• ARB_extension_query
• ARB_vertex_program / ARB_fragment_program
• NV_half_float, perhaps splitting storage format away from the vertex specification API.
• Floating point rendering/buffers - ATI_pixel_format_float and comparable NVIDIA and Apple extensions).

2) Superbuffers. Doesn't exist as an ARB extension yet. Conceivable it could make it into the core but a March ARB extension makes that problematic for a June spec update.

3) High level shading language. Jon asked for current interest / support information from IHVs. NVIDIA knows of only one major ISV using it today, and they're in a fetal stage (not shipping). 3Dlabs has about half a dozen ISVs implementing or already running; not sure if any have released yet. Similar for ATI. Mostly functionality has been available in betas, but ATI/3Dlabs now have publically available drivers with support.

Straw poll of attendees for thoughts on promotion of shading language into the core. There was a roughly equal split between those supporting immediate inclusion in the next OpenGL release, and those wanting additional ISV feedback and deployment experience as ARB extensions before moving to this stage, with slightly more people advocating the latter position. Nick noted that better tools and interchange formats would be valuable, in addition to the base shading functionality.

Pat notes that promotion to the core gives us an opportunity for changes beyond minor spec updates. The arb-gl2 working group is still operating, but just dealing with clarifications and minor functionality changes at present - no major new topics on the table.

The marketing WG should engage on this decision ASAP.

VBOs and Multiple Contexts

Kent says they've implemented this and have issues about sharing of VBOs. Example: VBO bound to object 1 in context A and B, but not current in B. Now context A can delete the VBO successfully; what happens in B, where there's still a reference to the object ID?

Nick noted (as rephrased later by Matt Craighead in email) that the best behavior is for the ARRAY_BUFFER binding to contribute to the reference count. This isn't strictly necessary because this binding is never used directly, but it would be somewhat odd if ELEMENT_ARRAY_BUFFER did contribute to the reference count (it needs to) and ARRAY_BUFFER didn't. Rob noted that the bind call was poorly named.

Rob suggests clarifying this distinction in the spec. Nick will propose language changes for this and circulate to the participants' list.

Next Meeting Plans

Apple will host in Cupertino on March 2nd/3rd; Howard Miller is again the contact at Apple. Intel will probably host in Sacramento in June. Volunteers for September and December are needed.

Thanks to Sun for hosting this meeting!