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

• Agreed on minor 2.0 spec edits to fix various small spec bugs.
• Reviewed new bylaws and voting / transition plans to them.
• Initial discussion of 2.1 feature list.
• Float buffer and superbuffers working group status reviews.

Summary of Discussion Topics

• Working Group Process Discussion
• OpenGL 2.0 Specification Issues
• OpenGL 2.1 Planning
• Vertex Attribute Aliasing & GLX Protocol
• New Bylaws Status
• GLP Updates
• Metafile Working Group
• Floating Point Buffer Extensions
• Superbuffers
• EXT_texture_coordinate_unnormalized
• Next Meeting Plans

Tuesday, September 21

Working Group Process Discussion

Michael - process breaks down in presence of contentious issues. Get out a "majority of minority" result which is then rubber-stamped by the ARB. If there are non-minor issues to be resolved, better to bring multiple proposals to the table and let the ARB select one which the WG would then refine.

JohnR - shouldn't reverse WG votes without all the original voters being involved.

Jeremy - effective WGs have agreement on high-level principles, which guide the overall process.

Bill notes that common practice in other groups is that a motion to reconsider must be brought by the group that won the previous vote, not the group that lost.

Takeaway points after discussion (some of this may go into the updated WG operating guidelines document):

• When establishing WGs, we should be very clear about scope and expected results.
• Need periodic checkins with the ARB, probably more often than quarterly. Jon suggests monthly email status reports. General feeling more in favor of telecons.

  Action: will institute regular monthly full ARB teleconferences. First will be Friday, 11 AM, October 22.

• If a minority strongly objects to something in the WG, they should ask the WG chair to push the issue up to the ARB. The WG chair should not prevent this.
• If a minority feels strongly enough to split into an alternative WG, it should be as official as the original, with its own agenda and scope. The ARB ends up with a choice of two competing specs. Possible problems: more than two competing specs (say, a third party cherry-picking both specs), and someone who simply can't implement the functionality.
• WG votes should be arrived at in a fashion expected to be representative of the ARB as a whole.
• Would like more formal rules for WG operations - Dave Zenz is willing to lead this effort. JohnR would like a protocol covering voting to reconsider previous decisions made within a WG.
• Would like to see implementations prior to ARB votes.
• Pat notes there are two types of extensions, "little" (e.g. two-sided stencil, point sprite) and "big" (GLSL, superbuffers). The problems we're discussing have emerged with "big" extensions. Suggestion: "big" extensions will take more than 3 months. At the first quarterly ARB meeting following setting up a WG, review what the problem being solved is and the chosen strategy.
• Dave will have a preliminary WG operating principles document for the October teleconference.

OpenGL 2.0 Specification Issues

• Benj - CLAMP_TO_NEAREST and t>1 Suggest rewriting eqs. 3.23-25 to enumerate different clamping cases.
• JohnR - glPixelZoom. Paul Ramsey will post his document describing glDrawPixels in context of multisampling. JohnR will post proposed changes to the section (using the same language as the multisample version).
• Mark and Brian - misc. bugs posted by email previously.
• Dale - fog bug (not emailed out). Change FOG_COORDINATE -> FOG_COORD on p. 191. Change FOG_COORD_SRC -> FOG_COORD in glRasterPos/glWindowPos section (pp. 54-55)
• Bill - Insert paragraph preceding last line in next-to-last paragraph of section 4.4.2 (Accum) to make clear that INVALID_OPERATION is always generated by glAccum if there's no accum buffer, or if in color index mode.

Jon will work up a revised 2.0 spec addressing these issues prior to the October teleconference.

OpenGL 2.1 Planning

Candidate features for summer 2005 release

• Superbuffers - may go directly from EXT to core?
• Float buffer extensions.
• Pixel buffer objects - still working on spec. Adding bounds checks, which is more reasonable performancewise for PBOs than VBOs. Kent is concerned about specifying data in formats the hardware may not accelerate. Jeremy asked about interactions with (to be developed) render-to-vertex-array layering on superbuffers. Dale sees PBOs as a non-blocking way to move data into the HW for e.g. video capture.
• Precompiled shaders.
• Timing/async extensions from 3Dlabs
• HLSL extensions.
• Instancing geometry - DX has an explicit API with per-primitive and per-vertex attributes, allowing instancing primitives with different per-vertex attributes. Pat notes that setting attributes in immediate mode outside of glBegin/glEnd can accomplish this already.
• EXT_texture_coordinate_unnormalized - to be discussed tomorrow afternoon.

Jon wants to try and settle the candidate feature list at the December ARB meeting and move the spec writing process up several months relative to last year. This would not absolutely preclude later additions, it is simply an accelerated schedule to work towards.

Vertex Attribute Aliasing & GLX Protocol

Brian brought slides to explain problem w/2.0 core/ARB/NV feature interactions. It appears that NVIDIA has not yet implemented indirect rendering for NV_vertex_program, so it will be possible to break the GLX protocol aliasing between the NV and ARB glVertexAttrib* calls now, using the existing assigned opcodes for the ARB functions. Pat will allocate new NV opcodes from their reserved block.

GLX protocol WG needs to reconvene to sort out 2.0 protocol.

New Bylaws Status

Nearly complete, finally. Set date for the transition to the new bylaws to November 23, 2004.

Jon will circulate the updated bylaws vote document and call an email vote. Dependent on a few last minor updates from Doug Crisman.

After this vote passes (if it does), we will hold votes for dues for CY04 ($0) and CY05 ($10000).

GLP Updates

Paul Ramsey brought slides describing the GLP project, for OpenGL rendering without a window system. Sun is working with U. Texas on a TeraGrid project - internal implementation only at this point.

Paul will look at the Khronos EGL API as a possible existing alternative - Jon thinks Khronos will be receptive to minor changes to help support GLP-type functionality.

Metafile Working Group

GLSL alone doesn't define bindings to other attributes apps need. Metafiles include additional attributes/information about shaders and their parameters. Primary question: does this even belong in the ARB?

Sony has introduced "Collada" XML schema which has been picked up by some content providers. Also inclues a rudimentary scene graph. This could be extended for GLSL.

• Does it belong in the GL core?
• What are the application needs?
• Will Collada suffice, or do we need something new?

Agreed with no objections to create "arb-metafile" WG with Michael Gold as chair, initial charter: determine whether to use Collada or develop our own format, and report back. Mailing list is already setup, people can subscribe via majordomo as usual.

Wednesday, September 22

Floating Point Buffer Extensions

Dale distributed updated specs. Merged pixel_format_float into color_buffer_float, so there are both core GL and WGL/GLX name strings associated with the latter now. Still have separate half_float_pixel and texture_float extensions. Open issues remaining with each extension, but the WG is as close as it's going to get and resolution from the whole ARB is needed.

Discussion of SGI IP issue. Dale will include SGI statement regarding their patent into each extension spec.

Review of open issues. Will add an invariant to cover passing through clamped fixed-point colors intact (section 2.14.9, issue 32). Noted we need a new issue discussing clamping behavior with MRT where some targets are FP and some are fixed-point. Current render_target spec doesn't allow setting this up, but it may be possible in the future.

If FP texture formats are required to be respected (e.g. not being allowed to choose a fixed-point internal format when FP requested), there are consequences for making texture_float a core feature - implication is that either a cap bit is needed, or 2.1 implementations will be required to support FP textures, with implications relative to the IP issues.

texture_float spec allows implementation to choose internal FP format (e.g. could choose FP16 when FP32 requested), and allowed varying clamping behavior (supporting NaN/Inf or not) to accomodate differences between ATI/NVIDIA HW. Apps will need to check that they got the FP format they asked for if they care about these range/representation issues.

Specs are basically ready to be implemented for testing purposes. Will review status on the October teleconference and, probably, call an email vote at that point.

Superbuffers

Barthold brought slides describing the current WG status and plans, and updated the ARB on WG progress. Now developing a pbuffer replacement for render-to-texture and offscreen rendering - all in GL, no context switching required. ATI and NVIDIA are co-authoring the EXT_framebuffer_object spec, with 3Dlabs leading the WG. Still on schedule from last meeting: draft spec September 2004, final spec December 2004. Hopefully promote to ARB status in December.

Texture/render buffers attached to framebuffer objects, which are in turn bound to the current state. Initial spec supports color & depth textures, multiple attachments for MRT, stencil render buffers. Does not support accum / multisample buffers or format groups.

Reviewed state that can or might change as a result of changing the FB attachments (thereby effectively changing the pixel format). Will probably move some "implementation dependent state" to a new "framebuffer dependent state" table. Agreed to move subpixel bits and maximum line/point sizes from "undecided" back to "implementation dependent". Agreed that the supported set of extensions cannot change.

Long discussion about multi-context behavior when a shared framebuffer object is invalid. Should errors be generated by glBegin? Straw poll: YES, error should be generated. Straw poll: a new GL error code should be introduced for invalid framebuffers (WG will decide on a name). Note this affects e.g. gluErrorString, which will return NULL if it doesn't recognize an error code.

Agreed to put the framebuffer valid bit into the context state as a query, rather than a separate ValidateFramebuffer call - the separate call was a holdover from uber buffers, where the call actually did work as well as returning valid status.

EXT_texture_coordinate_unnormalized

This is an Intel extension proposal - offers both regular [0,1] and unnormalized (texel indexed) texcoord access to all texture types. Pat doesn't think it's worth extending normalized access to RECTANGLE targets. Limited interest from others - several are willing to review the extension but may not choose to implement it.

Next Meeting Plans

Apple will host the December meeting in Cupertino - dates TBD.

Thanks to 3Dlabs for hosting this meeting!