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 this Meeting

• Added disclaimer to minutes. Participants should review minutes for accuracy prior to posting on opengl.org.
• Bylaws working group making excellent progress on identifying everyone's issues. Next step is to make new language proposals to address issues.
• GL2 working groups has selected glslang as the starting point, and is actively identifying and reducing issues from both the Cg and glslang documents. Still on track for a final spec by end of CY 2002.
• Advanced vertex program working group will start up now that fragment program working group is winding down.
• NVIDIA described their OpenGL 1.4 conformance test proposals. A new working group will begin with these proposals and work towards final tests.
• ARB_fragment_program working group presented their final specification, which was voted in as an official ARB extension!
• Two new vertex array object extension proposals were presented. This has become time-critical, and a working group will be formed immediately to converge on an ARB VAO specification.
• An ARB issue tracking list will be created on the member website.
• GLX 1.4, GLX 1.4 protocol, and proposed GLX_ARB_render_texture extension specifications will be circulated to the participants list shortly
• NVIDIA and Sun were renewed as Auxiliary Members for another year.
• The Bylaws, GL2, and VAO working groups held face-to-face breakout sessions in the remainder of our time.

Summary of Discussion Topics

• Introductions
• Working Group Summaries
  • Bylaws Working Group
  • GL2 Working Group
  • Vertex Program Working Group
• OpenGL 1.4 Conformance Update
• ARB_fragment_program
• New Extension Proposals
• Miscellaneous Issues
• Membership Renewals

September 18, 2002

Introductions

Next ARB Meeting

NVIDIA has offered to host the December meeting in Silicon Valley. Dates to be determined on the mailing list; December 3/4 or 10/11 are likely, 17/18 less so but still possible.

Publication of the ARB Minutes

Some attendees had concerns resulting from last June's minutes and inaccuracy of the resulting press coverage. Reporters seemed to assume that miscellaneous individual comments in the minutes represented official corporate positions, which was often not the case. ARB attendees are concerned about this, since they are not legal or PR representatives of their companies.

Some people suggested that instead of releasing the raw minutes, we just summarize conclusions / action items. However, it's not clear who would do that additional work. For the time being, the minutes will be written with fewer direct attributions of statements to individuals/companies, and everyone is reminded to sanity check the minutes and provide corrections to the Secretary prior to their posting on opengl.org.

Working Group Summaries

Bylaws Working Group

Dave Zenz reported lots of active participation, which is great. WG is still in the data-gathering phase, waiting on some feedback - closure on issues list is September 26th. Next step is agreement on prioritization of issues. Breakout session tomorrow will go over the issues list and make another pass at prioritization and identification of easily handled issues.

GL2 Working Group

Bill Licea-Kane reported vigorous activity on the mailing list. Main accomplishments: accepted contributions with signed contributor agreements from 3Dlabs & NVIDIA, and decided to use glslang as the basis for the HLSL. Now merging issues from both glslang and Cg drafts into a single issues list. Working hard to conclude draft spec by end of CY 2002.

Vertex Program Working Group

Pat Brown noted the purpose of continuing the group is to address advanced low-level programming, allowing capabilities such as looping and branching to the instruction set. Architectures are a little more divergent here, but there should be a new level of common denominator that can be reached. This group will pick up as the ARB_fragment_program group winds down.

OpenGL 1.4 Conformance Update

Pat Brown said that NVIDIA had previously committed to developing 1.4 test proposals, and got proposed 1.3 tests from Brian Paul. There's now a set of tests to contribute as a starting point. We will now setup a conformance working group to review tests. NVIDIA also did general cleanup to eliminate compiler warnings. A brief description of the new tests:

• Mipmap generation - verify level generation, do general tests that some level of filtering is being applied.
• Blending features (blend square, blend func separate, imaging) - modify existing blend test to cover these capabilities.
• Depth/shadow textures - generate random depth texture, render 3D coords with constant r=0.5 and iterate over the internal depth formats and comparison functions.
• Fog coordinate - extends existing fog tests so that test will fail if the wrong coord (fog coord/computed fog) was specified.
• Multi draw arrays - make sure geometry drawn via MDA and through the conventional vertex interface is the same.
• Point parameters - ramp test varying point size, run with a variety of different parameter settings. Test that the covered area is monotonic and vithin gross bounds of what area should ideally be covered. Similar to old pntaa test.
• Color sum - generate random texture with DECAL mode and secondary color, iterate through possible color sum enable / lighting enable / separate specular color enable combinations to check that the right sum is being computed.
• Stencil wrap - extend existing stencil test.
• Texture env crossbar - create one random texture as texture 0, bind to one of the N texture units, set others to passthru, test that it gets rendered.
• LOD bias - add LOD bias to texture levels and use LOD bias to cancel out that effect, then rerun miplin / mipsel tests.
• Texture mirror repeat - define trivial texture and render with texcoords over -1..2 range, verify that texture is repeated three times. Tests 1D, 2D, 3D textures.
• Window pos - draw random bitmap at random locations (including offscreen), and make sure bitmap shows up. Verify current raster color.

We need a working group leader to follow up and finalize conformance tests. NVIDIA may be willing to do this if nobody else will.

ARB_fragment_program

Benj Lipchak has led the working group over the last few months, and summarized the end result being presented to the ARB. The extension is based on ARB_vertex_program (much of it word-for-word identical). The only meaningful deltas are:

• Pipeline stages that are subsumed are 3.8, 3.9, and 3.10 (texture, color sum, fog) of the OpenGL 1.3 Specification.
• Instruction set is similar, but adds sin/cos/sincos, linear interpolation, compare, and texture sampling instructions.
• Each instruction can be saturated (clamp to range [0,1]).
• Removed rough-approximation EXP and LOG instructions.
• No relative addressing for parameter arrays, due to universal lack of interest.
• New resource type, "texture indirection", and a minimum requirement on depth of the indirection (4) - some HW has limits on chain of dependent fetches.
• Vertex programs have arbitrary vertex attributes that introduce aliasing issues. Makes no sense to introduce arbitrary attributes at fragment level - only colors, texcoords, fogcoord.
• Most state bindings are the same, removed some such as texgen state and point state. Added texture environment, color, and depth range.
• Completed spec in 3 months, building on vertex_program experience!

Intel suggests moving depth invariance language earlier, since it affects mixing fragment programs with the fixed-function pipeline.

ATI has an implementation that's been tracking the ARB spec and is close. Pat summarized differences with NV_fragment_program. Benj kept the ARB spec very close, and the programming model is the same.

The spec separates out texture images and associated parameters as "texture units" from "texture coordinate state". This supports upcoming hardware which provides a different number of texcoords and texture units.

Added .rgba component selectors. Talked about adding similar selectors for texcoords, but eventually dropped that nomenclature.

Texture lookups specify both texture unit and texture target, so OpenGL 1D/2D/3D texture enables are irrelevant.

Left out multiple color outputs, aux buffers, etc. - possible in a future layered extension.

Texcoords now have same precision requirements as positions and normals (in section 2.1.1).

It's possible to look at the source of vertex programs and determine if it will run on the implementation; not so easy on fragment programs (see issue 21) due to native resource limits. Only way to really tell is load program and see if it exceeds native limits.

Added a precision hint late in the process - accomodates hardware like NVIDIA's, which supports multiple precisions.

Some hardware has fixed-function fog at the end, others do it programatically. Not subsuming fog into fragment programs means some vendors would have to add code at the end of all programs, and change code when fog parameters changed. Added options which allow using fixed-function fog (see issue 27). Some open discussion on this followed.

Added explicit sin/cos instructions with no limits on input range.

Bimal had some concerns about counting texture indirections vs. native resource consumption. Benj offered to add language giving examples of corner cases (e.g. moving 2nd-to-last paragraph of issue 24 to the native resource counting section), which works for Bimal.

Microsoft noted their previously mentioned IP claim. They were asked if they could be at all more specific as to what their claim was, and will follow up with their lawyers to determine this.

VOTE; 10 YES / 2 ABSTAIN / 0 NO / 1 NOT PRESENT (HP), PASSES. HP was not present, but their vote would not affect the outcome.

The final ARB_fragment_program extension spec will be posted to the registry shortly.

New Extension Proposals

Two vertex array objects proposals were presented, a GL2 proposal (slides) from 3Dlabs and a classic EXT proposal from ATI/NVIDIA. ATI noted that the ATI_map_object_buffer extension spec, which came up in discussion, has only been distributed privately to a few ISVs so far.

GL2 objects would be defined in terms of client state and client machine units (also true of the ATI/NV proposal), which makes it hard to move VAOs onto the server. Principal design goal is accelerating rendering for direct rendering clients, not enabling more efficient indirect rendering. ATI/NV proposal is oriented towards more dynamic data, as well.

Dale asked if ATI/NV proposal could be split into staging buffer and VAO parts. Pat agreed they are different functionality. GL2 proposal supports indexing, unlike ATI/NV proposal - ATI and NV already have their own indexing extensions. Contemporary GPUs all support index arrays directly.

Dale noted a significant difference between the two proposals is using the GL2 object management framework vs. having more object-specific allocation calls. Of course legacy support will have to exist, but do we want to perpetuate it going forward? Kurt noted that application-provided handles are critical for display lists and helpful for scripting and other purposes, which colors his opinions. Considerable discussion followed; folks on the phone had trouble hearing and asked that the issues be circulated on the list.

Nick noted two issues: the general, sweeping idea of object management, and the specific issue of which VAO extension we should back, which is time-critical. We will setup a working group immediately to try to resolve the object management issue and still get better VAO management out in the very near future.

Randi then discussed the GL2_async_core extension (slides). This proposes a general, unified synchronization mechanism (analogous to the general, unified object management mechanism). General acceptance of the sync and fence concepts, but FlushStreamGL2 drew considerable concern about its exact meaning and testability. New vblank/swap interfaces resulted in some discussion to clarify just how they would behave.

Nick noted that NV_fence has been shipping since 2000, and thinks it may be possible to implement most of this proposal on top of NV_fence. He'd like to see this happen for compatibility with existing apps, as would John Stauffer. For the time being, we will discuss this on the main participants list.

Misc. Issues

Motivated by Dave's example in the bylaws WG, Jon will create an overall ARB issue tracking list for the participant website, to aid in tracking smaller issues that don't have their own working groups.

Along these lines, several GLX spec issues have been outstanding for some time. Jon will post GLX_ARB_render_texture extension proposal shortly, along with the final GLX 1.4 Specification and a draft of the GLX 1.4 protocol document.

Membership Renewals

Meeting in closed session, the permanent members renewed NVIDIA and Sun as Auxiliary members for another one-year term. No Auxiliary members were promoted to Permanent status at this time.

This ended the core meeting; the late afternoon and second day of our face-to-face time were used for breakout sessions of the Bylaws, GL2, and Vertex Array Object working groups. Notes on those meetings were taken by their respective working group leaders.

Thanks to Dell for hosting this meeting!