Monday, December 8

OpenGL 1.2 Specification Review

Jon Leech led a review of the draft 1.2 specification. In addition to numerous suggested edits, corrections, rephrasings, and reorganizations, some points were decided or identified:

• Identify all the undefined or implementation-dependent parts of the spec; there shouldn't be many.
• Check where INVALID_ENUM errors are mentioned explicitly for individual calls; the discussion of errors on page 12 is intended as a catch-all.
• The secondary color used in separate specular lighting will now be treated as RGBA instead of RGB, with an A component of 0.
• Determine how histogram and minmax operate on pixels which are not generated (e.g. due to an invalid rasterpos in DrawPixels) or which are clipped.
• A table should be constructed showing which of the convolution filters is applied to which operations - does this depend on the source pixel dimensionality, destination dimensionality, or some other parameter? The current spec is vague about this.
• Determine how histogram and minmax sinking affects defining texture images. When pixels are sunk, is the texture state updated even though the texture image is not?
• Determine the supported range and resolution of components in the pixel transfer path (see the "conversion to floating point") subsection of chapter 3).
• Decide if minmax clamps before use - can it be restricted to [0,1]? How is it actually used - for range expansion, range reduction, or both?
• See if we can consistently treat internal colors as [0,1] floating point rather than fixed point with an effective scaling, during e.g. the blending operations. The spec language gets convoluted as it stands, with explicit scale factors in the blending equations.
• Resolve the convolution border mode issue between HP, Sun, and SGI.
• CopyTexImage and TexSubImage/CopyTexSubImage specifications are missing. This is probably because they are specified in EXT_texture, not EXT_texture3d. EXT_texture became a part of OpenGL 1.1, with the 3D parts removed. The missing calls must be put in.
• Craig Dunwoody at SGI should proofread the state tables in chapter 6, since he's the expert in this area.
• Decide if imaging subset pixel state should be in a separate, new attribute group.
• A glossary would be very useful. Terminology like "component" is heavily overloaded in the spec document.
• A separate "Image Definition" section preceding DrawPixels, TexImage, and the imaging calls which take image parameters would be very nice. This would describe the layout of 1D, 2D, and 3D images in host memory, and the packing/unpacking that takes place before pixels are processed.

Review Schedule

The following schedule for future review cycles was agreed upon:

Date	Event
January 1, 1998	2nd draft released for ARB and public comment
January 21	Comments due on second draft
February 1	3rd draft released
February 21	Comments due on third draft
March 1	Final draft released
March 9-10	ARB meeting, hosted by Evans & Sutherland in Salt Lake City. Plan to approve the OpenGL 1.2 specification at this meeting.

Tuesday, December 9

OpenGL ICD for Windows

Kurt Akeley distributed a press release announcing the SGI-Microsoft agreement, available at http://www-admin.corp.sgi.com/aeon/business/1297/comments.html. In essence, the SGI Installable Client Driver kit will now be made available to hardware vendors as a Device Driver Kit on both Windows 95 and NT, via Microsoft.

Steve Wright picked up the discussion. They killed the Microsoft Mini-Client Driver kit for Windows 95 for strategic reasons. Drivers based on the new DDK will require certification and quality assurance through WHQL. The Windows 9X logo program will also require IHVs to certify Direct3D drivers; however, D3D may be certified without requiring OpenGL. The ICD license is now being distributed through Microsoft Developers Network.

IHVs are encouraged to continue working with the current SGI DDK until the new ICD kit is available from Microsoft in spring 1998. DDK licensing is now handled through Microsoft only.

OpenGL 1.2 support is expected in a later version of the DDK.

The status of software drivers is undecided. SGI's OpenGL for Windows is faster in many cases, but the code would have to be merged with Microsoft's. SGI will continue working with and funneling code to Microsoft.

The MCD model will continue to be supported on NT for backwards compatibility, but otherwise is effectively dead.

Paula Womack discussed the code base. SGI wants to merge the SampleGL and ICD code bases in the future; this will reduce the support burden for the OpenGL team and lead to faster updates. There is some concern about the clarity of the ICD relative to the SampleGL. Some work will probably have to be done to keep the SampleGL easy to understand and retarget.

GLX 1.3 Specification Review

Paula Womack led a review of the GLX 1.3 specification. It will be revised and reviewed on the same schedule as the OpenGL 1.2 spec.

More about the GLX spec review may be posted later - minutes are not available yet.

OpenGL Conformance Testing

There are three categories of conformance tests: the coverage tests for GL and GLX, the mustpass tests, and the optional conformance tests, some of which an implementation may fail.

At the September meeting, we agreed in principle that licensees responsible for an extension which made its way into OpenGL 1.2 would also be responsible for specifying the conformance tests for that extension. So far, only Evans & Sutherland has produced a proposal, for the separate_specular_color extension. The following assignments were made:

Feature	Responsible
Convolution Border Modes	Sun and HP
BGRA Pixel Format Draw Range Elements	Microsoft (Steve Wright will verify that they can do this)
Packed Pixel Formats	Nvidia (Michael Gold - thanks for "volunteering"!)
Rescale Normal	IBM
Separate Specular Color	E & S - initial proposal presented at meeting, needs changes as suggested below
Blend Color Blend Equation Color Matrix Color Table & Color Subtable Convolution Histogram/Minmax Texture Edge Clamp Texture LOD	SGI has primary responsibility for texturing and imaging features, but will work with other vendors who have implemented the imaging subset to share the load.
Texture 3D	Real3D (Richard S. Wright - volunteered after the ARB meeting)
FB Config Make Current Read Pbuffer (GLX)	SGI

IBM and Intergraph are interested in better testing of existing GL 1.1 features, including resurrecting some previously dropped tests. They will make proposals about this.

Bill Armstrong discussed his separate specular color test proposal. Modifications proposed by committee members included:

• Testing saturation in color sum.
• Testing clipping of both colors and during flat shading.
• Be aware that future extensions may allow explicit specification of the secondary color when lighting is disabled.

We need proposals for conformance tests to be circulated by mid-February, so that ARB members will have a chance to review them before the next meeting.

After the last meeting, Dale Kirkland asked OPC to host conformance results on their Web site, along with the existing Viewperf results. They don't seem to be willing.

Paula suggested that conformance test binaries might be made available on opengl.org, to let people other than licensees run the tests.

Finally, some discussion of the limitations of "conformance" as a measure of driver quality.

Miscellaneous

3D Labs has offered to host the ARB meeting in June in the UK , along with the OPC meeting (June 22). This would bring the two meetings back together. This is unfortunately close to SIGGRAPH in mid-July; they will ask OPC if the OPC meeting might be moved back some to ease the pressure.

Paula discussed SampleGL. SGI hopes to move to a single source tree including both SampleGL and the DDK, so they would share the same software renderer and hardware interfaces. SampleGL may contain some reference code (for clarity) and will also contain GLX and some X Windows specific code; the DDK may contain some code for Windows integration, but there won't be any Windows specific code in SampleGL. A 1.2 SampleGL is expected sometime in summer 1998.

OpenGL++

David Blythe reported on OpenGL++. There's been lots of work, but relatively little communication. A new version of the spec was distributed last week, with big changes including consistent syntax, naming conventions, and terminology for the object concepts. Future spec revisions should come in a more timely fashion.

Implementation work includes:

• Two implementations of concepts for a multithread/multibuffer pipeline model like Performer's APP-CULL-DRAW. They are working on Irix and being ported to NT.
• File loading for VRML and Optimizer, and a generic object framework for loading images, geometry, audio, and video.
• A test suite co-developed with the source code, and documentation auto-generated from the source.
• Support for intersection testing and picking.
• Benchmarks - OpenGL++ generally comes out slightly faster than Optimizer and slightly slower than Performer.
• Currently working on state sorting and caching, tesselation, flattening transformations, invalidating state caches, and an event notification structure for invalidation and other purposes.

An alpha version of the spec is expected in mid-January. It will probably be sent to the entire ARB.

David will try to organize an OpenGL++ subcommittee meeting either this evening, or in early January.

Licensing terms are not yet decided.

Fortran 90 Bindings

Bill Mitchell of NIST reported by phone. Since the review last year, the F90 prefixes have been removed; entry points look like the C API. Module names were changed from F90GL_ to OPENGL_. The pointer type was changed from a character pointer to a structure (derived type).

There have been about 200 downloads/month since the initial release. Bill will update the bindings for OpenGL 1.2 and supply them for inclusion with the SampleGL.

The specification is available on the NIST website at http://math.nist.gov/f90gl/

Discussion ensued; there was some concern about the F90 bindings being a US government work in the public domain, not subject to ARB control. Bill will circulate copies; the ARB is expected to vote a ratify a snapshot of the bindings, whose specification will be stored on opengl.org.

Extension Status

Jon Leech reported on the current status of the shared_texture_palette, point_parameters, and scene_marker extensions. The first two are shipping from several vendors, including 3dfx, NEC, and Nvidia. Scene_marker needs to be revised to incorporate suggestions from email following the last ARB meeting. Dan McCabe volunteered to do this.

Jon also discussed a new WGL extension for controlling monitor color correction tables (aka gamma ramps). This is still being developed. It is intended to address problems with the Windows DDI gamma ramp command - primarily, that it's not available on NT4 or for secondary devices like Voodoo.

David Blythe reported on extensions for higher quality shading. Fragment lighting has been changed to add back in the N dot L specular term, which was previous accidentally omitted. It was also changed to let the diffuse material supply the ambient color, and to specify that fragment lighting doesn't implicitly renormalize N.

Multitexture was changed to decouple the number of texture coordinate sets from the number of textures. This is complex, and David's still working on it. He's also adding more texture environments, like SUBTRACT. SGI originally saw less need for this because they planned to support signed color components. John Tynefeld described the partial multitexture implementation for 3dfx, which only supports REPLACE mode in the final texture environment due to hardware limitations. David is looking for feedback from other IHVs.

Bump mapping (fragment light space) again needs a way to represent negative values. David is working on a scale and bias technique to map between the normal [0,1] color range and [-1,1].

Jack Middleton described Sun's MultiDrawArrays extension. This specifies multiple sets of (first,last) pointers to DrawArrays, and can significantly boost rendering speed for short triangle and line strips in Sun's OpenGL implementation.

Procedural Issues

At the spec editor's request, the ARB agreed that in the future, extensions would not be accepted into the core or as common "ARB" extensions unless they were written against the current specification. There was considerable difficulty integrating some extensions written against 1.0 into 1.1, including the above-mentioned problem with EXT_texture3d.

Java Bindings

David Yu from SGI gave a status report. Bindings have been written against both the SGI and Windows JDKs.

Jack Middleton from Sun reported on proposed AWT bindings. They've added classes to specify a "configuration" object describing a visual. Methods exist for defining configurations, querying for visuals matching them, and creating AWT canvases satisfying them. This work will be in the 1.2 JDK, shipping from Sun in 1998. Jack will send the spec out to the ARB after he and David work on it a bit more.

Printing

Tom Palmer of CEI reported on their product, GLpr (see their website at http://www.ceintl.com/products/GLpr). It captures the OpenGL command stream with feedback mode and converts to resolution independent PostScript using a painter's algorithm. Texturing is not supported; transparency is possible with screendoor, but not yet implemented. This is aimed at CAE/GIS/CAD/Molecular modelling domains.

GLpr supports OpenGL 1.1. It consists of about 20 new calls and is easy to integrate. It supports overlaid text annotation. It is licensed to ISVs, not end users, and doesn't incur runtime fees.

Thanks to Intel for hosting this meeting!