|
ARB Meeting Notes
September 9-10, 1997
Hosted by SGI in Santa Clara, CA
Meeting notes taken by Randi Rost, HP |
Attendees
List of Handouts
The handouts in the list below were distributed at the meeting.
If the the document title is not linked to the actual document,
you can request a copy of the document from the contact person
listed.
Discussion of Extensions
David Blythe led a discussion of the extension specs that he's
been working on: fragment lighting, light texture, and multitexture.
He summarized the philosophy behind each extension and the main
issues that remain with each spec. Several people expressed interest
in communicating outside of the meeting to resolve issues with
the fragment lighting spec.
Jon Leech described the thinking behind the shared texture palette
specification. He also described the point parameters extension,
which SGI proposed to support the need for rendering particle
systems. Next, Jon described the status of the scene marker spec.
Some of the folks developing tiled rendering hardware for PC's
are quite interested in seeing this capability get into OpenGL.
However, Jon. pointed out that there really are no implementations
of this spec yet, so it seems premature to make this part of OpenGL
1.2. A variety of issues were raised and discussed.
Conformance Tests for ARB-blessed Extensions
Should there be conformance tests for EXT extensions? Some ISV's
have raised the issue that there are differences between some
EXT extensions from different vendors. It seems like the company
(or companies) that are really driving the extension definition
should provide conformance tests. If we decide to require conformance
tests for EXT extensions, how do we deal with all the existing
EXT extensions? There is general consensus that it would be good
to "raise the bar" for compatibility of EXT extensions by requiring
conformance tests for them. Kurt suggests that the ARB could
provide the service of sanctioning conformance tests for EXT extensions.
No concrete decision was reached.
How should conformance test results be reported? This has become
an issue because of the OPC's desire to associate OpenGL performance
data with conformance data. SGI does not want to be responsible
for publishing conformance results for other vendors. Kurt has
suggested that the OPC should handle the publishing of the conformance
results. According to Dale, the OPC has struggled with this issue
and there is no real consensus about the best way to handle this.
Paula suggests that the best thing to do at this point is to have
a direct dialog with the OPC folks in order to make progress on
the issues.
OpenGL 1.2 Imaging
Randi discussed the issues that have been raised with the convolution
border modes spec. Phil Lacroute of SGI has provided equation-based
descriptions of the various modes. Phil also suggested removing
the clamp during the specification of the convolution border color.
The original motivation for including the clamp was to make it
similar to the specification of the texture border color. Phil
pointed out that these two cases are not parallel: texture colors
are clamped during download, but input to convolution is not.
Thus, it seems like it makes sense not to clamp the convolution border color. Robert Hoffman of Sun
suggested that the constant border mode be changed to apply the
constant color after the convolution kernel has been applied.
Randi forwarded Robert's mail together with a recommendation to
go ahead and make this change. It was reported that Phil thought
that making this change was not a good idea. Randi will contact
Robert and Phil to resolve this issue and send out the revised
spec for people to review.
OpenGL 1.2
Optional support for advanced imaging - Randi recapped the mechanism for including the imaging capabilities
in OpenGL 1.2. The conclusion reached at the last ARB meeting
was to add a new extension string, "ARB_imaging" that would indicate
the presence of the optional imaging capabilities. If the "ARB_imaging"
extension string is present, the entry points and tokens for the
optional imaging capabilities would also be present in the implementation.
The tokens and entry points would not contain any suffix or prefix that differentiates them from standard
OpenGL tokens and entry points. Furthermore, the descriptions
of the optional capabilities would be merged directly into the
OpenGL 1.2 specification. Conformance tests for the imaging capabilities
would also be included as part of the sample implementation for
OpenGL 1.2.
EXT_paletted_texture - Brian Hook and Chris Hecker sent email arguing that the ARB
should not include this extension as part of OpenGL 1.2. They
believe that texture compression will provide a long-term solution
for this problem. An ARB voted was conducted on the proposal to
remove the paletted texture extension from the list of items going
into OpenGL 1.2. The vote passes 7/1/1 (yes/no/abstain).
EXT_texture_lod - Kurt attempted to revise this spec to make it unambiguous.
As a result of this attempt, he encountered some additional issues
that hadn't been adequately resolved in the earlier spec. Some
wording still needs to be tightened up, but this revised spec
is intended to serve as the basis for what goes into OpenGL 1.2.
EXT_packed_pixels - There was a discussion of the definitions for the bits within
packed pixels. The following table was produced on an overhead
transparency slide in order to show the pixel types that are supported
by this specification:
| |
RGB(A)
|
BGR(A)
|
|
3_3_2
|
X
|
|
|
2_3_3_REV
|
X
|
|
|
5_6_5
|
X
|
|
|
5_6_5_REV
|
X
|
|
|
4_4_4_4
|
X
|
X
|
|
4_4_4_4_REV
|
X
|
X
|
|
5_5_5_1
|
X (See Example 1)
|
X
|
|
1_5_5_5_REV
|
X (See Example 2)
|
X (See Example 3)
|
|
8_8_8_8
|
X
|
X
|
|
8_8_8_8_REV
|
X
|
X
|
|
10_10_10_2
|
X
|
X
|
|
2_10_10_10_REV
|
X
|
X
|
The suffix "REV" indicates that the components are inserted in
the reverse order within the machine unit (e.g., 2_3_3_REV indicates
that the 3 bits of red, 3 bits of green and 2 bits of blue are
loaded into the byte from right to left).
(Example 1) Big endian
15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00
R R R R R G G G G G B B B B B A
(Example 2) PC - Little endian
15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00
A B B B B B G G G G G R R R R R
(Example 3) PC - Little endian
15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00
A R R R R R G G G G G B B B B B
SGIS_texture_edge_clamp - This extension needs to be changed to EXT. No other issues
were raised.
EXT_separate_specular_color - David B. proposed that the per fragment lighting spec be used
as a starting point to create a simple spec to support the desired
functionality. The concern at SGI is that introducing a second
color would lead us down a path that might not be right for the
long term of OpenGL. To SGI, the per fragment lighting spec feels
more like the right long-term direction, so it would be nice to
solve the specular color issue in a way that fits with the evolution
of OpenGL. David passed out a spec that embodies the proposed
alternative. Kurt feels like we shouldn't put something this
new in OpenGL 1.2. After lengthy soul-searching, the consensus
is that the separate specular color extension fills a definite
need, there is a desire to see this capability get into OpenGL
1.2, and as long as we resist the urge to embellish the capability,
it could be made to work with the per fragment lighting stuff
in the future.
EXT_rescale_normal - No new issues.
Vertex array proposals - Bimal updated folks on the various vertex array proposals:
vertex array lock, vertex array range, and vertex array set (aka
object). There are a number of proposals that have been floated
to address several different concerns with implementing vertex
arrays as efficiently as possible. Who wants to see vertex array
sets in the spec? (11/1/11)
End of Day 1
Meeting Schedule
Next OPC meetings are November, January, and April. The next OpenGL
ARB meeting is currently scheduled for the week of Dec. 9, hosted
by Sun. No other meeting times were established, but it feels
like we won't be able to sync up with the OPC meetings in January
or April. We'll probably have a meeting in March.
SGI's OpenGL for Windows
SGI has developed an ICD implementation of OpenGL for Windows,
and they are making it available to OpenGL licensees who sign
an addendum to their OpenGL license agreement. SGI is working
to make this implementation more amenable to porting to new hardware.
They hope to have these changes ready to release in a beta-quality
product at the end of October. Some of the pertinent points:
- The source code was first made available to licensees in July
'97
- Currently adding support for hardware acceleration
- Implementation is optimized for performance
- Source is available to OpenGL licensees
- An addendum to the OpenGL license agreement is required
- Currently working on clearly defining the interfaces to support
hardware acceleration
- Porting effort necessary for rasterizing hardware should be about
4-6 weeks
- Aiming for end of October '97 release (beta quality)
- Code supports all Microsoft OpenGL extensions
- A test suite will also be provided
- No date yet for final release
Vertex Array Set
Bimal recapped the reasons that people were interested in the
vertex array set proposals:
- Delayed reference of data
- Single copy of data
- Preprocess data
- state independent
- state dependent
- Cache data without copy
- Fast array configuration changes
- Sharing vertices between primitives
- Matching OpenGL++
- Address space control
David B. suggested that the vertex array set proposal doesn't
yet have enough consensus to be added in OpenGL 1.2 and that it
might make more sense to form a smaller group of people to go
off and come up with an EXT extension that really works. David
B., Paula, Bimal, Dale, Matt, and Bill A. indicated interest in
participating in this effort.
Specular After Texture
Kurt passed out a preliminary version of a spec that addresses
the need for specular highlights on textured surfaces. This spec
is similar to the definition of fog in the current spec, in that
it defines the behavior and states that implementations are not
required to support the capability. The behavior defined by this
spec is that when lighting and texture mapping are enabled, a
new mode allows the results of the RGBA lighting to be split into
two separate values: one specular and one that combines ambient,
diffuse, and emissive components. The latter value is used as
input to the texture environment, and the computed fragment color
is the sum of the specular color resulting from texture application.
Kurt felt that this proposal provided the capability that people
were after and left the door open for doing the right thing with
per-fragment lighting in the future. People are encouraged to
think about Kurt's proposal. The current plan of record is to
pursue the separate specular extension defined by Bill A.
OpenGL++
Mark Segal provided an update on the status of OpenGL++. He was
not able to have a new version of the spec ready in time for this
meeting, but he will be releasing a new version of the spec within
a month. Some simple demos and simple tests are running. A more
final draft specification that could be presented to the ARB should
be ready by the end of the year. Companies that are participating
in the effort are SGI, IBM, Intel, Digital. Mark expects that
an alpha-quality implementation will be ready by the end of the
year. A Beta release would be ready next summer. The group met
Monday and spent the day talking about procedural and technical
issues. A fair amount of time was spent on multithreading issus.
The plan is to use Pthreads, which should provide a good implementation
across UNIX systems. Mark promised to send out the minutes from
Monday's meeting. Mark answered several questions about how OpenGL++
compared to DirectModel and Java3D, and which OpenGL extensions
would be needed (answer: none, just OpenGL 1.1).
SampleGL Update
The next SampleGL release is planned for sometime after the OpenGL
1.2 spec is finalized. The SampleGL for OpenGL 1.2 will include
support for the optional imaging capabilities. Paula would like
to see the SampleGL support the "blessed" EXT extensions too.
The list of things under consideration for the 1.2 release of
SampleGL includes:
- OpenGL 1.2 capabilities
- Extensions
- More/better conformance
- Better reference code
Paula will develop a SampleGL plan for 1.2 that can be discussed
at the next meeting.
Conformance
Paula polled the room to see what people thought should be done
about conformance testing for OpenGL 1.2. Should we add back
some of the tests that were thrown out? Jim C. suggests that maybe
you can put tests into a kit that vendors could use to exercise
an OpenGL implementation. There seems to be more issues with differences
between OpenGL implementations in the PC space. Paula also suggested
that the following companies should determine whether they can
contribute conformance tests for the indicated extensions:
Microsoft
SGI
- Texture 3D
- Texture LOD
- Edge clamp
- Packed pixels
E&S
Intel or IBM?
HP and SGI?
People are encouraged to send specific proposals to Paula about
what they want added to the conformance test suite.
GLX 1.3
Paula passed out specs for SGIX_pbuffer, SGIX_fbconfig, and SGI_make_current_read.
Paula hopes to see the GLX 1.3 schedule follow the same schedule
as OpenGL 1.2, with a spec available in November. There are no
currently open issues in the pbuffer spec. Paula talked about
the changes made to the fbconfig spec (listed in the handout).
Everyone needs to review the changes and raise any issues as soon
as possible.
Conformance Test Issues
Change #1 (drawpix.c change #1 on handout): Passes 6/0/1
Change #2 (drawpix.c changes #2 and #4 on handout): Passes 7/0/0
Change #3 (feedback.c): This is a major change. People should
go home and think about this one...we may decide to just make
this change with OpenGL 1.2.
Java Bindings for OpenGL
David Yu passed out a handout listing the design history and issues
considered during the development of the Java bindings for OpenGL.
He went through the handout, verbally describing the issues and
resolutions that were reached. SGI hope to release this soon for
Windows and SGI operating environments, and to contribute the
code to the ARB after that. The code would be made available
to OpenGL licenses before the end of the year. A straw poll indicated
that six or seven companies besides SGI are interested in seeing
the Java bindings completed.
Thanks to SGI for hosting this meeting!