Siggraph 2001 BOF Report

Most of the BOF was concerned with upcoming Java3D work - both 1.3 and 1.4. There were 4 speakers. Michael Schulman introduced everything, then Kevin Rushforth spoke about 1.3 and Doug Twilleager about 1.4. At the end Julian Gomez spoke about some forthcoming efforts in combining 3D/J3D into business systems.

The following are notes based on the slides and what was discussed at various points.

Development timescales

• 1.3 public beta Q4 2001, probably late October.
• 1.3 FCS Q1 2002, dependent on JDK 1.4 release date
• 1.4 Call For Experts Q4 2001

1.3 Features

• Depth-sorted transparence
• Per-view objects
• GeometryArray enhancements
• Piles of texture enhancements
• Raster destination can now use screen pixel offsets
• Behaviour synchronisation and performance updates
• Alpha based behaviours can now be paused
• scale-invariant OrientedShape3Ds
• OrderedGroup can now re-order children through a mask
• Audii improvements and environment parameters

Performance improvements

• Improved memory usage by cutting down on internal structures, removing all dynamic memory allocation (no GC due to Sun code) and internal list handlers become optimised.
• Optimise scene graph updates. Major areas of focus are transform updates appearance changes, lights and LODs.
• Performance optimisations are specifically targeting:
  • Geometry by-reference
  • Texture image handling
  • Scene anti-aliasing
  • OrderedGroups
  • Behavior schedular
  • Reduction in scene setup time

2D Optimisations

In addition to this, Raster code will now accept screen pixel offfsets for smaller work. The effect of this, plus the other overlay work in development will be to effectively eliminate the need for Text2D.

Utility Classes

The second utility to be added is the Scenegraph I/O package that people have been introduced to in the J3DFly demo from Sun.

1.4 Discussion

Extensibility

• Visibility/Picking/Collision: Interfaces provided to allow changed or optimized versions of these functions. It is possible the picking utils will move into the core.
• Rendering devices. Adding new rendering devices such as a software renderer or haptic devices.
• Threads: Allowing the user to provide, extend or change some of the internal threads. For example haptic systems may need to run at hundreds or thousands of times a second rather than the 10-100 of the visual renderer.
• Native context information: One of the most requested features so that users can access the native OpenGl or DirectX system.
• The extensibility will not allow the user to override existing semantics, such as the meaning of a Node or structure.

Programmable Shaders

• Vertex and Pixel shaders will be supported.
• A high-level language that is hardware independent: This will likely be RenderMan-like and will most certainly not be Java classes. There are a couple of papers here convering ideas on this, but Sun seem to want to be even more generic.
• There will be a shader pipeline concept that allows you to insert operations at any step. You can just define rasterisation only code for example.
• For cross-platform capabilities, the language is compiled at runtime.

Advanced Rendering Techniques

• Shadows
• Reflections
• Ancillary buffers will be made available for things like stencil and frame buffers.
• Trying to decide whether buffer information should be implicit "Make it always available in this mode" or have an explicit multi-pass rendering API.

Miscellaneous

• Geometry picking may become part of the core
• Behavior semantic changes such as per-view behaviours and allowing them in shared groups.
• Suggestions for anything else to consider?

J3D and IT

The system needs to be cross platform and Julian is looking to start a working group to consider the issues. There is no direction for a formal new area such as a JSR at this stage although at some time in the future there may be. For example, this group would aim to provide suggestions for changes to the Java3D API both in 1.4 and the future.