John Kessenich of LunarG has posted a proposal for LunarGLASS on the mesa-dev mailing list. "LunarG has decided to work on an open source, long-term, highly-functional, and modular shader and kernel compiler stack. Attached is our high-level proposal for this compiler architecture (LunarGLASS). We would like to solicit feedback from the open source community on doing this."
The LunarGLASS proposal is a six-page PDF and some of the takeaway quotes include "The short-term goal is to leverage investments in existing IRs while the long-term goal is to reduce the number of IRs and not require optimization difficulties caused by losing information going through an IR...Although a variety of IRs could be used for this purpose, a major part of this proposal is to base
the IRs on LLVM...Finally, new languages like OpenCL C, as well as ever more complex uses of GLSL, are raising the level of sophistication needed in compiler stacks. This calls for a solution more in line with handling the full-fledged language features and optimizations that LLVM is already prepared for."
There's also a block diagram of what's involved with LunarGLASS.
What's largely being discussed right now on the mailing list in regards to LunarGLASS is whether LLVM (the Low-Level Virtual Machine) is right for the IR (intermediate representation) or not for both the high and low-level implementations.
Keith Whitwell of VMware / Tungsten Graphics says "I think this is the big question mark with this proposal -- basically can it be done?...At a high level, LLVM is very compelling - there's a lot of work going on for it, a lot of people enhancing it, etc. Now, if it's possible to leverage that for shader compilation, I think that's very interesting."
José Fonseca, also of VMware, says "I also think there are indeed challenges translating LLVM IR to something like TGSI, Mesa IR; and I was skeptical about standardizing on LLVM IR for quite some time, but lately I've been reaching the conclusion that there's so much momentum behind LLVM that the benefits/synergy one gets by leveraging it will most likely exceed the pitfalls."
We'll see how LunarGLASS comes about and will keep you posted. The Low-Level Virtual Machine is already used heavily by Gallium3D's LLVMpipe driver to provide a faster CPU-based, software driven renderer thanks to LLVM's optimizations. LLVM has also been long talked about for being used elsewhere by Linux graphics drivers.