There's just a few weeks to go until the release of
LLVM 3.2, but AMD is still trying to get its "R600" GPU back-end merged into this next compiler infrastructure release.
Going back to March, AMD has been
trying to merge its R600 GPU back-end that is optionally used by their open-source graphics driver stack and is a requirement for the
Radeon OpenCL support with the open-source driver. The LLVM back-end can be used as part of
the R600 Gallium3D shader compiler. (See
benchmarks of the R600 LLVM compiler back-end from several months ago.)
Up to this point, the R600 LLVM back-end has been living in
a branded LLVM repository by Tom Stellard at AMD. The code should now be in a good state to be merged and Tom's hoping to see this happen for the upcoming LLVM 3.2 release in mid-December. This LLVM back-end has been publicly
available since last December. Originally it lived within the Mesa code-base but then it was decided it should ultimately be merged into LLVM.
It's not clear at the moment if LLVM developers will pull it for the LLVM 3.2 release, but
the discussion is still taking place.
Other LLVM 3.2 features include
PowerPC improvements, an
auto vectorizer,
Polly optimizations, and much more.