One of the GSoC proposals for LLVM this year is automatic GPGPU code generation support.
A student developer out of China has plans to attempt automatic GPGPU code generation for LLVM, assuming his Google Summer of Code work is accepted. He wants to build an automatic GPGPU code generation framework for LLVM based upon LLVM's existing Polly and PTX back-ends.
From Yabin Hu's proposal
, "With the broad proliferation of GPU computing, it is very important to provide an easy and automatic tool to develop or port the applications to GPU for normal developers, especially for those domain experts who want to harness the huge computing power of GPU. Polly has implemented many transformations, such as tiling, auto-vectorization and openmp code generation. With the help of LLVM's PTX backend, I plan to extend Polly with the feature of GPGPU code generation."
This would be quite a feat to pull off, especially for someone without a lot of LLVM development experience, but we'll see if the project gets approved by Google and then how far the student can make it over the course of the summer.
Some other related reading that would be recommended would be GCC's StarPU for CPU/GPU hybrid task programming
, Gdev as an open-source CUDA implementation
, and Emscripten