HPC framework

I have been thinking of using LLVM to design a framework for parallel
programming of High Performance Computers. The concept is that a VM will
provide the necessary partition between software developers and the
underlying hardware. If a programmer could describe his program
sufficiently well so that parallelism can be easily extracted, and
hardware vendors could describe their systems sufficiently well so that
the parallelizing could be optimized, the result would be highly
portable and efficient.

I have created a GitHub repository to work on this idea, although at the
moment I have no code. I do have a wiki set up at
http://wiki.github.com/guillermin/FPPF/ with a lengthier description of
the idea, and the doubts I am having. I would be very interested in
hearing from LLVM developers, since I have only started to look at LLVM
and still don't fully grasp it.

Thank you for your time,

A good way to start would be to look at Chapel, Fortress, and X10,
three projects which have basically the same goals as those you
describe, and write down some thoughts about how your system would
compare and contrast with them. This would help people understand
what kind of system you're thinking about, which would help them
determine how LLVM might be used in it.