From a linear programming language to CUDA? building a map from ANSI C to C for CUDA

Well,our research group is doing some research on building a map from linear programming language to C for CUDA,it involves the optimizing methods of compilers as well as binary translation.
Now,we have designed a framework of it based on some existed compiler infrastructures,I would like to know that is there anyone who has done or is doing the same thing now?