Hi, I have a JIT implementation for Lua that uses LLVM. Creating the LLVM IR by hand is tedious and error prone, as well as makes it harder to implement things because it is difficult to make changes. I have started a new implementation that generates C "assembler" code and then uses dmrC to translate to machine code. This has several practical advantages: 1. Much easier to understand and maintain the intermediate C code. 2. Can re-target to LLVM or another JIT engine - in my experiments I am using NanoJIT to start with - without having to change the layer that compiles from Lua to JIT! Of course the added translation from C to the IR used by the JIT engine is an extra step, and I have not yet benchmarked this against hand-written IR. This is still work in progress, but it is looking promising already. One interesting thing is that the generated C code only has gotos - no higher level structured programming concepts! Regards Dibyendu -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
