Have you guys considered piggybacking off LLVM for the JITing? I'm seeing it used all over the place: Apple to speed up their OpenGL implementation in OS X, Google's "Unladen Swallow" project to speed up Python by a factor of 5, and many others. I haven't used it myself (yet), but to the best of my knowledge, it's a highly modular and generic compiler infrastucture which can be used for both static and dynamic just-in-time compilation, is (reportedly) incredibly easy to use, in C++, and fast, both at generating code and the generated code itself.
I don't know how much hand-optimizing has gone into the existing dynarec, but LLVM seemingly beats even projects which have been plugging along for decades (like GCC) at performance, so it sounds like it would be worth a look, assuming that it's applicable at all.
LibJIT also looks interesting.
I don't know how much hand-optimizing has gone into the existing dynarec, but LLVM seemingly beats even projects which have been plugging along for decades (like GCC) at performance, so it sounds like it would be worth a look, assuming that it's applicable at all.
LibJIT also looks interesting.
