Would it make sense to use LLVM (with its full-fledged JIT and optimizers) to run an interpreted language code like Matlab?
What are the concrete parts of execution that could be enhanced, by using an abstract compiler-optimizer instead of Matlab's current modus operandi? (I understand it probably has a simple LLVM-like optimizer, but I don't believe it is as powerful as LLVM itself)
The reason I thought of this is because using the runtime information a typical Matlab calculation has (size of arguments etc.) I think a full-fledged optimizer could increase execution speed in very simple scenarios where a Matlab program now needs to take special measures to ensure optimal performance (like explicit preallocation of variables etc.)
I understand there is currently no support for any of this at all, but I'm just wondering what the impact of such an interpreter would be.