You don't really need a JIT for those things, smart compile time analysis can get as good or better results. See Stalin scheme:
http://en.wikipedia.org/wiki/Stalin_(Scheme_implementation) On Wednesday, July 11, 2012 3:05:50 PM UTC+2, tbc++ wrote: > > >> I've been debating on working on a native Clojure > > I've also been debating this for over two years now,y but I keep > coming back to the same problems over and over again: > > 1) A from-scratch native Clojure either needs to be just a compiler > (as this project is), or you're going to need a decent GC, JIT and > object system > 2) The Boehm GC is easy to use, but it's not fast or accurate. The > JVMs can implement generational concurrent compacting GCs (and do > depending on your implementation). These are all very useful features, > and things that Boehm doesn't support. In addition, Boehm isn't > accurate, so memory leaks can still occur. > 3) Unless you plan on staying strictly AOT, and igoring the REPL, then > we're going to need a JIT, that's not exactly something you can hack > together in an afternoon > 4) Arithmetic must auto-unbox or it will be quite slow. This also > requires a JIT. > 5) Clojure is made up of hundreds of single line functions, a JIT > should be able to inline this, one more reason why its needed > > > All these reasons have brought me to the conclusion that the simplest > way to get all this, is to base a native version of Clojure on PyPy. > No, I didn't say write it in Python and have it run on the Python VM. > PyPy is technically speaking a "language agnostic JIT generator > written in Python". > > A good overview of this process is here: > > http://morepypy.blogspot.com/2011/04/tutorial-writing-interpreter-with-pypy.html > > > What you get with a JIT written in PyPy: > > 1) Your interpreter/JIT is written in a dynamic language with tons of > meta-programming abilities > 2) You get at least 3 GCs (reference counting, mark-and-sweep, Boehm) > 3) You get a tracing JIT, that for dynamic languages blows method > level JITs away > 4) You get STM abilities (as of the most recent work in PyPy) > 5) Automatic unboxing, inlining and constant propagation of JIT'ed > code (comes free with a tracing JIT). > 6) And you still have great interop with C and with some work C++ (via > Reflex) > > I don't really want to hi-jack this thread, I just wanted to make > people aware that writing a fast native Clojure takes a bit more work > than simply slapping LLVM and Boehm together. Do that and the JVM will > beat you every single time. > > Timothy > -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
