2009/6/18 Edward Kmett <[email protected]>: > What is interesting is in a lazy setting, if you are tracing a bytecode > representation that knows about allocation and thunks, you can do some > additional optimizations in here. If on every path to a side exit or the end > of the loop you find that the thunk is evaluated you can evaluate it > strictly and move its execution earlier in the trace. This gives you a weak > form of runtime strictness analysis. If the pointer to that thunk never > escapes, then you can unbox the contents of the thunk and operate on its > members in registers. Add constant folding, polyinline caching to improve > branch prediction for spineless tagless g-machine thunk evaluation, and code > migration to the side exits and it becomes an interesting runtime system.
This sounds absolutely awesome! Is the source code for your prototype publicly available anywhere? I'd love to take a look at the basic structure of something like this - trace JITing is something I keep meaning to look at in more depth. Cheers, Max _______________________________________________ Haskell-Cafe mailing list [email protected] http://www.haskell.org/mailman/listinfo/haskell-cafe
