I implemented a fair number of optimizations, but there aren't many large benchmarks to really crunch down on. An experimental register allocation pass was awkward (DynASM doesn't really support run-time register selection) and didn't improve performance much.
I implemented a fair number of optimizations, but there aren't many large benchmarks to really crunch down on. An experimental register allocation pass was awkward (DynASM doesn't really support run-time register selection) and didn't improve performance much.