My undergraduate dissertation, titled 'An optimising compiler from an ML-like language to Java bytecode'. Features include inlining, tail call elimination and tail call modulo context transformations.
- Install opam
- run
opam switch create 5.1.1 - run
opam install . --deps-only - Install Krakatau
- Install a copy of a JDK.
> bash scripts/build_and_run.sh -h
_build/default/bin/compile.exe: unknown option '-h'.
compile -s <program_string> -f <file> <options>
-s Set string to compile
-f Set file to compile
-c Set name of generated class
-o Set output file
-opt-all Enable all optimisations
-peep Enable peephole optimisations
-const-fp Enable constant folding and propagation
-inline Enable inlining
-inl-threshold Adjust threshold for expression score in order to apply inlining. Default 10
-tmm Enable tail recursion modulo monoid
-tco Enable tail call optimisation
-tmc Enable tail recursion modulo cons
-dyn-lambdas Compile lambas using invokedynamic
-peep-box Enable boxing peephole optimisation
-peep-push-pop Enable push-pop peephole optimisation
-peep-goto-label Enable goto-label peephole optimisation
-peep-store-load Enable store-load peephole optimisation
-dump_debug Dump debug information (currently just compile times) to json file
-help Display this list of options
To run the benchmarking programs, additional dependencies are required:
mlton, moscow ml, polyml, sml/nj, hyperfine, maven, python, matplotlib and numpy.