gc-bench

Benchmarks for evaluating MPL performance.

Setup

You need these in your path:

• mlton, the MLton compiler
• mpl, the MaPLe compiler
• mpl-cc, the MaPLe compiler with concurrent collection

Run

Run all experiments. By default, does each benchmark one. You can pass --repeat N to do N repetitions of each benchmark.

$ ./run

$ ./run --repeat 10

The run script produces a file results/XXX where XXX is the current date/time. The report script will print a summary. With no arguments, this selects the most recent results. Or, you can pick a particular file.

$ ./report

$ ./report results/XXX

Huh?

Here's how the repo is organized:

• bench/ has benchmark codes. Each benchmark has its own subdirectory and .mlb for compilation.
• lib/ is a shared library.
• config/ defines how to use different compilers to make a benchmark.
• inputs/ contains files which are used as input by some benchmarks.
• scripts/ has some utilities for building and running benchmarks.
• exp.json defines the experiments to be performed. It's fairly self-explanatory.

Individual benchmark binaries are named BENCH.CONFIG.bin, where BENCH is a benchmark name (one of the subdirectories of bench/) and CONFIG is a compiler configuration (one of the files in config/).

The top-level makefile takes a binary name, compiles it and puts the resulting binary in bin/. You can run a single benchmark this way, for example:

$ make fib.mpl.bin
$ bin/fib.mpl.bin @mpl procs 4 -- -N 42