It would be worth to set runtime options: https://v2.ocaml.org/manual/runtime.html#s%3Aocamlrun-options I have found that GC options have a very significant effect on overall performance in normalization/conversion checking. We'd also want to use tuned GC settings in real-world implementations, so it's more realistic IMO to benchmark with them. I usedexport OCAMLRUNPARAM="v=0x400,s=100000000,i=100000000" to significant effect, your mileage may vary.
It would be worth to set runtime options: https://v2.ocaml.org/manual/runtime.html#s%3Aocamlrun-options I have found that GC options have a very significant effect on overall performance in normalization/conversion checking. We'd also want to use tuned GC settings in real-world implementations, so it's more realistic IMO to benchmark with them. I used
export OCAMLRUNPARAM="v=0x400,s=100000000,i=100000000"to significant effect, your mileage may vary.