Using ocamlopt with ocamlbuild

Reading time ~1 minute

As your OCaml programs grow bigger, it is a good idea to use a build system. Ocamlbuild is quite good at this job ! It will take care of your dependencies, find libraries and more importantly compile the files in the right order…

It took me a while to find the answer to the simple question, how to use ocamlopt instead of the default compiler. Assuming you want to compile a main.ml file, the short answer is to use ocamlbuild main.native, instead of ocamlbuild main.byte

ocamlopt and ocamlc

OCaml comes with two compilers : ocamlc and ocamlopt. ocamlc is the bytecode compiler, and ocamlopt is the native code compiler. If you don’t know which one to use, use ocamlopt since it provides standalone executables that are normally faster than bytecode.

Example

For a very quick benchmark let’s sort a long list.

let a = List.init 1000000 (fun x -> Random.int 500) ;;
let b = List.sort compare a ;;

print_string "done;\n";

And compile this with the different compilers…

ocamlbuild main.native
echo "native"
time ./main.native
rm main.native
rm -rf _build/


ocamlbuild main.byte
echo "byte"
time ./main.byte
rm main.byte
rm -rf _build/

And the results are quite convincing.

Finished, 4 targets (0 cached) in 00:00:00.
native
done;

real  0m2.156s
user  0m2.064s
sys 0m0.088s
Finished, 3 targets (0 cached) in 00:00:00.
byte
done;

real  0m6.340s
user  0m6.268s
sys 0m0.068s

Learning more

Real World OCaml: Functional programming for the masses by Yaron Minsky, Anil Madhavapeddy and Jason Hickey is a good introduction to OCaml.

How to optimize PyTorch code ?

Optimizing some deep learning code may seem quite complicated. After all, [PyTorch](https://pytorch.org/) is already super optimized so w...… Continue reading

Acronyms of deep learning

Published on March 10, 2024

AI with OCaml : the tic tac toe game

Published on September 24, 2023