chore(runtime): Avoid branches in 164abs and 164neg in runtime/numbers

[spotandjake]

This is a small pr that changes the way we do abs and neg in the runtime for some of our helpers, normally this would not matter but given we call gcd a lot it makes sense to avoid branching.

[ospencer]

This is the sort of change that without some benchmarking I'd be reluctant to merge the change. It makes the code much harder to read, and this is the exact type of optimization that binaryen/the wasm runtimes would make for us.

[spotandjake]

This is the sort of change that without some benchmarking I'd be reluctant to merge the change. It makes the code much harder to read, and this is the exact type of optimization that binaryen/the wasm runtimes would make for us.

I compiled the program below:

module Main

ignore((1/4) * (2/4))

with grain types.gr --release --wat both with the changes applied and without them and as can be seen from the wat diff, Binaryen does not seem to be catching what is happening and produces the comparison still. (Maybe this is an indicator that we need to play around with the binaryen optimization settings a little more?

I was trying to benchmark this program with a simple loop like:

module Main

for (let mut i = 0; i < 100_000_000; i += 1) {
  ignore((1/4) * (1/8))
}

test()

But i kept running into OutOfMemory: Maximum memory size exceeded when running with wasmtime types.gr.wasm which kind of sounds like a memory leak, and I wanted to let test with a lot of iterations so the jit compiler could kick in and the time taken for the module to warmup would be less of an issue.