Static hermes is slow. Document how to optimize

[paulmillr]

Here is the small sha512 implementation.

https://gist.github.com/paulmillr/f57268b66c3af9adf2423ef232cb0533

We've discussed the speed-up on twitter with @tmikov. The question is: how to make this much faster vs JS? Note that bigints aren't used internally - K512 is split into u32 arrays.

Otherwise there doesn't seem to be much of a point in the "blazing fast" Static hermes now

[tmikov]

How is this supposed to be executed to measure its performance? Is there commented source, or should we use the bundled output?

[paulmillr]

Updated the code to simplify benchmarking. @tmikov

❯ node noble-hashes.js                                     
6 ms/1000 iters of 32B
result: 9a475dff7125942adcc2f8d6a1ee8a058cab348e044356abb98eaeeb74a2d9ff02a1291771bbd08200edc28b743be8861ad1df084557e8b2ddfc7b4b73adff59
82 ms/10 iters of 1MB
result: 81368a4fd197fa54ba8e7e48e51a027f26571c2aa8cd0931320912668816dc3d29a06aa1febd437b7de4bf0c4fd7cc6f09052fdb7253b93e2788c05eba82d6e9

❯ ./build_release/bin/shermes --typed -exec noble-hashes.js
32 ms/1000 iters of 32B
result: 9a475dff7125942adcc2f8d6a1ee8a058cab348e044356abb98eaeeb74a2d9ff02a1291771bbd08200edc28b743be8861ad1df084557e8b2ddfc7b4b73adff59
1149 ms/10 iters of 1MB
result: 81368a4fd197fa54ba8e7e48e51a027f26571c2aa8cd0931320912668816dc3d29a06aa1febd437b7de4bf0c4fd7cc6f09052fdb7253b93e2788c05eba82d6e9

[tmikov]

Thank you! Will start working on it.

[tmikov]

Did some preliminary investigation. The biggest slowdown is from lack of integer support.

[paulmillr]

What's integer in this context? u32? i32? non-float u64?

[tmikov]

It literally means integers. Don't worry about it, we are working on it. We appreciate the sample js.