Optimizing Performance

Generally, you will want to compile with a release build, sometimes even for development.

This is as simple as:

cargo build --release # you may need to check your release profile

This can oftentimes be the difference between 40% and 3% CPU usage.

The following tips generally can be of help:

Avoid allocation or system calls in the node.
Avoid any locking primitives. Prefer the rtrb crate for SPSC message passing.
If possible, branch at the entry of the function as opposed to hot, per sample paths, branchless may also help
Consider polynomial approximations or LUT, if it meets your accuracy demands.
Try to use nightly simd or auto-vectorization for math-heavy paths. Sometimes this kicks in for free when using chunking utilities that match the LANE size on your arch.