After evaluating what I believe this crates goals are (being simple) I have come up with (along with the help of browsing other hex crates performance boosting methods) a minimal code change that doesn't sacrifice a lot of readability, or dip into unsafe, while still giving a noticeable speedup to the decoder.
Its possible that even simpler changes could speed up the crate, but this is the least intrusive method I have found that still produces a real performance gain (over the past few days of looking into this).
This code passes all tests, fmt, and clippy (stable rust 1.72).
Relevant benchmarks (using benches/hex.rs):
Performed on a Ryzen 2600
Before change
hex_decode time: [139.11 µs 139.97 µs 140.95 µs]
After change
hex_decode time: [107.11 µs 108.24 µs 109.44 µs]
This crate is a candidate to include #![forbid(unsafe_code)], and I recommend doing so, but that is outside of the scope of this PR.
After evaluating what I believe this crates goals are (being simple) I have come up with (along with the help of browsing other hex crates performance boosting methods) a minimal code change that doesn't sacrifice a lot of readability, or dip into unsafe, while still giving a noticeable speedup to the decoder.
Its possible that even simpler changes could speed up the crate, but this is the least intrusive method I have found that still produces a real performance gain (over the past few days of looking into this).
This code passes all tests, fmt, and clippy (stable rust 1.72).
Relevant benchmarks (using
benches/hex.rs):This crate is a candidate to include
#![forbid(unsafe_code)], and I recommend doing so, but that is outside of the scope of this PR.