Cargo Clippy makes some good reccomendations

[blake3-team]

Cargo Clippy makes some good reccomendations

[oconnor663]

Yes, I've looked at the clippy warnings. Most of them refer to code blocks like this:

loadu(inputs[0].add(block_offset + 0 * 4 * DEGREE)),
loadu(inputs[1].add(block_offset + 0 * 4 * DEGREE)),
loadu(inputs[2].add(block_offset + 0 * 4 * DEGREE)),
loadu(inputs[3].add(block_offset + 0 * 4 * DEGREE)),
loadu(inputs[0].add(block_offset + 1 * 4 * DEGREE)),
loadu(inputs[1].add(block_offset + 1 * 4 * DEGREE)),
loadu(inputs[2].add(block_offset + 1 * 4 * DEGREE)),
loadu(inputs[3].add(block_offset + 1 * 4 * DEGREE)),
loadu(inputs[0].add(block_offset + 2 * 4 * DEGREE)),
loadu(inputs[1].add(block_offset + 2 * 4 * DEGREE)),
loadu(inputs[2].add(block_offset + 2 * 4 * DEGREE)),
loadu(inputs[3].add(block_offset + 2 * 4 * DEGREE)),
loadu(inputs[0].add(block_offset + 3 * 4 * DEGREE)),
loadu(inputs[1].add(block_offset + 3 * 4 * DEGREE)),
loadu(inputs[2].add(block_offset + 3 * 4 * DEGREE)),
loadu(inputs[3].add(block_offset + 3 * 4 * DEGREE)),

Inlining these sorts of things, rather than hoping that the compiler will unroll a loop, is often important for performance. Once you write it out inline, the version that keeps the 0 * ... and 1 * ... is a lot easier to read, because it all lines up, and you can see the pattern more clearly.

[Rudxain]

@oconnor663 I have a question about this. I've seen that src/portable.rs makes heavy use of manual/explicit loop unrolling, even for small loops that just use 4/6/8 iterations. Is there still a benefit to doing this nowadays? If so, I recommend using this crate