perf: mark dna::complement as inline

[kloetzl]

This speeds up computing the complement on a long string by ~2.3x. Not sure why rust doesn't inline such a trivial function on its own.

I didn't manage to compile your benchmarks, hence I can only point to the results from my own project.

dna4_revcomp time: [688.18 µs 690.09 µs 692.27 µs]
dnax_revcomp time: [3.0615 ms 3.0705 ms 3.0815 ms]

rust-bio time: [12.029 ms 12.105 ms 12.212 ms]
rust-bio inlined time: [5.0742 ms 5.0901 ms 5.1087 ms]

There is still a small overhead of Rust over C. I think Rust repeatedly checks that a as usize is a valid index in COMPLEMENT. While that is usually not a problem, it is here, because the loop is so small and the extra instructions clog the pipeline.

[coveralls]

Coverage: 80.38% (+0.0%) from 80.38% when pulling f43f02b on kloetzl:patch-1 into e7f5744 on rust-bio:master.

[dcroote]

Thanks for the contribution, this makes sense. Can you also do the same for the complement function in src/alphabet/rna.rs?

I didn't manage to compile your benchmarks

Benchmarks require using nightly - did you try that? It'd be nice to add an alphabets.rs file to the benchmarks directory for testing some of these string functions like complement and revcomp.

[kloetzl]

Can you also do the same for the complement function in src/alphabet/rna.rs?

Sure, will do.

Benchmarks require using nightly - did you try that?

Didn't know that. Will give it a go.

[dcroote]

Thanks but it looks like you added #[inline] to revcomp instead of complement in rna.rs

[kloetzl]

Thanks but it looks like you added #[inline] to revcomp instead of complement in rna.rs

Thanks for the thorough review. I blame my lingering cold.

[dcroote]

Thanks!

[kloetzl]

Thanks for merging!