rewrite of shootout-fasta.rs

[TeXitoi]

improvements:

• no managed box
• no virtual calls
• no useless copy
• optimizations (bisect is slower, limit tests, BufferedWriter...)
• pass shootout test
• 10 times faster

[emberian]

Please change 'self to 'a or some other name; it is no longer allowed on master.

[huonw]

I think you can write this function as:

letmut cum = 0.0;let data = aa.iter().map(|&(ch, p)| { cum += p;(cum, ch asu8)}).collect();AAGen{rng: rng,data: data }

(with line breaking to taste.)

[TeXitoi]

With all the comment and using integer arithmetic for a big speedup (should be about as fast as the best official implementation).

[huonw]

Why can't this just be LINE_LENGTH + 1?

[TeXitoi]

The previous commit does not include the integer arithmetic, this one is better.

If we want safe code, we can remove it at the cost of 5% here.

[TeXitoi]

Finally, unsafe_set is not visible, but the zip version by @huonw is approx 7% slower. So no unsafe code, but a quite ugly loop with indexing.

[TeXitoi]

@huonw version discussed in the outdated diff is basically for a in line.mut_iter().zip(it.by_ref()).take(nb) {…. It is 7% slower.

[huonw]

(cc @alexcrichton ^, since, iirc, one doesn't get notifications from the main PR after commenting on an individual commit.)

[alexcrichton]

huh, who knew!