libcollection : Switches from uint to u32 in BitV and BitVSet

[gamazeps]

Closes #16736
linked to #18009

[Gankra]

Did you run the benches (PLEASE_BENCH=1) on this? I'd be interested to see if there were perf consequences on 64-bit platforms.

CC @apoelstra

[gamazeps]

Yup, here are the result with u32

test bitv::tests::bench_bitv_big_iter ... bench: 17185 ns/iter (+/- 634)
test bitv::tests::bench_bitv_big_union ... bench: 1017 ns/iter (+/- 22)
test bitv::tests::bench_bitv_set_big_fixed ... bench: 681 ns/iter (+/- 5)
test bitv::tests::bench_bitv_set_big_variable ... bench: 1694 ns/iter (+/- 25)
test bitv::tests::bench_bitv_set_small ... bench: 643 ns/iter (+/- 11)
test bitv::tests::bench_bitv_small_iter ... bench: 474 ns/iter (+/- 23)
test bitv::tests::bench_bitvset_big ... bench: 993 ns/iter (+/- 5)
test bitv::tests::bench_bitvset_iter ... bench: 25473 ns/iter (+/- 726)
test bitv::tests::bench_bitvset_small ... bench: 989 ns/iter (+/- 14)
test bitv::tests::bench_uint_small ... bench: 572 ns/iter (+/- 7)

and the results with uint
test bitv::tests::bench_bitv_big_iter ... bench: 17347 ns/iter (+/- 650)
test bitv::tests::bench_bitv_big_union ... bench: 519 ns/iter (+/- 8)
test bitv::tests::bench_bitv_set_big_fixed ... bench: 685 ns/iter (+/- 10)
test bitv::tests::bench_bitv_set_big_variable ... bench: 1713 ns/iter (+/- 26)
test bitv::tests::bench_bitv_set_small ... bench: 674 ns/iter (+/- 11)
test bitv::tests::bench_bitv_small_iter ... bench: 782 ns/iter (+/- 22)
test bitv::tests::bench_bitvset_big ... bench: 1069 ns/iter (+/- 10)
test bitv::tests::bench_bitvset_iter ... bench: 40832 ns/iter (+/- 7077)
test bitv::tests::bench_bitvset_small ... bench: 1065 ns/iter (+/- 30)
test bitv::tests::bench_uint_small ... bench: 602 ns/iter (+/- 5)

u32 are faster for everything but unions, for which they are much slower.
I think it's a good tradeoff, what's your opinion on it ?

cc @thestinger @aturon

[Gankra]

Looks good to me! Thanks a bunch!

[gamazeps]

r? @aturon