623ad6e025
Differential Revision: https://phabricator.services.mozilla.com/D24557 --HG-- extra : moz-landing-system : lando |
||
---|---|---|
.. | ||
benches | ||
examples | ||
src | ||
.cargo-checksum.json | ||
Cargo.toml | ||
LICENSE-APACHE | ||
LICENSE-MIT | ||
README.md | ||
appveyor.yml |
README.md
bit_reverse
Library Objective
This library provides a number of ways to compute the bit reversal of all primitive integers. There are currently 3 different algorithms implemented: Bitwise, Parallel, and Lookup reversal.
Example
use bit_reverse::ParallelReverse;
assert_eq!(0xA0u8.swap_bits(), 0x05u8);
This library is very simple to uses just import the crate and the algorithm you want to use.
Then you can call swap_bits
() on any primitive integer. If you want to try a different
algorithm just change the use statement and now your program will use the algorithm instead.
YMMV Performance Comparison
I wouldn't use BitwiseReverse
as it is mainly there for completeness and is strictly inferior
to ParallelReverse
, which is a Bitwise Parallel Reverse and thus an order of magnitude faster.
For small sizes, <= 16 bits, LookupReverse
is the fastest but it doesn't scale as well as
ParallelReverse
this is because ParallelReverse
does a constant number of operations for
every size (assuming your cpu has a hardware byte swap instruction). LookupReverse
needs more
lookups, ANDs, and ORs for each size increase. Thus ParallelReverse
performs a little better
at 32 bits and much better at 64 bits. These runtime characteristics are based on a Intel(R)
Core(TM) i7-4770K CPU @ 3.50GHz.
Memory Consumption
BitwiseReverse
and ParallelReverse
both only use a couple of stack variables for their
computations. LookupReverse
on the other hand statically allocates 256 u8s or 256 bytes to
do its computations. LookupReverse
's memory cost is shared by all of the types
'LookupReverse` supports.
no_std Compatible
To link to core instead of STD, disable default features for this library in your Cargo.toml. Cargo choosing features