From 1b6a04ca5504955c571d1c97504fb45ea0befee4 Mon Sep 17 00:00:00 2001 From: Valentin Popov Date: Mon, 8 Jan 2024 01:21:28 +0400 Subject: Initial vendor packages Signed-off-by: Valentin Popov --- vendor/byteorder/benches/bench.rs | 326 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 326 insertions(+) create mode 100644 vendor/byteorder/benches/bench.rs (limited to 'vendor/byteorder/benches/bench.rs') diff --git a/vendor/byteorder/benches/bench.rs b/vendor/byteorder/benches/bench.rs new file mode 100644 index 0000000..963251c --- /dev/null +++ b/vendor/byteorder/benches/bench.rs @@ -0,0 +1,326 @@ +#![feature(test)] + +extern crate test; + +macro_rules! bench_num { + ($name:ident, $read:ident, $bytes:expr, $data:expr) => { + mod $name { + use byteorder::{ + BigEndian, ByteOrder, LittleEndian, NativeEndian, + }; + use test::black_box as bb; + use test::Bencher; + + const NITER: usize = 100_000; + + #[bench] + fn read_big_endian(b: &mut Bencher) { + let buf = $data; + b.iter(|| { + for _ in 0..NITER { + bb(BigEndian::$read(&buf, $bytes)); + } + }); + } + + #[bench] + fn read_little_endian(b: &mut Bencher) { + let buf = $data; + b.iter(|| { + for _ in 0..NITER { + bb(LittleEndian::$read(&buf, $bytes)); + } + }); + } + + #[bench] + fn read_native_endian(b: &mut Bencher) { + let buf = $data; + b.iter(|| { + for _ in 0..NITER { + bb(NativeEndian::$read(&buf, $bytes)); + } + }); + } + } + }; + ($ty:ident, $max:ident, + $read:ident, $write:ident, $size:expr, $data:expr) => { + mod $ty { + use byteorder::{ + BigEndian, ByteOrder, LittleEndian, NativeEndian, + }; + use std::$ty; + use test::black_box as bb; + use test::Bencher; + + const NITER: usize = 100_000; + + #[bench] + fn read_big_endian(b: &mut Bencher) { + let buf = $data; + b.iter(|| { + for _ in 0..NITER { + bb(BigEndian::$read(&buf)); + } + }); + } + + #[bench] + fn read_little_endian(b: &mut Bencher) { + let buf = $data; + b.iter(|| { + for _ in 0..NITER { + bb(LittleEndian::$read(&buf)); + } + }); + } + + #[bench] + fn read_native_endian(b: &mut Bencher) { + let buf = $data; + b.iter(|| { + for _ in 0..NITER { + bb(NativeEndian::$read(&buf)); + } + }); + } + + #[bench] + fn write_big_endian(b: &mut Bencher) { + let mut buf = $data; + let n = $ty::$max; + b.iter(|| { + for _ in 0..NITER { + bb(BigEndian::$write(&mut buf, n)); + } + }); + } + + #[bench] + fn write_little_endian(b: &mut Bencher) { + let mut buf = $data; + let n = $ty::$max; + b.iter(|| { + for _ in 0..NITER { + bb(LittleEndian::$write(&mut buf, n)); + } + }); + } + + #[bench] + fn write_native_endian(b: &mut Bencher) { + let mut buf = $data; + let n = $ty::$max; + b.iter(|| { + for _ in 0..NITER { + bb(NativeEndian::$write(&mut buf, n)); + } + }); + } + } + }; +} + +bench_num!(u16, MAX, read_u16, write_u16, 2, [1, 2]); +bench_num!(i16, MAX, read_i16, write_i16, 2, [1, 2]); +bench_num!(u32, MAX, read_u32, write_u32, 4, [1, 2, 3, 4]); +bench_num!(i32, MAX, read_i32, write_i32, 4, [1, 2, 3, 4]); +bench_num!(u64, MAX, read_u64, write_u64, 8, [1, 2, 3, 4, 5, 6, 7, 8]); +bench_num!(i64, MAX, read_i64, write_i64, 8, [1, 2, 3, 4, 5, 6, 7, 8]); +bench_num!(f32, MAX, read_f32, write_f32, 4, [1, 2, 3, 4]); +bench_num!(f64, MAX, read_f64, write_f64, 8, [1, 2, 3, 4, 5, 6, 7, 8]); + +bench_num!(uint_1, read_uint, 1, [1]); +bench_num!(uint_2, read_uint, 2, [1, 2]); +bench_num!(uint_3, read_uint, 3, [1, 2, 3]); +bench_num!(uint_4, read_uint, 4, [1, 2, 3, 4]); +bench_num!(uint_5, read_uint, 5, [1, 2, 3, 4, 5]); +bench_num!(uint_6, read_uint, 6, [1, 2, 3, 4, 5, 6]); +bench_num!(uint_7, read_uint, 7, [1, 2, 3, 4, 5, 6, 7]); +bench_num!(uint_8, read_uint, 8, [1, 2, 3, 4, 5, 6, 7, 8]); + +bench_num!(int_1, read_int, 1, [1]); +bench_num!(int_2, read_int, 2, [1, 2]); +bench_num!(int_3, read_int, 3, [1, 2, 3]); +bench_num!(int_4, read_int, 4, [1, 2, 3, 4]); +bench_num!(int_5, read_int, 5, [1, 2, 3, 4, 5]); +bench_num!(int_6, read_int, 6, [1, 2, 3, 4, 5, 6]); +bench_num!(int_7, read_int, 7, [1, 2, 3, 4, 5, 6, 7]); +bench_num!(int_8, read_int, 8, [1, 2, 3, 4, 5, 6, 7, 8]); + +bench_num!( + u128, + MAX, + read_u128, + write_u128, + 16, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] +); +bench_num!( + i128, + MAX, + read_i128, + write_i128, + 16, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] +); + +bench_num!(uint128_1, read_uint128, 1, [1]); +bench_num!(uint128_2, read_uint128, 2, [1, 2]); +bench_num!(uint128_3, read_uint128, 3, [1, 2, 3]); +bench_num!(uint128_4, read_uint128, 4, [1, 2, 3, 4]); +bench_num!(uint128_5, read_uint128, 5, [1, 2, 3, 4, 5]); +bench_num!(uint128_6, read_uint128, 6, [1, 2, 3, 4, 5, 6]); +bench_num!(uint128_7, read_uint128, 7, [1, 2, 3, 4, 5, 6, 7]); +bench_num!(uint128_8, read_uint128, 8, [1, 2, 3, 4, 5, 6, 7, 8]); +bench_num!(uint128_9, read_uint128, 9, [1, 2, 3, 4, 5, 6, 7, 8, 9]); +bench_num!(uint128_10, read_uint128, 10, [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]); +bench_num!(uint128_11, read_uint128, 11, [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]); +bench_num!( + uint128_12, + read_uint128, + 12, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12] +); +bench_num!( + uint128_13, + read_uint128, + 13, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13] +); +bench_num!( + uint128_14, + read_uint128, + 14, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] +); +bench_num!( + uint128_15, + read_uint128, + 15, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15] +); +bench_num!( + uint128_16, + read_uint128, + 16, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] +); + +bench_num!(int128_1, read_int128, 1, [1]); +bench_num!(int128_2, read_int128, 2, [1, 2]); +bench_num!(int128_3, read_int128, 3, [1, 2, 3]); +bench_num!(int128_4, read_int128, 4, [1, 2, 3, 4]); +bench_num!(int128_5, read_int128, 5, [1, 2, 3, 4, 5]); +bench_num!(int128_6, read_int128, 6, [1, 2, 3, 4, 5, 6]); +bench_num!(int128_7, read_int128, 7, [1, 2, 3, 4, 5, 6, 7]); +bench_num!(int128_8, read_int128, 8, [1, 2, 3, 4, 5, 6, 7, 8]); +bench_num!(int128_9, read_int128, 9, [1, 2, 3, 4, 5, 6, 7, 8, 9]); +bench_num!(int128_10, read_int128, 10, [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]); +bench_num!(int128_11, read_int128, 11, [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]); +bench_num!( + int128_12, + read_int128, + 12, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12] +); +bench_num!( + int128_13, + read_int128, + 13, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13] +); +bench_num!( + int128_14, + read_int128, + 14, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] +); +bench_num!( + int128_15, + read_int128, + 15, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15] +); +bench_num!( + int128_16, + read_int128, + 16, + [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] +); + +macro_rules! bench_slice { + ($name:ident, $numty:ty, $read:ident, $write:ident) => { + mod $name { + use std::mem::size_of; + + use byteorder::{BigEndian, ByteOrder, LittleEndian}; + use rand::distributions; + use rand::{self, Rng}; + use test::Bencher; + + #[bench] + fn read_big_endian(b: &mut Bencher) { + let mut numbers: Vec<$numty> = rand::thread_rng() + .sample_iter(&distributions::Standard) + .take(100000) + .collect(); + let mut bytes = vec![0; numbers.len() * size_of::<$numty>()]; + BigEndian::$write(&numbers, &mut bytes); + + b.bytes = bytes.len() as u64; + b.iter(|| { + BigEndian::$read(&bytes, &mut numbers); + }); + } + + #[bench] + fn read_little_endian(b: &mut Bencher) { + let mut numbers: Vec<$numty> = rand::thread_rng() + .sample_iter(&distributions::Standard) + .take(100000) + .collect(); + let mut bytes = vec![0; numbers.len() * size_of::<$numty>()]; + LittleEndian::$write(&numbers, &mut bytes); + + b.bytes = bytes.len() as u64; + b.iter(|| { + LittleEndian::$read(&bytes, &mut numbers); + }); + } + + #[bench] + fn write_big_endian(b: &mut Bencher) { + let numbers: Vec<$numty> = rand::thread_rng() + .sample_iter(&distributions::Standard) + .take(100000) + .collect(); + let mut bytes = vec![0; numbers.len() * size_of::<$numty>()]; + + b.bytes = bytes.len() as u64; + b.iter(|| { + BigEndian::$write(&numbers, &mut bytes); + }); + } + + #[bench] + fn write_little_endian(b: &mut Bencher) { + let numbers: Vec<$numty> = rand::thread_rng() + .sample_iter(&distributions::Standard) + .take(100000) + .collect(); + let mut bytes = vec![0; numbers.len() * size_of::<$numty>()]; + + b.bytes = bytes.len() as u64; + b.iter(|| { + LittleEndian::$write(&numbers, &mut bytes); + }); + } + } + }; +} + +bench_slice!(slice_u16, u16, read_u16_into, write_u16_into); +bench_slice!(slice_u64, u64, read_u64_into, write_u64_into); +bench_slice!(slice_i64, i64, read_i64_into, write_i64_into); -- cgit v1.2.3