mirror of
https://github.com/actix/actix-web.git
synced 2024-12-30 03:50:42 +00:00
add safe vs unsafe benchmarks
This commit is contained in:
Rob Ede
parent
82b2786d6b
commit
31a3515e90
2 changed files with 284 additions and 0 deletions
|
|
@ -93,8 +93,13 @@ actix-server = "1.0.0"
|
|||
actix-connect = { version = "1.0.0", features=["openssl"] }
|
||||
actix-http-test = { version = "1.0.0", features=["openssl"] }
|
||||
actix-tls = { version = "1.0.0", features=["openssl"] }
|
||||
criterion = "0.3"
|
||||
futures = "0.3.1"
|
||||
env_logger = "0.6"
|
||||
serde_derive = "1.0"
|
||||
open-ssl = { version="0.10", package = "openssl" }
|
||||
rust-tls = { version="0.16", package = "rustls" }
|
||||
|
||||
[[bench]]
|
||||
name = "content-length"
|
||||
harness = false
|
||||
|
|
|
|||
279
actix-http/benches/content-length.rs
Normal file
279
actix-http/benches/content-length.rs
Normal file
|
|
@ -0,0 +1,279 @@
|
|||
use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
|
||||
|
||||
use bytes::BytesMut;
|
||||
|
||||
// benchmark sending all requests at the same time
|
||||
fn bench_write_content_length(c: &mut Criterion) {
|
||||
let mut group = c.benchmark_group("write_content_length");
|
||||
|
||||
let sizes = [
|
||||
0, 1, 11, 83, 101, 653, 1001, 6323, 10001, 56329, 100001, 123456, 98724245,
|
||||
4294967202,
|
||||
];
|
||||
|
||||
for i in sizes.iter() {
|
||||
group.bench_with_input(BenchmarkId::new("Original (unsafe)", i), i, |b, &i| {
|
||||
b.iter(|| {
|
||||
let mut b = BytesMut::with_capacity(35);
|
||||
_original::write_content_length(i, &mut b)
|
||||
})
|
||||
});
|
||||
|
||||
group.bench_with_input(BenchmarkId::new("New (safe)", i), i, |b, &i| {
|
||||
b.iter(|| {
|
||||
let mut b = BytesMut::with_capacity(35);
|
||||
_new::write_content_length(i, &mut b)
|
||||
})
|
||||
});
|
||||
}
|
||||
|
||||
group.finish();
|
||||
}
|
||||
|
||||
criterion_group!(benches, bench_write_content_length);
|
||||
criterion_main!(benches);
|
||||
|
||||
mod _new {
|
||||
use std::{ptr, slice};
|
||||
|
||||
use bytes::{BufMut, BytesMut};
|
||||
|
||||
const DEC_DIGITS_LUT: &[u8] = b"0001020304050607080910111213141516171819\
|
||||
2021222324252627282930313233343536373839\
|
||||
4041424344454647484950515253545556575859\
|
||||
6061626364656667686970717273747576777879\
|
||||
8081828384858687888990919293949596979899";
|
||||
|
||||
const DIGITS_START: u8 = b'0';
|
||||
|
||||
/// NOTE: bytes object has to contain enough space
|
||||
pub fn write_content_length(n: usize, bytes: &mut BytesMut) {
|
||||
if n == 0 {
|
||||
bytes.put_slice(b"\r\ncontent-length: 0\r\n");
|
||||
return;
|
||||
}
|
||||
|
||||
bytes.put_slice(b"\r\ncontent-length: ");
|
||||
|
||||
if n < 10 {
|
||||
bytes.put_u8(DIGITS_START + (n as u8));
|
||||
} else if n < 100 {
|
||||
let n = n as u8;
|
||||
|
||||
let d10 = n / 10;
|
||||
let d1 = n % 10;
|
||||
|
||||
bytes.put_u8(DIGITS_START + d10);
|
||||
bytes.put_u8(DIGITS_START + d1);
|
||||
} else if n < 1000 {
|
||||
let n = n as u16;
|
||||
|
||||
let d100 = (n / 100) as u8;
|
||||
let d10 = ((n / 10) % 10) as u8;
|
||||
let d1 = (n % 10) as u8;
|
||||
|
||||
bytes.put_u8(DIGITS_START + d100);
|
||||
bytes.put_u8(DIGITS_START + d10);
|
||||
bytes.put_u8(DIGITS_START + d1);
|
||||
} else if n < 10000 {
|
||||
let n = n as u16;
|
||||
|
||||
let d1000 = (n / 1000) as u8;
|
||||
let d100 = ((n / 100) % 10) as u8;
|
||||
let d10 = ((n / 10) % 10) as u8;
|
||||
let d1 = (n % 10) as u8;
|
||||
|
||||
bytes.put_u8(DIGITS_START + d1000);
|
||||
bytes.put_u8(DIGITS_START + d100);
|
||||
bytes.put_u8(DIGITS_START + d10);
|
||||
bytes.put_u8(DIGITS_START + d1);
|
||||
} else {
|
||||
write_usize(n, bytes);
|
||||
}
|
||||
|
||||
bytes.put_slice(b"\r\n");
|
||||
}
|
||||
|
||||
pub(crate) fn write_usize(mut n: usize, bytes: &mut BytesMut) {
|
||||
let mut curr: isize = 39;
|
||||
let mut buf = [0u8; 39];
|
||||
|
||||
let buf_ptr = buf.as_mut_ptr();
|
||||
let lut_ptr = DEC_DIGITS_LUT.as_ptr();
|
||||
|
||||
// eagerly decode 4 characters at a time
|
||||
while n >= 10_000 {
|
||||
let rem = (n % 10_000) as isize;
|
||||
n /= 10_000;
|
||||
|
||||
let d1 = (rem / 100) << 1;
|
||||
let d2 = (rem % 100) << 1;
|
||||
curr -= 4;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
|
||||
ptr::copy_nonoverlapping(
|
||||
lut_ptr.offset(d2),
|
||||
buf_ptr.offset(curr + 2),
|
||||
2,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
// if we reach here numbers are <= 9999, so at most 4 chars long
|
||||
let mut n = n as isize; // possibly reduce 64bit math
|
||||
|
||||
// decode 2 more chars, if > 2 chars
|
||||
if n >= 100 {
|
||||
let d1 = (n % 100) << 1;
|
||||
n /= 100;
|
||||
curr -= 2;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
|
||||
}
|
||||
}
|
||||
|
||||
// decode last 1 or 2 chars
|
||||
if n < 10 {
|
||||
curr -= 1;
|
||||
unsafe {
|
||||
*buf_ptr.offset(curr) = (n as u8) + b'0';
|
||||
}
|
||||
} else {
|
||||
let d1 = n << 1;
|
||||
curr -= 2;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
|
||||
}
|
||||
}
|
||||
|
||||
unsafe {
|
||||
bytes.extend_from_slice(slice::from_raw_parts(
|
||||
buf_ptr.offset(curr),
|
||||
39 - curr as usize,
|
||||
));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
mod _original {
|
||||
use std::{mem, ptr, slice};
|
||||
|
||||
use bytes::{BufMut, BytesMut};
|
||||
|
||||
const DEC_DIGITS_LUT: &[u8] = b"0001020304050607080910111213141516171819\
|
||||
2021222324252627282930313233343536373839\
|
||||
4041424344454647484950515253545556575859\
|
||||
6061626364656667686970717273747576777879\
|
||||
8081828384858687888990919293949596979899";
|
||||
|
||||
/// NOTE: bytes object has to contain enough space
|
||||
pub fn write_content_length(mut n: usize, bytes: &mut BytesMut) {
|
||||
if n < 10 {
|
||||
let mut buf: [u8; 21] = [
|
||||
b'\r', b'\n', b'c', b'o', b'n', b't', b'e', b'n', b't', b'-', b'l',
|
||||
b'e', b'n', b'g', b't', b'h', b':', b' ', b'0', b'\r', b'\n',
|
||||
];
|
||||
buf[18] = (n as u8) + b'0';
|
||||
bytes.put_slice(&buf);
|
||||
} else if n < 100 {
|
||||
let mut buf: [u8; 22] = [
|
||||
b'\r', b'\n', b'c', b'o', b'n', b't', b'e', b'n', b't', b'-', b'l',
|
||||
b'e', b'n', b'g', b't', b'h', b':', b' ', b'0', b'0', b'\r', b'\n',
|
||||
];
|
||||
let d1 = n << 1;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(
|
||||
DEC_DIGITS_LUT.as_ptr().add(d1),
|
||||
buf.as_mut_ptr().offset(18),
|
||||
2,
|
||||
);
|
||||
}
|
||||
bytes.put_slice(&buf);
|
||||
} else if n < 1000 {
|
||||
let mut buf: [u8; 23] = [
|
||||
b'\r', b'\n', b'c', b'o', b'n', b't', b'e', b'n', b't', b'-', b'l',
|
||||
b'e', b'n', b'g', b't', b'h', b':', b' ', b'0', b'0', b'0', b'\r',
|
||||
b'\n',
|
||||
];
|
||||
// decode 2 more chars, if > 2 chars
|
||||
let d1 = (n % 100) << 1;
|
||||
n /= 100;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(
|
||||
DEC_DIGITS_LUT.as_ptr().add(d1),
|
||||
buf.as_mut_ptr().offset(19),
|
||||
2,
|
||||
)
|
||||
};
|
||||
|
||||
// decode last 1
|
||||
buf[18] = (n as u8) + b'0';
|
||||
|
||||
bytes.put_slice(&buf);
|
||||
} else {
|
||||
bytes.put_slice(b"\r\ncontent-length: ");
|
||||
convert_usize(n, bytes);
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn convert_usize(mut n: usize, bytes: &mut BytesMut) {
|
||||
let mut curr: isize = 39;
|
||||
let mut buf: [u8; 41] = unsafe { mem::MaybeUninit::uninit().assume_init() };
|
||||
buf[39] = b'\r';
|
||||
buf[40] = b'\n';
|
||||
let buf_ptr = buf.as_mut_ptr();
|
||||
let lut_ptr = DEC_DIGITS_LUT.as_ptr();
|
||||
|
||||
// eagerly decode 4 characters at a time
|
||||
while n >= 10_000 {
|
||||
let rem = (n % 10_000) as isize;
|
||||
n /= 10_000;
|
||||
|
||||
let d1 = (rem / 100) << 1;
|
||||
let d2 = (rem % 100) << 1;
|
||||
curr -= 4;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
|
||||
ptr::copy_nonoverlapping(
|
||||
lut_ptr.offset(d2),
|
||||
buf_ptr.offset(curr + 2),
|
||||
2,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
// if we reach here numbers are <= 9999, so at most 4 chars long
|
||||
let mut n = n as isize; // possibly reduce 64bit math
|
||||
|
||||
// decode 2 more chars, if > 2 chars
|
||||
if n >= 100 {
|
||||
let d1 = (n % 100) << 1;
|
||||
n /= 100;
|
||||
curr -= 2;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
|
||||
}
|
||||
}
|
||||
|
||||
// decode last 1 or 2 chars
|
||||
if n < 10 {
|
||||
curr -= 1;
|
||||
unsafe {
|
||||
*buf_ptr.offset(curr) = (n as u8) + b'0';
|
||||
}
|
||||
} else {
|
||||
let d1 = n << 1;
|
||||
curr -= 2;
|
||||
unsafe {
|
||||
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
|
||||
}
|
||||
}
|
||||
|
||||
unsafe {
|
||||
bytes.extend_from_slice(slice::from_raw_parts(
|
||||
buf_ptr.offset(curr),
|
||||
41 - curr as usize,
|
||||
));
|
||||
}
|
||||
}
|
||||
}
|
||||
Loading…
Reference in a new issue