speed up the tests to 249MB/s (!!) by copying over the MyMemWriter1

for some reason, having MyMemWriter1 in scope makes everything much
faster.
This commit is contained in:
Erick Tryzelaar 2014-09-07 01:19:36 -07:00
parent bb361086b3
commit 6b2adccf60
2 changed files with 282 additions and 93 deletions

View File

@ -563,7 +563,7 @@ impl MyMemWriter1 {
}
// LLVM isn't yet able to lower `Vec::push_all` into a memcpy, so this helps
// MemWriter eke out that last bit of performance.
// MemWriter eke out that last bit of performance.
//#[inline(always)]
fn push_all_bytes(dst: &mut Vec<u8>, src: &[u8]) {
let dst_len = dst.len();

View File

@ -1,19 +1,19 @@
#![allow(non_camel_case_types)]
#![feature(phase, macro_rules)]
#![allow(non_camel_case_types)]
#[phase(plugin)]
extern crate serde2_macros;
extern crate serde2;
extern crate serialize;
extern crate test;
#[phase(plugin)]
extern crate serde2_macros;
//use std::io;
//use std::io::MemWriter;
use std::io;
use std::io::MemWriter;
use test::Bencher;
use serde2::json;
//use serde2::de;
use serde2::json;
use serde2::Serialize;
use serde2::ser;
@ -445,7 +445,7 @@ impl<D: de::Deserializer<E>, E> de::Deserializable<D, E> for Country {
*/
#[deriving(Encodable, Decodable)]
//#[deriving_serializable]
#[deriving_serializable]
//#[deriving_deserializable]
struct Log {
timestamp: i64,
@ -497,82 +497,274 @@ impl Log {
}
impl <S: ::serde2::VisitorState<R>, R> ::serde2::Serialize<S, R> for Log {
#[inline]
fn serialize(&self, s: &mut S) -> R {
struct Log3033<'a> {
value: &'a Log,
state: u8,
}
impl<'a, S: ::serde2::VisitorState<R>, R> ::serde2::Visitor<S, R> for Log3033<'a> {
#[inline]
fn visit(&mut self, s: &mut S) -> Option<R> {
match self.state {
0 => {
self.state += 1;
Some(s.visit_map_elt(true, "timestamp",
&self.value.timestamp))
}
1 => {
self.state += 1;
Some(s.visit_map_elt(false, "zone_id", &self.value.zone_id))
}
2 => {
self.state += 1;
Some(s.visit_map_elt(false, "zone_plan",
&self.value.zone_plan))
}
3 => {
self.state += 1;
Some(s.visit_map_elt(false, "http", &self.value.http))
}
4 => {
self.state += 1;
Some(s.visit_map_elt(false, "origin", &self.value.origin))
}
5 => {
self.state += 1;
Some(s.visit_map_elt(false, "country", &self.value.country))
}
6 => {
self.state += 1;
Some(s.visit_map_elt(false, "cache_status",
&self.value.cache_status))
}
7 => {
self.state += 1;
Some(s.visit_map_elt(false, "server_ip",
&self.value.server_ip))
}
8 => {
self.state += 1;
Some(s.visit_map_elt(false, "server_name",
&self.value.server_name))
}
9 => {
self.state += 1;
Some(s.visit_map_elt(false, "remote_ip",
&self.value.remote_ip))
}
10 => {
self.state += 1;
Some(s.visit_map_elt(false, "bytes_dlv",
&self.value.bytes_dlv))
}
11 => {
self.state += 1;
Some(s.visit_map_elt(false, "ray_id", &self.value.ray_id))
}
_ => None,
}
macro_rules! likely(
($val:expr) => {
{
extern {
#[link_name = "llvm.expect.i8"]
fn expect(val: u8, expected_val: u8) -> u8;
}
let x: bool = $val;
unsafe { expect(x as u8, 1) != 0 }
}
}
)
s.visit_named_map("Log", Log3033{value: self, state: 0,})
macro_rules! unlikely(
($val:expr) => {
{
extern {
#[link_name = "llvm.expect.i8"]
fn expect(val: u8, expected_val: u8) -> u8;
}
let x: bool = $val;
unsafe { expect(x as u8, 0) != 0 }
}
}
)
struct MyMemWriter0 {
buf: Vec<u8>,
}
impl MyMemWriter0 {
/*
pub fn new() -> MyMemWriter0 {
MyMemWriter0::with_capacity(128)
}
*/
pub fn with_capacity(cap: uint) -> MyMemWriter0 {
MyMemWriter0 {
buf: Vec::with_capacity(cap)
}
}
#[inline]
pub fn unwrap(self) -> Vec<u8> { self.buf }
}
impl Writer for MyMemWriter0 {
#[inline]
fn write(&mut self, buf: &[u8]) -> io::IoResult<()> {
self.buf.push_all(buf);
Ok(())
}
}
struct MyMemWriter1 {
buf: Vec<u8>,
}
impl MyMemWriter1 {
/*
pub fn new() -> MyMemWriter1 {
MyMemWriter1::with_capacity(128)
}
*/
pub fn with_capacity(cap: uint) -> MyMemWriter1 {
MyMemWriter1 {
buf: Vec::with_capacity(cap)
}
}
#[inline]
pub fn unwrap(self) -> Vec<u8> { self.buf }
}
// LLVM isn't yet able to lower `Vec::push_all` into a memcpy, so this helps
// MemWriter eke out that last bit of performance.
//#[inline(always)]
fn push_all_bytes(dst: &mut Vec<u8>, src: &[u8]) {
let dst_len = dst.len();
let src_len = src.len();
dst.reserve_additional(src_len);
unsafe {
// we would have failed if `reserve_additional` overflowed.
dst.set_len(dst_len + src_len);
::std::ptr::copy_nonoverlapping_memory(
dst.as_mut_ptr().offset(dst_len as int),
src.as_ptr(),
src_len);
}
}
impl Writer for MyMemWriter1 {
#[inline]
fn write(&mut self, buf: &[u8]) -> io::IoResult<()> {
push_all_bytes(&mut self.buf, buf);
Ok(())
}
}
pub struct MyMemWriter2 {
buf: Vec<u8>,
pos: uint,
}
impl MyMemWriter2 {
/// Create a new `MemWriter`.
#[inline]
pub fn new() -> MyMemWriter2 {
MyMemWriter2::with_capacity(128)
}
/// Create a new `MemWriter`, allocating at least `n` bytes for
/// the internal buffer.
#[inline]
pub fn with_capacity(n: uint) -> MyMemWriter2 {
MyMemWriter2 { buf: Vec::with_capacity(n), pos: 0 }
}
/// Acquires an immutable reference to the underlying buffer of this
/// `MemWriter`.
///
/// No method is exposed for acquiring a mutable reference to the buffer
/// because it could corrupt the state of this `MemWriter`.
#[inline]
pub fn get_ref<'a>(&'a self) -> &'a [u8] { self.buf.as_slice() }
/// Unwraps this `MemWriter`, returning the underlying buffer
#[inline]
pub fn unwrap(self) -> Vec<u8> { self.buf }
}
impl Writer for MyMemWriter2 {
#[inline]
fn write(&mut self, buf: &[u8]) -> io::IoResult<()> {
if self.pos != self.buf.len() {
push_all_bytes(&mut self.buf, buf);
} else {
// Make sure the internal buffer is as least as big as where we
// currently are
let difference = self.pos as i64 - self.buf.len() as i64;
if difference > 0 {
self.buf.grow(difference as uint, &0);
}
// Figure out what bytes will be used to overwrite what's currently
// there (left), and what will be appended on the end (right)
let cap = self.buf.len() - self.pos;
let (left, right) = if cap <= buf.len() {
(buf.slice_to(cap), buf.slice_from(cap))
} else {
(buf, &[])
};
// Do the necessary writes
if left.len() > 0 {
::std::slice::bytes::copy_memory(self.buf.mut_slice_from(self.pos), left);
}
if right.len() > 0 {
push_all_bytes(&mut self.buf, right);
}
}
// Bump us forward
self.pos += buf.len();
Ok(())
}
}
fn combine(seek: io::SeekStyle, cur: uint, end: uint, offset: i64) -> io::IoResult<u64> {
// compute offset as signed and clamp to prevent overflow
let pos = match seek {
io::SeekSet => 0,
io::SeekEnd => end,
io::SeekCur => cur,
} as i64;
if offset + pos < 0 {
Err(io::IoError {
kind: io::InvalidInput,
desc: "invalid seek to a negative offset",
detail: None
})
} else {
Ok((offset + pos) as u64)
}
}
impl Seek for MyMemWriter2 {
#[inline]
fn tell(&self) -> io::IoResult<u64> { Ok(self.pos as u64) }
#[inline]
fn seek(&mut self, pos: i64, style: io::SeekStyle) -> io::IoResult<()> {
let new = try!(combine(style, self.pos, self.buf.len(), pos));
self.pos = new as uint;
Ok(())
}
}
pub struct MyMemWriter3 {
buf: Vec<u8>,
pos: uint,
}
impl MyMemWriter3 {
/// Create a new `MemWriter`.
#[inline]
pub fn new() -> MyMemWriter3 {
MyMemWriter3::with_capacity(128)
}
/// Create a new `MemWriter`, allocating at least `n` bytes for
/// the internal buffer.
#[inline]
pub fn with_capacity(n: uint) -> MyMemWriter3 {
MyMemWriter3 { buf: Vec::with_capacity(n), pos: 0 }
}
/// Acquires an immutable reference to the underlying buffer of this
/// `MemWriter`.
///
/// No method is exposed for acquiring a mutable reference to the buffer
/// because it could corrupt the state of this `MemWriter`.
#[inline]
pub fn get_ref<'a>(&'a self) -> &'a [u8] { self.buf.as_slice() }
/// Unwraps this `MemWriter`, returning the underlying buffer
#[inline]
pub fn unwrap(self) -> Vec<u8> { self.buf }
}
impl Writer for MyMemWriter3 {
#[inline]
fn write(&mut self, buf: &[u8]) -> io::IoResult<()> {
// Make sure the internal buffer is as least as big as where we
// currently are
let difference = self.pos as i64 - self.buf.len() as i64;
if difference > 0 {
self.buf.grow(difference as uint, &0);
}
// Figure out what bytes will be used to overwrite what's currently
// there (left), and what will be appended on the end (right)
let cap = self.buf.len() - self.pos;
let (left, right) = if cap <= buf.len() {
(buf.slice_to(cap), buf.slice_from(cap))
} else {
(buf, &[])
};
// Do the necessary writes
if left.len() > 0 {
::std::slice::bytes::copy_memory(self.buf.mut_slice_from(self.pos), left);
}
if right.len() > 0 {
push_all_bytes(&mut self.buf, right);
}
// Bump us forward
self.pos += buf.len();
Ok(())
}
}
#[bench]
fn bench_encoder(b: &mut Bencher) {
@ -600,7 +792,6 @@ fn bench_serializer(b: &mut Bencher) {
});
}
/*
#[bench]
fn bench_copy(b: &mut Bencher) {
let s = r#"{"timestamp":2837513946597,"zone_id":123456,"zone_plan":"FREE","http":{"protocol":"HTTP11","status":200,"host_status":503,"up_status":520,"method":"GET","content_type":"text/html","user_agent":"Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/33.0.1750.146 Safari/537.36","referer":"https://www.cloudflare.com/","request_uri":"/cdn-cgi/trace"},"origin":{"ip":"1.2.3.4","port":8000,"hostname":"www.example.com","protocol":"HTTPS"},"country":"US","cache_status":"Hit","server_ip":"192.168.1.1","server_name":"metal.cloudflare.com","remote_ip":"10.1.2.3","bytes_dlv":123456,"ray_id":"10c73629cce30078-LAX"}"#;
@ -787,7 +978,6 @@ fn bench_manual_mem_writer_no_escape(b: &mut Bencher) {
b.iter(|| {
let mut wr = MemWriter::with_capacity(1024);
manual_no_escape(wr.by_ref(), &log);
let _json = wr.unwrap();
//let _json = String::from_utf8(_json).unwrap();
@ -987,14 +1177,15 @@ fn bench_manual_my_mem_writer3_escape(b: &mut Bencher) {
});
}
/*
fn direct<W: Writer>(wr: W, log: &Log) {
use ser::VisitorState;
use serde2::ser::VisitorState;
let mut serializer = json::Serializer::new(wr);
serializer.visit_struct_start("Log", 12).unwrap();
serializer.serialize_struct_start("Log", 12).unwrap();
serializer.visit_struct_elt("timestamp", &log.timestamp).unwrap();
serializer.visit_struct_elt("zone_id", &log.zone_id).unwrap();
serializer.serialize_struct_elt("timestamp", &log.timestamp).unwrap();
serializer.serialize_struct_elt("zone_id", &log.zone_id).unwrap();
serializer.serialize_struct_elt("zone_plan", &(log.zone_plan as uint)).unwrap();
serializer.serialize_struct_start("Http", 9).unwrap();
@ -1040,10 +1231,8 @@ fn bench_direct_mem_writer(b: &mut Bencher) {
direct(wr.by_ref(), &log);
let _json = wr.unwrap();
//let _json = String::from_utf8(wr.unwrap()).unwrap();
/*
assert_eq!(_s, _json.as_slice());
*/
//let _json = String::from_utf8(_json).unwrap();
//assert_eq!(_s, _json.as_slice());
});
}
@ -1061,13 +1250,13 @@ fn bench_direct_my_mem_writer0(b: &mut Bencher) {
direct(wr.by_ref(), &log);
let _json = wr.unwrap();
//let _json = String::from_utf8(wr.unwrap()).unwrap();
/*
assert_eq!(_s, _json.as_slice());
*/
//let _json = String::from_utf8(_json).unwrap();
//assert_eq!(_s, _json.as_slice());
});
}
*/
/*
#[bench]
fn bench_decoder(b: &mut Bencher) {
let s = r#"{"timestamp":2837513946597,"zone_id":123456,"zone_plan":"FREE","http":{"protocol":"HTTP11","status":200,"host_status":503,"up_status":520,"method":"GET","content_type":"text/html","user_agent":"Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/33.0.1750.146 Safari/537.36","referer":"https://www.cloudflare.com/","request_uri":"/cdn-cgi/trace"},"origin":{"ip":"1.2.3.4","port":8000,"hostname":"www.example.com","protocol":"HTTPS"},"country":"US","cache_status":"Hit","server_ip":"192.168.1.1","server_name":"metal.cloudflare.com","remote_ip":"10.1.2.3","bytes_dlv":123456,"ray_id":"10c73629cce30078-LAX"}"#;