Skip to content
Permalink
Browse files

Faster Base64 Encoding for NIOWebsocket connections (#1328)

Motivation:

In performance tests was shown that the current base 64 encoding function doesn’t perform very well.

#1322

Modifications:

This commit replaces the current Base64 implementation with a more performant one (by using UInt8 bytes instead of Unicode.Scalars).

Result:

The API will stay the same. Base64 encoding is just faster.
  • Loading branch information
fabianfett authored and Lukasa committed Dec 24, 2019
1 parent f1a7acc commit 653492753c817bdb5a4fe28370819f1a7450c90d
Showing with 105 additions and 59 deletions.
  1. +9 −0 NOTICE.txt
  2. +96 −59 Sources/NIOWebSocket/Base64.swift
@@ -77,3 +77,12 @@ This product contains "ifaddrs-android.c" and "ifaddrs-android.h" from dxr.mozil
* https://webrtc.googlesource.com/src/+/master/LICENSE
* HOMEPAGE:
* https://webrtc.googlesource.com/

---

This product contains a derivation of Fabian Fett's 'Base64.swift'.

* LICENSE (Apache License 2.0):
* https://github.com/fabianfett/swift-base64-kit/blob/master/LICENSE
* HOMEPAGE:
* https://github.com/fabianfett/swift-base64-kit
@@ -12,75 +12,112 @@
//
//===----------------------------------------------------------------------===//
// The base64 unicode table.
private let base64Table: [Unicode.Scalar] = [
"A", "B", "C", "D", "E", "F", "G", "H",
"I", "J", "K", "L", "M", "N", "O", "P",
"Q", "R", "S", "T", "U", "V", "W", "X",
"Y", "Z", "a", "b", "c", "d", "e", "f",
"g", "h", "i", "j", "k", "l", "m", "n",
"o", "p", "q", "r", "s", "t", "u", "v",
"w", "x", "y", "z", "0", "1", "2", "3",
"4", "5", "6", "7", "8", "9", "+", "/",
]
// This is a simplified vendored version from:
// https://github.com/fabianfett/swift-base64-kit
extension String {
/// Base64 encode an array of UInt8 to a string, without the use of Foundation.
///
/// This function performs the world's most naive Base64 encoding: no attempts to use a larger
/// lookup table or anything intelligent like that, just shifts and masks. This works fine, for
/// now: the purpose of this encoding is to avoid round-tripping through Data, and the perf gain
/// from avoiding that is more than enough to outweigh the silliness of this code.
init(base64Encoding array: Array<UInt8>) {
// In Base64, 3 bytes become 4 output characters, and we pad to the nearest multiple
// of four.
var outputString = String()
outputString.reserveCapacity(((array.count + 2) / 3) * 4)

var bytes = array.makeIterator()
while let firstByte = bytes.next() {
let secondByte = bytes.next()
let thirdByte = bytes.next()
outputString.unicodeScalars.append(String.encode(firstByte: firstByte))
outputString.unicodeScalars.append(String.encode(firstByte: firstByte, secondByte: secondByte))
outputString.unicodeScalars.append(String.encode(secondByte: secondByte, thirdByte: thirdByte))
outputString.unicodeScalars.append(String.encode(thirdByte: thirdByte))
}
/// Base64 encode a collection of UInt8 to a string, without the use of Foundation.
init<Buffer: Collection>(base64Encoding bytes: Buffer)
where Buffer.Element == UInt8
{
self = Base64.encode(bytes: bytes)
}
}

self = outputString
}

private static func encode(firstByte: UInt8) -> Unicode.Scalar {
let index = firstByte >> 2
return base64Table[Int(index)]
fileprivate struct Base64 {

static func encode<Buffer: Collection>(bytes: Buffer)
-> String where Buffer.Element == UInt8
{
// In Base64, 3 bytes become 4 output characters, and we pad to the
// nearest multiple of four.
let newCapacity = ((bytes.count + 2) / 3) * 4
let alphabet = Base64.encodeBase64

// NOTE: Once SE-263 lands we should replace this implementation with one
// that makes use of the non copy initializer. For more information
// please see:
// https://github.com/apple/swift-evolution/blob/master/proposals/0263-string-uninitialized-initializer.md
var outputBytes = [UInt8]()
outputBytes.reserveCapacity(newCapacity)

var input = bytes.makeIterator()

while let firstByte = input.next() {
let secondByte = input.next()
let thirdByte = input.next()

let firstChar = Base64.encode(alphabet: alphabet, firstByte: firstByte)
let secondChar = Base64.encode(alphabet: alphabet, firstByte: firstByte, secondByte: secondByte)
let thirdChar = Base64.encode(alphabet: alphabet, secondByte: secondByte, thirdByte: thirdByte)
let forthChar = Base64.encode(alphabet: alphabet, thirdByte: thirdByte)

outputBytes.append(firstChar)
outputBytes.append(secondChar)
outputBytes.append(thirdChar)
outputBytes.append(forthChar)
}

private static func encode(firstByte: UInt8, secondByte: UInt8?) -> Unicode.Scalar {
var index = (firstByte & 0b00000011) << 4
if let secondByte = secondByte {
index += (secondByte & 0b11110000) >> 4
}
return base64Table[Int(index)]
return String(decoding: outputBytes, as: Unicode.UTF8.self)
}

// MARK: Internal
// The base64 unicode table.
static let encodeBase64: [UInt8] = [
UInt8(ascii: "A"), UInt8(ascii: "B"), UInt8(ascii: "C"), UInt8(ascii: "D"),
UInt8(ascii: "E"), UInt8(ascii: "F"), UInt8(ascii: "G"), UInt8(ascii: "H"),
UInt8(ascii: "I"), UInt8(ascii: "J"), UInt8(ascii: "K"), UInt8(ascii: "L"),
UInt8(ascii: "M"), UInt8(ascii: "N"), UInt8(ascii: "O"), UInt8(ascii: "P"),
UInt8(ascii: "Q"), UInt8(ascii: "R"), UInt8(ascii: "S"), UInt8(ascii: "T"),
UInt8(ascii: "U"), UInt8(ascii: "V"), UInt8(ascii: "W"), UInt8(ascii: "X"),
UInt8(ascii: "Y"), UInt8(ascii: "Z"), UInt8(ascii: "a"), UInt8(ascii: "b"),
UInt8(ascii: "c"), UInt8(ascii: "d"), UInt8(ascii: "e"), UInt8(ascii: "f"),
UInt8(ascii: "g"), UInt8(ascii: "h"), UInt8(ascii: "i"), UInt8(ascii: "j"),
UInt8(ascii: "k"), UInt8(ascii: "l"), UInt8(ascii: "m"), UInt8(ascii: "n"),
UInt8(ascii: "o"), UInt8(ascii: "p"), UInt8(ascii: "q"), UInt8(ascii: "r"),
UInt8(ascii: "s"), UInt8(ascii: "t"), UInt8(ascii: "u"), UInt8(ascii: "v"),
UInt8(ascii: "w"), UInt8(ascii: "x"), UInt8(ascii: "y"), UInt8(ascii: "z"),
UInt8(ascii: "0"), UInt8(ascii: "1"), UInt8(ascii: "2"), UInt8(ascii: "3"),
UInt8(ascii: "4"), UInt8(ascii: "5"), UInt8(ascii: "6"), UInt8(ascii: "7"),
UInt8(ascii: "8"), UInt8(ascii: "9"), UInt8(ascii: "+"), UInt8(ascii: "/"),
]

static let encodePaddingCharacter: UInt8 = UInt8(ascii: "=")

static func encode(alphabet: [UInt8], firstByte: UInt8) -> UInt8 {
let index = firstByte >> 2
return alphabet[Int(index)]
}

static func encode(alphabet: [UInt8], firstByte: UInt8, secondByte: UInt8?) -> UInt8 {
var index = (firstByte & 0b00000011) << 4
if let secondByte = secondByte {
index += (secondByte & 0b11110000) >> 4
}
return alphabet[Int(index)]
}

private static func encode(secondByte: UInt8?, thirdByte: UInt8?) -> Unicode.Scalar {
guard let secondByte = secondByte else {
// No second byte means we are just emitting padding.
return "="
}
var index = (secondByte & 0b00001111) << 2
if let thirdByte = thirdByte {
index += (thirdByte & 0b11000000) >> 6
}
return base64Table[Int(index)]
static func encode(alphabet: [UInt8], secondByte: UInt8?, thirdByte: UInt8?) -> UInt8 {
guard let secondByte = secondByte else {
// No second byte means we are just emitting padding.
return Base64.encodePaddingCharacter
}
var index = (secondByte & 0b00001111) << 2
if let thirdByte = thirdByte {
index += (thirdByte & 0b11000000) >> 6
}
return alphabet[Int(index)]
}

private static func encode(thirdByte: UInt8?) -> Unicode.Scalar {
guard let thirdByte = thirdByte else {
// No third byte means just padding.
return "="
}
let index = thirdByte & 0b00111111
return base64Table[Int(index)]
static func encode(alphabet: [UInt8], thirdByte: UInt8?) -> UInt8 {
guard let thirdByte = thirdByte else {
// No third byte means just padding.
return Base64.encodePaddingCharacter
}
let index = thirdByte & 0b00111111
return alphabet[Int(index)]
}
}

0 comments on commit 6534927

Please sign in to comment.
You can’t perform that action at this time.