Skip to content

Commit

Permalink
More (shameless) writer over-optimization
Browse files Browse the repository at this point in the history
  • Loading branch information
@dcodeIO
dcodeIO committed Dec 5, 2016
1 parent 8a2dbc6 commit d83f799
Show file tree
Hide file tree
Showing 11 changed files with 324 additions and 202 deletions.
30 changes: 15 additions & 15 deletions README.md
View file
Expand Up @@ -364,33 +364,33 @@ The package includes a [benchmark](https://github.com/dcodeIO/protobuf.js/tree/m
```
benchmarking encoding performance ...
Type.encode to buffer x 514,048 ops/sec ±0.75% (93 runs sampled)
JSON.stringify to string x 355,935 ops/sec ±0.79% (91 runs sampled)
JSON.stringify to buffer x 191,023 ops/sec ±1.39% (86 runs sampled)
Type.encode to buffer x 469,818 ops/sec ±0.58% (89 runs sampled)
JSON.stringify to string x 309,883 ops/sec ±0.81% (92 runs sampled)
JSON.stringify to buffer x 174,440 ops/sec ±1.46% (87 runs sampled)
Type.encode to buffer was fastest
JSON.stringify to string was 30.8% slower
JSON.stringify to buffer was 63.1% slower
JSON.stringify to string was 34.2% slower
JSON.stringify to buffer was 63.2% slower
benchmarking decoding performance ...
Type.decode from buffer x 1,238,587 ops/sec ±1.73% (87 runs sampled)
JSON.parse from string x 312,168 ops/sec ±2.22% (83 runs sampled)
JSON.parse from buffer x 272,975 ops/sec ±2.45% (82 runs sampled)
Type.decode from buffer x 976,639 ops/sec ±0.81% (91 runs sampled)
JSON.parse from string x 294,282 ops/sec ±0.69% (89 runs sampled)
JSON.parse from buffer x 263,440 ops/sec ±0.84% (93 runs sampled)
Type.decode from buffer was fastest
JSON.parse from string was 74.9% slower
JSON.parse from buffer was 78.1% slower
JSON.parse from string was 69.8% slower
JSON.parse from buffer was 73.0% slower
benchmarking combined performance ...
Type to/from buffer x 246,428 ops/sec ±1.52% (89 runs sampled)
JSON to/from string x 136,380 ops/sec ±1.50% (80 runs sampled)
JSON to/from buffer x 95,229 ops/sec ±1.93% (86 runs sampled)
Type to/from buffer x 205,029 ops/sec ±1.59% (88 runs sampled)
JSON to/from string x 127,198 ops/sec ±0.76% (91 runs sampled)
JSON to/from buffer x 90,980 ops/sec ±0.77% (91 runs sampled)
Type to/from buffer was fastest
JSON to/from string was 44.6% slower
JSON to/from buffer was 61.5% slower
JSON to/from string was 37.5% slower
JSON to/from buffer was 55.3% slower
```

Note that JSON is a native binding nowadays and as such is *really* fast. So, how can protobuf.js be faster?
Expand Down
222 changes: 172 additions & 50 deletions bench/write.js
View file
@@ -1,75 +1,197 @@
var protobuf = require("../src/index"),
newSuite = require("./suite");
newSuite = require("./suite"),
ieee754 = require("../lib/ieee754");

// This benchmark compares raw data type performance of Uint8Array and Buffer.
// This uses internal finish machinery because overall difference is otherwise mainly caused by allocation.

var array = new Uint8Array(8);
/* var array = new Uint8Array(8);
var buffer = new Buffer(8);
// The following is a possible alternative to float / double writing - where supported. Quite a bit faster.

var f32 = new Float32Array(1);
var i8 = new Uint8Array(f32.buffer);
// raw fixed32 write speed
newSuite("fixed32")
.add("shift array", function() {
var val = 0x7fffffff;
array[0] = val & 255;
array[1] = val >>> 8 & 255;
array[2] = val >>> 16 & 255;
array[3] = val >>> 24;
})
.add("shift buffer", function() {
var val = 0x7fffffff;
buffer[0] = val & 255;
buffer[1] = val >>> 8 & 255;
buffer[2] = val >>> 16 & 255;
buffer[3] = val >>> 24;
})
.add("writeUInt32LE buffer", function() {
var val = 0x7fffffff;
buffer.writeUInt32LE(val, 0, true);
})
.run();
function writeF32Array(buf, pos, val) {
f32[0] = val;
for (var i = 0; i < 4; ++i)
buf[pos + i] = i8[i];
}
var f64 = new Float64Array(1);
var f32 = new Float32Array(f64.buffer);
var f8b = new Uint8Array(f64.buffer);
// raw float write speed
newSuite("float")
.add("Writer#float", function() {
var writer = new protobuf.Writer();
writer.float(0.1);
writer._finish(writer.head, array);
})
.add("Writer#writeF32Array", function() {
var writer = new protobuf.Writer();
writer.push(writeF32Array, 4, 0.1);
writer._finish(writer.head, array);
})
.add("BufferWriter#float", function() {
var writer = new protobuf.BufferWriter();
writer.float(0.1);
writer._finish(writer.head, buffer);
})
.add("BufferWriter#writeF32Array", function() {
var writer = new protobuf.BufferWriter();
writer.push(writeF32Array, 4, 0.1);
writer._finish(writer.head, buffer);
.add("ieee754 array", function() {
ieee754.write(array, 0.1, 0, false, 23, 4);
})
.add("ieee754 buffer", function() {
ieee754.write(buffer, 0.1, 0, false, 23, 4);
})
.add("f32 array", function() {
f32[0] = 0.1;
array[0] = f8b[0];
array[0+1] = f8b[1];
array[0+2] = f8b[2];
array[0+3] = f8b[3];
})
.add("f32 buffer", function() {
f32[0] = 0.1;
buffer[0] = f8b[0];
buffer[0+1] = f8b[1];
buffer[0+2] = f8b[2];
buffer[0+3] = f8b[3];
})
.add("writeFloatLE buffer", function() {
buffer.writeFloatLE(0.1, 0, true);
})
.run();
// raw double write speed
newSuite("double")
.add("Writer#double", function() {
var writer = new protobuf.Writer();
writer.double(0.1);
writer._finish(writer.head, array);
.add("ieee754 array", function() {
ieee754.write(array, 0.1, 0, false, 52, 8);
})
.add("ieee754 buffer", function() {
ieee754.write(buffer, 0.1, 0, false, 52, 8);
})
.add("BufferWriter#double", function() {
var writer = new protobuf.BufferWriter();
writer.double(0.1);
writer._finish(writer.head, buffer);
.add("f64 array", function() {
f64[0] = 0.1;
array[0] = f8b[0];
array[0+1] = f8b[1];
array[0+2] = f8b[2];
array[0+3] = f8b[3];
array[0+4] = f8b[4];
array[0+5] = f8b[5];
array[0+6] = f8b[6];
array[0+7] = f8b[7];
})
.add("f64 buffer", function() {
f64[0] = 0.1;
buffer[0] = f8b[0];
buffer[0+1] = f8b[1];
buffer[0+2] = f8b[2];
buffer[0+3] = f8b[3];
buffer[0+4] = f8b[4];
buffer[0+5] = f8b[5];
buffer[0+6] = f8b[6];
buffer[0+7] = f8b[7];
})
.add("writeDoubleLE buffer", function() {
buffer.writeDoubleLE(0.1, 0, true);
})
.run();
var arrayPlus1 = new Uint8Array(array.length + 1);
var bufferPlus1 = new Buffer(buffer.length + 1);
var source = Buffer.from([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]);
var array = new Uint8Array(16);
var buffer = new Buffer(16);
// raw bytes write speed
// interestingly, Uint8Array#set is faster than Buffer#copy, but only on actual Uint8Arrays.
newSuite("bytes")
.add("Writer#bytes", function() {
var writer = new protobuf.Writer();
writer.bytes(array);
writer._finish(writer.head, arrayPlus1);
.add("set array", function() {
array.set(source, 0);
})
.add("for array", function() {
for (var i = 0; i < source.length; ++i)
array[i] = source[i];
})
.add("BufferWriter#bytes", function() {
var writer = new protobuf.BufferWriter();
writer.bytes(buffer);
writer._finish(writer.head, bufferPlus1);
.add("set buffer", function() {
buffer.set(source, 0);
})
.add("for buffer", function() {
for (var i = 0; i < source.length; ++i)
buffer[i] = source[i];
})
.add("copy buffer", function() {
source.copy(buffer, 0);
})
.run();
*/

function writeString(buf, pos, val) {
for (var i = 0; i < val.length; ++i) {
var c1 = val.charCodeAt(i), c2;
if (c1 < 128) {
buf[pos++] = c1;
} else if (c1 < 2048) {
buf[pos++] = c1 >> 6 | 192;
buf[pos++] = c1 & 63 | 128;
} else if ((c1 & 0xFC00) === 0xD800 && ((c2 = val.charCodeAt(i + 1)) & 0xFC00) === 0xDC00) {
c1 = 0x10000 + ((c1 & 0x03FF) << 10) + (c2 & 0x03FF);
++i;
buf[pos++] = c1 >> 18 | 240;
buf[pos++] = c1 >> 12 & 63 | 128;
buf[pos++] = c1 >> 6 & 63 | 128;
buf[pos++] = c1 & 63 | 128;
} else {
buf[pos++] = c1 >> 12 | 224;
buf[pos++] = c1 >> 6 & 63 | 128;
buf[pos++] = c1 & 63 | 128;
}
}
}

function byteLength(val) {
var strlen = val.length >>> 0;
var len = 0;
for (var i = 0; i < strlen; ++i) {
var c1 = val.charCodeAt(i);
if (c1 < 128)
len += 1;
else if (c1 < 2048)
len += 2;
else if ((c1 & 0xFC00) === 0xD800 && (val.charCodeAt(i + 1) & 0xFC00) === 0xDC00) {
++i;
len += 4;
} else
len += 3;
}
return len;
}

var array = new Uint8Array(1000);
var buffer = new Buffer(1000);

[
"Lorem ipsu",
"Lorem ipsum dolo",
"Lorem ipsum dolor ",
"Lorem ipsum dolor s",
"Lorem ipsum dolor si",
"Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Duis autem vel eum iriure dolor in hendrerit in vulputate velit esse molestie consequat, vel illum dolore ipsum."
].forEach(function(str) {
// raw string write speed
newSuite("string[" + str.length + "]")
.add("writeString array", function() {
writeString(array, 0, str);
})
.add("writeString buffer", function() {
writeString(buffer, 0, str)
})
.add("write buffer", function() {
buffer.write(str, 0)
})
.add("utf8Write buffer", function() {
buffer.utf8Write(str, 0)
})
/* .add("byteLength array", function() {
byteLength(str)
})
.add("byteLength buffer", function() {
Buffer.byteLength(str)
}) */
.run();
});

0 comments on commit d83f799

Please sign in to comment.