buffer: speed up Buffer.isEncoding() method by JacksonTian · Pull Request #5256 · nodejs/node

JacksonTian · 2016-02-16T08:27:34Z

Use automata to avoid toLowerCase(), faster, but more dirty.

JacksonTian · 2016-02-16T08:28:23Z

Following is benchmark result:

$ ./node benchmark/compare.js ./node ~/.tnvm/versions/node/v5.5.0/bin/node -- buffers buffer-is-encoding.js
running ./node
buffers/buffer-is-encoding.js
running /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node
buffers/buffer-is-encoding.js
buffers/buffer-is-encoding.js encoding=hex: ./node: 37865000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 53626000 ......... -29.39%
buffers/buffer-is-encoding.js encoding=utf8: ./node: 32338000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 46584000 ........ -30.58%
buffers/buffer-is-encoding.js encoding=utf-8: ./node: 29117000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 29546000 ........ -1.45%
buffers/buffer-is-encoding.js encoding=ascii: ./node: 32724000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 26424000 ........ 23.84%
buffers/buffer-is-encoding.js encoding=binary: ./node: 26486000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 22335000 ....... 18.59%
buffers/buffer-is-encoding.js encoding=base64: ./node: 29564000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 17563000 ....... 68.33%
buffers/buffer-is-encoding.js encoding=ucs2: ./node: 36896000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 17559000 ........ 110.12%
buffers/buffer-is-encoding.js encoding=ucs-2: ./node: 31430000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 15970000 ........ 96.81%
buffers/buffer-is-encoding.js encoding=utf16le: ./node: 25419000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 14396000 ...... 76.57%
buffers/buffer-is-encoding.js encoding=utf-16le: ./node: 25545000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 12160000 .... 110.07%
buffers/buffer-is-encoding.js encoding=HEX: ./node: 38260000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 6280700 .......... 509.18%
buffers/buffer-is-encoding.js encoding=UTF8: ./node: 33291000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 6278300 ......... 430.25%
buffers/buffer-is-encoding.js encoding=UTF-8: ./node: 28115000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 5482000 ........ 412.87%
buffers/buffer-is-encoding.js encoding=ASCII: ./node: 33288000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4861600 ........ 584.71%
buffers/buffer-is-encoding.js encoding=BINARY: ./node: 25741000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 5067200 ....... 408.00%
buffers/buffer-is-encoding.js encoding=BASE64: ./node: 32574000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 5095100 ....... 539.33%
buffers/buffer-is-encoding.js encoding=UCS2: ./node: 32299000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 5052700 ......... 539.24%
buffers/buffer-is-encoding.js encoding=UCS-2: ./node: 32000000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4389800 ........ 628.97%
buffers/buffer-is-encoding.js encoding=UTF16LE: ./node: 23653000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4427300 ...... 434.26%
buffers/buffer-is-encoding.js encoding=UTF-16LE: ./node: 25788000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4756400 ..... 442.17%
buffers/buffer-is-encoding.js encoding=utf9: ./node: 33145000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4541200 ......... 629.87%
buffers/buffer-is-encoding.js encoding=utf-7: ./node: 31347000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4464600 ........ 602.13%
buffers/buffer-is-encoding.js encoding=utf17le: ./node: 31211000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4309600 ...... 624.23%
buffers/buffer-is-encoding.js encoding=utf-17le: ./node: 21892000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 5136700 ..... 326.19%
buffers/buffer-is-encoding.js encoding=Unicode-FTW: ./node: 61616000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4847600 . 1171.06%
buffers/buffer-is-encoding.js encoding=new gnu gun: ./node: 68657000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 4702100 . 1360.13%

Use automata to avoid toLowerCase(), faster, but more dirty.

ChALkeR · 2016-02-16T08:52:35Z

Ah, ignore my previous comment, I misread your results =). I removed it.

ronkorving · 2016-02-16T08:54:39Z

Is this not getting out of hand? I appreciate all the performance work being done, and being quite the performance geek myself, I really enjoy this. But how is this maintainable?

JacksonTian · 2016-02-16T08:54:44Z

: )

ronkorving · 2016-02-16T08:57:06Z

If I may make a suggestion that would keep maintainability at a higher level. Why not generate a function that generates the very code you wrote here (as a string), based on an array of strings you feed into it, then feed that into new Function and reap the benefits from that? See also acorn

ChALkeR · 2016-02-16T08:59:36Z

I'm not exactly sure that this is worth it.
By selecting only the popular encodings, we get:

buffers/buffer-is-encoding.js encoding=hex: ./node: 37865000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 53626000 ......... -29.39%
buffers/buffer-is-encoding.js encoding=utf8: ./node: 32338000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 46584000 ........ -30.58%
buffers/buffer-is-encoding.js encoding=utf-8: ./node: 29117000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 29546000 ........ -1.45%
buffers/buffer-is-encoding.js encoding=ascii: ./node: 32724000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 26424000 ........ 23.84%
buffers/buffer-is-encoding.js encoding=binary: ./node: 26486000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 22335000 ....... 18.59%
buffers/buffer-is-encoding.js encoding=base64: ./node: 29564000 /Users/jacksontian/.tnvm/versions/node/v5.5.0/bin/node: 17563000 ....... 68.33%

Also, I doubt that it's the bottleneck in some actual code.

JacksonTian · 2016-02-16T09:01:33Z

@ronkorving

So we have to trade performance off against maintainable.

Feel free to discuss it.

mscdex · 2016-02-16T09:14:37Z

I think I'd be more in favor of what @ronkorving is suggesting. It shouldn't be that hard to do either. Also the generated and generator function could be re-used in other places where case-insensitive (static) string matching is done.

ChALkeR · 2016-02-16T09:38:10Z

My results, alternative approach (this is compared with your PR, not with the current master):

buffers/buffer-is-encoding.js encoding=hex: ./node: 45905000 ./out/Release/node-pr5256: 35969000 .......... 27.62%
buffers/buffer-is-encoding.js encoding=utf8: ./node: 43540000 ./out/Release/node-pr5256: 30779000 ......... 41.46%
buffers/buffer-is-encoding.js encoding=utf-8: ./node: 42315000 ./out/Release/node-pr5256: 28372000 ........ 49.15%
buffers/buffer-is-encoding.js encoding=ascii: ./node: 32530000 ./out/Release/node-pr5256: 31952000 ......... 1.81%
buffers/buffer-is-encoding.js encoding=binary: ./node: 37935000 ./out/Release/node-pr5256: 25886000 ....... 46.55%
buffers/buffer-is-encoding.js encoding=base64: ./node: 29954000 ./out/Release/node-pr5256: 27233000 ........ 9.99%
buffers/buffer-is-encoding.js encoding=ucs2: ./node: 33028000 ./out/Release/node-pr5256: 29890000 ......... 10.50%
buffers/buffer-is-encoding.js encoding=ucs-2: ./node: 25978000 ./out/Release/node-pr5256: 27828000 ........ -6.65%
buffers/buffer-is-encoding.js encoding=utf16le: ./node: 38147000 ./out/Release/node-pr5256: 27242000 ...... 40.03%
buffers/buffer-is-encoding.js encoding=utf-16le: ./node: 37962000 ./out/Release/node-pr5256: 23881000 ..... 58.96%
buffers/buffer-is-encoding.js encoding=HEX: ./node: 11735000 ./out/Release/node-pr5256: 34252000 ......... -65.74%
buffers/buffer-is-encoding.js encoding=UTF8: ./node: 10640000 ./out/Release/node-pr5256: 27859000 ........ -61.81%
buffers/buffer-is-encoding.js encoding=UTF-8: ./node: 9671900 ./out/Release/node-pr5256: 26836000 ........ -63.96%
buffers/buffer-is-encoding.js encoding=ASCII: ./node: 9061500 ./out/Release/node-pr5256: 24353000 ........ -62.79%
buffers/buffer-is-encoding.js encoding=BINARY: ./node: 9730300 ./out/Release/node-pr5256: 22144000 ....... -56.06%
buffers/buffer-is-encoding.js encoding=BASE64: ./node: 9383700 ./out/Release/node-pr5256: 23397000 ....... -59.89%
buffers/buffer-is-encoding.js encoding=UCS2: ./node: 9742700 ./out/Release/node-pr5256: 28104000 ......... -65.33%
buffers/buffer-is-encoding.js encoding=UCS-2: ./node: 8520900 ./out/Release/node-pr5256: 26971000 ........ -68.41%
buffers/buffer-is-encoding.js encoding=UTF16LE: ./node: 10237000 ./out/Release/node-pr5256: 24840000 ..... -58.79%
buffers/buffer-is-encoding.js encoding=UTF-16LE: ./node: 10102000 ./out/Release/node-pr5256: 20327000 .... -50.30%
buffers/buffer-is-encoding.js encoding=utf9: ./node: 9463400 ./out/Release/node-pr5256: 32503000 ......... -70.88%
buffers/buffer-is-encoding.js encoding=utf-7: ./node: 7779000 ./out/Release/node-pr5256: 28436000 ........ -72.64%
buffers/buffer-is-encoding.js encoding=utf17le: ./node: 10116000 ./out/Release/node-pr5256: 26293000 ..... -61.53%
buffers/buffer-is-encoding.js encoding=utf-17le: ./node: 10322000 ./out/Release/node-pr5256: 23956000 .... -56.91%
buffers/buffer-is-encoding.js encoding=Unicode-FTW: ./node: 12716000 ./out/Release/node-pr5256: 50317000 . -74.73%
buffers/buffer-is-encoding.js encoding=new gnu gun: ./node: 13034000 ./out/Release/node-pr5256: 47377000 . -72.49%

Code:

Buffer.isEncoding=function(encoding){if(!encoding.length)encoding=''+encoding;varloweredCase=false;for(;;){switch(encoding.length){case3:
switch(encoding){case'hex':
returntrue;}break;case4:
switch(encoding){case'utf8':
case'ucs2':
returntrue;}break;case5:
switch(encoding){case'utf-8':
case'ascii':
case'ucs-2':
returntrue;}break;case6:
switch(encoding){case'binary':
case'base64':
returntrue;}break;case7:
switch(encoding){case'utf16le':
returntrue;}break;case8:
switch(encoding){case'utf-16le':
returntrue;}break;}if(loweredCase)returnfalse;encoding=(''+encoding).toLowerCase();loweredCase=true;}};

Note that all the actively used encodings are actually faster this way, and that the code is a bit cleaner.

Results compared with master:

buffers/buffer-is-encoding.js encoding=hex: ./node: 48271000 ./out/Release/node-master: 67270000 ........ -28.24%
buffers/buffer-is-encoding.js encoding=utf8: ./node: 43292000 ./out/Release/node-master: 43090000 ......... 0.47%
buffers/buffer-is-encoding.js encoding=utf-8: ./node: 42688000 ./out/Release/node-master: 29407000 ....... 45.17%
buffers/buffer-is-encoding.js encoding=ascii: ./node: 27191000 ./out/Release/node-master: 26880000 ........ 1.15%
buffers/buffer-is-encoding.js encoding=binary: ./node: 38716000 ./out/Release/node-master: 22406000 ...... 72.79%
buffers/buffer-is-encoding.js encoding=base64: ./node: 29703000 ./out/Release/node-master: 19464000 ...... 52.61%
buffers/buffer-is-encoding.js encoding=ucs2: ./node: 32250000 ./out/Release/node-master: 18487000 ........ 74.44%
buffers/buffer-is-encoding.js encoding=ucs-2: ./node: 25865000 ./out/Release/node-master: 15130000 ....... 70.95%
buffers/buffer-is-encoding.js encoding=utf16le: ./node: 38206000 ./out/Release/node-master: 12219000 .... 212.69%
buffers/buffer-is-encoding.js encoding=utf-16le: ./node: 37975000 ./out/Release/node-master: 9336700 .... 306.72%
buffers/buffer-is-encoding.js encoding=HEX: ./node: 11703000 ./out/Release/node-master: 7395300 .......... 58.24%
buffers/buffer-is-encoding.js encoding=UTF8: ./node: 10564000 ./out/Release/node-master: 6881200 ......... 53.52%
buffers/buffer-is-encoding.js encoding=UTF-8: ./node: 9603500 ./out/Release/node-master: 6410600 ......... 49.81%
buffers/buffer-is-encoding.js encoding=ASCII: ./node: 9010300 ./out/Release/node-master: 6396200 ......... 40.87%
buffers/buffer-is-encoding.js encoding=BINARY: ./node: 9724800 ./out/Release/node-master: 5432600 ........ 79.01%
buffers/buffer-is-encoding.js encoding=BASE64: ./node: 9248300 ./out/Release/node-master: 5398700 ........ 71.30%
buffers/buffer-is-encoding.js encoding=UCS2: ./node: 9863800 ./out/Release/node-master: 5778700 .......... 70.69%
buffers/buffer-is-encoding.js encoding=UCS-2: ./node: 8581300 ./out/Release/node-master: 5278000 ......... 62.59%
buffers/buffer-is-encoding.js encoding=UTF16LE: ./node: 10150000 ./out/Release/node-master: 4933200 ..... 105.75%
buffers/buffer-is-encoding.js encoding=UTF-16LE: ./node: 10312000 ./out/Release/node-master: 4579800 .... 125.16%
buffers/buffer-is-encoding.js encoding=utf9: ./node: 9374900 ./out/Release/node-master: 4416800 ......... 112.26%
buffers/buffer-is-encoding.js encoding=utf-7: ./node: 7697000 ./out/Release/node-master: 4292900 ......... 79.30%
buffers/buffer-is-encoding.js encoding=utf17le: ./node: 9994000 ./out/Release/node-master: 4237500 ...... 135.85%
buffers/buffer-is-encoding.js encoding=utf-17le: ./node: 10442000 ./out/Release/node-master: 4244100 .... 146.04%
buffers/buffer-is-encoding.js encoding=Unicode-FTW: ./node: 12759000 ./out/Release/node-master: 4875500 . 161.70%
buffers/buffer-is-encoding.js encoding=new gnu gun: ./node: 13005000 ./out/Release/node-master: 4993200 . 160.46%

I still doubt that this is worth it, because hex is still slower, and utf8 is sometimes slower, sometimes faster (tending to be slower, actually).

For reference, your PR against master on my PC:

buffers/buffer-is-encoding.js encoding=hex: ./out/Release/node-pr5256: 33124000 ./out/Release/node-master: 67186000 ......... -50.70%
buffers/buffer-is-encoding.js encoding=utf8: ./out/Release/node-pr5256: 32003000 ./out/Release/node-master: 43807000 ........ -26.95%
buffers/buffer-is-encoding.js encoding=utf-8: ./out/Release/node-pr5256: 27405000 ./out/Release/node-master: 29570000 ........ -7.32%
buffers/buffer-is-encoding.js encoding=ascii: ./out/Release/node-pr5256: 30777000 ./out/Release/node-master: 27544000 ........ 11.74%
buffers/buffer-is-encoding.js encoding=binary: ./out/Release/node-pr5256: 27291000 ./out/Release/node-master: 22253000 ....... 22.64%
buffers/buffer-is-encoding.js encoding=base64: ./out/Release/node-pr5256: 29385000 ./out/Release/node-master: 19128000 ....... 53.62%
buffers/buffer-is-encoding.js encoding=ucs2: ./out/Release/node-pr5256: 32516000 ./out/Release/node-master: 18361000 ......... 77.09%
buffers/buffer-is-encoding.js encoding=ucs-2: ./out/Release/node-pr5256: 28762000 ./out/Release/node-master: 14970000 ........ 92.13%
buffers/buffer-is-encoding.js encoding=utf16le: ./out/Release/node-pr5256: 24719000 ./out/Release/node-master: 12204000 ..... 102.55%
buffers/buffer-is-encoding.js encoding=utf-16le: ./out/Release/node-pr5256: 22317000 ./out/Release/node-master: 9641300 ..... 131.47%
buffers/buffer-is-encoding.js encoding=HEX: ./out/Release/node-pr5256: 35672000 ./out/Release/node-master: 7263700 .......... 391.09%
buffers/buffer-is-encoding.js encoding=UTF8: ./out/Release/node-pr5256: 28220000 ./out/Release/node-master: 6985000 ......... 304.01%
buffers/buffer-is-encoding.js encoding=UTF-8: ./out/Release/node-pr5256: 25079000 ./out/Release/node-master: 6383700 ........ 292.87%
buffers/buffer-is-encoding.js encoding=ASCII: ./out/Release/node-pr5256: 22696000 ./out/Release/node-master: 6173500 ........ 267.64%
buffers/buffer-is-encoding.js encoding=BINARY: ./out/Release/node-pr5256: 22915000 ./out/Release/node-master: 5689200 ....... 302.78%
buffers/buffer-is-encoding.js encoding=BASE64: ./out/Release/node-pr5256: 23843000 ./out/Release/node-master: 5414300 ....... 340.37%
buffers/buffer-is-encoding.js encoding=UCS2: ./out/Release/node-pr5256: 29847000 ./out/Release/node-master: 5532300 ......... 439.51%
buffers/buffer-is-encoding.js encoding=UCS-2: ./out/Release/node-pr5256: 27359000 ./out/Release/node-master: 5274800 ........ 418.66%
buffers/buffer-is-encoding.js encoding=UTF16LE: ./out/Release/node-pr5256: 25104000 ./out/Release/node-master: 4864700 ...... 416.03%
buffers/buffer-is-encoding.js encoding=UTF-16LE: ./out/Release/node-pr5256: 22025000 ./out/Release/node-master: 4567800 ..... 382.19%
buffers/buffer-is-encoding.js encoding=utf9: ./out/Release/node-pr5256: 29698000 ./out/Release/node-master: 4326000 ......... 586.50%
buffers/buffer-is-encoding.js encoding=utf-7: ./out/Release/node-pr5256: 26996000 ./out/Release/node-master: 4328400 ........ 523.69%
buffers/buffer-is-encoding.js encoding=utf17le: ./out/Release/node-pr5256: 27189000 ./out/Release/node-master: 4212900 ...... 545.37%
buffers/buffer-is-encoding.js encoding=utf-17le: ./out/Release/node-pr5256: 24378000 ./out/Release/node-master: 4246500 ..... 474.08%
buffers/buffer-is-encoding.js encoding=Unicode-FTW: ./out/Release/node-pr5256: 52608000 ./out/Release/node-master: 4990900 .. 954.07%
buffers/buffer-is-encoding.js encoding=new gnu gun: ./out/Release/node-pr5256: 55166000 ./out/Release/node-master: 4982400 . 1007.20%

Edit: code and results updated.

ChALkeR · 2016-02-16T09:57:40Z

Also note that the results should probably be rechecked on v8 4.9, this looks like something v8 should optimize on it's own.

jasnell · 2016-02-16T17:12:07Z

A generator function for this kind of parsing is a very good idea. I have deep concerns over making these kinds of micro-optimizations as it does make the code much more opaque and difficult to manage. If we can put this behind a generator function that optimizes either at compile time or module-load time, then we can balance those needs quite well.

trevnorris · 2016-02-16T19:08:03Z

I'm not -1 for performance improvements, but we need to recognize what we're actually optimizing. Look in lib/buffer.js and you'll notice that there are several locations that have a encoding type lookup. May I suggest something like:

functionnormalizeEncoding(encoding){// place all the black magic wizardy you want here}

Then use it like so:

Buffer.isEncoding=functionisEncoding(encoding){returnnormalizeEncoding(encoding)!==null;};

Which would also be useful for cases like Buffer#write(). e.g.:

switch(normalizeEncoding(encoding)){case'hex':
returnthis.hexWrite(string,offset,length);
...

Make sense?

brendanashworth · 2016-02-17T02:17:13Z

Which would also be useful for cases like Buffer#write(). e.g.:
switch(normalizeEncoding(encoding)) {
case 'hex':
return this.hexWrite(string, offset, length);
...

There are a few spots in core where encoding names are accepted and some where they aren't (e.g. 'UTF-8' (turned into buffer), so this benefits in more than one way.

mscdex · 2016-02-17T04:10:09Z

I had a little extra time today so I wrote a case-insensitive string comparison function generator and it seems like you pretty much need to keep the existing switch there that checks for the lowered cases in order to keep the performance from regressing for those (common) cases and just use the custom comparison function in place of the .toLowerCase() code.

I've also added a pre-optimize step in the benchmark in this PR before calling bench.start() to make things as fair as possible (you can see how to do this in the path and other benchmarks).

FWIW here are the results I get with my changes:

buffers/buffer-isencoding.js encoding=hex n=100000000: ./node: 42032000 ./node-before: 39292000 .......... 6.97%
buffers/buffer-isencoding.js encoding=utf8 n=100000000: ./node: 37179000 ./node-before: 35457000 ......... 4.86%
buffers/buffer-isencoding.js encoding=utf-8 n=100000000: ./node: 30648000 ./node-before: 29186000 ........ 5.01%
buffers/buffer-isencoding.js encoding=ascii n=100000000: ./node: 29270000 ./node-before: 26816000 ........ 9.15%
buffers/buffer-isencoding.js encoding=binary n=100000000: ./node: 26409000 ./node-before: 24497000 ....... 7.80%
buffers/buffer-isencoding.js encoding=base64 n=100000000: ./node: 23973000 ./node-before: 21901000 ....... 9.46%
buffers/buffer-isencoding.js encoding=ucs2 n=100000000: ./node: 21117000 ./node-before: 19959000 ......... 5.80%
buffers/buffer-isencoding.js encoding=ucs-2 n=100000000: ./node: 19334000 ./node-before: 18743000 ........ 3.16%
buffers/buffer-isencoding.js encoding=utf16le n=100000000: ./node: 17433000 ./node-before: 16605000 ...... 4.99%
buffers/buffer-isencoding.js encoding=utf-16le n=100000000: ./node: 15808000 ./node-before: 15745000 ..... 0.40%
buffers/buffer-isencoding.js encoding=HEX n=100000000: ./node: 14810000 ./node-before: 9373600 .......... 58.00%
buffers/buffer-isencoding.js encoding=UTF8 n=100000000: ./node: 14805000 ./node-before: 8956500 ......... 65.30%
buffers/buffer-isencoding.js encoding=UTF-8 n=100000000: ./node: 14587000 ./node-before: 8565300 ........ 70.30%
buffers/buffer-isencoding.js encoding=ASCII n=100000000: ./node: 14361000 ./node-before: 8739700 ........ 64.31%
buffers/buffer-isencoding.js encoding=BINARY n=100000000: ./node: 14236000 ./node-before: 8297500 ....... 71.56%
buffers/buffer-isencoding.js encoding=BASE64 n=100000000: ./node: 14234000 ./node-before: 8109800 ....... 75.52%
buffers/buffer-isencoding.js encoding=UCS2 n=100000000: ./node: 14652000 ./node-before: 7721100 ......... 89.76%
buffers/buffer-isencoding.js encoding=UCS-2 n=100000000: ./node: 14613000 ./node-before: 7525100 ........ 94.19%
buffers/buffer-isencoding.js encoding=UTF16LE n=100000000: ./node: 13613000 ./node-before: 7112700 ...... 91.40%
buffers/buffer-isencoding.js encoding=UTF-16LE n=100000000: ./node: 13367000 ./node-before: 7211300 ..... 85.37%
buffers/buffer-isencoding.js encoding=utf9 n=100000000: ./node: 12552000 ./node-before: 6386300 ......... 96.54%
buffers/buffer-isencoding.js encoding=utf-7 n=100000000: ./node: 12450000 ./node-before: 6395700 ........ 94.66%
buffers/buffer-isencoding.js encoding=utf17le n=100000000: ./node: 12738000 ./node-before: 6399700 ...... 99.05%
buffers/buffer-isencoding.js encoding=utf-17le n=100000000: ./node: 12333000 ./node-before: 6793400 ..... 81.54%
buffers/buffer-isencoding.js encoding=Unicode-FTW n=100000000: ./node: 17596000 ./node-before: 7299400 . 141.06%
buffers/buffer-isencoding.js encoding=new gnu gun n=100000000: ./node: 17258000 ./node-before: 7583100 . 127.58%

ChALkeR · 2016-02-17T06:01:50Z

@mscdex Have you tried to split the existing switch by the length of the input string?

mscdex · 2016-02-17T06:28:49Z

@ChALkeR I just tested that additional change and it does improve things:

buffers/buffer-isencoding.js encoding=hex n=100000000: ./node: 40356000 ./node-pre-string: 39489000 .......... 2.20%
buffers/buffer-isencoding.js encoding=utf8 n=100000000: ./node: 39834000 ./node-pre-string: 36809000 ......... 8.22%
buffers/buffer-isencoding.js encoding=utf-8 n=100000000: ./node: 38004000 ./node-pre-string: 29532000 ....... 28.69%
buffers/buffer-isencoding.js encoding=ascii n=100000000: ./node: 33636000 ./node-pre-string: 28023000 ....... 20.03%
buffers/buffer-isencoding.js encoding=binary n=100000000: ./node: 35756000 ./node-pre-string: 24973000 ...... 43.18%
buffers/buffer-isencoding.js encoding=base64 n=100000000: ./node: 31033000 ./node-pre-string: 22689000 ...... 36.77%
buffers/buffer-isencoding.js encoding=ucs2 n=100000000: ./node: 34393000 ./node-pre-string: 19936000 ........ 72.52%
buffers/buffer-isencoding.js encoding=ucs-2 n=100000000: ./node: 29396000 ./node-pre-string: 18719000 ....... 57.04%
buffers/buffer-isencoding.js encoding=utf16le n=100000000: ./node: 36172000 ./node-pre-string: 16855000 .... 114.60%
buffers/buffer-isencoding.js encoding=utf-16le n=100000000: ./node: 34160000 ./node-pre-string: 15633000 ... 118.52%
buffers/buffer-isencoding.js encoding=HEX n=100000000: ./node: 30110000 ./node-pre-string: 9410000 ......... 219.98%
buffers/buffer-isencoding.js encoding=UTF8 n=100000000: ./node: 26229000 ./node-pre-string: 9099000 ........ 188.26%
buffers/buffer-isencoding.js encoding=UTF-8 n=100000000: ./node: 22442000 ./node-pre-string: 8706300 ....... 157.77%
buffers/buffer-isencoding.js encoding=ASCII n=100000000: ./node: 22335000 ./node-pre-string: 8691300 ....... 156.98%
buffers/buffer-isencoding.js encoding=BINARY n=100000000: ./node: 22724000 ./node-pre-string: 8394200 ...... 170.70%
buffers/buffer-isencoding.js encoding=BASE64 n=100000000: ./node: 22631000 ./node-pre-string: 7945700 ...... 184.82%
buffers/buffer-isencoding.js encoding=UCS2 n=100000000: ./node: 25254000 ./node-pre-string: 7735400 ........ 226.47%
buffers/buffer-isencoding.js encoding=UCS-2 n=100000000: ./node: 22591000 ./node-pre-string: 7618800 ....... 196.51%
buffers/buffer-isencoding.js encoding=UTF16LE n=100000000: ./node: 23471000 ./node-pre-string: 7152400 ..... 228.16%
buffers/buffer-isencoding.js encoding=UTF-16LE n=100000000: ./node: 22622000 ./node-pre-string: 7259100 .... 211.64%
buffers/buffer-isencoding.js encoding=utf9 n=100000000: ./node: 22312000 ./node-pre-string: 6422100 ........ 247.42%
buffers/buffer-isencoding.js encoding=utf-7 n=100000000: ./node: 20268000 ./node-pre-string: 6468300 ....... 213.35%
buffers/buffer-isencoding.js encoding=utf17le n=100000000: ./node: 23559000 ./node-pre-string: 6395500 ..... 268.37%
buffers/buffer-isencoding.js encoding=utf-17le n=100000000: ./node: 21861000 ./node-pre-string: 6777700 .... 222.54%
buffers/buffer-isencoding.js encoding=Unicode-FTW n=100000000: ./node: 39908000 ./node-pre-string: 7257100 . 449.92%
buffers/buffer-isencoding.js encoding=new gnu gun n=100000000: ./node: 39264000 ./node-pre-string: 7596700 . 416.86%

EDIT: updated results, there was a typo previously which affected things a bit

ChALkeR · 2016-02-17T20:13:58Z

@mscdex Can I take a look at the code? ;-)

mscdex · 2016-02-17T20:59:51Z

@ChALkeR I just pushed it here.

JacksonTian · 2016-02-18T02:19:17Z

I am confuse why my first version faster than master, but look like slower after force optimization with following code:

// Force optimization before starting the benchmark
Buffer.isEncoding(encoding);
v8.setFlagsFromString('--allow_natives_syntax');
eval('%OptimizeFunctionOnNextCall(Buffer.isEncoding)');
Buffer.isEncoding(encoding);

JacksonTian · 2016-02-18T02:37:08Z

Here is an another implementation. It use template to generate isEncoding method.

mscdex · 2016-02-18T03:26:53Z

@JacksonTian Did you check the output when adding at least --trace-opt --trace-deopt to the command line to see if there is anything obvious happening (e.g. permanent deopt, constant re-optimizations, etc.)? Otherwise you can view/compare the resulting instructions in IRHydra2.

trevnorris · 2016-02-18T10:45:57Z

Buffer.isEncoding() is not a common code path. Though the code itself is common throughout Buffer. If this can be generalized for other similar cases into its own function and used everywhere applicable then I'll +1 this. But while I respect the performance tuning done here, it's significant enough on an uncommon code path that I may say it's not worth it.

jasnell · 2016-02-19T20:42:15Z

+1 to what @trevnorris said.

ronkorving · 2016-02-20T05:02:07Z

Is there something to say for the Node community to play a bigger role in V8 development directly? It seems to me that this type of optimization should be completely unnecessary if V8 was able to do it itself (in fact, it should be even much faster I expect). Now I'm no V8 expert in the slightest, so don't see myself lead an effort like this, but the Node community has a lot of working groups. Does it not make sense to have a group dedicated to the JS engine that Node so deeply depends upon?

bnoordhuis · 2016-02-20T11:15:06Z

Does it not make sense to have a group dedicated to the JS engine that Node so deeply depends upon?

#3741 - which, admittedly, I dropped the ball on. I wasn't (and still am not) quite sure how to flesh it out properly. It doesn't help that I currently don't have much time to dedicate.

ronkorving · 2016-02-20T11:49:34Z

Ah, I didn't know that was already ongoing. Great to see so much interest. I hope that process can resume at some point.

jasnell · 2016-03-21T22:51:59Z

@JacksonTian ... ping ... what's the status on this one?

RReverser · 2016-06-08T10:40:27Z

It this going to live in the light of #7207?

brendanashworth · 2016-08-01T22:10:12Z

I'm going to go ahead and close this — after #7207 the PR must be entirely redone if it is even still relevant, and the discussion over a generator function should probably occur upon code landing in a new PR. Thanks!

buffer: speed up Buffer.isEncoding() method
0c4bb94
Use automata to avoid toLowerCase(), faster, but more dirty.

JacksonTian force-pushed the faster_is_encoding branch from b9c4f27 to 0c4bb94Compare February 16, 2016 08:32

ChALkeR added the buffer Issues and PRs related to the buffer subsystem. label Feb 16, 2016

ChALkeR added the performance Issues and PRs related to the performance of Node.js. label Feb 16, 2016

ronkorving mentioned this pull request Feb 27, 2016
lib: improve module loading performance #5172
Merged

estliberitas force-pushed the master branch 2 times, most recently from 7da4fd4 to c7066fbCompare April 26, 2016 05:23

brendanashworth closed this Aug 1, 2016

JacksonTian deleted the faster_is_encoding branch May 4, 2017 08:09

Uh oh!

Uh oh!

Conversation

JacksonTian commented Feb 16, 2016

Uh oh!

JacksonTian commented Feb 16, 2016

Uh oh!

ChALkeR commented Feb 16, 2016

Uh oh!

ronkorving commented Feb 16, 2016

Uh oh!

JacksonTian commented Feb 16, 2016

Uh oh!

ronkorving commented Feb 16, 2016

Uh oh!

ChALkeR commented Feb 16, 2016

Uh oh!

JacksonTian commented Feb 16, 2016

Uh oh!

mscdex commented Feb 16, 2016

Uh oh!

ChALkeR commented Feb 16, 2016

Uh oh!

ChALkeR commented Feb 16, 2016

Uh oh!

jasnell commented Feb 16, 2016

Uh oh!

trevnorris commented Feb 16, 2016

Uh oh!

brendanashworth commented Feb 17, 2016

Uh oh!

mscdex commented Feb 17, 2016

Uh oh!

ChALkeR commented Feb 17, 2016

Uh oh!

mscdex commented Feb 17, 2016

Uh oh!

ChALkeR commented Feb 17, 2016

Uh oh!

mscdex commented Feb 17, 2016

Uh oh!

JacksonTian commented Feb 18, 2016

Uh oh!

JacksonTian commented Feb 18, 2016

Uh oh!

mscdex commented Feb 18, 2016

Uh oh!

trevnorris commented Feb 18, 2016

Uh oh!

jasnell commented Feb 19, 2016

Uh oh!

ronkorving commented Feb 20, 2016

Uh oh!

bnoordhuis commented Feb 20, 2016

Uh oh!

ronkorving commented Feb 20, 2016

Uh oh!

jasnell commented Mar 21, 2016

Uh oh!

RReverser commented Jun 8, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

brendanashworth commented Aug 1, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

RReverser commented Jun 8, 2016•
edited
Loading