1156886 - Optimize ToLowerCase and ToUpperCase in vm/Unicode.h

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jan de Mooij [:jandem]]

I was just comparing our Dromaeo object-string numbers with V8 (see also bug 1088293), and noticed we're like 8x slower on the toLowerCase/toUpperCase tests.

There are a few things we should do:

(1) Optimize the ch < 128 case in ToLowerCase(char16_t ch) and ToLowerCase(char16_t ch), similar to what we do for IsSpace, IsIdentifierPart and others.

(2) V8 has an even more optimal algorithm for Latin1 strings that can compare/convert multiple characters at once.

Just doing (1) should help a good deal, (2) might be nice as well.

Comment 1

[Jan de Mooij [:jandem]]

Attachment: Patch

This is (1) in comment 0. For the microbenchmark below I get:

js before: 561 ms
js after:  197 ms
d8:        118 ms

Still a bit slower than V8 but 2.8x faster than before and probably good enough for now.

function f() {
    var s = "abcdefg0123456789";
    for (var i=0; i<4; i++)
        s += s;
    var t = new Date;
    for (var i=0; i<1000000; i++)
        s = s.toLowerCase();
    print(new Date - t);
    return s;
}
f();

Comment 2

[Luke Wagner [:luke]]

Comment on attachment 8595542 [details] [diff] [review]
Patch

Nice

Comment 3

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/8d4f212cc9b8

Comment 4

[Guilherme Lima]

Unfortunately, on AWFY at least, this patch made no difference.

Comment 5

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/8d4f212cc9b8