794117 - IonMonkey: string.split, array.map and array.filter slower with Ion

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jan de Mooij [:jandem]]

cers reported this on IRC. For the "split" micro-benchmark below we have:

JM+TI : 1198 ms
Ion   : 2594 ms

The problem is that these functions call GetTypeCallerInitObject, which calls currentScript. We want to self-host map and filter eventually, and we could specialize split.

function f() {
    var s = "";
    var t = new Date;
    for (var i=0; i<10000000; i++) {
        s.split(s);
    }
    print(new Date - t);
}
f();

Comment 1

[Boris Zbarsky [:bzbarsky]]

The real issue is that we end up in GetPcScript.  See bug 728045.

Comment 2

[David Anderson [:dvander] - inactive, e-mail if emergency]

Bleh - once again, I think this is indicative of a design flaw in the engine, and not a problem with GetPcScript. Though if there's no easy workaround for these maybe we should hack something up, like a one-entry cache.

Comment 3

[Boris Zbarsky [:bzbarsky]]

Well, by "real issue" I meant "where the time gets spent".  No opinion on how to fix it.  ;)

Comment 4

[Sean Stangl [:sstangl]]

With Bug 794679 fixed, local results are 1488 ms with --no-ion, 1579 with ion.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Well, it's still slower, just not as much slower, right?