595391 - JM: Fix empty script performance

Status: RESOLVED DUPLICATE of bug 614333

(Core :: JavaScript Engine, defect)

Description

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

Per bug 594972 comment #2 through 4, we should figure out how deal with the empty script.

Right now we can't JIT it easily because it sits in read-only memory and is shared across all threads.

Comment 1

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

Potential options:
 1) Special case it in ICs.
 2) Remove it.
 3) Pre-assemble something and attach it, make the empty script readable.
 4) Put an emptyScript bit in JSScript and move the empty script per-compartment.

Comment 2

[David Mandelin [:dmandelin]]

(In reply to comment #1)
> Potential options:
>  1) Special case it in ICs.

This is the most appealing option to me. It seems that the call IC is the natural place for any kind of special-case action when calling.

>  3) Pre-assemble something and attach it, make the empty script readable.

This seems to be the easiest dumb solution, so I would go for that if #1 is no-go. But I don't understand #4 and its advantages too clearly, so maybe that one is better.

Comment 3

[Chris Leary [:cdleary] (not checking bugmail)]

Dumb background question: when do we actually see the empty script in performant code?

Comment 4

[Chris Leary [:cdleary] (not checking bugmail)]

Nevermind, reading through bugmail and I saw bug 594972 comment 4.

Comment 6

[Jan de Mooij [:jandem]]

I can't reproduce this anymore (likely bug 614333). Calling an empty function many times is now as fast as calling a function with "var x" in it.