608823 - JM: "x & double" is slow

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jan de Mooij [:jandem]]

Attachment: Test case

For code like x & 0xffffffff JM is very slow. It does a stub call and spends a lot of time converting that double constant to int32. For the attached test case I get these numbers:

-m: 339, 3926 (> 11x slower)
-j: 221, 185  (first is always faster, seems unrelated)

Comment 1

[Tom S [:evilpie]]

Attachment: WIP 1

This patches makes bitop use tryBinaryConstantFold, and therefor extends this function.
I also changed it in a manner every primitive constant is handled like an normal integer value.

Comment 2

[Tom S [:evilpie]]

Attachment: Bitop with constant primitives

Comment 3

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

Comment on attachment 495483 [details] [diff] [review]
Bitop with constant primitives

Looks good, I'll r+ this with some tests :) if you need help writing them or getting them into the harness, just give a shout in #jsapi

Comment 4

[Tom S [:evilpie]]

Attachment: Bitop with constant primitives

Comment 5

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

Comment on attachment 498126 [details] [diff] [review]
Bitop with constant primitives

I am totally sorry about the lateness of this review. I suck. Patch looks great, and thanks for the tests!

Comment 6

[Boris Zbarsky [:bzbarsky]]

http://hg.mozilla.org/tracemonkey/rev/46a2d2d34ecf

Comment 7

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

http://hg.mozilla.org/mozilla-central/rev/46a2d2d34ecf