1176451 - Optimize Array.concat when used on different types of objects

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Brian Hackett [Laid off!]]

Attachment: patch

Currently we only optimize Array.concat when the 'this' and argument object have the same boxed/unboxed representation.  It would be better if we could optimize concat'ing different kinds of arrays; this affects performance a lot on kraken-stanford-crypto-pbkdf2, where double and int32 unboxed arrays are frequently concatenated.  The attached patch fixes this, and also fixes a couple other performance issues.  The unboxed array analysis is not forced the first time we try to attach a baseline cache for concat and other array natives (on pbkdf2 we would do this too early and end up using an unboxed int32 representation for a group that should really use an unboxed double representation) and the basic concat() native is augmented to use a similar optimized path to the specialized array_concat_dense and avoid some of the perf cost of calling this native.

Comment 1

[Jan de Mooij [:jandem]]

Comment on attachment 8625004 [details] [diff] [review]
patch

Review of attachment 8625004 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/jsarray.cpp
@@ +2565,2 @@
>  DenseElementResult
>  ArrayConcatDenseKernel(JSContext* cx, JSObject* obj1, JSObject* obj2, JSObject* result)

Hm I'm a bit worried about template bloat; this will emit like 36 copies. What's the increase in binary size, debug and opt build?

Comment 2

[Brian Hackett [Laid off!]]

On 10.9 x64 an opt shell is .25% bigger, and a --enable-debug --disable-optimize shell is .23% bigger.

Comment 3

[Jan de Mooij [:jandem]]

Comment on attachment 8625004 [details] [diff] [review]
patch

Review of attachment 8625004 [details] [diff] [review]:
-----------------------------------------------------------------

Sorry for the delay.

Comment 4

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/fdf5862a8c00

Comment 5

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/fdf5862a8c00

Comment 6

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

I backed this out in https://hg.mozilla.org/mozilla-central/rev/d6ea652c5799 to fix bug 1191492.

Comment 7

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/e671afb66591

Comment 8

[Brian Hackett [Laid off!]]

I relanded this patch as is because it looks like another bug caused bug 1191492.

Comment 9

[[Back July 11] Ryan VanderMeulen]

https://hg.mozilla.org/mozilla-central/rev/e671afb66591