441479 - for-in loops should use one backward branch (with downward goto on entry)

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect, P1)

Description

[Brendan Eich [:brendan]]

See bug 371802, which fixed while and do-while loops to use one backward branch, and bug 441477, which will be fixed shortly so for(;;) loops avoid branch at top and goto backward at bottom.

/be

Comment 1

[Brendan Eich [:brendan]]

Attachment: patch in progress, no decompiler changes yet

... so not ready for prime time. The downward goto to iterator-next call is not eliminated in the case of any empty loop body and no header-located main- and side-effects-committing loop variable update code (e.g., JSOP_FORLOCAL). When the patch for bug 458851 is done it will always put code up at the header.

Andreas, this patch leaves us with no loops using JSOP_IFEQ*, so is there any reason for TraceRecorder::flipIf and js_IsLoopEdge in jstracer.cpp to handle IFEQ?

/be

Comment 2

[Brendan Eich [:brendan]]

Attachment: fix

Jason, your review is more than welcome too.

This cleans up some old spaghetti in the decompiler, and paves the way for the elimination of the JSOP_FOR* bytecodes in favor of JSOP_NEXTITER followed by an appropriate JSOP_SET* op.

/be

Comment 3

[Blake Kaplan (:mrbkap) (inactive)]

Comment on attachment 344228 [details] [diff] [review]
fix

r=me with the ss_format fix I pointed out in person.

Comment 4

[Brendan Eich [:brendan]]

Fixed in tm:

http://hg.mozilla.org/tracemonkey/rev/ea1ffb1c0d93

We should sync with m-c today if we don't regress perf.

/be

Comment 5

[Brendan Eich [:brendan]]

Fixed in m-c today.

/be