Inlined load of string.length and array.length....

Inlined load of string.length and array.length.

http://compute1.aar:9013/golem/r4583-v8-serya-length-inlined-vs-4583-v8.html


Committed: http://code.google.com/p/v8/source/detail?r=4601

[SeRya, 2010-05-05 11:55:17]

Changes in debug.cc needs to pass cctest/test-debug/DebugStepKeyedLoadLoop. It
looks like breakpoints set on a named load instruction did't work when it had
been  inlined. Also it looks for me like DebugStepKeyedLoadLoop tests that it
doesn't intend to test.

Soren, do you think I'm right? If so I would change this test to set a
breakpoint on the line "x = a[i];" and add a similar test for a breakpoint on a
named load statement.

[Søren Thygesen Gjesse, 2010-05-05 12:33:12]

The debugger change looks right, and please add a test with a inlined named
load.

Regarding the actual inlining of string.length and array.length isn't this
change missing some files?

[SeRya, 2010-05-05 14:50:09]

Added the DebugStepNamedLoadLoop test.

I checked that all the modified files are here. This change is based on commited
http://codereview.chromium.org/1706013 which changed string length
representation to SMI.

I was wrong about DebugStepKeyedLoadLoop, it indeed checks that inlined keyed
load reverted. But I still don't understand how it could depend on named load of
array length. Could V8 replace 'i < len' by 'i < a.length'?

[Søren Thygesen Gjesse, 2010-05-06 09:16:39]

LGTM

It is quite funny that we have had unused inlined code sitting there for .length
access on arrays and strings for so long.