Fixed performance regression in v8 regexp benchmark on ARM.

Fixed performance regression in v8 regexp benchmark on ARM.

Committed: http://code.google.com/p/v8/source/detail?r=9119

[Yang, 2011-09-02 13:15:22]

Please take a look.

The performance regression was introduced in r9027 when I committed my tentative
implementation for sliced strings. In particular, the bug was introduced in line
4500:
http://code.google.com/p/v8/source/diff?spec=svn9027&r=9027&format=side&path=...

http://codereview.chromium.org/7824033/diff/1/src/arm/code-stubs-arm.cc
File src/arm/code-stubs-arm.cc (left):

http://codereview.chromium.org/7824033/diff/1/src/arm/code-stubs-arm.cc#oldco...
src/arm/code-stubs-arm.cc:4530: __ b(ne, &runtime);
even though subject is not the result of the regexp, comparing it with SUCCESS,
FAILURE and EXCEPTION would eventually result in a jump into runtime, so that
the final result is correct, but performance takes a hit.

http://codereview.chromium.org/7824033/diff/1/src/arm/code-stubs-arm.cc
File src/arm/code-stubs-arm.cc (right):

http://codereview.chromium.org/7824033/diff/1/src/arm/code-stubs-arm.cc#newco...
src/arm/code-stubs-arm.cc:4490: __ ldr(subject, MemOperand(fp, kSubjectOffset +
2 * kPointerSize));
the underlying subject string is no longer needed, and can be replaced by the
original subject string.

http://codereview.chromium.org/7824033/diff/1/src/arm/code-stubs-arm.cc#newco...
src/arm/code-stubs-arm.cc:4523: __ cmp(r0,
Operand(NativeRegExpMacroAssembler::SUCCESS));
not subject (r4) but r0 is the return value of the regexp.

[Erik Corry, 2011-09-02 13:25:19]

LGTM

http://codereview.chromium.org/7824033/diff/1/src/arm/code-stubs-arm.cc
File src/arm/code-stubs-arm.cc (right):

http://codereview.chromium.org/7824033/diff/1/src/arm/code-stubs-arm.cc#newco...
src/arm/code-stubs-arm.cc:4546: __ LoadRoot(ip,
Heap::kTerminationExceptionRootIndex);
We have CompareRoot for this.