Add inlining of keyed store on ARM...

Add inlining of keyed store on ARM

This ports the inlining of keyed store to the ARM port. As the inlined code does not handle the write barrier it only supports storing of smis.

Committed: http://code.google.com/p/v8/source/detail?r=4531

[Erik Corry, 2010-04-28 09:40:03]

LGTM

http://codereview.chromium.org/1719021/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/1719021/diff/1/2#newcode5305
src/arm/codegen-arm.cc:5305: __ DecrementCounter(&Counters::keyed_store_inline,
1, r1, r2);
Let's use VirtualFrame::scratch0() and scratch1() for these.

http://codereview.chromium.org/1719021/diff/1/2#newcode5310
src/arm/codegen-arm.cc:5310: // Call keyed load IC. It has all arguments on the
stack.
Comment out of date.  Is the value still in r0 here?

http://codereview.chromium.org/1719021/diff/1/2#newcode5463
src/arm/codegen-arm.cc:5463: // Generate inlined version of the keyed store if
the code is in a loop
We should assert that registers are spilled.

http://codereview.chromium.org/1719021/diff/1/2#newcode5489
src/arm/codegen-arm.cc:5489: // Load the receiver from the stack.
Perhaps we could hoist this load so it can execute in parallel with the tst and
branch above.

http://codereview.chromium.org/1719021/diff/1/2#newcode5659
src/arm/codegen-arm.cc:5659: frame->EmitPop(r0);  // Value.
Use FlushAllButCopyTOSToR0.

[Søren Thygesen Gjesse, 2010-04-28 11:08:23]

http://codereview.chromium.org/1719021/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/1719021/diff/1/2#newcode5305
src/arm/codegen-arm.cc:5305: __ DecrementCounter(&Counters::keyed_store_inline,
1, r1, r2);
On 2010/04/28 09:40:04, Erik Corry wrote:
> Let's use VirtualFrame::scratch0() and scratch1() for these.

Done.

http://codereview.chromium.org/1719021/diff/1/2#newcode5310
src/arm/codegen-arm.cc:5310: // Call keyed load IC. It has all arguments on the
stack.
On 2010/04/28 09:40:04, Erik Corry wrote:
> Comment out of date.  Is the value still in r0 here?

Done.

http://codereview.chromium.org/1719021/diff/1/2#newcode5463
src/arm/codegen-arm.cc:5463: // Generate inlined version of the keyed store if
the code is in a loop
On 2010/04/28 09:40:04, Erik Corry wrote:
> We should assert that registers are spilled.

Done.

http://codereview.chromium.org/1719021/diff/1/2#newcode5489
src/arm/codegen-arm.cc:5489: // Load the receiver from the stack.
On 2010/04/28 09:40:04, Erik Corry wrote:
> Perhaps we could hoist this load so it can execute in parallel with the tst
and
> branch above.

Done.

http://codereview.chromium.org/1719021/diff/1/2#newcode5659
src/arm/codegen-arm.cc:5659: frame->EmitPop(r0);  // Value.
On 2010/04/28 09:40:04, Erik Corry wrote:
> Use FlushAllButCopyTOSToR0.

That doesn't work. We need SpillAllButTOSWithTOSInR0 (I think).