X64 Crankshaft: Add Typeof and TypeofIs to x64 lithium.

X64 Crankshaft: Add Typeof and TypeofIs to x64 lithium.  Fix presubmit.

Committed: http://code.google.com/p/v8/source/detail?r=6912

[William Hesse, 2011-02-23 10:36:34]

ConstructCall functions are just moved below EmitTypeOf functions.

[Lasse Reichstein, 2011-02-23 11:22:36]

LGTM

http://codereview.chromium.org/6568004/diff/1/src/x64/lithium-codegen-x64.cc
File src/x64/lithium-codegen-x64.cc (right):

http://codereview.chromium.org/6568004/diff/1/src/x64/lithium-codegen-x64.cc#...
src/x64/lithium-codegen-x64.cc:3207: __ LoadRoot(result,
Heap::kTrueValueRootIndex);
Would it be faster to make a macro for loading true/false that uses the fact
that they are adjacent in the root array?
I.e., something like

 setcc(condition, kScratchRegister);
 movzxbl(kScratchRegister, kScratchRegister);
 movq(result, Operand(kRootRegister, kScratchRegister, times_pointer_size,
Heap::kTrueValueRootIndex * kPointerSize);

That would avoid the conditional jumps (which, in this case, wouldn't help us
since the EmitTypeofIs already uses labels for control, so that would need to be
changed too).

[William Hesse, 2011-02-23 11:27:51]

On 2011/02/23 11:22:36, Lasse Reichstein wrote:
> LGTM
> 
> http://codereview.chromium.org/6568004/diff/1/src/x64/lithium-codegen-x64.cc
> File src/x64/lithium-codegen-x64.cc (right):
> 

We do this at some other place, but, since the Typeof code is based
on jumps, it does not help much here.
>
http://codereview.chromium.org/6568004/diff/1/src/x64/lithium-codegen-x64.cc#...
> src/x64/lithium-codegen-x64.cc:3207: __ LoadRoot(result,
> Heap::kTrueValueRootIndex);
> Would it be faster to make a macro for loading true/false that uses the fact
> that they are adjacent in the root array?
> I.e., something like
> 
>  setcc(condition, kScratchRegister);
>  movzxbl(kScratchRegister, kScratchRegister);
>  movq(result, Operand(kRootRegister, kScratchRegister, times_pointer_size,
> Heap::kTrueValueRootIndex * kPointerSize);
> 
> That would avoid the conditional jumps (which, in this case, wouldn't help us
> since the EmitTypeofIs already uses labels for control, so that would need to
be
> changed too).