Port optimization of comparison with a trivial LHS from ia32 to x64.

Port optimization of comparison with a trivial LHS from ia32 to x64.

Committed: http://code.google.com/p/v8/source/detail?r=4943

[Lasse Reichstein, 2010-06-24 08:59:48]

LGTM.
Does it apply in more cases? E.g. GenerateMathPow where one of the operands is
often a constant, and I'm guessing the left operand is most likely to be one.

http://codereview.chromium.org/2868028/diff/3001/4001
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/2868028/diff/3001/4001#newcode3839
src/x64/codegen-x64.cc:3839: Load(right);
The optimization is likely to apply here too, because left is almost certainly a
literal string.

http://codereview.chromium.org/2868028/diff/3001/4001#newcode3846
src/x64/codegen-x64.cc:3846: Load(right);
... but it's probably not worth it here, since left should be an object.

http://codereview.chromium.org/2868028/diff/3001/4001#newcode3863
src/x64/codegen-x64.cc:3863: frame_->Push(&right_result);
Could we reverse the comparison instead, and keep left on top of the stack?

[Lasse Reichstein, 2010-06-24 09:00:41]

Ignore the comment on GenerateMathPow - it's only called from our
Math.prototype.pow function with variables as arguments.

[William Hesse, 2010-06-24 14:43:53]

http://codereview.chromium.org/2868028/diff/3001/4001
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/2868028/diff/3001/4001#newcode3839
src/x64/codegen-x64.cc:3839: Load(right);
On 2010/06/24 08:59:48, Lasse Reichstein wrote:
> The optimization is likely to apply here too, because left is almost certainly
a
> literal string.

But since we are going to a builtin, it will be spilled anyway.

http://codereview.chromium.org/2868028/diff/3001/4001#newcode3863
src/x64/codegen-x64.cc:3863: frame_->Push(&right_result);
On 2010/06/24 08:59:48, Lasse Reichstein wrote:
> Could we reverse the comparison instead, and keep left on top of the stack?
> 
I don't think so, because the ToNumber calls inside comparison have to happen in
a certain order.
>