Copy of changelist from issue 16450. Converts compare operation...

Copy of changelist from issue 16450.  Converts compare operation
to use virtual frames, adds xchg instruction to assembler.

Committed: http://code.google.com/p/v8/source/detail?r=1021

[William Hesse, 2008-12-29 11:47:25]

Here is the new version of 16450, with your comments addressed.

[Kevin Millikin (Chromium), 2008-12-29 13:30:37]

LGTM with some minor comments below.

http://codereview.chromium.org/16481/diff/1/8
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/16481/diff/1/8#newcode1212
Line 1212: is_smi.Branch(zero, &right_side, &left_side, taken);
It doesn't really matter, but would rather have these passed in the other order.

http://codereview.chromium.org/16481/diff/1/8#newcode1230
Line 1230: Result answer = frame_->CallStub(&stub, &right_side, &left_side, 0);
Ditto.

http://codereview.chromium.org/16481/diff/1/8#newcode1238
Line 1238: // This is currenly sub-optimal, I think.
A multiline comment should not begin as an end of line comment.

Don't be so personal in the comments and try to be more explicit.  The reader
doesn't know who the "I" in question is if they want to ask why this is
suboptimal.

You should probably hint at what's suboptimal here: the slow case frame will be
spilled by the call to the stub, which will currently cause the fast case frame
to be spilled too.

http://codereview.chromium.org/16481/diff/1/3
File src/virtual-frame-ia32.cc (right):

http://codereview.chromium.org/16481/diff/1/3#newcode135
Line 135: if (!frame_registers_.is_used(target.code())) return;
Use the overloaded version of is_used that takes a Register argument.

[William Hesse, 2008-12-29 13:43:27]

All changes made.

http://codereview.chromium.org/16481/diff/1/8
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/16481/diff/1/8#newcode1212
Line 1212: is_smi.Branch(zero, &right_side, &left_side, taken);
On 2008/12/29 13:30:37, Kevin Millikin wrote:
> It doesn't really matter, but would rather have these passed in the other
order.

Done.

http://codereview.chromium.org/16481/diff/1/8#newcode1230
Line 1230: Result answer = frame_->CallStub(&stub, &right_side, &left_side, 0);
On 2008/12/29 13:30:37, Kevin Millikin wrote:
> Ditto.

Done.

http://codereview.chromium.org/16481/diff/1/8#newcode1238
Line 1238: // This is currenly sub-optimal, I think.
On 2008/12/29 13:30:37, Kevin Millikin wrote:
> A multiline comment should not begin as an end of line comment.
> 
> Don't be so personal in the comments and try to be more explicit.  The reader
> doesn't know who the "I" in question is if they want to ask why this is
> suboptimal.
> 
> You should probably hint at what's suboptimal here: the slow case frame will
be
> spilled by the call to the stub, which will currently cause the fast case
frame
> to be spilled too.

Done.

http://codereview.chromium.org/16481/diff/1/3
File src/virtual-frame-ia32.cc (right):

http://codereview.chromium.org/16481/diff/1/3#newcode135
Line 135: if (!frame_registers_.is_used(target.code())) return;
On 2008/12/29 13:30:37, Kevin Millikin wrote:
> Use the overloaded version of is_used that takes a Register argument.

Done.