Remove spills from most inlined runtime calls.

Remove spills from most inlined runtime calls.

Committed: http://code.google.com/p/v8/source/detail?r=1182

[William Hesse, 2009-01-28 12:18:10]

This change is independent of the change to CallCodeObject.

[William Hesse, 2009-01-28 12:18:35]

This change is independent of the change to CallCodeObject.

[Kevin Millikin (Chromium), 2009-01-29 09:42:36]

This LGTM, but I think you should get rid of the drop/push sequence in a couple
of places and clean up some comments.

http://codereview.chromium.org/19622/diff/1/2
File src/codegen-ia32.cc (left):

http://codereview.chromium.org/19622/diff/1/2#oldcode3591
Line 3591: // We need the CC bits to come out as not_equal in the case where the
It's nice to get rid of this.

http://codereview.chromium.org/19622/diff/1/2
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/19622/diff/1/2#newcode3604
Line 3604: Result temp = allocator()->Allocate();
We will need to sort out the decision to spill an existing register (eg, value)
or allocate a fresh one.  A simple solution is if the existing register is
unaliased in the frame, reuse it, otherwise allocate.

A more sophisticated solution considers the cost of spilling vs. allocating. 
Allocating will never consider the actually unneeded off-frame register, because
it thinks it is needed.

http://codereview.chromium.org/19622/diff/1/2#newcode3620
Line 3620: // Seed the result with the formal parameters count, which will be
I would reword this comment to be more explicit.  The real point is that the
(read length version of the) arguments access stub takes the parameter count as
an argument passed in eax.

http://codereview.chromium.org/19622/diff/1/2#newcode3657
Line 3657: frame_->Push(&temp);
Can we avoid the drop/push sequence here by setting the top element?

http://codereview.chromium.org/19622/diff/1/2#newcode3705
Line 3705: frame_->Drop();
Can we avoid the drop/push sequence here by setting the top element?

http://codereview.chromium.org/19622/diff/1/2#newcode3719
Line 3719: left.ToRegister();
I would rather not move left to a register, but generate a difference cmp
instruction in the case it's a constant.

http://codereview.chromium.org/19622/diff/1/2#newcode3732
Line 3732: VirtualFrame::SpilledScope spilled_scope(this);
Double check that this is scope is gone once you update to include your
immediately previous change list.

[William Hesse, 2012-05-07 09:36:09]

http://codereview.chromium.org/19622/diff/1/src/codegen-ia32.cc
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/19622/diff/1/src/codegen-ia32.cc#newcode3620
src/codegen-ia32.cc:3620: // Seed the result with the formal parameters count,
which will be
On 2009/01/29 09:42:36, Kevin Millikin wrote:
> I would reword this comment to be more explicit.  The real point is that the
> (read length version of the) arguments access stub takes the parameter count
as
> an argument passed in eax.

Done.

http://codereview.chromium.org/19622/diff/1/src/codegen-ia32.cc#newcode3657
src/codegen-ia32.cc:3657: frame_->Push(&temp);
On 2009/01/29 09:42:36, Kevin Millikin wrote:
> Can we avoid the drop/push sequence here by setting the top element?

Done.

http://codereview.chromium.org/19622/diff/1/src/codegen-ia32.cc#newcode3705
src/codegen-ia32.cc:3705: frame_->Drop();
On 2009/01/29 09:42:36, Kevin Millikin wrote:
> Can we avoid the drop/push sequence here by setting the top element?

Done.

http://codereview.chromium.org/19622/diff/1/src/codegen-ia32.cc#newcode3719
src/codegen-ia32.cc:3719: left.ToRegister();
On 2009/01/29 09:42:36, Kevin Millikin wrote:
> I would rather not move left to a register, but generate a difference cmp
> instruction in the case it's a constant.

What if right is a constant?  Let's do this when we start handling constants
differently.