Experimental: thread the virtual frame through the deferred code...

Experimental: thread the virtual frame through the deferred code
generated at the end of functions.  It is piggybacked on the jump target
infrastructure pairing the label of a block with the epected frame on
entry.

This is simple and works for now but eventually we want to get the code
to make a frame mergable off the fast path or do away with it
completely.

The inlined code for the smi part of binary operations is duplicated,
but that is a temporary situation.

Committed: http://code.google.com/p/v8/source/detail?r=957

[Mads Ager (chromium), 2008-12-10 11:48:23]

LGTM

[William Hesse, 2008-12-10 12:02:51]

Looks good to me, with one comment.

http://codereview.chromium.org/13339/diff/1/6
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/13339/diff/1/6#newcode708
Line 708: void GenerateInlineCode();
Where is this function now defined?  Or is it dead?

[Kevin Millikin (Chromium), 2008-12-10 12:09:50]

http://codereview.chromium.org/13339/diff/1/6
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/13339/diff/1/6#newcode708
Line 708: void GenerateInlineCode();
On 2008/12/10 12:02:51, William Hesse wrote:
> Where is this function now defined?  Or is it dead?

Not dead, we call it in the generic binary operation.  It's a clone of
GenericBinaryOpStub::GenerateSmiCode, so I defined them together to emphasize
that fact.

http://codereview.chromium.org/13339/diff/1/6#newcode4386
Line 4386: void DeferredInlineBinaryOperation::GenerateInlineCode() {
...is defined right here.