Incomplete change for inlining more smi cases for

Refactor the smi case inlining for binary operations, so
it's easier to inline the code on demand. Right now, we still
only inline the smi case code for bitwise operations.

Committed: http://code.google.com/p/v8/source/detail?r=547

[iposva, 2008-10-21 16:59:22]

Other than the VirtualFrame vs native frame confusion below, this change LGTM.

-Ivan

http://codereview.chromium.org/7669/diff/401/404
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/7669/diff/401/404#newcode774
Line 774: __ push(eax);
This push here relies on the fact that the code generators VirtualFrame frame_
maps directly to the machine frame. I think this should be changed to return the
result in eax.

http://codereview.chromium.org/7669/diff/401/404#newcode819
Line 819: __ mov(eax, frame_->Top());  // get x
This should be a Pop(reg) to balance it with the requirement for the Push(eax)
below.

http://codereview.chromium.org/7669/diff/401/404#newcode827
Line 827: __ mov(frame_->Top(), eax);
We should frame_->Push(eax) here, since the deferred code does not know the
location of the virtual frame top. It works now, because the virtual frame and
the machine frame are exact mappings.