Initial port of VirtualFrame to the ARM architecture. To reduce the...

Initial port of VirtualFrame to the ARM architecture.  To reduce the
size of the change, the initial implementation provides read access to
the frame (eg, locals, parameters, and expression stack elements) but
no mutation.

Committed: http://code.google.com/p/v8/source/detail?r=694

[Kevin Millikin (Chromium), 2008-10-14 15:30:13]

This will be broken into two changes.

[iposva, 2008-10-16 04:48:47]

Apart from the below comments LGTM.

-Ivan

http://codereview.chromium.org/7302/diff/1/2
File src/codegen-arm.cc (right):

http://codereview.chromium.org/7302/diff/1/2#newcode1671
Line 1671: __ ldr(r1, frame_->Element(kNextOffset / kPointerSize));  // read
next_sp
It would be less error prone if the division by kPointerSize would be done once
at the local variable initialization. Maybe you want to rename the local to not
be a Offset because that does imply byte offsets.

http://codereview.chromium.org/7302/diff/1/2#newcode1690
Line 1690: __ ldr(r1, frame_->Element(kNextOffset / kPointerSize));
ditto

http://codereview.chromium.org/7302/diff/1/2#newcode1780
Line 1780: __ ldr(r1, frame_->Element(kNextOffset / kPointerSize));
ditto

http://codereview.chromium.org/7302/diff/1/3
File src/codegen-arm.h (right):

http://codereview.chromium.org/7302/diff/1/3#newcode68
Line 68: // Index -1 corresponds to the receiver.
Can you please add a
MemOperand Receiver() const {
  return Parameter(-1);
}

This will increase the readability tremendously because then you are not left
wondering what the -1 parameter is. Same should be done for the ia32 file.
Although at this point I do not see a single explicit reference to Parameter(-1)
in the CodeGenerator at all. So this might be just a cosmetic nit.

[Kevin Millikin (Chromium), 2008-11-04 08:39:59]

Ivan,

I'd like to land this change (and the next part, pushing and popping through the
virtual frame) on the bleeding_edge branch.  Right now, the two code generators
are too far out of synch.

http://codereview.chromium.org/7302/diff/401/601
File src/codegen-arm.cc (right):

http://codereview.chromium.org/7302/diff/401/601#newcode217
Line 217: __ CallStub(&stub);
This is the only place that "ReceiverOperand" would be explicitly used (the
index -1 also appears in slots apparently).  I'm not quite sure how to best
rewrite this snippet to make it write through the frame.

http://codereview.chromium.org/7302/diff/401/601#newcode1700
Line 1700: / kPointerSize;
Changed as per request.

[iposva, 2008-11-05 00:33:56]

This looks as it is pretty much equivalent to the ia32 stuff.

LGTM

-Ivan