Emit pushes and pops through the virtual frame on ARM. Merging of...

Emit pushes and pops through the virtual frame on ARM.  Merging of
frames is not yet handled.  The ARM code generator should be back in
line with the IA32 one.

Committed: http://code.google.com/p/v8/source/detail?r=706

[Kevin Millikin (Chromium), 2008-11-05 13:39:00]

This is the second part of the change to move all frame accesses to the virtual
frame object.  The ARM code generator should be essentially the same as the
Intel one on bleeding_edge after this change.

[iposva, 2008-11-06 18:18:34]

LGTM

Thanks for also cleaning up the comments.

-Ivan

http://codereview.chromium.org/9182/diff/201/202
File src/codegen-arm.cc (right):

http://codereview.chromium.org/9182/diff/201/202#newcode104
Line 104: __ add(sp, sp, Operand(kPointerSize));
Shouldn't this just call Drop(1)?

http://codereview.chromium.org/9182/diff/201/202#newcode172
Line 172: // cp: callee's context
Drive by comment not related to this change:
This is redundant to the "calling convention" comment below and I do believe
that both are wrong now that the calling convention changed!

Filed issue 141: http://code.google.com/p/v8/issues/detail?id=141

http://codereview.chromium.org/9182/diff/201/202#newcode193
Line 193: // cp: callee's context
ditto

http://codereview.chromium.org/9182/diff/201/202#newcode574
Line 574: if (size <= 0) {
Can a reference size ever be negative? If not, then we should ASSERT that
instead. Would make the code slightly more readable.

[Kevin Millikin (Chromium), 2008-11-07 08:19:10]

http://codereview.chromium.org/9182/diff/201/202
File src/codegen-arm.cc (right):

http://codereview.chromium.org/9182/diff/201/202#newcode104
Line 104: __ add(sp, sp, Operand(kPointerSize));
On 2008/11/06 18:18:34, iposva wrote:
> Shouldn't this just call Drop(1)?

Fixed.

http://codereview.chromium.org/9182/diff/201/202#newcode574
Line 574: if (size <= 0) {
On 2008/11/06 18:18:34, iposva wrote:
> Can a reference size ever be negative? If not, then we should ASSERT that
> instead. Would make the code slightly more readable.

Yes it can be negative, but we should probably change that.  The size is just
the (integer) value of the type enum, and ILLEGAL has value -1.  I will
investigate and clean up as a separate change.