Experimental: introduce a helper class to delmite a scope where the...

Experimental: introduce a helper class to delmite a scope where the
virtual frame is expected to be spilled.  Add versions of code
generator helper functions that spill the virtual frame before
returning.

This change should preserve semantics (one redundant spill was
removed) and is mostly for documentation purposes.  It should make
reading and reviewing the code easier.

Committed: http://code.google.com/p/v8/source/detail?r=980

[Kevin Millikin (Chromium), 2008-12-15 09:13:56]

Mads, can you take a quick look at this?  I'm hoping it makes it easier to work
(and do code reviews) on the code generator.

[Mads Ager (chromium), 2008-12-15 09:43:12]

LGTM

http://codereview.chromium.org/14423/diff/1/2
File src/codegen-ia32.h (right):

http://codereview.chromium.org/14423/diff/1/2#newcode86
Line 86: // stack and then spill the stack frame.  This function is used
temporarily while
Is this line too long?

http://codereview.chromium.org/14423/diff/1/3
File src/virtual-frame-ia32.h (right):

http://codereview.chromium.org/14423/diff/1/3#newcode187
Line 187: class SpilledScope BASE_EMBEDDED {
As we discussed, it would be nice to have a spilled scope block calls to the
Load functions that do not spill at the end.

[Kevin Millikin (Chromium), 2008-12-15 15:46:18]

http://codereview.chromium.org/14423/diff/1/2
File src/codegen-ia32.h (right):

http://codereview.chromium.org/14423/diff/1/2#newcode86
Line 86: // stack and then spill the stack frame.  This function is used
temporarily while
On 2008/12/15 09:43:12, Mads Ager wrote:
> Is this line too long?

Yes, fixed.

http://codereview.chromium.org/14423/diff/1/3
File src/virtual-frame-ia32.h (right):

http://codereview.chromium.org/14423/diff/1/3#newcode187
Line 187: class SpilledScope BASE_EMBEDDED {
On 2008/12/15 09:43:12, Mads Ager wrote:
> As we discussed, it would be nice to have a spilled scope block calls to the
> Load functions that do not spill at the end.

Done.  SpilledScope now sets a flag in the code generator.  The ...AndSpill
versions should only be called when it's set, and the raw versions when it's
cleared.  It adds some extra (usually no-op) spills where we were either using
an ...AndSpill function to establish a region of spill code, or allowing one of
the raw functions to end one.

Since all of those will go away eventually, I think it's OK.