Added support for property loads to the CFG builder and fast-mode

Added support for property loads to the CFG builder and fast-mode
compiler.


Committed: http://code.google.com/p/v8/source/detail?r=2653

[Kevin Millikin (Chromium), 2009-08-07 12:02:00]

Issues to be addressed in the future:

1. It's clear that much of the graph building code can be shared.  Might be
enough to have superclasses for one-operand, two-operand, and three-operand
instructions.

2. Same with codegen code.  We could go a long way by adding things like Push
and Call with a common semantics to the macro assemblers and using
platform-specific constant symbolic registers (eg, kAccumulator, kScratch, ...).

3. There is a pop/push sequence in the property loads which could be eliminated,
but it's probably better to make the ICs drop the receiver and key if there is
one.

[William Hesse, 2009-08-07 12:50:26]

LGTM.

http://codereview.chromium.org/165129/diff/1/2
File src/cfg.h (right):

http://codereview.chromium.org/165129/diff/1/2#newcode422
Line 422: // Load a property from a receiver, leaving the result in a location.
I like PropertyLoad better.  A PropertyRef to me means the EcmaScript standard
Reference internal type.

http://codereview.chromium.org/165129/diff/1/5
File src/x64/cfg-x64.cc (right):

http://codereview.chromium.org/165129/diff/1/5#newcode151
Line 151: __ pop(kScratchRegister);  // Discard key.
If you don't care what you pop to, popping to a legacy register is shorter than
popping to kScratchRegister (r10),
because pops are 64-bit by default, so you don't need a REX prefix except to
address a new register.

[Kevin Millikin (Chromium), 2009-08-07 13:25:33]

http://codereview.chromium.org/165129/diff/1/2
File src/cfg.h (right):

http://codereview.chromium.org/165129/diff/1/2#newcode422
Line 422: // Load a property from a receiver, leaving the result in a location.
On 2009/08/07 12:50:26, William Hesse wrote:
> I like PropertyLoad better.  A PropertyRef to me means the EcmaScript standard
> Reference internal type.
> 

Going with PropLoad.

http://codereview.chromium.org/165129/diff/1/5
File src/x64/cfg-x64.cc (right):

http://codereview.chromium.org/165129/diff/1/5#newcode151
Line 151: __ pop(kScratchRegister);  // Discard key.
On 2009/08/07 12:50:26, William Hesse wrote:
> If you don't care what you pop to, popping to a legacy register is shorter
than
> popping to kScratchRegister (r10),
> because pops are 64-bit by default, so you don't need a REX prefix except to
> address a new register.

Thanks for the tip, using rbx.