Implement fast calls of functions in the presence of eval (if the eval...

Implement fast calls of functions in the presence of eval (if the eval
calls do not introduce new bindings).

The infrastructure is already in place for fast loads from context
slots in the presence of eval.  This change simply uses that
infrastructure for calls as well as loads.


Committed: http://code.google.com/p/v8/source/detail?r=4609

[Erik Corry, 2010-05-07 09:39:05]

LGTM apart from the behaviour with const.

http://codereview.chromium.org/2027002/diff/5001/6004
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2027002/diff/5001/6004#newcode3705
src/arm/codegen-arm.cc:3705: // JavaScript example: 'with (obj) foo(1, 2, 3)' 
// foo is in obj
Is this the right comment.  The code looks more related to 'eval' than 'with'. 
Perhaps should say "foo may be in obj."

http://codereview.chromium.org/2027002/diff/5001/6004#newcode3727
src/arm/codegen-arm.cc:3727: // This rules out argument loads.
This rules out argument loads since the eval forces arguments into an arguments
array.

http://codereview.chromium.org/2027002/diff/5001/6004#newcode3757
src/arm/codegen-arm.cc:3757: done.Bind();
At this point everything is spilled but the function and receiver are in r0 and
r1.  I think a comment to that effect is in order.

http://codereview.chromium.org/2027002/diff/5001/6003
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/2027002/diff/5001/6003#newcode2905
src/x64/codegen-x64.cc:2905: done.Branch(not_equal, &function);
Should this skip the Push and the LoadGlobalReceiver.

[Mads Ager (chromium), 2010-05-07 10:18:56]

Thanks Erik, good catch on the const issue! :)

http://codereview.chromium.org/2027002/diff/5001/6004
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2027002/diff/5001/6004#newcode3705
src/arm/codegen-arm.cc:3705: // JavaScript example: 'with (obj) foo(1, 2, 3)' 
// foo is in obj
On 2010/05/07 09:39:05, Erik Corry wrote:
> Is this the right comment.  The code looks more related to 'eval' than 'with'.

> Perhaps should say "foo may be in obj."

Well, this branch is for anything where the function variable is DYNAMIC, so the
example is a fine example for that.  I'll add an example with eval as well.

http://codereview.chromium.org/2027002/diff/5001/6004#newcode3727
src/arm/codegen-arm.cc:3727: // This rules out argument loads.
On 2010/05/07 09:39:05, Erik Corry wrote:
> This rules out argument loads since the eval forces arguments into an
arguments
> array.

Done.

http://codereview.chromium.org/2027002/diff/5001/6004#newcode3757
src/arm/codegen-arm.cc:3757: done.Bind();
On 2010/05/07 09:39:05, Erik Corry wrote:
> At this point everything is spilled but the function and receiver are in r0
and
> r1.  I think a comment to that effect is in order.

Done.

http://codereview.chromium.org/2027002/diff/5001/6003
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/2027002/diff/5001/6003#newcode2905
src/x64/codegen-x64.cc:2905: done.Branch(not_equal, &function);
On 2010/05/07 09:39:05, Erik Corry wrote:
> Should this skip the Push and the LoadGlobalReceiver.

Oops, no, it definitely should not.  Added regression test and fixed this on x64
and ia32.  Thanks!

[Erik Corry, 2010-05-07 10:38:31]

LGTM

http://codereview.chromium.org/2027002/diff/11001/12001
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2027002/diff/11001/12001#newcode3708
src/arm/codegen-arm.cc:3708: //  with (obj) foo(1, 2, 3)  // foo is in obj
As I understand it the point of this code is to go fast when foo could be in
obj, but in fact it isn't.  So I would prefer "foo may be in obj."  Note the
full stop.

http://codereview.chromium.org/2027002/diff/11001/12001#newcode3713
src/arm/codegen-arm.cc:3713: //    f();  // f could be in extension object
Full stop.