X64: Enable inline caching for calls, with some stubs stubbed out.

X64: Enable inline caching for calls, with some stubs stubbed out.

Committed: http://code.google.com/p/v8/source/detail?r=2481

[William Hesse, 2009-07-15 15:07:24]

A tiny review for you.

[Mads Ager (chromium), 2009-07-15 15:15:34]

LGTM

What does this do to the tests though?  I would hope that this change would make
a number of tests fail?  Need to add them to the test expectations?

[Mads Ager (chromium), 2009-07-15 15:16:25]

I hit the wrong version of reply.  Here is my comment.

http://codereview.chromium.org/149674/diff/1/2
File src/ic.cc (right):

http://codereview.chromium.org/149674/diff/1/2#newcode1251
Line 1251: /*
I guess we don't use the premonomorphic state for call ICs.  Either leave in the
method or completely remove it (also from the header file).

[William Hesse, 2009-07-16 08:24:32]

Comment addressed.
This change makes no tests fail, since returning InternalError from a stub
generator just leaves the IC state unchanged, as if there was an allocation
error while creating the stub.  So calls go from intialized to premonomorphic,
and then try to go to monomorphic or constant function or whatever the case is,
and fail to create the monomorphic stub, and remain premonomorphic.  This makes
it easy to implement the stubs one-by-one.

Do I have a LGTM?

http://codereview.chromium.org/149674/diff/1/2
File src/ic.cc (right):

http://codereview.chromium.org/149674/diff/1/2#newcode1251
Line 1251: /*
On 2009/07/15 15:16:25, Mads Ager wrote:
> I guess we don't use the premonomorphic state for call ICs.  Either leave in
the
> method or completely remove it (also from the header file).

We do have a premonomorphic state for call ICs, it just has the same code as the
initialized state, and differs only in the flags on the code object, I think.  I
will remove the function, and comment the use of GenerateInitialize for
premonomorphic.

[Kasper Lund, 2009-07-16 08:28:46]

LGTM.