RFC: Switch to ast ids (instead of positions) for type feedback.

RFC: Switch to ast ids (instead of positions) for type feedback.

The goal here is to avoid having several pieces of type feedback tied
to a single position. Since everything else already relied on ast ids,
it seemed natural to use them for type feedback as well[*]. The
implementation squeezes in a new reloc info mode (CODE_TARGET_WITH_ID)
which is used for calls to IC stubs from the code generated by the
full codegen. The new mode should (in most cases) use just one extra
byte per call in the encoded form. The type feedback oracle can simply
focus on this reloc info mode.

[*]: I did play with disambiguating positions, but because of the
implicit way they are set, it's hard to be sure all the cases are
fixed.

[fschneider, 2011-02-11 11:28:18]

Overall the change lgtm. Could you upload a rebased version?

It would be see how much it changes the size of our relocation info overall.

I'd consider not changing the type of ast ids and leave them as integers, since
it would make the change a lot smaller without changing the semantics.

http://codereview.chromium.org/5699002/diff/3001/src/assembler.cc
File src/assembler.cc (left):

http://codereview.chromium.org/5699002/diff/3001/src/assembler.cc#oldcode92
src/assembler.cc:92: // The encoding relies on the fact that there are less than
14
I'm confused by this comment in the existing code since the ASSERT below says <=
14

http://codereview.chromium.org/5699002/diff/3001/src/assembler.cc
File src/assembler.cc (right):

http://codereview.chromium.org/5699002/diff/3001/src/assembler.cc#newcode97
src/assembler.cc:97: // code_taget:         [6 bits pc delta] 01
--> code_target

http://codereview.chromium.org/5699002/diff/3001/src/assembler.cc#newcode99
src/assembler.cc:99: // code_taget_with_id: [6 bits pc delta] 10,
--> code_target

http://codereview.chromium.org/5699002/diff/3001/src/assembler.h
File src/assembler.h (right):

http://codereview.chromium.org/5699002/diff/3001/src/assembler.h#newcode202
src/assembler.h:202: NUMBER_OF_MODES,  // must be no greater than ?? - see
RelocInfoWriter
Update comment. Maybe a STATIC_ASSERT would be better than just a comment or
normal ASSERT.

http://codereview.chromium.org/5699002/diff/3001/src/v8globals.h
File src/v8globals.h (right):

http://codereview.chromium.org/5699002/diff/3001/src/v8globals.h#newcode108
src/v8globals.h:108: typedef unsigned AstId;
Maybe consider leaving ast id as integers (like before) - although I like having
a separate type AstId, I'm not sure if the typedef alone gives us much benefit
overall.

It would also affect less source lines if we leave the type for ast ids
unchanged (i.e. making it easier to merge).