Second attempt: Improve our type feedback by recogizining never-executed IC c...

Second attempt: Improve our type feedback by recogizining never-executed IC calls for binary operations.

This is an improved version of my earlier change r5970. It avoids degrading the
non-optimized code.

Initially we emit a conditional branch that is either always- or never-taken
after a smi-check (depending on whether we test for smi for for non-smi)
Since test-eax always sets the carry-flag to 0 we use jump-if-carry and
jump-if-not-carry.

The first invocation of the stub patches a jc with a jz and
jnc with a jnz-instruction so that the code looks exactly as it was
without patching. The only difference is the test- or nop-instruction
after the IC-call.


Committed: http://code.google.com/p/v8/source/detail?r=6030

[Vitaly Repeshko, 2010-12-14 18:28:49]

I really like the approach!

How about hiding a bit more of the implementation details? We now rely pretty
heavily on the previous instruction being test, so maybe we should wrap this
stuff into a single function, e.g.:
JumpPatchSite patch_site(masm);
patch_site.JumpIfSmi(&call_stub);
or
patch_site.JumpIfNotSmi(&call_stub);
I think it'll also be easier to read.

-- Vitaly

[fschneider, 2010-12-15 11:39:26]

New version uploaded. Now I added two helpers JumpIfSmi and JumpIfNotSmi.

[Vitaly Repeshko, 2010-12-15 12:18:50]

LGTM

http://codereview.chromium.org/5763004/diff/16001/src/ia32/ic-ia32.cc
File src/ia32/ic-ia32.cc (right):

http://codereview.chromium.org/5763004/diff/16001/src/ia32/ic-ia32.cc#newcode...
src/ia32/ic-ia32.cc:2120: Condition cc = *jmp_address == 
Assembler::kJncShortOpcode
Nit: "(*jmp_address ==  Assembler::kJncShortOpcode)".