Implement a type recording ToBoolean IC.

Implement a type recording ToBoolean IC.

The IC records the set of types it has seen, e.g. {String} or {Boolean,
Undefined}, etc.  Note that in theory this could lead to a large number of
different ToBoolean ICs (512, to be exact, because we distinguish 9 types),
but in practice only a small handful of them are actually generated.

Currently the type recording part is only implemented on ia32, other platforms
continue to work like they did before, though.

Removed some dead code on the way.
Committed: http://code.google.com/p/v8/source/detail?r=8716

[danno, 2011-07-21 12:39:06]

LGTM with comments.

http://codereview.chromium.org/7473028/diff/1/src/code-stubs.cc
File src/code-stubs.cc (right):

http://codereview.chromium.org/7473028/diff/1/src/code-stubs.cc#newcode341
src/code-stubs.cc:341: if (Contains(UNDEFINED)) stream->Add("Undefined");
Doesn't the lack of separation between types end up in unreadable strings in the
output?

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc
File src/ia32/code-stubs-ia32.cc (right):

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:316: __ mov(eax, FieldOperand(eax,
String::kLengthOffset));
eax -> tos_?

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:362: __ cmp(eax, value);
This should be duplicated in the two if branches so that the else at the bottom
doesn't generate a spurious cmp.

[Mads Ager (chromium), 2011-07-21 12:52:47]

From a very quick look it LGTM too.

[fschneider, 2011-07-21 13:16:44]

DBC:

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc
File src/ia32/code-stubs-ia32.cc (right):

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:346: if (types_.Contains(INTERNAL_OBJECT)) {
Can you explain why we need the INTERNAL_OBJECT case?

[Vitaly Repeshko, 2011-07-21 13:17:22]

Drive by comment.

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc
File src/ia32/code-stubs-ia32.cc (right):

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:288: __ xor_(tos_, Operand(tos_));
The macro assembler has Set macro that emits the optimal instruction without
hurting readability.

[Sven Panne, 2011-07-21 13:49:28]

http://codereview.chromium.org/7473028/diff/1/src/code-stubs.cc
File src/code-stubs.cc (right):

http://codereview.chromium.org/7473028/diff/1/src/code-stubs.cc#newcode341
src/code-stubs.cc:341: if (Contains(UNDEFINED)) stream->Add("Undefined");
On 2011/07/21 12:39:07, danno wrote:
> Doesn't the lack of separation between types end up in unreadable strings in
the output?

This is intended and consistent with other conversion routines. The output is
nicely readable.

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc
File src/ia32/code-stubs-ia32.cc (right):

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:288: __ xor_(tos_, Operand(tos_));
On 2011/07/21 13:17:23, Vitaly Repeshko wrote:
> The macro assembler has Set macro that emits the optimal instruction without
hurting readability.

I didn't know that one, thanks. :-)

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:316: __ mov(eax, FieldOperand(eax,
String::kLengthOffset));
On 2011/07/21 12:39:07, danno wrote:
> eax -> tos_?

Ooops, fixed, same for Smi part above.

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:346: if (types_.Contains(INTERNAL_OBJECT)) {
On 2011/07/21 13:16:44, fschneider wrote:
> Can you explain why we need the INTERNAL_OBJECT case?

While we probably can't encounter internal objects at "normal" runtime (at least
I think so), they *do* occur at v8 startup, e.g. object/function templates etc.

http://codereview.chromium.org/7473028/diff/1/src/ia32/code-stubs-ia32.cc#new...
src/ia32/code-stubs-ia32.cc:362: __ cmp(eax, value);
On 2011/07/21 12:39:07, danno wrote:
> This should be duplicated in the two if branches so that the else at the
bottom
> doesn't generate a spurious cmp.

Done.