Derive synthetic type bounds for expressions

Derive synthetic type bounds for expressions

Currently synthesizes Smi as the lower bound for numeric operations (except for +, which might result in a string). That would need to change in places where we want to track constants as type feedback.

Does not do anything about variables yet.

R=jkummerow@chromium.org
BUG=

Committed: http://code.google.com/p/v8/source/detail?r=15422

[Jakob Kummerow, 2013-06-26 15:09:23]

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc
File src/typing.cc (right):

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode251
src/typing.cc:251: MergeLowerType(expr, Type::Union(
As discussed offline, inferring the lower bound as the union may be a little too
pessimistic. Also, it should be consistent with OR and AND below.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode319
src/typing.cc:319: expr->RecordTypeFeedback(oracle(), zone());
nit: {} while you're here

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode326
src/typing.cc:326: expr->RecordTypeFeedback(oracle(), zone());
nit: {}

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode346
src/typing.cc:346: // Lower type is None already
nit: trailing full stop please.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode413
src/typing.cc:413: if (expr->op() == Token::NOT) {
Combine this with the switch below maybe? If you want to keep calls to
RecordFooTypeFeedback separate from the logic related to default bounds that's
fine too.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode457
src/typing.cc:457: if (expr->is_prefix()) {
I don't think we need all this logic here. Regardless of prefix or postfix, the
lower types is always Smi and the upper type is always Number.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode501
src/typing.cc:501: // No lower type
nit: trailing full stop please.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode513
src/typing.cc:513: case Token::SHR:
I think you're mixing up SHR and SAR here.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode522
src/typing.cc:522: // No unique lower type (could be Smi or string)
nit: trailing full stop please.

[rossberg, 2013-06-26 16:18:59]

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc
File src/typing.cc (right):

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode251
src/typing.cc:251: MergeLowerType(expr, Type::Union(
On 2013/06/26 15:09:23, Jakob wrote:
> As discussed offline, inferring the lower bound as the union may be a little
too
> pessimistic. Also, it should be consistent with OR and AND below.

Actually, the correct thing to use here is intersection.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode319
src/typing.cc:319: expr->RecordTypeFeedback(oracle(), zone());
On 2013/06/26 15:09:23, Jakob wrote:
> nit: {} while you're here

Done.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode326
src/typing.cc:326: expr->RecordTypeFeedback(oracle(), zone());
On 2013/06/26 15:09:23, Jakob wrote:
> nit: {}

Done.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode346
src/typing.cc:346: // Lower type is None already
On 2013/06/26 15:09:23, Jakob wrote:
> nit: trailing full stop please.

Done.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode413
src/typing.cc:413: if (expr->op() == Token::NOT) {
On 2013/06/26 15:09:23, Jakob wrote:
> Combine this with the switch below maybe? If you want to keep calls to
> RecordFooTypeFeedback separate from the logic related to default bounds that's
> fine too.

Yeah, I'd like to keep it separate. (And we should perhaps get rid of the
conditional.)

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode457
src/typing.cc:457: if (expr->is_prefix()) {
On 2013/06/26 15:09:23, Jakob wrote:
> I don't think we need all this logic here. Regardless of prefix or postfix,
the
> lower types is always Smi and the upper type is always Number.

Good catch. Done.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode501
src/typing.cc:501: // No lower type
On 2013/06/26 15:09:23, Jakob wrote:
> nit: trailing full stop please.

Comment is obsolete now.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode513
src/typing.cc:513: case Token::SHR:
On 2013/06/26 15:09:23, Jakob wrote:
> I think you're mixing up SHR and SAR here.

Done.

https://codereview.chromium.org/17842004/diff/11001/src/typing.cc#newcode522
src/typing.cc:522: // No unique lower type (could be Smi or string)
On 2013/06/26 15:09:23, Jakob wrote:
> nit: trailing full stop please.

Done.

[Jakob Kummerow, 2013-06-27 14:41:07]

lgtm

[rossberg, 2013-06-27 14:54:50]

PTAL. Made the case for addition more precise.

[Jakob Kummerow, 2013-06-27 16:23:25]

still LGTM

[rossberg, 2013-07-01 13:59:24]

Committed patchset #7 manually as r15422 (presubmit successful).