Skip some conditional deopts for Div/Mul when all uses are truncating

Skip some conditional deopts for Div/Mul when all uses are truncating.

- set "can be minus zero" flag properly so minus-zero checks are skipped
- skip "integer result?" check in division code when uses are truncating
- drive-by cleanup: consolidated computation of kCanOverflow flag for Add/Sub into range inference phase

BUG=v8:2132
R=svenpanne@chromium.org

Committed: http://code.google.com/p/v8/source/detail?r=15060

[Jakob Kummerow, 2013-06-10 15:00:52]

PTAL.

[Sven Panne, 2013-06-11 07:59:29]

LGTM in the sense that it will probably work. Nevertheless, I don't like the
current chaos regarding -0. We mix up 2 totally unrelated things IMHO: If the
result of an operation can be -0 (by the definition of the operation in the
spec, can_be_minus_zero_) and if we actually care about the difference between
-0 and 0 (kBailoutOnMinusZero). We should clean this up in some different CL,
the current state of affairs makes me cry... :-P

https://codereview.chromium.org/16741002/diff/11001/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/16741002/diff/11001/src/hydrogen-instructions...
src/hydrogen-instructions.h:1003: bool
HasAtLeastOneUseWithFlagAndNoneWithout(Flag f);
This helper predicate is totally confusing and hard to make sense of. Remove it
and simply use a combination of CheckUsesForFlag and either uses()->Done() (more
efficient) or UseCount() at its single call site.

[Jakob Kummerow, 2013-06-11 11:44:06]

Committed patchset #2 manually as r15060 (presubmit successful).