Hydrogen binop improvements

Hydrogen binop improvements

- Truncate oddball if possible.
- Support for StringAdd with only one String argument.
- Use constructor macro for HMul.
- Add ForceNumberType for HydrogenStubs to enforce input representations.

BUG=
R=svenpanne@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=16849

[Sven Panne, 2013-09-16 13:38:09]

https://codereview.chromium.org/23503058/diff/1/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/23503058/diff/1/src/hydrogen-instructions.h#n...
src/hydrogen-instructions.h:3884: if (left->IsForceRepresentation()) {
Why do we propagate back only to HForceRepresentation? This looks weird...

https://codereview.chromium.org/23503058/diff/1/src/hydrogen.cc
File src/hydrogen.cc (right):

https://codereview.chromium.org/23503058/diff/1/src/hydrogen.cc#newcode7698
src/hydrogen.cc:7698: } else if (expected->Is(Type::Signed32())) {
Just drop the "else", it's cleaner... :-)

https://codereview.chromium.org/23503058/diff/1/src/hydrogen.h
File src/hydrogen.h (right):

https://codereview.chromium.org/23503058/diff/1/src/hydrogen.h#newcode1326
src/hydrogen.h:1326: HControlInstruction* compare = new(zone()) Condition(p1);
This template and the following 2 should probably be rewritten by calling the
non-negated version plus a SwapSuccessors (or whatever you call it). Another
possibility would be an optional argument so AddCompare could do the right thing
without a fixup afterwards.

[oliv, 2013-09-16 15:44:43]

hey, ptal

https://codereview.chromium.org/23503058/diff/1/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/23503058/diff/1/src/hydrogen-instructions.h#n...
src/hydrogen-instructions.h:3884: if (left->IsForceRepresentation()) {
On 2013/09/16 13:38:09, Sven Panne wrote:
> Why do we propagate back only to HForceRepresentation? This looks weird...

It is actually even wrong in the general case, sorry

[oliv, 2013-09-19 16:44:57]

bump ;)

[Sven Panne, 2013-09-20 06:55:36]

lgtm

[oliv, 2013-09-20 07:21:41]

Committed patchset #5 manually as r16849 (presubmit successful).