Add support for unary plus in hydrogen compiler.

Add support for unary plus in hydrogen compiler.

Also strength-reduction of unary minus.
Fixes issue 1248.

BUG=1248
Committed: http://code.google.com/p/v8/source/detail?r=7217

[Kevin Millikin (Chromium), 2011-03-14 14:18:46]

http://codereview.chromium.org/6685045/diff/1/src/hydrogen.cc
File src/hydrogen.cc (right):

http://codereview.chromium.org/6685045/diff/1/src/hydrogen.cc#newcode4536
src/hydrogen.cc:4536: instr = new HSub(graph_->GetConstant0(), value);
I'm suspicious of this case.  Isn't it true that:

-(0) = 0-0 = 0+(-0) = +0

?  We should probably have a test that checks the corner cases for unary minus.

There is type feedback for the stub in unoptimized code, so it seems better to
call HGraphBuilder::BuildBinaryOperation and get our hands on that feedback.

http://codereview.chromium.org/6685045/diff/1/src/hydrogen.cc#newcode4539
src/hydrogen.cc:4539: instr = new HSub(value, graph_->GetConstant0());
I'm also suspicious of this:

+(-0) = 0-(-0) = 0+0 = +0

[Lasse Reichstein, 2011-03-15 12:18:33]

I have changed it to using HMul, reorganized the function to avoid the long if
followed by the similar switch, and tried a micro-optimization of LMulI.

http://codereview.chromium.org/6685045/diff/1/src/hydrogen.cc
File src/hydrogen.cc (right):

http://codereview.chromium.org/6685045/diff/1/src/hydrogen.cc#newcode4536
src/hydrogen.cc:4536: instr = new HSub(graph_->GetConstant0(), value);
Argh, foiled again by -0.
Yes, you are correct.
Incidentally, it would work if we used a ConstantMinus0 instead, but then we
would lose smi-ness.
I guess multiplication is safer (and the slight speedup I saw was probably due
to suddently being able to ignore -0)

http://codereview.chromium.org/6685045/diff/1/src/hydrogen.cc#newcode4539
src/hydrogen.cc:4539: instr = new HSub(value, graph_->GetConstant0());
Agree, that's bad too.

[Kevin Millikin (Chromium), 2011-03-15 12:34:00]

LGTM.