Enables optimization on MIPS

Enables optimization on MIPS

- Exposes double precision rather than single precision floating point registers
  to architecture generic code. This entails a mechanical change in the assembler
  to floating point instructions.
- Adds slti, sltiu instructions for comparison with small immediates, and uses
  them in the Branch(Un)Signed* macros.
- Fixes a bug in the intermediate language comparison/branching instructions
  caused by using subu for comparison. Now it uses two slt instructions.
- Fixes register use bugs in intermediate language.
- More uses of NULLREG instead of loading Object::null()

R=regis@google.com

Committed: https://code.google.com/p/dart/source/detail?r=23190

[regis, 2013-05-24 22:32:59]

LGTM

https://codereview.chromium.org/16022011/diff/7001/runtime/vm/assembler_mips.cc
File runtime/vm/assembler_mips.cc (right):

https://codereview.chromium.org/16022011/diff/7001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.cc:214: slt(rd2, TMP1, rn);
Do you always need both results? Would it make sense to split this function?

[zra, 2013-05-24 23:25:47]

Thanks!

https://codereview.chromium.org/16022011/diff/7001/runtime/vm/assembler_mips.cc
File runtime/vm/assembler_mips.cc (right):

https://codereview.chromium.org/16022011/diff/7001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.cc:214: slt(rd2, TMP1, rn);
On 2013/05/24 22:32:59, regis wrote:
> Do you always need both results? Would it make sense to split this function?

Not always, no. Both are needed only when the object is a Smi and the comparison
is something other than == or !=. The problem is that in
ControlInstruction::EmitBranchOnCondition you won't know which of the split
CompareObjects came before, if I understand your suggestion correctly.

[zra, 2013-05-24 23:45:48]

Committed patchset #3 manually as r23190 (presubmit successful).