VM: Improve and fix double negation on MIPS.

VM: Improve and fix double negation on MIPS.

Add neg-instruction to the MIPS backend and use it instead of mul with -1.

Fixes the test that requires that identical(nan, -nan) == false.

TEST=language/nan_identical_test
R=zra@google.com

Committed: https://github.com/dart-lang/sdk/commit/64261cbb9da9e41a81d8d4ecdb7681817b71df97

[Florian Schneider, 2015-10-29 20:27:04]

fschneider@google.com changed reviewers:
+ zra@google.com

[zra, 2015-10-29 21:16:14]

lgtm

https://codereview.chromium.org/1425143003/diff/1/tests/language/nan_identica...
File tests/language/nan_identical_test.dart (right):

https://codereview.chromium.org/1425143003/diff/1/tests/language/nan_identica...
tests/language/nan_identical_test.dart:1: // Copyright (c) 2015, the Dart
project authors.  Please see the AUTHORS file
It looks like this test will fail if we ever try to run on an architecture
configured to canonicalize NaNs. Whatever we pull out of the byte array into an
FP register will be the canonicalized value. V8 has chosen to turn that bit on
on ARM64 as an optimization, if I understand this CL correctly:
https://codereview.chromium.org/255343004/

I guess this test failing well let us know. Not sure if that's the right way to
find out. Maybe add a comment here about the problem?

[Florian Schneider, 2015-10-29 21:34:50]

Committed patchset #1 (id:1) manually as
64261cbb9da9e41a81d8d4ecdb7681817b71df97 (presubmit successful).