Floating point negate instructions are still present in release 6, only
one format of neg is removed, NEG.PS. Others formats can be used and in
r6 they also change the sign of NaN-like operands as well. This makes r6
generated code simpler for Neg_d and Neg_s macroassembler functions.

BUG=

Review-Url: https://codereview.chromium.org/2285703002
Cr-Commit-Position: refs/heads/master@{#38943}