Mask result with qNaN if input to floor is NaN

This CL masks the output with qNaN for floor of FP16 NaN values
i.e. mask the most significant mantissa bit with 1. In ARM hardware,
the rest of the mantissa bits are still processed when input is NaN,
therefore the mask comes at the end of the floor function.

Test: cts-tradefed  run singleCommand cts-dev --module CtsUtilTestCases --test android.util.cts.HalfTest
Test: Exhaustive testing of 0x0..0xFFFF input bits between the java
      implementation and FRINTM ARM instruction

Change-Id: Ida1a67fcff876b3d42546f97d12bf8cb26346d0e
Signed-off-by: Pranav Vashi <neobuddy89@gmail.com>
1 file changed