v2.5.0

Added

• gcc: Add xfhalf, xfalthalf, xfhalfinx and xfhalthalfinx support. This
  means code generation in the backend, libgcc changes to make softfloat
  routines set the fcsr, itanium style name mangling rules for and some helper
  routines when truncating doubles.
• gcc: Add xfhalf and xfalthalf compile test
• binutils: Add xfhalfinx and xfalthalfinx support
• gdb/sim: Add xfhalf and xfhalfinx support
• gdb/sim: Add float and half float tests generated by patched TestFloat-3e
• newlib: Add support for zfinx (POSIX fenv)

Changed

• gcc, binutils: Conversion instructions between half float and other float
  modes are automatically inferred. Internally, they are stil handled separately
  to make changes such as selectively disabling them easier.

Fixed

• binutils: Add fcsr alias instructions for zfinx
• gdb/sim: NaN boxing for floats in double float registers
• gdb/sim: dynamic rounding mode didn't work at all
• gdb/sim: Add RISC-V NaN behavior (canonical NaN instead of complicated rules)
• gcc: Various typos in tests
• gcc: Hardware loops for the xcorev target are now disabled (due to wrong
  code generation)
• gcc: Nested looping code now properly results in nested hardware loop
  instructions