• Add a new Float16 value-based class encapsulating a short field holding IEEE 754 binary16 encoded value.
• Float16 type should have operation level parity with the existing Float type.
• For cleaner interface develop a helper class Float16Math containing intrinsic entry points for various intrinsified Float16 operations.
• Intrinsic entry points will receive and return unwrapped ‘short values’ holding IEEE 754 binary16 values.
• Add compiler support encompassing inline-expander, new IR, auto-vectorization, and efficient backend implementation for x86 and AArch64 targets.
• Develop new functional and performance validation tests covering newly added support.