README

Float16 - half-precision floating point (binary16 from IEEE 754-2008)

This library implements simple float16 ⇔ float32 conversations that do not require special processor instructions.

Sequence functions:

• UnpackFloat16SequenceAuto(): float16 → float32 conversion for sequences
• PackFloat16SequenceAuto(): float32 → float16 conversion for sequences

For both of the functions above, {Pack,Unpack}Float16SequenceIntrisincs() will be called if F16C flag is set for CPU. Otherwise, a fallback function will be used.

SEARCH TAGS: float16, half precision floating point, fp16