The Half-precision extension is an optional architecture that extends the VFPv3 architecture. It provides VFP instructions that perform conversion between single-precision (32-bit) and half-precision (16-bit) floating-point numbers.

The half-precision instructions are only available on VFP systems that implement the half-precision extension. The VFP variants that implement the half-precision extension are VFPv3-FP16, VFPv3-D16-FP16, and VFPv4.