* POSIX thread functions that deal with thread scheduling.

* Helper functions to convert between fp16 (stored as uint16_t) and float. * The half-precision format is assumed to follow IEEE 754.

C++/Linux高性能的技术我之前整理了一些，包括SIMD/CUDA之类的技术，但是和deepseek确认了一下它说SIMD/CUDA不适合高频交易，假设它说的是对的，那我就随便说说其它的可选技术。