[high_byte]

is HFT really measured by assembly op-level performance? I'd assume the algorithms efficiency is a magnitude more crucial than branch-stuff optimization and the likes. better hardware, better parallelism. lots of ground to cover before LIKELY/UNLIKELY macros are likely (lol) to contribute. at least to my educated (but unfamiliar with HFT) experience.

[chas]

I also don't have any direct HFT experience, but I do know that HFT firms hire FPGA designers (https://www.indeed.com/q-Fpga-Engineer-Trading-jobs.html) which suggests they really care about counting cycles and basically everything in is play from a performance perspective.

To contribute to the OP, https://godbolt.org/ is indispensible for high-performance C++ development. In specific, you can see under what circumstances the compiler optimizations apply to your code.