>In general, try to use the compiler to your advantage, but make sure to check the resulting assembly. In this case eqsub emitted a perfect sequence of cmeqs and subs, but you won’t always be so lucky. Occasionally it just won’t emit what you want, even if you use intrinsics, and in those cases writing in raw assembly is an option.<p>I strongly disagree with such a conclusion. If you're gonna twist and bend your source code so that the compiler generates the "desired assembly" for your target architecture, you should be instead writing assembly.
> It turns out that the AArch64 execution state (or is it the A64 instruction set? or the Armv8-A architecture? I hate this<p><a href="https://nickdesaulniers.github.io/blog/2023/03/10/disambiguating-arm/" rel="nofollow">https://nickdesaulniers.github.io/blog/2023/03/10/disambigua...</a>