Jim Huang

#640 is considered as a feasible solution. @anthony-linaro, please check if it fits.

> I still need some time to investigate the cause that makes `_mm_dp_pd` and `_mm_dp_ps` failed. The above seems to be working. Is my understanding correct?

> If this PR is ok, I will remove the optimization option and turn this PR into ready for review Except for the `recast_f64` and `recast_i64` functions (the naming should...

> How about renaming them as `sse2neon_recast_u64_f64` and `sse2neon_recast_f64_s64`? It sounds pretty well.

> I think I can only separate it into a commit that introducing `OPTNONE` for some tests/intrinsics and another one handling the strict-aliasing errors. It makes sense to decouple `FORCE_INLINE_OPTNONE`...

Is it time to close #635 and #639 ? If so, refine the git commit messages accordingly.

Can you add some test cases in `tests/driver.sh` to reflect the proposed changes?

> Currently, left-to-right evaluation cannot be achieved. Per the C99 standard, left-to-right evaluation should be guaranteed. You should investigate the root cause preventing the proposed changes from being compliant with...

Thank @nosba0957 for contributing!

Reported by CI: > examples/syscall-steal2.c:35:15: warning: symbol 'syscall_kprobe' was not declared. Should it be static?