_mm_storeu_si16 and _mm_storeu_si64 are missing but not documented as such

[Lokathor]

• _mm_storeu_si16
• _mm_storeu_si64

They're both missing from core::arch::x86 and core::arch::x86_64.

It might be fine in a sense that they're missing, since there's still also core::arch::x86_64::_mm_storeu_si128 which is basically the same thing, but either they should be added in for completeness or there should be documentation on the fact that they're deliberately missing.

[lassade]

It isn't exactly the same thing, _mm_storeu_si64 stores only the frist 8 bytes, while _mm_storeu_si128 store all 16;

Some algorithms may require it, like a base64 decoder: processing 16 hexadecimal chars yields 8 bytes of data, having this intrinsic will greatly reduce the complexity of the final code;

[lassade]

After bit of digging, I figure out that we only need these functions for completeness.

Who ever need these functions should instead use core::ptr::copy_nonoverlapping the assembly generated is optimum:

1. clang with intrinsic: https://godbolt.org/z/3GcGvE
2. rust version: https://play.rust-lang.org/?version=stable&mode=release&edition=2018&gist=9019494e9cdeae78aa76bdb1674c08fb

[TDecking]

These intrinsics were added in rust-lang/stdarch#1594.

[workingjubilee]

cool! closing this issue then?