Why does Go not have intrinsic functions?

[Arthur Comte]

Intrinsic functions (functions that compile down to one hardware instruction) are common in (relatively) low-level languages. Go even supports embed ASM (though a weird flavour of it)
Can someone explain why we do not have an intrinsic function package? This seems like it would greatly ease the use of SIMD
Thanks in advance

[Axel Wagner]

Go does have them.

math/big compiles down to single instructions on platforms where it's supported. sync/atomic does as well.

--
You received this message because you are subscribed to the Google Groups "golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/golang-nuts/7d95b9b9-9b94-48c1-9784-c5b7c54bc6b9n%40googlegroups.com.

[Ian Lance Taylor]

On Thu, Apr 7, 2022 at 11:03 AM Arthur Comte <ad...@arthurcomte.com> wrote:
>
> Intrinsic functions (functions that compile down to one hardware instruction) are common in (relatively) low-level languages. Go even supports embed ASM (though a weird flavour of it)
> Can someone explain why we do not have an intrinsic function package? This seems like it would greatly ease the use of SIMD

Go does have a couple of intrinsic function packages: sync/atomic and
math/bits. Many of the functions in those packages are implemented
directly in the compiler.

What Go does not have is a generalized way to write intrinsic
functions apart from implementing them in the compiler. That was
https://go.dev/issue/17373, which was declined.

Go also does not have a SIMD package, but that is more because nobody
has figured out the right approach for such a thing in Go. There was
at least one attempt at https://go.dev/issue/35307, but that too was
declined.

Ian

[Sam Hughes]

FWIW, I took a stab at a SIMD-oriented feature (https://go.dev/issue/48499), but as @Ian%20Lance%20Taylor put it, it's about the right approach. I skewed too far towards convenience in what I proposed, gaining significant maintainability concerns.