Missing ldar intrinsics for clang-cl targeting arm64 windows

[BenMoran98]

Hello Developers!

This is an issue regarding compilation of c source code that contains ARM64 based intrinsic using clang-cl from llvm toolchain. Does intrinsic like '__ldar8' and '__ldar64' are supported by clang-cl tool? I am trying to compile the source code that involve these intrinsic on ARM64 based PC.

My source :

Error:

Thanks!!!

[DavidSpickett]

I think we are missing these. Full list from MSVC is here - https://learn.microsoft.com/en-us/cpp/intrinsics/arm64-intrinsics?view=msvc-170.

Judging by https://github.com/numpy/numpy/blob/main/numpy/_core/src/common/npy_atomic.h, we need:

• __ldar8
• __ldar64
• __ldar32
• __ldar16, just to complete the set.

https://clang.llvm.org/docs/MSVCCompatibility.html#simd-and-instruction-set-intrinsic-behavior may be relevant if these instructions are part of an optional extension.

[DavidSpickett]

This gives all the supported intriniscs:

$ cat clang/include/clang/Basic/BuiltinsAArch64.def | grep ALL_MS | cut -d'(' -f2 | cut -d',' -f1
<...>
_CountLeadingOnes64
_CountLeadingSigns
_CountLeadingSigns64
_CountLeadingZeros
_CountLeadingZeros64
_CountOneBits
_CountOneBits64
__prefetch
__hlt

(normally I'd look for a header file but these are compiler builtins as I understand it)
Or look in ./lib/clang/20/include/intrin.h.

[DavidSpickett]

Related: #65405

[zafeerali943]

Related: #65405

Do you have any plans to add support for the above intrinsic?

[DavidSpickett]

The ones referenced in that issue are in fact already implemented, it's still open because it's referring to msvc intriniscs generally (I assume).

I don't see a reason not to support these ldar ones, I will have a go at implementing them.

[DavidSpickett]

Also requested by numpy/numpy#28106.

[DavidSpickett]

I will have a go at implementing them.

Unfortunately I no longer have time to look at this.

(and if you are here because you searched for the error message, the solution that numpy implemented for the issue linked above may be interesting to you)

[zafeerali943]

@DavidSpickett this issue can be closed as the following intrinsics are related with CPP atomics and this issue has been fixed!
Thanks

[DavidSpickett]

As I understand it, a solution was found for numpy that avoids the need for these specific intrinsics. clang-cl is still missing them, so I will keep this issue open.