[Clang] Missing intrinsic for VU64 x VU64 + VU128 vec_msum in Clang altivec.h

[johnplatts]

GCC 11 and later has the vec_msum(__vector unsigned long long, __vector unsigned long long, __vector unsigned __int128) and vec_msum(__vector signed long long, __vector signed long long, __vector signed __int128) intrinsics that are missing from the Clang version of altivec.h.

The vec_msum(__vector unsigned long long, __vector unsigned long long, __vector unsigned __int128) is equivalent to the POWER9 vmsumudm instruction.

The GCC vec_msum(__vector signed long long, __vector signed long long, __vector signed __int128) intrinsic actually does vec_msum((__vector unsigned long)a, (__vector unsigned long)b, (__vector unsigned __int128)c).

[llvmorg-github-actions]

@llvm/issue-subscribers-clang-frontend

Author: John Platts (johnplatts)

Details

GCC 11 and later has the `vec_msum(__vector unsigned long long, __vector unsigned long long, __vector unsigned __int128)` and `vec_msum(__vector signed long long, __vector signed long long, __vector signed __int128)` intrinsics that are missing from the Clang version of altivec.h.

The vec_msum(__vector unsigned long long, __vector unsigned long long, __vector unsigned __int128) is equivalent to the POWER9 vmsumudm instruction.

The GCC vec_msum(__vector signed long long, __vector signed long long, __vector signed __int128) intrinsic actually does vec_msum((__vector unsigned long)a, (__vector unsigned long)b, (__vector unsigned __int128)c).