[libcxx-commits] [libcxx] [libc++] Optimize num_get integral functions (PR #121795)

[Louis Dionne via libcxx-commits]

================
@@ -116,11 +116,36 @@ template <class _VecT, class _Iter>
   }(make_index_sequence<__simd_vector_size_v<_VecT>>{});
 }
 
+// Load the first _Np elements, zero the rest
+_LIBCPP_DIAGNOSTIC_PUSH
+_LIBCPP_CLANG_DIAGNOSTIC_IGNORED("-Wpsabi")
+template <class _VecT, size_t _Np, class _Iter>
+[[__nodiscard__]] _LIBCPP_ALWAYS_INLINE _LIBCPP_HIDE_FROM_ABI _VecT __partial_load(_Iter __iter) noexcept {
+  return [=]<size_t... _LoadIndices, size_t... _ZeroIndices>(
+             index_sequence<_LoadIndices...>, index_sequence<_ZeroIndices...>) _LIBCPP_ALWAYS_INLINE noexcept {
+    return _VecT{__iter[_LoadIndices]...};
+  }(make_index_sequence<_Np>{}, make_index_sequence<__simd_vector_size_v<_VecT> - _Np>{});
+}
+
+template <class _VecT>
+[[__nodiscard__]] _LIBCPP_ALWAYS_INLINE _LIBCPP_HIDE_FROM_ABI _VecT
+__broadcast(__simd_vector_underlying_type_t<_VecT> __val) {
----------------
ldionne wrote:

Documentation?

https://github.com/llvm/llvm-project/pull/121795