[libc++] Optimize num_get integral functions

Author: philnik777

libcxx/docs/ReleaseNotes/20.rst

@@ -73,6 +73,8 @@ Improvements and New Features
   optimized, resulting in a performance improvement of up to 2x for trivial element types (e.g., `std::vector<int>`),
   and up to 3.4x for non-trivial element types (e.g., `std::vector<std::vector<int>>`).
 
+- The ``num_get::do_get`` integral overloads have been optimized. resulting in a performance improvement of up to 2.8x.
+
 Deprecations and Removals
 -------------------------
 

libcxx/include/__algorithm/simd_utils.h

@@ -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) {
+  return [&]<std::size_t... _Indices>(index_sequence<_Indices...>) {
+    return _VecT{((void)_Indices, __val)...};
+  }(make_index_sequence<__simd_vector_size_v<_VecT>>());
+}
+_LIBCPP_DIAGNOSTIC_POP
+
 template <size_t _Np>
 [[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI bool __all_of(__simd_vector<bool, _Np> __vec) noexcept {
   return __builtin_reduce_and(__vec);
 }
 
+template <size_t _Np>
+[[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI bool __none_of(__simd_vector<bool, _Np> __vec) noexcept {
+  return !__builtin_reduce_or(__vec);
+}
+
 template <class _Tp, size_t _Np>
 [[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI auto __as_mask(__simd_vector<_Tp, _Np> __vec) noexcept {
   static_assert(!is_same<_Tp, bool>::value, "vector type should not be a bool!");

libcxx/include/__configuration/abi.h

@@ -57,8 +57,6 @@
 // because it changes the mangling of the virtual function located in the vtable, which
 // changes how it gets signed.
 #  define _LIBCPP_ABI_BAD_FUNCTION_CALL_GOOD_WHAT_MESSAGE
-// Enable optimized version of __do_get_(un)signed which avoids redundant copies.
-#  define _LIBCPP_ABI_OPTIMIZED_LOCALE_NUM_GET
 // Give reverse_iterator<T> one data member of type T, not two.
 // Also, in C++17 and later, don't derive iterator types from std::iterator.
 #  define _LIBCPP_ABI_NO_ITERATOR_BASES

libcxx/include/__locale_dir/locale_base_api.h

@@ -40,8 +40,6 @@
 //  float               __strtof(const char*, char**, __locale_t);
 //  double              __strtod(const char*, char**, __locale_t);
 //  long double         __strtold(const char*, char**, __locale_t);
-//  long long           __strtoll(const char*, char**, __locale_t);
-//  unsigned long long  __strtoull(const char*, char**, __locale_t);
 // }
 //
 // Character manipulation functions
@@ -160,15 +158,6 @@ inline _LIBCPP_HIDE_FROM_ABI long double __strtold(const char* __nptr, char** __
   return strtold_l(__nptr, __endptr, __loc);
 }
 
-inline _LIBCPP_HIDE_FROM_ABI long long __strtoll(const char* __nptr, char** __endptr, int __base, __locale_t __loc) {
-  return strtoll_l(__nptr, __endptr, __base, __loc);
-}
-
-inline _LIBCPP_HIDE_FROM_ABI unsigned long long
-__strtoull(const char* __nptr, char** __endptr, int __base, __locale_t __loc) {
-  return strtoull_l(__nptr, __endptr, __base, __loc);
-}
-
 //
 // Character manipulation functions
 //

libcxx/include/__locale_dir/support/bsd_like.h

@@ -61,15 +61,6 @@ inline _LIBCPP_HIDE_FROM_ABI long double __strtold(const char* __nptr, char** __
   return ::strtold_l(__nptr, __endptr, __loc);
 }
 
-inline _LIBCPP_HIDE_FROM_ABI long long __strtoll(const char* __nptr, char** __endptr, int __base, __locale_t __loc) {
-  return ::strtoll_l(__nptr, __endptr, __base, __loc);
-}
-
-inline _LIBCPP_HIDE_FROM_ABI unsigned long long
-__strtoull(const char* __nptr, char** __endptr, int __base, __locale_t __loc) {
-  return ::strtoull_l(__nptr, __endptr, __base, __loc);
-}
-
 //
 // Character manipulation functions
 //

libcxx/include/__locale_dir/support/windows.h

@@ -174,14 +174,6 @@ inline _LIBCPP_HIDE_FROM_ABI double __strtod(const char* __nptr, char** __endptr
   return ::_strtod_l(__nptr, __endptr, __loc);
 }
 
-inline _LIBCPP_HIDE_FROM_ABI long long __strtoll(const char* __nptr, char** __endptr, int __base, __locale_t __loc) {
-  return ::_strtoi64_l(__nptr, __endptr, __base, __loc);
-}
-inline _LIBCPP_HIDE_FROM_ABI unsigned long long
-__strtoull(const char* __nptr, char** __endptr, int __base, __locale_t __loc) {
-  return ::_strtoui64_l(__nptr, __endptr, __base, __loc);
-}
-
 //
 // Character manipulation functions
 //