[libc++] Optimize num_get integral functions

Author: philnik777

libcxx/docs/ReleaseNotes/21.rst

@@ -72,6 +72,8 @@ Improvements and New Features
 
 - The ``num_put::do_put`` integral overloads have been optimized, resulting in a performance improvement of up to 2.4x.
 
+- The ``num_get::do_get`` integral overloads have been optimized, resulting in a performance improvement of up to 2.8x.
+
 - The ``std::stable_sort`` algorithm uses radix sort for floating-point types now, which can improve the performance
   up to 10x, depending on type of sorted elements and the initial state of the sorted array.
 

libcxx/include/CMakeLists.txt

@@ -516,7 +516,6 @@ set(files
   __locale_dir/locale_base_api.h
   __locale_dir/locale_base_api/bsd_locale_fallbacks.h
   __locale_dir/locale_base_api/ibm.h
-  __locale_dir/locale_base_api/musl.h
   __locale_dir/locale_base_api/openbsd.h
   __locale_dir/messages.h
   __locale_dir/money.h

libcxx/include/__algorithm/simd_utils.h

@@ -114,11 +114,37 @@ 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]..., ((void)_ZeroIndices, 0)...};
+  }(make_index_sequence<_Np>{}, make_index_sequence<__simd_vector_size_v<_VecT> - _Np>{});
+}
+
+// Create a vector where every elements is __val
+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 <class _Tp, size_t _Np>
 [[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI bool __all_of(__simd_vector<_Tp, _Np> __vec) noexcept {
   return __builtin_reduce_and(__builtin_convertvector(__vec, __simd_vector<bool, _Np>));
 }
 
+template <class _Tp, size_t _Np>
+[[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI bool __none_of(__simd_vector<_Tp, _Np> __vec) noexcept {
+  return !__builtin_reduce_or(__builtin_convertvector(__vec, __simd_vector<bool, _Np>));
+}
+
 template <class _Tp, size_t _Np>
 [[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI size_t __find_first_set(__simd_vector<_Tp, _Np> __vec) noexcept {
   using __mask_vec = __simd_vector<bool, _Np>;

libcxx/include/__locale_dir/locale_base_api.h

@@ -57,8 +57,6 @@
 //  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
@@ -106,7 +104,6 @@
 //
 //  int     __snprintf(char*, size_t, __locale_t, const char*, ...); // required by the headers
 //  int     __asprintf(char**, __locale_t, const char*, ...);        // required by the headers
-//  int     __sscanf(const char*, __locale_t, const char*, ...);     // required by the headers
 // }
 
 #if _LIBCPP_HAS_LOCALIZATION
@@ -131,8 +128,6 @@
 #      include <__locale_dir/locale_base_api/ibm.h>
 #    elif defined(__OpenBSD__)
 #      include <__locale_dir/locale_base_api/openbsd.h>
-#    elif defined(__wasi__) || _LIBCPP_HAS_MUSL_LIBC
-#      include <__locale_dir/locale_base_api/musl.h>
 #    endif
 
 #    include <__locale_dir/locale_base_api/bsd_locale_fallbacks.h>
@@ -192,15 +187,6 @@ 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
 //
@@ -302,11 +288,6 @@ _LIBCPP_HIDE_FROM_ABI _LIBCPP_VARIADIC_ATTRIBUTE_FORMAT(__printf__, 3, 4) int __
     char** __s, __locale_t __loc, const char* __format, _Args&&... __args) {
   return std::__libcpp_asprintf_l(__s, __loc, __format, std::forward<_Args>(__args)...);
 }
-template <class... _Args>
-_LIBCPP_HIDE_FROM_ABI _LIBCPP_VARIADIC_ATTRIBUTE_FORMAT(__scanf__, 3, 4) int __sscanf(
-    const char* __s, __locale_t __loc, const char* __format, _Args&&... __args) {
-  return std::__libcpp_sscanf_l(__s, __loc, __format, std::forward<_Args>(__args)...);
-}
 _LIBCPP_DIAGNOSTIC_POP
 #    undef _LIBCPP_VARIADIC_ATTRIBUTE_FORMAT
 

libcxx/include/__locale_dir/locale_base_api/bsd_locale_fallbacks.h

@@ -125,16 +125,6 @@ inline _LIBCPP_ATTRIBUTE_FORMAT(__printf__, 3, 4) int __libcpp_asprintf_l(
   return __res;
 }
 
-inline _LIBCPP_ATTRIBUTE_FORMAT(__scanf__, 3, 4) int __libcpp_sscanf_l(
-    const char* __s, locale_t __l, const char* __format, ...) {
-  va_list __va;
-  va_start(__va, __format);
-  __locale_guard __current(__l);
-  int __res = vsscanf(__s, __format, __va);
-  va_end(__va);
-  return __res;
-}
-
 _LIBCPP_END_NAMESPACE_STD
 
 #endif // _LIBCPP___LOCALE_DIR_LOCALE_BASE_API_BSD_LOCALE_FALLBACKS_H

libcxx/include/__locale_dir/locale_base_api/ibm.h

@@ -53,11 +53,6 @@ struct __setAndRestore {
 
 // The following are not POSIX routines.  These are quick-and-dirty hacks
 // to make things pretend to work
-inline _LIBCPP_HIDE_FROM_ABI long long strtoll_l(const char* __nptr, char** __endptr, int __base, locale_t locale) {
-  __setAndRestore __newloc(locale);
-  return ::strtoll(__nptr, __endptr, __base);
-}
-
 inline _LIBCPP_HIDE_FROM_ABI double strtod_l(const char* __nptr, char** __endptr, locale_t locale) {
   __setAndRestore __newloc(locale);
   return ::strtod(__nptr, __endptr);
@@ -73,12 +68,6 @@ inline _LIBCPP_HIDE_FROM_ABI long double strtold_l(const char* __nptr, char** __
   return ::strtold(__nptr, __endptr);
 }
 
-inline _LIBCPP_HIDE_FROM_ABI unsigned long long
-strtoull_l(const char* __nptr, char** __endptr, int __base, locale_t locale) {
-  __setAndRestore __newloc(locale);
-  return ::strtoull(__nptr, __endptr, __base);
-}
-
 inline _LIBCPP_HIDE_FROM_ABI
 _LIBCPP_ATTRIBUTE_FORMAT(__printf__, 2, 0) int vasprintf(char** strp, const char* fmt, va_list ap) {
   const size_t buff_size = 256;