[libcxx] Use generic builtins for popcount, clz and ctz

Use __builtin_popcountg instead of __buildin_popcount{l|ll}
Use __builtin_clzg instead of __buildin_clz{l|ll}
Use __builtin_ctzg instead of __builtin_ctz{l|ll}

Author: marcauberer

libcxx/include/__bit/countl.h

@@ -25,15 +25,27 @@ _LIBCPP_PUSH_MACROS
 _LIBCPP_BEGIN_NAMESPACE_STD
 
 _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_clz(unsigned __x) _NOEXCEPT {
+#if __has_builtin(__builtin_clzg)
+  return __builtin_clzg(__x);
+#else
   return __builtin_clz(__x);
+#endif
 }
 
 _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_clz(unsigned long __x) _NOEXCEPT {
+#if __has_builtin(__builtin_clzg)
+  return __builtin_clzg(__x);
+#else
   return __builtin_clzl(__x);
+#endif
 }
 
 _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_clz(unsigned long long __x) _NOEXCEPT {
+#if __has_builtin(__builtin_clzg)
+  return __builtin_clzg(__x);
+#else
   return __builtin_clzll(__x);
+#endif
 }
 
 #ifndef _LIBCPP_HAS_NO_INT128
@@ -47,8 +59,13 @@ inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_clz(__uint128_t __x)
   // - Any bits set:
   //   - The number of leading zeros of the input is the number of leading
   //     zeros in the high 64-bits.
+#  if __has_builtin(__builtin_clzg)
+  return ((__x >> 64) == 0) ? (64 + __builtin_clzg(static_cast<unsigned long long>(__x)))
+                            : __builtin_clzg(static_cast<unsigned long long>(__x >> 64));
+#  else
   return ((__x >> 64) == 0) ? (64 + __builtin_clzll(static_cast<unsigned long long>(__x)))
                             : __builtin_clzll(static_cast<unsigned long long>(__x >> 64));
+#  endif
 }
 #endif // _LIBCPP_HAS_NO_INT128
 

libcxx/include/__bit/countr.h

@@ -24,15 +24,27 @@ _LIBCPP_PUSH_MACROS
 _LIBCPP_BEGIN_NAMESPACE_STD
 
 _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_ctz(unsigned __x) _NOEXCEPT {
+#if __has_builtin(__builtin_ctzg)
+  return __builtin_ctzg(__x);
+#else
   return __builtin_ctz(__x);
+#endif
 }
 
 _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_ctz(unsigned long __x) _NOEXCEPT {
+#if __has_builtin(__builtin_ctzg)
+  return __builtin_ctzg(__x);
+#else
   return __builtin_ctzl(__x);
+#endif
 }
 
 _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_ctz(unsigned long long __x) _NOEXCEPT {
+#if __has_builtin(__builtin_ctzg)
+  return __builtin_ctzg(__x);
+#else
   return __builtin_ctzll(__x);
+#endif
 }
 
 template <class _Tp>

libcxx/include/__bit/popcount.h

@@ -24,15 +24,27 @@ _LIBCPP_PUSH_MACROS
 _LIBCPP_BEGIN_NAMESPACE_STD
 
 inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_popcount(unsigned __x) _NOEXCEPT {
+#if __has_builtin(__builtin_popcountg)
+  return __builtin_popcountg(__x);
+#else
   return __builtin_popcount(__x);
+#endif
 }
 
 inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_popcount(unsigned long __x) _NOEXCEPT {
+#if __has_builtin(__builtin_popcountg)
+  return __builtin_popcountg(__x);
+#else
   return __builtin_popcountl(__x);
+#endif
 }
 
 inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_popcount(unsigned long long __x) _NOEXCEPT {
+#if __has_builtin(__builtin_popcountg)
+  return __builtin_popcountg(__x);
+#else
   return __builtin_popcountll(__x);
+#endif
 }
 
 #if _LIBCPP_STD_VER >= 20

libcxx/src/include/ryu/d2s_intrinsics.h

@@ -249,7 +249,7 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline bool __multipleOfPowerOf2(const uint64_t __value, const uint32_t __p) {
   _LIBCPP_ASSERT_INTERNAL(__value != 0, "");
   _LIBCPP_ASSERT_INTERNAL(__p < 64, "");
-  // __builtin_ctzll doesn't appear to be faster here.
+  // __builtin_ctzll/__builtin_ctzg doesn't appear to be faster here.
   return (__value & ((1ull << __p) - 1)) == 0;
 }
 

libcxx/src/include/ryu/ryu.h

@@ -43,6 +43,7 @@
 // Avoid formatting to keep the changes with the original code minimal.
 // clang-format off
 
+#include <__bit/countr.h>
 #include <__charconv/chars_format.h>
 #include <__charconv/to_chars_result.h>
 #include <__config>
@@ -72,15 +73,15 @@ _LIBCPP_HIDE_FROM_ABI inline unsigned char _BitScanForward64(unsigned long* __in
   if (__mask == 0) {
     return false;
   }
-  *__index = __builtin_ctzll(__mask);
+  *__index = __libcpp_ctz(__mask);
   return true;
 }
 
 _LIBCPP_HIDE_FROM_ABI inline unsigned char _BitScanForward(unsigned long* __index, unsigned int __mask) {
   if (__mask == 0) {
     return false;
   }
-  *__index = __builtin_ctz(__mask);
+  *__index = __libcpp_ctz(__mask);
   return true;
 }
 #endif  // !_MSC_VER

libcxx/src/ryu/f2s.cpp

@@ -107,7 +107,7 @@ inline constexpr uint64_t __FLOAT_POW5_SPLIT[47] = {
 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline bool __multipleOfPowerOf2(const uint32_t __value, const uint32_t __p) {
   _LIBCPP_ASSERT_INTERNAL(__value != 0, "");
   _LIBCPP_ASSERT_INTERNAL(__p < 32, "");
-  // __builtin_ctz doesn't appear to be faster here.
+  // __builtin_ctz/__builtin_ctzg doesn't appear to be faster here.
   return (__value & ((1u << __p) - 1)) == 0;
 }