@zigc/lib 0.17.0-dev.135 → 0.17.0-dev.215

package/libcxx/src/atomic.cpp

@@ -9,8 +9,13 @@
 #include <__thread/timed_backoff_policy.h>
 #include <atomic>
 #include <climits>
+#include <cstddef>
+#include <cstdint>
+#include <cstring>
 #include <functional>
+#include <new>
 #include <thread>
+#include <type_traits>
 
 #include "include/apple_availability.h"
 
@@ -41,6 +46,11 @@
 // OpenBSD has no indirect syscalls
 #  define _LIBCPP_FUTEX(...) futex(__VA_ARGS__)
 
+#elif defined(_WIN32)
+
+#  include <memory>
+#  include <windows.h>
+
 #else // <- Add other operating systems here
 
 // Baseline needs no new headers
@@ -49,17 +59,34 @@
 
 #endif
 
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
 _LIBCPP_BEGIN_NAMESPACE_STD
 
+struct NoTimeout {};
+
 #ifdef __linux__
 
-static void
-__libcpp_platform_wait_on_address(__cxx_atomic_contention_t const volatile* __ptr, __cxx_contention_t __val) {
-  static constexpr timespec __timeout = {2, 0};
-  _LIBCPP_FUTEX(__ptr, FUTEX_WAIT_PRIVATE, __val, &__timeout, 0, 0);
+template <std::size_t _Size, class MaybeTimeout>
+static void __platform_wait_on_address(void const* __ptr, void const* __val, MaybeTimeout maybe_timeout_ns) {
+  static_assert(_Size == 4, "Can only wait on 4 bytes value");
+  alignas(__cxx_contention_t) char buffer[_Size];
+  std::memcpy(&buffer, const_cast<const void*>(__val), _Size);
+  static constexpr timespec __default_timeout = {2, 0};
+  timespec __timeout;
+  if constexpr (is_same_v<MaybeTimeout, NoTimeout>) {
+    __timeout = __default_timeout;
+  } else {
+    __timeout.tv_sec  = maybe_timeout_ns / 1'000'000'000;
+    __timeout.tv_nsec = maybe_timeout_ns % 1'000'000'000;
+  }
+  _LIBCPP_FUTEX(__ptr, FUTEX_WAIT_PRIVATE, *reinterpret_cast<__cxx_contention_t const*>(&buffer), &__timeout, 0, 0);
 }
 
-static void __libcpp_platform_wake_by_address(__cxx_atomic_contention_t const volatile* __ptr, bool __notify_one) {
+template <std::size_t _Size>
+static void __platform_wake_by_address(void const* __ptr, bool __notify_one) {
+  static_assert(_Size == 4, "Can only wake up on 4 bytes value");
   _LIBCPP_FUTEX(__ptr, FUTEX_WAKE_PRIVATE, __notify_one ? 1 : INT_MAX, 0, 0, 0);
 }
 
@@ -70,19 +97,41 @@ extern "C" int __ulock_wait(
 extern "C" int __ulock_wake(uint32_t operation, void* addr, uint64_t wake_value);
 
 // https://github.com/apple/darwin-xnu/blob/2ff845c2e033bd0ff64b5b6aa6063a1f8f65aa32/bsd/sys/ulock.h#L82
+#  define UL_COMPARE_AND_WAIT 1
 #  define UL_COMPARE_AND_WAIT64 5
 #  define ULF_WAKE_ALL 0x00000100
 
-static void
-__libcpp_platform_wait_on_address(__cxx_atomic_contention_t const volatile* __ptr, __cxx_contention_t __val) {
-  static_assert(sizeof(__cxx_atomic_contention_t) == 8, "Waiting on 8 bytes value");
-  __ulock_wait(UL_COMPARE_AND_WAIT64, const_cast<__cxx_atomic_contention_t*>(__ptr), __val, 0);
+template <std::size_t _Size, class MaybeTimeout>
+static void __platform_wait_on_address(void const* __ptr, void const* __val, MaybeTimeout maybe_timeout_ns) {
+  static_assert(_Size == 8 || _Size == 4, "Can only wait on 8 bytes or 4 bytes value");
+  auto __timeout_us = [&] {
+    if constexpr (is_same_v<MaybeTimeout, NoTimeout>) {
+      return uint32_t(0);
+    } else {
+      return std::max(static_cast<uint32_t>(maybe_timeout_ns / 1000), uint32_t(1));
+    }
+  }();
+  if constexpr (_Size == 4) {
+    alignas(uint32_t) char buffer[_Size];
+    std::memcpy(&buffer, const_cast<const void*>(__val), _Size);
+    __ulock_wait(
+        UL_COMPARE_AND_WAIT, const_cast<void*>(__ptr), *reinterpret_cast<uint32_t const*>(&buffer), __timeout_us);
+  } else {
+    alignas(uint64_t) char buffer[_Size];
+    std::memcpy(&buffer, const_cast<const void*>(__val), _Size);
+    __ulock_wait(
+        UL_COMPARE_AND_WAIT64, const_cast<void*>(__ptr), *reinterpret_cast<uint64_t const*>(&buffer), __timeout_us);
+  }
 }
 
-static void __libcpp_platform_wake_by_address(__cxx_atomic_contention_t const volatile* __ptr, bool __notify_one) {
-  static_assert(sizeof(__cxx_atomic_contention_t) == 8, "Waking up on 8 bytes value");
-  __ulock_wake(
-      UL_COMPARE_AND_WAIT64 | (__notify_one ? 0 : ULF_WAKE_ALL), const_cast<__cxx_atomic_contention_t*>(__ptr), 0);
+template <std::size_t _Size>
+static void __platform_wake_by_address(void const* __ptr, bool __notify_one) {
+  static_assert(_Size == 8 || _Size == 4, "Can only wake up on 8 bytes or 4 bytes value");
+
+  if constexpr (_Size == 4)
+    __ulock_wake(UL_COMPARE_AND_WAIT | (__notify_one ? 0 : ULF_WAKE_ALL), const_cast<void*>(__ptr), 0);
+  else
+    __ulock_wake(UL_COMPARE_AND_WAIT64 | (__notify_one ? 0 : ULF_WAKE_ALL), const_cast<void*>(__ptr), 0);
 }
 
 #elif defined(__FreeBSD__) && __SIZEOF_LONG__ == 8
@@ -92,119 +141,355 @@ static void __libcpp_platform_wake_by_address(__cxx_atomic_contention_t const vo
  * limit its use to architectures where long and int64_t are synonyms.
  */
 
-static void
-__libcpp_platform_wait_on_address(__cxx_atomic_contention_t const volatile* __ptr, __cxx_contention_t __val) {
-  _umtx_op(const_cast<__cxx_atomic_contention_t*>(__ptr), UMTX_OP_WAIT, __val, nullptr, nullptr);
+template <std::size_t _Size, class MaybeTimeout>
+static void __platform_wait_on_address(void const* __ptr, void const* __val, MaybeTimeout maybe_timeout_ns) {
+  static_assert(_Size == 8, "Can only wait on 8 bytes value");
+  // At the moment, FreeBSD is stuck on stable ABI, which only supports platform wait with __cxx_contention_t
+  // It is safe to reinterpret_cast the val as it is ever going to be passed a __cxx_contention_t under this ABI
+  // If in the future FreeBSD decides to experiment unstable ABI to support more types, this cast will no longer be
+  // safe.
+  __cxx_contention_t value = *reinterpret_cast<const __cxx_contention_t*>(__val);
+  if constexpr (is_same_v<MaybeTimeout, NoTimeout>) {
+    _umtx_op(const_cast<void*>(__ptr), UMTX_OP_WAIT, value, nullptr, nullptr);
+  } else {
+    timespec timeout{};
+    timeout.tv_sec  = maybe_timeout_ns / 1'000'000'000;
+    timeout.tv_nsec = maybe_timeout_ns % 1'000'000'000;
+
+    _umtx_op(const_cast<void*>(__ptr),
+             UMTX_OP_WAIT,
+             value,
+             reinterpret_cast<void*>(static_cast<uintptr_t>(sizeof(timeout))),
+             &timeout);
+  }
+}
+
+template <std::size_t _Size>
+static void __platform_wake_by_address(void const* __ptr, bool __notify_one) {
+  static_assert(_Size == 8, "Can only wake up on 8 bytes value");
+  _umtx_op(const_cast<void*>(__ptr), UMTX_OP_WAKE, __notify_one ? 1 : INT_MAX, nullptr, nullptr);
+}
+
+#elif defined(_WIN32)
+
+static void* win32_get_synch_api_function(const char* function_name) {
+  // Attempt to load the API set. Note that as per the Microsoft STL implementation, we assume this API is already
+  // loaded and accessible. While this isn't explicitly guaranteed by publicly available Win32 API documentation, it is
+  // true in practice, and may be guaranteed by internal documentation not released publicly. In any case the fact that
+  // the Microsoft STL made this assumption is reasonable basis to say that we can too. The alternative to this would be
+  // to use LoadLibrary, but then leak the module handle. We can't call FreeLibrary, as this would have to be triggered
+  // by a global static destructor, which would hang off DllMain, and calling FreeLibrary from DllMain is explicitly
+  // mentioned as not being allowed:
+  // https://learn.microsoft.com/en-us/windows/win32/dlls/dllmain
+  // Given the range of bad options here, we have chosen to mirror what Microsoft did, as it seems fair to assume that
+  // Microsoft will guarantee compatibility for us, as we are exposed to the same conditions as all existing Windows
+  // apps using the Microsoft STL VS2015/2017/2019/2022 runtimes, where Windows 7 support has not been excluded at
+  // compile time.
+  static auto module_handle = GetModuleHandleW(L"api-ms-win-core-synch-l1-2-0.dll");
+  if (module_handle == nullptr) {
+    return nullptr;
+  }
+
+  // Attempt to locate the function in the API and return the result to the caller. Note that the NULL return from this
+  // method is documented as being interchangeable with nullptr.
+  // https://devblogs.microsoft.com/oldnewthing/20180307-00/?p=98175
+  return reinterpret_cast<void*>(GetProcAddress(module_handle, function_name));
 }
 
-static void __libcpp_platform_wake_by_address(__cxx_atomic_contention_t const volatile* __ptr, bool __notify_one) {
-  _umtx_op(const_cast<__cxx_atomic_contention_t*>(__ptr), UMTX_OP_WAKE, __notify_one ? 1 : INT_MAX, nullptr, nullptr);
+template <std::size_t _Size, class MaybeTimeout>
+static void __platform_wait_on_address(void const* __ptr, void const* __val, MaybeTimeout maybe_timeout_ns) {
+  static_assert(_Size == 8, "Can only wait on 8 bytes value");
+  // WaitOnAddress was added in Windows 8 (build 9200)
+  static auto wait_on_address =
+      reinterpret_cast<BOOL(WINAPI*)(void*, PVOID, SIZE_T, DWORD)>(win32_get_synch_api_function("WaitOnAddress"));
+  if (wait_on_address != nullptr) {
+    auto timeout_ms = [&]() -> DWORD {
+      if constexpr (is_same_v<MaybeTimeout, NoTimeout>) {
+        return INFINITE;
+      } else {
+        uint64_t ms = maybe_timeout_ns / 1'000'000;
+        if (ms == 0 && maybe_timeout_ns > 100'000)
+          // Round up to 1ms if requested between 100us - 1ms
+          return 1;
+
+        return static_cast<DWORD>(std::min(static_cast<uint64_t>(INFINITE), ms));
+      }
+    }();
+    wait_on_address(const_cast<void*>(__ptr), const_cast<void*>(__val), _Size, timeout_ms);
+  } else {
+    std::chrono::nanoseconds timeout = std::chrono::nanoseconds(0);
+    if constexpr (!is_same_v<MaybeTimeout, NoTimeout>) {
+      timeout = std::chrono::nanoseconds(maybe_timeout_ns);
+    }
+    __libcpp_thread_poll_with_backoff(
+        [=]() -> bool { return std::memcmp(const_cast<const void*>(__ptr), __val, _Size) != 0; },
+        __libcpp_timed_backoff_policy(),
+        timeout);
+  }
+}
+
+template <std::size_t _Size>
+static void __platform_wake_by_address(void const* __ptr, bool __notify_one) {
+  static_assert(_Size == 8, "Can only wake up on 8 bytes value");
+  if (__notify_one) {
+    // WakeByAddressSingle was added in Windows 8 (build 9200)
+    static auto wake_by_address_single =
+        reinterpret_cast<void(WINAPI*)(PVOID)>(win32_get_synch_api_function("WakeByAddressSingle"));
+    if (wake_by_address_single != nullptr) {
+      wake_by_address_single(const_cast<void*>(__ptr));
+    } else {
+      // The fallback implementation of waking does nothing, as the fallback wait implementation just does polling, so
+      // there's nothing to do here.
+    }
+  } else {
+    // WakeByAddressAll was added in Windows 8 (build 9200)
+    static auto wake_by_address_all =
+        reinterpret_cast<void(WINAPI*)(PVOID)>(win32_get_synch_api_function("WakeByAddressAll"));
+    if (wake_by_address_all != nullptr) {
+      wake_by_address_all(const_cast<void*>(__ptr));
+    } else {
+      // The fallback implementation of waking does nothing, as the fallback wait implementation just does polling, so
+      // there's nothing to do here.
+    }
+  }
 }
 
 #else // <- Add other operating systems here
 
 // Baseline is just a timed backoff
 
-static void
-__libcpp_platform_wait_on_address(__cxx_atomic_contention_t const volatile* __ptr, __cxx_contention_t __val) {
+template <std::size_t _Size, class MaybeTimeout>
+static void __platform_wait_on_address(void const* __ptr, void const* __val, MaybeTimeout maybe_timeout_ns) {
+  std::chrono::nanoseconds timeout = std::chrono::nanoseconds(0);
+  if constexpr (!is_same_v<MaybeTimeout, NoTimeout>) {
+    timeout = std::chrono::nanoseconds(maybe_timeout_ns);
+  }
   __libcpp_thread_poll_with_backoff(
-      [=]() -> bool { return !__cxx_nonatomic_compare_equal(__cxx_atomic_load(__ptr, memory_order_relaxed), __val); },
-      __libcpp_timed_backoff_policy());
+      [=]() -> bool { return std::memcmp(const_cast<const void*>(__ptr), __val, _Size) != 0; },
+      __libcpp_timed_backoff_policy(),
+      timeout);
 }
 
-static void __libcpp_platform_wake_by_address(__cxx_atomic_contention_t const volatile*, bool) {}
+template <std::size_t _Size>
+static void __platform_wake_by_address(void const*, bool) {}
 
 #endif // __linux__
 
-static constexpr size_t __libcpp_contention_table_size = (1 << 8); /* < there's no magic in this number */
-
-struct alignas(64) /*  aim to avoid false sharing */ __libcpp_contention_table_entry {
-  __cxx_atomic_contention_t __contention_state;
-  __cxx_atomic_contention_t __platform_state;
-  inline constexpr __libcpp_contention_table_entry() : __contention_state(0), __platform_state(0) {}
-};
-
-static __libcpp_contention_table_entry __libcpp_contention_table[__libcpp_contention_table_size];
-
-static hash<void const volatile*> __libcpp_contention_hasher;
-
-static __libcpp_contention_table_entry* __libcpp_contention_state(void const volatile* p) {
-  return &__libcpp_contention_table[__libcpp_contention_hasher(p) & (__libcpp_contention_table_size - 1)];
-}
+// =============================
+// Local hidden helper functions
+// =============================
 
 /* Given an atomic to track contention and an atomic to actually wait on, which may be
    the same atomic, we try to detect contention to avoid spuriously calling the platform. */
 
-static void __libcpp_contention_notify(__cxx_atomic_contention_t volatile* __contention_state,
-                                       __cxx_atomic_contention_t const volatile* __platform_state,
-                                       bool __notify_one) {
-  if (0 != __cxx_atomic_load(__contention_state, memory_order_seq_cst))
+template <std::size_t _Size>
+static void
+__contention_notify(__cxx_atomic_contention_t* __waiter_count, void const* __address_to_notify, bool __notify_one) {
+  if (0 != __cxx_atomic_load(__waiter_count, memory_order_seq_cst))
     // We only call 'wake' if we consumed a contention bit here.
-    __libcpp_platform_wake_by_address(__platform_state, __notify_one);
-}
-static __cxx_contention_t
-__libcpp_contention_monitor_for_wait(__cxx_atomic_contention_t volatile* /*__contention_state*/,
-                                     __cxx_atomic_contention_t const volatile* __platform_state) {
-  // We will monitor this value.
-  return __cxx_atomic_load(__platform_state, memory_order_acquire);
-}
-static void __libcpp_contention_wait(__cxx_atomic_contention_t volatile* __contention_state,
-                                     __cxx_atomic_contention_t const volatile* __platform_state,
-                                     __cxx_contention_t __old_value) {
-  __cxx_atomic_fetch_add(__contention_state, __cxx_contention_t(1), memory_order_relaxed);
-  // https://github.com/llvm/llvm-project/issues/109290
+    __platform_wake_by_address<_Size>(__address_to_notify, __notify_one);
+}
+
+template <std::size_t _Size, class MaybeTimeout>
+static void __contention_wait(__cxx_atomic_contention_t* __waiter_count,
+                              void const* __address_to_wait,
+                              void const* __old_value,
+                              MaybeTimeout maybe_timeout_ns) {
+  __cxx_atomic_fetch_add(__waiter_count, __cxx_contention_t(1), memory_order_relaxed);
+  // https://llvm.org/PR109290
   // There are no platform guarantees of a memory barrier in the platform wait implementation
   __cxx_atomic_thread_fence(memory_order_seq_cst);
   // We sleep as long as the monitored value hasn't changed.
-  __libcpp_platform_wait_on_address(__platform_state, __old_value);
-  __cxx_atomic_fetch_sub(__contention_state, __cxx_contention_t(1), memory_order_release);
+  __platform_wait_on_address<_Size>(__address_to_wait, __old_value, maybe_timeout_ns);
+  __cxx_atomic_fetch_sub(__waiter_count, __cxx_contention_t(1), memory_order_release);
+}
+
+static constexpr size_t __contention_table_size = (1 << 8); /* < there's no magic in this number */
+
+static constexpr hash<void const*> __contention_hasher;
+
+// Waiter count table for all atomics with the correct size that use itself as the wait/notify address.
+
+struct alignas(
+    std::hardware_constructive_interference_size) /*  aim to avoid false sharing */ __contention_state_native {
+  __cxx_atomic_contention_t __waiter_count;
+  constexpr __contention_state_native() : __waiter_count(0) {}
+};
+
+static __contention_state_native __contention_table_native[__contention_table_size];
+
+static __cxx_atomic_contention_t* __get_native_waiter_count(void const* p) {
+  return &__contention_table_native[__contention_hasher(p) & (__contention_table_size - 1)].__waiter_count;
+}
+
+// Global contention table for all atomics with the wrong size that use the global table's atomic as wait/notify
+// address.
+
+struct alignas(
+    std::hardware_constructive_interference_size) /*  aim to avoid false sharing */ __contention_state_global {
+  __cxx_atomic_contention_t __waiter_count;
+  __cxx_atomic_contention_t __platform_state;
+  constexpr __contention_state_global() : __waiter_count(0), __platform_state(0) {}
+};
+
+static __contention_state_global __contention_table_global[__contention_table_size];
+
+static __contention_state_global* __get_global_contention_state(void const* p) {
+  return &__contention_table_global[__contention_hasher(p) & (__contention_table_size - 1)];
 }
 
 /* When the incoming atomic is the wrong size for the platform wait size, need to
    launder the value sequence through an atomic from our table. */
 
-static void __libcpp_atomic_notify(void const volatile* __location) {
-  auto const __entry = __libcpp_contention_state(__location);
+static void __atomic_notify_global_table(void const* __location) {
+  auto const __entry = __get_global_contention_state(__location);
   // The value sequence laundering happens on the next line below.
   __cxx_atomic_fetch_add(&__entry->__platform_state, __cxx_contention_t(1), memory_order_seq_cst);
-  __libcpp_contention_notify(
-      &__entry->__contention_state,
-      &__entry->__platform_state,
-      false /* when laundering, we can't handle notify_one */);
+  __contention_notify<sizeof(__cxx_atomic_contention_t)>(
+      &__entry->__waiter_count, &__entry->__platform_state, false /* when laundering, we can't handle notify_one */);
+}
+
+// =============================
+// New dylib exported symbols
+// =============================
+
+// global
+_LIBCPP_EXPORTED_FROM_ABI __cxx_contention_t __atomic_monitor_global(void const* __location) noexcept {
+  auto const __entry = __get_global_contention_state(__location);
+  return __cxx_atomic_load(&__entry->__platform_state, memory_order_acquire);
+}
+
+_LIBCPP_EXPORTED_FROM_ABI void
+__atomic_wait_global_table(void const* __location, __cxx_contention_t __old_value) noexcept {
+  auto const __entry = __get_global_contention_state(__location);
+  __contention_wait<sizeof(__cxx_atomic_contention_t)>(
+      &__entry->__waiter_count, &__entry->__platform_state, &__old_value, NoTimeout{});
+}
+
+_LIBCPP_EXPORTED_FROM_ABI void __atomic_wait_global_table_with_timeout(
+    void const* __location, __cxx_contention_t __old_value, uint64_t __timeout_ns) _NOEXCEPT {
+  auto const __entry = __get_global_contention_state(__location);
+  __contention_wait<sizeof(__cxx_atomic_contention_t)>(
+      &__entry->__waiter_count, &__entry->__platform_state, &__old_value, __timeout_ns);
+}
+
+_LIBCPP_EXPORTED_FROM_ABI void __atomic_notify_one_global_table(void const* __location) noexcept {
+  __atomic_notify_global_table(__location);
+}
+
+_LIBCPP_EXPORTED_FROM_ABI void __atomic_notify_all_global_table(void const* __location) noexcept {
+  __atomic_notify_global_table(__location);
+}
+
+// native
+
+template <std::size_t _Size>
+_LIBCPP_EXPORTED_FROM_ABI void __atomic_wait_native(void const* __address, void const* __old_value) noexcept {
+  __contention_wait<_Size>(__get_native_waiter_count(__address), __address, __old_value, NoTimeout{});
 }
+
+template <std::size_t _Size>
+_LIBCPP_EXPORTED_FROM_ABI void
+__atomic_wait_native_with_timeout(void const* __address, void const* __old_value, uint64_t __timeout_ns) noexcept {
+  __contention_wait<_Size>(__get_native_waiter_count(__address), __address, __old_value, __timeout_ns);
+}
+
+template <std::size_t _Size>
+_LIBCPP_EXPORTED_FROM_ABI void __atomic_notify_one_native(void const* __location) noexcept {
+  __contention_notify<_Size>(__get_native_waiter_count(__location), __location, true);
+}
+
+template <std::size_t _Size>
+_LIBCPP_EXPORTED_FROM_ABI void __atomic_notify_all_native(void const* __location) noexcept {
+  __contention_notify<_Size>(__get_native_waiter_count(__location), __location, false);
+}
+
+// ==================================================
+// Instantiation of the templates with supported size
+// ==================================================
+
+#if defined(_LIBCPP_ABI_ATOMIC_WAIT_NATIVE_BY_SIZE)
+
+#  define _INSTANTIATE(_SIZE)                                                                                          \
+    template _LIBCPP_EXPORTED_FROM_ABI void __atomic_wait_native<_SIZE>(void const*, void const*) noexcept;            \
+    template _LIBCPP_EXPORTED_FROM_ABI void __atomic_wait_native_with_timeout<_SIZE>(                                  \
+        void const*, void const*, uint64_t) noexcept;                                                                  \
+    template _LIBCPP_EXPORTED_FROM_ABI void __atomic_notify_one_native<_SIZE>(void const*) noexcept;                   \
+    template _LIBCPP_EXPORTED_FROM_ABI void __atomic_notify_all_native<_SIZE>(void const*) noexcept;
+
+_LIBCPP_NATIVE_PLATFORM_WAIT_SIZES(_INSTANTIATE)
+
+#  undef _INSTANTIATE
+
+#else // _LIBCPP_ABI_ATOMIC_WAIT_NATIVE_BY_SIZE
+
+template _LIBCPP_EXPORTED_FROM_ABI void
+__atomic_wait_native<sizeof(__cxx_contention_t)>(void const* __address, void const* __old_value) noexcept;
+
+template _LIBCPP_EXPORTED_FROM_ABI void __atomic_wait_native_with_timeout<sizeof(__cxx_contention_t)>(
+    void const* __address, void const* __old_value, uint64_t) noexcept;
+
+template _LIBCPP_EXPORTED_FROM_ABI void
+__atomic_notify_one_native<sizeof(__cxx_contention_t)>(void const* __location) noexcept;
+
+template _LIBCPP_EXPORTED_FROM_ABI void
+__atomic_notify_all_native<sizeof(__cxx_contention_t)>(void const* __location) noexcept;
+
+#endif // _LIBCPP_ABI_ATOMIC_WAIT_NATIVE_BY_SIZE
+
+// =============================================================
+// Old dylib exported symbols, for backwards compatibility
+// =============================================================
+_LIBCPP_DIAGNOSTIC_PUSH
+_LIBCPP_CLANG_DIAGNOSTIC_IGNORED("-Wmissing-prototypes")
+
 _LIBCPP_EXPORTED_FROM_ABI void __cxx_atomic_notify_one(void const volatile* __location) noexcept {
-  __libcpp_atomic_notify(__location);
+  __atomic_notify_global_table(const_cast<void const*>(__location));
 }
+
 _LIBCPP_EXPORTED_FROM_ABI void __cxx_atomic_notify_all(void const volatile* __location) noexcept {
-  __libcpp_atomic_notify(__location);
+  __atomic_notify_global_table(const_cast<void const*>(__location));
 }
+
 _LIBCPP_EXPORTED_FROM_ABI __cxx_contention_t __libcpp_atomic_monitor(void const volatile* __location) noexcept {
-  auto const __entry = __libcpp_contention_state(__location);
-  return __libcpp_contention_monitor_for_wait(&__entry->__contention_state, &__entry->__platform_state);
+  auto const __entry = __get_global_contention_state(const_cast<void const*>(__location));
+  return __cxx_atomic_load(&__entry->__platform_state, memory_order_acquire);
 }
+
 _LIBCPP_EXPORTED_FROM_ABI void
 __libcpp_atomic_wait(void const volatile* __location, __cxx_contention_t __old_value) noexcept {
-  auto const __entry = __libcpp_contention_state(__location);
-  __libcpp_contention_wait(&__entry->__contention_state, &__entry->__platform_state, __old_value);
+  auto const __entry = __get_global_contention_state(const_cast<void const*>(__location));
+  __contention_wait<sizeof(__cxx_atomic_contention_t)>(
+      &__entry->__waiter_count, &__entry->__platform_state, &__old_value, NoTimeout{});
 }
 
-/* When the incoming atomic happens to be the platform wait size, we still need to use the
-   table for the contention detection, but we can use the atomic directly for the wait. */
-
 _LIBCPP_EXPORTED_FROM_ABI void __cxx_atomic_notify_one(__cxx_atomic_contention_t const volatile* __location) noexcept {
-  __libcpp_contention_notify(&__libcpp_contention_state(__location)->__contention_state, __location, true);
+  auto __location_cast = const_cast<const void*>(static_cast<const volatile void*>(__location));
+  __contention_notify<sizeof(__cxx_atomic_contention_t)>(
+      __get_native_waiter_count(__location_cast), __location_cast, true);
 }
+
 _LIBCPP_EXPORTED_FROM_ABI void __cxx_atomic_notify_all(__cxx_atomic_contention_t const volatile* __location) noexcept {
-  __libcpp_contention_notify(&__libcpp_contention_state(__location)->__contention_state, __location, false);
-}
-// This function is never used, but still exported for ABI compatibility.
-_LIBCPP_EXPORTED_FROM_ABI __cxx_contention_t
-__libcpp_atomic_monitor(__cxx_atomic_contention_t const volatile* __location) noexcept {
-  return __libcpp_contention_monitor_for_wait(&__libcpp_contention_state(__location)->__contention_state, __location);
+  auto __location_cast = const_cast<const void*>(static_cast<const volatile void*>(__location));
+  __contention_notify<sizeof(__cxx_atomic_contention_t)>(
+      __get_native_waiter_count(__location_cast), __location_cast, false);
 }
+
 _LIBCPP_EXPORTED_FROM_ABI void
 __libcpp_atomic_wait(__cxx_atomic_contention_t const volatile* __location, __cxx_contention_t __old_value) noexcept {
-  __libcpp_contention_wait(&__libcpp_contention_state(__location)->__contention_state, __location, __old_value);
+  auto __location_cast = const_cast<const void*>(static_cast<const volatile void*>(__location));
+  __contention_wait<sizeof(__cxx_atomic_contention_t)>(
+      __get_native_waiter_count(__location_cast), __location_cast, &__old_value, NoTimeout{});
 }
 
+// this function is even unused in the old ABI
+_LIBCPP_EXPORTED_FROM_ABI __cxx_contention_t
+__libcpp_atomic_monitor(__cxx_atomic_contention_t const volatile* __location) noexcept {
+  return __cxx_atomic_load(__location, memory_order_acquire);
+}
+
+_LIBCPP_DIAGNOSTIC_POP
+
 _LIBCPP_END_NAMESPACE_STD
+
+_LIBCPP_POP_MACROS

package/libcxx/src/barrier.cpp

@@ -60,11 +60,12 @@ public:
 _LIBCPP_EXPORTED_FROM_ABI __barrier_algorithm_base* __construct_barrier_algorithm_base(ptrdiff_t& __expected) {
   return new __barrier_algorithm_base(__expected);
 }
-_LIBCPP_EXPORTED_FROM_ABI bool
-__arrive_barrier_algorithm_base(__barrier_algorithm_base* __barrier, __barrier_phase_t __old_phase) noexcept {
+_LIBCPP_EXPORTED_FROM_ABI bool __arrive_barrier_algorithm_base(
+    _LIBCPP_NOESCAPE __barrier_algorithm_base* __barrier, __barrier_phase_t __old_phase) noexcept {
   return __barrier->__arrive(__old_phase);
 }
-_LIBCPP_EXPORTED_FROM_ABI void __destroy_barrier_algorithm_base(__barrier_algorithm_base* __barrier) noexcept {
+_LIBCPP_EXPORTED_FROM_ABI void
+__destroy_barrier_algorithm_base(_LIBCPP_NOESCAPE __barrier_algorithm_base* __barrier) noexcept {
   delete __barrier;
 }
 

package/libcxx/src/charconv.cpp

@@ -14,17 +14,20 @@
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
-#ifndef _LIBCPP_ABI_DO_NOT_EXPORT_TO_CHARS_BASE_10
+#if _LIBCPP_AVAILABILITY_MINIMUM_HEADER_VERSION < 15
 
 namespace __itoa {
 
+_LIBCPP_DIAGNOSTIC_PUSH
+_LIBCPP_CLANG_DIAGNOSTIC_IGNORED("-Wmissing-prototypes")
+// These functions exist for ABI compatibility, so we don't ever want a declaration prior to the definition.
 _LIBCPP_EXPORTED_FROM_ABI char* __u32toa(uint32_t value, char* buffer) noexcept { return __base_10_u32(buffer, value); }
-
 _LIBCPP_EXPORTED_FROM_ABI char* __u64toa(uint64_t value, char* buffer) noexcept { return __base_10_u64(buffer, value); }
+_LIBCPP_DIAGNOSTIC_POP
 
 } // namespace __itoa
 
-#endif // _LIBCPP_ABI_DO_NOT_EXPORT_TO_CHARS_BASE_10
+#endif // _LIBCPP_AVAILABILITY_MINIMUM_HEADER_VERSION < 15
 
 // The original version of floating-point to_chars was written by Microsoft and
 // contributed with the following license.

package/libcxx/src/condition_variable_destructor.cpp

@@ -14,7 +14,7 @@
 #include <__config>
 #include <__thread/support.h>
 
-#if _LIBCPP_ABI_VERSION == 1 || !_LIBCPP_HAS_TRIVIAL_CONDVAR_DESTRUCTION
+#if _LIBCPP_AVAILABILITY_MINIMUM_HEADER_VERSION < 9 || !_LIBCPP_HAS_TRIVIAL_CONDVAR_DESTRUCTION
 #  define NEEDS_CONDVAR_DESTRUCTOR
 #endif
 

package/libcxx/src/error_category.cpp

@@ -8,7 +8,9 @@
 
 #include <__config>
 
-#ifdef _LIBCPP_DEPRECATED_ABI_LEGACY_LIBRARY_DEFINITIONS_FOR_INLINE_FUNCTIONS
+// This has technically been removed in LLVM 3.4
+#if !defined(_LIBCPP_OBJECT_FORMAT_COFF) && !defined(_LIBCPP_OBJECT_FORMAT_XCOFF) &&                                   \
+    _LIBCPP_AVAILABILITY_MINIMUM_HEADER_VERSION < 4
 #  define _LIBCPP_ERROR_CATEGORY_DEFINE_LEGACY_INLINE_FUNCTIONS
 #endif
 

package/libcxx/src/exception.cpp

@@ -9,20 +9,12 @@
 #define _LIBCPP_ENABLE_CXX20_REMOVED_UNCAUGHT_EXCEPTION
 #define _LIBCPP_DISABLE_DEPRECATION_WARNINGS
 
-#include <exception>
-#include <new>
-#include <typeinfo>
-
-#if defined(LIBCXXRT) || defined(LIBCXX_BUILDING_LIBCXXABI)
-#  include <cxxabi.h>
-using namespace __cxxabiv1;
-#  define HAVE_DEPENDENT_EH_ABI 1
-#endif
+#include <__config>
 
 #if defined(_LIBCPP_ABI_MICROSOFT)
 #  include "support/runtime/exception_msvc.ipp"
 #  include "support/runtime/exception_pointer_msvc.ipp"
-#elif defined(_LIBCPPABI_VERSION)
+#elif defined(LIBCXX_BUILDING_LIBCXXABI)
 #  include "support/runtime/exception_libcxxabi.ipp"
 #  include "support/runtime/exception_pointer_cxxabi.ipp"
 #elif defined(LIBCXXRT)

package/libcxx/src/experimental/time_zone.cpp

@@ -720,7 +720,7 @@ __get_sys_info(sys_seconds __time,
 // Iff the "offsets" are the same '__current.__end' is replaced with
 // '__next.__end', which effectively merges the two objects in one object. The
 // function returns true if a merge occurred.
-[[nodiscard]] bool __merge_continuation(sys_info& __current, const sys_info& __next) {
+[[nodiscard]] static bool __merge_continuation(sys_info& __current, const sys_info& __next) {
   if (__current.end != __next.begin)
     return false;
 

package/libcxx/src/experimental/tzdb.cpp

@@ -49,6 +49,8 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 
 namespace chrono {
 
+_LIBCPP_DIAGNOSTIC_PUSH
+_LIBCPP_CLANG_DIAGNOSTIC_IGNORED("-Wmissing-prototypes")
 // This function is weak so it can be overriden in the tests. The
 // declaration is in the test header test/support/test_tzdb.h
 _LIBCPP_WEAK string_view __libcpp_tzdb_directory() {
@@ -59,6 +61,7 @@ _LIBCPP_WEAK string_view __libcpp_tzdb_directory() {
   abort();
 #endif
 }
+_LIBCPP_DIAGNOSTIC_POP
 
 //===----------------------------------------------------------------------===//
 //                           Details
@@ -767,7 +770,7 @@ void __init_tzdb(tzdb& __tzdb, __tz::__rules_storage_type& __rules) {
   // On Linux systems it seems /etc/timezone is deprecated and being phased out.
   // This file is used when /etc/localtime does not exist, or when it exists but
   // is not a symlink. For more information and links see
-  // https://github.com/llvm/llvm-project/issues/105634
+  // https://llvm.org/PR105634
 
   string __name = chrono::__current_zone_environment();