[libc++] Make std::align an inline function

libcxx/docs/ReleaseNotes/22.rst

@@ -83,6 +83,8 @@ Improvements and New Features
   iterators, resulting in a performance improvement for ``std::deque<short>`` and
   ``std::join_view<vector<vector<short>>>`` iterators.
 
+- The performance of ``align`` has been improved about 3x by making it an inline function.
+
 Deprecations and Removals
 -------------------------
 
@@ -99,6 +101,8 @@ Potentially breaking changes
   ``_LIBCPP_ABI_NO_ITERATOR_BASES``. If you are using this flag and care about ABI stability, you should set
   ``_LIBCPP_ABI_NO_REVERSE_ITERATOR_SECOND_MEMBER`` as well.
 
+- ``align`` has been removed from libc++ dynamic library in ABI v2. You could use the ABI flag ``_LIBCPP_DEPRECATED_ABI_LEGACY_LIBRARY_DEFINITIONS_FOR_INLINE_FUNCTIONS`` to reserve it.
+
 Announcements About Future Releases
 -----------------------------------
 
@@ -114,5 +118,7 @@ ABI Affecting Changes
   potentially inheriting from the types they wrap. At this point in time we are not aware of any ABI changes caused by
   this.
 
+- ``align`` has been removed from libc++ dynamic library in ABI v2. You could use the ABI flag ``_LIBCPP_DEPRECATED_ABI_LEGACY_LIBRARY_DEFINITIONS_FOR_INLINE_FUNCTIONS`` to reserve it.
+
 Build System Changes
 --------------------

libcxx/include/__memory/align.h

@@ -11,15 +11,41 @@
 
 #include <__config>
 #include <__cstddef/size_t.h>
+#include <cstdint>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
 #endif
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
+__attribute__((always_inline)) inline void* __align_impl(size_t __align, size_t __sz, void*& __ptr, size_t& __space) {
+  void* __r = nullptr;
+  if (__sz <= __space) {
+    char* __p1 = static_cast<char*>(__ptr);
+    char* __p2 = reinterpret_cast<char*>(reinterpret_cast<uintptr_t>(__p1 + (__align - 1)) & -__align);
+    size_t __d = static_cast<size_t>(__p2 - __p1);
+    if (__d <= __space - __sz) {
+      __r   = __p2;
+      __ptr = __r;
+      __space -= __d;
+    }
+  }
+  return __r;
+}
+
+#ifdef _LIBCPP_ALIGN_DEFINE_LEGACY_INLINE_FUNCTIONS
+
 _LIBCPP_EXPORTED_FROM_ABI void* align(size_t __align, size_t __sz, void*& __ptr, size_t& __space);
 
+#else
+
+inline _LIBCPP_HIDE_FROM_ABI void* align(size_t __align, size_t __sz, void*& __ptr, size_t& __space) {
+  return __align_impl(__align, __sz, __ptr, __space);
+}
+
+#endif // _LIBCPP_ALIGN_DEFINE_LEGACY_INLINE_FUNCTIONS
+
 _LIBCPP_END_NAMESPACE_STD
 
 #endif // _LIBCPP___MEMORY_ALIGN_H

libcxx/src/memory.cpp

@@ -8,6 +8,7 @@
 
 #include <__config>
 #ifdef _LIBCPP_DEPRECATED_ABI_LEGACY_LIBRARY_DEFINITIONS_FOR_INLINE_FUNCTIONS
+#  define _LIBCPP_ALIGN_DEFINE_LEGACY_INLINE_FUNCTIONS
 #  define _LIBCPP_SHARED_PTR_DEFINE_LEGACY_INLINE_FUNCTIONS
 #endif
 
@@ -132,19 +133,11 @@ __sp_mut& __get_sp_mut(const void* p) {
 
 #endif // _LIBCPP_HAS_THREADS
 
+#if defined(_LIBCPP_ALIGN_DEFINE_LEGACY_INLINE_FUNCTIONS)
+
 void* align(size_t alignment, size_t size, void*& ptr, size_t& space) {
-  void* r = nullptr;
-  if (size <= space) {
-    char* p1 = static_cast<char*>(ptr);
-    char* p2 = reinterpret_cast<char*>(reinterpret_cast<uintptr_t>(p1 + (alignment - 1)) & -alignment);
-    size_t d = static_cast<size_t>(p2 - p1);
-    if (d <= space - size) {
-      r   = p2;
-      ptr = r;
-      space -= d;
-    }
-  }
-  return r;
+  return __align_impl(alignment, size, ptr, space);
 }
+#endif // _LIBCPP_ALIGN_DEFINE_LEGACY_INLINE_FUNCTIONS
 
 _LIBCPP_END_NAMESPACE_STD

libcxx/test/benchmarks/memory/align.bench.cpp

@@ -0,0 +1,27 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+// UNSUPPORTED: c++03
+
+#include <memory>
+
+#include "benchmark/benchmark.h"
+#include "test_macros.h"
+
+static void BM_align(benchmark::State& state) {
+  char buffer[1024];
+  void* data = buffer + 123;
+  std::size_t sz{sizeof(buffer) - 123};
+
+  for (auto _ : state) {
+    benchmark::DoNotOptimize(std::align(state.range(), state.range(), data, sz));
+  }
+}
+BENCHMARK(BM_align)->Range(1, 256);
+
+BENCHMARK_MAIN();