[libcxx-commits] [libcxx] [libc++] Add __pointer_int_pair (PR #94324)

[Louis Dionne via libcxx-commits]

================
@@ -0,0 +1,154 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___UTILITY_POINTER_INT_PAIR_H
+#define _LIBCPP___UTILITY_POINTER_INT_PAIR_H
+
+#include <__assert>
+#include <__bit/bit_log2.h>
+#include <__concepts/derived_from.h>
+#include <__config>
+#include <__tuple/tuple_element.h>
+#include <__tuple/tuple_size.h>
+#include <__type_traits/is_pointer.h>
+#include <__type_traits/is_unsigned.h>
+#include <__type_traits/is_void.h>
+#include <__type_traits/remove_pointer.h>
+#include <__utility/swap.h>
+#include <cstddef>
+#include <cstdint>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+#if _LIBCPP_STD_VER >= 23
+
+// A __pointer_int_pair is a pair of a pointer and an integral type. The lower bits of the pointer that are free
+// due to the alignment requirement of the pointee are used to store the integral type.
+//
+// This imposes a constraint on the number of bits available for the integral type -- the integral type can use
+// at most log2(alignof(T)) bits. This technique allows storing the integral type without additional storage
+// beyond that of the pointer itself, at the cost of some bit twiddling.
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+template <class>
+struct _PointerLikeTraits;
+
+template <class _Tp>
+  requires(!is_void_v<_Tp>)
+struct _PointerLikeTraits<_Tp*> {
+  static constexpr size_t __low_bits_available = std::__bit_log2(alignof(_Tp));
----------------
ldionne wrote:

MTE is an excellent observation. However, it seems they are using the high bits of pointers instead of the low bits like we're doing here. Furthermore, I think it is unlikely that MTE (or anything else) would break us given that LLVM and a bunch of other software is using tricks like these ones. Whatever they do, they have to make it work with code that uses these tricks. So I'm not overly concerned about this.

https://github.com/llvm/llvm-project/pull/94324