[clang] [libcxx] [clang & libcxx] constexpr pointer tagging (DO NOT MERGE) (PR #111861)

[Louis Dionne via cfe-commits]

Hana =?utf-8?q?Dusíková?= <hanicka at hanicka.net>,
Hana =?utf-8?q?Dusíková?= <hanicka at hanicka.net>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/111861 at github.com>


================
@@ -0,0 +1,517 @@
+// -*- C++ -*-
+//===----------------------------------------------------------------------===//
+//
+// 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___TAGGED_PTR_H
+#define _LIBCPP___TAGGED_PTR_H
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+#if _LIBCPP_STD_VER >= 26
+  
+#include <__config>
+#include <__type_traits/is_trivially_copyable.h>
+#include <__assert>
+#include "__bit/has_single_bit.h"
+#include <__type_traits/rank.h>
+#include "pointer_traits.h"
+#include <compare>
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+template <typename T, typename Y> concept convertible_to_from = std::convertible_to<Y, T> && std::convertible_to<T, Y>;
+  
+template <typename T> concept pointer_tagging_schema = requires(T::dirty_pointer payload, T::clean_pointer clean, T::tag_type tag) {
+  requires convertible_to_from<typename T::tag_type, uintptr_t>;
+  requires std::is_pointer_v<typename T::clean_pointer>;
+  
+  { T::encode_pointer_with_tag(clean, tag) } noexcept -> std::same_as<typename T::dirty_pointer>;
+  { T::recover_pointer(payload) } noexcept -> std::same_as<typename T::clean_pointer>;
+  { T::recover_value(payload) } noexcept -> std::same_as<typename T::tag_type>;
+};
+
+template <typename T> concept pointer_tagging_schema_with_aliasing = pointer_tagging_schema<T> && requires(T::dirty_pointer payload) {
+  { T::recover_aliasing_pointer(payload) } noexcept -> std::same_as<typename T::clean_pointer>;
+};
+
+// no-op schema so I can better explain how schemas work
+struct no_tag {
+  template <typename T, typename Tag> struct schema {
+    using clean_pointer = T *;
+    using dirty_pointer = void *;
+    using tag_type = Tag;
+
+    [[clang::always_inline]] static constexpr dirty_pointer encode_pointer_with_tag(clean_pointer _ptr, tag_type) noexcept {
+      return (dirty_pointer)_ptr;
+    }
+    [[clang::always_inline]] static constexpr clean_pointer recover_pointer(dirty_pointer _ptr) noexcept {
+      return (clean_pointer)_ptr;
+    }
+    [[clang::always_inline]] static constexpr clean_pointer recover_aliasing_pointer(dirty_pointer _ptr) noexcept {
+      return (clean_pointer)_ptr;
+    }
+    [[clang::always_inline]] static constexpr tag_type recover_value(dirty_pointer) noexcept {
+      return {};
+    }
+  };
+};
+
+// most basic schema for tagging
+// it lets user to provide their own mask
+template <uintptr_t Mask> struct bitmask_tag {
+  static constexpr uintptr_t _mask = Mask;
+
+  template <typename T, typename Tag> struct schema {
+    using clean_pointer = T *;
+    using dirty_pointer = void *;
+    using tag_type = Tag;
+
+    [[clang::always_inline]] static constexpr dirty_pointer encode_pointer_with_tag(clean_pointer _ptr, tag_type _value) noexcept {
+#if __has_builtin(__builtin_tag_pointer_mask_or)
+      return static_cast<dirty_pointer>(__builtin_tag_pointer_mask_or((void *)(_ptr), static_cast<uintptr_t>(_value), _mask));
+#else
+      return reinterpret_cast<dirty_pointer>((reinterpret_cast<uintptr_t>(_ptr) & static_cast<uintptr_t>(_mask)) | (static_cast<uintptr_t>(_value) & ~static_cast<uintptr_t>(_mask)));
+#endif
+    }
+    [[clang::always_inline]] static constexpr clean_pointer recover_pointer(dirty_pointer _ptr) noexcept {
+#if __has_builtin(__builtin_tag_pointer_mask)
+      return static_cast<clean_pointer>(__builtin_tag_pointer_mask((void *)_ptr, ~_mask));
+#else
+      return reinterpret_cast<clean_pointer>(reinterpret_cast<uintptr_t>(_ptr) & ~static_cast<uintptr_t>(_mask));
+#endif
+    }
+    [[clang::always_inline]] static constexpr tag_type recover_value(dirty_pointer _ptr) noexcept {
+#if __has_builtin(__builtin_tag_pointer_mask_as_int)
+      return static_cast<tag_type>(__builtin_tag_pointer_mask_as_int((void *)_ptr, _mask));
+#else
+      return static_cast<tag_type>(reinterpret_cast<uintptr_t>(_ptr) & static_cast<uintptr_t>(_mask));
+#endif
+    }
+  };
+};
+
+// schema which allows only pointer of custom provided minimal alignment 
+// otherwise it behaves as custom mask schema
+template <unsigned Alignment> struct custom_alignment_tag {
+  static constexpr uintptr_t mask = (static_cast<uintptr_t>(1u) << static_cast<uintptr_t>(Alignment)) - 1ull;
+  template <typename T, typename Tag> struct schema: bitmask_tag<mask>::template schema<T, Tag> {
+    using _underlying_schema =bitmask_tag<mask>::template schema<T, Tag>;
+  
+    using clean_pointer = _underlying_schema::clean_pointer;
+    using dirty_pointer = _underlying_schema::dirty_pointer;
+    using tag_type = _underlying_schema::tag_type;
+    
+    [[clang::always_inline]] static constexpr dirty_pointer encode_pointer_with_tag(clean_pointer _ptr, tag_type _value) noexcept {
+#if __has_builtin(__builtin_is_aligned)
+      _LIBCPP_ASSERT_ARGUMENT_WITHIN_DOMAIN(__builtin_is_aligned(_ptr, Alignment), "Pointer must be aligned by provided alignemt for tagging");
+#else
+      if !consteval {
+        _LIBCPP_ASSERT_ARGUMENT_WITHIN_DOMAIN(reinterpret_cast<uintptr_t>(std::addressof(_ptr)) % Alignment == 0, "Pointer must be aligned by provided alignemt for tagging");
+      }
+#endif
+      return _underlying_schema::encode_pointer_with_tag(_ptr, _value);
+    }
+    
+    using _underlying_schema::recover_pointer;
+    using _underlying_schema::recover_value;
+  };
+};
+
+// default scheme which gives only bits from alignment
+struct alignment_low_bits_tag {
+  template <typename T> static constexpr unsigned alignment = alignof(T);
+  template <typename T, typename Tag> using schema = typename custom_alignment_tag<alignment<T>>::template schema<T, Tag>;
+};
+
+// scheme which shifts bits to left by Bits bits and gives the space for tagging
+template <unsigned Bits> struct shift_tag {
+  static constexpr unsigned _shift = Bits;
+  static constexpr uintptr_t _mask = (uintptr_t{1u} << _shift) - 1u;
+
+  template <typename T, typename Tag> struct schema {
+    using clean_pointer = T *;
+    using dirty_pointer = void *;
+    using tag_type = Tag;
+
+    [[clang::always_inline]] static constexpr dirty_pointer encode_pointer_with_tag(clean_pointer _ptr, tag_type _value) noexcept {
+#if __has_builtin(__builtin_tag_pointer_shift_or)
+      return static_cast<dirty_pointer>(__builtin_tag_pointer_shift_or((void *)(_ptr), (uintptr_t)_value, _shift));
+#else
+      return reinterpret_cast<dirty_pointer>((reinterpret_cast<uintptr_t>(_ptr) << _shift) | (static_cast<uintptr_t>(_value) & ((1ull << static_cast<uintptr_t>(_shift)) - 1ull)));
+#endif
+    }
+    [[clang::always_inline]] static constexpr clean_pointer recover_pointer(dirty_pointer _ptr) noexcept {
+#if __has_builtin(__builtin_tag_pointer_unshift)
+      return static_cast<clean_pointer>(__builtin_tag_pointer_unshift((void *)_ptr, _shift));
+#else
+      return reinterpret_cast<clean_pointer>(reinterpret_cast<uintptr_t>(_ptr) >> _shift);
+#endif
+    }
+    [[clang::always_inline]] static constexpr tag_type recover_value(dirty_pointer _ptr) noexcept {
+#if __has_builtin(__builtin_tag_pointer_mask_as_int)
+      return static_cast<tag_type>(__builtin_tag_pointer_mask_as_int((void *)_ptr, _mask));
+#else
+      return static_cast<tag_type>(reinterpret_cast<uintptr_t>(_ptr) & static_cast<uintptr_t>(_mask));
+#endif      
+    }
+  };
+};
+
+// scheme which shifts pointer to left by 8 bits and give this space as guaranteed space for tagging
+struct low_byte_tag {
+  template <typename T, typename Tag> using schema = typename shift_tag<8>::template schema<T, Tag>;
+};
+
+// this will give user access to upper byte of pointer on aarch64
+// also it supports recovering aliasing pointer as no-op (fast-path)
+struct upper_byte_tag {
+  template <typename T> static constexpr unsigned _shift = sizeof(T *) * 8ull - 8ull;
+  template <typename T> static constexpr uintptr_t _mask = 0b1111'1111ull << _shift<T>;
+  
+  template <typename T, typename Tag> struct schema: bitmask_tag<_mask<T>>::template schema<T, Tag> {
+    using _underlying_schema = bitmask_tag<_mask<T>>::template schema<T, Tag>;
+    
+    using clean_pointer = _underlying_schema::clean_pointer;
+    using dirty_pointer = _underlying_schema::dirty_pointer;
+    using tag_type = _underlying_schema::tag_type;
+  
+    [[clang::always_inline]] static constexpr clean_pointer recover_aliasing_pointer(dirty_pointer _ptr) noexcept {
+      return (clean_pointer)_ptr;
+    }
+    
+    using _underlying_schema::encode_pointer_with_tag;
+    using _underlying_schema::recover_pointer;
+    using _underlying_schema::recover_value;
+  };
+};
+
+// improved version of previous aarch64 upper byte scheme
+// with added shifting tag value into position, so the tag doesn't need to know about exact position
+struct upper_byte_shifted_tag: upper_byte_tag { 
+  template <typename T, typename Tag> struct schema: upper_byte_tag::template schema<T, uintptr_t> {
+    using _underlying_schema = upper_byte_tag::template schema<T, uintptr_t>;
+    
+    using clean_pointer = _underlying_schema::clean_pointer;
+    using dirty_pointer = _underlying_schema::dirty_pointer;
+    using tag_type = Tag;
+  
+    [[clang::always_inline]] static constexpr dirty_pointer encode_pointer_with_tag(clean_pointer _ptr, tag_type _value) noexcept {
+      return _underlying_schema::encode_pointer_with_tag(_ptr, static_cast<uintptr_t>(_value) << upper_byte_tag::_shift<T>);
+    }
+    [[clang::always_inline]] static constexpr tag_type recover_value(dirty_pointer _ptr) noexcept {
+      return static_cast<tag_type>(_underlying_schema::recover_value(_ptr) >> upper_byte_tag::_shift<T>);
+    }
+    
+    using _underlying_schema::recover_pointer;
+    using _underlying_schema::recover_aliasing_pointer;
+  };
+};
+
+// forward declaration
+template <typename _T, typename _Tag = uintptr_t, typename _Schema = alignment_low_bits_tag> class tagged_ptr;
+
+
+template <typename _Schema, typename _T, typename _Tag = uintptr_t> constexpr auto tag_ptr(_T * _ptr, _Tag _tag = {}) noexcept {
+  return tagged_ptr<_T, _Tag, _Schema>{_ptr, _tag};
+}
+
+template <typename _T, typename _Tag, typename _Schema = alignment_low_bits_tag> constexpr auto tagged_pointer_cast(typename _Schema::template schema<_T, _Tag>::dirty_pointer _ptr) noexcept -> tagged_ptr<_T, _Tag, _Schema> {
+  using result_type = tagged_ptr<_T, _Tag, _Schema>;
+  return result_type{typename result_type::already_tagged_tag{_ptr}};
+}
+
+template <typename _Schema2, typename _T, typename _Tag, typename _Schema> constexpr auto scheme_pointer_cast(tagged_ptr<_T, _Tag, _Schema> in) noexcept {
+  return tagged_ptr<_T, _Tag, _Schema2>{in.pointer(), in.tag()};
+}
+
+template <typename _Y, typename _T, typename _Tag, typename _Schema> constexpr auto const_pointer_cast(tagged_ptr<_T, _Tag, _Schema> in) noexcept {
+  // TODO we can just use native pointer here
+  return tagged_ptr<_Y, _Tag, _Schema>{const_cast<_Y*>(in.pointer()), in.tag()};
+}
+
+template <typename _Y, typename _T, typename _Tag, typename _Schema> constexpr auto static_pointer_cast(tagged_ptr<_T, _Tag, _Schema> in) noexcept {
+  return tagged_ptr<_Y, _Tag, _Schema>{static_cast<_Y*>(in.pointer()), in.tag()};
+}
+
+template <typename _Y, typename _T, typename _Tag, typename _Schema> constexpr auto dynamic_pointer_cast(tagged_ptr<_T, _Tag, _Schema> in) noexcept {
+  return tagged_ptr<_Y, _Tag, _Schema>{dynamic_cast<_Y*>(in.pointer()), in.tag()};
+}
+
+template <typename _Y, typename _T, typename _Tag, typename _Schema> auto reinterpret_pointer_cast(tagged_ptr<_T, _Tag, _Schema> in) noexcept {
+  return tagged_ptr<_Y, _Tag, _Schema>{reinterpret_cast<_Y*>(in.pointer()), in.tag()};
+}
+
+
+// wrapper class containing the pointer value and provides access
+template <typename _T, typename _Tag, typename _Schema> class tagged_ptr {
+public:
+  using schema = typename _Schema::template schema<_T, _Tag>;
+  using dirty_pointer = typename schema::dirty_pointer;
+  using clean_pointer = typename schema::clean_pointer;
+  using tag_type = typename schema::tag_type;
+  
+  using value_type = std::remove_cvref_t<decltype(*std::declval<clean_pointer>())>;
----------------
ldionne wrote:

```suggestion
  using element_type = std::remove_cvref_t<decltype(*std::declval<clean_pointer>())>;
```

To be consistent with all other smart ptrs. Also you probably need to use `pointer_traits<clean_pointer>::element_type` here.

https://github.com/llvm/llvm-project/pull/111861