Squashed 'externals/mcl/' changes from 761b7c05e..0172df743

0172df743 CMakeLists: Only add tests if MASTER_PROJECT
52e8dff62 0.1.11
fc8d745cc container: hmap fixups
5b5c0130d memory: Add overaligned_unique_ptr
c7c9bbd17 mcl: Increment version to 0.1.10
678aa32a8 assert: Handle expr strings separately
b38a9d2ef tests: Update to Catch 3.0.1
8aeacfe32 mcl: Increment version to 0.1.9
b468a2ab5 mcl: meta_byte: Split off meta_byte_group
d3ae1ae47 mcl: ihmap: Implement inline variant of hmap
5cbfe6eed mcl: hmap: Split detail into headers
ee7467677 mcl: hmap: Better default hash
f1d902ce9 mcl: hash: Add xmrx
322a221f0 mcl: hmap: Bugfix skip_empty_or_tombstone
689f393f7 mcl: hmap: x64 implementation
fa6ff746a mcl: hmap: Add generic meta_byte_group implementation
91e3073ad mcl: hmap: Add more member functions
4998335a5 mcl: Install only if master project
7ff4d2549 mcl: hmap prototype
416a2c6b5 mcl: clang-format: Adopt WebKit style bracing
d5a46fa70 mcl/assert: Flush stderr
e3b6cc79e externals: Update mcl to 0.1.7
190c68475 mcl: Build as PIC

git-subtree-dir: externals/mcl
git-subtree-split: 0172df74316351868c215f735e5a2538b10d71fb

.clang-format

@@ -34,7 +34,7 @@ BraceWrapping:
   AfterClass:      false
   AfterControlStatement: Never
   AfterEnum:       false
-  AfterFunction:   false
+  AfterFunction:   true
   AfterNamespace:  false
   AfterObjCDeclaration: false
   AfterStruct:     false
@@ -62,7 +62,7 @@ ColumnLimit:     0
 CommentPragmas:  '^ IWYU pragma:'
 CompactNamespaces: false
 ConstructorInitializerAllOnOneLineOrOnePerLine: true
-ConstructorInitializerIndentWidth: 8
+ConstructorInitializerIndentWidth: 4
 ContinuationIndentWidth: 4
 Cpp11BracedListStyle: true
 DeriveLineEnding: true

CMakeLists.txt

@@ -1,10 +1,18 @@
 cmake_minimum_required(VERSION 3.12 FATAL_ERROR)
 include(GNUInstallDirs)
 
-project(mcl LANGUAGES CXX VERSION 0.1.7)
+project(mcl LANGUAGES CXX VERSION 0.1.11)
+
+# Determine if we're built as a subproject (using add_subdirectory)
+# or if this is the master project.
+set(MASTER_PROJECT OFF)
+if (CMAKE_CURRENT_SOURCE_DIR STREQUAL CMAKE_SOURCE_DIR)
+  set(MASTER_PROJECT ON)
+endif()
 
 # Project options
-option(MCL_WARNINGS_AS_ERRORS "Warnings as errors" ON)
+option(MCL_WARNINGS_AS_ERRORS "Warnings as errors" ${MASTER_PROJECT})
+option(MCL_INSTALL "Enable installation" ${MASTER_PROJECT})
 
 # Default to a Release build
 if (NOT CMAKE_BUILD_TYPE)
@@ -85,7 +93,7 @@ endif()
 # Dependencies
 
 if (NOT TARGET Catch2::Catch2)
-    find_package(Catch2 QUIET)
+    find_package(Catch2 3 QUIET)
 endif()
 
 if (NOT TARGET fmt::fmt)
@@ -95,32 +103,33 @@ endif()
 # Project files
 
 add_subdirectory(src)
-if (TARGET Catch2::Catch2)
+if (TARGET Catch2::Catch2 AND MASTER_PROJECT)
     add_subdirectory(tests)
 endif()
 
 # Install instructions
+if (MCL_INSTALL)
+    include(GNUInstallDirs)
+    include(CMakePackageConfigHelpers)
 
-include(GNUInstallDirs)
+    install(TARGETS mcl EXPORT mclTargets)
-include(CMakePackageConfigHelpers)
+    install(EXPORT mclTargets
-
-install(TARGETS mcl EXPORT mclTargets)
-install(EXPORT mclTargets
         NAMESPACE merry::
         DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/mcl"
-)
+    )
 
-configure_package_config_file(CMakeModules/mclConfig.cmake.in
+    configure_package_config_file(CMakeModules/mclConfig.cmake.in
         mclConfig.cmake
         INSTALL_DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/mcl"
-)
+    )
-write_basic_package_version_file(mclConfigVersion.cmake
+    write_basic_package_version_file(mclConfigVersion.cmake
         COMPATIBILITY SameMajorVersion
-)
+    )
-install(FILES
+    install(FILES
         "${CMAKE_CURRENT_BINARY_DIR}/mclConfig.cmake"
         "${CMAKE_CURRENT_BINARY_DIR}/mclConfigVersion.cmake"
         DESTINATION "${CMAKE_INSTALL_LIBDIR}/cmake/mcl"
-)
+    )
 
-install(DIRECTORY include/ TYPE INCLUDE FILES_MATCHING PATTERN "*.hpp")
+    install(DIRECTORY include/ TYPE INCLUDE FILES_MATCHING PATTERN "*.hpp")
+endif()

include/mcl/assert.hpp

@@ -13,11 +13,12 @@
 
 namespace mcl::detail {
 
-[[noreturn]] void assert_terminate_impl(fmt::string_view msg, fmt::format_args args);
+[[noreturn]] void assert_terminate_impl(const char* expr_str, fmt::string_view msg, fmt::format_args args);
 
 template<typename... Ts>
-[[noreturn]] void assert_terminate(fmt::string_view msg, Ts... args) {
+[[noreturn]] void assert_terminate(const char* expr_str, fmt::string_view msg, Ts... args)
-    assert_terminate_impl(msg, fmt::make_format_args(args...));
+{
+    assert_terminate_impl(expr_str, msg, fmt::make_format_args(args...));
 }
 
 }  // namespace mcl::detail
@@ -32,7 +33,7 @@ template<typename... Ts>
             }                                                            \
         } else {                                                         \
             if (!(expr)) [[unlikely]] {                                  \
-                ::mcl::detail::assert_terminate(#expr);                  \
+                ::mcl::detail::assert_terminate(#expr, "(none)");        \
             }                                                            \
         }                                                                \
     }()
@@ -45,12 +46,12 @@ template<typename... Ts>
             }                                                                \
         } else {                                                             \
             if (!(expr)) [[unlikely]] {                                      \
-                ::mcl::detail::assert_terminate(#expr "\nMessage: " __VA_ARGS__); \
+                ::mcl::detail::assert_terminate(#expr, __VA_ARGS__);         \
             }                                                                \
         }                                                                    \
     }()
 
-#define ASSERT_FALSE(...) ::mcl::detail::assert_terminate("false\nMessage: " __VA_ARGS__)
+#define ASSERT_FALSE(...) ::mcl::detail::assert_terminate("false", __VA_ARGS__)
 
 #if defined(NDEBUG) || defined(MCL_IGNORE_ASSERTS)
 #    define DEBUG_ASSERT(expr) ASSUME(expr)

include/mcl/bit/bit_count.hpp

@@ -13,12 +13,14 @@
 namespace mcl::bit {
 
 template<BitIntegral T>
-inline size_t count_ones(T x) {
+inline size_t count_ones(T x)
+{
     return std::bitset<bitsizeof<T>>(x).count();
 }
 
 template<BitIntegral T>
-constexpr size_t count_leading_zeros(T x) {
+constexpr size_t count_leading_zeros(T x)
+{
     size_t result = bitsizeof<T>;
     while (x != 0) {
         x >>= 1;
@@ -28,7 +30,8 @@ constexpr size_t count_leading_zeros(T x) {
 }
 
 template<BitIntegral T>
-constexpr int highest_set_bit(T x) {
+constexpr int highest_set_bit(T x)
+{
     int result = -1;
     while (x != 0) {
         x >>= 1;
@@ -38,7 +41,8 @@ constexpr int highest_set_bit(T x) {
 }
 
 template<BitIntegral T>
-constexpr size_t lowest_set_bit(T x) {
+constexpr size_t lowest_set_bit(T x)
+{
     if (x == 0) {
         return bitsizeof<T>;
     }

include/mcl/bit/bit_field.hpp

@@ -13,7 +13,8 @@ namespace mcl::bit {
 
 /// Create a mask with `count` number of one bits.
 template<size_t count, BitIntegral T>
-constexpr T ones() {
+constexpr T ones()
+{
     static_assert(count <= bitsizeof<T>, "count larger than bitsize of T");
 
     if constexpr (count == 0) {
@@ -25,7 +26,8 @@ constexpr T ones() {
 
 /// Create a mask with `count` number of one bits.
 template<BitIntegral T>
-constexpr T ones(size_t count) {
+constexpr T ones(size_t count)
+{
     ASSERT_MSG(count <= bitsizeof<T>, "count larger than bitsize of T");
 
     if (count == 0) {
@@ -36,7 +38,8 @@ constexpr T ones(size_t count) {
 
 /// Create a mask of type T for bits [begin_bit, end_bit] inclusive.
 template<size_t begin_bit, size_t end_bit, BitIntegral T>
-constexpr T mask() {
+constexpr T mask()
+{
     static_assert(begin_bit <= end_bit, "invalid bit range (position of beginning bit cannot be greater than that of end bit)");
     static_assert(begin_bit < bitsizeof<T>, "begin_bit must be smaller than size of T");
     static_assert(end_bit < bitsizeof<T>, "end_bit must be smaller than size of T");
@@ -46,7 +49,8 @@ constexpr T mask() {
 
 /// Create a mask of type T for bits [begin_bit, end_bit] inclusive.
 template<BitIntegral T>
-constexpr T mask(size_t begin_bit, size_t end_bit) {
+constexpr T mask(size_t begin_bit, size_t end_bit)
+{
     ASSERT_MSG(begin_bit <= end_bit, "invalid bit range (position of beginning bit cannot be greater than that of end bit)");
     ASSERT_MSG(begin_bit < bitsizeof<T>, "begin_bit must be smaller than size of T");
     ASSERT_MSG(end_bit < bitsizeof<T>, "end_bit must be smaller than size of T");
@@ -56,91 +60,104 @@ constexpr T mask(size_t begin_bit, size_t end_bit) {
 
 /// Extract bits [begin_bit, end_bit] inclusive from value of type T.
 template<size_t begin_bit, size_t end_bit, BitIntegral T>
-constexpr T get_bits(T value) {
+constexpr T get_bits(T value)
+{
     constexpr T m = mask<begin_bit, end_bit, T>();
     return (value & m) >> begin_bit;
 }
 
 /// Extract bits [begin_bit, end_bit] inclusive from value of type T.
 template<BitIntegral T>
-constexpr T get_bits(size_t begin_bit, size_t end_bit, T value) {
+constexpr T get_bits(size_t begin_bit, size_t end_bit, T value)
+{
     const T m = mask<T>(begin_bit, end_bit);
     return (value & m) >> begin_bit;
 }
 
 /// Clears bits [begin_bit, end_bit] inclusive of value of type T.
 template<size_t begin_bit, size_t end_bit, BitIntegral T>
-constexpr T clear_bits(T value) {
+constexpr T clear_bits(T value)
+{
     constexpr T m = mask<begin_bit, end_bit, T>();
     return value & ~m;
 }
 
 /// Clears bits [begin_bit, end_bit] inclusive of value of type T.
 template<BitIntegral T>
-constexpr T clear_bits(size_t begin_bit, size_t end_bit, T value) {
+constexpr T clear_bits(size_t begin_bit, size_t end_bit, T value)
+{
     const T m = mask<T>(begin_bit, end_bit);
     return value & ~m;
 }
 
 /// Modifies bits [begin_bit, end_bit] inclusive of value of type T.
 template<size_t begin_bit, size_t end_bit, BitIntegral T>
-constexpr T set_bits(T value, T new_bits) {
+constexpr T set_bits(T value, T new_bits)
+{
     constexpr T m = mask<begin_bit, end_bit, T>();
     return (value & ~m) | ((new_bits << begin_bit) & m);
 }
 
 /// Modifies bits [begin_bit, end_bit] inclusive of value of type T.
 template<BitIntegral T>
-constexpr T set_bits(size_t begin_bit, size_t end_bit, T value, T new_bits) {
+constexpr T set_bits(size_t begin_bit, size_t end_bit, T value, T new_bits)
+{
     const T m = mask<T>(begin_bit, end_bit);
     return (value & ~m) | ((new_bits << begin_bit) & m);
 }
 
 /// Extract bit at bit_position from value of type T.
 template<size_t bit_position, BitIntegral T>
-constexpr bool get_bit(T value) {
+constexpr bool get_bit(T value)
+{
     constexpr T m = mask<bit_position, bit_position, T>();
     return (value & m) != 0;
 }
 
 /// Extract bit at bit_position from value of type T.
 template<BitIntegral T>
-constexpr bool get_bit(size_t bit_position, T value) {
+constexpr bool get_bit(size_t bit_position, T value)
+{
     const T m = mask<T>(bit_position, bit_position);
     return (value & m) != 0;
 }
 
 /// Clears bit at bit_position of value of type T.
 template<size_t bit_position, BitIntegral T>
-constexpr T clear_bit(T value) {
+constexpr T clear_bit(T value)
+{
     constexpr T m = mask<bit_position, bit_position, T>();
     return value & ~m;
 }
 
 /// Clears bit at bit_position of value of type T.
 template<BitIntegral T>
-constexpr T clear_bit(size_t bit_position, T value) {
+constexpr T clear_bit(size_t bit_position, T value)
+{
     const T m = mask<T>(bit_position, bit_position);
     return value & ~m;
 }
 
 /// Modifies bit at bit_position of value of type T.
 template<size_t bit_position, BitIntegral T>
-constexpr T set_bit(T value, bool new_bit) {
+constexpr T set_bit(T value, bool new_bit)
+{
     constexpr T m = mask<bit_position, bit_position, T>();
     return (value & ~m) | (new_bit ? m : static_cast<T>(0));
 }
 
 /// Modifies bit at bit_position of value of type T.
 template<BitIntegral T>
-constexpr T set_bit(size_t bit_position, T value, bool new_bit) {
+constexpr T set_bit(size_t bit_position, T value, bool new_bit)
+{
     const T m = mask<T>(bit_position, bit_position);
     return (value & ~m) | (new_bit ? m : static_cast<T>(0));
 }
 
 /// Sign-extends a value that has bit_count bits to the full bitwidth of type T.
 template<size_t bit_count, BitIntegral T>
-constexpr T sign_extend(T value) {
+constexpr T sign_extend(T value)
+{
     static_assert(bit_count != 0, "cannot sign-extend zero-sized value");
 
     using S = std::make_signed_t<T>;
@@ -150,7 +167,8 @@ constexpr T sign_extend(T value) {
 
 /// Sign-extends a value that has bit_count bits to the full bitwidth of type T.
 template<BitIntegral T>
-constexpr T sign_extend(size_t bit_count, T value) {
+constexpr T sign_extend(size_t bit_count, T value)
+{
     ASSERT_MSG(bit_count != 0, "cannot sign-extend zero-sized value");
 
     using S = std::make_signed_t<T>;
@@ -160,7 +178,8 @@ constexpr T sign_extend(size_t bit_count, T value) {
 
 /// Replicate an element across a value of type T.
 template<size_t element_size, BitIntegral T>
-constexpr T replicate_element(T value) {
+constexpr T replicate_element(T value)
+{
     static_assert(element_size <= bitsizeof<T>, "element_size is too large");
     static_assert(bitsizeof<T> % element_size == 0, "bitsize of T not divisible by element_size");
 
@@ -173,7 +192,8 @@ constexpr T replicate_element(T value) {
 
 /// Replicate an element of type U across a value of type T.
 template<BitIntegral U, BitIntegral T>
-constexpr T replicate_element(T value) {
+constexpr T replicate_element(T value)
+{
     static_assert(bitsizeof<U> <= bitsizeof<T>, "element_size is too large");
 
     return replicate_element<bitsizeof<U>, T>(value);
@@ -181,7 +201,8 @@ constexpr T replicate_element(T value) {
 
 /// Replicate an element across a value of type T.
 template<BitIntegral T>
-constexpr T replicate_element(size_t element_size, T value) {
+constexpr T replicate_element(size_t element_size, T value)
+{
     ASSERT_MSG(element_size <= bitsizeof<T>, "element_size is too large");
     ASSERT_MSG(bitsizeof<T> % element_size == 0, "bitsize of T not divisible by element_size");
 
@@ -192,7 +213,8 @@ constexpr T replicate_element(size_t element_size, T value) {
 }
 
 template<BitIntegral T>
-constexpr bool most_significant_bit(T value) {
+constexpr bool most_significant_bit(T value)
+{
     return get_bit<bitsizeof<T> - 1, T>(value);
 }
 

include/mcl/bit/rotate.hpp

@@ -11,7 +11,8 @@
 namespace mcl::bit {
 
 template<BitIntegral T>
-constexpr T rotate_right(T x, size_t amount) {
+constexpr T rotate_right(T x, size_t amount)
+{
     amount %= bitsizeof<T>;
     if (amount == 0) {
         return x;
@@ -20,7 +21,8 @@ constexpr T rotate_right(T x, size_t amount) {
 }
 
 template<BitIntegral T>
-constexpr T rotate_left(T x, size_t amount) {
+constexpr T rotate_left(T x, size_t amount)
+{
     amount %= bitsizeof<T>;
     if (amount == 0) {
         return x;

include/mcl/bit/swap.hpp

@@ -8,18 +8,21 @@
 
 namespace mcl::bit {
 
-constexpr u16 swap_bytes_16(u16 value) {
+constexpr u16 swap_bytes_16(u16 value)
+{
     return static_cast<u16>(u32{value} >> 8 | u32{value} << 8);
 }
 
-constexpr u32 swap_bytes_32(u32 value) {
+constexpr u32 swap_bytes_32(u32 value)
+{
     return ((value & 0xff000000u) >> 24)
          | ((value & 0x00ff0000u) >> 8)
          | ((value & 0x0000ff00u) << 8)
          | ((value & 0x000000ffu) << 24);
 }
 
-constexpr u64 swap_bytes_64(u64 value) {
+constexpr u64 swap_bytes_64(u64 value)
+{
     return ((value & 0xff00000000000000ull) >> 56)
          | ((value & 0x00ff000000000000ull) >> 40)
          | ((value & 0x0000ff0000000000ull) >> 24)
@@ -30,19 +33,22 @@ constexpr u64 swap_bytes_64(u64 value) {
          | ((value & 0x00000000000000ffull) << 56);
 }
 
-constexpr u32 swap_halves_32(u32 value) {
+constexpr u32 swap_halves_32(u32 value)
+{
     return ((value & 0xffff0000u) >> 16)
          | ((value & 0x0000ffffu) << 16);
 }
 
-constexpr u64 swap_halves_64(u64 value) {
+constexpr u64 swap_halves_64(u64 value)
+{
     return ((value & 0xffff000000000000ull) >> 48)
          | ((value & 0x0000ffff00000000ull) >> 16)
          | ((value & 0x00000000ffff0000ull) << 16)
          | ((value & 0x000000000000ffffull) << 48);
 }
 
-constexpr u64 swap_words_64(u64 value) {
+constexpr u64 swap_words_64(u64 value)
+{
     return ((value & 0xffffffff00000000ull) >> 32)
          | ((value & 0x00000000ffffffffull) << 32);
 }

include/mcl/bit_cast.hpp

@@ -11,7 +11,8 @@ namespace mcl {
 
 /// Reinterpret objects of one type as another by bit-casting between object representations.
 template<class Dest, class Source>
-inline Dest bit_cast(const Source& source) noexcept {
+inline Dest bit_cast(const Source& source) noexcept
+{
     static_assert(sizeof(Dest) == sizeof(Source), "size of destination and source objects must be equal");
     static_assert(std::is_trivially_copyable_v<Dest>, "destination type must be trivially copyable.");
     static_assert(std::is_trivially_copyable_v<Source>, "source type must be trivially copyable");
@@ -24,7 +25,8 @@ inline Dest bit_cast(const Source& source) noexcept {
 /// Reinterpret objects of any arbitrary type as another type by bit-casting between object representations.
 /// Note that here we do not verify if source pointed to by source_ptr has enough bytes to read from.
 template<class Dest, class SourcePtr>
-inline Dest bit_cast_pointee(const SourcePtr source_ptr) noexcept {
+inline Dest bit_cast_pointee(const SourcePtr source_ptr) noexcept
+{
     static_assert(sizeof(SourcePtr) == sizeof(void*), "source pointer must have size of a pointer");
     static_assert(std::is_trivially_copyable_v<Dest>, "destination type must be trivially copyable.");
 

include/mcl/container/detail/meta_byte.hpp

@@ -0,0 +1,35 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include "mcl/bitsizeof.hpp"
+#include "mcl/stdint.hpp"
+
+namespace mcl::detail {
+
+/// if MSB is 0, this is a full slot. remaining 7 bits is a partial hash of the key.
+/// if MSB is 1, this is a non-full slot.
+enum class meta_byte : u8 {
+    empty = 0xff,
+    tombstone = 0x80,
+    end_sentinel = 0x88,
+};
+
+inline bool is_full(meta_byte mb)
+{
+    return (static_cast<u8>(mb) & 0x80) == 0;
+}
+
+inline meta_byte meta_byte_from_hash(size_t hash)
+{
+    return static_cast<meta_byte>(hash >> (bitsizeof<size_t> - 7));
+}
+
+inline size_t group_index_from_hash(size_t hash, size_t group_index_mask)
+{
+    return hash & group_index_mask;
+}
+
+}  // namespace mcl::detail

include/mcl/container/detail/meta_byte_group.hpp

@@ -0,0 +1,263 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <array>
+#include <bit>
+
+#include "mcl/assert.hpp"
+#include "mcl/container/detail/meta_byte.hpp"
+#include "mcl/macro/architecture.hpp"
+#include "mcl/stdint.hpp"
+
+#if defined(MCL_ARCHITECTURE_ARM64)
+#    include <arm_neon.h>
+#elif defined(MCL_ARCHITECTURE_X86_64)
+#    include <emmintrin.h>
+
+#    include "mcl/bit_cast.hpp"
+#else
+#    include <cstring>
+#endif
+
+namespace mcl::detail {
+
+#if defined(MCL_ARCHITECTURE_ARM64)
+
+struct meta_byte_group {
+    static constexpr size_t max_group_size{16};
+
+    explicit meta_byte_group(meta_byte* ptr)
+        : data{vld1q_u8(reinterpret_cast<u8*>(ptr))}
+    {}
+
+    explicit meta_byte_group(const std::array<meta_byte, 16>& array)
+        : data{vld1q_u8(reinterpret_cast<const u8*>(array.data()))}
+    {}
+
+    uint64x2_t match(meta_byte cmp) const
+    {
+        return vreinterpretq_u64_u8(vandq_u8(vceqq_u8(data,
+                                                      vdupq_n_u8(static_cast<u8>(cmp))),
+                                             vdupq_n_u8(0x80)));
+    }
+
+    uint64x2_t match_empty_or_tombstone() const
+    {
+        return vreinterpretq_u64_u8(vandq_u8(data,
+                                             vdupq_n_u8(0x80)));
+    }
+
+    bool is_any_empty() const
+    {
+        static_assert(meta_byte::empty == static_cast<meta_byte>(0xff), "empty must be maximal u8 value");
+        return vmaxvq_u8(data) == 0xff;
+    }
+
+    bool is_all_empty_or_tombstone() const
+    {
+        return vminvq_u8(vandq_u8(data, vdupq_n_u8(0x80))) == 0x80;
+    }
+
+    meta_byte get(size_t index) const
+    {
+        return static_cast<meta_byte>(data[index]);
+    }
+
+    void set(size_t index, meta_byte value)
+    {
+        data[index] = static_cast<u8>(value);
+    }
+
+    uint8x16_t data;
+};
+
+#    define MCL_HMAP_MATCH_META_BYTE_GROUP(MATCH, ...)                                                             \
+        {                                                                                                          \
+            const uint64x2_t match_result{MATCH};                                                                  \
+                                                                                                                   \
+            for (u64 match_result_v{match_result[0]}; match_result_v != 0; match_result_v &= match_result_v - 1) { \
+                const size_t match_index{static_cast<size_t>(std::countr_zero(match_result_v) / 8)};               \
+                __VA_ARGS__                                                                                        \
+            }                                                                                                      \
+                                                                                                                   \
+            for (u64 match_result_v{match_result[1]}; match_result_v != 0; match_result_v &= match_result_v - 1) { \
+                const size_t match_index{static_cast<size_t>(8 + std::countr_zero(match_result_v) / 8)};           \
+                __VA_ARGS__                                                                                        \
+            }                                                                                                      \
+        }
+
+#    define MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(MATCH, ...)                                                                      \
+        {                                                                                                                               \
+            const uint64x2_t match_result{MATCH};                                                                                       \
+                                                                                                                                        \
+            for (u64 match_result_v{match_result[0]}; match_result_v != 0; match_result_v &= match_result_v - 1) {                      \
+                const size_t match_index{static_cast<size_t>(std::countr_zero(match_result_v) / 8)};                                    \
+                __VA_ARGS__                                                                                                             \
+            }                                                                                                                           \
+                                                                                                                                        \
+            for (u64 match_result_v{match_result[1] & 0x00ffffffffffffff}; match_result_v != 0; match_result_v &= match_result_v - 1) { \
+                const size_t match_index{static_cast<size_t>(8 + std::countr_zero(match_result_v) / 8)};                                \
+                __VA_ARGS__                                                                                                             \
+            }                                                                                                                           \
+        }
+
+#elif defined(MCL_ARCHITECTURE_X86_64)
+
+struct meta_byte_group {
+    static constexpr size_t max_group_size{16};
+
+    explicit meta_byte_group(meta_byte* ptr)
+        : data{_mm_load_si128(reinterpret_cast<__m128i const*>(ptr))}
+    {}
+
+    explicit meta_byte_group(const std::array<meta_byte, 16>& array)
+        : data{_mm_loadu_si128(reinterpret_cast<__m128i const*>(array.data()))}
+    {}
+
+    u16 match(meta_byte cmp) const
+    {
+        return _mm_movemask_epi8(_mm_cmpeq_epi8(data, _mm_set1_epi8(static_cast<u8>(cmp))));
+    }
+
+    u16 match_empty_or_tombstone() const
+    {
+        return _mm_movemask_epi8(data);
+    }
+
+    bool is_any_empty() const
+    {
+        return match(meta_byte::empty);
+    }
+
+    bool is_all_empty_or_tombstone() const
+    {
+        return match_empty_or_tombstone() == 0xffff;
+    }
+
+    meta_byte get(size_t index) const
+    {
+        return mcl::bit_cast<std::array<meta_byte, max_group_size>>(data)[index];
+    }
+
+    void set(size_t index, meta_byte value)
+    {
+        auto array = mcl::bit_cast<std::array<meta_byte, max_group_size>>(data);
+        array[index] = value;
+        data = mcl::bit_cast<__m128i>(array);
+    }
+
+    __m128i data;
+};
+
+#    define MCL_HMAP_MATCH_META_BYTE_GROUP(MATCH, ...)                                           \
+        {                                                                                        \
+            for (u16 match_result{MATCH}; match_result != 0; match_result &= match_result - 1) { \
+                const size_t match_index{static_cast<size_t>(std::countr_zero(match_result))};   \
+                __VA_ARGS__                                                                      \
+            }                                                                                    \
+        }
+
+#    define MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(MATCH, ...)                                                              \
+        {                                                                                                                       \
+            for (u16 match_result{static_cast<u16>((MATCH) & (0x7fff))}; match_result != 0; match_result &= match_result - 1) { \
+                const size_t match_index{static_cast<size_t>(std::countr_zero(match_result))};                                  \
+                __VA_ARGS__                                                                                                     \
+            }                                                                                                                   \
+        }
+
+#else
+
+struct meta_byte_group {
+    static constexpr size_t max_group_size{16};
+
+    static constexpr u64 msb{0x8080808080808080};
+    static constexpr u64 lsb{0x0101010101010101};
+    static constexpr u64 not_msb{0x7f7f7f7f7f7f7f7f};
+    static constexpr u64 not_lsb{0xfefefefefefefefe};
+
+    explicit meta_byte_group(meta_byte* ptr)
+    {
+        std::memcpy(data.data(), ptr, sizeof(data));
+    }
+
+    explicit meta_byte_group(const std::array<meta_byte, 16>& array)
+        : data{array}
+    {}
+
+    std::array<u64, 2> match(meta_byte cmp) const
+    {
+        DEBUG_ASSERT(is_full(cmp));
+
+        const u64 vcmp{lsb * static_cast<u64>(cmp)};
+        return {(msb - ((data[0] ^ vcmp) & not_msb)) & ~data[0] & msb, (msb - ((data[1] ^ vcmp) & not_msb)) & ~data[1] & msb};
+    }
+
+    std::array<u64, 2> match_empty_or_tombstone() const
+    {
+        return {data[0] & msb, data[1] & msb};
+    }
+
+    bool is_any_empty() const
+    {
+        static_assert((static_cast<u8>(meta_byte::empty) & 0xc0) == 0xc0);
+        static_assert((static_cast<u8>(meta_byte::tombstone) & 0xc0) == 0x80);
+
+        return (data[0] & (data[0] << 1) & msb) || (data[1] & (data[1] << 1) & msb);
+    }
+
+    bool is_all_empty_or_tombstone() const
+    {
+        return (data[0] & data[1] & msb) == msb;
+    }
+
+    meta_byte get(size_t index) const
+    {
+        return mcl::bit_cast<std::array<meta_byte, max_group_size>>(data)[index];
+    }
+
+    void set(size_t index, meta_byte value)
+    {
+        auto array = mcl::bit_cast<std::array<meta_byte, max_group_size>>(data);
+        array[index] = value;
+        data = mcl::bit_cast<std::array<u64, 2>>(array);
+    }
+
+    std::array<u64, 2> data;
+};
+
+#    define MCL_HMAP_MATCH_META_BYTE_GROUP(MATCH, ...)                                                             \
+        {                                                                                                          \
+            const std::array<u64, 2> match_result{MATCH};                                                          \
+                                                                                                                   \
+            for (u64 match_result_v{match_result[0]}; match_result_v != 0; match_result_v &= match_result_v - 1) { \
+                const size_t match_index{static_cast<size_t>(std::countr_zero(match_result_v) / 8)};               \
+                __VA_ARGS__                                                                                        \
+            }                                                                                                      \
+                                                                                                                   \
+            for (u64 match_result_v{match_result[1]}; match_result_v != 0; match_result_v &= match_result_v - 1) { \
+                const size_t match_index{static_cast<size_t>(8 + std::countr_zero(match_result_v) / 8)};           \
+                __VA_ARGS__                                                                                        \
+            }                                                                                                      \
+        }
+
+#    define MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(MATCH, ...)                                                                      \
+        {                                                                                                                               \
+            const std::array<u64, 2> match_result{MATCH};                                                                               \
+                                                                                                                                        \
+            for (u64 match_result_v{match_result[0]}; match_result_v != 0; match_result_v &= match_result_v - 1) {                      \
+                const size_t match_index{static_cast<size_t>(std::countr_zero(match_result_v) / 8)};                                    \
+                __VA_ARGS__                                                                                                             \
+            }                                                                                                                           \
+                                                                                                                                        \
+            for (u64 match_result_v{match_result[1] & 0x00ffffffffffffff}; match_result_v != 0; match_result_v &= match_result_v - 1) { \
+                const size_t match_index{static_cast<size_t>(8 + std::countr_zero(match_result_v) / 8)};                                \
+                __VA_ARGS__                                                                                                             \
+            }                                                                                                                           \
+        }
+
+#endif
+
+}  // namespace mcl::detail

include/mcl/container/detail/slot_union.hpp

@@ -0,0 +1,16 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+namespace mcl::detail {
+
+template<typename ValueType>
+union slot_union {
+    slot_union() {}
+    ~slot_union() {}
+    ValueType value;
+};
+
+}  // namespace mcl::detail

include/mcl/container/hmap.hpp

@@ -0,0 +1,532 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <cstddef>
+#include <functional>
+#include <limits>
+#include <type_traits>
+#include <utility>
+
+#include "mcl/assert.hpp"
+#include "mcl/container/detail/meta_byte.hpp"
+#include "mcl/container/detail/meta_byte_group.hpp"
+#include "mcl/container/detail/slot_union.hpp"
+#include "mcl/hash/xmrx.hpp"
+#include "mcl/hint/assume.hpp"
+#include "mcl/memory/overaligned_unique_ptr.hpp"
+
+namespace mcl {
+
+template<typename KeyType, typename MappedType, typename Hash, typename Pred>
+class hmap;
+
+template<bool IsConst, typename KeyType, typename MappedType, typename Hash, typename Pred>
+class hmap_iterator {
+    using base_value_type = std::pair<const KeyType, MappedType>;
+    using slot_type = detail::slot_union<base_value_type>;
+
+public:
+    using key_type = KeyType;
+    using mapped_type = MappedType;
+    using iterator_category = std::forward_iterator_tag;
+    using difference_type = std::ptrdiff_t;
+    using value_type = std::conditional_t<IsConst, std::add_const_t<base_value_type>, base_value_type>;
+    using pointer = value_type*;
+    using const_pointer = const value_type*;
+    using reference = value_type&;
+    using const_reference = const value_type&;
+
+    hmap_iterator() = default;
+    hmap_iterator(const hmap_iterator& other) = default;
+    hmap_iterator& operator=(const hmap_iterator& other) = default;
+
+    hmap_iterator& operator++()
+    {
+        if (mb_ptr == nullptr)
+            return *this;
+
+        ++mb_ptr;
+        ++slot_ptr;
+
+        skip_empty_or_tombstone();
+
+        return *this;
+    }
+    hmap_iterator operator++(int)
+    {
+        hmap_iterator it(*this);
+        ++*this;
+        return it;
+    }
+
+    bool operator==(const hmap_iterator& other) const
+    {
+        return std::tie(mb_ptr, slot_ptr) == std::tie(other.mb_ptr, other.slot_ptr);
+    }
+    bool operator!=(const hmap_iterator& other) const
+    {
+        return !operator==(other);
+    }
+
+    reference operator*() const
+    {
+        return static_cast<reference>(slot_ptr->value);
+    }
+    pointer operator->() const
+    {
+        return std::addressof(operator*());
+    }
+
+private:
+    friend class hmap<KeyType, MappedType, Hash, Pred>;
+
+    hmap_iterator(detail::meta_byte* mb_ptr, slot_type* slot_ptr)
+        : mb_ptr{mb_ptr}, slot_ptr{slot_ptr}
+    {
+        ASSUME(mb_ptr != nullptr);
+        ASSUME(slot_ptr != nullptr);
+    }
+
+    void skip_empty_or_tombstone()
+    {
+        if (!mb_ptr)
+            return;
+
+        while (*mb_ptr == detail::meta_byte::empty || *mb_ptr == detail::meta_byte::tombstone) {
+            ++mb_ptr;
+            ++slot_ptr;
+        }
+
+        if (*mb_ptr == detail::meta_byte::end_sentinel) {
+            mb_ptr = nullptr;
+            slot_ptr = nullptr;
+        }
+    }
+
+    detail::meta_byte* mb_ptr{nullptr};
+    slot_type* slot_ptr{nullptr};
+};
+
+template<typename KeyType, typename MappedType, typename Hash = hash::avalanche_xmrx<KeyType>, typename Pred = std::equal_to<KeyType>>
+class hmap {
+public:
+    using key_type = KeyType;
+    using mapped_type = MappedType;
+    using hasher = Hash;
+    using key_equal = Pred;
+    using value_type = std::pair<const key_type, mapped_type>;
+    using reference = value_type&;
+    using const_reference = const value_type&;
+    using pointer = value_type*;
+    using const_pointer = const value_type*;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
+
+    using iterator = hmap_iterator<false, key_type, mapped_type, hasher, key_equal>;
+    using const_iterator = hmap_iterator<true, key_type, mapped_type, hasher, key_equal>;
+
+private:
+    static constexpr size_t group_size{detail::meta_byte_group::max_group_size};
+    static constexpr size_t average_max_group_load{group_size - 2};
+
+    using slot_type = detail::slot_union<value_type>;
+    using slot_ptr = std::unique_ptr<slot_type[]>;
+    using meta_byte_ptr = overaligned_unique_ptr<group_size, detail::meta_byte[]>;
+    static_assert(!std::is_reference_v<key_type>);
+    static_assert(!std::is_reference_v<mapped_type>);
+
+public:
+    hmap()
+    {
+        initialize_members(1);
+    }
+    hmap(const hmap& other)
+    {
+        deep_copy(other);
+    }
+    hmap(hmap&& other)
+        : group_index_mask{std::exchange(other.group_index_mask, 0)}
+        , empty_slots{std::exchange(other.empty_slots, 0)}
+        , full_slots{std::exchange(other.full_slots, 0)}
+        , mbs{std::move(other.mbs)}
+        , slots{std::move(other.slots)}
+    {
+    }
+    hmap& operator=(const hmap& other)
+    {
+        deep_copy(other);
+        return *this;
+    }
+    hmap& operator=(hmap&& other)
+    {
+        group_index_mask = std::exchange(other.group_index_mask, 0);
+        empty_slots = std::exchange(other.empty_slots, 0);
+        full_slots = std::exchange(other.full_slots, 0);
+        mbs = std::move(other.mbs);
+        slots = std::move(other.slots);
+        return *this;
+    }
+
+    ~hmap()
+    {
+        if (!mbs)
+            return;
+
+        clear();
+    }
+
+    [[nodiscard]] bool empty() const noexcept { return full_slots == 0; }
+    size_type size() const noexcept { return full_slots; }
+    size_type max_size() const noexcept { return static_cast<size_type>(std::numeric_limits<difference_type>::max()); }
+
+    iterator begin()
+    {
+        iterator result{iterator_at(0)};
+        result.skip_empty_or_tombstone();
+        return result;
+    }
+    iterator end()
+    {
+        return {};
+    }
+    const_iterator cbegin() const
+    {
+        const_iterator result{const_iterator_at(0)};
+        result.skip_empty_or_tombstone();
+        return result;
+    }
+    const_iterator cend() const
+    {
+        return {};
+    }
+    const_iterator begin() const
+    {
+        return cbegin();
+    }
+    const_iterator end() const
+    {
+        return cend();
+    }
+
+    template<typename K = key_type, typename... Args>
+    std::pair<iterator, bool> try_emplace(K&& k, Args&&... args)
+    {
+        auto [item_index, item_found] = find_key_or_empty_slot(k);
+        if (!item_found) {
+            new (&slots[item_index].value) value_type(
+                std::piecewise_construct,
+                std::forward_as_tuple(std::forward<K>(k)),
+                std::forward_as_tuple(std::forward<Args>(args)...));
+        }
+        return {iterator_at(item_index), !item_found};
+    }
+
+    template<typename K = key_type, typename V = mapped_type>
+    std::pair<iterator, bool> insert_or_assign(K&& k, V&& v)
+    {
+        auto [item_index, item_found] = find_key_or_empty_slot(k);
+        if (item_found) {
+            slots[item_index].value.second = std::forward<V>(v);
+        } else {
+            new (&slots[item_index].value) value_type(
+                std::forward<K>(k),
+                std::forward<V>(v));
+        }
+        return {iterator_at(item_index), !item_found};
+    }
+
+    void erase(const_iterator position)
+    {
+        if (position == cend()) {
+            return;
+        }
+
+        const std::size_t item_index{static_cast<std::size_t>(std::distance(mbs.get(), position.mb_ptr))};
+        const std::size_t group_index{item_index / group_size};
+        const detail::meta_byte_group g{mbs.get() + group_index * group_size};
+
+        erase_impl(item_index, std::move(g));
+    }
+    void erase(iterator position)
+    {
+        if (position == end()) {
+            return;
+        }
+
+        const std::size_t item_index{static_cast<std::size_t>(std::distance(mbs.get(), position.mb_ptr))};
+        const std::size_t group_index{item_index / group_size};
+        const detail::meta_byte_group g{mbs.get() + group_index * group_size};
+
+        erase_impl(item_index, std::move(g));
+    }
+    template<typename K = key_type>
+    size_t erase(const K& key)
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            detail::meta_byte_group g{mbs.get() + group_index * group_size};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP(g.match(mb), {
+                const std::size_t item_index{group_index * group_size + match_index};
+
+                if (key_equal{}(slots[item_index].value.first, key)) [[likely]] {
+                    erase_impl(item_index, std::move(g));
+                    return 1;
+                }
+            });
+
+            if (g.is_any_empty()) [[likely]] {
+                return 0;
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+
+    template<typename K = key_type>
+    iterator find(const K& key)
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            detail::meta_byte_group g{mbs.get() + group_index * group_size};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP(g.match(mb), {
+                const std::size_t item_index{group_index * group_size + match_index};
+
+                if (key_equal{}(slots[item_index].value.first, key)) [[likely]] {
+                    return iterator_at(item_index);
+                }
+            });
+
+            if (g.is_any_empty()) [[likely]] {
+                return {};
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+    template<typename K = key_type>
+    const_iterator find(const K& key) const
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            detail::meta_byte_group g{mbs.get() + group_index * group_size};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP(g.match(mb), {
+                const std::size_t item_index{group_index * group_size + match_index};
+
+                if (key_equal{}(slots[item_index].value.first, key)) [[likely]] {
+                    return const_iterator_at(item_index);
+                }
+            });
+
+            if (g.is_any_empty()) [[likely]] {
+                return {};
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+    template<typename K = key_type>
+    bool contains(const K& key) const
+    {
+        return find(key) != end();
+    }
+    template<typename K = key_type>
+    size_t count(const K& key) const
+    {
+        return contains(key) ? 1 : 0;
+    }
+
+    template<typename K = key_type>
+    mapped_type& operator[](K&& k)
+    {
+        return try_emplace(std::forward<K>(k)).first->second;
+    }
+    template<typename K = key_type>
+    mapped_type& at(K&& k)
+    {
+        const auto iter{find(k)};
+        if (iter == end()) {
+            throw std::out_of_range("hmap::at: key not found");
+        }
+        return iter->second;
+    }
+    template<typename K = key_type>
+    const mapped_type& at(K&& k) const
+    {
+        const auto iter{find(k)};
+        if (iter == end()) {
+            throw std::out_of_range("hmap::at: key not found");
+        }
+        return iter->second;
+    }
+
+    void clear()
+    {
+        for (auto iter{begin()}; iter != end(); ++iter) {
+            iter->~value_type();
+        }
+
+        clear_metadata();
+    }
+
+private:
+    iterator iterator_at(std::size_t item_index)
+    {
+        return {mbs.get() + item_index, slots.get() + item_index};
+    }
+    const_iterator const_iterator_at(std::size_t item_index) const
+    {
+        return {mbs.get() + item_index, slots.get() + item_index};
+    }
+
+    std::pair<std::size_t, bool> find_key_or_empty_slot(const key_type& key)
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        std::size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            detail::meta_byte_group g{mbs.get() + group_index * group_size};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP(g.match(mb), {
+                const std::size_t item_index{group_index * group_size + match_index};
+
+                if (key_equal{}(slots[item_index].value.first, key)) [[likely]] {
+                    return {item_index, true};
+                }
+            });
+
+            if (g.is_any_empty()) [[likely]] {
+                return {find_empty_slot_to_insert(hash), false};
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+
+    std::size_t find_empty_slot_to_insert(const std::size_t hash)
+    {
+        if (empty_slots == 0) [[unlikely]] {
+            grow_and_rehash();
+        }
+
+        std::size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            detail::meta_byte_group g{mbs.get() + group_index * group_size};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP(g.match_empty_or_tombstone(), {
+                const std::size_t item_index{group_index * group_size + match_index};
+
+                if (mbs[item_index] == detail::meta_byte::empty) [[likely]] {
+                    --empty_slots;
+                }
+                ++full_slots;
+
+                mbs[item_index] = detail::meta_byte_from_hash(hash);
+
+                return item_index;
+            });
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+
+    void erase_impl(std::size_t item_index, detail::meta_byte_group&& g)
+    {
+        slots[item_index].value->~value_type();
+
+        --full_slots;
+        if (g.is_any_empty()) {
+            mbs[item_index] = detail::meta_byte::empty;
+            ++empty_slots;
+        } else {
+            mbs[item_index] = detail::meta_byte::tombstone;
+        }
+    }
+
+    void grow_and_rehash()
+    {
+        const std::size_t new_group_count{2 * (group_index_mask + 1)};
+
+        pow2_resize(new_group_count);
+    }
+
+    void pow2_resize(std::size_t new_group_count)
+    {
+        auto iter{begin()};
+
+        const auto old_mbs{std::move(mbs)};
+        const auto old_slots{std::move(slots)};
+
+        initialize_members(new_group_count);
+
+        for (; iter != end(); ++iter) {
+            const std::size_t hash{hasher{}(iter->first)};
+            const std::size_t item_index{find_empty_slot_to_insert(hash)};
+
+            new (&slots[item_index].value) value_type(std::move(iter.slot_ptr->value));
+            iter.slot_ptr->value.~value_type();
+        }
+    }
+
+    void deep_copy(const hmap& other)
+    {
+        initialize_members(other.group_index_mask + 1);
+
+        for (auto iter = other.begin(); iter != other.end(); ++iter) {
+            const std::size_t hash{hasher{}(iter->first)};
+            const std::size_t item_index{find_empty_slot_to_insert(hash)};
+
+            new (&slots[item_index].value) value_type(iter.slot_ptr->value);
+        }
+    }
+
+    void initialize_members(std::size_t group_count)
+    {
+        // DEBUG_ASSERT(group_count != 0 && std::ispow2(group_count));
+
+        group_index_mask = group_count - 1;
+        mbs = make_overaligned_unique_ptr_array<group_size, detail::meta_byte>(group_count * group_size + 1);
+        slots = slot_ptr{new slot_type[group_count * group_size]};
+
+        clear_metadata();
+    }
+
+    void clear_metadata()
+    {
+        const std::size_t group_count{group_index_mask + 1};
+
+        empty_slots = group_count * average_max_group_load;
+        full_slots = 0;
+
+        std::memset(mbs.get(), static_cast<int>(detail::meta_byte::empty), group_count * group_size);
+        mbs[group_count * group_size] = detail::meta_byte::end_sentinel;
+    }
+
+    std::size_t group_index_mask;
+    std::size_t empty_slots;
+    std::size_t full_slots;
+    meta_byte_ptr mbs;
+    slot_ptr slots;
+};
+
+}  // namespace mcl

include/mcl/container/ihmap.hpp

@@ -0,0 +1,549 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <array>
+#include <cstddef>
+#include <functional>
+#include <limits>
+#include <type_traits>
+#include <utility>
+
+#include "mcl/assert.hpp"
+#include "mcl/container/detail/meta_byte.hpp"
+#include "mcl/container/detail/meta_byte_group.hpp"
+#include "mcl/container/detail/slot_union.hpp"
+#include "mcl/hash/xmrx.hpp"
+#include "mcl/hint/assume.hpp"
+
+namespace mcl {
+
+template<typename KeyType, typename MappedType, typename Hash, typename Pred>
+class ihmap;
+
+namespace detail {
+
+constexpr std::array<meta_byte, 16> ihmap_default_meta{
+    meta_byte::empty, meta_byte::empty, meta_byte::empty, meta_byte::empty,
+    meta_byte::empty, meta_byte::empty, meta_byte::empty, meta_byte::empty,
+    meta_byte::empty, meta_byte::empty, meta_byte::empty, meta_byte::empty,
+    meta_byte::empty, meta_byte::empty, meta_byte::empty, meta_byte::tombstone};
+
+template<typename KeyType, typename MappedType>
+struct ihmap_group {
+    using base_value_type = std::pair<const KeyType, MappedType>;
+    using slot_type = detail::slot_union<base_value_type>;
+
+    static constexpr std::size_t group_size{meta_byte_group::max_group_size - 1};
+
+    meta_byte_group meta{ihmap_default_meta};
+    std::array<slot_type, group_size> slots{};
+};
+
+}  // namespace detail
+
+template<bool IsConst, typename KeyType, typename MappedType, typename Hash, typename Pred>
+class ihmap_iterator {
+    using group_type = detail::ihmap_group<KeyType, MappedType>;
+    using base_value_type = typename group_type::base_value_type;
+
+public:
+    using key_type = KeyType;
+    using mapped_type = MappedType;
+    using iterator_category = std::forward_iterator_tag;
+    using difference_type = std::ptrdiff_t;
+    using value_type = std::conditional_t<IsConst, std::add_const_t<base_value_type>, base_value_type>;
+    using pointer = value_type*;
+    using const_pointer = const value_type*;
+    using reference = value_type&;
+    using const_reference = const value_type&;
+
+    ihmap_iterator() = default;
+    ihmap_iterator(const ihmap_iterator& other) = default;
+    ihmap_iterator& operator=(const ihmap_iterator& other) = default;
+
+    ihmap_iterator& operator++()
+    {
+        if (group_ptr == nullptr)
+            return *this;
+
+        ++slot_index;
+
+        skip_empty_or_tombstone();
+
+        return *this;
+    }
+    ihmap_iterator operator++(int)
+    {
+        ihmap_iterator it(*this);
+        ++*this;
+        return it;
+    }
+
+    bool operator==(const ihmap_iterator& other) const
+    {
+        return std::tie(group_ptr, slot_index) == std::tie(other.group_ptr, other.slot_index);
+    }
+    bool operator!=(const ihmap_iterator& other) const
+    {
+        return !operator==(other);
+    }
+
+    reference operator*() const
+    {
+        return static_cast<reference>(group_ptr->slots[slot_index].value);
+    }
+    pointer operator->() const
+    {
+        return std::addressof(operator*());
+    }
+
+private:
+    friend class ihmap<KeyType, MappedType, Hash, Pred>;
+
+    ihmap_iterator(group_type* group_ptr, size_t slot_index)
+        : group_ptr{group_ptr}, slot_index{slot_index}
+    {
+        ASSUME(group_ptr != nullptr);
+    }
+
+    void skip_empty_or_tombstone()
+    {
+        if (!group_ptr)
+            return;
+
+        while (true) {
+            const detail::meta_byte mb = group_ptr->meta.get(slot_index);
+            if (slot_index == group_type::group_size) {
+                slot_index = 0;
+                ++group_ptr;
+
+                if (mb == detail::meta_byte::end_sentinel) {
+                    group_ptr = nullptr;
+                    return;
+                }
+
+                continue;
+            }
+            if (is_full(mb)) {
+                break;
+            }
+            ++slot_index;
+        }
+    }
+
+    group_type* group_ptr{nullptr};
+    std::size_t slot_index{0};
+};
+
+template<typename KeyType, typename MappedType, typename Hash = hash::avalanche_xmrx<KeyType>, typename Pred = std::equal_to<KeyType>>
+class ihmap {
+    using group_type = detail::ihmap_group<KeyType, MappedType>;
+
+public:
+    using key_type = KeyType;
+    using mapped_type = MappedType;
+    using hasher = Hash;
+    using key_equal = Pred;
+    using value_type = typename group_type::base_value_type;
+    using reference = value_type&;
+    using const_reference = const value_type&;
+    using pointer = value_type*;
+    using const_pointer = const value_type*;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
+
+    using iterator = ihmap_iterator<false, key_type, mapped_type, hasher, key_equal>;
+    using const_iterator = ihmap_iterator<true, key_type, mapped_type, hasher, key_equal>;
+
+private:
+    static_assert(!std::is_reference_v<key_type>);
+    static_assert(!std::is_reference_v<mapped_type>);
+
+    static constexpr std::size_t group_size{group_type::group_size};
+    static constexpr std::size_t average_max_group_load{group_size - 2};
+
+    struct position {
+        std::size_t group_index;
+        std::size_t slot_index;
+    };
+
+public:
+    ihmap()
+    {
+        initialize_members(1);
+    }
+    ihmap(const ihmap& other)
+    {
+        deep_copy(other);
+    }
+    ihmap(ihmap&& other)
+        : group_index_mask{std::exchange(other.group_index_mask, 0)}
+        , empty_slots{std::exchange(other.empty_slots, 0)}
+        , full_slots{std::exchange(other.full_slots, 0)}
+        , groups{std::move(other.groups)}
+    {
+    }
+    ihmap& operator=(const ihmap& other)
+    {
+        deep_copy(other);
+        return *this;
+    }
+    ihmap& operator=(ihmap&& other)
+    {
+        group_index_mask = std::exchange(other.group_index_mask, 0);
+        empty_slots = std::exchange(other.empty_slots, 0);
+        full_slots = std::exchange(other.full_slots, 0);
+        groups = std::move(other.groups);
+        return *this;
+    }
+
+    ~ihmap()
+    {
+        if (!groups)
+            return;
+
+        clear();
+    }
+
+    [[nodiscard]] bool empty() const noexcept { return full_slots == 0; }
+    size_type size() const noexcept { return full_slots; }
+    size_type max_size() const noexcept { return static_cast<size_type>(std::numeric_limits<difference_type>::max()); }
+
+    iterator begin()
+    {
+        iterator result{iterator_at({0, 0})};
+        result.skip_empty_or_tombstone();
+        return result;
+    }
+    iterator end()
+    {
+        return {};
+    }
+    const_iterator cbegin() const
+    {
+        const_iterator result{const_iterator_at({0, 0})};
+        result.skip_empty_or_tombstone();
+        return result;
+    }
+    const_iterator cend() const
+    {
+        return {};
+    }
+    const_iterator begin() const
+    {
+        return cbegin();
+    }
+    const_iterator end() const
+    {
+        return cend();
+    }
+
+    template<typename K = key_type, typename... Args>
+    std::pair<iterator, bool> try_emplace(K&& k, Args&&... args)
+    {
+        auto [pos, item_found] = find_key_or_empty_slot(k);
+        if (!item_found) {
+            new (&groups[pos.group_index].slots[pos.slot_index].value) value_type(
+                std::piecewise_construct,
+                std::forward_as_tuple(std::forward<K>(k)),
+                std::forward_as_tuple(std::forward<Args>(args)...));
+        }
+        return {iterator_at(pos), !item_found};
+    }
+
+    template<typename K = key_type, typename V = mapped_type>
+    std::pair<iterator, bool> insert_or_assign(K&& k, V&& v)
+    {
+        auto [pos, item_found] = find_key_or_empty_slot(k);
+        if (item_found) {
+            groups[pos.group_index].slots[pos.slot_index].value.second = std::forward<V>(v);
+        } else {
+            new (&groups[pos.group_index].slots[pos.slot_index].value) value_type(
+                std::forward<K>(k),
+                std::forward<V>(v));
+        }
+        return {iterator_at(pos), !item_found};
+    }
+
+    void erase(const_iterator iter)
+    {
+        if (iter == cend()) {
+            return;
+        }
+
+        const std::size_t group_index{static_cast<std::size_t>(std::distance(groups.get(), iter.group_ptr))};
+
+        erase_impl({group_index, iter.slot_index});
+    }
+    void erase(iterator iter)
+    {
+        if (iter == end()) {
+            return;
+        }
+
+        const std::size_t group_index{static_cast<std::size_t>(std::distance(groups.get(), iter.group_ptr))};
+
+        erase_impl({group_index, iter.slot_index});
+    }
+    template<typename K = key_type>
+    std::size_t erase(const K& key)
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        std::size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            const group_type& g{groups[group_index]};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(g.meta.match(mb), {
+                if (key_equal{}(g.slots[match_index].value.first, key)) [[likely]] {
+                    erase_impl({group_index, match_index});
+                    return 1;
+                }
+            });
+
+            if (g.meta.is_any_empty()) [[likely]] {
+                return 0;
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+
+    template<typename K = key_type>
+    iterator find(const K& key)
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        std::size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            const group_type& g{groups[group_index]};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(g.meta.match(mb), {
+                if (key_equal{}(g.slots[match_index].value.first, key)) [[likely]] {
+                    return iterator_at({group_index, match_index});
+                }
+            });
+
+            if (g.meta.is_any_empty()) [[likely]] {
+                return {};
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+    template<typename K = key_type>
+    const_iterator find(const K& key) const
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        std::size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            const group_type& g{groups[group_index]};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(g.meta.match(mb), {
+                if (key_equal{}(g.slots[match_index].value.first, key)) [[likely]] {
+                    return const_iterator_at({group_index, match_index});
+                }
+            });
+
+            if (g.meta.is_any_empty()) [[likely]] {
+                return {};
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+    template<typename K = key_type>
+    bool contains(const K& key) const
+    {
+        return find(key) != end();
+    }
+    template<typename K = key_type>
+    std::size_t count(const K& key) const
+    {
+        return contains(key) ? 1 : 0;
+    }
+
+    template<typename K = key_type>
+    mapped_type& operator[](K&& k)
+    {
+        return try_emplace(std::forward<K>(k)).first->second;
+    }
+    template<typename K = key_type>
+    mapped_type& at(K&& k)
+    {
+        const auto iter{find(k)};
+        if (iter == end()) {
+            throw std::out_of_range("ihmap::at: key not found");
+        }
+        return iter->second;
+    }
+    template<typename K = key_type>
+    const mapped_type& at(K&& k) const
+    {
+        const auto iter{find(k)};
+        if (iter == end()) {
+            throw std::out_of_range("ihmap::at: key not found");
+        }
+        return iter->second;
+    }
+
+    void clear()
+    {
+        for (auto iter{begin()}; iter != end(); ++iter) {
+            iter->~value_type();
+        }
+
+        clear_metadata();
+    }
+
+private:
+    iterator iterator_at(position pos)
+    {
+        return {groups.get() + pos.group_index, pos.slot_index};
+    }
+    const_iterator const_iterator_at(position pos) const
+    {
+        return {groups.get() + pos.group_index, pos.slot_index};
+    }
+
+    std::pair<position, bool> find_key_or_empty_slot(const key_type& key)
+    {
+        const std::size_t hash{hasher{}(key)};
+        const detail::meta_byte mb{detail::meta_byte_from_hash(hash)};
+
+        std::size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            const group_type& g{groups[group_index]};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(g.meta.match(mb), {
+                if (key_equal{}(g.slots[match_index].value.first, key)) [[likely]] {
+                    return {{group_index, match_index}, true};
+                }
+            });
+
+            if (g.meta.is_any_empty()) [[likely]] {
+                return {find_empty_slot_to_insert(hash), false};
+            }
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+
+    position find_empty_slot_to_insert(const std::size_t hash)
+    {
+        if (empty_slots == 0) [[unlikely]] {
+            grow_and_rehash();
+        }
+
+        std::size_t group_index{detail::group_index_from_hash(hash, group_index_mask)};
+
+        while (true) {
+            group_type& g{groups[group_index]};
+
+            MCL_HMAP_MATCH_META_BYTE_GROUP_EXCEPT_LAST(g.meta.match_empty_or_tombstone(), {
+                if (g.meta.get(match_index) == detail::meta_byte::empty) [[likely]] {
+                    --empty_slots;
+                }
+                ++full_slots;
+
+                g.meta.set(match_index, detail::meta_byte_from_hash(hash));
+
+                return {group_index, match_index};
+            });
+
+            group_index = (group_index + 1) & group_index_mask;
+        }
+    }
+
+    void erase_impl(position pos)
+    {
+        group_type& g{groups[pos.group_index]};
+
+        g.slots[pos.slot_index].value.~value_type();
+
+        --full_slots;
+        if (g.meta.is_any_empty()) {
+            g.meta.set(pos.slot_index, detail::meta_byte::empty);
+            ++empty_slots;
+        } else {
+            g.meta.set(pos.slot_index, detail::meta_byte::tombstone);
+        }
+    }
+
+    void grow_and_rehash()
+    {
+        const std::size_t new_group_count{2 * (group_index_mask + 1)};
+
+        pow2_resize(new_group_count);
+    }
+
+    void pow2_resize(std::size_t new_group_count)
+    {
+        auto iter{begin()};
+
+        const auto old_groups{std::move(groups)};
+
+        initialize_members(new_group_count);
+
+        for (; iter != end(); ++iter) {
+            const std::size_t hash{hasher{}(iter->first)};
+            const position pos{find_empty_slot_to_insert(hash)};
+
+            new (&groups[pos.group_index].slots[pos.slot_index].value) value_type(std::move(iter.group_ptr->slots[iter.slot_index].value));
+            iter.group_ptr->slots[iter.slot_index].value.~value_type();
+        }
+    }
+
+    void deep_copy(const ihmap& other)
+    {
+        initialize_members(other.group_index_mask + 1);
+
+        for (auto iter = other.begin(); iter != other.end(); ++iter) {
+            const std::size_t hash{hasher{}(iter->first)};
+            const position pos{find_empty_slot_to_insert(hash)};
+
+            new (&groups[pos.group_index].slots[pos.slot_index].value) value_type(iter.group_ptr->slots[iter.slot_index].value);
+        }
+    }
+
+    void initialize_members(std::size_t group_count)
+    {
+        // DEBUG_ASSERT(group_count != 0 && std::ispow2(group_count));
+
+        group_index_mask = group_count - 1;
+        groups = std::unique_ptr<group_type[]>{new group_type[group_count]};
+
+        clear_metadata();
+    }
+
+    void clear_metadata()
+    {
+        const std::size_t group_count{group_index_mask + 1};
+
+        empty_slots = group_count * average_max_group_load;
+        full_slots = 0;
+
+        for (size_t i{0}; i < group_count; ++i) {
+            groups[i].meta = detail::meta_byte_group{detail::ihmap_default_meta};
+        }
+        groups[group_count - 1].meta.set(group_size, detail::meta_byte::end_sentinel);
+    }
+
+    std::size_t group_index_mask;
+    std::size_t empty_slots;
+    std::size_t full_slots;
+    std::unique_ptr<group_type[]> groups;
+};
+
+}  // namespace mcl

include/mcl/container/intrusive_list.hpp

@@ -21,7 +21,8 @@ class intrusive_list_iterator;
 template<typename T>
 class intrusive_list_node {
 public:
-    bool is_sentinel() const {
+    bool is_sentinel() const
+    {
         return is_sentinel_;
     }
 
@@ -42,7 +43,8 @@ class intrusive_list_sentinel final : public intrusive_list_node<T> {
     using intrusive_list_node<T>::is_sentinel_;
 
 public:
-    intrusive_list_sentinel() {
+    intrusive_list_sentinel()
+    {
         next = this;
         prev = this;
         is_sentinel_ = true;
@@ -72,50 +74,56 @@ public:
     intrusive_list_iterator& operator=(const intrusive_list_iterator& other) = default;
 
     explicit intrusive_list_iterator(node_pointer list_node)
-            : node(list_node) {
+        : node(list_node) {}
-    }
     explicit intrusive_list_iterator(pointer data)
-            : node(data) {
+        : node(data) {}
-    }
     explicit intrusive_list_iterator(reference data)
-            : node(&data) {
+        : node(&data) {}
-    }
 
-    intrusive_list_iterator& operator++() {
+    intrusive_list_iterator& operator++()
+    {
         node = node->next;
         return *this;
     }
-    intrusive_list_iterator& operator--() {
+    intrusive_list_iterator& operator--()
+    {
         node = node->prev;
         return *this;
     }
-    intrusive_list_iterator operator++(int) {
+    intrusive_list_iterator operator++(int)
+    {
         intrusive_list_iterator it(*this);
         ++*this;
         return it;
     }
-    intrusive_list_iterator operator--(int) {
+    intrusive_list_iterator operator--(int)
+    {
         intrusive_list_iterator it(*this);
         --*this;
         return it;
     }
 
-    bool operator==(const intrusive_list_iterator& other) const {
+    bool operator==(const intrusive_list_iterator& other) const
+    {
         return node == other.node;
     }
-    bool operator!=(const intrusive_list_iterator& other) const {
+    bool operator!=(const intrusive_list_iterator& other) const
+    {
         return !operator==(other);
     }
 
-    reference operator*() const {
+    reference operator*() const
+    {
         DEBUG_ASSERT(!node->is_sentinel());
         return static_cast<reference>(*node);
     }
-    pointer operator->() const {
+    pointer operator->() const
+    {
         return std::addressof(operator*());
     }
 
-    node_pointer AsNodePointer() const {
+    node_pointer AsNodePointer() const
+    {
         return node;
     }
 
@@ -145,7 +153,8 @@ public:
      * @param location The location to insert the node.
      * @param new_node The node to add.
      */
-    iterator insert(iterator location, pointer new_node) {
+    iterator insert(iterator location, pointer new_node)
+    {
         return insert_before(location, new_node);
     }
 
@@ -156,7 +165,8 @@ public:
      * @param location The location to insert the new node.
      * @param new_node The node to insert into the list.
      */
-    iterator insert_before(iterator location, pointer new_node) {
+    iterator insert_before(iterator location, pointer new_node)
+    {
         auto existing_node = location.AsNodePointer();
 
         new_node->next = existing_node;
@@ -173,7 +183,8 @@ public:
      * @param position Location to insert the node in front of.
      * @param new_node The node to be inserted into the list.
      */
-    iterator insert_after(iterator position, pointer new_node) {
+    iterator insert_after(iterator position, pointer new_node)
+    {
         if (empty())
             return insert(begin(), new_node);
 
@@ -184,7 +195,8 @@ public:
      * Add an entry to the start of the list.
      * @param node Node to add to the list.
      */
-    void push_front(pointer node) {
+    void push_front(pointer node)
+    {
         insert(begin(), node);
     }
 
@@ -192,7 +204,8 @@ public:
      * Add an entry to the end of the list
      * @param node Node to add to the list.
      */
-    void push_back(pointer node) {
+    void push_back(pointer node)
+    {
         insert(end(), node);
     }
 
@@ -200,7 +213,8 @@ public:
      * Erases the node at the front of the list.
      * @note Must not be called on an empty list.
      */
-    void pop_front() {
+    void pop_front()
+    {
         DEBUG_ASSERT(!empty());
         erase(begin());
     }
@@ -209,7 +223,8 @@ public:
      * Erases the node at the back of the list.
      * @note Must not be called on an empty list.
      */
-    void pop_back() {
+    void pop_back()
+    {
         DEBUG_ASSERT(!empty());
         erase(--end());
     }
@@ -218,7 +233,8 @@ public:
      * Removes a node from this list
      * @param it An iterator that points to the node to remove from list.
      */
-    pointer remove(iterator& it) {
+    pointer remove(iterator& it)
+    {
         DEBUG_ASSERT(it != end());
 
         pointer node = &*it++;
@@ -237,7 +253,8 @@ public:
      * Removes a node from this list
      * @param it A constant iterator that points to the node to remove from list.
      */
-    pointer remove(const iterator& it) {
+    pointer remove(const iterator& it)
+    {
         iterator copy = it;
         return remove(copy);
     }
@@ -246,7 +263,8 @@ public:
      * Removes a node from this list.
      * @param node A pointer to the node to remove.
      */
-    pointer remove(pointer node) {
+    pointer remove(pointer node)
+    {
         return remove(iterator(node));
     }
 
@@ -254,7 +272,8 @@ public:
      * Removes a node from this list.
      * @param node A reference to the node to remove.
      */
-    pointer remove(reference node) {
+    pointer remove(reference node)
+    {
         return remove(iterator(node));
     }
 
@@ -262,7 +281,8 @@ public:
      * Is this list empty?
      * @returns true if there are no nodes in this list.
      */
-    bool empty() const {
+    bool empty() const
+    {
         return root->next == root.get();
     }
 
@@ -270,7 +290,8 @@ public:
      * Gets the total number of elements within this list.
      * @return the number of elements in this list.
      */
-    size_type size() const {
+    size_type size() const
+    {
         return static_cast<size_type>(std::distance(begin(), end()));
     }
 
@@ -278,7 +299,8 @@ public:
      * Retrieves a reference to the node at the front of the list.
      * @note Must not be called on an empty list.
      */
-    reference front() {
+    reference front()
+    {
         DEBUG_ASSERT(!empty());
         return *begin();
     }
@@ -287,7 +309,8 @@ public:
      * Retrieves a constant reference to the node at the front of the list.
      * @note Must not be called on an empty list.
      */
-    const_reference front() const {
+    const_reference front() const
+    {
         DEBUG_ASSERT(!empty());
         return *begin();
     }
@@ -296,7 +319,8 @@ public:
      * Retrieves a reference to the node at the back of the list.
      * @note Must not be called on an empty list.
      */
-    reference back() {
+    reference back()
+    {
         DEBUG_ASSERT(!empty());
         return *--end();
     }
@@ -305,7 +329,8 @@ public:
      * Retrieves a constant reference to the node at the back of the list.
      * @note Must not be called on an empty list.
      */
-    const_reference back() const {
+    const_reference back() const
+    {
         DEBUG_ASSERT(!empty());
         return *--end();
     }
@@ -331,7 +356,8 @@ public:
      * Erases a node from the list, indicated by an iterator.
      * @param it The iterator that points to the node to erase.
      */
-    iterator erase(iterator it) {
+    iterator erase(iterator it)
+    {
         remove(it);
         return it;
     }
@@ -340,7 +366,8 @@ public:
      * Erases a node from this list.
      * @param node A pointer to the node to erase from this list.
      */
-    iterator erase(pointer node) {
+    iterator erase(pointer node)
+    {
         return erase(iterator(node));
     }
 
@@ -348,7 +375,8 @@ public:
      * Erases a node from this list.
      * @param node A reference to the node to erase from this list.
      */
-    iterator erase(reference node) {
+    iterator erase(reference node)
+    {
         return erase(iterator(node));
     }
 
@@ -356,7 +384,8 @@ public:
      * Exchanges contents of this list with another list instance.
      * @param other The other list to swap with.
      */
-    void swap(intrusive_list& other) noexcept {
+    void swap(intrusive_list& other) noexcept
+    {
         root.swap(other.root);
     }
 
@@ -371,7 +400,8 @@ private:
  * @param rhs The second list.
  */
 template<typename T>
-void swap(intrusive_list<T>& lhs, intrusive_list<T>& rhs) noexcept {
+void swap(intrusive_list<T>& lhs, intrusive_list<T>& rhs) noexcept
+{
     lhs.swap(rhs);
 }
 

include/mcl/hash/xmrx.hpp

@@ -0,0 +1,32 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+// Reference: http://jonkagstrom.com/bit-mixer-construction/
+
+#pragma once
+
+#include <functional>
+
+#include "mcl/bit/rotate.hpp"
+#include "mcl/stdint.hpp"
+
+namespace mcl::hash {
+
+constexpr size_t xmrx(size_t x)
+{
+    x ^= x >> 32;
+    x *= 0xff51afd7ed558ccd;
+    x ^= bit::rotate_right(x, 47) ^ bit::rotate_right(x, 23);
+    return x;
+}
+
+template<typename T>
+struct avalanche_xmrx {
+    size_t operator()(const T& value)
+    {
+        return xmrx(std::hash<T>{}(value));
+    }
+};
+
+}  // namespace mcl::hash

include/mcl/iterator/reverse.hpp

@@ -13,12 +13,14 @@ template<typename T>
 struct reverse_adapter {
     T& iterable;
 
-    constexpr auto begin() {
+    constexpr auto begin()
+    {
         using namespace std;
         return rbegin(iterable);
     }
 
-    constexpr auto end() {
+    constexpr auto end()
+    {
         using namespace std;
         return rend(iterable);
     }
@@ -27,7 +29,8 @@ struct reverse_adapter {
 }  // namespace detail
 
 template<typename T>
-constexpr detail::reverse_adapter<T> reverse(T&& iterable) {
+constexpr detail::reverse_adapter<T> reverse(T&& iterable)
+{
     return detail::reverse_adapter<T>{iterable};
 }
 

include/mcl/memory/overaligned_unique_ptr.hpp

@@ -0,0 +1,46 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <type_traits>
+
+#ifdef _MSC_VER
+#    include <malloc.h>
+#else
+#    include <cstdlib>
+#endif
+
+namespace mcl {
+
+namespace detail {
+struct aligned_alloc_deleter {
+    template<typename T>
+    void operator()(T* p) const
+    {
+#ifdef _MSC_VER
+        _aligned_free(const_cast<std::remove_const_t<T>*>(p));
+#else
+        std::free(const_cast<std::remove_const_t<T>*>(p));
+#endif
+    }
+};
+}  // namespace detail
+
+template<size_t, typename T>
+using overaligned_unique_ptr = std::unique_ptr<T, detail::aligned_alloc_deleter>;
+
+template<size_t alignment, typename T>
+auto make_overaligned_unique_ptr_array(size_t element_count)
+{
+    const size_t min_size = element_count * sizeof(T);
+    const size_t alloc_size = (min_size + alignment - 1) / alignment * alignment;
+#ifdef _MSC_VER
+    return overaligned_unique_ptr<alignment, T[]>{static_cast<T*>(_aligned_malloc(alloc_size, alignment))};
+#else
+    return overaligned_unique_ptr<alignment, T[]>{static_cast<T*>(std::aligned_alloc(alignment, alloc_size))};
+#endif
+}
+
+}  // namespace mcl

include/mcl/scope_exit.hpp

@@ -21,7 +21,9 @@ class scope_exit final {
 public:
     explicit scope_exit(Function&& fn)
         : function(std::move(fn)) {}
-    ~scope_exit() noexcept {
+
+    ~scope_exit() noexcept
+    {
         function();
     }
 
@@ -34,7 +36,9 @@ class scope_fail final {
 public:
     explicit scope_fail(Function&& fn)
         : function(std::move(fn)), exception_count(std::uncaught_exceptions()) {}
-    ~scope_fail() noexcept {
+
+    ~scope_fail() noexcept
+    {
         if (std::uncaught_exceptions() > exception_count) {
             function();
         }
@@ -50,7 +54,9 @@ class scope_success final {
 public:
     explicit scope_success(Function&& fn)
         : function(std::move(fn)), exception_count(std::uncaught_exceptions()) {}
-    ~scope_success() {
+
+    ~scope_success()
+    {
         if (std::uncaught_exceptions() <= exception_count) {
             function();
         }
@@ -64,17 +70,20 @@ private:
 // We use ->* here as it has the highest precedence of the operators we can use.
 
 template<typename Function>
-auto operator->*(scope_exit_tag, Function&& function) {
+auto operator->*(scope_exit_tag, Function&& function)
+{
     return scope_exit<std::decay_t<Function>>{std::forward<Function>(function)};
 }
 
 template<typename Function>
-auto operator->*(scope_fail_tag, Function&& function) {
+auto operator->*(scope_fail_tag, Function&& function)
+{
     return scope_fail<std::decay_t<Function>>{std::forward<Function>(function)};
 }
 
 template<typename Function>
-auto operator->*(scope_success_tag, Function&& function) {
+auto operator->*(scope_success_tag, Function&& function)
+{
     return scope_success<std::decay_t<Function>>{std::forward<Function>(function)};
 }
 

src/assert.cpp

@@ -11,9 +11,11 @@
 
 namespace mcl::detail {
 
-[[noreturn]] void assert_terminate_impl(fmt::string_view msg, fmt::format_args args) {
+[[noreturn]] void assert_terminate_impl(const char* expr_str, fmt::string_view msg, fmt::format_args args)
-    fmt::print(stderr, "assertion failed: ");
+{
+    fmt::print(stderr, "assertion failed: {}\nMessage:", expr_str);
     fmt::vprint(stderr, msg, args);
+    std::fflush(stderr);
     std::terminate();
 }
 

tests/CMakeLists.txt

@@ -1,13 +1,14 @@
 add_executable(mcl-tests
     bit/bit_field_tests.cpp
-    main.cpp
+    container/hmap.cpp
+    container/ihmap.cpp
     mp/metavalue_tests.cpp
     mp/typelist_tests.cpp
     type_traits/type_traits_tests.cpp
 )
 target_include_directories(mcl-tests PUBLIC .)
 target_compile_options(mcl-tests PRIVATE ${STAMINA_CXX_FLAGS})
-target_link_libraries(mcl-tests PRIVATE Catch2::Catch2 mcl)
+target_link_libraries(mcl-tests PRIVATE Catch2::Catch2WithMain mcl)
 
 include(CTest)
 include(Catch)

tests/bit/bit_field_tests.cpp

@@ -5,11 +5,12 @@
 #include <array>
 #include <tuple>
 
-#include <catch2/catch.hpp>
+#include <catch2/catch_test_macros.hpp>
 #include <mcl/bit/bit_field.hpp>
 #include <mcl/stdint.hpp>
 
-TEST_CASE("mcl::bit::ones", "[bit]") {
+TEST_CASE("mcl::bit::ones", "[bit]")
+{
     const std::array cases{
         std::make_tuple<size_t, u8>(0, 0x00),
         std::make_tuple<size_t, u8>(1, 0x01),

tests/container/hmap.cpp

@@ -0,0 +1,66 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#include <unordered_map>
+
+#include <catch2/catch_test_macros.hpp>
+#include <fmt/core.h>
+#include <mcl/container/hmap.hpp>
+#include <mcl/stdint.hpp>
+
+TEST_CASE("mcl::hmap", "[hmap]")
+{
+    mcl::hmap<u64, u64> double_map;
+
+    constexpr int count = 100000;
+
+    REQUIRE(double_map.empty());
+
+    for (int i = 0; i < count; ++i) {
+        double_map[i] = i * 2;
+        REQUIRE(double_map.size() == i + 1);
+    }
+
+    for (int i = 0; i < count; ++i) {
+        REQUIRE(double_map[i] == i * 2);
+        REQUIRE(double_map.contains(i));
+    }
+
+    for (int i = 0; i < count; ++i) {
+        auto iter = double_map.find(i);
+        REQUIRE(iter->first == i);
+        REQUIRE(iter->second == i * 2);
+    }
+
+    for (int i = count; i < count * 2; ++i) {
+        REQUIRE(!double_map.contains(i));
+    }
+
+    for (int i = 0; i < count; ++i) {
+        auto result = double_map.try_emplace(i, 0);
+        REQUIRE(!result.second);
+    }
+
+    for (auto [k, v] : double_map) {
+        REQUIRE(k * 2 == v);
+    }
+
+    std::unordered_map<u64, size_t> indexes_count;
+    for (auto [k, v] : double_map) {
+        (void)v;
+        indexes_count[k]++;
+    }
+    for (auto [k, v] : indexes_count) {
+        (void)k;
+        REQUIRE(v == 1);
+    }
+
+    REQUIRE(!double_map.empty());
+    double_map.clear();
+    REQUIRE(double_map.empty());
+
+    for (auto [k, v] : double_map) {
+        REQUIRE(false);
+    }
+}

tests/container/ihmap.cpp

@@ -0,0 +1,66 @@
+// This file is part of the mcl project.
+// Copyright (c) 2022 merryhime
+// SPDX-License-Identifier: MIT
+
+#include <unordered_map>
+
+#include <catch2/catch_test_macros.hpp>
+#include <fmt/core.h>
+#include <mcl/container/ihmap.hpp>
+#include <mcl/stdint.hpp>
+
+TEST_CASE("mcl::ihmap", "[ihmap]")
+{
+    mcl::ihmap<u64, u64> double_map;
+
+    constexpr int count = 100000;
+
+    REQUIRE(double_map.empty());
+
+    for (int i = 0; i < count; ++i) {
+        double_map[i] = i * 2;
+        REQUIRE(double_map.size() == i + 1);
+    }
+
+    for (int i = 0; i < count; ++i) {
+        REQUIRE(double_map[i] == i * 2);
+        REQUIRE(double_map.contains(i));
+    }
+
+    for (int i = 0; i < count; ++i) {
+        auto iter = double_map.find(i);
+        REQUIRE(iter->first == i);
+        REQUIRE(iter->second == i * 2);
+    }
+
+    for (int i = count; i < count * 2; ++i) {
+        REQUIRE(!double_map.contains(i));
+    }
+
+    for (int i = 0; i < count; ++i) {
+        auto result = double_map.try_emplace(i, 0);
+        REQUIRE(!result.second);
+    }
+
+    for (auto [k, v] : double_map) {
+        REQUIRE(k * 2 == v);
+    }
+
+    std::unordered_map<u64, size_t> indexes_count;
+    for (auto [k, v] : double_map) {
+        (void)v;
+        indexes_count[k]++;
+    }
+    for (auto [k, v] : indexes_count) {
+        (void)k;
+        REQUIRE(v == 1);
+    }
+
+    REQUIRE(!double_map.empty());
+    double_map.clear();
+    REQUIRE(double_map.empty());
+
+    for (auto [k, v] : double_map) {
+        REQUIRE(false);
+    }
+}

tests/main.cpp

@@ -1,6 +0,0 @@
-// This file is part of the mcl project.
-// Copyright (c) 2022 merryhime
-// SPDX-License-Identifier: MIT
-
-#define CATCH_CONFIG_MAIN
-#include "catch2/catch.hpp"