| 1 | /* |
| 2 | Copyright (c) 2005-2021 Intel Corporation |
| 3 | |
| 4 | Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | you may not use this file except in compliance with the License. |
| 6 | You may obtain a copy of the License at |
| 7 | |
| 8 | http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | |
| 10 | Unless required by applicable law or agreed to in writing, software |
| 11 | distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | See the License for the specific language governing permissions and |
| 14 | limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | #ifndef __TBB_detail__utils_H |
| 18 | #define __TBB_detail__utils_H |
| 19 | |
| 20 | #include <type_traits> |
| 21 | #include <cstdint> |
| 22 | #include <atomic> |
| 23 | |
| 24 | #include "_config.h" |
| 25 | #include "_assert.h" |
| 26 | #include "_machine.h" |
| 27 | |
| 28 | namespace tbb { |
| 29 | namespace detail { |
| 30 | inline namespace d0 { |
| 31 | |
| 32 | //! Utility template function to prevent "unused" warnings by various compilers. |
| 33 | template<typename... T> void suppress_unused_warning(T&&...) {} |
| 34 | |
| 35 | //! Compile-time constant that is upper bound on cache line/sector size. |
| 36 | /** It should be used only in situations where having a compile-time upper |
| 37 | bound is more useful than a run-time exact answer. |
| 38 | @ingroup memory_allocation */ |
| 39 | constexpr size_t max_nfs_size = 128; |
| 40 | constexpr std::size_t max_nfs_size_exp = 7; |
| 41 | static_assert(1 << max_nfs_size_exp == max_nfs_size, "max_nfs_size_exp must be a log2(max_nfs_size)" ); |
| 42 | |
| 43 | //! Class that implements exponential backoff. |
| 44 | class atomic_backoff { |
| 45 | //! Time delay, in units of "pause" instructions. |
| 46 | /** Should be equal to approximately the number of "pause" instructions |
| 47 | that take the same time as an context switch. Must be a power of two.*/ |
| 48 | static constexpr std::int32_t LOOPS_BEFORE_YIELD = 16; |
| 49 | std::int32_t count; |
| 50 | |
| 51 | public: |
| 52 | // In many cases, an object of this type is initialized eagerly on hot path, |
| 53 | // as in for(atomic_backoff b; ; b.pause()) { /*loop body*/ } |
| 54 | // For this reason, the construction cost must be very small! |
| 55 | atomic_backoff() : count(1) {} |
| 56 | // This constructor pauses immediately; do not use on hot paths! |
| 57 | atomic_backoff(bool) : count(1) { pause(); } |
| 58 | |
| 59 | //! No Copy |
| 60 | atomic_backoff(const atomic_backoff&) = delete; |
| 61 | atomic_backoff& operator=(const atomic_backoff&) = delete; |
| 62 | |
| 63 | //! Pause for a while. |
| 64 | void pause() { |
| 65 | if (count <= LOOPS_BEFORE_YIELD) { |
| 66 | machine_pause(delay: count); |
| 67 | // Pause twice as long the next time. |
| 68 | count *= 2; |
| 69 | } else { |
| 70 | // Pause is so long that we might as well yield CPU to scheduler. |
| 71 | yield(); |
| 72 | } |
| 73 | } |
| 74 | |
| 75 | //! Pause for a few times and return false if saturated. |
| 76 | bool bounded_pause() { |
| 77 | machine_pause(delay: count); |
| 78 | if (count < LOOPS_BEFORE_YIELD) { |
| 79 | // Pause twice as long the next time. |
| 80 | count *= 2; |
| 81 | return true; |
| 82 | } else { |
| 83 | return false; |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | void reset() { |
| 88 | count = 1; |
| 89 | } |
| 90 | }; |
| 91 | |
| 92 | //! Spin WHILE the condition is true. |
| 93 | /** T and U should be comparable types. */ |
| 94 | template <typename T, typename C> |
| 95 | T spin_wait_while(const std::atomic<T>& location, C comp, std::memory_order order) { |
| 96 | atomic_backoff backoff; |
| 97 | T snapshot = location.load(order); |
| 98 | while (comp(snapshot)) { |
| 99 | backoff.pause(); |
| 100 | snapshot = location.load(order); |
| 101 | } |
| 102 | return snapshot; |
| 103 | } |
| 104 | |
| 105 | //! Spin WHILE the value of the variable is equal to a given value |
| 106 | /** T and U should be comparable types. */ |
| 107 | template <typename T, typename U> |
| 108 | T spin_wait_while_eq(const std::atomic<T>& location, const U value, std::memory_order order = std::memory_order_acquire) { |
| 109 | return spin_wait_while(location, [&value](T t) { return t == value; }, order); |
| 110 | } |
| 111 | |
| 112 | //! Spin UNTIL the value of the variable is equal to a given value |
| 113 | /** T and U should be comparable types. */ |
| 114 | template<typename T, typename U> |
| 115 | T spin_wait_until_eq(const std::atomic<T>& location, const U value, std::memory_order order = std::memory_order_acquire) { |
| 116 | return spin_wait_while(location, [&value](T t) { return t != value; }, order); |
| 117 | } |
| 118 | |
| 119 | //! Spin UNTIL the condition returns true or spinning time is up. |
| 120 | /** Returns what the passed functor returned last time it was invoked. */ |
| 121 | template <typename Condition> |
| 122 | bool timed_spin_wait_until(Condition condition) { |
| 123 | // 32 pauses + 32 yields are meausered as balanced spin time before sleep. |
| 124 | bool finish = condition(); |
| 125 | for (int i = 1; !finish && i < 32; finish = condition(), i *= 2) { |
| 126 | machine_pause(delay: i); |
| 127 | } |
| 128 | for (int i = 32; !finish && i < 64; finish = condition(), ++i) { |
| 129 | yield(); |
| 130 | } |
| 131 | return finish; |
| 132 | } |
| 133 | |
| 134 | template <typename T> |
| 135 | std::uintptr_t log2(T in) { |
| 136 | __TBB_ASSERT(in > 0, "The logarithm of a non-positive value is undefined." ); |
| 137 | return machine_log2(in); |
| 138 | } |
| 139 | |
| 140 | template<typename T> |
| 141 | T reverse_bits(T src) { |
| 142 | return machine_reverse_bits(src); |
| 143 | } |
| 144 | |
| 145 | template<typename T> |
| 146 | T reverse_n_bits(T src, std::size_t n) { |
| 147 | __TBB_ASSERT(n != 0, "Reverse for 0 bits is undefined behavior." ); |
| 148 | return reverse_bits(src) >> (number_of_bits<T>() - n); |
| 149 | } |
| 150 | |
| 151 | // A function to check if passed integer is a power of two |
| 152 | template <typename IntegerType> |
| 153 | constexpr bool is_power_of_two( IntegerType arg ) { |
| 154 | static_assert(std::is_integral<IntegerType>::value, |
| 155 | "An argument for is_power_of_two should be integral type" ); |
| 156 | return arg && (0 == (arg & (arg - 1))); |
| 157 | } |
| 158 | |
| 159 | // A function to determine if passed integer is a power of two |
| 160 | // at least as big as another power of two, i.e. for strictly positive i and j, |
| 161 | // with j being a power of two, determines whether i==j<<k for some nonnegative k |
| 162 | template <typename ArgIntegerType, typename DivisorIntegerType> |
| 163 | constexpr bool is_power_of_two_at_least(ArgIntegerType arg, DivisorIntegerType divisor) { |
| 164 | // Divisor should be a power of two |
| 165 | static_assert(std::is_integral<ArgIntegerType>::value, |
| 166 | "An argument for is_power_of_two_at_least should be integral type" ); |
| 167 | return 0 == (arg & (arg - divisor)); |
| 168 | } |
| 169 | |
| 170 | // A function to compute arg modulo divisor where divisor is a power of 2. |
| 171 | template<typename ArgIntegerType, typename DivisorIntegerType> |
| 172 | inline ArgIntegerType modulo_power_of_two(ArgIntegerType arg, DivisorIntegerType divisor) { |
| 173 | __TBB_ASSERT( is_power_of_two(divisor), "Divisor should be a power of two" ); |
| 174 | return arg & (divisor - 1); |
| 175 | } |
| 176 | |
| 177 | //! A function to check if passed in pointer is aligned on a specific border |
| 178 | template<typename T> |
| 179 | constexpr bool is_aligned(T* pointer, std::uintptr_t alignment) { |
| 180 | return 0 == ((std::uintptr_t)pointer & (alignment - 1)); |
| 181 | } |
| 182 | |
| 183 | #if TBB_USE_ASSERT |
| 184 | static void* const poisoned_ptr = reinterpret_cast<void*>(-1); |
| 185 | |
| 186 | //! Set p to invalid pointer value. |
| 187 | template<typename T> |
| 188 | inline void poison_pointer( T* &p ) { p = reinterpret_cast<T*>(poisoned_ptr); } |
| 189 | |
| 190 | template<typename T> |
| 191 | inline void poison_pointer(std::atomic<T*>& p) { p.store(reinterpret_cast<T*>(poisoned_ptr), std::memory_order_relaxed); } |
| 192 | |
| 193 | /** Expected to be used in assertions only, thus no empty form is defined. **/ |
| 194 | template<typename T> |
| 195 | inline bool is_poisoned( T* p ) { return p == reinterpret_cast<T*>(poisoned_ptr); } |
| 196 | |
| 197 | template<typename T> |
| 198 | inline bool is_poisoned(const std::atomic<T*>& p) { return is_poisoned(p.load(std::memory_order_relaxed)); } |
| 199 | #else |
| 200 | template<typename T> |
| 201 | inline void poison_pointer(T&) {/*do nothing*/} |
| 202 | #endif /* !TBB_USE_ASSERT */ |
| 203 | |
| 204 | template <std::size_t alignment = 0, typename T> |
| 205 | bool assert_pointer_valid(T* p, const char* = nullptr) { |
| 206 | suppress_unused_warning(p, comment); |
| 207 | __TBB_ASSERT(p != nullptr, comment); |
| 208 | __TBB_ASSERT(!is_poisoned(p), comment); |
| 209 | #if !(_MSC_VER && _MSC_VER <= 1900 && !__INTEL_COMPILER) |
| 210 | __TBB_ASSERT(is_aligned(p, alignment == 0 ? alignof(T) : alignment), comment); |
| 211 | #endif |
| 212 | // Returns something to simplify assert_pointers_valid implementation. |
| 213 | return true; |
| 214 | } |
| 215 | |
| 216 | template <typename... Args> |
| 217 | void assert_pointers_valid(Args*... p) { |
| 218 | // suppress_unused_warning is used as an evaluation context for the variadic pack. |
| 219 | suppress_unused_warning(assert_pointer_valid(p)...); |
| 220 | } |
| 221 | |
| 222 | //! Base class for types that should not be assigned. |
| 223 | class no_assign { |
| 224 | public: |
| 225 | void operator=(const no_assign&) = delete; |
| 226 | no_assign(const no_assign&) = default; |
| 227 | no_assign() = default; |
| 228 | }; |
| 229 | |
| 230 | //! Base class for types that should not be copied or assigned. |
| 231 | class no_copy: no_assign { |
| 232 | public: |
| 233 | no_copy(const no_copy&) = delete; |
| 234 | no_copy() = default; |
| 235 | }; |
| 236 | |
| 237 | template <typename T> |
| 238 | void swap_atomics_relaxed(std::atomic<T>& lhs, std::atomic<T>& rhs){ |
| 239 | T tmp = lhs.load(std::memory_order_relaxed); |
| 240 | lhs.store(rhs.load(std::memory_order_relaxed), std::memory_order_relaxed); |
| 241 | rhs.store(tmp, std::memory_order_relaxed); |
| 242 | } |
| 243 | |
| 244 | //! One-time initialization states |
| 245 | enum class do_once_state { |
| 246 | uninitialized = 0, ///< No execution attempts have been undertaken yet |
| 247 | pending, ///< A thread is executing associated do-once routine |
| 248 | executed, ///< Do-once routine has been executed |
| 249 | initialized = executed ///< Convenience alias |
| 250 | }; |
| 251 | |
| 252 | //! One-time initialization function |
| 253 | /** /param initializer Pointer to function without arguments |
| 254 | The variant that returns bool is used for cases when initialization can fail |
| 255 | and it is OK to continue execution, but the state should be reset so that |
| 256 | the initialization attempt was repeated the next time. |
| 257 | /param state Shared state associated with initializer that specifies its |
| 258 | initialization state. Must be initially set to #uninitialized value |
| 259 | (e.g. by means of default static zero initialization). **/ |
| 260 | template <typename F> |
| 261 | void atomic_do_once( const F& initializer, std::atomic<do_once_state>& state ) { |
| 262 | // The loop in the implementation is necessary to avoid race when thread T2 |
| 263 | // that arrived in the middle of initialization attempt by another thread T1 |
| 264 | // has just made initialization possible. |
| 265 | // In such a case T2 has to rely on T1 to initialize, but T1 may already be past |
| 266 | // the point where it can recognize the changed conditions. |
| 267 | do_once_state expected_state; |
| 268 | while ( state.load( m: std::memory_order_acquire ) != do_once_state::executed ) { |
| 269 | if( state.load( m: std::memory_order_relaxed ) == do_once_state::uninitialized ) { |
| 270 | expected_state = do_once_state::uninitialized; |
| 271 | #if defined(__INTEL_COMPILER) && __INTEL_COMPILER <= 1910 |
| 272 | using enum_type = typename std::underlying_type<do_once_state>::type; |
| 273 | if( ((std::atomic<enum_type>&)state).compare_exchange_strong( (enum_type&)expected_state, (enum_type)do_once_state::pending ) ) { |
| 274 | #else |
| 275 | if( state.compare_exchange_strong( e&: expected_state, i: do_once_state::pending ) ) { |
| 276 | #endif |
| 277 | run_initializer( initializer, state ); |
| 278 | break; |
| 279 | } |
| 280 | } |
| 281 | spin_wait_while_eq( location: state, value: do_once_state::pending ); |
| 282 | } |
| 283 | } |
| 284 | |
| 285 | // Run the initializer which can not fail |
| 286 | template<typename Functor> |
| 287 | void run_initializer(const Functor& f, std::atomic<do_once_state>& state ) { |
| 288 | f(); |
| 289 | state.store(i: do_once_state::executed, m: std::memory_order_release); |
| 290 | } |
| 291 | |
| 292 | #if __TBB_CPP20_CONCEPTS_PRESENT |
| 293 | template <typename T> |
| 294 | concept boolean_testable_impl = std::convertible_to<T, bool>; |
| 295 | |
| 296 | template <typename T> |
| 297 | concept boolean_testable = boolean_testable_impl<T> && requires( T&& t ) { |
| 298 | { !std::forward<T>(t) } -> boolean_testable_impl; |
| 299 | }; |
| 300 | |
| 301 | #if __TBB_CPP20_COMPARISONS_PRESENT |
| 302 | struct synthesized_three_way_comparator { |
| 303 | template <typename T1, typename T2> |
| 304 | auto operator()( const T1& lhs, const T2& rhs ) const |
| 305 | requires requires { |
| 306 | { lhs < rhs } -> boolean_testable; |
| 307 | { rhs < lhs } -> boolean_testable; |
| 308 | } |
| 309 | { |
| 310 | if constexpr (std::three_way_comparable_with<T1, T2>) { |
| 311 | return lhs <=> rhs; |
| 312 | } else { |
| 313 | if (lhs < rhs) { |
| 314 | return std::weak_ordering::less; |
| 315 | } |
| 316 | if (rhs < lhs) { |
| 317 | return std::weak_ordering::greater; |
| 318 | } |
| 319 | return std::weak_ordering::equivalent; |
| 320 | } |
| 321 | } |
| 322 | }; // struct synthesized_three_way_comparator |
| 323 | |
| 324 | template <typename T1, typename T2 = T1> |
| 325 | using synthesized_three_way_result = decltype(synthesized_three_way_comparator{}(std::declval<T1&>(), |
| 326 | std::declval<T2&>())); |
| 327 | |
| 328 | #endif // __TBB_CPP20_COMPARISONS_PRESENT |
| 329 | |
| 330 | // Check if the type T is implicitly OR explicitly convertible to U |
| 331 | template <typename T, typename U> |
| 332 | concept relaxed_convertible_to = std::constructible_from<U, T>; |
| 333 | |
| 334 | template <typename T, typename U> |
| 335 | concept adaptive_same_as = |
| 336 | #if __TBB_STRICT_CONSTRAINTS |
| 337 | std::same_as<T, U>; |
| 338 | #else |
| 339 | std::convertible_to<T, U>; |
| 340 | #endif |
| 341 | #endif // __TBB_CPP20_CONCEPTS_PRESENT |
| 342 | |
| 343 | } // namespace d0 |
| 344 | |
| 345 | namespace d1 { |
| 346 | |
| 347 | class delegate_base { |
| 348 | public: |
| 349 | virtual bool operator()() const = 0; |
| 350 | virtual ~delegate_base() {} |
| 351 | }; |
| 352 | |
| 353 | template <typename FuncType> |
| 354 | class delegated_function : public delegate_base { |
| 355 | public: |
| 356 | delegated_function(FuncType& f) : my_func(f) {} |
| 357 | |
| 358 | bool operator()() const override { |
| 359 | return my_func(); |
| 360 | } |
| 361 | |
| 362 | private: |
| 363 | FuncType &my_func; |
| 364 | }; |
| 365 | } // namespace d1 |
| 366 | |
| 367 | } // namespace detail |
| 368 | } // namespace tbb |
| 369 | |
| 370 | #endif // __TBB_detail__utils_H |
| 371 | |