include/llvm/Support/Parallel.h

   1 //===- llvm/Support/Parallel.h - Parallel algorithms ----------------------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8
   9 #ifndef LLVM_SUPPORT_PARALLEL_H
  10 #define LLVM_SUPPORT_PARALLEL_H
  11
  12 #include "llvm/ADT/STLExtras.h"
  13 #include "llvm/Config/llvm-config.h"
  14 #include "llvm/Support/MathExtras.h"
  15
  16 #include <algorithm>
  17 #include <condition_variable>
  18 #include <functional>
  19 #include <mutex>
  20
  21 #if defined(_MSC_VER) && LLVM_ENABLE_THREADS
  22 #pragma warning(push)
  23 #pragma warning(disable : 4530)
  24 #include <concrt.h>
  25 #include <ppl.h>
  26 #pragma warning(pop)
  27 #endif
  28
  29 namespace llvm {
  30
  31 namespace parallel {
  32 struct sequential_execution_policy {};
  33 struct parallel_execution_policy {};
  34
  35 template <typename T>
  36 struct is_execution_policy
  37     : public std::integral_constant<
  38           bool, llvm::is_one_of<T, sequential_execution_policy,
  39                                 parallel_execution_policy>::value> {};
  40
  41 constexpr sequential_execution_policy seq{};
  42 constexpr parallel_execution_policy par{};
  43
  44 namespace detail {
  45
  46 #if LLVM_ENABLE_THREADS
  47
  48 class Latch {
  49   uint32_t Count;
  50   mutable std::mutex Mutex;
  51   mutable std::condition_variable Cond;
  52
  53 public:
  54   explicit Latch(uint32_t Count = 0) : Count(Count) {}
  55   ~Latch() { sync(); }
  56
  57   void inc() {
  58     std::lock_guard<std::mutex> lock(Mutex);
  59     ++Count;
  60   }
  61
  62   void dec() {
  63     std::lock_guard<std::mutex> lock(Mutex);
  64     if (--Count == 0)
  65       Cond.notify_all();
  66   }
  67
  68   void sync() const {
  69     std::unique_lock<std::mutex> lock(Mutex);
  70     Cond.wait(lock, [&] { return Count == 0; });
  71   }
  72 };
  73
  74 class TaskGroup {
  75   Latch L;
  76   bool Parallel;
  77
  78 public:
  79   TaskGroup();
  80   ~TaskGroup();
  81
  82   void spawn(std::function<void()> f);
  83
  84   void sync() const { L.sync(); }
  85 };
  86
  87 #if defined(_MSC_VER)
  88 template <class RandomAccessIterator, class Comparator>
  89 void parallel_sort(RandomAccessIterator Start, RandomAccessIterator End,
  90                    const Comparator &Comp) {
  91   concurrency::parallel_sort(Start, End, Comp);
  92 }
  93 template <class IterTy, class FuncTy>
  94 void parallel_for_each(IterTy Begin, IterTy End, FuncTy Fn) {
  95   concurrency::parallel_for_each(Begin, End, Fn);
  96 }
  97
  98 template <class IndexTy, class FuncTy>
  99 void parallel_for_each_n(IndexTy Begin, IndexTy End, FuncTy Fn) {
 100   concurrency::parallel_for(Begin, End, Fn);
 101 }
 102
 103 #else
 104 const ptrdiff_t MinParallelSize = 1024;
 105
 106 /// Inclusive median.
 107 template <class RandomAccessIterator, class Comparator>
 108 RandomAccessIterator medianOf3(RandomAccessIterator Start,
 109                                RandomAccessIterator End,
 110                                const Comparator &Comp) {
 111   RandomAccessIterator Mid = Start + (std::distance(Start, End) / 2);
 112   return Comp(*Start, *(End - 1))
 113              ? (Comp(*Mid, *(End - 1)) ? (Comp(*Start, *Mid) ? Mid : Start)
 114                                        : End - 1)
 115              : (Comp(*Mid, *Start) ? (Comp(*(End - 1), *Mid) ? Mid : End - 1)
 116                                    : Start);
 117 }
 118
 119 template <class RandomAccessIterator, class Comparator>
 120 void parallel_quick_sort(RandomAccessIterator Start, RandomAccessIterator End,
 121                          const Comparator &Comp, TaskGroup &TG, size_t Depth) {
 122   // Do a sequential sort for small inputs.
 123   if (std::distance(Start, End) < detail::MinParallelSize || Depth == 0) {
 124     llvm::sort(Start, End, Comp);
 125     return;
 126   }
 127
 128   // Partition.
 129   auto Pivot = medianOf3(Start, End, Comp);
 130   // Move Pivot to End.
 131   std::swap(*(End - 1), *Pivot);
 132   Pivot = std::partition(Start, End - 1, [&Comp, End](decltype(*Start) V) {
 133     return Comp(V, *(End - 1));
 134   });
 135   // Move Pivot to middle of partition.
 136   std::swap(*Pivot, *(End - 1));
 137
 138   // Recurse.
 139   TG.spawn([=, &Comp, &TG] {
 140     parallel_quick_sort(Start, Pivot, Comp, TG, Depth - 1);
 141   });
 142   parallel_quick_sort(Pivot + 1, End, Comp, TG, Depth - 1);
 143 }
 144
 145 template <class RandomAccessIterator, class Comparator>
 146 void parallel_sort(RandomAccessIterator Start, RandomAccessIterator End,
 147                    const Comparator &Comp) {
 148   TaskGroup TG;
 149   parallel_quick_sort(Start, End, Comp, TG,
 150                       llvm::Log2_64(std::distance(Start, End)) + 1);
 151 }
 152
 153 template <class IterTy, class FuncTy>
 154 void parallel_for_each(IterTy Begin, IterTy End, FuncTy Fn) {
 155   // TaskGroup has a relatively high overhead, so we want to reduce
 156   // the number of spawn() calls. We'll create up to 1024 tasks here.
 157   // (Note that 1024 is an arbitrary number. This code probably needs
 158   // improving to take the number of available cores into account.)
 159   ptrdiff_t TaskSize = std::distance(Begin, End) / 1024;
 160   if (TaskSize == 0)
 161     TaskSize = 1;
 162
 163   TaskGroup TG;
 164   while (TaskSize < std::distance(Begin, End)) {
 165     TG.spawn([=, &Fn] { std::for_each(Begin, Begin + TaskSize, Fn); });
 166     Begin += TaskSize;
 167   }
 168   std::for_each(Begin, End, Fn);
 169 }
 170
 171 template <class IndexTy, class FuncTy>
 172 void parallel_for_each_n(IndexTy Begin, IndexTy End, FuncTy Fn) {
 173   ptrdiff_t TaskSize = (End - Begin) / 1024;
 174   if (TaskSize == 0)
 175     TaskSize = 1;
 176
 177   TaskGroup TG;
 178   IndexTy I = Begin;
 179   for (; I + TaskSize < End; I += TaskSize) {
 180     TG.spawn([=, &Fn] {
 181       for (IndexTy J = I, E = I + TaskSize; J != E; ++J)
 182         Fn(J);
 183     });
 184   }
 185   for (IndexTy J = I; J < End; ++J)
 186     Fn(J);
 187 }
 188
 189 #endif
 190
 191 #endif
 192
 193 template <typename Iter>
 194 using DefComparator =
 195     std::less<typename std::iterator_traits<Iter>::value_type>;
 196
 197 } // namespace detail
 198
 199 // sequential algorithm implementations.
 200 template <class Policy, class RandomAccessIterator,
 201           class Comparator = detail::DefComparator<RandomAccessIterator>>
 202 void sort(Policy policy, RandomAccessIterator Start, RandomAccessIterator End,
 203           const Comparator &Comp = Comparator()) {
 204   static_assert(is_execution_policy<Policy>::value,
 205                 "Invalid execution policy!");
 206   llvm::sort(Start, End, Comp);
 207 }
 208
 209 template <class Policy, class IterTy, class FuncTy>
 210 void for_each(Policy policy, IterTy Begin, IterTy End, FuncTy Fn) {
 211   static_assert(is_execution_policy<Policy>::value,
 212                 "Invalid execution policy!");
 213   std::for_each(Begin, End, Fn);
 214 }
 215
 216 template <class Policy, class IndexTy, class FuncTy>
 217 void for_each_n(Policy policy, IndexTy Begin, IndexTy End, FuncTy Fn) {
 218   static_assert(is_execution_policy<Policy>::value,
 219                 "Invalid execution policy!");
 220   for (IndexTy I = Begin; I != End; ++I)
 221     Fn(I);
 222 }
 223
 224 // Parallel algorithm implementations, only available when LLVM_ENABLE_THREADS
 225 // is true.
 226 #if LLVM_ENABLE_THREADS
 227 template <class RandomAccessIterator,
 228           class Comparator = detail::DefComparator<RandomAccessIterator>>
 229 void sort(parallel_execution_policy policy, RandomAccessIterator Start,
 230           RandomAccessIterator End, const Comparator &Comp = Comparator()) {
 231   detail::parallel_sort(Start, End, Comp);
 232 }
 233
 234 template <class IterTy, class FuncTy>
 235 void for_each(parallel_execution_policy policy, IterTy Begin, IterTy End,
 236               FuncTy Fn) {
 237   detail::parallel_for_each(Begin, End, Fn);
 238 }
 239
 240 template <class IndexTy, class FuncTy>
 241 void for_each_n(parallel_execution_policy policy, IndexTy Begin, IndexTy End,
 242                 FuncTy Fn) {
 243   detail::parallel_for_each_n(Begin, End, Fn);
 244 }
 245 #endif
 246
 247 } // namespace parallel
 248 } // namespace llvm
 249
 250 #endif // LLVM_SUPPORT_PARALLEL_H