LLVM  8.0.1
Parallel.h
Go to the documentation of this file.
1 //===- llvm/Support/Parallel.h - Parallel algorithms ----------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #ifndef LLVM_SUPPORT_PARALLEL_H
11 #define LLVM_SUPPORT_PARALLEL_H
12 
13 #include "llvm/ADT/STLExtras.h"
14 #include "llvm/Config/llvm-config.h"
16 
17 #include <algorithm>
18 #include <condition_variable>
19 #include <functional>
20 #include <mutex>
21 
22 #if defined(_MSC_VER) && LLVM_ENABLE_THREADS
23 #pragma warning(push)
24 #pragma warning(disable : 4530)
25 #include <concrt.h>
26 #include <ppl.h>
27 #pragma warning(pop)
28 #endif
29 
30 namespace llvm {
31 
32 namespace parallel {
35 
36 template <typename T>
38  : public std::integral_constant<
39  bool, llvm::is_one_of<T, sequential_execution_policy,
40  parallel_execution_policy>::value> {};
41 
44 
45 namespace detail {
46 
47 #if LLVM_ENABLE_THREADS
48 
49 class Latch {
50  uint32_t Count;
51  mutable std::mutex Mutex;
52  mutable std::condition_variable Cond;
53 
54 public:
55  explicit Latch(uint32_t Count = 0) : Count(Count) {}
56  ~Latch() { sync(); }
57 
58  void inc() {
59  std::lock_guard<std::mutex> lock(Mutex);
60  ++Count;
61  }
62 
63  void dec() {
64  std::lock_guard<std::mutex> lock(Mutex);
65  if (--Count == 0)
66  Cond.notify_all();
67  }
68 
69  void sync() const {
70  std::unique_lock<std::mutex> lock(Mutex);
71  Cond.wait(lock, [&] { return Count == 0; });
72  }
73 };
74 
75 class TaskGroup {
76  Latch L;
77 
78 public:
79  void spawn(std::function<void()> f);
80 
81  void sync() const { L.sync(); }
82 };
83 
84 #if defined(_MSC_VER)
85 template <class RandomAccessIterator, class Comparator>
86 void parallel_sort(RandomAccessIterator Start, RandomAccessIterator End,
87  const Comparator &Comp) {
88  concurrency::parallel_sort(Start, End, Comp);
89 }
90 template <class IterTy, class FuncTy>
91 void parallel_for_each(IterTy Begin, IterTy End, FuncTy Fn) {
92  concurrency::parallel_for_each(Begin, End, Fn);
93 }
94 
95 template <class IndexTy, class FuncTy>
96 void parallel_for_each_n(IndexTy Begin, IndexTy End, FuncTy Fn) {
97  concurrency::parallel_for(Begin, End, Fn);
98 }
99 
100 #else
102 
103 /// Inclusive median.
104 template <class RandomAccessIterator, class Comparator>
105 RandomAccessIterator medianOf3(RandomAccessIterator Start,
106  RandomAccessIterator End,
107  const Comparator &Comp) {
108  RandomAccessIterator Mid = Start + (std::distance(Start, End) / 2);
109  return Comp(*Start, *(End - 1))
110  ? (Comp(*Mid, *(End - 1)) ? (Comp(*Start, *Mid) ? Mid : Start)
111  : End - 1)
112  : (Comp(*Mid, *Start) ? (Comp(*(End - 1), *Mid) ? Mid : End - 1)
113  : Start);
114 }
115 
116 template <class RandomAccessIterator, class Comparator>
117 void parallel_quick_sort(RandomAccessIterator Start, RandomAccessIterator End,
118  const Comparator &Comp, TaskGroup &TG, size_t Depth) {
119  // Do a sequential sort for small inputs.
120  if (std::distance(Start, End) < detail::MinParallelSize || Depth == 0) {
121  llvm::sort(Start, End, Comp);
122  return;
123  }
124 
125  // Partition.
126  auto Pivot = medianOf3(Start, End, Comp);
127  // Move Pivot to End.
128  std::swap(*(End - 1), *Pivot);
129  Pivot = std::partition(Start, End - 1, [&Comp, End](decltype(*Start) V) {
130  return Comp(V, *(End - 1));
131  });
132  // Move Pivot to middle of partition.
133  std::swap(*Pivot, *(End - 1));
134 
135  // Recurse.
136  TG.spawn([=, &Comp, &TG] {
137  parallel_quick_sort(Start, Pivot, Comp, TG, Depth - 1);
138  });
139  parallel_quick_sort(Pivot + 1, End, Comp, TG, Depth - 1);
140 }
141 
142 template <class RandomAccessIterator, class Comparator>
143 void parallel_sort(RandomAccessIterator Start, RandomAccessIterator End,
144  const Comparator &Comp) {
145  TaskGroup TG;
146  parallel_quick_sort(Start, End, Comp, TG,
147  llvm::Log2_64(std::distance(Start, End)) + 1);
148 }
149 
150 template <class IterTy, class FuncTy>
151 void parallel_for_each(IterTy Begin, IterTy End, FuncTy Fn) {
152  // TaskGroup has a relatively high overhead, so we want to reduce
153  // the number of spawn() calls. We'll create up to 1024 tasks here.
154  // (Note that 1024 is an arbitrary number. This code probably needs
155  // improving to take the number of available cores into account.)
156  ptrdiff_t TaskSize = std::distance(Begin, End) / 1024;
157  if (TaskSize == 0)
158  TaskSize = 1;
159 
160  TaskGroup TG;
161  while (TaskSize < std::distance(Begin, End)) {
162  TG.spawn([=, &Fn] { std::for_each(Begin, Begin + TaskSize, Fn); });
163  Begin += TaskSize;
164  }
165  std::for_each(Begin, End, Fn);
166 }
167 
168 template <class IndexTy, class FuncTy>
169 void parallel_for_each_n(IndexTy Begin, IndexTy End, FuncTy Fn) {
170  ptrdiff_t TaskSize = (End - Begin) / 1024;
171  if (TaskSize == 0)
172  TaskSize = 1;
173 
174  TaskGroup TG;
175  IndexTy I = Begin;
176  for (; I + TaskSize < End; I += TaskSize) {
177  TG.spawn([=, &Fn] {
178  for (IndexTy J = I, E = I + TaskSize; J != E; ++J)
179  Fn(J);
180  });
181  }
182  for (IndexTy J = I; J < End; ++J)
183  Fn(J);
184 }
185 
186 #endif
187 
188 #endif
189 
190 template <typename Iter>
191 using DefComparator =
192  std::less<typename std::iterator_traits<Iter>::value_type>;
193 
194 } // namespace detail
195 
196 // sequential algorithm implementations.
197 template <class Policy, class RandomAccessIterator,
199 void sort(Policy policy, RandomAccessIterator Start, RandomAccessIterator End,
200  const Comparator &Comp = Comparator()) {
202  "Invalid execution policy!");
203  llvm::sort(Start, End, Comp);
204 }
205 
206 template <class Policy, class IterTy, class FuncTy>
207 void for_each(Policy policy, IterTy Begin, IterTy End, FuncTy Fn) {
209  "Invalid execution policy!");
210  std::for_each(Begin, End, Fn);
211 }
212 
213 template <class Policy, class IndexTy, class FuncTy>
214 void for_each_n(Policy policy, IndexTy Begin, IndexTy End, FuncTy Fn) {
216  "Invalid execution policy!");
217  for (IndexTy I = Begin; I != End; ++I)
218  Fn(I);
219 }
220 
221 // Parallel algorithm implementations, only available when LLVM_ENABLE_THREADS
222 // is true.
223 #if LLVM_ENABLE_THREADS
224 template <class RandomAccessIterator,
226 void sort(parallel_execution_policy policy, RandomAccessIterator Start,
227  RandomAccessIterator End, const Comparator &Comp = Comparator()) {
228  detail::parallel_sort(Start, End, Comp);
229 }
230 
231 template <class IterTy, class FuncTy>
232 void for_each(parallel_execution_policy policy, IterTy Begin, IterTy End,
233  FuncTy Fn) {
234  detail::parallel_for_each(Begin, End, Fn);
235 }
236 
237 template <class IndexTy, class FuncTy>
238 void for_each_n(parallel_execution_policy policy, IndexTy Begin, IndexTy End,
239  FuncTy Fn) {
240  detail::parallel_for_each_n(Begin, End, Fn);
241 }
242 #endif
243 
244 } // namespace parallel
245 } // namespace llvm
246 
247 #endif // LLVM_SUPPORT_PARALLEL_H
const ptrdiff_t MinParallelSize
Definition: Parallel.h:101
void for_each_n(Policy policy, IndexTy Begin, IndexTy End, FuncTy Fn)
Definition: Parallel.h:214
This class represents lattice values for constants.
Definition: AllocatorList.h:24
void parallel_for_each_n(IndexTy Begin, IndexTy End, FuncTy Fn)
Definition: Parallel.h:169
constexpr sequential_execution_policy seq
Definition: Parallel.h:42
Latch(uint32_t Count=0)
Definition: Parallel.h:55
SmartMutex< false > Mutex
Mutex - A standard, always enforced mutex.
Definition: Mutex.h:139
auto partition(R &&Range, UnaryPredicate P) -> decltype(adl_begin(Range))
Provide wrappers to std::partition which take ranges instead of having to pass begin/end explicitly...
Definition: STLExtras.h:1275
void spawn(std::function< void()> f)
Definition: Parallel.cpp:122
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
void sort(IteratorTy Start, IteratorTy End)
Definition: STLExtras.h:1116
constexpr parallel_execution_policy par
Definition: Parallel.h:43
void for_each(Policy policy, IterTy Begin, IterTy End, FuncTy Fn)
Definition: Parallel.h:207
void swap(llvm::BitVector &LHS, llvm::BitVector &RHS)
Implement std::swap in terms of BitVector swap.
Definition: BitVector.h:941
std::less< typename std::iterator_traits< Iter >::value_type > DefComparator
Definition: Parallel.h:192
void parallel_for_each(IterTy Begin, IterTy End, FuncTy Fn)
Definition: Parallel.h:151
#define I(x, y, z)
Definition: MD5.cpp:58
void parallel_sort(RandomAccessIterator Start, RandomAccessIterator End, const Comparator &Comp)
Definition: Parallel.h:143
RandomAccessIterator medianOf3(RandomAccessIterator Start, RandomAccessIterator End, const Comparator &Comp)
Inclusive median.
Definition: Parallel.h:105
void for_each(parallel_execution_policy policy, IterTy Begin, IterTy End, FuncTy Fn)
Definition: Parallel.h:232
void parallel_quick_sort(RandomAccessIterator Start, RandomAccessIterator End, const Comparator &Comp, TaskGroup &TG, size_t Depth)
Definition: Parallel.h:117
print Print MemDeps of function
void sort(Policy policy, RandomAccessIterator Start, RandomAccessIterator End, const Comparator &Comp=Comparator())
Definition: Parallel.h:199
unsigned Log2_64(uint64_t Value)
Return the floor log base 2 of the specified value, -1 if the value is zero.
Definition: MathExtras.h:545