Subversion Repositories QNX 8.QNX8 LLVM/Clang compiler suite

//===- llvm/Support/Parallel.h - Parallel algorithms ----------------------===//

//

// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

// See https://llvm.org/LICENSE.txt for license information.

// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

//

//===----------------------------------------------------------------------===//

#ifndef LLVM_SUPPORT_PARALLEL_H

#define LLVM_SUPPORT_PARALLEL_H

#include "llvm/ADT/STLExtras.h"

#include "llvm/Config/llvm-config.h"

#include "llvm/Support/Error.h"

#include "llvm/Support/MathExtras.h"

#include "llvm/Support/Threading.h"

#include <algorithm>

#include <condition_variable>

#include <functional>

#include <mutex>

namespace llvm {

namespace parallel {

// Strategy for the default executor used by the parallel routines provided by

// this file. It defaults to using all hardware threads and should be

// initialized before the first use of parallel routines.

extern ThreadPoolStrategy strategy;

#if LLVM_ENABLE_THREADS

#ifdef _WIN32

// Direct access to thread_local variables from a different DLL isn't

// possible with Windows Native TLS.

unsigned getThreadIndex();

#else

// Don't access this directly, use the getThreadIndex wrapper.

extern thread_local unsigned threadIndex;

inline unsigned getThreadIndex() { return threadIndex; }

#endif

#else

inline unsigned getThreadIndex() { return 0; }

#endif

namespace detail {

class Latch {

  uint32_t Count;

  mutable std::mutex Mutex;

  mutable std::condition_variable Cond;

public:

  explicit Latch(uint32_t Count = 0) : Count(Count) {}

  ~Latch() {

    // Ensure at least that sync() was called.

    assert(Count == 0);

  }

  void inc() {

    std::lock_guard<std::mutex> lock(Mutex);

    ++Count;

  }

  void dec() {

    std::lock_guard<std::mutex> lock(Mutex);

    if (--Count == 0)

      Cond.notify_all();

  }

  void sync() const {

    std::unique_lock<std::mutex> lock(Mutex);

    Cond.wait(lock, [&] { return Count == 0; });

  }

};

} // namespace detail

class TaskGroup {

  detail::Latch L;

  bool Parallel;

public:

  TaskGroup();

  ~TaskGroup();

  // Spawn a task, but does not wait for it to finish.

  void spawn(std::function<void()> f);

  // Similar to spawn, but execute the task immediately when ThreadsRequested ==

  // 1. The difference is to give the following pattern a more intuitive order

  // when single threading is requested.

  //

  // for (size_t begin = 0, i = 0, taskSize = 0;;) {

  //   taskSize += ...

  //   bool done = ++i == end;

  //   if (done || taskSize >= taskSizeLimit) {

  //     tg.execute([=] { fn(begin, i); });

  //     if (done)

  //       break;

  //     begin = i;

  //     taskSize = 0;

  //   }

  // }

  void execute(std::function<void()> f);

  void sync() const { L.sync(); }

};

namespace detail {

#if LLVM_ENABLE_THREADS

const ptrdiff_t MinParallelSize = 1024;

/// Inclusive median.

template <class RandomAccessIterator, class Comparator>

RandomAccessIterator medianOf3(RandomAccessIterator Start,

                               RandomAccessIterator End,

                               const Comparator &Comp) {

  RandomAccessIterator Mid = Start + (std::distance(Start, End) / 2);

  return Comp(*Start, *(End - 1))

             ? (Comp(*Mid, *(End - 1)) ? (Comp(*Start, *Mid) ? Mid : Start)

                                       : End - 1)

             : (Comp(*Mid, *Start) ? (Comp(*(End - 1), *Mid) ? Mid : End - 1)

                                   : Start);

}

template <class RandomAccessIterator, class Comparator>

void parallel_quick_sort(RandomAccessIterator Start, RandomAccessIterator End,

                         const Comparator &Comp, TaskGroup &TG, size_t Depth) {

  // Do a sequential sort for small inputs.

  if (std::distance(Start, End) < detail::MinParallelSize || Depth == 0) {

    llvm::sort(Start, End, Comp);

    return;

  }

  // Partition.

  auto Pivot = medianOf3(Start, End, Comp);

  // Move Pivot to End.

  std::swap(*(End - 1), *Pivot);

  Pivot = std::partition(Start, End - 1, [&Comp, End](decltype(*Start) V) {

    return Comp(V, *(End - 1));

  });

  // Move Pivot to middle of partition.

  std::swap(*Pivot, *(End - 1));

  // Recurse.

  TG.spawn([=, &Comp, &TG] {

    parallel_quick_sort(Start, Pivot, Comp, TG, Depth - 1);

  });

  parallel_quick_sort(Pivot + 1, End, Comp, TG, Depth - 1);

}

template <class RandomAccessIterator, class Comparator>

void parallel_sort(RandomAccessIterator Start, RandomAccessIterator End,

                   const Comparator &Comp) {

  TaskGroup TG;

  parallel_quick_sort(Start, End, Comp, TG,

                      llvm::Log2_64(std::distance(Start, End)) + 1);

}

// TaskGroup has a relatively high overhead, so we want to reduce

// the number of spawn() calls. We'll create up to 1024 tasks here.

// (Note that 1024 is an arbitrary number. This code probably needs

// improving to take the number of available cores into account.)

enum { MaxTasksPerGroup = 1024 };

template <class IterTy, class ResultTy, class ReduceFuncTy,

          class TransformFuncTy>

ResultTy parallel_transform_reduce(IterTy Begin, IterTy End, ResultTy Init,

                                   ReduceFuncTy Reduce,

                                   TransformFuncTy Transform) {

  // Limit the number of tasks to MaxTasksPerGroup to limit job scheduling

  // overhead on large inputs.

  size_t NumInputs = std::distance(Begin, End);

  if (NumInputs == 0)

    return std::move(Init);

  size_t NumTasks = std::min(static_cast<size_t>(MaxTasksPerGroup), NumInputs);

  std::vector<ResultTy> Results(NumTasks, Init);

  {

    // Each task processes either TaskSize or TaskSize+1 inputs. Any inputs

    // remaining after dividing them equally amongst tasks are distributed as

    // one extra input over the first tasks.

    TaskGroup TG;

    size_t TaskSize = NumInputs / NumTasks;

    size_t RemainingInputs = NumInputs % NumTasks;

    IterTy TBegin = Begin;

    for (size_t TaskId = 0; TaskId < NumTasks; ++TaskId) {

      IterTy TEnd = TBegin + TaskSize + (TaskId < RemainingInputs ? 1 : 0);

      TG.spawn([=, &Transform, &Reduce, &Results] {

        // Reduce the result of transformation eagerly within each task.

        ResultTy R = Init;

        for (IterTy It = TBegin; It != TEnd; ++It)

          R = Reduce(R, Transform(*It));

        Results[TaskId] = R;

      });

      TBegin = TEnd;

    }

    assert(TBegin == End);

  }

  // Do a final reduction. There are at most 1024 tasks, so this only adds

  // constant single-threaded overhead for large inputs. Hopefully most

  // reductions are cheaper than the transformation.

  ResultTy FinalResult = std::move(Results.front());

  for (ResultTy &PartialResult :

       MutableArrayRef(Results.data() + 1, Results.size() - 1))

    FinalResult = Reduce(FinalResult, std::move(PartialResult));

  return std::move(FinalResult);

}

#endif

} // namespace detail

} // namespace parallel

template <class RandomAccessIterator,

          class Comparator = std::less<

              typename std::iterator_traits<RandomAccessIterator>::value_type>>

void parallelSort(RandomAccessIterator Start, RandomAccessIterator End,

                  const Comparator &Comp = Comparator()) {

#if LLVM_ENABLE_THREADS

  if (parallel::strategy.ThreadsRequested != 1) {

    parallel::detail::parallel_sort(Start, End, Comp);

    return;

  }

#endif

  llvm::sort(Start, End, Comp);

}

void parallelFor(size_t Begin, size_t End, function_ref<void(size_t)> Fn);

template <class IterTy, class FuncTy>

void parallelForEach(IterTy Begin, IterTy End, FuncTy Fn) {

  parallelFor(0, End - Begin, [&](size_t I) { Fn(Begin[I]); });

}

template <class IterTy, class ResultTy, class ReduceFuncTy,

          class TransformFuncTy>

ResultTy parallelTransformReduce(IterTy Begin, IterTy End, ResultTy Init,

                                 ReduceFuncTy Reduce,

                                 TransformFuncTy Transform) {

#if LLVM_ENABLE_THREADS

  if (parallel::strategy.ThreadsRequested != 1) {

    return parallel::detail::parallel_transform_reduce(Begin, End, Init, Reduce,

                                                       Transform);

  }

#endif

  for (IterTy I = Begin; I != End; ++I)

    Init = Reduce(std::move(Init), Transform(*I));

  return std::move(Init);

}

// Range wrappers.

template <class RangeTy,

          class Comparator = std::less<decltype(*std::begin(RangeTy()))>>

void parallelSort(RangeTy &&R, const Comparator &Comp = Comparator()) {

  parallelSort(std::begin(R), std::end(R), Comp);

}

template <class RangeTy, class FuncTy>

void parallelForEach(RangeTy &&R, FuncTy Fn) {

  parallelForEach(std::begin(R), std::end(R), Fn);

}

template <class RangeTy, class ResultTy, class ReduceFuncTy,

          class TransformFuncTy>

ResultTy parallelTransformReduce(RangeTy &&R, ResultTy Init,

                                 ReduceFuncTy Reduce,

                                 TransformFuncTy Transform) {

  return parallelTransformReduce(std::begin(R), std::end(R), Init, Reduce,

                                 Transform);

}

// Parallel for-each, but with error handling.

template <class RangeTy, class FuncTy>

Error parallelForEachError(RangeTy &&R, FuncTy Fn) {

  // The transform_reduce algorithm requires that the initial value be copyable.

  // Error objects are uncopyable. We only need to copy initial success values,

  // so work around this mismatch via the C API. The C API represents success

  // values with a null pointer. The joinErrors discards null values and joins

  // multiple errors into an ErrorList.

  return unwrap(parallelTransformReduce(

      std::begin(R), std::end(R), wrap(Error::success()),

      [](LLVMErrorRef Lhs, LLVMErrorRef Rhs) {

        return wrap(joinErrors(unwrap(Lhs), unwrap(Rhs)));

      },

      [&Fn](auto &&V) { return wrap(Fn(V)); }));

}

} // namespace llvm

#endif // LLVM_SUPPORT_PARALLEL_H