//===- LoopGeneratorsKMP.h - IR helper to create loops ----------*- C++ -*-===//
 
//
 
// 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
 
//
 
//===----------------------------------------------------------------------===//
 
//
 
// This file contains functions to create scalar and OpenMP parallel loops
 
// as LLVM-IR.
 
//
 
//===----------------------------------------------------------------------===//
 
#ifndef POLLY_LOOP_GENERATORS_KMP_H
 
#define POLLY_LOOP_GENERATORS_KMP_H
 
 
 
#include "polly/CodeGen/IRBuilder.h"
 
#include "polly/CodeGen/LoopGenerators.h"
 
#include "polly/Support/ScopHelper.h"
 
#include "llvm/ADT/SetVector.h"
 
 
 
namespace polly {
 
using llvm::GlobalValue;
 
using llvm::GlobalVariable;
 
 
 
/// This ParallelLoopGenerator subclass handles the generation of parallelized
 
/// code, utilizing the LLVM OpenMP library.
 
class ParallelLoopGeneratorKMP final : public ParallelLoopGenerator {
 
public:
 
  /// Create a parallel loop generator for the current function.
 
  ParallelLoopGeneratorKMP(PollyIRBuilder &Builder, LoopInfo &LI,
 
                           DominatorTree &DT, const DataLayout &DL)
 
      : ParallelLoopGenerator(Builder, LI, DT, DL) {
 
    SourceLocationInfo = createSourceLocation();
 
  }
 
 
 
protected:
 
  /// The source location struct of this loop.
 
  /// ident_t = type { i32, i32, i32, i32, i8* }
 
  GlobalValue *SourceLocationInfo;
 
 
 
  /// Convert the combination of given chunk size and scheduling type (which
 
  /// might have been set via the command line) into the corresponding
 
  /// scheduling type. This may result (e.g.) in a 'change' from
 
  /// "static chunked" scheduling to "static non-chunked" (regarding the
 
  /// provided and returned scheduling types).
 
  ///
 
  /// @param ChunkSize    The chunk size, set via command line or its default.
 
  /// @param Scheduling   The scheduling, set via command line or its default.
 
  ///
 
  /// @return The corresponding OMPGeneralSchedulingType.
 
  OMPGeneralSchedulingType
 
  getSchedType(int ChunkSize, OMPGeneralSchedulingType Scheduling) const;
 
 
 
  /// Returns True if 'LongType' is 64bit wide, otherwise: False.
 
  bool is64BitArch();
 
 
 
public:
 
  // The functions below may be used if one does not want to generate a
 
  // specific OpenMP parallel loop, but generate individual parts of it
 
  // (e.g. the subfunction definition).
 
 
 
  /// Create a runtime library call to spawn the worker threads.
 
  ///
 
  /// @param SubFn      The subfunction which holds the loop body.
 
  /// @param SubFnParam The parameter for the subfunction (basically the struct
 
  ///                   filled with the outside values).
 
  /// @param LB         The lower bound for the loop we parallelize.
 
  /// @param UB         The upper bound for the loop we parallelize.
 
  /// @param Stride     The stride of the loop we parallelize.
 
  void createCallSpawnThreads(Value *SubFn, Value *SubFnParam, Value *LB,
 
                              Value *UB, Value *Stride);
 
 
 
  void deployParallelExecution(Function *SubFn, Value *SubFnParam, Value *LB,
 
                               Value *UB, Value *Stride) override;
 
 
 
  Function *prepareSubFnDefinition(Function *F) const override;
 
 
 
  std::tuple<Value *, Function *> createSubFn(Value *Stride, AllocaInst *Struct,
 
                                              SetVector<Value *> UsedValues,
 
                                              ValueMapT &VMap) override;
 
 
 
  /// Create a runtime library call to get the current global thread number.
 
  ///
 
  /// @return A Value ref which holds the current global thread number.
 
  Value *createCallGlobalThreadNum();
 
 
 
  /// Create a runtime library call to request a number of threads.
 
  /// Which will be used in the next OpenMP section (by the next fork).
 
  ///
 
  /// @param GlobalThreadID   The global thread ID.
 
  /// @param NumThreads       The number of threads to use.
 
  void createCallPushNumThreads(Value *GlobalThreadID, Value *NumThreads);
 
 
 
  /// Create a runtime library call to prepare the OpenMP runtime.
 
  /// For dynamically scheduled loops, saving the loop arguments.
 
  ///
 
  /// @param GlobalThreadID   The global thread ID.
 
  /// @param LB               The loop's lower bound.
 
  /// @param UB               The loop's upper bound.
 
  /// @param Inc              The loop increment.
 
  /// @param ChunkSize        The chunk size of the parallel loop.
 
  void createCallDispatchInit(Value *GlobalThreadID, Value *LB, Value *UB,
 
                              Value *Inc, Value *ChunkSize);
 
 
 
  /// Create a runtime library call to retrieve the next (dynamically)
 
  /// allocated chunk of work for this thread.
 
  ///
 
  /// @param GlobalThreadID   The global thread ID.
 
  /// @param IsLastPtr        Pointer to a flag, which is set to 1 if this is
 
  ///                         the last chunk of work, or 0 otherwise.
 
  /// @param LBPtr            Pointer to the lower bound for the next chunk.
 
  /// @param UBPtr            Pointer to the upper bound for the next chunk.
 
  /// @param StridePtr        Pointer to the stride for the next chunk.
 
  ///
 
  /// @return A Value which holds 1 if there is work to be done, 0 otherwise.
 
  Value *createCallDispatchNext(Value *GlobalThreadID, Value *IsLastPtr,
 
                                Value *LBPtr, Value *UBPtr, Value *StridePtr);
 
 
 
  /// Create a runtime library call to prepare the OpenMP runtime.
 
  /// For statically scheduled loops, saving the loop arguments.
 
  ///
 
  /// @param GlobalThreadID   The global thread ID.
 
  /// @param IsLastPtr        Pointer to a flag, which is set to 1 if this is
 
  ///                         the last chunk of work, or 0 otherwise.
 
  /// @param LBPtr            Pointer to the lower bound for the next chunk.
 
  /// @param UBPtr            Pointer to the upper bound for the next chunk.
 
  /// @param StridePtr        Pointer to the stride for the next chunk.
 
  /// @param ChunkSize        The chunk size of the parallel loop.
 
  void createCallStaticInit(Value *GlobalThreadID, Value *IsLastPtr,
 
                            Value *LBPtr, Value *UBPtr, Value *StridePtr,
 
                            Value *ChunkSize);
 
 
 
  /// Create a runtime library call to mark the end of
 
  /// a statically scheduled loop.
 
  ///
 
  /// @param GlobalThreadID   The global thread ID.
 
  void createCallStaticFini(Value *GlobalThreadID);
 
 
 
  /// Create the current source location.
 
  ///
 
  /// TODO: Generates only(!) dummy values.
 
  GlobalVariable *createSourceLocation();
 
};
 
} // end namespace polly
 
#endif