The ParallelLoopGenerator allows to create parallelized loops. More...

#include <LoopGenerators.h>

Inheritance diagram for polly::ParallelLoopGenerator:

Public Member Functions
	ParallelLoopGenerator (PollyIRBuilder &Builder, LoopInfo &LI, DominatorTree &DT, const DataLayout &DL)
	Create a parallel loop generator for the current function.

virtual	~ParallelLoopGenerator ()

Value *	createParallelLoop (Value LB, Value UB, Value Stride, SetVector< Value > &Values, ValueMapT &VMap, BasicBlock::iterator *LoopBody)
	Create a parallel loop.

AllocaInst *	storeValuesIntoStruct (SetVector< Value * > &Values)
	Create a struct for all `Values` and store them in there.

void	extractValuesFromStruct (SetVector< Value * > Values, Type Ty, Value Struct, ValueMapT &VMap)
	Extract all values from the `Struct` and construct the mapping.

Function *	createSubFnDefinition ()
	Create the definition of the parallel subfunction.

virtual void	deployParallelExecution (Function SubFn, Value SubFnParam, Value LB, Value UB, Value *Stride)=0
	Create the runtime library calls for spawn and join of the worker threads.

virtual Function *	prepareSubFnDefinition (Function *F) const =0
	Prepare the definition of the parallel subfunction.

virtual std::tuple< Value , Function >	createSubFn (Value Stride, AllocaInst Struct, SetVector< Value * > UsedValues, ValueMapT &VMap)=0
	Create the parallel subfunction.

Protected Attributes
PollyIRBuilder &	Builder
	The IR builder we use to create instructions.

LoopInfo &	LI
	The loop info of the current function we need to update.

DominatorTree &	DT
	The dominance tree of the current function we need to update.

Type *	LongType
	The type of a "long" on this hardware used for backend calls.

Module *	M
	The current module.

llvm::DebugLoc	DLGenerated
	Debug location for generated code without direct link to any specific line.

Detailed Description

The ParallelLoopGenerator allows to create parallelized loops.

To parallelize a loop, we perform the following steps: o Generate a subfunction which will hold the loop body. o Create a struct to hold all outer values needed in the loop body. o Create calls to a runtime library to achieve the actual parallelism. These calls will spawn and join threads, define how the work (here the iterations) are distributed between them and make sure each has access to the struct holding all needed values.

At the moment we support only one parallel runtime, OpenMP.

If we parallelize the outer loop of the following loop nest,

S0; for (int i = 0; i < N; i++) for (int j = 0; j < M; j++) S1(i, j); S2;

we will generate the following code (with different runtime function names):

S0; auto *values = storeValuesIntoStruct(); // Execute subfunction with multiple threads spawn_threads(subfunction, values); join_threads(); S2;

// This function is executed in parallel by different threads void subfunction(values) { while (auto *WorkItem = getWorkItem()) { int LB = WorkItem.begin(); int UB = WorkItem.end(); for (int i = LB; i < UB; i++) for (int j = 0; j < M; j++) S1(i, j); } cleanup_thread(); }

Definition at line 128 of file LoopGenerators.h.

Constructor & Destructor Documentation

◆ ParallelLoopGenerator()

polly::ParallelLoopGenerator::ParallelLoopGenerator	(	PollyIRBuilder &	Builder,
		LoopInfo &	LI,
		DominatorTree &	DT,
		const DataLayout &	DL
	)

inline

Create a parallel loop generator for the current function.

Definition at line 131 of file LoopGenerators.h.

◆ ~ParallelLoopGenerator()

virtual polly::ParallelLoopGenerator::~ParallelLoopGenerator ( )

inlinevirtual

Definition at line 140 of file LoopGenerators.h.

Member Function Documentation

◆ createParallelLoop()

Value * ParallelLoopGenerator::createParallelLoop	(	Value *	LB,
		Value *	UB,
		Value *	Stride,
		SetVector< Value * > &	Values,
		ValueMapT &	VMap,
		BasicBlock::iterator *	LoopBody
	)

Create a parallel loop.

This function is the main function to automatically generate a parallel loop with all its components.

Parameters

LB	The lower bound for the loop we parallelize.
UB	The upper bound for the loop we parallelize.
Stride	The stride of the loop we parallelize.
Values	A set of LLVM-IR Values that should be available in the new loop body.
VMap	A map to allow outside access to the new versions of the values in `Values`.
LoopBody	A pointer to an iterator that is set to point to the body of the created loop. It should be used to insert instructions that form the actual loop body.

Returns: The newly created induction variable for this loop.

Definition at line 176 of file LoopGenerators.cpp.

References Builder, createSubFn(), deployParallelExecution(), Function, LongType, storeValuesIntoStruct(), and polly::Value.

◆ createSubFn()

virtual std::tuple< Value , Function > polly::ParallelLoopGenerator::createSubFn	(	Value *	Stride,
		AllocaInst *	Struct,
		SetVector< Value * >	UsedValues,
		ValueMapT &	VMap
	)

pure virtual

Create the parallel subfunction.

Parameters

Stride	The induction variable increment.
Struct	A struct holding all values in `Values`.
Values	A set of LLVM-IR Values that should be available in the new loop body.
VMap	A map to allow outside access to the new versions of the values in `Values`.
SubFn	The newly created subfunction is returned here.

Returns: The newly created induction variable.

Implemented in polly::ParallelLoopGeneratorGOMP, and polly::ParallelLoopGeneratorKMP.

Referenced by createParallelLoop().

◆ createSubFnDefinition()

Function * ParallelLoopGenerator::createSubFnDefinition ( )

Create the definition of the parallel subfunction.

Returns: A pointer to the subfunction.

Definition at line 199 of file LoopGenerators.cpp.

References Builder, Function, polly::PollySkipFnAttr, and prepareSubFnDefinition().

Referenced by polly::ParallelLoopGeneratorGOMP::createSubFn(), and polly::ParallelLoopGeneratorKMP::createSubFn().

◆ deployParallelExecution()

virtual void polly::ParallelLoopGenerator::deployParallelExecution	(	Function *	SubFn,
		Value *	SubFnParam,
		Value *	LB,
		Value *	UB,
		Value *	Stride
	)

pure virtual

Create the runtime library calls for spawn and join of the worker threads.

Additionally, places a call to the specified subfunction.

Parameters

SubFn	The subfunction which holds the loop body.
SubFnParam	The parameter for the subfunction (basically the struct filled with the outside values).
LB	The lower bound for the loop we parallelize.
UB	The upper bound for the loop we parallelize.
Stride	The stride of the loop we parallelize.

Implemented in polly::ParallelLoopGeneratorGOMP, and polly::ParallelLoopGeneratorKMP.

Referenced by createParallelLoop().

◆ extractValuesFromStruct()

void ParallelLoopGenerator::extractValuesFromStruct	(	SetVector< Value * >	Values,
		Type *	Ty,
		Value *	Struct,
		ValueMapT &	VMap
	)

Extract all values from the Struct and construct the mapping.

Parameters

Values	The values which were stored in the struct.
Struct	The struct holding all the values in `Values`.
VMap	A map to associate every element of `Values` with the new llvm value loaded from the `Struct`.

Definition at line 242 of file LoopGenerators.cpp.

References Builder, and polly::Value.

Referenced by polly::ParallelLoopGeneratorGOMP::createSubFn(), and polly::ParallelLoopGeneratorKMP::createSubFn().

◆ prepareSubFnDefinition()

virtual Function * polly::ParallelLoopGenerator::prepareSubFnDefinition ( Function * F ) const

pure virtual

Prepare the definition of the parallel subfunction.

Creates the argument list and names them (as well as the subfunction).

Parameters

F	A pointer to the (parallel) subfunction's parent function.

Returns: The pointer to the (parallel) subfunction.

Implemented in polly::ParallelLoopGeneratorGOMP, and polly::ParallelLoopGeneratorKMP.

References Function.

Referenced by createSubFnDefinition().

◆ storeValuesIntoStruct()

AllocaInst * ParallelLoopGenerator::storeValuesIntoStruct ( SetVector< Value * > & Values )

Create a struct for all Values and store them in there.

Parameters

Values The values which should be stored in the struct.

Returns: The created struct.

Definition at line 216 of file LoopGenerators.cpp.

References Builder, and polly::Value.

Referenced by createParallelLoop().

Member Data Documentation

◆ Builder

PollyIRBuilder& polly::ParallelLoopGenerator::Builder

protected

The IR builder we use to create instructions.

Definition at line 165 of file LoopGenerators.h.

◆ DLGenerated

llvm::DebugLoc polly::ParallelLoopGenerator::DLGenerated

protected

Debug location for generated code without direct link to any specific line.

We only set the DebugLoc where the IR Verifier requires us to. Otherwise, absent debug location for optimized code should be fine.

Definition at line 184 of file LoopGenerators.h.

◆ DT

DominatorTree& polly::ParallelLoopGenerator::DT

protected

The dominance tree of the current function we need to update.

Definition at line 171 of file LoopGenerators.h.

Referenced by polly::ParallelLoopGeneratorGOMP::createSubFn(), and polly::ParallelLoopGeneratorKMP::createSubFn().

◆ LI

LoopInfo& polly::ParallelLoopGenerator::LI

protected

The loop info of the current function we need to update.

Definition at line 168 of file LoopGenerators.h.

Referenced by polly::ParallelLoopGeneratorGOMP::createSubFn(), and polly::ParallelLoopGeneratorKMP::createSubFn().

◆ LongType

Type* polly::ParallelLoopGenerator::LongType

protected

The type of a "long" on this hardware used for backend calls.

Definition at line 174 of file LoopGenerators.h.

◆ M

Module* polly::ParallelLoopGenerator::M

protected

The current module.

Definition at line 177 of file LoopGenerators.h.

The documentation for this class was generated from the following files:

Public Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ ParallelLoopGenerator()

◆ ~ParallelLoopGenerator()

Member Function Documentation

◆ createParallelLoop()

◆ createSubFn()

◆ createSubFnDefinition()

◆ deployParallelExecution()

◆ extractValuesFromStruct()

◆ prepareSubFnDefinition()

◆ storeValuesIntoStruct()

Member Data Documentation

◆ Builder

◆ DLGenerated

◆ DT

◆ LI

◆ LongType

◆ M