#include <DependenceAnalysis.h>

Inheritance diagram for llvm::DependenceAnalysis:

Collaboration diagram for llvm::DependenceAnalysis:

Classes
struct	BoundInfo
struct	CoefficientInfo
class	Constraint
struct	Subscript
Public Member Functions
std::unique_ptr< Dependence >	depends (Instruction Src, Instruction Dst, bool PossiblyLoopIndependent)
const SCEV *	getSplitIteration (const Dependence &Dep, unsigned Level)
	DependenceAnalysis ()
bool	runOnFunction (Function &F) override
void	releaseMemory () override
void	getAnalysisUsage (AnalysisUsage &) const override
void	print (raw_ostream &, const Module *=nullptr) const override
Static Public Attributes
static char	ID = 0

Detailed Description

DependenceAnalysis - This class is the main dependence-analysis driver.

Definition at line 280 of file DependenceAnalysis.h.

Constructor & Destructor Documentation

llvm::DependenceAnalysis::DependenceAnalysis ( ) [inline]

Definition at line 918 of file DependenceAnalysis.h.

References llvm::PassRegistry::getPassRegistry(), and llvm::initializeDependenceAnalysisPass().

Member Function Documentation

std::unique_ptr< Dependence > DependenceAnalysis::depends	(	Instruction *	Src,
		Instruction *	Dst,
		bool	PossiblyLoopIndependent
	)

depends - Tests for a dependence between the Src and Dst instructions. Returns NULL if no dependence; otherwise, returns a Dependence (or a FullDependence) with as much information as can be gleaned. The flag PossiblyLoopIndependent should be set by the caller if it appears that control flow can reach from Src to Dst without traversing a loop back edge.

Definition at line 3280 of file DependenceAnalysis.cpp.

References llvm::SmallBitVector::any(), llvm::SmallBitVector::count(), llvm::dbgs(), DEBUG, Delinearize, llvm::Dependence::DVEntry::Direction, dumpSmallBitVector(), llvm::dyn_cast(), EQ, llvm::SmallBitVector::find_first(), llvm::SmallBitVector::find_next(), llvm::FullDependence::getDirection(), llvm::ScalarEvolution::getElementSize(), llvm::Instruction::getParent(), llvm::GEPOperator::getPointerOperand(), getPointerOperand(), llvm::GEPOperator::getPointerOperandType(), llvm::ScalarEvolution::getSCEV(), llvm::GEPOperator::idx_begin(), llvm::GEPOperator::idx_end(), isLoadOrStore(), isLoopInvariant(), LI, llvm_unreachable, llvm::AliasAnalysis::MayAlias, llvm::Instruction::mayReadFromMemory(), llvm::Instruction::mayWriteToMemory(), llvm::AliasAnalysis::MustAlias, llvm::AliasAnalysis::NoAlias, llvm::Dependence::DVEntry::NONE, P, llvm::AliasAnalysis::PartialAlias, llvm::SmallBitVector::reset(), llvm::SmallVectorImpl< T >::resize(), llvm::Dependence::DVEntry::Scalar, llvm::SmallBitVector::set(), llvm::SmallVectorTemplateCommon< T >::size(), and underlyingObjectsAlias().

Referenced by dumpExampleDependence().

void DependenceAnalysis::getAnalysisUsage ( AnalysisUsage & ) const [override, virtual]

getAnalysisUsage - This function should be overriden by passes that need analysis information to do their job. If a pass specifies that it uses a particular analysis result to this function, it can then use the getAnalysis<AnalysisType>() function, below.

Reimplemented from llvm::Pass.

Definition at line 144 of file DependenceAnalysis.cpp.

References llvm::AnalysisUsage::addRequiredTransitive(), and llvm::AnalysisUsage::setPreservesAll().

const SCEV * DependenceAnalysis::getSplitIteration	(	const Dependence &	Dep,
		unsigned	Level
	)

getSplitIteration - Give a dependence that's splittable at some particular level, return the iteration that should be used to split the loop.

Generally, the dependence analyzer will be used to build a dependence graph for a function (basically a map from instructions to dependences). Looking for cycles in the graph shows us loops that cannot be trivially vectorized/parallelized.

We can try to improve the situation by examining all the dependences that make up the cycle, looking for ones we can break. Sometimes, peeling the first or last iteration of a loop will break dependences, and there are flags for those possibilities. Sometimes, splitting a loop at some other iteration will do the trick, and we've got a flag for that case. Rather than waste the space to record the exact iteration (since we rarely know), we provide a method that calculates the iteration. It's a drag that it must work from scratch, but wonderful in that it's possible.

Here's an example:

for (i = 0; i < 10; i++) A[i] = ... ... = A[11 - i]

There's a loop-carried flow dependence from the store to the load, found by the weak-crossing SIV test. The dependence will have a flag, indicating that the dependence can be broken by splitting the loop. Calling getSplitIteration will return 5. Splitting the loop breaks the dependence, like so:

for (i = 0; i <= 5; i++) A[i] = ... ... = A[11 - i] for (i = 6; i < 10; i++) A[i] = ... ... = A[11 - i]

breaks the dependence and allows us to vectorize/parallelize both loops.

Definition at line 3731 of file DependenceAnalysis.cpp.

References llvm::SmallBitVector::any(), llvm::SmallBitVector::count(), llvm::dbgs(), DEBUG, Delinearize, llvm::dyn_cast(), llvm::SmallBitVector::find_first(), llvm::SmallBitVector::find_next(), llvm::Dependence::getDst(), llvm::ScalarEvolution::getElementSize(), llvm::Instruction::getParent(), llvm::GEPOperator::getPointerOperand(), getPointerOperand(), llvm::GEPOperator::getPointerOperandType(), llvm::ScalarEvolution::getSCEV(), llvm::Dependence::getSrc(), llvm::GEPOperator::idx_begin(), llvm::GEPOperator::idx_end(), isLoadOrStore(), isLoopInvariant(), llvm::Dependence::isSplitable(), LI, llvm_unreachable, llvm::Instruction::mayReadFromMemory(), llvm::Instruction::mayWriteToMemory(), llvm::AliasAnalysis::MustAlias, P, llvm::SmallBitVector::reset(), llvm::SmallVectorImpl< T >::resize(), llvm::SmallBitVector::set(), llvm::SmallVectorTemplateCommon< T >::size(), and underlyingObjectsAlias().

Referenced by dumpExampleDependence().

void DependenceAnalysis::print	(	raw_ostream &	O,
		const Module *	M = `nullptr`
	)		const `[override, virtual]`

print - Print out the internal state of the pass. This is called by Analyze to print out the contents of an analysis. Otherwise it is not necessary to implement this method. Beware that the module pointer MAY be null. This automatically forwards to a virtual function that does not provide the Module* in case the analysis doesn't need it it can just be ignored.

Reimplemented from llvm::Pass.

Definition at line 185 of file DependenceAnalysis.cpp.

References dumpExampleDependence(), and F().

void DependenceAnalysis::releaseMemory ( ) [override, virtual]

releaseMemory() - This member can be implemented by a pass if it wants to be able to release its memory when it is no longer needed. The default behavior of passes is to hold onto memory for the entire duration of their lifetime (which is the entire compile time). For pipelined passes, this is not a big deal because that memory gets recycled every time the pass is invoked on another program unit. For IP passes, it is more important to free memory when it is unused.

Optionally implement this function to release pass memory when it is no longer used.

Reimplemented from llvm::Pass.

Definition at line 140 of file DependenceAnalysis.cpp.

bool DependenceAnalysis::runOnFunction ( Function & F ) [override, virtual]