llvm::OptimizationRemarkEmitter Class Reference

The optimization diagnostic interface. More...

#include "llvm/Analysis/OptimizationRemarkEmitter.h"

Public Member Functions
	OptimizationRemarkEmitter (const Function *F, BlockFrequencyInfo *BFI)
	OptimizationRemarkEmitter (const Function *F)
	This variant can be used to generate ORE on demand (without the analysis pass).
	OptimizationRemarkEmitter (OptimizationRemarkEmitter &&Arg)
OptimizationRemarkEmitter &	operator= (OptimizationRemarkEmitter &&RHS)
bool	invalidate (Function &F, const PreservedAnalyses &PA, FunctionAnalysisManager::Invalidator &Inv)
	Handle invalidation events in the new pass manager.
bool	enabled () const
	Return true iff at least some remarks are enabled.
void	emit (DiagnosticInfoOptimizationBase &OptDiag)
	Output the remark via the diagnostic handler and to the optimization record file.
template<typename T >
void	emit (T RemarkBuilder, decltype(RemarkBuilder()) *=nullptr)
	Take a lambda that returns a remark which will be emitted.
bool	allowExtraAnalysis (StringRef PassName) const
	Whether we allow for extra compile-time budget to perform more analysis to produce fewer false positives.

Static Public Member Functions
static bool	allowExtraAnalysis (const Function &F, StringRef PassName)
static bool	allowExtraAnalysis (LLVMContext &Ctx, StringRef PassName)

Detailed Description

The optimization diagnostic interface.

It allows reporting when optimizations are performed and when they are not along with the reasons for it. Hotness information of the corresponding code region can be included in the remark if DiagnosticsHotnessRequested is enabled in the LLVM context.

Definition at line 32 of file OptimizationRemarkEmitter.h.

Constructor & Destructor Documentation

OptimizationRemarkEmitter() [1/3]

llvm::OptimizationRemarkEmitter::OptimizationRemarkEmitter ( const Function *  F, BlockFrequencyInfo *  BFI  )

inline

Definition at line 34 of file OptimizationRemarkEmitter.h.

References F.

OptimizationRemarkEmitter() [2/3]

OptimizationRemarkEmitter::OptimizationRemarkEmitter

(

const Function *

F

)

This variant can be used to generate ORE on demand (without the analysis pass).

Note that this ctor has a very different cost depending on whether F->getContext().getDiagnosticsHotnessRequested() is on or not. If it's off the operation is free.

Whereas if DiagnosticsHotnessRequested is on, it is fairly expensive operation since BFI and all its required analyses are computed. This is for example useful for CGSCC passes that can't use function analyses passes in the old PM.

Definition at line 27 of file OptimizationRemarkEmitter.cpp.

References llvm::LoopInfoBase< BlockT, LoopT >::analyze(), F, and llvm::DominatorTreeBase< NodeT, IsPostDom >::recalculate().

OptimizationRemarkEmitter() [3/3]

llvm::OptimizationRemarkEmitter::OptimizationRemarkEmitter ( OptimizationRemarkEmitter &&  Arg )

inline

Definition at line 50 of file OptimizationRemarkEmitter.h.

Member Function Documentation

allowExtraAnalysis() [1/3]

static bool llvm::OptimizationRemarkEmitter::allowExtraAnalysis ( const Function &  F, StringRef  PassName  )

inlinestatic

Definition at line 100 of file OptimizationRemarkEmitter.h.

References allowExtraAnalysis(), F, and PassName.

allowExtraAnalysis() [2/3]

static bool llvm::OptimizationRemarkEmitter::allowExtraAnalysis ( LLVMContext &  Ctx, StringRef  PassName  )

inlinestatic

Definition at line 103 of file OptimizationRemarkEmitter.h.

References llvm::LLVMContext::getDiagHandlerPtr(), llvm::LLVMContext::getLLVMRemarkStreamer(), llvm::DiagnosticHandler::isAnyRemarkEnabled(), and PassName.

allowExtraAnalysis() [3/3]

bool llvm::OptimizationRemarkEmitter::allowExtraAnalysis ( StringRef  PassName ) const

inline

Whether we allow for extra compile-time budget to perform more analysis to produce fewer false positives.

This is useful when reporting missed optimizations. In this case we can use the extra analysis (1) to filter trivial false positives or (2) to provide more context so that non-trivial false positives can be quickly detected by the user.

Definition at line 97 of file OptimizationRemarkEmitter.h.

References allowExtraAnalysis(), and PassName.

Referenced by allowExtraAnalysis(), llvm::LoopVectorizationLegality::canVectorize(), convertAnnotation2Metadata(), llvm::LoopVectorizePass::processLoop(), and runImpl().

emit() [1/2]

void OptimizationRemarkEmitter::emit

(

DiagnosticInfoOptimizationBase &

OptDiag

)

Output the remark via the diagnostic handler and to the optimization record file.

Definition at line 79 of file OptimizationRemarkEmitter.cpp.

References llvm::LLVMContext::diagnose(), llvm::Function::getContext(), llvm::LLVMContext::getDiagnosticsHotnessThreshold(), and llvm::DiagnosticInfoOptimizationBase::getHotness().

Referenced by llvm::LoopVectorizeHints::allowVectorization(), llvm::canSinkOrHoistInst(), checkMixedPrecision(), llvm::computeUnrollCount(), createFunctionClones(), deleteLoopIfDead(), DoFlattenLoopPair(), doPromotion(), emit(), EmitAndPrintRemark(), llvm::emitInlinedInto(), llvm::InnerLoopVectorizer::emitMemRuntimeChecks(), emitRemark(), llvm::LoopVectorizeHints::emitRemarkWithHints(), expandToSwitch(), llvm::BasicTTIImplBase< T >::getUnrollingPreferences(), hoist(), isSafeToExecuteUnconditionally(), markTails(), llvm::LoopVectorizePass::processLoop(), llvm::pgo::promoteIndirectCall(), llvm::promoteLoopAccessesToScalars(), llvm::MLInlineAdvice::recordInliningImpl(), llvm::MLInlineAdvice::recordInliningWithCalleeDeletedImpl(), llvm::MLInlineAdvice::recordUnattemptedInliningImpl(), llvm::MLInlineAdvice::recordUnsuccessfulInliningImpl(), reportFastISelFailure(), reportLoadElim(), reportMayClobberedLoad(), llvm::reportVectorization(), llvm::reportVectorizationFailure(), llvm::SSPLayoutAnalysis::requiresStackProtector(), runImpl(), llvm::setProfMetadata(), llvm::SITargetLowering::shouldExpandAtomicRMWInIR(), llvm::shouldInline(), sink(), tryToUnrollLoop(), llvm::UnrollAndJamLoop(), llvm::UnrollLoop(), verifyFuncBFI(), llvm::memtag::StackInfoBuilder::visit(), and warnAboutLeftoverTransformations().

emit() [2/2]

template<typename T >

void llvm::OptimizationRemarkEmitter::emit ( T  RemarkBuilder, decltype(RemarkBuilder()) *  = nullptr  )

inline

Take a lambda that returns a remark which will be emitted.

Second argument is only used to restrict this to functions.

Definition at line 76 of file OptimizationRemarkEmitter.h.

References emit(), enabled(), and value.

enabled()

bool llvm::OptimizationRemarkEmitter::enabled ( ) const

inline

Return true iff at least some remarks are enabled.

Definition at line 64 of file OptimizationRemarkEmitter.h.

References llvm::Function::getContext(), llvm::LLVMContext::getDiagHandlerPtr(), llvm::LLVMContext::getLLVMRemarkStreamer(), and llvm::DiagnosticHandler::isAnyRemarkEnabled().

Referenced by emit().

invalidate()

bool OptimizationRemarkEmitter::invalidate	(	Function &	F,
		const PreservedAnalyses &	PA,
		FunctionAnalysisManager::Invalidator &	Inv
	)

Handle invalidation events in the new pass manager.

Definition at line 48 of file OptimizationRemarkEmitter.cpp.

References F, and llvm::AnalysisManager< IRUnitT, ExtraArgTs >::Invalidator::invalidate().

operator=()

OptimizationRemarkEmitter & llvm::OptimizationRemarkEmitter::operator= ( OptimizationRemarkEmitter &&  RHS )

inline

Definition at line 53 of file OptimizationRemarkEmitter.h.

References RHS.

---

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

• include/llvm/Analysis/OptimizationRemarkEmitter.h
• lib/Analysis/OptimizationRemarkEmitter.cpp