AMDGPULibCalls.cpp File Reference

#include "AMDGPU.h"
#include "AMDGPULibFunc.h"
#include "GCNSubtarget.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/InitializePasses.h"
#include "llvm/Target/TargetMachine.h"

Include dependency graph for AMDGPULibCalls.cpp:

Go to the source code of this file.

Classes
class	llvm::AMDGPULibCalls
struct	TableRef

Namespaces
	llvm

Macros
#define	DEBUG_TYPE   "amdgpu-simplifylib"
#define	MATH_PI   numbers::pi
#define	MATH_E   numbers::e
#define	MATH_SQRT2   numbers::sqrt2
#define	MATH_SQRT1_2   numbers::inv_sqrt2

Functions
	INITIALIZE_PASS_BEGIN (AMDGPUSimplifyLibCalls, "amdgpu-simplifylib", "Simplify well-known AMD library calls", false, false) INITIALIZE_PASS_END(AMDGPUSimplifyLibCalls
amdgpu Simplify well known AMD library false	INITIALIZE_PASS (AMDGPUUseNativeCalls, "amdgpu-usenative", "Replace builtin math calls with that native versions.", false, false) template< typename IRB > static CallInst *CreateCallEx(IRB &B
static bool	HasNative (AMDGPULibFunc::EFuncId id)
static TableRef	getOptTable (AMDGPULibFunc::EFuncId id)
static int	getVecSize (const AMDGPULibFunc &FInfo)
static AMDGPULibFunc::EType	getArgType (const AMDGPULibFunc &FInfo)
static double	llvm::log2 (double V)

Variables
static cl::opt< bool >	EnablePreLink ("amdgpu-prelink", cl::desc("Enable pre-link mode optimizations"), cl::init(false), cl::Hidden)
static cl::list< std::string >	UseNative ("amdgpu-use-native", cl::desc("Comma separated list of functions to replace with native, or all"), cl::CommaSeparated, cl::ValueOptional, cl::Hidden)
amdgpu	simplifylib
amdgpu Simplify well known AMD library	calls
amdgpu Simplify well known AMD library	false
amdgpu Simplify well known AMD library false FunctionCallee	Callee
amdgpu Simplify well known AMD library false FunctionCallee Value *	Arg
amdgpu Simplify well known AMD library false FunctionCallee Value const Twine &	Name
static const TableEntry	tbl_acos []
static const TableEntry	tbl_acosh []
static const TableEntry	tbl_acospi []
static const TableEntry	tbl_asin []
static const TableEntry	tbl_asinh []
static const TableEntry	tbl_asinpi []
static const TableEntry	tbl_atan []
static const TableEntry	tbl_atanh []
static const TableEntry	tbl_atanpi []
static const TableEntry	tbl_cbrt []
static const TableEntry	tbl_cos []
static const TableEntry	tbl_cosh []
static const TableEntry	tbl_cospi []
static const TableEntry	tbl_erfc []
static const TableEntry	tbl_erf []
static const TableEntry	tbl_exp []
static const TableEntry	tbl_exp2 []
static const TableEntry	tbl_exp10 []
static const TableEntry	tbl_expm1 []
static const TableEntry	tbl_log []
static const TableEntry	tbl_log2 []
static const TableEntry	tbl_log10 []
static const TableEntry	tbl_rsqrt []
static const TableEntry	tbl_sin []
static const TableEntry	tbl_sinh []
static const TableEntry	tbl_sinpi []
static const TableEntry	tbl_sqrt []
static const TableEntry	tbl_tan []
static const TableEntry	tbl_tanh []
static const TableEntry	tbl_tanpi []
static const TableEntry	tbl_tgamma []

Detailed Description

This file does AMD library function optimizations.

Definition in file AMDGPULibCalls.cpp.

Macro Definition Documentation

DEBUG_TYPE

#define DEBUG_TYPE   "amdgpu-simplifylib"

Definition at line 23 of file AMDGPULibCalls.cpp.

MATH_E

#define MATH_E   numbers::e

Definition at line 38 of file AMDGPULibCalls.cpp.

MATH_PI

#define MATH_PI   numbers::pi

Definition at line 37 of file AMDGPULibCalls.cpp.

MATH_SQRT1_2

#define MATH_SQRT1_2   numbers::inv_sqrt2

Definition at line 40 of file AMDGPULibCalls.cpp.

MATH_SQRT2

#define MATH_SQRT2   numbers::sqrt2

Definition at line 39 of file AMDGPULibCalls.cpp.

Function Documentation

getArgType()

static AMDGPULibFunc::EType getArgType ( const AMDGPULibFunc &  FInfo )

inlinestatic

Definition at line 456 of file AMDGPULibCalls.cpp.

References llvm::AMDGPULibFuncBase::Param::ArgType, and llvm::AMDGPULibFunc::getLeads().

Referenced by llvm::AMDGPULibCalls::fold(), and llvm::AMDGPULibCalls::useNative().

getOptTable()

static TableRef getOptTable ( AMDGPULibFunc::EFuncId  id )

static

Definition at line 408 of file AMDGPULibCalls.cpp.

References llvm::AMDGPULibFuncBase::EI_ACOS, llvm::AMDGPULibFuncBase::EI_ACOSH, llvm::AMDGPULibFuncBase::EI_ACOSPI, llvm::AMDGPULibFuncBase::EI_ASIN, llvm::AMDGPULibFuncBase::EI_ASINH, llvm::AMDGPULibFuncBase::EI_ASINPI, llvm::AMDGPULibFuncBase::EI_ATAN, llvm::AMDGPULibFuncBase::EI_ATANH, llvm::AMDGPULibFuncBase::EI_ATANPI, llvm::AMDGPULibFuncBase::EI_CBRT, llvm::AMDGPULibFuncBase::EI_COS, llvm::AMDGPULibFuncBase::EI_COSH, llvm::AMDGPULibFuncBase::EI_COSPI, llvm::AMDGPULibFuncBase::EI_ERF, llvm::AMDGPULibFuncBase::EI_ERFC, llvm::AMDGPULibFuncBase::EI_EXP, llvm::AMDGPULibFuncBase::EI_EXP10, llvm::AMDGPULibFuncBase::EI_EXP2, llvm::AMDGPULibFuncBase::EI_EXPM1, llvm::AMDGPULibFuncBase::EI_LOG, llvm::AMDGPULibFuncBase::EI_LOG10, llvm::AMDGPULibFuncBase::EI_LOG2, llvm::AMDGPULibFuncBase::EI_NCOS, llvm::AMDGPULibFuncBase::EI_NEXP2, llvm::AMDGPULibFuncBase::EI_NLOG2, llvm::AMDGPULibFuncBase::EI_NRSQRT, llvm::AMDGPULibFuncBase::EI_NSIN, llvm::AMDGPULibFuncBase::EI_NSQRT, llvm::AMDGPULibFuncBase::EI_RSQRT, llvm::AMDGPULibFuncBase::EI_SIN, llvm::AMDGPULibFuncBase::EI_SINH, llvm::AMDGPULibFuncBase::EI_SINPI, llvm::AMDGPULibFuncBase::EI_SQRT, llvm::AMDGPULibFuncBase::EI_TAN, llvm::AMDGPULibFuncBase::EI_TANH, llvm::AMDGPULibFuncBase::EI_TANPI, llvm::AMDGPULibFuncBase::EI_TGAMMA, tbl_acos, tbl_acosh, tbl_acospi, tbl_asin, tbl_asinh, tbl_asinpi, tbl_atan, tbl_atanh, tbl_atanpi, tbl_cbrt, tbl_cos, tbl_cosh, tbl_cospi, tbl_erf, tbl_erfc, tbl_exp, tbl_exp10, tbl_exp2, tbl_expm1, tbl_log, tbl_log10, tbl_log2, tbl_rsqrt, tbl_sin, tbl_sinh, tbl_sinpi, tbl_sqrt, tbl_tan, tbl_tanh, tbl_tanpi, and tbl_tgamma.

getVecSize()

static int getVecSize ( const AMDGPULibFunc &  FInfo )

inlinestatic

Definition at line 452 of file AMDGPULibCalls.cpp.

References llvm::AMDGPULibFunc::getLeads(), and llvm::AMDGPULibFuncBase::Param::VectorSize.

Referenced by llvm::AMDGPULibCalls::fold().

HasNative()

static bool HasNative ( AMDGPULibFunc::EFuncId  id )

static

Definition at line 375 of file AMDGPULibCalls.cpp.

References llvm::AMDGPULibFuncBase::EI_COS, llvm::AMDGPULibFuncBase::EI_DIVIDE, llvm::AMDGPULibFuncBase::EI_EXP, llvm::AMDGPULibFuncBase::EI_EXP10, llvm::AMDGPULibFuncBase::EI_EXP2, llvm::AMDGPULibFuncBase::EI_LOG, llvm::AMDGPULibFuncBase::EI_LOG10, llvm::AMDGPULibFuncBase::EI_LOG2, llvm::AMDGPULibFuncBase::EI_POWR, llvm::AMDGPULibFuncBase::EI_RECIP, llvm::AMDGPULibFuncBase::EI_RSQRT, llvm::AMDGPULibFuncBase::EI_SIN, llvm::AMDGPULibFuncBase::EI_SINCOS, llvm::AMDGPULibFuncBase::EI_SQRT, and llvm::AMDGPULibFuncBase::EI_TAN.

Referenced by llvm::AMDGPULibCalls::useNative().

INITIALIZE_PASS()

amdgpu Simplify well known AMD library false INITIALIZE_PASS	(	AMDGPUUseNativeCalls	,
		"amdgpu-usenative"	,
		"Replace builtin math calls with that native versions."	,
		false	,
		false
	)		&

INITIALIZE_PASS_BEGIN()

INITIALIZE_PASS_BEGIN	(	AMDGPUSimplifyLibCalls	,
		"amdgpu-simplifylib"	,
		"Simplify well-known AMD library calls"	,
		false	,
		false
	)

Variable Documentation

Arg

amdgpu Simplify well known AMD library false FunctionCallee Value* Arg

Definition at line 205 of file AMDGPULibCalls.cpp.

Referenced by llvm::AAEvaluator::AAEvaluator(), llvm::opt::Option::accept(), llvm::AliasSetTracker::add(), AddAliasScopeMetadata(), AddAlignmentAssumptions(), addArgumentReturnedAttrs(), addConditions(), llvm::AnalysisUsage::addPreserved(), addVCallToSet(), llvm::AMDGPUAAResult::alias(), allCallersPassValidPointerForArgument(), allocateSGPR32Input(), allocateSGPR64Input(), llvm::SITargetLowering::allocateSpecialInputVGPRs(), allocateVGPR32Input(), llvm::CCState::AnalyzeArgumentsSecondPass(), llvm::AMDGPUTargetLowering::analyzeFormalArgumentsCompute(), appendArg(), appendString(), llvm::BitVector::apply(), llvm::xray::CallArgRecord::arg(), llvm::MachineFunction::ArgRegPair::ArgRegPair(), llvm::IRPosition::argument(), llvm::CallLowering::ValueHandler::assignValueToAddress(), llvm::BTFDebug::beginFunctionImpl(), CalculateTailCallArgDest(), llvm::orc::shared::ParallelCallGroup::call(), llvm::CallGraph::CallGraph(), llvm::orc::remote::OrcRemoteTargetClient::callIntInt(), checkFunctionMemoryAccess(), llvm::CoroIdInst::clearPromise(), combineFneg(), llvm::computeConstantRange(), llvm::computeExpressionSize(), computeKnownBitsFromAssume(), llvm::computeSignatureVTs(), llvm::DwarfUnit::constructSubprogramArguments(), convertArgumentInfo(), llvm::ArgDescriptor::createArg(), llvm::IRBuilderBase::CreateIsNotNull(), llvm::IRBuilderBase::CreateIsNull(), llvm::Attributor::createShallowWrapper(), llvm::orc::LocalCXXRuntimeOverridesBase::CXAAtExitOverride(), llvm::objcarc::Depends(), llvm::DivergenceInfo::DivergenceInfo(), llvm::dlltoolDriverMain(), doList(), llvm::DominatorTreeBase< BasicBlock, IsPostDom >::DominatorTreeBase(), eliminateSwiftError(), eliminateSwiftErrorArgument(), llvm::AMDGPU::HSAMD::MetadataStreamerV3::emitKernelArg(), llvm::AMDGPU::HSAMD::MetadataStreamerV3::emitKernelArgs(), llvm::remarks::BitstreamRemarkSerializerHelper::emitRemarkBlock(), ExpandResponseFile(), findArgumentCopyElisionCandidates(), findDependencies(), FindPredecessorAutoreleaseWithSafePath(), FindPredecessorRetainWithSafePath(), llvm::objcarc::findSingleDependency(), FindSingleUseIdentifiedObject(), fitArgInto64Bits(), llvm::VarDefInit::Fold(), llvm::FoldingSetBase::FoldingSetBase(), llvm::InstCombinerImpl::foldIntegerTypedPHI(), ForeachDagApply(), llvm::detail::frexp(), llvm::GlobalsAAResult::FunctionInfo::FunctionInfo(), llvm::DagInit::get(), llvm::DILocalVariable::getArg(), llvm::remarks::Remark::getArgsAsMsg(), getArgumentTypeAlign(), llvm::IRPosition::getAssociatedFunction(), llvm::VarDefInit::getAsString(), llvm::AbstractCallSite::getCallArgOperand(), llvm::AbstractCallSite::getCallArgOperandNo(), llvm::CoroSuspendInst::getCoroSave(), llvm::IRPosition::getCtxI(), llvm::AMDGPUSubtarget::getExplicitKernArgSize(), llvm::MemoryLocation::getForArgument(), llvm::cflaa::getGlobalOrArgAttrFromValue(), llvm::DenseMapInfo< PointerEmbeddedInt< IntT, Bits > >::getHashValue(), llvm::DenseMapInfo< MemoryLocOrCall >::getHashValue(), llvm::MDNodeKeyImpl< DILocalVariable >::getHashValue(), getInterfaceValue(), llvm::DiagnosticInfoOptimizationBase::getMsg(), llvm::X86TargetLowering::getNegatedExpression(), llvm::PassRegistry::getPassInfo(), llvm::SIMachineFunctionInfo::getPreloadedReg(), llvm::CoroIdInst::getPromise(), getSearchPaths(), llvm::ARM_AM::getSOImmVal(), llvm::CoroIdAsyncInst::getStorageArgumentIndex(), llvm::CoroSuspendAsyncInst::getStorageArgumentIndex(), llvm::ARM_AM::getT2SOImmVal(), llvm::getTotalArgumentsSizeInBytes(), getVectorCallCosts(), llvm::GlobalsAAResult::GlobalsAAResult(), HandleByValArgument(), handlePhiDef(), HandlePrefixedOrGroupedOption(), llvm::detail::hash_value(), llvm::hash_value(), HasSafePathToPredecessorCall(), llvm::AArch64RegisterInfo::hasSVEArgsOrReturn(), llvm::HexagonEvaluator::HexagonEvaluator(), llvm::Attributor::identifyDefaultAbstractAttributes(), llvm::detail::ilogb(), initSlots2Values(), llvm::yaml::CustomMappingTraits< std::map< std::vector< uint64_t >, WholeProgramDevirtResolution::ByArg > >::inputOne(), llvm::SIInstrInfo::insertNoops(), insertNoopsInBundle(), insertParsePoints(), insertSinCosCall(), insertSpills(), llvm::ARMTTIImpl::instCombineIntrinsic(), llvm::X86TTIImpl::instCombineIntrinsic(), llvm::remarks::StringTable::internalize(), internalizeFunction(), llvm::PreservedAnalyses::intersect(), llvm::invertCondition(), llvm::IRSimilarity::IRSimilarityCandidate::IRSimilarityCandidate(), isCString(), llvm::SITargetLowering::isEligibleForTailCallOptimization(), isIntExtFree(), llvm::InformationCache::isInvolvedInMustTailCall(), llvm::MDNodeKeyImpl< DILocalVariable >::isKeyOf(), isKnownNonNullFromDominatingCondition(), llvm::DILocalVariable::isParameter(), llvm::objcarc::IsPotentialRetainableObjPtr(), isPredicatedOnPHI(), isSafeToPromoteArgument(), llvm::NVPTXTTIImpl::isSourceOfDivergence(), llvm::Value::isSwiftError(), isSwiftError(), llvm::AMDGPUInstrInfo::isUniformMMO(), llvm::Attributor::isValidFunctionSignatureRewrite(), llvm::LazyValueInfo::LazyValueInfo(), LLVMGetNextParam(), LLVMGetPreviousParam(), LLVMRemarkArgGetDebugLoc(), LLVMRemarkArgGetKey(), LLVMRemarkArgGetValue(), LLVMSetParamAlignment(), llvm::AMDGPULegalizerInfo::loadInputValue(), llvm::AMDGPUTargetLowering::loadInputValue(), LookupNearestOption(), llvm::Pass::lookupPassInfo(), llvm::LoopInfoBase< BasicBlock, Loop >::LoopInfoBase(), lowerBALLOTIntrinsic(), llvm::ARMCallLowering::lowerCall(), llvm::MipsCallLowering::lowerCall(), llvm::VETargetLowering::LowerCall(), llvm::HexagonTargetLowering::LowerCall(), llvm::SITargetLowering::LowerCall(), llvm::RISCVTargetLowering::LowerCall(), llvm::CallLowering::lowerCall(), llvm::SparcTargetLowering::LowerCall_32(), llvm::SparcTargetLowering::LowerCall_64(), llvm::FastISel::lowerCallTo(), llvm::SparcTargetLowering::LowerF128_LibCallArg(), llvm::X86CallLowering::lowerFormalArguments(), llvm::ARMCallLowering::lowerFormalArguments(), llvm::R600TargetLowering::LowerFormalArguments(), llvm::AArch64CallLowering::lowerFormalArguments(), llvm::AMDGPUCallLowering::lowerFormalArguments(), llvm::MipsCallLowering::lowerFormalArguments(), llvm::VETargetLowering::LowerFormalArguments(), llvm::SITargetLowering::LowerFormalArguments(), llvm::NVPTXTargetLowering::LowerFormalArguments(), llvm::SparcTargetLowering::LowerFormalArguments_32(), llvm::SparcTargetLowering::LowerFormalArguments_64(), llvm::AMDGPUCallLowering::lowerFormalArgumentsKernel(), LowerFSINCOS(), LowerMemOpCallTo(), llvm::SITargetLowering::LowerReturn(), llvm::SparcTargetLowering::LowerReturn_32(), lowerStatepointMetaArgs(), makeStatepointExplicitImpl(), llvm::coro::LowererBase::makeSubFnCall(), llvm::X86TargetLowering::markLibCallAttributes(), markTails(), MatchingStackOffset(), needsFPFromSig(), onlySingleValue(), llvm::GVNExpression::BasicExpression::op_push_back(), std::hash< llvm::pdb::PDB_SymType >::operator()(), llvm::orc::shared::detail::ReadArgs< ArgT, ArgTs... >::operator()(), llvm::operator<<(), llvm::detail::value_sequence_iterator< ValueT >::operator=(), llvm::LazyValueInfo::operator=(), llvm::MIBundleOperandIteratorBase< const MachineOperand >::operator==(), llvm::MCRegisterInfo::mc_difflist_iterator< mc_subreg_iterator >::operator==(), llvm::BasicBlock::phi_iterator_impl< PHINodeT, BBIteratorT >::operator==(), llvm::LazyCallGraph::postorder_ref_scc_iterator::operator==(), llvm::BlotMapVector< KeyT, ValueT >::operator[](), llvm::yaml::CustomMappingTraits< std::map< std::vector< uint64_t >, WholeProgramDevirtResolution::ByArg > >::output(), llvm::cflaa::parentFunctionOfValue(), llvm::remarks::HotnessThresholdParser::parse(), llvm::cl::parser< const PassInfo * >::parse(), llvm::cl::parser< std::string >::parse(), llvm::cl::parser< char >::parse(), llvm::parseCommandLineOptions(), parseDouble(), AbstractManglingParser< ManglingParser< Alloc >, Alloc >::parseEncoding(), AbstractManglingParser< ManglingParser< Alloc >, Alloc >::parseExpr(), ParseFunctionArgs(), llvm::remarks::parseHotnessThresholdOption(), llvm::GCNTargetMachine::parseMachineFunctionInfo(), llvm::opt::OptTable::ParseOneArg(), llvm::remarks::YAMLRemarkParser::parseRemark(), AbstractManglingParser< ManglingParser< Alloc >, Alloc >::parseSpecialName(), AbstractManglingParser< ManglingParser< Alloc >, Alloc >::parseTemplateArg(), AbstractManglingParser< ManglingParser< Alloc >, Alloc >::parseTemplateArgs(), passingValueIsAlwaysUndefined(), Passv64i1ArgInRegs(), llvm::R600TargetLowering::PerformDAGCombine(), llvm::AMDGPUAAResult::pointsToConstantMemory(), populateExternalRelations(), llvm::LegacyDivergenceAnalysis::print(), llvm::sys::printArg(), processArg(), processDbgDeclares(), processPSInputArgs(), ProfileDagInit(), programUndefinedIfUndefOrPoison(), llvm::promoteCall(), llvm::cl::ProvidePositionalOption(), llvm::Attributor::registerFunctionSignatureRewrite(), llvm::DIExpression::replaceArg(), replaceArgumentUses(), replaceConstants(), replaceSwiftErrorOps(), replaceWithCallToVeclib(), llvm::VarDefInit::resolveReferences(), llvm::DagInit::resolveReferences(), llvm::LoopAnalysisManagerFunctionProxy::Result::Result(), llvm::InnerAnalysisManagerProxy< AnalysisManagerT, IRUnitT, ExtraArgTs >::Result::Result(), llvm::DivergenceAnalysisPrinterPass::run(), llvm::RequireAnalysisPass< AnalysisT, IRUnitT, AnalysisManagerT, ExtraArgTs >::run(), llvm::orc::runAsMain(), llvm::runFuzzerOnInputs(), llvm::runIPSCCP(), llvm::IRTranslator::runOnMachineFunction(), llvm::coro::salvageDebugInfo(), llvm::ScalarEvolution::ScalarEvolution(), llvm::detail::scalbn(), llvm::FastISel::selectIntrinsicCall(), llvm::CallLowering::setArgFlags(), llvm::SIMachineFunctionInfo::setWorkItemIDX(), llvm::SIMachineFunctionInfo::setWorkItemIDY(), llvm::SIMachineFunctionInfo::setWorkItemIDZ(), llvm::X86TTIImpl::simplifyDemandedUseBitsIntrinsic(), simplifyInvariantGroupIntrinsic(), simplifyX86movmsk(), StoreTailCallArgumentsToStackSlot(), llvm::SubsumingPositionIterator::SubsumingPositionIterator(), tryConstantFoldCall(), tryToElideArgumentCopy(), llvm::UpgradeARCRuntime(), llvm::IRPosition::value(), llvm::ObjectSizeOffsetVisitor::visitCallBase(), llvm::InstCombinerImpl::visitCallInst(), llvm::InnerLoopVectorizer::widenCallInstruction(), llvm::wouldInstructionBeTriviallyDead(), and writeFunctionTypeMetadataRecords().

Callee

amdgpu Simplify well known AMD library false FunctionCallee Callee

Definition at line 205 of file AMDGPULibCalls.cpp.

Referenced by llvm::AbstractCallSite::AbstractCallSite(), AddCalls(), addCallToCallGraph(), addNoRecurseAttrs(), adjustCallerSSPLevel(), adjustCallerStackProbes(), adjustCallerStackProbeSize(), adjustMinLegalVectorWidth(), adjustNullPointerValidAttr(), allCallersPassValidPointerForArgument(), analyzeLoopUnrollCost(), llvm::ArgumentPromotionPass::areFunctionArgsABICompatible(), llvm::X86TTIImpl::areFunctionArgsABICompatible(), llvm::TargetTransformInfoImplBase::areFunctionArgsABICompatible(), llvm::TargetTransformInfo::areFunctionArgsABICompatible(), llvm::GCNTTIImpl::areInlineCompatible(), llvm::TargetTransformInfoImplBase::areInlineCompatible(), llvm::AttributeFuncs::areInlineCompatible(), llvm::TargetTransformInfo::areInlineCompatible(), llvm::AVRDAGToDAGISel::select< AVRISD::CALL >(), buildCallOperands(), CallExpr::CallExpr(), callsShareTOCBase(), computeFunctionSummary(), copyFeaturesToFunction(), llvm::IRBuilderBase::CreateCall(), llvm::IRBuilderBase::CreateCallBr(), llvm::objcarc::createCallInstWithColors(), llvm::IRBuilderBase::CreateConstrainedFPCall(), llvm::IRBuilderBase::CreateInvoke(), DeleteBasicBlock(), doCallSiteSplitting(), emitBinaryFloatFnCallHelper(), EmitCall(), emitDirectiveRelocJalr(), llvm::emitInlinedInto(), emitLibCall(), emitUnaryFloatFnCallHelper(), findInitTrampoline(), llvm::sampleprof::FunctionSamples::findInlinedFunctions(), findUses(), llvm::AMDGPULibCalls::fold(), for(), llvm::FunctionCallee::FunctionCallee(), functionsHaveCompatibleAttributes(), llvm::ReplayInlineAdvisor::getAdviceImpl(), getAllocationData(), getAllocationDataForFunction(), getAllocationSize(), llvm::IRPosition::getAssociatedArgument(), llvm::getAttributeBasedInliningDecision(), llvm::AbstractCallSite::getCallbackUses(), getCalledFunction(), getCallOpcode(), getEdgeKind(), llvm::FunctionPropertiesInfo::getFunctionPropertiesInfo(), llvm::getInlineCost(), llvm::InlineAdvisor::getMandatoryKind(), llvm::AArch64InstrInfo::getOutliningType(), llvm::ARMBaseInstrInfo::getOutliningType(), llvm::MipsCCState::getSpecialCallingConvForCallee(), handleAttr(), hasReturnsTwiceAttr(), llvm::Attributor::identifyDefaultAbstractAttributes(), INITIALIZE_PASS(), llvm::InformationCache::initializeModuleSlice(), inlineCallIfPossible(), insertSinCosCall(), InstrBreaksNoFree(), InstrBreaksNonThrowing(), InstrBreaksNoSync(), instructionDoesNotReturn(), isCalleeLoad(), llvm::CodeExtractor::isEligible(), llvm::HexagonTargetLowering::IsEligibleForTailCallOptimization(), isEqual(), llvm::isFreeCall(), isFunctionGlobalAddress(), isIndirectCall(), llvm::isInlineViable(), llvm::isLegalToPromote(), isReportingError(), isReturnNonNull(), llvm::isSafeToSpeculativelyExecute(), llvm::VETargetLowering::LowerCall(), llvm::HexagonTargetLowering::LowerCall(), llvm::SITargetLowering::LowerCall(), llvm::RISCVTargetLowering::LowerCall(), llvm::NVPTXTargetLowering::LowerCall(), llvm::SystemZTargetLowering::LowerCall(), llvm::SparcTargetLowering::LowerCall_32(), llvm::SparcTargetLowering::LowerCall_64(), llvm::SelectionDAGBuilder::LowerCallSiteWithDeoptBundle(), llvm::SelectionDAGBuilder::LowerCallSiteWithDeoptBundleImpl(), llvm::SelectionDAGBuilder::LowerDeoptimizeCall(), llvm::SparcTargetLowering::LowerF128Compare(), llvm::SparcTargetLowering::LowerF128Op(), llvm::SparcTargetLowering::LowerGlobalTLSAddress(), llvm::IntrinsicLowering::LowerIntrinsicCall(), llvm::SelectionDAGBuilder::LowerStatepoint(), llvm::AMDGPUTargetLowering::lowerUnhandledCall(), markAliveBlocks(), CallExpr::match(), llvm::AttributeFuncs::mergeAttributesForInlining(), llvm::pdb::operator<<(), llvm::FortifiedLibCallSimplifier::optimizeCall(), llvm::LibCallSimplifier::optimizeCall(), AbstractManglingParser< ManglingParser< Alloc >, Alloc >::parseExpr(), prepareDescriptorIndirectCall(), prepareIndirectCall(), CallExpr::printLeft(), llvm::sampleprof::ProfiledCallGraph::ProfiledCallGraph(), llvm::promoteCall(), llvm::promoteCallWithIfThenElse(), propagateCallAttrsFromCallee(), llvm::ImportedFunctionsInliningStatistics::recordInline(), llvm::CallGraphNode::removeAnyCallEdgeTo(), llvm::CallGraphNode::removeOneAbstractEdgeTo(), llvm::ReplayInlineAdvisor::ReplayInlineAdvisor(), llvm::AMDGPUSimplifyLibCallsPass::run(), llvm::AMDGPUUseNativeCallsPass::run(), llvm::InlinerPass::run(), runImpl(), runNVVMIntrRange(), llvm::FastISel::selectPatchpoint(), setAND(), setOR(), llvm::shouldInline(), llvm::SimplifyCall(), simplifySuspendPoint(), llvm::SubsumingPositionIterator::SubsumingPositionIterator(), transformCallee(), llvm::tryPromoteCall(), llvm::coro::updateCallGraph(), UpdateCallGraphAfterInlining(), updateCallProfile(), llvm::updateProfileCallee(), llvm::AMDGPULibCalls::useNative(), versionCallSite(), and llvm::sampleprof::SampleProfileWriterBinary::writeBody().

calls

amdgpu Simplify well known AMD library calls

Definition at line 198 of file AMDGPULibCalls.cpp.

EnablePreLink

cl::opt<bool> EnablePreLink("amdgpu-prelink", cl::desc("Enable pre-link mode optimizations"), cl::init(false), cl::Hidden)

static

Referenced by llvm::AMDGPULibCalls::fold().

false

amdgpu Simplify well known AMD library false

Definition at line 198 of file AMDGPULibCalls.cpp.

Referenced by llvm::AMDGPULibCalls::fold().

Name

amdgpu Simplify well known AMD library false FunctionCallee Value const Twine& Name

Initial value:

= "") {
  CallInst *R = B.CreateCall(Callee, Arg, Name);
  if (Function *F = dyn_cast<Function>(Callee.getCallee()))
    R->setCallingConv(F->getCallingConv());
  return R;
}
 
template <typename IRB>
static CallInst *CreateCallEx2(IRB &B, FunctionCallee Callee, Value *Arg1,
                               Value *Arg2, const Twine &Name = "") {
  CallInst *R = B.CreateCall(Callee, {Arg1, Arg2}, Name);
  if (Function *F = dyn_cast<Function>(Callee.getCallee()))
    R->setCallingConv(F->getCallingConv());
  return R;
}
 
 
 
 
struct TableEntry {
  double   result;
  double   input;
}

Definition at line 206 of file AMDGPULibCalls.cpp.

simplifylib

amdgpu simplifylib

Definition at line 197 of file AMDGPULibCalls.cpp.

tbl_acos

const TableEntry tbl_acos[]

static

Initial value:

= {
  {MATH_PI / 2.0, 0.0},
  {MATH_PI / 2.0, -0.0},
  {0.0, 1.0},
  {MATH_PI, -1.0}
}

Definition at line 231 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_acosh

const TableEntry tbl_acosh[]

static

Initial value:

= {
  {0.0, 1.0}
}

Definition at line 237 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_acospi

const TableEntry tbl_acospi[]

static

Initial value:

= {
  {0.5, 0.0},
  {0.5, -0.0},
  {0.0, 1.0},
  {1.0, -1.0}
}

Definition at line 240 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_asin

const TableEntry tbl_asin[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0},
  {MATH_PI / 2.0, 1.0},
  {-MATH_PI / 2.0, -1.0}
}

Definition at line 246 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_asinh

const TableEntry tbl_asinh[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 252 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_asinpi

const TableEntry tbl_asinpi[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0},
  {0.5, 1.0},
  {-0.5, -1.0}
}

Definition at line 256 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_atan

const TableEntry tbl_atan[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0},
  {MATH_PI / 4.0, 1.0},
  {-MATH_PI / 4.0, -1.0}
}

Definition at line 262 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_atanh

const TableEntry tbl_atanh[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 268 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_atanpi

const TableEntry tbl_atanpi[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0},
  {0.25, 1.0},
  {-0.25, -1.0}
}

Definition at line 272 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_cbrt

const TableEntry tbl_cbrt[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0},
  {1.0, 1.0},
  {-1.0, -1.0},
}

Definition at line 278 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_cos

const TableEntry tbl_cos[]

static

Initial value:

= {
  {1.0, 0.0},
  {1.0, -0.0}
}

Definition at line 284 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_cosh

const TableEntry tbl_cosh[]

static

Initial value:

= {
  {1.0, 0.0},
  {1.0, -0.0}
}

Definition at line 288 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_cospi

const TableEntry tbl_cospi[]

static

Initial value:

= {
  {1.0, 0.0},
  {1.0, -0.0}
}

Definition at line 292 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_erf

const TableEntry tbl_erf[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 300 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_erfc

const TableEntry tbl_erfc[]

static

Initial value:

= {
  {1.0, 0.0},
  {1.0, -0.0}
}

Definition at line 296 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_exp

const TableEntry tbl_exp[]

static

Initial value:

= {
  {1.0, 0.0},
  {1.0, -0.0},
  {MATH_E, 1.0}
}

Definition at line 304 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_exp10

const TableEntry tbl_exp10[]

static

Initial value:

= {
  {1.0, 0.0},
  {1.0, -0.0},
  {10.0, 1.0}
}

Definition at line 314 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_exp2

const TableEntry tbl_exp2[]

static

Initial value:

= {
  {1.0, 0.0},
  {1.0, -0.0},
  {2.0, 1.0}
}

Definition at line 309 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_expm1

const TableEntry tbl_expm1[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 319 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_log

const TableEntry tbl_log[]

static

Initial value:

= {
  {0.0, 1.0},
  {1.0, MATH_E}
}

Definition at line 323 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_log10

const TableEntry tbl_log10[]

static

Initial value:

= {
  {0.0, 1.0},
  {1.0, 10.0}
}

Definition at line 331 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_log2

const TableEntry tbl_log2[]

static

Initial value:

= {
  {0.0, 1.0},
  {1.0, 2.0}
}

Definition at line 327 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_rsqrt

const TableEntry tbl_rsqrt[]

static

Initial value:

= {
  {1.0, 1.0},
  {MATH_SQRT1_2, 2.0}
}

Definition at line 335 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_sin

const TableEntry tbl_sin[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 339 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_sinh

const TableEntry tbl_sinh[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 343 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_sinpi

const TableEntry tbl_sinpi[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 347 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_sqrt

const TableEntry tbl_sqrt[]

static

Initial value:

= {
  {0.0, 0.0},
  {1.0, 1.0},
  {MATH_SQRT2, 2.0}
}

Definition at line 351 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_tan

const TableEntry tbl_tan[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 356 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_tanh

const TableEntry tbl_tanh[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 360 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_tanpi

const TableEntry tbl_tanpi[]

static

Initial value:

= {
  {0.0, 0.0},
  {-0.0, -0.0}
}

Definition at line 364 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

tbl_tgamma

const TableEntry tbl_tgamma[]

static

Initial value:

= {
  {1.0, 1.0},
  {1.0, 2.0},
  {2.0, 3.0},
  {6.0, 4.0}
}

Definition at line 368 of file AMDGPULibCalls.cpp.

Referenced by getOptTable().

UseNative

cl::list<std::string> UseNative("amdgpu-use-native", cl::desc("Comma separated list of functions to replace with native, or all"), cl::CommaSeparated, cl::ValueOptional, cl::Hidden)

static

Referenced by llvm::AMDGPULibCalls::initNativeFuncs(), and llvm::AMDGPUUseNativeCallsPass::run().