AMDGPUCodeGenPrepare.cpp File Reference

This pass does misc. More...

#include "AMDGPU.h"
#include "AMDGPUTargetMachine.h"
#include "SIModeRegisterDefaults.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/UniformityAnalysis.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/TargetPassConfig.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InstVisitor.h"
#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
#include "llvm/Support/KnownBits.h"
#include "llvm/Transforms/Utils/IntegerDivision.h"
#include "llvm/Transforms/Utils/Local.h"

Go to the source code of this file.

Classes
class	VectorSlice
	Helper class for "break large PHIs" (visitPHINode). More...

Macros
#define	DEBUG_TYPE   "amdgpu-codegenprepare"

Functions
static bool	promotedOpIsNSW (const Instruction &I)
static bool	promotedOpIsNUW (const Instruction &I)
static void	extractValues (IRBuilder<> &Builder, SmallVectorImpl< Value * > &Values, Value *V)
static Value *	insertValues (IRBuilder<> &Builder, Type *Ty, SmallVectorImpl< Value * > &Values)
static SelectInst *	findSelectThroughCast (Value *V, CastInst *&Cast)
static Value *	emitRsqIEEE1ULP (IRBuilder<> &Builder, Value *Src, bool IsNegative)
	Emit an expansion of 1.0 / sqrt(Src) good for 1ulp that supports denormals.
static bool	hasUnsafeFPMath (const Function &F)
static std::pair< Value *, Value * >	getMul64 (IRBuilder<> &Builder, Value *LHS, Value *RHS)
static Value *	getMulHu (IRBuilder<> &Builder, Value *LHS, Value *RHS)
static Value *	getSign32 (Value *V, IRBuilder<> &Builder, const DataLayout *DL)
static bool	areInSameBB (const Value *A, const Value *B)
static bool	isInterestingPHIIncomingValue (const Value *V)
static void	collectPHINodes (const PHINode &I, SmallPtrSet< const PHINode *, 8 > &SeenPHIs)
static bool	isOneOrNegOne (const Value *Val)
	INITIALIZE_PASS_BEGIN (AMDGPUCodeGenPrepare, DEBUG_TYPE, "AMDGPU IR optimizations", false, false) INITIALIZE_PASS_END(AMDGPUCodeGenPrepare

Variables
	DEBUG_TYPE
AMDGPU IR	optimizations
AMDGPU IR	false

Detailed Description

This pass does misc.

AMDGPU optimizations on IR before instruction selection.

Definition in file AMDGPUCodeGenPrepare.cpp.

Macro Definition Documentation

DEBUG_TYPE

#define DEBUG_TYPE   "amdgpu-codegenprepare"

Definition at line 35 of file AMDGPUCodeGenPrepare.cpp.

Function Documentation

areInSameBB()

static bool areInSameBB ( const Value *  A, const Value *  B  )

static

Definition at line 1724 of file AMDGPUCodeGenPrepare.cpp.

References A, and B.

Referenced by isInterestingPHIIncomingValue().

collectPHINodes()

static void collectPHINodes ( const PHINode &  I, SmallPtrSet< const PHINode *, 8 > &  SeenPHIs  )

static

Definition at line 1787 of file AMDGPUCodeGenPrepare.cpp.

References collectPHINodes(), I, and llvm::SmallPtrSetImpl< PtrType >::insert().

Referenced by collectPHINodes().

emitRsqIEEE1ULP()

static Value * emitRsqIEEE1ULP ( IRBuilder<> &  Builder, Value *  Src, bool  IsNegative  )

static

Emit an expansion of 1.0 / sqrt(Src) good for 1ulp that supports denormals.

Definition at line 861 of file AMDGPUCodeGenPrepare.cpp.

References llvm::IRBuilderBase::CreateFCmpOLT(), llvm::IRBuilderBase::CreateFMul(), llvm::IRBuilderBase::CreateSelect(), llvm::IRBuilderBase::CreateUnaryIntrinsic(), llvm::ConstantFP::get(), llvm::Type::getFltSemantics(), and llvm::APFloat::getSmallestNormalized().

extractValues()

static void extractValues ( IRBuilder<> &  Builder, SmallVectorImpl< Value * > &  Values, Value *  V  )

static

Definition at line 600 of file AMDGPUCodeGenPrepare.cpp.

References llvm::IRBuilderBase::CreateExtractElement(), E, I, and llvm::SmallVectorTemplateBase< T, bool >::push_back().

findSelectThroughCast()

static SelectInst * findSelectThroughCast ( Value *  V, CastInst *&  Cast  )

static

Definition at line 692 of file AMDGPUCodeGenPrepare.cpp.

References llvm::User::getOperand().

getMul64()

static std::pair< Value *, Value * > getMul64 ( IRBuilder<> &  Builder, Value *  LHS, Value *  RHS  )

static

Definition at line 1170 of file AMDGPUCodeGenPrepare.cpp.

References llvm::IRBuilderBase::CreateLShr(), llvm::IRBuilderBase::CreateMul(), llvm::IRBuilderBase::CreateTrunc(), llvm::IRBuilderBase::CreateZExt(), llvm::IRBuilderBase::getInt32Ty(), llvm::IRBuilderBase::getInt64(), llvm::IRBuilderBase::getInt64Ty(), Hi, LHS, Lo, and RHS.

Referenced by getMulHu().

getMulHu()

static Value * getMulHu ( IRBuilder<> &  Builder, Value *  LHS, Value *  RHS  )

static

Definition at line 1184 of file AMDGPUCodeGenPrepare.cpp.

References getMul64(), LHS, and RHS.

getSign32()

static Value * getSign32 ( Value *  V, IRBuilder<> &  Builder, const DataLayout *  DL  )

static

Definition at line 1357 of file AMDGPUCodeGenPrepare.cpp.

References llvm::computeKnownBits(), llvm::IRBuilderBase::CreateAShr(), DL, llvm::Constant::getAllOnesValue(), llvm::IRBuilderBase::getInt32(), llvm::Constant::getNullValue(), llvm::KnownBits::isNegative(), and llvm::KnownBits::isNonNegative().

hasUnsafeFPMath()

static bool hasUnsafeFPMath ( const Function &  F )

static

Definition at line 1165 of file AMDGPUCodeGenPrepare.cpp.

References F, and llvm::Attribute::getValueAsBool().

Referenced by llvm::AMDGPUCodeGenPreparePass::run().

INITIALIZE_PASS_BEGIN()

INITIALIZE_PASS_BEGIN	(	AMDGPUCodeGenPrepare	,
		DEBUG_TYPE	,
		"AMDGPU IR optimizations"	,
		false	,
		false
	)

insertValues()

static Value * insertValues ( IRBuilder<> &  Builder, Type *  Ty, SmallVectorImpl< Value * > &  Values  )

static

Definition at line 612 of file AMDGPUCodeGenPrepare.cpp.

References assert(), llvm::IRBuilderBase::CreateInsertElement(), E, llvm::PoisonValue::get(), I, llvm::Type::isVectorTy(), and llvm::SmallVectorBase< Size_T >::size().

isInterestingPHIIncomingValue()

static bool isInterestingPHIIncomingValue ( const Value *  V )

static

Definition at line 1732 of file AMDGPUCodeGenPrepare.cpp.

References llvm::BitVector::all(), areInSameBB(), Idx, and llvm::BitVector::set().

isOneOrNegOne()

static bool isOneOrNegOne ( const Value *  Val )

static

Definition at line 2114 of file AMDGPUCodeGenPrepare.cpp.

References llvm::CallingConv::C, llvm::PatternMatch::m_APFloat(), and llvm::PatternMatch::match().

promotedOpIsNSW()

static bool promotedOpIsNSW ( const Instruction &  I )

static

Definition at line 427 of file AMDGPUCodeGenPrepare.cpp.

References I.

promotedOpIsNUW()

static bool promotedOpIsNUW ( const Instruction &  I )

static

Definition at line 441 of file AMDGPUCodeGenPrepare.cpp.

References I.

Variable Documentation

DEBUG_TYPE

DEBUG_TYPE

Definition at line 2233 of file AMDGPUCodeGenPrepare.cpp.

false

AMDGPU IR false

Definition at line 2234 of file AMDGPUCodeGenPrepare.cpp.

optimizations

AMDGPU IR optimizations

Definition at line 2233 of file AMDGPUCodeGenPrepare.cpp.