AMDGPUPromoteAlloca.cpp File Reference

#include "AMDGPU.h"
#include "GCNSubtarget.h"
#include "Utils/AMDGPUBaseInfo.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Analysis/CaptureTracking.h"
#include "llvm/Analysis/InstSimplifyFolder.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/TargetPassConfig.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/IR/IntrinsicsR600.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Transforms/Utils/SSAUpdater.h"

Go to the source code of this file.

Classes
struct	MemTransferInfo

Macros
#define	DEBUG_TYPE   "amdgpu-promote-alloca"

Functions
	INITIALIZE_PASS_BEGIN (AMDGPUPromoteAlloca, DEBUG_TYPE, "AMDGPU promote alloca to vector or LDS", false, false) INITIALIZE_PASS_END(AMDGPUPromoteAlloca
AMDGPU promote alloca to vector or false	INITIALIZE_PASS_BEGIN (AMDGPUPromoteAllocaToVector, DEBUG_TYPE "-to-vector", "AMDGPU promote alloca to vector", false, false) INITIALIZE_PASS_END(AMDGPUPromoteAllocaToVector
static void	collectAllocaUses (AllocaInst &Alloca, SmallVectorImpl< Use * > &Uses)
static bool	isSupportedMemset (MemSetInst *I, AllocaInst *AI, const DataLayout &DL)
static Value *	calculateVectorIndex (Value *Ptr, const std::map< GetElementPtrInst *, Value * > &GEPIdx)
static Value *	GEPToVectorIndex (GetElementPtrInst *GEP, AllocaInst *Alloca, Type *VecElemTy, const DataLayout &DL)
static Value *	promoteAllocaUserToVector (Instruction *Inst, const DataLayout &DL, FixedVectorType *VectorTy, unsigned VecStoreSize, unsigned ElementSize, DenseMap< MemTransferInst *, MemTransferInfo > &TransferInfo, std::map< GetElementPtrInst *, Value * > &GEPVectorIdx, Value *CurVal, SmallVectorImpl< LoadInst * > &DeferredLoads)
	Promotes a single user of the alloca to a vector form.
static bool	isSupportedAccessType (FixedVectorType *VecTy, Type *AccessTy, const DataLayout &DL)
template<typename InstContainer >
static void	forEachWorkListItem (const InstContainer &WorkList, std::function< void(Instruction *)> Fn)
	Iterates over an instruction worklist that may contain multiple instructions from the same basic block, but in a different order.
static bool	isCallPromotable (CallInst *CI)

Variables
	DEBUG_TYPE
AMDGPU promote alloca to vector or	LDS
AMDGPU promote alloca to vector or	false
AMDGPU promote alloca to vector or false DEBUG_TYPE to	vector

Macro Definition Documentation

DEBUG_TYPE

#define DEBUG_TYPE   "amdgpu-promote-alloca"

Definition at line 48 of file AMDGPUPromoteAlloca.cpp.

Function Documentation

calculateVectorIndex()

static Value * calculateVectorIndex ( Value *  Ptr, const std::map< GetElementPtrInst *, Value * > &  GEPIdx  )

static

Definition at line 385 of file AMDGPUPromoteAlloca.cpp.

References assert(), GEP, llvm::Type::getInt32Ty(), llvm::Constant::getNullValue(), I, and Ptr.

Referenced by promoteAllocaUserToVector().

collectAllocaUses()

static void collectAllocaUses ( AllocaInst &  Alloca, SmallVectorImpl< Use * > &  Uses  )

static

Definition at line 252 of file AMDGPUPromoteAlloca.cpp.

References llvm::SmallVectorImpl< T >::pop_back_val(), and Uses.

forEachWorkListItem()

template<typename InstContainer >

static void forEachWorkListItem ( const InstContainer &  WorkList, std::function< void(Instruction *)>  Fn  )

static

Iterates over an instruction worklist that may contain multiple instructions from the same basic block, but in a different order.

Definition at line 686 of file AMDGPUPromoteAlloca.cpp.

References llvm::DenseMapBase< DerivedT, KeyT, ValueT, KeyInfoT, BucketT >::insert().

GEPToVectorIndex()

static Value * GEPToVectorIndex ( GetElementPtrInst *  GEP, AllocaInst *  Alloca, Type *  VecElemTy, const DataLayout &  DL  )

static

Definition at line 396 of file AMDGPUPromoteAlloca.cpp.

References DL, llvm::MapVector< KeyT, ValueT, MapType, VectorType >::front(), GEP, llvm::APInt::isZero(), llvm::MapVector< KeyT, ValueT, MapType, VectorType >::size(), and llvm::APInt::udivrem().

INITIALIZE_PASS_BEGIN() [1/2]

INITIALIZE_PASS_BEGIN	(	AMDGPUPromoteAlloca	,
		DEBUG_TYPE	,
		"AMDGPU promote alloca to vector or LDS"	,
		false	,
		false
	)

INITIALIZE_PASS_BEGIN() [2/2]

AMDGPU promote alloca to vector or false INITIALIZE_PASS_BEGIN	(	AMDGPUPromoteAllocaToVector	,
		DEBUG_TYPE "-to-vector"	,
		"AMDGPU promote alloca to vector"	,
		false	,
		false
	)

isCallPromotable()

static bool isCallPromotable ( CallInst *  CI )

static

Definition at line 1091 of file AMDGPUPromoteAlloca.cpp.

References llvm::IntrinsicInst::getIntrinsicID().

isSupportedAccessType()

static bool isSupportedAccessType ( FixedVectorType *  VecTy, Type *  AccessTy, const DataLayout &  DL  )

static

Definition at line 660 of file AMDGPUPromoteAlloca.cpp.

References DL, llvm::VectorType::getElementType(), llvm::CastInst::isBitOrNoopPointerCastable(), and llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isKnownMultipleOf().

isSupportedMemset()

static bool isSupportedMemset ( MemSetInst *  I, AllocaInst *  AI, const DataLayout &  DL  )

static

Definition at line 370 of file AMDGPUPromoteAlloca.cpp.

References DL, llvm::AllocaInst::getAllocatedType(), I, llvm::PatternMatch::m_SpecificInt(), llvm::PatternMatch::match(), and Size.

promoteAllocaUserToVector()

static Value * promoteAllocaUserToVector ( Instruction *  Inst, const DataLayout &  DL, FixedVectorType *  VectorTy, unsigned  VecStoreSize, unsigned  ElementSize, DenseMap< MemTransferInst *, MemTransferInfo > &  TransferInfo, std::map< GetElementPtrInst *, Value * > &  GEPVectorIdx, Value *  CurVal, SmallVectorImpl< LoadInst * > &  DeferredLoads  )

static

Promotes a single user of the alloca to a vector form.

Parameters

	Inst	Instruction to be promoted.
	DL	Module Data Layout.
	VectorTy	Vectorized Type.
	VecStoreSize	Size of VectorTy in bytes.
	ElementSize	Size of VectorTy element type in bytes.
	TransferInfo	MemTransferInst info map.
	GEPVectorIdx	GEP -> VectorIdx cache.
	CurVal	Current value of the vector (e.g. last stored value)
[out]	DeferredLoads	Inst is added to this vector if it can't be promoted now. This happens when promoting requires CurVal, but CurVal is nullptr.

Returns

the stored value if Inst would have written to the alloca, or nullptr otherwise.

Definition at line 444 of file AMDGPUPromoteAlloca.cpp.

References assert(), calculateVectorIndex(), llvm::IRBuilderBase::CreateAdd(), llvm::IRBuilderBase::CreateBitCast(), llvm::IRBuilderBase::CreateBitOrPointerCast(), llvm::IRBuilderBase::CreateExtractElement(), llvm::IRBuilderBase::CreateInsertElement(), llvm::IRBuilderBase::CreateIntToPtr(), llvm::IRBuilderBase::CreateLoad(), llvm::IRBuilderBase::CreateShuffleVector(), llvm::IRBuilderBase::CreateVectorSplat(), MemTransferInfo::DestIndex, DL, llvm::FixedVectorType::get(), llvm::PoisonValue::get(), llvm::Value::getContext(), llvm::VectorType::getElementCount(), llvm::VectorType::getElementType(), llvm::IRBuilderBase::getIntN(), llvm::IRBuilderBase::getIntNTy(), llvm::FixedVectorType::getNumElements(), llvm::Instruction::getOpcode(), llvm::getPointerOperand(), llvm::IRBuilderBase::getPtrTy(), llvm::Value::getType(), llvm::ConstantInt::getZExtValue(), Idx, Intr, llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isKnownMultipleOf(), llvm::Type::isPointerTy(), llvm::Type::isPtrOrPtrVectorTy(), llvm::Length, llvm_unreachable, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::Value::replaceAllUsesWith(), llvm::IRBuilderBase::SetInsertPoint(), Size, and MemTransferInfo::SrcIndex.

Variable Documentation

DEBUG_TYPE

DEBUG_TYPE

Definition at line 208 of file AMDGPUPromoteAlloca.cpp.

false

AMDGPU promote alloca to vector or false DEBUG_TYPE to AMDGPU promote alloca to false

Definition at line 209 of file AMDGPUPromoteAlloca.cpp.

LDS

AMDGPU promote alloca to vector or LDS

Definition at line 209 of file AMDGPUPromoteAlloca.cpp.

Referenced by llvm::AMDGPUTargetLowering::getTargetNodeName().

vector

AMDGPU promote alloca to vector or false DEBUG_TYPE to AMDGPU promote alloca to vector

Definition at line 214 of file AMDGPUPromoteAlloca.cpp.

Referenced by llvm::LLT::fixed_vector(), and llvm::LLT::scalable_vector().