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
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 *, WeakTrackingVH > &GEPIdx)
static Value *	GEPToVectorIndex (GetElementPtrInst *GEP, AllocaInst *Alloca, Type *VecElemTy, const DataLayout &DL, SmallVector< Instruction * > &NewInsts)
static Value *	promoteAllocaUserToVector (Instruction *Inst, const DataLayout &DL, FixedVectorType *VectorTy, unsigned VecStoreSize, unsigned ElementSize, DenseMap< MemTransferInst *, MemTransferInfo > &TransferInfo, std::map< GetElementPtrInst *, WeakTrackingVH > &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 BasicBlock::iterator	skipToNonAllocaInsertPt (BasicBlock &BB, BasicBlock::iterator I)
	Find an insert point after an alloca, after all other allocas clustered at the start of the block.
static bool	isCallPromotable (CallInst *CI)

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

Macro Definition Documentation

DEBUG_TYPE

#define DEBUG_TYPE   "amdgpu-promote-alloca"

Definition at line 48 of file AMDGPUPromoteAlloca.cpp.

Function Documentation

calculateVectorIndex()

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

static

Definition at line 390 of file AMDGPUPromoteAlloca.cpp.

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

Referenced by promoteAllocaUserToVector().

collectAllocaUses()

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

static

Definition at line 239 of file AMDGPUPromoteAlloca.cpp.

References llvm::cast(), llvm::isa(), and Uses.

forEachWorkListItem()

template<typename InstContainer>

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 749 of file AMDGPUPromoteAlloca.cpp.

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

GEPToVectorIndex()

Value * GEPToVectorIndex ( GetElementPtrInst * GEP, AllocaInst * Alloca, Type * VecElemTy, const DataLayout & DL, SmallVector< Instruction * > & NewInsts )

static

Definition at line 404 of file AMDGPUPromoteAlloca.cpp.

References assert(), DL, llvm::dyn_cast(), llvm::MapVector< KeyT, ValueT, MapType, VectorType >::front(), GEP, llvm::APInt::isOne(), llvm::APInt::isZero(), llvm::Offset, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::APInt::sdivrem(), llvm::APInt::sext(), and llvm::MapVector< KeyT, ValueT, MapType, VectorType >::size().

INITIALIZE_PASS_BEGIN()

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

References DEBUG_TYPE, and INITIALIZE_PASS_DEPENDENCY.

isCallPromotable()

bool isCallPromotable ( CallInst * CI )

static

Definition at line 1197 of file AMDGPUPromoteAlloca.cpp.

References llvm::dyn_cast(), and II.

isSupportedAccessType()

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

static

Definition at line 718 of file AMDGPUPromoteAlloca.cpp.

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

isSupportedMemset()

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

static

Definition at line 376 of file AMDGPUPromoteAlloca.cpp.

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

promoteAllocaUserToVector()

Value * promoteAllocaUserToVector ( Instruction * Inst, const DataLayout & DL, FixedVectorType * VectorTy, unsigned VecStoreSize, unsigned ElementSize, DenseMap< MemTransferInst *, MemTransferInfo > & TransferInfo, std::map< GetElementPtrInst *, WeakTrackingVH > & 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 500 of file AMDGPUPromoteAlloca.cpp.

References assert(), calculateVectorIndex(), llvm::cast(), MemTransferInfo::DestIndex, DL, llvm::dyn_cast(), llvm::FixedVectorType::get(), llvm::PoisonValue::get(), llvm::Value::getContext(), llvm::VectorType::getElementCount(), llvm::VectorType::getElementType(), llvm::Type::getIntNTy(), getNumElements(), llvm::FixedVectorType::getNumElements(), llvm::Instruction::getOpcode(), llvm::getPointerOperand(), llvm::Value::getType(), llvm::ConstantInt::getZExtValue(), llvm::isa(), llvm::details::FixedOrScalableQuantity< LeafTy, ValueTy >::isKnownMultipleOf(), llvm::Type::isPointerTy(), llvm::Type::isPtrOrPtrVectorTy(), llvm::Length, llvm_unreachable, llvm::PoisonMaskElem, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::Value::replaceAllUsesWith(), Size, and MemTransferInfo::SrcIndex.

skipToNonAllocaInsertPt()

BasicBlock::iterator skipToNonAllocaInsertPt ( BasicBlock & BB, BasicBlock::iterator I )

static

Find an insert point after an alloca, after all other allocas clustered at the start of the block.

Definition at line 787 of file AMDGPUPromoteAlloca.cpp.

References E(), llvm::BasicBlock::end(), I, and llvm::isa().

Variable Documentation

DEBUG_TYPE

DEBUG_TYPE

Definition at line 206 of file AMDGPUPromoteAlloca.cpp.

false

AMDGPU promote alloca to vector or false

Definition at line 207 of file AMDGPUPromoteAlloca.cpp.

LDS

AMDGPU promote alloca to vector or LDS

Definition at line 207 of file AMDGPUPromoteAlloca.cpp.

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