ARMParallelDSP.cpp File Reference

Armv6 introduced instructions to perform 32-bit SIMD operations. More...

#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/LoopAccessAnalysis.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/NoFolder.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/LoopUtils.h"
#include "llvm/Pass.h"
#include "llvm/PassRegistry.h"
#include "llvm/PassSupport.h"
#include "llvm/Support/Debug.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/CodeGen/TargetPassConfig.h"
#include "ARM.h"
#include "ARMSubtarget.h"

Include dependency graph for ARMParallelDSP.cpp:

Go to the source code of this file.

Macros
#define	DEBUG_TYPE   "arm-parallel-dsp"

Functions
	STATISTIC (NumSMLAD, "Number of smlad instructions generated")
template<unsigned MaxBitWidth>
static bool	IsNarrowSequence (Value *V, ValueList &VL)
template<typename MemInst >
static bool	AreSequentialAccesses (MemInst *MemOp0, MemInst *MemOp1, const DataLayout &DL, ScalarEvolution &SE)
static void	MatchReductions (Function &F, Loop *TheLoop, BasicBlock *Header, ReductionList &Reductions)
static void	AddMACCandidate (OpChainList &Candidates, Instruction *Mul, Value *MulOp0, Value *MulOp1)
static void	MatchParallelMACSequences (Reduction &R, OpChainList &Candidates)
static void	AliasCandidates (BasicBlock *Header, Instructions &Reads, Instructions &Writes)
static bool	AreAliased (AliasAnalysis *AA, Instructions &Reads, Instructions &Writes, OpChainList &MACCandidates)
static bool	CheckMACMemory (OpChainList &Candidates)
static LoadInst *	CreateLoadIns (IRBuilder< NoFolder > &IRB, LoadInst &BaseLoad, const Type *LoadTy)
	INITIALIZE_PASS_BEGIN (ARMParallelDSP, "arm-parallel-dsp", "Transform loops to use DSP intrinsics", false, false) INITIALIZE_PASS_END(ARMParallelDSP

Variables
static cl::opt< bool >	DisableParallelDSP ("disable-arm-parallel-dsp", cl::Hidden, cl::init(false), cl::desc("Disable the ARM Parallel DSP pass"))
arm parallel	dsp
arm parallel Transform loops to use DSP	intrinsics
arm parallel Transform loops to use DSP	false

Detailed Description

Armv6 introduced instructions to perform 32-bit SIMD operations.

The purpose of this pass is do some IR pattern matching to create ACLE DSP intrinsics, which map on these 32-bit SIMD operations. This pass runs only when unaligned accesses is supported/enabled.

Definition in file ARMParallelDSP.cpp.

Macro Definition Documentation

DEBUG_TYPE

#define DEBUG_TYPE   "arm-parallel-dsp"

Definition at line 41 of file ARMParallelDSP.cpp.

Function Documentation

AddMACCandidate()

static void AddMACCandidate ( OpChainList &  Candidates, Instruction *  Mul, Value *  MulOp0, Value *  MulOp1  )

static

Definition at line 500 of file ARMParallelDSP.cpp.

References assert(), llvm::dbgs(), llvm::Value::dump(), llvm::Instruction::getOpcode(), and LLVM_DEBUG.

Referenced by MatchParallelMACSequences().

AliasCandidates()

static void AliasCandidates ( BasicBlock *  Header, Instructions &  Reads, Instructions &  Writes  )

static

Definition at line 553 of file ARMParallelDSP.cpp.

References llvm::Instruction::mayReadFromMemory(), and llvm::Instruction::mayWriteToMemory().

Referenced by CheckMACMemory().

AreAliased()

static bool AreAliased ( AliasAnalysis *  AA, Instructions &  Reads, Instructions &  Writes, OpChainList &  MACCandidates  )

static

Definition at line 566 of file ARMParallelDSP.cpp.

References assert(), llvm::dbgs(), llvm::Value::dump(), llvm::AAResults::getModRefInfo(), llvm::intersectModRef(), llvm::isModOrRefSet(), LLVM_DEBUG, and llvm::ModRef.

Referenced by CheckMACMemory().

AreSequentialAccesses()

template<typename MemInst >

static bool AreSequentialAccesses ( MemInst *  MemOp0, MemInst *  MemOp1, const DataLayout &  DL, ScalarEvolution &  SE  )

static

Definition at line 269 of file ARMParallelDSP.cpp.

References assert(), llvm::SmallPtrSetImpl< PtrType >::count(), llvm::dbgs(), llvm::Value::dump(), llvm::dyn_cast(), llvm::Value::hasOneUse(), I, llvm::SmallPtrSetImpl< PtrType >::insert(), llvm::isConsecutiveAccess(), LLVM_DEBUG, llvm::SmallVectorTemplateBase< T >::push_back(), and Reduction.

CheckMACMemory()

static bool CheckMACMemory ( OpChainList &  Candidates )

static

Definition at line 597 of file ARMParallelDSP.cpp.

References AliasCandidates(), AreAliased(), C, llvm::dbgs(), llvm::Value::dump(), llvm::Loop::dump(), llvm::Module::dump(), llvm::LoopBase< BlockT, LoopT >::getHeader(), LLVM_DEBUG, MatchParallelMACSequences(), and MatchReductions().

CreateLoadIns()

static LoadInst* CreateLoadIns ( IRBuilder< NoFolder > &  IRB, LoadInst &  BaseLoad, const Type *  LoadTy  )

static

Definition at line 691 of file ARMParallelDSP.cpp.

References llvm::AMDGPU::HSAMD::Kernel::Key::Args, llvm::Intrinsic::arm_smlad, llvm::Intrinsic::arm_smladx, llvm::Intrinsic::arm_smlald, llvm::Intrinsic::arm_smlaldx, llvm::IRBuilder< T, Inserter >::CreateAlignedLoad(), llvm::IRBuilder< T, Inserter >::CreateBitCast(), llvm::dbgs(), llvm::Value::dump(), llvm::dyn_cast(), llvm::IntegerType::get(), llvm::LoadInst::getAlignment(), llvm::Module::getContext(), llvm::Intrinsic::getDeclaration(), llvm::Instruction::getParent(), llvm::LoadInst::getPointerAddressSpace(), llvm::LoadInst::getPointerOperand(), llvm::Type::getPointerTo(), llvm::GlobalValue::getType(), llvm::Type::isIntegerTy(), LLVM_DEBUG, llvm::PatternMatch::m_APInt(), llvm::PatternMatch::match(), and Other.

INITIALIZE_PASS_BEGIN()

INITIALIZE_PASS_BEGIN	(	ARMParallelDSP	,
		"arm-parallel-dsp"	,
		"Transform loops to use DSP intrinsics"	,
		false	,
		false
	)

Referenced by llvm::createARMParallelDSPPass().

IsNarrowSequence()

template<unsigned MaxBitWidth>

static bool IsNarrowSequence ( Value *  V, ValueList &  VL  )

static

Definition at line 228 of file ARMParallelDSP.cpp.

References llvm::dbgs(), llvm::Value::dump(), llvm::dyn_cast(), LLVM_DEBUG, llvm::PatternMatch::m_Add(), llvm::PatternMatch::m_ConstantInt(), llvm::PatternMatch::m_Load(), llvm::PatternMatch::m_Trunc(), llvm::PatternMatch::m_Value(), llvm::PatternMatch::m_ZExtOrSExt(), and llvm::PatternMatch::match().

MatchParallelMACSequences()

static void MatchParallelMACSequences ( Reduction &  R, OpChainList &  Candidates  )

static

Definition at line 515 of file ARMParallelDSP.cpp.

References llvm::MCID::Add, AddMACCandidate(), llvm::dbgs(), llvm::dyn_cast(), llvm::Instruction::getOpcode(), llvm::User::getOperand(), LLVM_DEBUG, and llvm::Match.

Referenced by CheckMACMemory().

MatchReductions()

static void MatchReductions ( Function &  F, Loop *  TheLoop, BasicBlock *  Header, ReductionList &  Reductions  )

static

Definition at line 459 of file ARMParallelDSP.cpp.

References llvm::RecurrenceDescriptor::AddReductionVar(), llvm::dbgs(), llvm::dyn_cast(), llvm::Function::getFnAttribute(), llvm::LoopBase< BlockT, LoopT >::getLoopLatch(), llvm::LoopBase< BlockT, LoopT >::getLoopPreheader(), llvm::Attribute::getValueAsString(), LLVM_DEBUG, llvm::BasicBlock::phis(), Reduction, and llvm::RecurrenceDescriptor::RK_IntegerAdd.

Referenced by CheckMACMemory().

STATISTIC()

STATISTIC	(	NumSMLAD	,
		"Number of smlad instructions generated"
	)

Variable Documentation

DisableParallelDSP

cl::opt<bool> DisableParallelDSP("disable-arm-parallel-dsp", cl::Hidden, cl::init(false), cl::desc("Disable the ARM Parallel DSP pass"))

static

dsp

arm parallel dsp

Definition at line 785 of file ARMParallelDSP.cpp.

false

arm parallel Transform loops to use DSP false

Definition at line 785 of file ARMParallelDSP.cpp.

intrinsics

arm parallel Transform loops to use DSP intrinsics

Definition at line 785 of file ARMParallelDSP.cpp.