LLVM
15.0.0git
|
#include "llvm/Transforms/Scalar/LoopStrengthReduce.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/Hashing.h"
#include "llvm/ADT/PointerIntPair.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallBitVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/iterator_range.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/IVUsers.h"
#include "llvm/Analysis/LoopAnalysisManager.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/MemorySSA.h"
#include "llvm/Analysis/MemorySSAUpdater.h"
#include "llvm/Analysis/ScalarEvolution.h"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/Analysis/ScalarEvolutionNormalization.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/BinaryFormat/Dwarf.h"
#include "llvm/Config/llvm-config.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DebugInfoMetadata.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/PassManager.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Use.h"
#include "llvm/IR/User.h"
#include "llvm/IR/Value.h"
#include "llvm/IR/ValueHandle.h"
#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Transforms/Utils.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/LoopUtils.h"
#include "llvm/Transforms/Utils/ScalarEvolutionExpander.h"
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <cstdint>
#include <iterator>
#include <limits>
#include <map>
#include <numeric>
#include <utility>
Go to the source code of this file.
Macros | |
#define | DEBUG_TYPE "loop-reduce" |
Functions | |
static void | DoInitialMatch (const SCEV *S, Loop *L, SmallVectorImpl< const SCEV * > &Good, SmallVectorImpl< const SCEV * > &Bad, ScalarEvolution &SE) |
Recursion helper for initialMatch. More... | |
static bool | containsAddRecDependentOnLoop (const SCEV *S, const Loop &L) |
static bool | isAddRecSExtable (const SCEVAddRecExpr *AR, ScalarEvolution &SE) |
Return true if the given addrec can be sign-extended without changing its value. More... | |
static bool | isAddSExtable (const SCEVAddExpr *A, ScalarEvolution &SE) |
Return true if the given add can be sign-extended without changing its value. More... | |
static bool | isMulSExtable (const SCEVMulExpr *M, ScalarEvolution &SE) |
Return true if the given mul can be sign-extended without changing its value. More... | |
static const SCEV * | getExactSDiv (const SCEV *LHS, const SCEV *RHS, ScalarEvolution &SE, bool IgnoreSignificantBits=false) |
Return an expression for LHS /s RHS, if it can be determined and if the remainder is known to be zero, or null otherwise. More... | |
static int64_t | ExtractImmediate (const SCEV *&S, ScalarEvolution &SE) |
If S involves the addition of a constant integer value, return that integer value, and mutate S to point to a new SCEV with that value excluded. More... | |
static GlobalValue * | ExtractSymbol (const SCEV *&S, ScalarEvolution &SE) |
If S involves the addition of a GlobalValue address, return that symbol, and mutate S to point to a new SCEV with that value excluded. More... | |
static bool | isAddressUse (const TargetTransformInfo &TTI, Instruction *Inst, Value *OperandVal) |
Returns true if the specified instruction is using the specified value as an address. More... | |
static MemAccessTy | getAccessType (const TargetTransformInfo &TTI, Instruction *Inst, Value *OperandVal) |
Return the type of the memory being accessed. More... | |
static bool | isExistingPhi (const SCEVAddRecExpr *AR, ScalarEvolution &SE) |
Return true if this AddRec is already a phi in its loop. More... | |
static bool | isHighCostExpansion (const SCEV *S, SmallPtrSetImpl< const SCEV * > &Processed, ScalarEvolution &SE) |
Check if expanding this expression is likely to incur significant cost. More... | |
static bool | isAMCompletelyFolded (const TargetTransformInfo &TTI, const LSRUse &LU, const Formula &F) |
Check if the addressing mode defined by F is completely folded in LU at isel time. More... | |
static InstructionCost | getScalingFactorCost (const TargetTransformInfo &TTI, const LSRUse &LU, const Formula &F, const Loop &L) |
static bool | isAMCompletelyFolded (const TargetTransformInfo &TTI, LSRUse::KindType Kind, MemAccessTy AccessTy, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale, Instruction *Fixup=nullptr) |
static unsigned | getSetupCost (const SCEV *Reg, unsigned Depth) |
static bool | isAMCompletelyFolded (const TargetTransformInfo &TTI, int64_t MinOffset, int64_t MaxOffset, LSRUse::KindType Kind, MemAccessTy AccessTy, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale) |
static bool | isAMCompletelyFolded (const TargetTransformInfo &TTI, int64_t MinOffset, int64_t MaxOffset, LSRUse::KindType Kind, MemAccessTy AccessTy, const Formula &F, const Loop &L) |
static bool | isLegalUse (const TargetTransformInfo &TTI, int64_t MinOffset, int64_t MaxOffset, LSRUse::KindType Kind, MemAccessTy AccessTy, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale) |
Test whether we know how to expand the current formula. More... | |
static bool | isLegalUse (const TargetTransformInfo &TTI, int64_t MinOffset, int64_t MaxOffset, LSRUse::KindType Kind, MemAccessTy AccessTy, const Formula &F) |
static bool | isAlwaysFoldable (const TargetTransformInfo &TTI, LSRUse::KindType Kind, MemAccessTy AccessTy, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg) |
static bool | isAlwaysFoldable (const TargetTransformInfo &TTI, ScalarEvolution &SE, int64_t MinOffset, int64_t MaxOffset, LSRUse::KindType Kind, MemAccessTy AccessTy, const SCEV *S, bool HasBaseReg) |
static User::op_iterator | findIVOperand (User::op_iterator OI, User::op_iterator OE, Loop *L, ScalarEvolution &SE) |
Helper for CollectChains that finds an IV operand (computed by an AddRec in this loop) within [OI,OE) or returns OE. More... | |
static Value * | getWideOperand (Value *Oper) |
IVChain logic must consistently peek base TruncInst operands, so wrap it in a convenient helper. More... | |
static bool | isCompatibleIVType (Value *LVal, Value *RVal) |
Return true if we allow an IV chain to include both types. More... | |
static const SCEV * | getExprBase (const SCEV *S) |
Return an approximation of this SCEV expression's "base", or NULL for any constant. More... | |
static bool | isProfitableChain (IVChain &Chain, SmallPtrSetImpl< Instruction * > &Users, ScalarEvolution &SE, const TargetTransformInfo &TTI) |
Return true if the number of registers needed for the chain is estimated to be less than the number required for the individual IV users. More... | |
static bool | canFoldIVIncExpr (const SCEV *IncExpr, Instruction *UserInst, Value *Operand, const TargetTransformInfo &TTI) |
Return true if the IVInc can be folded into an addressing mode. More... | |
static const SCEV * | CollectSubexprs (const SCEV *S, const SCEVConstant *C, SmallVectorImpl< const SCEV * > &Ops, const Loop *L, ScalarEvolution &SE, unsigned Depth=0) |
Split S into subexpressions which can be pulled out into separate registers. More... | |
static bool | mayUsePostIncMode (const TargetTransformInfo &TTI, LSRUse &LU, const SCEV *S, const Loop *L, ScalarEvolution &SE) |
Return true if the SCEV represents a value that may end up as a post-increment operation. More... | |
static bool | LoopExitValHasSingleUse (Loop *L) |
static unsigned | numLLVMArgOps (SmallVectorImpl< uint64_t > &Expr) |
Returns the total number of DW_OP_llvm_arg operands in the expression. More... | |
static void | updateDVIWithLocation (DbgValueInst &DVI, Value *Location, SmallVectorImpl< uint64_t > &Ops) |
Overwrites DVI with the location and Ops as the DIExpression. More... | |
static void | updateDVIWithLocations (DbgValueInst &DVI, SmallVectorImpl< Value * > &Locations, SmallVectorImpl< uint64_t > &Ops) |
Overwrite DVI with locations placed into a DIArglist. More... | |
static void | UpdateDbgValueInst (DVIRecoveryRec &DVIRec, SmallVectorImpl< Value * > &NewLocationOps, SmallVectorImpl< uint64_t > &NewExpr) |
Write the new expression and new location ops for the dbg.value. More... | |
static Value * | getValueOrUndef (WeakVH &VH, LLVMContext &C) |
Cached location ops may be erased during LSR, in which case an undef is required when restoring from the cache. More... | |
static void | restorePreTransformState (DVIRecoveryRec &DVIRec) |
Restore the DVI's pre-LSR arguments. Substitute undef for any erased values. More... | |
static bool | SalvageDVI (llvm::Loop *L, ScalarEvolution &SE, llvm::PHINode *LSRInductionVar, DVIRecoveryRec &DVIRec, const SCEV *SCEVInductionVar, SCEVDbgValueBuilder IterCountExpr) |
static void | DbgRewriteSalvageableDVIs (llvm::Loop *L, ScalarEvolution &SE, llvm::PHINode *LSRInductionVar, SmallVector< std::unique_ptr< DVIRecoveryRec >, 2 > &DVIToUpdate) |
Obtain an expression for the iteration count, then attempt to salvage the dbg.value intrinsics. More... | |
static void | DbgGatherSalvagableDVI (Loop *L, ScalarEvolution &SE, SmallVector< std::unique_ptr< DVIRecoveryRec >, 2 > &SalvageableDVISCEVs, SmallSet< AssertingVH< DbgValueInst >, 2 > &DVIHandles) |
Identify and cache salvageable DVI locations and expressions along with the corresponding SCEV(s). More... | |
static llvm::PHINode * | GetInductionVariable (const Loop &L, ScalarEvolution &SE, const LSRInstance &LSR) |
Ideally pick the PHI IV inserted by ScalarEvolutionExpander. More... | |
static bool | ReduceLoopStrength (Loop *L, IVUsers &IU, ScalarEvolution &SE, DominatorTree &DT, LoopInfo &LI, const TargetTransformInfo &TTI, AssumptionCache &AC, TargetLibraryInfo &TLI, MemorySSA *MSSA) |
INITIALIZE_PASS_BEGIN (LoopStrengthReduce, "loop-reduce", "Loop Strength Reduction", false, false) INITIALIZE_PASS_END(LoopStrengthReduce | |
Variables | |
static const unsigned | MaxIVUsers = 200 |
MaxIVUsers is an arbitrary threshold that provides an early opportunity for bail out. More... | |
static const unsigned | MaxSCEVSalvageExpressionSize = 64 |
Limit the size of expression that SCEV-based salvaging will attempt to translate into a DIExpression. More... | |
static cl::opt< bool > | EnablePhiElim ("enable-lsr-phielim", cl::Hidden, cl::init(true), cl::desc("Enable LSR phi elimination")) |
static cl::opt< bool > | InsnsCost ("lsr-insns-cost", cl::Hidden, cl::init(true), cl::desc("Add instruction count to a LSR cost model")) |
static cl::opt< bool > | LSRExpNarrow ("lsr-exp-narrow", cl::Hidden, cl::init(false), cl::desc("Narrow LSR complex solution using" " expectation of registers number")) |
static cl::opt< bool > | FilterSameScaledReg ("lsr-filter-same-scaled-reg", cl::Hidden, cl::init(true), cl::desc("Narrow LSR search space by filtering non-optimal formulae" " with the same ScaledReg and Scale")) |
static cl::opt< TTI::AddressingModeKind > | PreferredAddresingMode ("lsr-preferred-addressing-mode", cl::Hidden, cl::init(TTI::AMK_None), cl::desc("A flag that overrides the target's preferred addressing mode."), cl::values(clEnumValN(TTI::AMK_None, "none", "Don't prefer any addressing mode"), clEnumValN(TTI::AMK_PreIndexed, "preindexed", "Prefer pre-indexed addressing mode"), clEnumValN(TTI::AMK_PostIndexed, "postindexed", "Prefer post-indexed addressing mode"))) |
static cl::opt< unsigned > | ComplexityLimit ("lsr-complexity-limit", cl::Hidden, cl::init(std::numeric_limits< uint16_t >::max()), cl::desc("LSR search space complexity limit")) |
static cl::opt< unsigned > | SetupCostDepthLimit ("lsr-setupcost-depth-limit", cl::Hidden, cl::init(7), cl::desc("The limit on recursion depth for LSRs setup cost")) |
static cl::opt< bool > | StressIVChain ("stress-ivchain", cl::Hidden, cl::init(false), cl::desc("Stress test LSR IV chains")) |
loop | reduce |
loop Loop Strength | Reduction |
loop Loop Strength | false |
#define DEBUG_TYPE "loop-reduce" |
Definition at line 130 of file LoopStrengthReduce.cpp.
|
static |
Return true if the IVInc can be folded into an addressing mode.
Definition at line 3166 of file LoopStrengthReduce.cpp.
References getAccessType(), llvm::SCEVConstant::getAPInt(), llvm::APInt::getMinSignedBits(), llvm::ConstantInt::getSExtValue(), llvm::SCEVConstant::getValue(), isAddressUse(), and isAlwaysFoldable().
|
static |
Split S into subexpressions which can be pulled out into separate registers.
If C is non-null, multiply each subexpression by C.
Return remainder expression after factoring the subexpressions captured by Ops. If Ops is complete, return NULL.
Definition at line 3564 of file LoopStrengthReduce.cpp.
References llvm::Depth, llvm::SCEV::FlagAnyWrap, llvm::ScalarEvolution::getAddRecExpr(), llvm::ScalarEvolution::getConstant(), llvm::SCEVAddRecExpr::getLoop(), llvm::ScalarEvolution::getMulExpr(), llvm::User::getNumOperands(), llvm::User::getOperand(), llvm::SCEVAddRecExpr::getStart(), llvm::SCEVAddRecExpr::getStepRecurrence(), llvm::SCEVAddRecExpr::getType(), llvm::SCEVAddRecExpr::isAffine(), llvm::SCEV::isZero(), Mul, and S.
Definition at line 480 of file LoopStrengthReduce.cpp.
References S, and llvm::SCEVExprContains().
|
static |
Identify and cache salvageable DVI locations and expressions along with the corresponding SCEV(s).
Also ensure that the DVI is not deleted between cacheing and salvaging.
Definition at line 6499 of file LoopStrengthReduce.cpp.
References B, llvm::ScalarEvolution::containsUndefs(), llvm::LoopBase< BlockT, LoopT >::getBlocks(), llvm::DbgVariableIntrinsic::getNumVariableLocationOps(), llvm::ScalarEvolution::getSCEV(), llvm::DbgVariableIntrinsic::hasArgList(), I, llvm::ScalarEvolution::isSCEVable(), llvm::DbgVariableIntrinsic::isUndef(), llvm::DbgVariableIntrinsic::location_ops(), move, and S.
Referenced by ReduceLoopStrength().
|
static |
Obtain an expression for the iteration count, then attempt to salvage the dbg.value intrinsics.
Definition at line 6461 of file LoopStrengthReduce.cpp.
References assert(), llvm::dbgs(), llvm::ScalarEvolution::getSCEV(), LLVM_DEBUG, MaxSCEVSalvageExpressionSize, and SalvageDVI().
Referenced by ReduceLoopStrength().
|
static |
Recursion helper for initialMatch.
Definition at line 407 of file LoopStrengthReduce.cpp.
References llvm::drop_begin(), llvm::SCEV::FlagAnyWrap, llvm::ScalarEvolution::getAddRecExpr(), llvm::Constant::getAllOnesValue(), llvm::ScalarEvolution::getConstant(), llvm::ScalarEvolution::getEffectiveSCEVType(), llvm::LoopBase< BlockT, LoopT >::getHeader(), llvm::ScalarEvolution::getMulExpr(), llvm::User::getOperand(), llvm::ScalarEvolution::getSCEV(), llvm::SCEV::getType(), Mul, llvm::User::operands(), llvm::ScalarEvolution::properlyDominates(), and S.
|
static |
If S involves the addition of a constant integer value, return that integer value, and mutate S to point to a new SCEV with that value excluded.
Definition at line 787 of file LoopStrengthReduce.cpp.
References llvm::SCEV::FlagAnyWrap, llvm::ScalarEvolution::getAddExpr(), llvm::ScalarEvolution::getAddRecExpr(), llvm::ScalarEvolution::getConstant(), and S.
Referenced by isAlwaysFoldable().
|
static |
If S involves the addition of a GlobalValue address, return that symbol, and mutate S to point to a new SCEV with that value excluded.
Definition at line 813 of file LoopStrengthReduce.cpp.
References llvm::SCEV::FlagAnyWrap, llvm::ScalarEvolution::getAddExpr(), llvm::ScalarEvolution::getAddRecExpr(), llvm::ScalarEvolution::getConstant(), and S.
Referenced by isAlwaysFoldable().
|
static |
Helper for CollectChains that finds an IV operand (computed by an AddRec in this loop) within [OI,OE) or returns OE.
If IVUsers mapped Instructions to IVStrideUses, we could partially skip this.
Definition at line 2760 of file LoopStrengthReduce.cpp.
References llvm::SCEVAddRecExpr::getLoop(), llvm::ScalarEvolution::getSCEV(), and llvm::ScalarEvolution::isSCEVable().
|
static |
Return the type of the memory being accessed.
Definition at line 884 of file LoopStrengthReduce.cpp.
References llvm::IntegerType::get(), llvm::PointerType::get(), llvm::Type::getPointerAddressSpace(), llvm::TargetTransformInfo::getTgtMemIntrinsic(), llvm::Value::getType(), memcpy(), prefetch, llvm::MemIntrinsicInfo::PtrVal, SI, and UnknownAddressSpace.
Referenced by canFoldIVIncExpr().
|
static |
Return an expression for LHS /s RHS, if it can be determined and if the remainder is known to be zero, or null otherwise.
If IgnoreSignificantBits is true, expressions like (X * Y) /s Y are simplified to X, ignoring that the multiplication may overflow, which is useful when the result will be used in a context where the most significant bits are ignored.
Definition at line 682 of file LoopStrengthReduce.cpp.
References llvm::drop_begin(), llvm::SCEV::FlagAnyWrap, llvm::ScalarEvolution::getAddExpr(), llvm::ScalarEvolution::getAddRecExpr(), llvm::SCEVConstant::getAPInt(), llvm::ScalarEvolution::getConstant(), llvm::ScalarEvolution::getMulExpr(), llvm::User::getOperand(), llvm::Value::getType(), isAddRecSExtable(), isAddSExtable(), isMulSExtable(), llvm::Type::isPointerTy(), llvm::RISCVISD::LA, LHS, Mul, llvm::User::operands(), RA, RHS, and S.
Return an approximation of this SCEV expression's "base", or NULL for any constant.
Returning the expression itself is conservative. Returning a deeper subexpression is more precise and valid as long as it isn't less complex than another subexpression. For expressions involving multiple unscaled values, we need to return the pointer-type SCEVUnknown. This avoids forming chains across objects, such as: PrevOper==a[i], IVOper==b[i], IVInc==b-a.
Since SCEVUnknown is the rightmost type, and pointers are the rightmost SCEVUnknown, we simply return the rightmost SCEV operand.
Definition at line 2807 of file LoopStrengthReduce.cpp.
References llvm_unreachable, llvm::reverse(), S, llvm::scAddExpr, llvm::scAddRecExpr, llvm::scConstant, llvm::scMulExpr, llvm::scSignExtend, llvm::scTruncate, and llvm::scZeroExtend.
|
static |
Ideally pick the PHI IV inserted by ScalarEvolutionExpander.
As a fallback any PHi from the loop header is usable, but may have less chance of surviving subsequent transforms.
Definition at line 6554 of file LoopStrengthReduce.cpp.
References llvm::ScalarEvolution::containsUndefs(), llvm::LoopBase< BlockT, LoopT >::getHeader(), llvm::ScalarEvolution::getSCEV(), llvm::ScalarEvolution::isSCEVable(), IV, llvm::AArch64_AM::LSR, P, and llvm::BasicBlock::phis().
Referenced by ReduceLoopStrength().
|
static |
Definition at line 1791 of file LoopStrengthReduce.cpp.
References assert(), llvm::codeview::Basic, F, llvm::TargetTransformInfo::getScalingFactorCost(), isAMCompletelyFolded(), llvm::InstructionCost::isValid(), llvm_unreachable, and llvm::max().
Definition at line 1248 of file LoopStrengthReduce.cpp.
References llvm::Depth, i, and S.
|
static |
Cached location ops may be erased during LSR, in which case an undef is required when restoring from the cache.
The type of that location is no longer available, so just use int8. The undef will be replaced by one or more locations later when a SCEVDbgValueBuilder selects alternative locations to use for the salvage.
Definition at line 6321 of file LoopStrengthReduce.cpp.
References get, and llvm::Type::getInt8Ty().
Referenced by restorePreTransformState().
IVChain logic must consistently peek base TruncInst operands, so wrap it in a convenient helper.
Definition at line 2779 of file LoopStrengthReduce.cpp.
INITIALIZE_PASS_BEGIN | ( | LoopStrengthReduce | , |
"loop-reduce" | , | ||
"Loop Strength Reduction" | , | ||
false | , | ||
false | |||
) |
|
static |
Return true if the given addrec can be sign-extended without changing its value.
Definition at line 654 of file LoopStrengthReduce.cpp.
References llvm::IntegerType::get(), llvm::ScalarEvolution::getContext(), llvm::ScalarEvolution::getSignExtendExpr(), llvm::SCEVAddRecExpr::getType(), and llvm::ScalarEvolution::getTypeSizeInBits().
Referenced by getExactSDiv().
|
static |
Returns true if the specified instruction is using the specified value as an address.
Definition at line 839 of file LoopStrengthReduce.cpp.
References llvm::TargetTransformInfo::getTgtMemIntrinsic(), memcpy(), prefetch, llvm::MemIntrinsicInfo::PtrVal, and SI.
Referenced by canFoldIVIncExpr().
|
static |
Return true if the given add can be sign-extended without changing its value.
Definition at line 662 of file LoopStrengthReduce.cpp.
References llvm::IntegerType::get(), llvm::ScalarEvolution::getContext(), llvm::ScalarEvolution::getSignExtendExpr(), and llvm::ScalarEvolution::getTypeSizeInBits().
Referenced by getExactSDiv().
|
static |
Definition at line 1828 of file LoopStrengthReduce.cpp.
References isAMCompletelyFolded().
Referenced by canFoldIVIncExpr().
|
static |
Definition at line 1850 of file LoopStrengthReduce.cpp.
References ExtractImmediate(), ExtractSymbol(), isAMCompletelyFolded(), and S.
|
static |
Check if the addressing mode defined by F
is completely folded in LU
at isel time.
This includes address-mode folding and special icmp tricks. This function returns true if LU
can accommodate what F
defines and up to 1 base + 1 scaled + offset. In other words, if F
has several base registers, this function may still return true. Therefore, users still need to account for additional base registers and/or unfolded offsets to derive an accurate cost model.
Definition at line 1774 of file LoopStrengthReduce.cpp.
References F, Fixup, and llvm::TargetTransformInfo::LSRWithInstrQueries().
Referenced by getScalingFactorCost(), isAlwaysFoldable(), isAMCompletelyFolded(), and isLegalUse().
|
static |
Definition at line 1736 of file LoopStrengthReduce.cpp.
References assert(), F, and isAMCompletelyFolded().
|
static |
Definition at line 1715 of file LoopStrengthReduce.cpp.
References isAMCompletelyFolded().
|
static |
Definition at line 1661 of file LoopStrengthReduce.cpp.
References llvm::codeview::Basic, Fixup, llvm::TargetTransformInfo::isLegalAddressingMode(), llvm::TargetTransformInfo::isLegalICmpImmediate(), and llvm_unreachable.
Return true if we allow an IV chain to include both types.
Definition at line 2786 of file LoopStrengthReduce.cpp.
References llvm::Type::getPointerAddressSpace(), llvm::Value::getType(), and llvm::Type::isPointerTy().
|
static |
Return true if this AddRec is already a phi in its loop.
Definition at line 939 of file LoopStrengthReduce.cpp.
References llvm::ScalarEvolution::getEffectiveSCEVType(), llvm::LoopBase< BlockT, LoopT >::getHeader(), llvm::SCEVAddRecExpr::getLoop(), llvm::ScalarEvolution::getSCEV(), llvm::SCEVAddRecExpr::getType(), llvm::ScalarEvolution::isSCEVable(), and llvm::BasicBlock::phis().
Referenced by isHighCostExpansion().
|
static |
Check if expanding this expression is likely to incur significant cost.
This is tricky because SCEV doesn't track which expressions are actually computed by the current IR.
We currently allow expansion of IV increments that involve adds, multiplication by constants, and AddRecs from existing phis.
TODO: Allow UDivExpr if we can find an existing IV increment that is an obvious multiple of the UDivExpr.
Definition at line 959 of file LoopStrengthReduce.cpp.
References llvm::User::getNumOperands(), llvm::Instruction::getOpcode(), llvm::User::getOperand(), llvm::ScalarEvolution::getSCEV(), llvm::Value::getType(), llvm::SmallPtrSetImpl< PtrType >::insert(), isExistingPhi(), llvm::ScalarEvolution::isSCEVable(), Mul, S, llvm::scConstant, llvm::scSignExtend, llvm::scTruncate, llvm::scUnknown, llvm::scZeroExtend, and llvm::Value::users().
|
static |
Definition at line 1767 of file LoopStrengthReduce.cpp.
References F, and isLegalUse().
|
static |
Test whether we know how to expand the current formula.
Definition at line 1753 of file LoopStrengthReduce.cpp.
References isAMCompletelyFolded().
Referenced by isLegalUse().
|
static |
Return true if the given mul can be sign-extended without changing its value.
Definition at line 670 of file LoopStrengthReduce.cpp.
References llvm::IntegerType::get(), llvm::ScalarEvolution::getContext(), llvm::ScalarEvolution::getSignExtendExpr(), llvm::ScalarEvolution::getTypeSizeInBits(), and M.
Referenced by getExactSDiv().
|
static |
Return true if the number of registers needed for the chain is estimated to be less than the number required for the individual IV users.
First prohibit any IV users that keep the IV live across increments (the Users set should be empty). Next count the number and type of increments in the chain.
Chaining IVs can lead to considerable code bloat if ISEL doesn't effectively use postinc addressing modes. Only consider it profitable it the increments can be computed in fewer registers when chained.
TODO: Consider IVInc free if it's already used in another chains.
Definition at line 2873 of file LoopStrengthReduce.cpp.
References assert(), llvm::dbgs(), for, llvm::ScalarEvolution::getSCEV(), llvm::TargetTransformInfo::isProfitableLSRChainElement(), LLVM_DEBUG, StressIVChain, and Users.
|
static |
Definition at line 5606 of file LoopStrengthReduce.cpp.
References llvm::LoopBase< BlockT, LoopT >::getExitBlock(), llvm::Value::getNumUses(), and llvm::BasicBlock::phis().
Referenced by ReduceLoopStrength().
|
static |
Return true if the SCEV represents a value that may end up as a post-increment operation.
Definition at line 3622 of file LoopStrengthReduce.cpp.
References llvm::SCEVAddRecExpr::getStart(), llvm::SCEVAddRecExpr::getStepRecurrence(), llvm::SCEVAddRecExpr::getType(), llvm::TargetTransformInfo::isIndexedLoadLegal(), llvm::TargetTransformInfo::isIndexedStoreLegal(), llvm::ScalarEvolution::isLoopInvariant(), llvm::TargetTransformInfo::MIM_PostInc, and S.
|
static |
Returns the total number of DW_OP_llvm_arg operands in the expression.
This helps in determining if a DIArglist is necessary or can be omitted from the dbg.value.
Definition at line 6246 of file LoopStrengthReduce.cpp.
References llvm::dwarf::DW_OP_LLVM_arg.
Referenced by UpdateDbgValueInst(), updateDVIWithLocation(), and updateDVIWithLocations().
|
static |
Definition at line 6586 of file LoopStrengthReduce.cpp.
References llvm::SmallSet< T, N, C >::clear(), DbgGatherSalvagableDVI(), DbgRewriteSalvageableDVIs(), llvm::dbgs(), DEBUG_TYPE, llvm::DeleteDeadPHIs(), DL, EnablePhiElim, llvm::Module::getDataLayout(), llvm::LoopBase< BlockT, LoopT >::getHeader(), GetInductionVariable(), llvm::BasicBlock::getModule(), llvm::Loop::isLoopSimplifyForm(), llvm::Loop::isRecursivelyLCSSAForm(), IV, LLVM_DEBUG, LoopExitValHasSingleUse(), llvm::OnlyCheapRepl, llvm::RecursivelyDeleteTriviallyDeadInstructionsPermissive(), llvm::rewriteLoopExitValues(), and Rewriter.
Referenced by llvm::LoopStrengthReducePass::run().
|
static |
Restore the DVI's pre-LSR arguments. Substitute undef for any erased values.
Definition at line 6326 of file LoopStrengthReduce.cpp.
References assert(), llvm::dbgs(), get, llvm::MDNode::get(), getValueOrUndef(), and LLVM_DEBUG.
Referenced by SalvageDVI().
|
static |
Definition at line 6356 of file LoopStrengthReduce.cpp.
References assert(), llvm::SmallVectorImpl< T >::assign(), B, llvm::ScalarEvolution::computeConstantDifference(), llvm::ScalarEvolution::containsErasedValue(), llvm::ScalarEvolution::containsUndefs(), llvm::dbgs(), llvm::dwarf::DW_OP_LLVM_arg, i, LLVM_DEBUG, Offset, restorePreTransformState(), and UpdateDbgValueInst().
Referenced by DbgRewriteSalvageableDVIs().
|
static |
Write the new expression and new location ops for the dbg.value.
If possible reduce the szie of the dbg.value intrinsic by omitting DIArglist. This can be omitted if:
Definition at line 6287 of file LoopStrengthReduce.cpp.
References llvm::sys::path::append(), assert(), llvm::drop_begin(), llvm::dwarf::DW_OP_LLVM_arg, llvm::DIExpression::isComplex(), numLLVMArgOps(), updateDVIWithLocation(), and updateDVIWithLocations().
Referenced by SalvageDVI().
|
static |
Overwrites DVI with the location and Ops as the DIExpression.
This will create an invalid expression if Ops has any dwarf::DW_OP_llvm_arg operands, because a DIArglist is not created for the first argument of the dbg.value.
Definition at line 6258 of file LoopStrengthReduce.cpp.
References assert(), get, llvm::Value::getContext(), numLLVMArgOps(), llvm::DbgVariableIntrinsic::setExpression(), and llvm::DbgVariableIntrinsic::setRawLocation().
Referenced by UpdateDbgValueInst().
|
static |
Overwrite DVI with locations placed into a DIArglist.
Definition at line 6268 of file LoopStrengthReduce.cpp.
References assert(), get, llvm::MDNode::get(), llvm::Value::getContext(), numLLVMArgOps(), llvm::DbgVariableIntrinsic::setExpression(), and llvm::DbgVariableIntrinsic::setRawLocation().
Referenced by UpdateDbgValueInst().
|
static |
|
static |
Referenced by ReduceLoopStrength().
Definition at line 6711 of file LoopStrengthReduce.cpp.
|
static |
|
static |
|
static |
|
static |
MaxIVUsers is an arbitrary threshold that provides an early opportunity for bail out.
This threshold is far beyond the number of users that LSR can conceivably solve, so it should not affect generated code, but catches the worst cases before LSR burns too much compile time and stack space.
Definition at line 136 of file LoopStrengthReduce.cpp.
|
static |
Limit the size of expression that SCEV-based salvaging will attempt to translate into a DIExpression.
Choose a maximum size such that debuginfo is not excessively increased and the salvaging is not too expensive for the compiler.
Definition at line 142 of file LoopStrengthReduce.cpp.
Referenced by DbgRewriteSalvageableDVIs().
|
static |
loop reduce |
Definition at line 6710 of file LoopStrengthReduce.cpp.
Definition at line 6711 of file LoopStrengthReduce.cpp.
Referenced by llvm::LoopVectorizationLegality::canVectorizeFPMath(), llvm::LoopVectorizationCostModel::canVectorizeReductions(), llvm::LoopVectorizationCostModel::collectInLoopReductions(), llvm::LoopVectorizationCostModel::collectValuesToIgnore(), llvm::LoopVectorizationLegality::isInvariantAddressOfReduction(), llvm::LoopVectorizationLegality::isInvariantStoreOfReduction(), PerformSELECTCombine(), llvm::LoopVectorizationLegality::prepareToFoldTailByMasking(), and llvm::LoopVectorizationCostModel::selectInterleaveCount().
|
static |