1 //===- InlineCost.cpp - Cost analysis for inliner -------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements inline cost analysis.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/Analysis/InlineCost.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/ADT/SetVector.h"
17 #include "llvm/ADT/SmallPtrSet.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/ADT/Statistic.h"
20 #include "llvm/Analysis/AssumptionTracker.h"
21 #include "llvm/Analysis/ConstantFolding.h"
22 #include "llvm/Analysis/CodeMetrics.h"
23 #include "llvm/Analysis/InstructionSimplify.h"
24 #include "llvm/Analysis/TargetTransformInfo.h"
25 #include "llvm/IR/CallSite.h"
26 #include "llvm/IR/CallingConv.h"
27 #include "llvm/IR/DataLayout.h"
28 #include "llvm/IR/GetElementPtrTypeIterator.h"
29 #include "llvm/IR/GlobalAlias.h"
30 #include "llvm/IR/InstVisitor.h"
31 #include "llvm/IR/IntrinsicInst.h"
32 #include "llvm/IR/Operator.h"
33 #include "llvm/Support/Debug.h"
34 #include "llvm/Support/raw_ostream.h"
38 #define DEBUG_TYPE "inline-cost"
40 STATISTIC(NumCallsAnalyzed, "Number of call sites analyzed");
44 class CallAnalyzer : public InstVisitor<CallAnalyzer, bool> {
45 typedef InstVisitor<CallAnalyzer, bool> Base;
46 friend class InstVisitor<CallAnalyzer, bool>;
48 // DataLayout if available, or null.
49 const DataLayout *const DL;
51 /// The TargetTransformInfo available for this compilation.
52 const TargetTransformInfo &TTI;
54 /// The cache of @llvm.assume intrinsics.
55 AssumptionTracker *AT;
57 // The called function.
63 bool IsCallerRecursive;
65 bool ExposesReturnsTwice;
66 bool HasDynamicAlloca;
67 bool ContainsNoDuplicateCall;
71 /// Number of bytes allocated statically by the callee.
72 uint64_t AllocatedSize;
73 unsigned NumInstructions, NumVectorInstructions;
74 int FiftyPercentVectorBonus, TenPercentVectorBonus;
77 // While we walk the potentially-inlined instructions, we build up and
78 // maintain a mapping of simplified values specific to this callsite. The
79 // idea is to propagate any special information we have about arguments to
80 // this call through the inlinable section of the function, and account for
81 // likely simplifications post-inlining. The most important aspect we track
82 // is CFG altering simplifications -- when we prove a basic block dead, that
83 // can cause dramatic shifts in the cost of inlining a function.
84 DenseMap<Value *, Constant *> SimplifiedValues;
86 // Keep track of the values which map back (through function arguments) to
87 // allocas on the caller stack which could be simplified through SROA.
88 DenseMap<Value *, Value *> SROAArgValues;
90 // The mapping of caller Alloca values to their accumulated cost savings. If
91 // we have to disable SROA for one of the allocas, this tells us how much
92 // cost must be added.
93 DenseMap<Value *, int> SROAArgCosts;
95 // Keep track of values which map to a pointer base and constant offset.
96 DenseMap<Value *, std::pair<Value *, APInt> > ConstantOffsetPtrs;
98 // Custom simplification helper routines.
99 bool isAllocaDerivedArg(Value *V);
100 bool lookupSROAArgAndCost(Value *V, Value *&Arg,
101 DenseMap<Value *, int>::iterator &CostIt);
102 void disableSROA(DenseMap<Value *, int>::iterator CostIt);
103 void disableSROA(Value *V);
104 void accumulateSROACost(DenseMap<Value *, int>::iterator CostIt,
105 int InstructionCost);
106 bool isGEPOffsetConstant(GetElementPtrInst &GEP);
107 bool accumulateGEPOffset(GEPOperator &GEP, APInt &Offset);
108 bool simplifyCallSite(Function *F, CallSite CS);
109 ConstantInt *stripAndComputeInBoundsConstantOffsets(Value *&V);
111 // Custom analysis routines.
112 bool analyzeBlock(BasicBlock *BB, SmallPtrSetImpl<const Value *> &EphValues);
114 // Disable several entry points to the visitor so we don't accidentally use
115 // them by declaring but not defining them here.
116 void visit(Module *); void visit(Module &);
117 void visit(Function *); void visit(Function &);
118 void visit(BasicBlock *); void visit(BasicBlock &);
120 // Provide base case for our instruction visit.
121 bool visitInstruction(Instruction &I);
123 // Our visit overrides.
124 bool visitAlloca(AllocaInst &I);
125 bool visitPHI(PHINode &I);
126 bool visitGetElementPtr(GetElementPtrInst &I);
127 bool visitBitCast(BitCastInst &I);
128 bool visitPtrToInt(PtrToIntInst &I);
129 bool visitIntToPtr(IntToPtrInst &I);
130 bool visitCastInst(CastInst &I);
131 bool visitUnaryInstruction(UnaryInstruction &I);
132 bool visitCmpInst(CmpInst &I);
133 bool visitSub(BinaryOperator &I);
134 bool visitBinaryOperator(BinaryOperator &I);
135 bool visitLoad(LoadInst &I);
136 bool visitStore(StoreInst &I);
137 bool visitExtractValue(ExtractValueInst &I);
138 bool visitInsertValue(InsertValueInst &I);
139 bool visitCallSite(CallSite CS);
140 bool visitReturnInst(ReturnInst &RI);
141 bool visitBranchInst(BranchInst &BI);
142 bool visitSwitchInst(SwitchInst &SI);
143 bool visitIndirectBrInst(IndirectBrInst &IBI);
144 bool visitResumeInst(ResumeInst &RI);
145 bool visitUnreachableInst(UnreachableInst &I);
148 CallAnalyzer(const DataLayout *DL, const TargetTransformInfo &TTI,
149 AssumptionTracker *AT, Function &Callee, int Threshold)
150 : DL(DL), TTI(TTI), AT(AT), F(Callee), Threshold(Threshold), Cost(0),
151 IsCallerRecursive(false), IsRecursiveCall(false),
152 ExposesReturnsTwice(false), HasDynamicAlloca(false),
153 ContainsNoDuplicateCall(false), HasReturn(false), HasIndirectBr(false),
154 AllocatedSize(0), NumInstructions(0), NumVectorInstructions(0),
155 FiftyPercentVectorBonus(0), TenPercentVectorBonus(0), VectorBonus(0),
156 NumConstantArgs(0), NumConstantOffsetPtrArgs(0), NumAllocaArgs(0),
157 NumConstantPtrCmps(0), NumConstantPtrDiffs(0),
158 NumInstructionsSimplified(0), SROACostSavings(0),
159 SROACostSavingsLost(0) {}
161 bool analyzeCall(CallSite CS);
163 int getThreshold() { return Threshold; }
164 int getCost() { return Cost; }
166 // Keep a bunch of stats about the cost savings found so we can print them
167 // out when debugging.
168 unsigned NumConstantArgs;
169 unsigned NumConstantOffsetPtrArgs;
170 unsigned NumAllocaArgs;
171 unsigned NumConstantPtrCmps;
172 unsigned NumConstantPtrDiffs;
173 unsigned NumInstructionsSimplified;
174 unsigned SROACostSavings;
175 unsigned SROACostSavingsLost;
182 /// \brief Test whether the given value is an Alloca-derived function argument.
183 bool CallAnalyzer::isAllocaDerivedArg(Value *V) {
184 return SROAArgValues.count(V);
187 /// \brief Lookup the SROA-candidate argument and cost iterator which V maps to.
188 /// Returns false if V does not map to a SROA-candidate.
189 bool CallAnalyzer::lookupSROAArgAndCost(
190 Value *V, Value *&Arg, DenseMap<Value *, int>::iterator &CostIt) {
191 if (SROAArgValues.empty() || SROAArgCosts.empty())
194 DenseMap<Value *, Value *>::iterator ArgIt = SROAArgValues.find(V);
195 if (ArgIt == SROAArgValues.end())
199 CostIt = SROAArgCosts.find(Arg);
200 return CostIt != SROAArgCosts.end();
203 /// \brief Disable SROA for the candidate marked by this cost iterator.
205 /// This marks the candidate as no longer viable for SROA, and adds the cost
206 /// savings associated with it back into the inline cost measurement.
207 void CallAnalyzer::disableSROA(DenseMap<Value *, int>::iterator CostIt) {
208 // If we're no longer able to perform SROA we need to undo its cost savings
209 // and prevent subsequent analysis.
210 Cost += CostIt->second;
211 SROACostSavings -= CostIt->second;
212 SROACostSavingsLost += CostIt->second;
213 SROAArgCosts.erase(CostIt);
216 /// \brief If 'V' maps to a SROA candidate, disable SROA for it.
217 void CallAnalyzer::disableSROA(Value *V) {
219 DenseMap<Value *, int>::iterator CostIt;
220 if (lookupSROAArgAndCost(V, SROAArg, CostIt))
224 /// \brief Accumulate the given cost for a particular SROA candidate.
225 void CallAnalyzer::accumulateSROACost(DenseMap<Value *, int>::iterator CostIt,
226 int InstructionCost) {
227 CostIt->second += InstructionCost;
228 SROACostSavings += InstructionCost;
231 /// \brief Check whether a GEP's indices are all constant.
233 /// Respects any simplified values known during the analysis of this callsite.
234 bool CallAnalyzer::isGEPOffsetConstant(GetElementPtrInst &GEP) {
235 for (User::op_iterator I = GEP.idx_begin(), E = GEP.idx_end(); I != E; ++I)
236 if (!isa<Constant>(*I) && !SimplifiedValues.lookup(*I))
242 /// \brief Accumulate a constant GEP offset into an APInt if possible.
244 /// Returns false if unable to compute the offset for any reason. Respects any
245 /// simplified values known during the analysis of this callsite.
246 bool CallAnalyzer::accumulateGEPOffset(GEPOperator &GEP, APInt &Offset) {
250 unsigned IntPtrWidth = DL->getPointerSizeInBits();
251 assert(IntPtrWidth == Offset.getBitWidth());
253 for (gep_type_iterator GTI = gep_type_begin(GEP), GTE = gep_type_end(GEP);
255 ConstantInt *OpC = dyn_cast<ConstantInt>(GTI.getOperand());
257 if (Constant *SimpleOp = SimplifiedValues.lookup(GTI.getOperand()))
258 OpC = dyn_cast<ConstantInt>(SimpleOp);
261 if (OpC->isZero()) continue;
263 // Handle a struct index, which adds its field offset to the pointer.
264 if (StructType *STy = dyn_cast<StructType>(*GTI)) {
265 unsigned ElementIdx = OpC->getZExtValue();
266 const StructLayout *SL = DL->getStructLayout(STy);
267 Offset += APInt(IntPtrWidth, SL->getElementOffset(ElementIdx));
271 APInt TypeSize(IntPtrWidth, DL->getTypeAllocSize(GTI.getIndexedType()));
272 Offset += OpC->getValue().sextOrTrunc(IntPtrWidth) * TypeSize;
277 bool CallAnalyzer::visitAlloca(AllocaInst &I) {
278 // Check whether inlining will turn a dynamic alloca into a static
279 // alloca, and handle that case.
280 if (I.isArrayAllocation()) {
281 if (Constant *Size = SimplifiedValues.lookup(I.getArraySize())) {
282 ConstantInt *AllocSize = dyn_cast<ConstantInt>(Size);
283 assert(AllocSize && "Allocation size not a constant int?");
284 Type *Ty = I.getAllocatedType();
285 AllocatedSize += Ty->getPrimitiveSizeInBits() * AllocSize->getZExtValue();
286 return Base::visitAlloca(I);
290 // Accumulate the allocated size.
291 if (I.isStaticAlloca()) {
292 Type *Ty = I.getAllocatedType();
293 AllocatedSize += (DL ? DL->getTypeAllocSize(Ty) :
294 Ty->getPrimitiveSizeInBits());
297 // We will happily inline static alloca instructions.
298 if (I.isStaticAlloca())
299 return Base::visitAlloca(I);
301 // FIXME: This is overly conservative. Dynamic allocas are inefficient for
302 // a variety of reasons, and so we would like to not inline them into
303 // functions which don't currently have a dynamic alloca. This simply
304 // disables inlining altogether in the presence of a dynamic alloca.
305 HasDynamicAlloca = true;
309 bool CallAnalyzer::visitPHI(PHINode &I) {
310 // FIXME: We should potentially be tracking values through phi nodes,
311 // especially when they collapse to a single value due to deleted CFG edges
314 // FIXME: We need to propagate SROA *disabling* through phi nodes, even
315 // though we don't want to propagate it's bonuses. The idea is to disable
316 // SROA if it *might* be used in an inappropriate manner.
318 // Phi nodes are always zero-cost.
322 bool CallAnalyzer::visitGetElementPtr(GetElementPtrInst &I) {
324 DenseMap<Value *, int>::iterator CostIt;
325 bool SROACandidate = lookupSROAArgAndCost(I.getPointerOperand(),
328 // Try to fold GEPs of constant-offset call site argument pointers. This
329 // requires target data and inbounds GEPs.
330 if (DL && I.isInBounds()) {
331 // Check if we have a base + offset for the pointer.
332 Value *Ptr = I.getPointerOperand();
333 std::pair<Value *, APInt> BaseAndOffset = ConstantOffsetPtrs.lookup(Ptr);
334 if (BaseAndOffset.first) {
335 // Check if the offset of this GEP is constant, and if so accumulate it
337 if (!accumulateGEPOffset(cast<GEPOperator>(I), BaseAndOffset.second)) {
338 // Non-constant GEPs aren't folded, and disable SROA.
344 // Add the result as a new mapping to Base + Offset.
345 ConstantOffsetPtrs[&I] = BaseAndOffset;
347 // Also handle SROA candidates here, we already know that the GEP is
348 // all-constant indexed.
350 SROAArgValues[&I] = SROAArg;
356 if (isGEPOffsetConstant(I)) {
358 SROAArgValues[&I] = SROAArg;
360 // Constant GEPs are modeled as free.
364 // Variable GEPs will require math and will disable SROA.
370 bool CallAnalyzer::visitBitCast(BitCastInst &I) {
371 // Propagate constants through bitcasts.
372 Constant *COp = dyn_cast<Constant>(I.getOperand(0));
374 COp = SimplifiedValues.lookup(I.getOperand(0));
376 if (Constant *C = ConstantExpr::getBitCast(COp, I.getType())) {
377 SimplifiedValues[&I] = C;
381 // Track base/offsets through casts
382 std::pair<Value *, APInt> BaseAndOffset
383 = ConstantOffsetPtrs.lookup(I.getOperand(0));
384 // Casts don't change the offset, just wrap it up.
385 if (BaseAndOffset.first)
386 ConstantOffsetPtrs[&I] = BaseAndOffset;
388 // Also look for SROA candidates here.
390 DenseMap<Value *, int>::iterator CostIt;
391 if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt))
392 SROAArgValues[&I] = SROAArg;
394 // Bitcasts are always zero cost.
398 bool CallAnalyzer::visitPtrToInt(PtrToIntInst &I) {
399 const DataLayout *DL = I.getDataLayout();
400 // Propagate constants through ptrtoint.
401 Constant *COp = dyn_cast<Constant>(I.getOperand(0));
403 COp = SimplifiedValues.lookup(I.getOperand(0));
405 if (Constant *C = ConstantExpr::getPtrToInt(COp, I.getType())) {
406 SimplifiedValues[&I] = C;
410 // Track base/offset pairs when converted to a plain integer provided the
411 // integer is large enough to represent the pointer.
412 unsigned IntegerSize = I.getType()->getScalarSizeInBits();
413 if (DL && IntegerSize >= DL->getPointerSizeInBits()) {
414 std::pair<Value *, APInt> BaseAndOffset
415 = ConstantOffsetPtrs.lookup(I.getOperand(0));
416 if (BaseAndOffset.first)
417 ConstantOffsetPtrs[&I] = BaseAndOffset;
420 // This is really weird. Technically, ptrtoint will disable SROA. However,
421 // unless that ptrtoint is *used* somewhere in the live basic blocks after
422 // inlining, it will be nuked, and SROA should proceed. All of the uses which
423 // would block SROA would also block SROA if applied directly to a pointer,
424 // and so we can just add the integer in here. The only places where SROA is
425 // preserved either cannot fire on an integer, or won't in-and-of themselves
426 // disable SROA (ext) w/o some later use that we would see and disable.
428 DenseMap<Value *, int>::iterator CostIt;
429 if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt))
430 SROAArgValues[&I] = SROAArg;
432 return TargetTransformInfo::TCC_Free == TTI.getUserCost(&I);
435 bool CallAnalyzer::visitIntToPtr(IntToPtrInst &I) {
436 const DataLayout *DL = I.getDataLayout();
437 // Propagate constants through ptrtoint.
438 Constant *COp = dyn_cast<Constant>(I.getOperand(0));
440 COp = SimplifiedValues.lookup(I.getOperand(0));
442 if (Constant *C = ConstantExpr::getIntToPtr(COp, I.getType())) {
443 SimplifiedValues[&I] = C;
447 // Track base/offset pairs when round-tripped through a pointer without
448 // modifications provided the integer is not too large.
449 Value *Op = I.getOperand(0);
450 unsigned IntegerSize = Op->getType()->getScalarSizeInBits();
451 if (DL && IntegerSize <= DL->getPointerSizeInBits()) {
452 std::pair<Value *, APInt> BaseAndOffset = ConstantOffsetPtrs.lookup(Op);
453 if (BaseAndOffset.first)
454 ConstantOffsetPtrs[&I] = BaseAndOffset;
457 // "Propagate" SROA here in the same manner as we do for ptrtoint above.
459 DenseMap<Value *, int>::iterator CostIt;
460 if (lookupSROAArgAndCost(Op, SROAArg, CostIt))
461 SROAArgValues[&I] = SROAArg;
463 return TargetTransformInfo::TCC_Free == TTI.getUserCost(&I);
466 bool CallAnalyzer::visitCastInst(CastInst &I) {
467 // Propagate constants through ptrtoint.
468 Constant *COp = dyn_cast<Constant>(I.getOperand(0));
470 COp = SimplifiedValues.lookup(I.getOperand(0));
472 if (Constant *C = ConstantExpr::getCast(I.getOpcode(), COp, I.getType())) {
473 SimplifiedValues[&I] = C;
477 // Disable SROA in the face of arbitrary casts we don't whitelist elsewhere.
478 disableSROA(I.getOperand(0));
480 return TargetTransformInfo::TCC_Free == TTI.getUserCost(&I);
483 bool CallAnalyzer::visitUnaryInstruction(UnaryInstruction &I) {
484 Value *Operand = I.getOperand(0);
485 Constant *COp = dyn_cast<Constant>(Operand);
487 COp = SimplifiedValues.lookup(Operand);
489 if (Constant *C = ConstantFoldInstOperands(I.getOpcode(), I.getType(),
491 SimplifiedValues[&I] = C;
495 // Disable any SROA on the argument to arbitrary unary operators.
496 disableSROA(Operand);
501 bool CallAnalyzer::visitCmpInst(CmpInst &I) {
502 Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
503 // First try to handle simplified comparisons.
504 if (!isa<Constant>(LHS))
505 if (Constant *SimpleLHS = SimplifiedValues.lookup(LHS))
507 if (!isa<Constant>(RHS))
508 if (Constant *SimpleRHS = SimplifiedValues.lookup(RHS))
510 if (Constant *CLHS = dyn_cast<Constant>(LHS)) {
511 if (Constant *CRHS = dyn_cast<Constant>(RHS))
512 if (Constant *C = ConstantExpr::getCompare(I.getPredicate(), CLHS, CRHS)) {
513 SimplifiedValues[&I] = C;
518 if (I.getOpcode() == Instruction::FCmp)
521 // Otherwise look for a comparison between constant offset pointers with
523 Value *LHSBase, *RHSBase;
524 APInt LHSOffset, RHSOffset;
525 std::tie(LHSBase, LHSOffset) = ConstantOffsetPtrs.lookup(LHS);
527 std::tie(RHSBase, RHSOffset) = ConstantOffsetPtrs.lookup(RHS);
528 if (RHSBase && LHSBase == RHSBase) {
529 // We have common bases, fold the icmp to a constant based on the
531 Constant *CLHS = ConstantInt::get(LHS->getContext(), LHSOffset);
532 Constant *CRHS = ConstantInt::get(RHS->getContext(), RHSOffset);
533 if (Constant *C = ConstantExpr::getICmp(I.getPredicate(), CLHS, CRHS)) {
534 SimplifiedValues[&I] = C;
535 ++NumConstantPtrCmps;
541 // If the comparison is an equality comparison with null, we can simplify it
542 // for any alloca-derived argument.
543 if (I.isEquality() && isa<ConstantPointerNull>(I.getOperand(1)))
544 if (isAllocaDerivedArg(I.getOperand(0))) {
545 // We can actually predict the result of comparisons between an
546 // alloca-derived value and null. Note that this fires regardless of
548 bool IsNotEqual = I.getPredicate() == CmpInst::ICMP_NE;
549 SimplifiedValues[&I] = IsNotEqual ? ConstantInt::getTrue(I.getType())
550 : ConstantInt::getFalse(I.getType());
554 // Finally check for SROA candidates in comparisons.
556 DenseMap<Value *, int>::iterator CostIt;
557 if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt)) {
558 if (isa<ConstantPointerNull>(I.getOperand(1))) {
559 accumulateSROACost(CostIt, InlineConstants::InstrCost);
569 bool CallAnalyzer::visitSub(BinaryOperator &I) {
570 // Try to handle a special case: we can fold computing the difference of two
571 // constant-related pointers.
572 Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
573 Value *LHSBase, *RHSBase;
574 APInt LHSOffset, RHSOffset;
575 std::tie(LHSBase, LHSOffset) = ConstantOffsetPtrs.lookup(LHS);
577 std::tie(RHSBase, RHSOffset) = ConstantOffsetPtrs.lookup(RHS);
578 if (RHSBase && LHSBase == RHSBase) {
579 // We have common bases, fold the subtract to a constant based on the
581 Constant *CLHS = ConstantInt::get(LHS->getContext(), LHSOffset);
582 Constant *CRHS = ConstantInt::get(RHS->getContext(), RHSOffset);
583 if (Constant *C = ConstantExpr::getSub(CLHS, CRHS)) {
584 SimplifiedValues[&I] = C;
585 ++NumConstantPtrDiffs;
591 // Otherwise, fall back to the generic logic for simplifying and handling
593 return Base::visitSub(I);
596 bool CallAnalyzer::visitBinaryOperator(BinaryOperator &I) {
597 Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
598 if (!isa<Constant>(LHS))
599 if (Constant *SimpleLHS = SimplifiedValues.lookup(LHS))
601 if (!isa<Constant>(RHS))
602 if (Constant *SimpleRHS = SimplifiedValues.lookup(RHS))
604 Value *SimpleV = SimplifyBinOp(I.getOpcode(), LHS, RHS, DL);
605 if (Constant *C = dyn_cast_or_null<Constant>(SimpleV)) {
606 SimplifiedValues[&I] = C;
610 // Disable any SROA on arguments to arbitrary, unsimplified binary operators.
617 bool CallAnalyzer::visitLoad(LoadInst &I) {
619 DenseMap<Value *, int>::iterator CostIt;
620 if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt)) {
622 accumulateSROACost(CostIt, InlineConstants::InstrCost);
632 bool CallAnalyzer::visitStore(StoreInst &I) {
634 DenseMap<Value *, int>::iterator CostIt;
635 if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt)) {
637 accumulateSROACost(CostIt, InlineConstants::InstrCost);
647 bool CallAnalyzer::visitExtractValue(ExtractValueInst &I) {
648 // Constant folding for extract value is trivial.
649 Constant *C = dyn_cast<Constant>(I.getAggregateOperand());
651 C = SimplifiedValues.lookup(I.getAggregateOperand());
653 SimplifiedValues[&I] = ConstantExpr::getExtractValue(C, I.getIndices());
657 // SROA can look through these but give them a cost.
661 bool CallAnalyzer::visitInsertValue(InsertValueInst &I) {
662 // Constant folding for insert value is trivial.
663 Constant *AggC = dyn_cast<Constant>(I.getAggregateOperand());
665 AggC = SimplifiedValues.lookup(I.getAggregateOperand());
666 Constant *InsertedC = dyn_cast<Constant>(I.getInsertedValueOperand());
668 InsertedC = SimplifiedValues.lookup(I.getInsertedValueOperand());
669 if (AggC && InsertedC) {
670 SimplifiedValues[&I] = ConstantExpr::getInsertValue(AggC, InsertedC,
675 // SROA can look through these but give them a cost.
679 /// \brief Try to simplify a call site.
681 /// Takes a concrete function and callsite and tries to actually simplify it by
682 /// analyzing the arguments and call itself with instsimplify. Returns true if
683 /// it has simplified the callsite to some other entity (a constant), making it
685 bool CallAnalyzer::simplifyCallSite(Function *F, CallSite CS) {
686 // FIXME: Using the instsimplify logic directly for this is inefficient
687 // because we have to continually rebuild the argument list even when no
688 // simplifications can be performed. Until that is fixed with remapping
689 // inside of instsimplify, directly constant fold calls here.
690 if (!canConstantFoldCallTo(F))
693 // Try to re-map the arguments to constants.
694 SmallVector<Constant *, 4> ConstantArgs;
695 ConstantArgs.reserve(CS.arg_size());
696 for (CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
698 Constant *C = dyn_cast<Constant>(*I);
700 C = dyn_cast_or_null<Constant>(SimplifiedValues.lookup(*I));
702 return false; // This argument doesn't map to a constant.
704 ConstantArgs.push_back(C);
706 if (Constant *C = ConstantFoldCall(F, ConstantArgs)) {
707 SimplifiedValues[CS.getInstruction()] = C;
714 bool CallAnalyzer::visitCallSite(CallSite CS) {
715 if (CS.hasFnAttr(Attribute::ReturnsTwice) &&
716 !F.getAttributes().hasAttribute(AttributeSet::FunctionIndex,
717 Attribute::ReturnsTwice)) {
718 // This aborts the entire analysis.
719 ExposesReturnsTwice = true;
723 cast<CallInst>(CS.getInstruction())->cannotDuplicate())
724 ContainsNoDuplicateCall = true;
726 if (Function *F = CS.getCalledFunction()) {
727 // When we have a concrete function, first try to simplify it directly.
728 if (simplifyCallSite(F, CS))
731 // Next check if it is an intrinsic we know about.
732 // FIXME: Lift this into part of the InstVisitor.
733 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction())) {
734 switch (II->getIntrinsicID()) {
736 return Base::visitCallSite(CS);
738 case Intrinsic::memset:
739 case Intrinsic::memcpy:
740 case Intrinsic::memmove:
741 // SROA can usually chew through these intrinsics, but they aren't free.
746 if (F == CS.getInstruction()->getParent()->getParent()) {
747 // This flag will fully abort the analysis, so don't bother with anything
749 IsRecursiveCall = true;
753 if (TTI.isLoweredToCall(F)) {
754 // We account for the average 1 instruction per call argument setup
756 Cost += CS.arg_size() * InlineConstants::InstrCost;
758 // Everything other than inline ASM will also have a significant cost
759 // merely from making the call.
760 if (!isa<InlineAsm>(CS.getCalledValue()))
761 Cost += InlineConstants::CallPenalty;
764 return Base::visitCallSite(CS);
767 // Otherwise we're in a very special case -- an indirect function call. See
768 // if we can be particularly clever about this.
769 Value *Callee = CS.getCalledValue();
771 // First, pay the price of the argument setup. We account for the average
772 // 1 instruction per call argument setup here.
773 Cost += CS.arg_size() * InlineConstants::InstrCost;
775 // Next, check if this happens to be an indirect function call to a known
776 // function in this inline context. If not, we've done all we can.
777 Function *F = dyn_cast_or_null<Function>(SimplifiedValues.lookup(Callee));
779 return Base::visitCallSite(CS);
781 // If we have a constant that we are calling as a function, we can peer
782 // through it and see the function target. This happens not infrequently
783 // during devirtualization and so we want to give it a hefty bonus for
784 // inlining, but cap that bonus in the event that inlining wouldn't pan
785 // out. Pretend to inline the function, with a custom threshold.
786 CallAnalyzer CA(DL, TTI, AT, *F, InlineConstants::IndirectCallThreshold);
787 if (CA.analyzeCall(CS)) {
788 // We were able to inline the indirect call! Subtract the cost from the
789 // bonus we want to apply, but don't go below zero.
790 Cost -= std::max(0, InlineConstants::IndirectCallThreshold - CA.getCost());
793 return Base::visitCallSite(CS);
796 bool CallAnalyzer::visitReturnInst(ReturnInst &RI) {
797 // At least one return instruction will be free after inlining.
798 bool Free = !HasReturn;
803 bool CallAnalyzer::visitBranchInst(BranchInst &BI) {
804 // We model unconditional branches as essentially free -- they really
805 // shouldn't exist at all, but handling them makes the behavior of the
806 // inliner more regular and predictable. Interestingly, conditional branches
807 // which will fold away are also free.
808 return BI.isUnconditional() || isa<ConstantInt>(BI.getCondition()) ||
809 dyn_cast_or_null<ConstantInt>(
810 SimplifiedValues.lookup(BI.getCondition()));
813 bool CallAnalyzer::visitSwitchInst(SwitchInst &SI) {
814 // We model unconditional switches as free, see the comments on handling
816 if (isa<ConstantInt>(SI.getCondition()))
818 if (Value *V = SimplifiedValues.lookup(SI.getCondition()))
819 if (isa<ConstantInt>(V))
822 // Otherwise, we need to accumulate a cost proportional to the number of
823 // distinct successor blocks. This fan-out in the CFG cannot be represented
824 // for free even if we can represent the core switch as a jumptable that
825 // takes a single instruction.
827 // NB: We convert large switches which are just used to initialize large phi
828 // nodes to lookup tables instead in simplify-cfg, so this shouldn't prevent
829 // inlining those. It will prevent inlining in cases where the optimization
830 // does not (yet) fire.
831 SmallPtrSet<BasicBlock *, 8> SuccessorBlocks;
832 SuccessorBlocks.insert(SI.getDefaultDest());
833 for (auto I = SI.case_begin(), E = SI.case_end(); I != E; ++I)
834 SuccessorBlocks.insert(I.getCaseSuccessor());
835 // Add cost corresponding to the number of distinct destinations. The first
836 // we model as free because of fallthrough.
837 Cost += (SuccessorBlocks.size() - 1) * InlineConstants::InstrCost;
841 bool CallAnalyzer::visitIndirectBrInst(IndirectBrInst &IBI) {
842 // We never want to inline functions that contain an indirectbr. This is
843 // incorrect because all the blockaddress's (in static global initializers
844 // for example) would be referring to the original function, and this
845 // indirect jump would jump from the inlined copy of the function into the
846 // original function which is extremely undefined behavior.
847 // FIXME: This logic isn't really right; we can safely inline functions with
848 // indirectbr's as long as no other function or global references the
849 // blockaddress of a block within the current function.
850 HasIndirectBr = true;
854 bool CallAnalyzer::visitResumeInst(ResumeInst &RI) {
855 // FIXME: It's not clear that a single instruction is an accurate model for
856 // the inline cost of a resume instruction.
860 bool CallAnalyzer::visitUnreachableInst(UnreachableInst &I) {
861 // FIXME: It might be reasonably to discount the cost of instructions leading
862 // to unreachable as they have the lowest possible impact on both runtime and
864 return true; // No actual code is needed for unreachable.
867 bool CallAnalyzer::visitInstruction(Instruction &I) {
868 // Some instructions are free. All of the free intrinsics can also be
869 // handled by SROA, etc.
870 if (TargetTransformInfo::TCC_Free == TTI.getUserCost(&I))
873 // We found something we don't understand or can't handle. Mark any SROA-able
874 // values in the operand list as no longer viable.
875 for (User::op_iterator OI = I.op_begin(), OE = I.op_end(); OI != OE; ++OI)
882 /// \brief Analyze a basic block for its contribution to the inline cost.
884 /// This method walks the analyzer over every instruction in the given basic
885 /// block and accounts for their cost during inlining at this callsite. It
886 /// aborts early if the threshold has been exceeded or an impossible to inline
887 /// construct has been detected. It returns false if inlining is no longer
888 /// viable, and true if inlining remains viable.
889 bool CallAnalyzer::analyzeBlock(BasicBlock *BB,
890 SmallPtrSetImpl<const Value *> &EphValues) {
891 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
892 // FIXME: Currently, the number of instructions in a function regardless of
893 // our ability to simplify them during inline to constants or dead code,
894 // are actually used by the vector bonus heuristic. As long as that's true,
895 // we have to special case debug intrinsics here to prevent differences in
896 // inlining due to debug symbols. Eventually, the number of unsimplified
897 // instructions shouldn't factor into the cost computation, but until then,
898 // hack around it here.
899 if (isa<DbgInfoIntrinsic>(I))
902 // Skip ephemeral values.
903 if (EphValues.count(I))
907 if (isa<ExtractElementInst>(I) || I->getType()->isVectorTy())
908 ++NumVectorInstructions;
910 // If the instruction simplified to a constant, there is no cost to this
911 // instruction. Visit the instructions using our InstVisitor to account for
912 // all of the per-instruction logic. The visit tree returns true if we
913 // consumed the instruction in any way, and false if the instruction's base
914 // cost should count against inlining.
916 ++NumInstructionsSimplified;
918 Cost += InlineConstants::InstrCost;
920 // If the visit this instruction detected an uninlinable pattern, abort.
921 if (IsRecursiveCall || ExposesReturnsTwice || HasDynamicAlloca ||
925 // If the caller is a recursive function then we don't want to inline
926 // functions which allocate a lot of stack space because it would increase
927 // the caller stack usage dramatically.
928 if (IsCallerRecursive &&
929 AllocatedSize > InlineConstants::TotalAllocaSizeRecursiveCaller)
932 if (NumVectorInstructions > NumInstructions/2)
933 VectorBonus = FiftyPercentVectorBonus;
934 else if (NumVectorInstructions > NumInstructions/10)
935 VectorBonus = TenPercentVectorBonus;
939 // Check if we've past the threshold so we don't spin in huge basic
940 // blocks that will never inline.
941 if (Cost > (Threshold + VectorBonus))
948 /// \brief Compute the base pointer and cumulative constant offsets for V.
950 /// This strips all constant offsets off of V, leaving it the base pointer, and
951 /// accumulates the total constant offset applied in the returned constant. It
952 /// returns 0 if V is not a pointer, and returns the constant '0' if there are
953 /// no constant offsets applied.
954 ConstantInt *CallAnalyzer::stripAndComputeInBoundsConstantOffsets(Value *&V) {
955 if (!DL || !V->getType()->isPointerTy())
958 unsigned IntPtrWidth = DL->getPointerSizeInBits();
959 APInt Offset = APInt::getNullValue(IntPtrWidth);
961 // Even though we don't look through PHI nodes, we could be called on an
962 // instruction in an unreachable block, which may be on a cycle.
963 SmallPtrSet<Value *, 4> Visited;
966 if (GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
967 if (!GEP->isInBounds() || !accumulateGEPOffset(*GEP, Offset))
969 V = GEP->getPointerOperand();
970 } else if (Operator::getOpcode(V) == Instruction::BitCast) {
971 V = cast<Operator>(V)->getOperand(0);
972 } else if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
973 if (GA->mayBeOverridden())
975 V = GA->getAliasee();
979 assert(V->getType()->isPointerTy() && "Unexpected operand type!");
980 } while (Visited.insert(V));
982 Type *IntPtrTy = DL->getIntPtrType(V->getContext());
983 return cast<ConstantInt>(ConstantInt::get(IntPtrTy, Offset));
986 /// \brief Analyze a call site for potential inlining.
988 /// Returns true if inlining this call is viable, and false if it is not
989 /// viable. It computes the cost and adjusts the threshold based on numerous
990 /// factors and heuristics. If this method returns false but the computed cost
991 /// is below the computed threshold, then inlining was forcibly disabled by
992 /// some artifact of the routine.
993 bool CallAnalyzer::analyzeCall(CallSite CS) {
996 // Track whether the post-inlining function would have more than one basic
997 // block. A single basic block is often intended for inlining. Balloon the
998 // threshold by 50% until we pass the single-BB phase.
999 bool SingleBB = true;
1000 int SingleBBBonus = Threshold / 2;
1001 Threshold += SingleBBBonus;
1003 // Perform some tweaks to the cost and threshold based on the direct
1004 // callsite information.
1006 // We want to more aggressively inline vector-dense kernels, so up the
1007 // threshold, and we'll lower it if the % of vector instructions gets too
1009 assert(NumInstructions == 0);
1010 assert(NumVectorInstructions == 0);
1011 FiftyPercentVectorBonus = Threshold;
1012 TenPercentVectorBonus = Threshold / 2;
1014 // Give out bonuses per argument, as the instructions setting them up will
1015 // be gone after inlining.
1016 for (unsigned I = 0, E = CS.arg_size(); I != E; ++I) {
1017 if (DL && CS.isByValArgument(I)) {
1018 // We approximate the number of loads and stores needed by dividing the
1019 // size of the byval type by the target's pointer size.
1020 PointerType *PTy = cast<PointerType>(CS.getArgument(I)->getType());
1021 unsigned TypeSize = DL->getTypeSizeInBits(PTy->getElementType());
1022 unsigned PointerSize = DL->getPointerSizeInBits();
1023 // Ceiling division.
1024 unsigned NumStores = (TypeSize + PointerSize - 1) / PointerSize;
1026 // If it generates more than 8 stores it is likely to be expanded as an
1027 // inline memcpy so we take that as an upper bound. Otherwise we assume
1028 // one load and one store per word copied.
1029 // FIXME: The maxStoresPerMemcpy setting from the target should be used
1030 // here instead of a magic number of 8, but it's not available via
1032 NumStores = std::min(NumStores, 8U);
1034 Cost -= 2 * NumStores * InlineConstants::InstrCost;
1036 // For non-byval arguments subtract off one instruction per call
1038 Cost -= InlineConstants::InstrCost;
1042 // If there is only one call of the function, and it has internal linkage,
1043 // the cost of inlining it drops dramatically.
1044 bool OnlyOneCallAndLocalLinkage = F.hasLocalLinkage() && F.hasOneUse() &&
1045 &F == CS.getCalledFunction();
1046 if (OnlyOneCallAndLocalLinkage)
1047 Cost += InlineConstants::LastCallToStaticBonus;
1049 // If the instruction after the call, or if the normal destination of the
1050 // invoke is an unreachable instruction, the function is noreturn. As such,
1051 // there is little point in inlining this unless there is literally zero
1053 Instruction *Instr = CS.getInstruction();
1054 if (InvokeInst *II = dyn_cast<InvokeInst>(Instr)) {
1055 if (isa<UnreachableInst>(II->getNormalDest()->begin()))
1057 } else if (isa<UnreachableInst>(++BasicBlock::iterator(Instr)))
1060 // If this function uses the coldcc calling convention, prefer not to inline
1062 if (F.getCallingConv() == CallingConv::Cold)
1063 Cost += InlineConstants::ColdccPenalty;
1065 // Check if we're done. This can happen due to bonuses and penalties.
1066 if (Cost > Threshold)
1072 Function *Caller = CS.getInstruction()->getParent()->getParent();
1073 // Check if the caller function is recursive itself.
1074 for (User *U : Caller->users()) {
1078 Instruction *I = Site.getInstruction();
1079 if (I->getParent()->getParent() == Caller) {
1080 IsCallerRecursive = true;
1085 // Populate our simplified values by mapping from function arguments to call
1086 // arguments with known important simplifications.
1087 CallSite::arg_iterator CAI = CS.arg_begin();
1088 for (Function::arg_iterator FAI = F.arg_begin(), FAE = F.arg_end();
1089 FAI != FAE; ++FAI, ++CAI) {
1090 assert(CAI != CS.arg_end());
1091 if (Constant *C = dyn_cast<Constant>(CAI))
1092 SimplifiedValues[FAI] = C;
1094 Value *PtrArg = *CAI;
1095 if (ConstantInt *C = stripAndComputeInBoundsConstantOffsets(PtrArg)) {
1096 ConstantOffsetPtrs[FAI] = std::make_pair(PtrArg, C->getValue());
1098 // We can SROA any pointer arguments derived from alloca instructions.
1099 if (isa<AllocaInst>(PtrArg)) {
1100 SROAArgValues[FAI] = PtrArg;
1101 SROAArgCosts[PtrArg] = 0;
1105 NumConstantArgs = SimplifiedValues.size();
1106 NumConstantOffsetPtrArgs = ConstantOffsetPtrs.size();
1107 NumAllocaArgs = SROAArgValues.size();
1109 // FIXME: If a caller has multiple calls to a callee, we end up recomputing
1110 // the ephemeral values multiple times (and they're completely determined by
1111 // the callee, so this is purely duplicate work).
1112 SmallPtrSet<const Value *, 32> EphValues;
1113 CodeMetrics::collectEphemeralValues(&F, AT, EphValues);
1115 // The worklist of live basic blocks in the callee *after* inlining. We avoid
1116 // adding basic blocks of the callee which can be proven to be dead for this
1117 // particular call site in order to get more accurate cost estimates. This
1118 // requires a somewhat heavyweight iteration pattern: we need to walk the
1119 // basic blocks in a breadth-first order as we insert live successors. To
1120 // accomplish this, prioritizing for small iterations because we exit after
1121 // crossing our threshold, we use a small-size optimized SetVector.
1122 typedef SetVector<BasicBlock *, SmallVector<BasicBlock *, 16>,
1123 SmallPtrSet<BasicBlock *, 16> > BBSetVector;
1124 BBSetVector BBWorklist;
1125 BBWorklist.insert(&F.getEntryBlock());
1126 // Note that we *must not* cache the size, this loop grows the worklist.
1127 for (unsigned Idx = 0; Idx != BBWorklist.size(); ++Idx) {
1128 // Bail out the moment we cross the threshold. This means we'll under-count
1129 // the cost, but only when undercounting doesn't matter.
1130 if (Cost > (Threshold + VectorBonus))
1133 BasicBlock *BB = BBWorklist[Idx];
1137 // Disallow inlining a blockaddress. A blockaddress only has defined
1138 // behavior for an indirect branch in the same function, and we do not
1139 // currently support inlining indirect branches. But, the inliner may not
1140 // see an indirect branch that ends up being dead code at a particular call
1141 // site. If the blockaddress escapes the function, e.g., via a global
1142 // variable, inlining may lead to an invalid cross-function reference.
1143 if (BB->hasAddressTaken())
1146 // Analyze the cost of this block. If we blow through the threshold, this
1147 // returns false, and we can bail on out.
1148 if (!analyzeBlock(BB, EphValues)) {
1149 if (IsRecursiveCall || ExposesReturnsTwice || HasDynamicAlloca ||
1153 // If the caller is a recursive function then we don't want to inline
1154 // functions which allocate a lot of stack space because it would increase
1155 // the caller stack usage dramatically.
1156 if (IsCallerRecursive &&
1157 AllocatedSize > InlineConstants::TotalAllocaSizeRecursiveCaller)
1163 TerminatorInst *TI = BB->getTerminator();
1165 // Add in the live successors by first checking whether we have terminator
1166 // that may be simplified based on the values simplified by this call.
1167 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
1168 if (BI->isConditional()) {
1169 Value *Cond = BI->getCondition();
1170 if (ConstantInt *SimpleCond
1171 = dyn_cast_or_null<ConstantInt>(SimplifiedValues.lookup(Cond))) {
1172 BBWorklist.insert(BI->getSuccessor(SimpleCond->isZero() ? 1 : 0));
1176 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
1177 Value *Cond = SI->getCondition();
1178 if (ConstantInt *SimpleCond
1179 = dyn_cast_or_null<ConstantInt>(SimplifiedValues.lookup(Cond))) {
1180 BBWorklist.insert(SI->findCaseValue(SimpleCond).getCaseSuccessor());
1185 // If we're unable to select a particular successor, just count all of
1187 for (unsigned TIdx = 0, TSize = TI->getNumSuccessors(); TIdx != TSize;
1189 BBWorklist.insert(TI->getSuccessor(TIdx));
1191 // If we had any successors at this point, than post-inlining is likely to
1192 // have them as well. Note that we assume any basic blocks which existed
1193 // due to branches or switches which folded above will also fold after
1195 if (SingleBB && TI->getNumSuccessors() > 1) {
1196 // Take off the bonus we applied to the threshold.
1197 Threshold -= SingleBBBonus;
1202 // If this is a noduplicate call, we can still inline as long as
1203 // inlining this would cause the removal of the caller (so the instruction
1204 // is not actually duplicated, just moved).
1205 if (!OnlyOneCallAndLocalLinkage && ContainsNoDuplicateCall)
1208 Threshold += VectorBonus;
1210 return Cost < Threshold;
1213 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
1214 /// \brief Dump stats about this call's analysis.
1215 void CallAnalyzer::dump() {
1216 #define DEBUG_PRINT_STAT(x) dbgs() << " " #x ": " << x << "\n"
1217 DEBUG_PRINT_STAT(NumConstantArgs);
1218 DEBUG_PRINT_STAT(NumConstantOffsetPtrArgs);
1219 DEBUG_PRINT_STAT(NumAllocaArgs);
1220 DEBUG_PRINT_STAT(NumConstantPtrCmps);
1221 DEBUG_PRINT_STAT(NumConstantPtrDiffs);
1222 DEBUG_PRINT_STAT(NumInstructionsSimplified);
1223 DEBUG_PRINT_STAT(SROACostSavings);
1224 DEBUG_PRINT_STAT(SROACostSavingsLost);
1225 DEBUG_PRINT_STAT(ContainsNoDuplicateCall);
1226 DEBUG_PRINT_STAT(Cost);
1227 DEBUG_PRINT_STAT(Threshold);
1228 DEBUG_PRINT_STAT(VectorBonus);
1229 #undef DEBUG_PRINT_STAT
1233 INITIALIZE_PASS_BEGIN(InlineCostAnalysis, "inline-cost", "Inline Cost Analysis",
1235 INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
1236 INITIALIZE_PASS_DEPENDENCY(AssumptionTracker)
1237 INITIALIZE_PASS_END(InlineCostAnalysis, "inline-cost", "Inline Cost Analysis",
1240 char InlineCostAnalysis::ID = 0;
1242 InlineCostAnalysis::InlineCostAnalysis() : CallGraphSCCPass(ID) {}
1244 InlineCostAnalysis::~InlineCostAnalysis() {}
1246 void InlineCostAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
1247 AU.setPreservesAll();
1248 AU.addRequired<AssumptionTracker>();
1249 AU.addRequired<TargetTransformInfo>();
1250 CallGraphSCCPass::getAnalysisUsage(AU);
1253 bool InlineCostAnalysis::runOnSCC(CallGraphSCC &SCC) {
1254 TTI = &getAnalysis<TargetTransformInfo>();
1255 AT = &getAnalysis<AssumptionTracker>();
1259 InlineCost InlineCostAnalysis::getInlineCost(CallSite CS, int Threshold) {
1260 return getInlineCost(CS, CS.getCalledFunction(), Threshold);
1263 /// \brief Test that two functions either have or have not the given attribute
1264 /// at the same time.
1265 static bool attributeMatches(Function *F1, Function *F2,
1266 Attribute::AttrKind Attr) {
1267 return F1->hasFnAttribute(Attr) == F2->hasFnAttribute(Attr);
1270 /// \brief Test that there are no attribute conflicts between Caller and Callee
1271 /// that prevent inlining.
1272 static bool functionsHaveCompatibleAttributes(Function *Caller,
1274 return attributeMatches(Caller, Callee, Attribute::SanitizeAddress) &&
1275 attributeMatches(Caller, Callee, Attribute::SanitizeMemory) &&
1276 attributeMatches(Caller, Callee, Attribute::SanitizeThread);
1279 InlineCost InlineCostAnalysis::getInlineCost(CallSite CS, Function *Callee,
1281 // Cannot inline indirect calls.
1283 return llvm::InlineCost::getNever();
1285 // Calls to functions with always-inline attributes should be inlined
1286 // whenever possible.
1287 if (CS.hasFnAttr(Attribute::AlwaysInline)) {
1288 if (isInlineViable(*Callee))
1289 return llvm::InlineCost::getAlways();
1290 return llvm::InlineCost::getNever();
1293 // Never inline functions with conflicting attributes (unless callee has
1294 // always-inline attribute).
1295 if (!functionsHaveCompatibleAttributes(CS.getCaller(), Callee))
1296 return llvm::InlineCost::getNever();
1298 // Don't inline this call if the caller has the optnone attribute.
1299 if (CS.getCaller()->hasFnAttribute(Attribute::OptimizeNone))
1300 return llvm::InlineCost::getNever();
1302 // Don't inline functions which can be redefined at link-time to mean
1303 // something else. Don't inline functions marked noinline or call sites
1305 if (Callee->mayBeOverridden() ||
1306 Callee->hasFnAttribute(Attribute::NoInline) || CS.isNoInline())
1307 return llvm::InlineCost::getNever();
1309 DEBUG(llvm::dbgs() << " Analyzing call of " << Callee->getName()
1312 CallAnalyzer CA(Callee->getDataLayout(), *TTI, AT, *Callee, Threshold);
1313 bool ShouldInline = CA.analyzeCall(CS);
1317 // Check if there was a reason to force inlining or no inlining.
1318 if (!ShouldInline && CA.getCost() < CA.getThreshold())
1319 return InlineCost::getNever();
1320 if (ShouldInline && CA.getCost() >= CA.getThreshold())
1321 return InlineCost::getAlways();
1323 return llvm::InlineCost::get(CA.getCost(), CA.getThreshold());
1326 bool InlineCostAnalysis::isInlineViable(Function &F) {
1328 F.getAttributes().hasAttribute(AttributeSet::FunctionIndex,
1329 Attribute::ReturnsTwice);
1330 for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
1331 // Disallow inlining of functions which contain indirect branches or
1333 if (isa<IndirectBrInst>(BI->getTerminator()) || BI->hasAddressTaken())
1336 for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE;
1342 // Disallow recursive calls.
1343 if (&F == CS.getCalledFunction())
1346 // Disallow calls which expose returns-twice to a function not previously
1347 // attributed as such.
1348 if (!ReturnsTwice && CS.isCall() &&
1349 cast<CallInst>(CS.getInstruction())->canReturnTwice())