1 //===- InlineCost.cpp - Cost analysis for inliner -------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements inline cost analysis.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "llvm/Analysis/InlineCost.h"
15 #include "llvm/Support/CallSite.h"
16 #include "llvm/CallingConv.h"
17 #include "llvm/IntrinsicInst.h"
18 #include "llvm/Target/TargetData.h"
19 #include "llvm/ADT/SmallPtrSet.h"
20
21 using namespace llvm;
22
23 /// callIsSmall - If a call is likely to lower to a single target instruction,
24 /// or is otherwise deemed small return true.
25 /// TODO: Perhaps calls like memcpy, strcpy, etc?
callIsSmall(const Function * F)26 bool llvm::callIsSmall(const Function *F) {
27 if (!F) return false;
28
29 if (F->hasLocalLinkage()) return false;
30
31 if (!F->hasName()) return false;
32
33 StringRef Name = F->getName();
34
35 // These will all likely lower to a single selection DAG node.
36 if (Name == "copysign" || Name == "copysignf" || Name == "copysignl" ||
37 Name == "fabs" || Name == "fabsf" || Name == "fabsl" ||
38 Name == "sin" || Name == "sinf" || Name == "sinl" ||
39 Name == "cos" || Name == "cosf" || Name == "cosl" ||
40 Name == "sqrt" || Name == "sqrtf" || Name == "sqrtl" )
41 return true;
42
43 // These are all likely to be optimized into something smaller.
44 if (Name == "pow" || Name == "powf" || Name == "powl" ||
45 Name == "exp2" || Name == "exp2l" || Name == "exp2f" ||
46 Name == "floor" || Name == "floorf" || Name == "ceil" ||
47 Name == "round" || Name == "ffs" || Name == "ffsl" ||
48 Name == "abs" || Name == "labs" || Name == "llabs")
49 return true;
50
51 return false;
52 }
53
54 /// analyzeBasicBlock - Fill in the current structure with information gleaned
55 /// from the specified block.
analyzeBasicBlock(const BasicBlock * BB,const TargetData * TD)56 void CodeMetrics::analyzeBasicBlock(const BasicBlock *BB,
57 const TargetData *TD) {
58 ++NumBlocks;
59 unsigned NumInstsBeforeThisBB = NumInsts;
60 for (BasicBlock::const_iterator II = BB->begin(), E = BB->end();
61 II != E; ++II) {
62 if (isa<PHINode>(II)) continue; // PHI nodes don't count.
63
64 // Special handling for calls.
65 if (isa<CallInst>(II) || isa<InvokeInst>(II)) {
66 if (isa<DbgInfoIntrinsic>(II))
67 continue; // Debug intrinsics don't count as size.
68
69 ImmutableCallSite CS(cast<Instruction>(II));
70
71 if (const Function *F = CS.getCalledFunction()) {
72 // If a function is both internal and has a single use, then it is
73 // extremely likely to get inlined in the future (it was probably
74 // exposed by an interleaved devirtualization pass).
75 if (F->hasInternalLinkage() && F->hasOneUse())
76 ++NumInlineCandidates;
77
78 // If this call is to function itself, then the function is recursive.
79 // Inlining it into other functions is a bad idea, because this is
80 // basically just a form of loop peeling, and our metrics aren't useful
81 // for that case.
82 if (F == BB->getParent())
83 isRecursive = true;
84 }
85
86 if (!isa<IntrinsicInst>(II) && !callIsSmall(CS.getCalledFunction())) {
87 // Each argument to a call takes on average one instruction to set up.
88 NumInsts += CS.arg_size();
89
90 // We don't want inline asm to count as a call - that would prevent loop
91 // unrolling. The argument setup cost is still real, though.
92 if (!isa<InlineAsm>(CS.getCalledValue()))
93 ++NumCalls;
94 }
95 }
96
97 if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
98 if (!AI->isStaticAlloca())
99 this->usesDynamicAlloca = true;
100 }
101
102 if (isa<ExtractElementInst>(II) || II->getType()->isVectorTy())
103 ++NumVectorInsts;
104
105 if (const CastInst *CI = dyn_cast<CastInst>(II)) {
106 // Noop casts, including ptr <-> int, don't count.
107 if (CI->isLosslessCast() || isa<IntToPtrInst>(CI) ||
108 isa<PtrToIntInst>(CI))
109 continue;
110 // trunc to a native type is free (assuming the target has compare and
111 // shift-right of the same width).
112 if (isa<TruncInst>(CI) && TD &&
113 TD->isLegalInteger(TD->getTypeSizeInBits(CI->getType())))
114 continue;
115 // Result of a cmp instruction is often extended (to be used by other
116 // cmp instructions, logical or return instructions). These are usually
117 // nop on most sane targets.
118 if (isa<CmpInst>(CI->getOperand(0)))
119 continue;
120 } else if (const GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(II)){
121 // If a GEP has all constant indices, it will probably be folded with
122 // a load/store.
123 if (GEPI->hasAllConstantIndices())
124 continue;
125 }
126
127 ++NumInsts;
128 }
129
130 if (isa<ReturnInst>(BB->getTerminator()))
131 ++NumRets;
132
133 // We never want to inline functions that contain an indirectbr. This is
134 // incorrect because all the blockaddress's (in static global initializers
135 // for example) would be referring to the original function, and this indirect
136 // jump would jump from the inlined copy of the function into the original
137 // function which is extremely undefined behavior.
138 if (isa<IndirectBrInst>(BB->getTerminator()))
139 containsIndirectBr = true;
140
141 // Remember NumInsts for this BB.
142 NumBBInsts[BB] = NumInsts - NumInstsBeforeThisBB;
143 }
144
145 // CountCodeReductionForConstant - Figure out an approximation for how many
146 // instructions will be constant folded if the specified value is constant.
147 //
CountCodeReductionForConstant(Value * V)148 unsigned CodeMetrics::CountCodeReductionForConstant(Value *V) {
149 unsigned Reduction = 0;
150 for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
151 User *U = *UI;
152 if (isa<BranchInst>(U) || isa<SwitchInst>(U)) {
153 // We will be able to eliminate all but one of the successors.
154 const TerminatorInst &TI = cast<TerminatorInst>(*U);
155 const unsigned NumSucc = TI.getNumSuccessors();
156 unsigned Instrs = 0;
157 for (unsigned I = 0; I != NumSucc; ++I)
158 Instrs += NumBBInsts[TI.getSuccessor(I)];
159 // We don't know which blocks will be eliminated, so use the average size.
160 Reduction += InlineConstants::InstrCost*Instrs*(NumSucc-1)/NumSucc;
161 } else {
162 // Figure out if this instruction will be removed due to simple constant
163 // propagation.
164 Instruction &Inst = cast<Instruction>(*U);
165
166 // We can't constant propagate instructions which have effects or
167 // read memory.
168 //
169 // FIXME: It would be nice to capture the fact that a load from a
170 // pointer-to-constant-global is actually a *really* good thing to zap.
171 // Unfortunately, we don't know the pointer that may get propagated here,
172 // so we can't make this decision.
173 if (Inst.mayReadFromMemory() || Inst.mayHaveSideEffects() ||
174 isa<AllocaInst>(Inst))
175 continue;
176
177 bool AllOperandsConstant = true;
178 for (unsigned i = 0, e = Inst.getNumOperands(); i != e; ++i)
179 if (!isa<Constant>(Inst.getOperand(i)) && Inst.getOperand(i) != V) {
180 AllOperandsConstant = false;
181 break;
182 }
183
184 if (AllOperandsConstant) {
185 // We will get to remove this instruction...
186 Reduction += InlineConstants::InstrCost;
187
188 // And any other instructions that use it which become constants
189 // themselves.
190 Reduction += CountCodeReductionForConstant(&Inst);
191 }
192 }
193 }
194 return Reduction;
195 }
196
197 // CountCodeReductionForAlloca - Figure out an approximation of how much smaller
198 // the function will be if it is inlined into a context where an argument
199 // becomes an alloca.
200 //
CountCodeReductionForAlloca(Value * V)201 unsigned CodeMetrics::CountCodeReductionForAlloca(Value *V) {
202 if (!V->getType()->isPointerTy()) return 0; // Not a pointer
203 unsigned Reduction = 0;
204 for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
205 Instruction *I = cast<Instruction>(*UI);
206 if (isa<LoadInst>(I) || isa<StoreInst>(I))
207 Reduction += InlineConstants::InstrCost;
208 else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
209 // If the GEP has variable indices, we won't be able to do much with it.
210 if (GEP->hasAllConstantIndices())
211 Reduction += CountCodeReductionForAlloca(GEP);
212 } else if (BitCastInst *BCI = dyn_cast<BitCastInst>(I)) {
213 // Track pointer through bitcasts.
214 Reduction += CountCodeReductionForAlloca(BCI);
215 } else {
216 // If there is some other strange instruction, we're not going to be able
217 // to do much if we inline this.
218 return 0;
219 }
220 }
221
222 return Reduction;
223 }
224
225 /// analyzeFunction - Fill in the current structure with information gleaned
226 /// from the specified function.
analyzeFunction(Function * F,const TargetData * TD)227 void CodeMetrics::analyzeFunction(Function *F, const TargetData *TD) {
228 // If this function contains a call to setjmp or _setjmp, never inline
229 // it. This is a hack because we depend on the user marking their local
230 // variables as volatile if they are live across a setjmp call, and they
231 // probably won't do this in callers.
232 if (F->callsFunctionThatReturnsTwice())
233 callsSetJmp = true;
234
235 // Look at the size of the callee.
236 for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
237 analyzeBasicBlock(&*BB, TD);
238 }
239
240 /// analyzeFunction - Fill in the current structure with information gleaned
241 /// from the specified function.
analyzeFunction(Function * F,const TargetData * TD)242 void InlineCostAnalyzer::FunctionInfo::analyzeFunction(Function *F,
243 const TargetData *TD) {
244 Metrics.analyzeFunction(F, TD);
245
246 // A function with exactly one return has it removed during the inlining
247 // process (see InlineFunction), so don't count it.
248 // FIXME: This knowledge should really be encoded outside of FunctionInfo.
249 if (Metrics.NumRets==1)
250 --Metrics.NumInsts;
251
252 // Check out all of the arguments to the function, figuring out how much
253 // code can be eliminated if one of the arguments is a constant.
254 ArgumentWeights.reserve(F->arg_size());
255 for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I)
256 ArgumentWeights.push_back(ArgInfo(Metrics.CountCodeReductionForConstant(I),
257 Metrics.CountCodeReductionForAlloca(I)));
258 }
259
260 /// NeverInline - returns true if the function should never be inlined into
261 /// any caller
NeverInline()262 bool InlineCostAnalyzer::FunctionInfo::NeverInline() {
263 return (Metrics.callsSetJmp || Metrics.isRecursive ||
264 Metrics.containsIndirectBr);
265 }
266 // getSpecializationBonus - The heuristic used to determine the per-call
267 // performance boost for using a specialization of Callee with argument
268 // specializedArgNo replaced by a constant.
getSpecializationBonus(Function * Callee,SmallVectorImpl<unsigned> & SpecializedArgNos)269 int InlineCostAnalyzer::getSpecializationBonus(Function *Callee,
270 SmallVectorImpl<unsigned> &SpecializedArgNos)
271 {
272 if (Callee->mayBeOverridden())
273 return 0;
274
275 int Bonus = 0;
276 // If this function uses the coldcc calling convention, prefer not to
277 // specialize it.
278 if (Callee->getCallingConv() == CallingConv::Cold)
279 Bonus -= InlineConstants::ColdccPenalty;
280
281 // Get information about the callee.
282 FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
283
284 // If we haven't calculated this information yet, do so now.
285 if (CalleeFI->Metrics.NumBlocks == 0)
286 CalleeFI->analyzeFunction(Callee, TD);
287
288 unsigned ArgNo = 0;
289 unsigned i = 0;
290 for (Function::arg_iterator I = Callee->arg_begin(), E = Callee->arg_end();
291 I != E; ++I, ++ArgNo)
292 if (ArgNo == SpecializedArgNos[i]) {
293 ++i;
294 Bonus += CountBonusForConstant(I);
295 }
296
297 // Calls usually take a long time, so they make the specialization gain
298 // smaller.
299 Bonus -= CalleeFI->Metrics.NumCalls * InlineConstants::CallPenalty;
300
301 return Bonus;
302 }
303
304 // ConstantFunctionBonus - Figure out how much of a bonus we can get for
305 // possibly devirtualizing a function. We'll subtract the size of the function
306 // we may wish to inline from the indirect call bonus providing a limit on
307 // growth. Leave an upper limit of 0 for the bonus - we don't want to penalize
308 // inlining because we decide we don't want to give a bonus for
309 // devirtualizing.
ConstantFunctionBonus(CallSite CS,Constant * C)310 int InlineCostAnalyzer::ConstantFunctionBonus(CallSite CS, Constant *C) {
311
312 // This could just be NULL.
313 if (!C) return 0;
314
315 Function *F = dyn_cast<Function>(C);
316 if (!F) return 0;
317
318 int Bonus = InlineConstants::IndirectCallBonus + getInlineSize(CS, F);
319 return (Bonus > 0) ? 0 : Bonus;
320 }
321
322 // CountBonusForConstant - Figure out an approximation for how much per-call
323 // performance boost we can expect if the specified value is constant.
CountBonusForConstant(Value * V,Constant * C)324 int InlineCostAnalyzer::CountBonusForConstant(Value *V, Constant *C) {
325 unsigned Bonus = 0;
326 for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
327 User *U = *UI;
328 if (CallInst *CI = dyn_cast<CallInst>(U)) {
329 // Turning an indirect call into a direct call is a BIG win
330 if (CI->getCalledValue() == V)
331 Bonus += ConstantFunctionBonus(CallSite(CI), C);
332 } else if (InvokeInst *II = dyn_cast<InvokeInst>(U)) {
333 // Turning an indirect call into a direct call is a BIG win
334 if (II->getCalledValue() == V)
335 Bonus += ConstantFunctionBonus(CallSite(II), C);
336 }
337 // FIXME: Eliminating conditional branches and switches should
338 // also yield a per-call performance boost.
339 else {
340 // Figure out the bonuses that wll accrue due to simple constant
341 // propagation.
342 Instruction &Inst = cast<Instruction>(*U);
343
344 // We can't constant propagate instructions which have effects or
345 // read memory.
346 //
347 // FIXME: It would be nice to capture the fact that a load from a
348 // pointer-to-constant-global is actually a *really* good thing to zap.
349 // Unfortunately, we don't know the pointer that may get propagated here,
350 // so we can't make this decision.
351 if (Inst.mayReadFromMemory() || Inst.mayHaveSideEffects() ||
352 isa<AllocaInst>(Inst))
353 continue;
354
355 bool AllOperandsConstant = true;
356 for (unsigned i = 0, e = Inst.getNumOperands(); i != e; ++i)
357 if (!isa<Constant>(Inst.getOperand(i)) && Inst.getOperand(i) != V) {
358 AllOperandsConstant = false;
359 break;
360 }
361
362 if (AllOperandsConstant)
363 Bonus += CountBonusForConstant(&Inst);
364 }
365 }
366
367 return Bonus;
368 }
369
getInlineSize(CallSite CS,Function * Callee)370 int InlineCostAnalyzer::getInlineSize(CallSite CS, Function *Callee) {
371 // Get information about the callee.
372 FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
373
374 // If we haven't calculated this information yet, do so now.
375 if (CalleeFI->Metrics.NumBlocks == 0)
376 CalleeFI->analyzeFunction(Callee, TD);
377
378 // InlineCost - This value measures how good of an inline candidate this call
379 // site is to inline. A lower inline cost make is more likely for the call to
380 // be inlined. This value may go negative.
381 //
382 int InlineCost = 0;
383
384 // Compute any size reductions we can expect due to arguments being passed into
385 // the function.
386 //
387 unsigned ArgNo = 0;
388 CallSite::arg_iterator I = CS.arg_begin();
389 for (Function::arg_iterator FI = Callee->arg_begin(), FE = Callee->arg_end();
390 FI != FE; ++I, ++FI, ++ArgNo) {
391
392 // If an alloca is passed in, inlining this function is likely to allow
393 // significant future optimization possibilities (like scalar promotion, and
394 // scalarization), so encourage the inlining of the function.
395 //
396 if (isa<AllocaInst>(I))
397 InlineCost -= CalleeFI->ArgumentWeights[ArgNo].AllocaWeight;
398
399 // If this is a constant being passed into the function, use the argument
400 // weights calculated for the callee to determine how much will be folded
401 // away with this information.
402 else if (isa<Constant>(I))
403 InlineCost -= CalleeFI->ArgumentWeights[ArgNo].ConstantWeight;
404 }
405
406 // Each argument passed in has a cost at both the caller and the callee
407 // sides. Measurements show that each argument costs about the same as an
408 // instruction.
409 InlineCost -= (CS.arg_size() * InlineConstants::InstrCost);
410
411 // Now that we have considered all of the factors that make the call site more
412 // likely to be inlined, look at factors that make us not want to inline it.
413
414 // Calls usually take a long time, so they make the inlining gain smaller.
415 InlineCost += CalleeFI->Metrics.NumCalls * InlineConstants::CallPenalty;
416
417 // Look at the size of the callee. Each instruction counts as 5.
418 InlineCost += CalleeFI->Metrics.NumInsts*InlineConstants::InstrCost;
419
420 return InlineCost;
421 }
422
getInlineBonuses(CallSite CS,Function * Callee)423 int InlineCostAnalyzer::getInlineBonuses(CallSite CS, Function *Callee) {
424 // Get information about the callee.
425 FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
426
427 // If we haven't calculated this information yet, do so now.
428 if (CalleeFI->Metrics.NumBlocks == 0)
429 CalleeFI->analyzeFunction(Callee, TD);
430
431 bool isDirectCall = CS.getCalledFunction() == Callee;
432 Instruction *TheCall = CS.getInstruction();
433 int Bonus = 0;
434
435 // If there is only one call of the function, and it has internal linkage,
436 // make it almost guaranteed to be inlined.
437 //
438 if (Callee->hasLocalLinkage() && Callee->hasOneUse() && isDirectCall)
439 Bonus += InlineConstants::LastCallToStaticBonus;
440
441 // If the instruction after the call, or if the normal destination of the
442 // invoke is an unreachable instruction, the function is noreturn. As such,
443 // there is little point in inlining this.
444 if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
445 if (isa<UnreachableInst>(II->getNormalDest()->begin()))
446 Bonus += InlineConstants::NoreturnPenalty;
447 } else if (isa<UnreachableInst>(++BasicBlock::iterator(TheCall)))
448 Bonus += InlineConstants::NoreturnPenalty;
449
450 // If this function uses the coldcc calling convention, prefer not to inline
451 // it.
452 if (Callee->getCallingConv() == CallingConv::Cold)
453 Bonus += InlineConstants::ColdccPenalty;
454
455 // Add to the inline quality for properties that make the call valuable to
456 // inline. This includes factors that indicate that the result of inlining
457 // the function will be optimizable. Currently this just looks at arguments
458 // passed into the function.
459 //
460 CallSite::arg_iterator I = CS.arg_begin();
461 for (Function::arg_iterator FI = Callee->arg_begin(), FE = Callee->arg_end();
462 FI != FE; ++I, ++FI)
463 // Compute any constant bonus due to inlining we want to give here.
464 if (isa<Constant>(I))
465 Bonus += CountBonusForConstant(FI, cast<Constant>(I));
466
467 return Bonus;
468 }
469
470 // getInlineCost - The heuristic used to determine if we should inline the
471 // function call or not.
472 //
getInlineCost(CallSite CS,SmallPtrSet<const Function *,16> & NeverInline)473 InlineCost InlineCostAnalyzer::getInlineCost(CallSite CS,
474 SmallPtrSet<const Function*, 16> &NeverInline) {
475 return getInlineCost(CS, CS.getCalledFunction(), NeverInline);
476 }
477
getInlineCost(CallSite CS,Function * Callee,SmallPtrSet<const Function *,16> & NeverInline)478 InlineCost InlineCostAnalyzer::getInlineCost(CallSite CS,
479 Function *Callee,
480 SmallPtrSet<const Function*, 16> &NeverInline) {
481 Instruction *TheCall = CS.getInstruction();
482 Function *Caller = TheCall->getParent()->getParent();
483
484 // Don't inline functions which can be redefined at link-time to mean
485 // something else. Don't inline functions marked noinline or call sites
486 // marked noinline.
487 if (Callee->mayBeOverridden() ||
488 Callee->hasFnAttr(Attribute::NoInline) || NeverInline.count(Callee) ||
489 CS.isNoInline())
490 return llvm::InlineCost::getNever();
491
492 // Get information about the callee.
493 FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
494
495 // If we haven't calculated this information yet, do so now.
496 if (CalleeFI->Metrics.NumBlocks == 0)
497 CalleeFI->analyzeFunction(Callee, TD);
498
499 // If we should never inline this, return a huge cost.
500 if (CalleeFI->NeverInline())
501 return InlineCost::getNever();
502
503 // FIXME: It would be nice to kill off CalleeFI->NeverInline. Then we
504 // could move this up and avoid computing the FunctionInfo for
505 // things we are going to just return always inline for. This
506 // requires handling setjmp somewhere else, however.
507 if (!Callee->isDeclaration() && Callee->hasFnAttr(Attribute::AlwaysInline))
508 return InlineCost::getAlways();
509
510 if (CalleeFI->Metrics.usesDynamicAlloca) {
511 // Get information about the caller.
512 FunctionInfo &CallerFI = CachedFunctionInfo[Caller];
513
514 // If we haven't calculated this information yet, do so now.
515 if (CallerFI.Metrics.NumBlocks == 0) {
516 CallerFI.analyzeFunction(Caller, TD);
517
518 // Recompute the CalleeFI pointer, getting Caller could have invalidated
519 // it.
520 CalleeFI = &CachedFunctionInfo[Callee];
521 }
522
523 // Don't inline a callee with dynamic alloca into a caller without them.
524 // Functions containing dynamic alloca's are inefficient in various ways;
525 // don't create more inefficiency.
526 if (!CallerFI.Metrics.usesDynamicAlloca)
527 return InlineCost::getNever();
528 }
529
530 // InlineCost - This value measures how good of an inline candidate this call
531 // site is to inline. A lower inline cost make is more likely for the call to
532 // be inlined. This value may go negative due to the fact that bonuses
533 // are negative numbers.
534 //
535 int InlineCost = getInlineSize(CS, Callee) + getInlineBonuses(CS, Callee);
536 return llvm::InlineCost::get(InlineCost);
537 }
538
539 // getSpecializationCost - The heuristic used to determine the code-size
540 // impact of creating a specialized version of Callee with argument
541 // SpecializedArgNo replaced by a constant.
getSpecializationCost(Function * Callee,SmallVectorImpl<unsigned> & SpecializedArgNos)542 InlineCost InlineCostAnalyzer::getSpecializationCost(Function *Callee,
543 SmallVectorImpl<unsigned> &SpecializedArgNos)
544 {
545 // Don't specialize functions which can be redefined at link-time to mean
546 // something else.
547 if (Callee->mayBeOverridden())
548 return llvm::InlineCost::getNever();
549
550 // Get information about the callee.
551 FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
552
553 // If we haven't calculated this information yet, do so now.
554 if (CalleeFI->Metrics.NumBlocks == 0)
555 CalleeFI->analyzeFunction(Callee, TD);
556
557 int Cost = 0;
558
559 // Look at the original size of the callee. Each instruction counts as 5.
560 Cost += CalleeFI->Metrics.NumInsts * InlineConstants::InstrCost;
561
562 // Offset that with the amount of code that can be constant-folded
563 // away with the given arguments replaced by constants.
564 for (SmallVectorImpl<unsigned>::iterator an = SpecializedArgNos.begin(),
565 ae = SpecializedArgNos.end(); an != ae; ++an)
566 Cost -= CalleeFI->ArgumentWeights[*an].ConstantWeight;
567
568 return llvm::InlineCost::get(Cost);
569 }
570
571 // getInlineFudgeFactor - Return a > 1.0 factor if the inliner should use a
572 // higher threshold to determine if the function call should be inlined.
getInlineFudgeFactor(CallSite CS)573 float InlineCostAnalyzer::getInlineFudgeFactor(CallSite CS) {
574 Function *Callee = CS.getCalledFunction();
575
576 // Get information about the callee.
577 FunctionInfo &CalleeFI = CachedFunctionInfo[Callee];
578
579 // If we haven't calculated this information yet, do so now.
580 if (CalleeFI.Metrics.NumBlocks == 0)
581 CalleeFI.analyzeFunction(Callee, TD);
582
583 float Factor = 1.0f;
584 // Single BB functions are often written to be inlined.
585 if (CalleeFI.Metrics.NumBlocks == 1)
586 Factor += 0.5f;
587
588 // Be more aggressive if the function contains a good chunk (if it mades up
589 // at least 10% of the instructions) of vector instructions.
590 if (CalleeFI.Metrics.NumVectorInsts > CalleeFI.Metrics.NumInsts/2)
591 Factor += 2.0f;
592 else if (CalleeFI.Metrics.NumVectorInsts > CalleeFI.Metrics.NumInsts/10)
593 Factor += 1.5f;
594 return Factor;
595 }
596
597 /// growCachedCostInfo - update the cached cost info for Caller after Callee has
598 /// been inlined.
599 void
growCachedCostInfo(Function * Caller,Function * Callee)600 InlineCostAnalyzer::growCachedCostInfo(Function *Caller, Function *Callee) {
601 CodeMetrics &CallerMetrics = CachedFunctionInfo[Caller].Metrics;
602
603 // For small functions we prefer to recalculate the cost for better accuracy.
604 if (CallerMetrics.NumBlocks < 10 && CallerMetrics.NumInsts < 1000) {
605 resetCachedCostInfo(Caller);
606 return;
607 }
608
609 // For large functions, we can save a lot of computation time by skipping
610 // recalculations.
611 if (CallerMetrics.NumCalls > 0)
612 --CallerMetrics.NumCalls;
613
614 if (Callee == 0) return;
615
616 CodeMetrics &CalleeMetrics = CachedFunctionInfo[Callee].Metrics;
617
618 // If we don't have metrics for the callee, don't recalculate them just to
619 // update an approximation in the caller. Instead, just recalculate the
620 // caller info from scratch.
621 if (CalleeMetrics.NumBlocks == 0) {
622 resetCachedCostInfo(Caller);
623 return;
624 }
625
626 // Since CalleeMetrics were already calculated, we know that the CallerMetrics
627 // reference isn't invalidated: both were in the DenseMap.
628 CallerMetrics.usesDynamicAlloca |= CalleeMetrics.usesDynamicAlloca;
629
630 // FIXME: If any of these three are true for the callee, the callee was
631 // not inlined into the caller, so I think they're redundant here.
632 CallerMetrics.callsSetJmp |= CalleeMetrics.callsSetJmp;
633 CallerMetrics.isRecursive |= CalleeMetrics.isRecursive;
634 CallerMetrics.containsIndirectBr |= CalleeMetrics.containsIndirectBr;
635
636 CallerMetrics.NumInsts += CalleeMetrics.NumInsts;
637 CallerMetrics.NumBlocks += CalleeMetrics.NumBlocks;
638 CallerMetrics.NumCalls += CalleeMetrics.NumCalls;
639 CallerMetrics.NumVectorInsts += CalleeMetrics.NumVectorInsts;
640 CallerMetrics.NumRets += CalleeMetrics.NumRets;
641
642 // analyzeBasicBlock counts each function argument as an inst.
643 if (CallerMetrics.NumInsts >= Callee->arg_size())
644 CallerMetrics.NumInsts -= Callee->arg_size();
645 else
646 CallerMetrics.NumInsts = 0;
647
648 // We are not updating the argument weights. We have already determined that
649 // Caller is a fairly large function, so we accept the loss of precision.
650 }
651
652 /// clear - empty the cache of inline costs
clear()653 void InlineCostAnalyzer::clear() {
654 CachedFunctionInfo.clear();
655 }
656