1 //===- CodeMetrics.cpp - Code cost measurements ---------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements code cost measurement utilities.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/Analysis/CodeMetrics.h"
14 #include "llvm/ADT/SmallPtrSet.h"
15 #include "llvm/Analysis/AssumptionCache.h"
16 #include "llvm/Analysis/LoopInfo.h"
17 #include "llvm/Analysis/TargetTransformInfo.h"
18 #include "llvm/Analysis/ValueTracking.h"
19 #include "llvm/IR/Function.h"
20 #include "llvm/Support/Debug.h"
21 #include "llvm/Support/InstructionCost.h"
22 
23 #define DEBUG_TYPE "code-metrics"
24 
25 using namespace llvm;
26 
27 static void
28 appendSpeculatableOperands(const Value *V,
29                            SmallPtrSetImpl<const Value *> &Visited,
30                            SmallVectorImpl<const Value *> &Worklist) {
31   const User *U = dyn_cast<User>(V);
32   if (!U)
33     return;
34 
35   for (const Value *Operand : U->operands())
36     if (Visited.insert(Operand).second)
37       if (isSafeToSpeculativelyExecute(Operand))
38         Worklist.push_back(Operand);
39 }
40 
41 static void completeEphemeralValues(SmallPtrSetImpl<const Value *> &Visited,
42                                     SmallVectorImpl<const Value *> &Worklist,
43                                     SmallPtrSetImpl<const Value *> &EphValues) {
44   // Note: We don't speculate PHIs here, so we'll miss instruction chains kept
45   // alive only by ephemeral values.
46 
47   // Walk the worklist using an index but without caching the size so we can
48   // append more entries as we process the worklist. This forms a queue without
49   // quadratic behavior by just leaving processed nodes at the head of the
50   // worklist forever.
51   for (int i = 0; i < (int)Worklist.size(); ++i) {
52     const Value *V = Worklist[i];
53 
54     assert(Visited.count(V) &&
55            "Failed to add a worklist entry to our visited set!");
56 
57     // If all uses of this value are ephemeral, then so is this value.
58     if (!all_of(V->users(), [&](const User *U) { return EphValues.count(U); }))
59       continue;
60 
61     EphValues.insert(V);
62     LLVM_DEBUG(dbgs() << "Ephemeral Value: " << *V << "\n");
63 
64     // Append any more operands to consider.
65     appendSpeculatableOperands(V, Visited, Worklist);
66   }
67 }
68 
69 // Find all ephemeral values.
70 void CodeMetrics::collectEphemeralValues(
71     const Loop *L, AssumptionCache *AC,
72     SmallPtrSetImpl<const Value *> &EphValues) {
73   SmallPtrSet<const Value *, 32> Visited;
74   SmallVector<const Value *, 16> Worklist;
75 
76   for (auto &AssumeVH : AC->assumptions()) {
77     if (!AssumeVH)
78       continue;
79     Instruction *I = cast<Instruction>(AssumeVH);
80 
81     // Filter out call sites outside of the loop so we don't do a function's
82     // worth of work for each of its loops (and, in the common case, ephemeral
83     // values in the loop are likely due to @llvm.assume calls in the loop).
84     if (!L->contains(I->getParent()))
85       continue;
86 
87     if (EphValues.insert(I).second)
88       appendSpeculatableOperands(I, Visited, Worklist);
89   }
90 
91   completeEphemeralValues(Visited, Worklist, EphValues);
92 }
93 
94 void CodeMetrics::collectEphemeralValues(
95     const Function *F, AssumptionCache *AC,
96     SmallPtrSetImpl<const Value *> &EphValues) {
97   SmallPtrSet<const Value *, 32> Visited;
98   SmallVector<const Value *, 16> Worklist;
99 
100   for (auto &AssumeVH : AC->assumptions()) {
101     if (!AssumeVH)
102       continue;
103     Instruction *I = cast<Instruction>(AssumeVH);
104     assert(I->getParent()->getParent() == F &&
105            "Found assumption for the wrong function!");
106 
107     if (EphValues.insert(I).second)
108       appendSpeculatableOperands(I, Visited, Worklist);
109   }
110 
111   completeEphemeralValues(Visited, Worklist, EphValues);
112 }
113 
114 /// Fill in the current structure with information gleaned from the specified
115 /// block.
116 void CodeMetrics::analyzeBasicBlock(
117     const BasicBlock *BB, const TargetTransformInfo &TTI,
118     const SmallPtrSetImpl<const Value *> &EphValues, bool PrepareForLTO) {
119   ++NumBlocks;
120   // Use a proxy variable for NumInsts of type InstructionCost, so that it can
121   // use InstructionCost's arithmetic properties such as saturation when this
122   // feature is added to InstructionCost.
123   // When storing the value back to NumInsts, we can assume all costs are Valid
124   // because the IR should not contain any nodes that cannot be costed. If that
125   // happens the cost-model is broken.
126   InstructionCost NumInstsProxy = NumInsts;
127   InstructionCost NumInstsBeforeThisBB = NumInsts;
128   for (const Instruction &I : *BB) {
129     // Skip ephemeral values.
130     if (EphValues.count(&I))
131       continue;
132 
133     // Special handling for calls.
134     if (const auto *Call = dyn_cast<CallBase>(&I)) {
135       if (const Function *F = Call->getCalledFunction()) {
136         bool IsLoweredToCall = TTI.isLoweredToCall(F);
137         // If a function is both internal and has a single use, then it is
138         // extremely likely to get inlined in the future (it was probably
139         // exposed by an interleaved devirtualization pass).
140         // When preparing for LTO, liberally consider calls as inline
141         // candidates.
142         if (!Call->isNoInline() && IsLoweredToCall &&
143             ((F->hasInternalLinkage() && F->hasOneUse()) || PrepareForLTO)) {
144           ++NumInlineCandidates;
145         }
146 
147         // If this call is to function itself, then the function is recursive.
148         // Inlining it into other functions is a bad idea, because this is
149         // basically just a form of loop peeling, and our metrics aren't useful
150         // for that case.
151         if (F == BB->getParent())
152           isRecursive = true;
153 
154         if (IsLoweredToCall)
155           ++NumCalls;
156       } else {
157         // We don't want inline asm to count as a call - that would prevent loop
158         // unrolling. The argument setup cost is still real, though.
159         if (!Call->isInlineAsm())
160           ++NumCalls;
161       }
162     }
163 
164     if (const AllocaInst *AI = dyn_cast<AllocaInst>(&I)) {
165       if (!AI->isStaticAlloca())
166         this->usesDynamicAlloca = true;
167     }
168 
169     if (isa<ExtractElementInst>(I) || I.getType()->isVectorTy())
170       ++NumVectorInsts;
171 
172     if (I.getType()->isTokenTy() && I.isUsedOutsideOfBlock(BB))
173       notDuplicatable = true;
174 
175     if (const CallInst *CI = dyn_cast<CallInst>(&I)) {
176       if (CI->cannotDuplicate())
177         notDuplicatable = true;
178       if (CI->isConvergent())
179         convergent = true;
180     }
181 
182     if (const InvokeInst *InvI = dyn_cast<InvokeInst>(&I))
183       if (InvI->cannotDuplicate())
184         notDuplicatable = true;
185 
186     NumInstsProxy += TTI.getUserCost(&I, TargetTransformInfo::TCK_CodeSize);
187     NumInsts = *NumInstsProxy.getValue();
188   }
189 
190   if (isa<ReturnInst>(BB->getTerminator()))
191     ++NumRets;
192 
193   // We never want to inline functions that contain an indirectbr.  This is
194   // incorrect because all the blockaddress's (in static global initializers
195   // for example) would be referring to the original function, and this indirect
196   // jump would jump from the inlined copy of the function into the original
197   // function which is extremely undefined behavior.
198   // FIXME: This logic isn't really right; we can safely inline functions
199   // with indirectbr's as long as no other function or global references the
200   // blockaddress of a block within the current function.  And as a QOI issue,
201   // if someone is using a blockaddress without an indirectbr, and that
202   // reference somehow ends up in another function or global, we probably
203   // don't want to inline this function.
204   notDuplicatable |= isa<IndirectBrInst>(BB->getTerminator());
205 
206   // Remember NumInsts for this BB.
207   InstructionCost NumInstsThisBB = NumInstsProxy - NumInstsBeforeThisBB;
208   NumBBInsts[BB] = *NumInstsThisBB.getValue();
209 }
210