1//===- ModuleInliner.cpp - Code related to module inliner -----------------===//
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 the mechanics required to implement inlining without
10// missing any calls in the module level. It doesn't need any infromation about
11// SCC or call graph, which is different from the SCC inliner. The decisions of
12// which calls are profitable to inline are implemented elsewhere.
13//
14//===----------------------------------------------------------------------===//
15
16#include "llvm/Transforms/IPO/ModuleInliner.h"
17#include "llvm/ADT/ScopeExit.h"
18#include "llvm/ADT/SmallVector.h"
19#include "llvm/ADT/Statistic.h"
20#include "llvm/Analysis/AliasAnalysis.h"
21#include "llvm/Analysis/AssumptionCache.h"
22#include "llvm/Analysis/BlockFrequencyInfo.h"
23#include "llvm/Analysis/CtxProfAnalysis.h"
24#include "llvm/Analysis/InlineAdvisor.h"
25#include "llvm/Analysis/InlineCost.h"
26#include "llvm/Analysis/InlineOrder.h"
27#include "llvm/Analysis/OptimizationRemarkEmitter.h"
28#include "llvm/Analysis/ProfileSummaryInfo.h"
29#include "llvm/Analysis/ReplayInlineAdvisor.h"
30#include "llvm/Analysis/TargetLibraryInfo.h"
31#include "llvm/IR/DiagnosticInfo.h"
32#include "llvm/IR/Function.h"
33#include "llvm/IR/InstIterator.h"
34#include "llvm/IR/Instruction.h"
35#include "llvm/IR/IntrinsicInst.h"
36#include "llvm/IR/Module.h"
37#include "llvm/IR/PassManager.h"
38#include "llvm/Support/CommandLine.h"
39#include "llvm/Support/Debug.h"
40#include "llvm/Support/raw_ostream.h"
41#include "llvm/Transforms/Utils/CallPromotionUtils.h"
42#include "llvm/Transforms/Utils/Cloning.h"
43#include <cassert>
44
45using namespace llvm;
46
47#define DEBUG_TYPE "module-inline"
48
49STATISTIC(NumInlined, "Number of functions inlined");
50STATISTIC(NumDeleted, "Number of functions deleted because all callers found");
51
52static cl::opt<bool> CtxProfPromoteAlwaysInline(
53 "ctx-prof-promote-alwaysinline", cl::init(Val: false), cl::Hidden,
54 cl::desc("If using a contextual profile in this module, and an indirect "
55 "call target is marked as alwaysinline, perform indirect call "
56 "promotion for that target. If multiple targets for an indirect "
57 "call site fit this description, they are all promoted."));
58
59InlineAdvisor &ModuleInlinerPass::getAdvisor(const ModuleAnalysisManager &MAM,
60 FunctionAnalysisManager &FAM,
61 Module &M) {
62 if (OwnedAdvisor)
63 return *OwnedAdvisor;
64
65 auto *IAA = MAM.getCachedResult<InlineAdvisorAnalysis>(IR&: M);
66 if (!IAA) {
67 // It should still be possible to run the inliner as a stand-alone module
68 // pass, for test scenarios. In that case, we default to the
69 // DefaultInlineAdvisor, which doesn't need to keep state between module
70 // pass runs. It also uses just the default InlineParams. In this case, we
71 // need to use the provided FAM, which is valid for the duration of the
72 // inliner pass, and thus the lifetime of the owned advisor. The one we
73 // would get from the MAM can be invalidated as a result of the inliner's
74 // activity.
75 OwnedAdvisor = std::make_unique<DefaultInlineAdvisor>(
76 args&: M, args&: FAM, args: Params, args: InlineContext{.LTOPhase: LTOPhase, .Pass: InlinePass::ModuleInliner});
77
78 return *OwnedAdvisor;
79 }
80 assert(IAA->getAdvisor() &&
81 "Expected a present InlineAdvisorAnalysis also have an "
82 "InlineAdvisor initialized");
83 return *IAA->getAdvisor();
84}
85
86static bool isKnownLibFunction(Function &F, TargetLibraryInfo &TLI) {
87 // Either this is a normal library function or a "vectorizable"
88 // function. Not using the VFDatabase here because this query
89 // is related only to libraries handled via the TLI.
90 return TLI.getLibFunc(FDecl: F) != NotLibFunc ||
91 TLI.isKnownVectorFunctionInLibrary(F: F.getName());
92}
93
94PreservedAnalyses ModuleInlinerPass::run(Module &M,
95 ModuleAnalysisManager &MAM) {
96 LLVM_DEBUG(dbgs() << "---- Module Inliner is Running ---- \n");
97
98 auto &IAA = MAM.getResult<InlineAdvisorAnalysis>(IR&: M);
99 if (!IAA.tryCreate(Params, Mode, ReplaySettings: {},
100 IC: InlineContext{.LTOPhase: LTOPhase, .Pass: InlinePass::ModuleInliner})) {
101 M.getContext().emitError(
102 ErrorStr: "Could not setup Inlining Advisor for the requested "
103 "mode and/or options");
104 return PreservedAnalyses::all();
105 }
106
107 auto &CtxProf = MAM.getResult<CtxProfAnalysis>(IR&: M);
108
109 bool Changed = false;
110
111 ProfileSummaryInfo *PSI = MAM.getCachedResult<ProfileSummaryAnalysis>(IR&: M);
112
113 FunctionAnalysisManager &FAM =
114 MAM.getResult<FunctionAnalysisManagerModuleProxy>(IR&: M).getManager();
115
116 auto GetTLI = [&FAM](Function &F) -> TargetLibraryInfo & {
117 return FAM.getResult<TargetLibraryAnalysis>(IR&: F);
118 };
119
120 InlineAdvisor &Advisor = getAdvisor(MAM, FAM, M);
121 Advisor.onPassEntry();
122
123 llvm::scope_exit AdvisorOnExit([&] { Advisor.onPassExit(); });
124
125 // In the module inliner, a priority-based worklist is used for calls across
126 // the entire Module. With this module inliner, the inline order is not
127 // limited to bottom-up order. More globally scope inline order is enabled.
128 // Also, the inline deferral logic become unnecessary in this module inliner.
129 // It is possible to use other priority heuristics, e.g. profile-based
130 // heuristic.
131 //
132 // TODO: Here is a huge amount duplicate code between the module inliner and
133 // the SCC inliner, which need some refactoring.
134 auto Calls = getInlineOrder(FAM, Params, MAM, M);
135 assert(Calls != nullptr && "Expected an initialized InlineOrder");
136
137 // Populate the initial list of calls in this module.
138 SetVector<std::pair<CallBase *, Function *>> ICPCandidates;
139 for (Function &F : M) {
140 if (F.isDeclaration())
141 continue;
142 auto &ORE = FAM.getResult<OptimizationRemarkEmitterAnalysis>(IR&: F);
143 for (Instruction &I : instructions(F)) {
144 if (auto *CB = dyn_cast<CallBase>(Val: &I)) {
145 if (Function *Callee = CB->getCalledFunction()) {
146 if (!Callee->isDeclaration())
147 Calls->push(Elt: CB);
148 else if (!isa<IntrinsicInst>(Val: I)) {
149 using namespace ore;
150 setInlineRemark(CB&: *CB, Message: "unavailable definition");
151 ORE.emit(RemarkBuilder: [&]() {
152 return OptimizationRemarkMissed(DEBUG_TYPE, "NoDefinition", &I)
153 << NV("Callee", Callee) << " will not be inlined into "
154 << NV("Caller", CB->getCaller())
155 << " because its definition is unavailable"
156 << setIsVerbose();
157 });
158 }
159 } else if (CtxProfPromoteAlwaysInline &&
160 CtxProf.isInSpecializedModule() && CB->isIndirectCall()) {
161 CtxProfAnalysis::collectIndirectCallPromotionList(IC&: *CB, Profile&: CtxProf,
162 Candidates&: ICPCandidates);
163 }
164 }
165 }
166 }
167 for (auto &[CB, Target] : ICPCandidates) {
168 if (auto *DirectCB = promoteCallWithIfThenElse(CB&: *CB, Callee&: *Target, CtxProf))
169 Calls->push(Elt: DirectCB);
170 }
171 if (Calls->empty())
172 return PreservedAnalyses::all();
173
174 // Track the dead functions to delete once finished with inlining calls. We
175 // defer deleting these to make it easier to handle the call graph updates.
176 SmallVector<Function *, 4> DeadFunctions;
177
178 // Loop forward over all of the calls.
179 while (!Calls->empty()) {
180 CallBase *CB = Calls->pop();
181 Function &F = *CB->getCaller();
182 Function &Callee = *CB->getCalledFunction();
183
184 LLVM_DEBUG(dbgs() << "Inlining calls in: " << F.getName() << "\n"
185 << " Function size: " << F.getInstructionCount()
186 << "\n");
187 (void)F;
188
189 auto GetAssumptionCache = [&](Function &F) -> AssumptionCache & {
190 return FAM.getResult<AssumptionAnalysis>(IR&: F);
191 };
192
193 auto Advice = Advisor.getAdvice(CB&: *CB, /*OnlyMandatory*/ MandatoryOnly: false);
194 // Check whether we want to inline this callsite.
195 if (!Advice->isInliningRecommended()) {
196 Advice->recordUnattemptedInlining();
197 continue;
198 }
199
200 // Setup the data structure used to plumb customization into the
201 // `InlineFunction` routine.
202 InlineFunctionInfo IFI(
203 GetAssumptionCache, PSI,
204 &FAM.getResult<BlockFrequencyAnalysis>(IR&: *(CB->getCaller())),
205 &FAM.getResult<BlockFrequencyAnalysis>(IR&: Callee));
206
207 InlineResult IR =
208 InlineFunction(CB&: *CB, IFI, CtxProf, /*MergeAttributes=*/true,
209 CalleeAAR: &FAM.getResult<AAManager>(IR&: *CB->getCaller()),
210 /*InsertLifetime=*/true,
211 /*TrackInlineHistory=*/true);
212 if (!IR.isSuccess()) {
213 Advice->recordUnsuccessfulInlining(Result: IR);
214 continue;
215 }
216
217 Changed = true;
218 ++NumInlined;
219
220 LLVM_DEBUG(dbgs() << " Size after inlining: " << F.getInstructionCount()
221 << "\n");
222
223 // Add any new callsites to defined functions to the worklist.
224 if (!IFI.InlinedCallSites.empty()) {
225 for (CallBase *ICB : reverse(C&: IFI.InlinedCallSites)) {
226 Function *NewCallee = ICB->getCalledFunction();
227 if (!NewCallee) {
228 // Try to promote an indirect (virtual) call without waiting for
229 // the post-inline cleanup and the next DevirtSCCRepeatedPass
230 // iteration because the next iteration may not happen and we may
231 // miss inlining it.
232 // FIXME: enable for ctxprof.
233 if (CtxProf.isInSpecializedModule())
234 if (tryPromoteCall(CB&: *ICB))
235 NewCallee = ICB->getCalledFunction();
236 }
237 if (NewCallee)
238 if (!NewCallee->isDeclaration())
239 Calls->push(Elt: ICB);
240 }
241 }
242
243 // For local functions, check whether this makes the callee trivially
244 // dead. In that case, we can drop the body of the function eagerly
245 // which may reduce the number of callers of other functions to one,
246 // changing inline cost thresholds.
247 bool CalleeWasDeleted = false;
248 if (Callee.hasLocalLinkage()) {
249 // To check this we also need to nuke any dead constant uses (perhaps
250 // made dead by this operation on other functions).
251 Callee.removeDeadConstantUsers();
252 // if (Callee.use_empty() && !CG.isLibFunction(Callee)) {
253 if (Callee.use_empty() && !isKnownLibFunction(F&: Callee, TLI&: GetTLI(Callee))) {
254 Calls->erase_if(
255 Pred: [&](const CallBase *CB) { return CB->getCaller() == &Callee; });
256
257 // Report inlining decision BEFORE deleting function contents, so we
258 // can still access e.g. the DebugLoc
259 Advice->recordInliningWithCalleeDeleted();
260 // Clear the body and queue the function itself for deletion when we
261 // finish inlining.
262 // Note that after this point, it is an error to do anything other
263 // than use the callee's address or delete it.
264 Callee.dropAllReferences();
265 assert(!is_contained(DeadFunctions, &Callee) &&
266 "Cannot put cause a function to become dead twice!");
267 DeadFunctions.push_back(Elt: &Callee);
268 CalleeWasDeleted = true;
269 }
270 }
271 if (!CalleeWasDeleted)
272 Advice->recordInlining();
273 }
274
275 // Now that we've finished inlining all of the calls across this module,
276 // delete all of the trivially dead functions.
277 //
278 // Note that this walks a pointer set which has non-deterministic order but
279 // that is OK as all we do is delete things and add pointers to unordered
280 // sets.
281 for (Function *DeadF : DeadFunctions) {
282 // Clear out any cached analyses.
283 FAM.clear(IR&: *DeadF, Name: DeadF->getName());
284
285 // And delete the actual function from the module.
286 M.getFunctionList().erase(IT: DeadF);
287
288 ++NumDeleted;
289 }
290
291 if (!Changed)
292 return PreservedAnalyses::all();
293
294 return PreservedAnalyses::none();
295}
296