1 //===- CGSCCPassManager.cpp - Managing & running CGSCC passes -------------===//
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 #include "llvm/Analysis/CGSCCPassManager.h"
11 #include "llvm/IR/CallSite.h"
12 #include "llvm/IR/InstIterator.h"
16 // Explicit template instantiations and specialization defininitions for core
20 // Explicit instantiations for the core proxy templates.
21 template class AllAnalysesOn<LazyCallGraph::SCC>;
22 template class AnalysisManager<LazyCallGraph::SCC, LazyCallGraph &>;
23 template class PassManager<LazyCallGraph::SCC, CGSCCAnalysisManager,
24 LazyCallGraph &, CGSCCUpdateResult &>;
25 template class InnerAnalysisManagerProxy<CGSCCAnalysisManager, Module>;
26 template class OuterAnalysisManagerProxy<ModuleAnalysisManager,
27 LazyCallGraph::SCC, LazyCallGraph &>;
28 template class OuterAnalysisManagerProxy<CGSCCAnalysisManager, Function>;
30 /// Explicitly specialize the pass manager run method to handle call graph
34 PassManager<LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &,
35 CGSCCUpdateResult &>::run(LazyCallGraph::SCC &InitialC,
36 CGSCCAnalysisManager &AM,
37 LazyCallGraph &G, CGSCCUpdateResult &UR) {
38 PreservedAnalyses PA = PreservedAnalyses::all();
41 dbgs() << "Starting CGSCC pass manager run.\n";
43 // The SCC may be refined while we are running passes over it, so set up
44 // a pointer that we can update.
45 LazyCallGraph::SCC *C = &InitialC;
47 for (auto &Pass : Passes) {
49 dbgs() << "Running pass: " << Pass->name() << " on " << *C << "\n";
51 PreservedAnalyses PassPA = Pass->run(*C, AM, G, UR);
53 // Update the SCC if necessary.
54 C = UR.UpdatedC ? UR.UpdatedC : C;
56 // Check that we didn't miss any update scenario.
57 assert(!UR.InvalidatedSCCs.count(C) && "Processing an invalid SCC!");
58 assert(C->begin() != C->end() && "Cannot have an empty SCC!");
60 // Update the analysis manager as each pass runs and potentially
61 // invalidates analyses.
62 AM.invalidate(*C, PassPA);
64 // Finally, we intersect the final preserved analyses to compute the
65 // aggregate preserved set for this pass manager.
66 PA.intersect(std::move(PassPA));
68 // FIXME: Historically, the pass managers all called the LLVM context's
69 // yield function here. We don't have a generic way to acquire the
70 // context and it isn't yet clear what the right pattern is for yielding
71 // in the new pass manager so it is currently omitted.
72 // ...getContext().yield();
75 // Invaliadtion was handled after each pass in the above loop for the current
76 // SCC. Therefore, the remaining analysis results in the AnalysisManager are
77 // preserved. We mark this with a set so that we don't need to inspect each
79 PA.preserveSet<AllAnalysesOn<LazyCallGraph::SCC>>();
82 dbgs() << "Finished CGSCC pass manager run.\n";
87 bool CGSCCAnalysisManagerModuleProxy::Result::invalidate(
88 Module &M, const PreservedAnalyses &PA,
89 ModuleAnalysisManager::Invalidator &Inv) {
90 // If literally everything is preserved, we're done.
91 if (PA.areAllPreserved())
92 return false; // This is still a valid proxy.
94 // If this proxy or the call graph is going to be invalidated, we also need
95 // to clear all the keys coming from that analysis.
97 // We also directly invalidate the FAM's module proxy if necessary, and if
98 // that proxy isn't preserved we can't preserve this proxy either. We rely on
99 // it to handle module -> function analysis invalidation in the face of
100 // structural changes and so if it's unavailable we conservatively clear the
101 // entire SCC layer as well rather than trying to do invalidation ourselves.
102 auto PAC = PA.getChecker<CGSCCAnalysisManagerModuleProxy>();
103 if (!(PAC.preserved() || PAC.preservedSet<AllAnalysesOn<Module>>()) ||
104 Inv.invalidate<LazyCallGraphAnalysis>(M, PA) ||
105 Inv.invalidate<FunctionAnalysisManagerModuleProxy>(M, PA)) {
108 // And the proxy itself should be marked as invalid so that we can observe
109 // the new call graph. This isn't strictly necessary because we cheat
110 // above, but is still useful.
114 // Directly check if the relevant set is preserved so we can short circuit
115 // invalidating SCCs below.
116 bool AreSCCAnalysesPreserved =
117 PA.allAnalysesInSetPreserved<AllAnalysesOn<LazyCallGraph::SCC>>();
119 // Ok, we have a graph, so we can propagate the invalidation down into it.
120 for (auto &RC : G->postorder_ref_sccs())
122 Optional<PreservedAnalyses> InnerPA;
124 // Check to see whether the preserved set needs to be adjusted based on
125 // module-level analysis invalidation triggering deferred invalidation
127 if (auto *OuterProxy =
128 InnerAM->getCachedResult<ModuleAnalysisManagerCGSCCProxy>(C))
129 for (const auto &OuterInvalidationPair :
130 OuterProxy->getOuterInvalidations()) {
131 AnalysisKey *OuterAnalysisID = OuterInvalidationPair.first;
132 const auto &InnerAnalysisIDs = OuterInvalidationPair.second;
133 if (Inv.invalidate(OuterAnalysisID, M, PA)) {
136 for (AnalysisKey *InnerAnalysisID : InnerAnalysisIDs)
137 InnerPA->abandon(InnerAnalysisID);
141 // Check if we needed a custom PA set. If so we'll need to run the inner
144 InnerAM->invalidate(C, *InnerPA);
148 // Otherwise we only need to do invalidation if the original PA set didn't
149 // preserve all SCC analyses.
150 if (!AreSCCAnalysesPreserved)
151 InnerAM->invalidate(C, PA);
154 // Return false to indicate that this result is still a valid proxy.
159 CGSCCAnalysisManagerModuleProxy::Result
160 CGSCCAnalysisManagerModuleProxy::run(Module &M, ModuleAnalysisManager &AM) {
161 // Force the Function analysis manager to also be available so that it can
162 // be accessed in an SCC analysis and proxied onward to function passes.
163 // FIXME: It is pretty awkward to just drop the result here and assert that
164 // we can find it again later.
165 (void)AM.getResult<FunctionAnalysisManagerModuleProxy>(M);
167 return Result(*InnerAM, AM.getResult<LazyCallGraphAnalysis>(M));
170 AnalysisKey FunctionAnalysisManagerCGSCCProxy::Key;
172 FunctionAnalysisManagerCGSCCProxy::Result
173 FunctionAnalysisManagerCGSCCProxy::run(LazyCallGraph::SCC &C,
174 CGSCCAnalysisManager &AM,
176 // Collect the FunctionAnalysisManager from the Module layer and use that to
177 // build the proxy result.
179 // This allows us to rely on the FunctionAnalysisMangaerModuleProxy to
180 // invalidate the function analyses.
181 auto &MAM = AM.getResult<ModuleAnalysisManagerCGSCCProxy>(C, CG).getManager();
182 Module &M = *C.begin()->getFunction().getParent();
183 auto *FAMProxy = MAM.getCachedResult<FunctionAnalysisManagerModuleProxy>(M);
184 assert(FAMProxy && "The CGSCC pass manager requires that the FAM module "
185 "proxy is run on the module prior to entering the CGSCC "
188 // Note that we special-case invalidation handling of this proxy in the CGSCC
189 // analysis manager's Module proxy. This avoids the need to do anything
190 // special here to recompute all of this if ever the FAM's module proxy goes
192 return Result(FAMProxy->getManager());
195 bool FunctionAnalysisManagerCGSCCProxy::Result::invalidate(
196 LazyCallGraph::SCC &C, const PreservedAnalyses &PA,
197 CGSCCAnalysisManager::Invalidator &Inv) {
198 for (LazyCallGraph::Node &N : C)
199 FAM->invalidate(N.getFunction(), PA);
201 // This proxy doesn't need to handle invalidation itself. Instead, the
202 // module-level CGSCC proxy handles it above by ensuring that if the
203 // module-level FAM proxy becomes invalid the entire SCC layer, which
204 // includes this proxy, is cleared.
208 } // End llvm namespace
211 /// Helper function to update both the \c CGSCCAnalysisManager \p AM and the \c
212 /// CGSCCPassManager's \c CGSCCUpdateResult \p UR based on a range of newly
215 /// The range of new SCCs must be in postorder already. The SCC they were split
216 /// out of must be provided as \p C. The current node being mutated and
217 /// triggering updates must be passed as \p N.
219 /// This function returns the SCC containing \p N. This will be either \p C if
220 /// no new SCCs have been split out, or it will be the new SCC containing \p N.
221 template <typename SCCRangeT>
223 incorporateNewSCCRange(const SCCRangeT &NewSCCRange, LazyCallGraph &G,
224 LazyCallGraph::Node &N, LazyCallGraph::SCC *C,
225 CGSCCAnalysisManager &AM, CGSCCUpdateResult &UR,
226 bool DebugLogging = false) {
227 typedef LazyCallGraph::SCC SCC;
229 if (NewSCCRange.begin() == NewSCCRange.end())
232 // Add the current SCC to the worklist as its shape has changed.
233 UR.CWorklist.insert(C);
235 dbgs() << "Enqueuing the existing SCC in the worklist:" << *C << "\n";
240 // Update the current SCC. Note that if we have new SCCs, this must actually
242 assert(C != &*NewSCCRange.begin() &&
243 "Cannot insert new SCCs without changing current SCC!");
244 C = &*NewSCCRange.begin();
245 assert(G.lookupSCC(N) == C && "Failed to update current SCC!");
248 reverse(make_range(std::next(NewSCCRange.begin()), NewSCCRange.end()))) {
249 assert(C != &NewC && "No need to re-visit the current SCC!");
250 assert(OldC != &NewC && "Already handled the original SCC!");
251 UR.CWorklist.insert(&NewC);
253 dbgs() << "Enqueuing a newly formed SCC:" << NewC << "\n";
259 LazyCallGraph::SCC &llvm::updateCGAndAnalysisManagerForFunctionPass(
260 LazyCallGraph &G, LazyCallGraph::SCC &InitialC, LazyCallGraph::Node &N,
261 CGSCCAnalysisManager &AM, CGSCCUpdateResult &UR, bool DebugLogging) {
262 typedef LazyCallGraph::Node Node;
263 typedef LazyCallGraph::Edge Edge;
264 typedef LazyCallGraph::SCC SCC;
265 typedef LazyCallGraph::RefSCC RefSCC;
267 RefSCC &InitialRC = InitialC.getOuterRefSCC();
269 RefSCC *RC = &InitialRC;
270 Function &F = N.getFunction();
272 // Walk the function body and build up the set of retained, promoted, and
274 SmallVector<Constant *, 16> Worklist;
275 SmallPtrSet<Constant *, 16> Visited;
276 SmallPtrSet<Function *, 16> RetainedEdges;
277 SmallSetVector<Function *, 4> PromotedRefTargets;
278 SmallSetVector<Function *, 4> DemotedCallTargets;
280 // First walk the function and handle all called functions. We do this first
281 // because if there is a single call edge, whether there are ref edges is
283 for (Instruction &I : instructions(F))
284 if (auto CS = CallSite(&I))
285 if (Function *Callee = CS.getCalledFunction())
286 if (Visited.insert(Callee).second && !Callee->isDeclaration()) {
287 const Edge *E = N.lookup(*Callee);
288 // FIXME: We should really handle adding new calls. While it will
289 // make downstream usage more complex, there is no fundamental
290 // limitation and it will allow passes within the CGSCC to be a bit
291 // more flexible in what transforms they can do. Until then, we
292 // verify that new calls haven't been introduced.
293 assert(E && "No function transformations should introduce *new* "
294 "call edges! Any new calls should be modeled as "
295 "promoted existing ref edges!");
296 RetainedEdges.insert(Callee);
298 PromotedRefTargets.insert(Callee);
301 // Now walk all references.
302 for (Instruction &I : instructions(F))
303 for (Value *Op : I.operand_values())
304 if (Constant *C = dyn_cast<Constant>(Op))
305 if (Visited.insert(C).second)
306 Worklist.push_back(C);
308 LazyCallGraph::visitReferences(Worklist, Visited, [&](Function &Referee) {
309 const Edge *E = N.lookup(Referee);
310 // FIXME: Similarly to new calls, we also currently preclude
311 // introducing new references. See above for details.
312 assert(E && "No function transformations should introduce *new* ref "
313 "edges! Any new ref edges would require IPO which "
314 "function passes aren't allowed to do!");
315 RetainedEdges.insert(&Referee);
317 DemotedCallTargets.insert(&Referee);
320 // First remove all of the edges that are no longer present in this function.
321 // We have to build a list of dead targets first and then remove them as the
322 // data structures will all be invalidated by removing them.
323 SmallVector<PointerIntPair<Node *, 1, Edge::Kind>, 4> DeadTargets;
325 if (!RetainedEdges.count(&E.getFunction()))
326 DeadTargets.push_back({E.getNode(), E.getKind()});
327 for (auto DeadTarget : DeadTargets) {
328 Node &TargetN = *DeadTarget.getPointer();
329 bool IsCall = DeadTarget.getInt() == Edge::Call;
330 SCC &TargetC = *G.lookupSCC(TargetN);
331 RefSCC &TargetRC = TargetC.getOuterRefSCC();
333 if (&TargetRC != RC) {
334 RC->removeOutgoingEdge(N, TargetN);
336 dbgs() << "Deleting outgoing edge from '" << N << "' to '" << TargetN
341 dbgs() << "Deleting internal " << (IsCall ? "call" : "ref")
342 << " edge from '" << N << "' to '" << TargetN << "'\n";
346 // For separate SCCs this is trivial.
347 RC->switchTrivialInternalEdgeToRef(N, TargetN);
349 // Otherwise we may end up re-structuring the call graph. First,
350 // invalidate any SCC analyses. We have to do this before we split
351 // functions into new SCCs and lose track of where their analyses are
353 // FIXME: We should accept a more precise preserved set here. For
354 // example, it might be possible to preserve some function analyses
355 // even as the SCC structure is changed.
356 AM.invalidate(*C, PreservedAnalyses::none());
357 // Now update the call graph.
358 C = incorporateNewSCCRange(RC->switchInternalEdgeToRef(N, TargetN), G,
359 N, C, AM, UR, DebugLogging);
363 auto NewRefSCCs = RC->removeInternalRefEdge(N, TargetN);
364 if (!NewRefSCCs.empty()) {
365 // Note that we don't bother to invalidate analyses as ref-edge
366 // connectivity is not really observable in any way and is intended
367 // exclusively to be used for ordering of transforms rather than for
368 // analysis conclusions.
370 // The RC worklist is in reverse postorder, so we first enqueue the
371 // current RefSCC as it will remain the parent of all split RefSCCs, then
372 // we enqueue the new ones in RPO except for the one which contains the
373 // source node as that is the "bottom" we will continue processing in the
375 UR.RCWorklist.insert(RC);
377 dbgs() << "Enqueuing the existing RefSCC in the update worklist: "
379 // Update the RC to the "bottom".
380 assert(G.lookupSCC(N) == C && "Changed the SCC when splitting RefSCCs!");
381 RC = &C->getOuterRefSCC();
382 assert(G.lookupRefSCC(N) == RC && "Failed to update current RefSCC!");
383 assert(NewRefSCCs.front() == RC &&
384 "New current RefSCC not first in the returned list!");
385 for (RefSCC *NewRC : reverse(
386 make_range(std::next(NewRefSCCs.begin()), NewRefSCCs.end()))) {
387 assert(NewRC != RC && "Should not encounter the current RefSCC further "
388 "in the postorder list of new RefSCCs.");
389 UR.RCWorklist.insert(NewRC);
391 dbgs() << "Enqueuing a new RefSCC in the update worklist: " << *NewRC
397 // Next demote all the call edges that are now ref edges. This helps make
398 // the SCCs small which should minimize the work below as we don't want to
399 // form cycles that this would break.
400 for (Function *RefTarget : DemotedCallTargets) {
401 Node &TargetN = *G.lookup(*RefTarget);
402 SCC &TargetC = *G.lookupSCC(TargetN);
403 RefSCC &TargetRC = TargetC.getOuterRefSCC();
405 // The easy case is when the target RefSCC is not this RefSCC. This is
406 // only supported when the target RefSCC is a child of this RefSCC.
407 if (&TargetRC != RC) {
408 assert(RC->isAncestorOf(TargetRC) &&
409 "Cannot potentially form RefSCC cycles here!");
410 RC->switchOutgoingEdgeToRef(N, TargetN);
412 dbgs() << "Switch outgoing call edge to a ref edge from '" << N
413 << "' to '" << TargetN << "'\n";
417 // We are switching an internal call edge to a ref edge. This may split up
420 // For separate SCCs this is trivial.
421 RC->switchTrivialInternalEdgeToRef(N, TargetN);
425 // Otherwise we may end up re-structuring the call graph. First, invalidate
426 // any SCC analyses. We have to do this before we split functions into new
427 // SCCs and lose track of where their analyses are cached.
428 // FIXME: We should accept a more precise preserved set here. For example,
429 // it might be possible to preserve some function analyses even as the SCC
430 // structure is changed.
431 AM.invalidate(*C, PreservedAnalyses::none());
432 // Now update the call graph.
433 C = incorporateNewSCCRange(RC->switchInternalEdgeToRef(N, TargetN), G,
434 N, C, AM, UR, DebugLogging);
437 // Now promote ref edges into call edges.
438 for (Function *CallTarget : PromotedRefTargets) {
439 Node &TargetN = *G.lookup(*CallTarget);
440 SCC &TargetC = *G.lookupSCC(TargetN);
441 RefSCC &TargetRC = TargetC.getOuterRefSCC();
443 // The easy case is when the target RefSCC is not this RefSCC. This is
444 // only supported when the target RefSCC is a child of this RefSCC.
445 if (&TargetRC != RC) {
446 assert(RC->isAncestorOf(TargetRC) &&
447 "Cannot potentially form RefSCC cycles here!");
448 RC->switchOutgoingEdgeToCall(N, TargetN);
450 dbgs() << "Switch outgoing ref edge to a call edge from '" << N
451 << "' to '" << TargetN << "'\n";
455 dbgs() << "Switch an internal ref edge to a call edge from '" << N
456 << "' to '" << TargetN << "'\n";
458 // Otherwise we are switching an internal ref edge to a call edge. This
459 // may merge away some SCCs, and we add those to the UpdateResult. We also
460 // need to make sure to update the worklist in the event SCCs have moved
461 // before the current one in the post-order sequence.
462 auto InitialSCCIndex = RC->find(*C) - RC->begin();
463 auto InvalidatedSCCs = RC->switchInternalEdgeToCall(N, TargetN);
464 if (!InvalidatedSCCs.empty()) {
466 assert(G.lookupSCC(N) == C && "Failed to update current SCC!");
468 // Any analyses cached for this SCC are no longer precise as the shape
469 // has changed by introducing this cycle.
470 AM.invalidate(*C, PreservedAnalyses::none());
472 for (SCC *InvalidatedC : InvalidatedSCCs) {
473 assert(InvalidatedC != C && "Cannot invalidate the current SCC!");
474 UR.InvalidatedSCCs.insert(InvalidatedC);
476 // Also clear any cached analyses for the SCCs that are dead. This
477 // isn't really necessary for correctness but can release memory.
478 AM.clear(*InvalidatedC);
481 auto NewSCCIndex = RC->find(*C) - RC->begin();
482 if (InitialSCCIndex < NewSCCIndex) {
483 // Put our current SCC back onto the worklist as we'll visit other SCCs
484 // that are now definitively ordered prior to the current one in the
485 // post-order sequence, and may end up observing more precise context to
486 // optimize the current SCC.
487 UR.CWorklist.insert(C);
489 dbgs() << "Enqueuing the existing SCC in the worklist: " << *C << "\n";
490 // Enqueue in reverse order as we pop off the back of the worklist.
491 for (SCC &MovedC : reverse(make_range(RC->begin() + InitialSCCIndex,
492 RC->begin() + NewSCCIndex))) {
493 UR.CWorklist.insert(&MovedC);
495 dbgs() << "Enqueuing a newly earlier in post-order SCC: " << MovedC
501 assert(!UR.InvalidatedSCCs.count(C) && "Invalidated the current SCC!");
502 assert(!UR.InvalidatedRefSCCs.count(RC) && "Invalidated the current RefSCC!");
503 assert(&C->getOuterRefSCC() == RC && "Current SCC not in current RefSCC!");
505 // Record the current RefSCC and SCC for higher layers of the CGSCC pass
506 // manager now that all the updates have been applied.
507 if (RC != &InitialRC)