xref: /freebsd/contrib/llvm-project/llvm/utils/TableGen/Common/CodeGenSchedule.cpp (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
1 //===- CodeGenSchedule.cpp - Scheduling MachineModels ---------------------===//
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 defines structures to encapsulate the machine model as described in
10 // the target description.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "CodeGenSchedule.h"
15 #include "CodeGenInstruction.h"
16 #include "CodeGenTarget.h"
17 #include "llvm/ADT/MapVector.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/ADT/SmallPtrSet.h"
20 #include "llvm/ADT/SmallVector.h"
21 #include "llvm/Support/Casting.h"
22 #include "llvm/Support/Debug.h"
23 #include "llvm/Support/Regex.h"
24 #include "llvm/Support/raw_ostream.h"
25 #include "llvm/TableGen/Error.h"
26 #include <algorithm>
27 #include <iterator>
28 #include <utility>
29 
30 using namespace llvm;
31 
32 #define DEBUG_TYPE "subtarget-emitter"
33 
34 #ifndef NDEBUG
dumpIdxVec(ArrayRef<unsigned> V)35 static void dumpIdxVec(ArrayRef<unsigned> V) {
36   for (unsigned Idx : V)
37     dbgs() << Idx << ", ";
38 }
39 #endif
40 
41 namespace {
42 
43 // (instrs a, b, ...) Evaluate and union all arguments. Identical to AddOp.
44 struct InstrsOp : public SetTheory::Operator {
apply__anona30522b50111::InstrsOp45   void apply(SetTheory &ST, DagInit *Expr, SetTheory::RecSet &Elts,
46              ArrayRef<SMLoc> Loc) override {
47     ST.evaluate(Expr->arg_begin(), Expr->arg_end(), Elts, Loc);
48   }
49 };
50 
51 // (instregex "OpcPat",...) Find all instructions matching an opcode pattern.
52 struct InstRegexOp : public SetTheory::Operator {
53   const CodeGenTarget &Target;
InstRegexOp__anona30522b50111::InstRegexOp54   InstRegexOp(const CodeGenTarget &t) : Target(t) {}
55 
56   /// Remove any text inside of parentheses from S.
removeParens__anona30522b50111::InstRegexOp57   static std::string removeParens(llvm::StringRef S) {
58     std::string Result;
59     unsigned Paren = 0;
60     // NB: We don't care about escaped parens here.
61     for (char C : S) {
62       switch (C) {
63       case '(':
64         ++Paren;
65         break;
66       case ')':
67         --Paren;
68         break;
69       default:
70         if (Paren == 0)
71           Result += C;
72       }
73     }
74     return Result;
75   }
76 
apply__anona30522b50111::InstRegexOp77   void apply(SetTheory &ST, DagInit *Expr, SetTheory::RecSet &Elts,
78              ArrayRef<SMLoc> Loc) override {
79     ArrayRef<const CodeGenInstruction *> Instructions =
80         Target.getInstructionsByEnumValue();
81 
82     unsigned NumGeneric = Target.getNumFixedInstructions();
83     unsigned NumPseudos = Target.getNumPseudoInstructions();
84     auto Generics = Instructions.slice(0, NumGeneric);
85     auto Pseudos = Instructions.slice(NumGeneric, NumPseudos);
86     auto NonPseudos = Instructions.slice(NumGeneric + NumPseudos);
87 
88     for (Init *Arg : Expr->getArgs()) {
89       StringInit *SI = dyn_cast<StringInit>(Arg);
90       if (!SI)
91         PrintFatalError(Loc, "instregex requires pattern string: " +
92                                  Expr->getAsString());
93       StringRef Original = SI->getValue();
94       // Drop an explicit ^ anchor to not interfere with prefix search.
95       bool HadAnchor = Original.consume_front("^");
96 
97       // Extract a prefix that we can binary search on.
98       static const char RegexMetachars[] = "()^$|*+?.[]\\{}";
99       auto FirstMeta = Original.find_first_of(RegexMetachars);
100       if (FirstMeta != StringRef::npos && FirstMeta > 0) {
101         // If we have a regex like ABC* we can only use AB as the prefix, as
102         // the * acts on C.
103         switch (Original[FirstMeta]) {
104         case '+':
105         case '*':
106         case '?':
107           --FirstMeta;
108           break;
109         default:
110           break;
111         }
112       }
113 
114       // Look for top-level | or ?. We cannot optimize them to binary search.
115       if (removeParens(Original).find_first_of("|?") != std::string::npos)
116         FirstMeta = 0;
117 
118       std::optional<Regex> Regexpr;
119       StringRef Prefix = Original.substr(0, FirstMeta);
120       StringRef PatStr = Original.substr(FirstMeta);
121       if (!PatStr.empty()) {
122         // For the rest use a python-style prefix match.
123         std::string pat = std::string(PatStr);
124         // Add ^ anchor. If we had one originally, don't need the group.
125         if (HadAnchor) {
126           pat.insert(0, "^");
127         } else {
128           pat.insert(0, "^(");
129           pat.insert(pat.end(), ')');
130         }
131         Regexpr = Regex(pat);
132       }
133 
134       int NumMatches = 0;
135 
136       // The generic opcodes are unsorted, handle them manually.
137       for (auto *Inst : Generics) {
138         StringRef InstName = Inst->TheDef->getName();
139         if (InstName.starts_with(Prefix) &&
140             (!Regexpr || Regexpr->match(InstName.substr(Prefix.size())))) {
141           Elts.insert(Inst->TheDef);
142           NumMatches++;
143         }
144       }
145 
146       // Target instructions are split into two ranges: pseudo instructions
147       // first, than non-pseudos. Each range is in lexicographical order
148       // sorted by name. Find the sub-ranges that start with our prefix.
149       struct Comp {
150         bool operator()(const CodeGenInstruction *LHS, StringRef RHS) {
151           return LHS->TheDef->getName() < RHS;
152         }
153         bool operator()(StringRef LHS, const CodeGenInstruction *RHS) {
154           return LHS < RHS->TheDef->getName() &&
155                  !RHS->TheDef->getName().starts_with(LHS);
156         }
157       };
158       auto Range1 =
159           std::equal_range(Pseudos.begin(), Pseudos.end(), Prefix, Comp());
160       auto Range2 = std::equal_range(NonPseudos.begin(), NonPseudos.end(),
161                                      Prefix, Comp());
162 
163       // For these ranges we know that instruction names start with the prefix.
164       // Check if there's a regex that needs to be checked.
165       const auto HandleNonGeneric = [&](const CodeGenInstruction *Inst) {
166         StringRef InstName = Inst->TheDef->getName();
167         if (!Regexpr || Regexpr->match(InstName.substr(Prefix.size()))) {
168           Elts.insert(Inst->TheDef);
169           NumMatches++;
170         }
171       };
172       std::for_each(Range1.first, Range1.second, HandleNonGeneric);
173       std::for_each(Range2.first, Range2.second, HandleNonGeneric);
174 
175       if (0 == NumMatches)
176         PrintFatalError(Loc, "instregex has no matches: " + Original);
177     }
178   }
179 };
180 
181 } // end anonymous namespace
182 
183 /// CodeGenModels ctor interprets machine model records and populates maps.
CodeGenSchedModels(RecordKeeper & RK,const CodeGenTarget & TGT)184 CodeGenSchedModels::CodeGenSchedModels(RecordKeeper &RK,
185                                        const CodeGenTarget &TGT)
186     : Records(RK), Target(TGT) {
187 
188   Sets.addFieldExpander("InstRW", "Instrs");
189 
190   // Allow Set evaluation to recognize the dags used in InstRW records:
191   // (instrs Op1, Op1...)
192   Sets.addOperator("instrs", std::make_unique<InstrsOp>());
193   Sets.addOperator("instregex", std::make_unique<InstRegexOp>(Target));
194 
195   // Instantiate a CodeGenProcModel for each SchedMachineModel with the values
196   // that are explicitly referenced in tablegen records. Resources associated
197   // with each processor will be derived later. Populate ProcModelMap with the
198   // CodeGenProcModel instances.
199   collectProcModels();
200 
201   // Instantiate a CodeGenSchedRW for each SchedReadWrite record explicitly
202   // defined, and populate SchedReads and SchedWrites vectors. Implicit
203   // SchedReadWrites that represent sequences derived from expanded variant will
204   // be inferred later.
205   collectSchedRW();
206 
207   // Instantiate a CodeGenSchedClass for each unique SchedRW signature directly
208   // required by an instruction definition, and populate SchedClassIdxMap. Set
209   // NumItineraryClasses to the number of explicit itinerary classes referenced
210   // by instructions. Set NumInstrSchedClasses to the number of itinerary
211   // classes plus any classes implied by instructions that derive from class
212   // Sched and provide SchedRW list. This does not infer any new classes from
213   // SchedVariant.
214   collectSchedClasses();
215 
216   // Find instruction itineraries for each processor. Sort and populate
217   // CodeGenProcModel::ItinDefList. (Cycle-to-cycle itineraries). This requires
218   // all itinerary classes to be discovered.
219   collectProcItins();
220 
221   // Find ItinRW records for each processor and itinerary class.
222   // (For per-operand resources mapped to itinerary classes).
223   collectProcItinRW();
224 
225   // Find UnsupportedFeatures records for each processor.
226   // (For per-operand resources mapped to itinerary classes).
227   collectProcUnsupportedFeatures();
228 
229   // Infer new SchedClasses from SchedVariant.
230   inferSchedClasses();
231 
232   // Populate each CodeGenProcModel's WriteResDefs, ReadAdvanceDefs, and
233   // ProcResourceDefs.
234   LLVM_DEBUG(
235       dbgs() << "\n+++ RESOURCE DEFINITIONS (collectProcResources) +++\n");
236   collectProcResources();
237 
238   // Collect optional processor description.
239   collectOptionalProcessorInfo();
240 
241   // Check MCInstPredicate definitions.
242   checkMCInstPredicates();
243 
244   // Check STIPredicate definitions.
245   checkSTIPredicates();
246 
247   // Find STIPredicate definitions for each processor model, and construct
248   // STIPredicateFunction objects.
249   collectSTIPredicates();
250 
251   checkCompleteness();
252 }
253 
checkSTIPredicates() const254 void CodeGenSchedModels::checkSTIPredicates() const {
255   DenseMap<StringRef, const Record *> Declarations;
256 
257   // There cannot be multiple declarations with the same name.
258   const RecVec Decls = Records.getAllDerivedDefinitions("STIPredicateDecl");
259   for (const Record *R : Decls) {
260     StringRef Name = R->getValueAsString("Name");
261     const auto It = Declarations.find(Name);
262     if (It == Declarations.end()) {
263       Declarations[Name] = R;
264       continue;
265     }
266 
267     PrintError(R->getLoc(), "STIPredicate " + Name + " multiply declared.");
268     PrintFatalNote(It->second->getLoc(), "Previous declaration was here.");
269   }
270 
271   // Disallow InstructionEquivalenceClasses with an empty instruction list.
272   const RecVec Defs =
273       Records.getAllDerivedDefinitions("InstructionEquivalenceClass");
274   for (const Record *R : Defs) {
275     RecVec Opcodes = R->getValueAsListOfDefs("Opcodes");
276     if (Opcodes.empty()) {
277       PrintFatalError(R->getLoc(), "Invalid InstructionEquivalenceClass "
278                                    "defined with an empty opcode list.");
279     }
280   }
281 }
282 
283 // Used by function `processSTIPredicate` to construct a mask of machine
284 // instruction operands.
constructOperandMask(ArrayRef<int64_t> Indices)285 static APInt constructOperandMask(ArrayRef<int64_t> Indices) {
286   APInt OperandMask;
287   if (Indices.empty())
288     return OperandMask;
289 
290   int64_t MaxIndex = *llvm::max_element(Indices);
291   assert(MaxIndex >= 0 && "Invalid negative indices in input!");
292   OperandMask = OperandMask.zext(MaxIndex + 1);
293   for (const int64_t Index : Indices) {
294     assert(Index >= 0 && "Invalid negative indices!");
295     OperandMask.setBit(Index);
296   }
297 
298   return OperandMask;
299 }
300 
processSTIPredicate(STIPredicateFunction & Fn,const ProcModelMapTy & ProcModelMap)301 static void processSTIPredicate(STIPredicateFunction &Fn,
302                                 const ProcModelMapTy &ProcModelMap) {
303   DenseMap<const Record *, unsigned> Opcode2Index;
304   using OpcodeMapPair = std::pair<const Record *, OpcodeInfo>;
305   std::vector<OpcodeMapPair> OpcodeMappings;
306   std::vector<std::pair<APInt, APInt>> OpcodeMasks;
307 
308   DenseMap<const Record *, unsigned> Predicate2Index;
309   unsigned NumUniquePredicates = 0;
310 
311   // Number unique predicates and opcodes used by InstructionEquivalenceClass
312   // definitions. Each unique opcode will be associated with an OpcodeInfo
313   // object.
314   for (const Record *Def : Fn.getDefinitions()) {
315     RecVec Classes = Def->getValueAsListOfDefs("Classes");
316     for (const Record *EC : Classes) {
317       const Record *Pred = EC->getValueAsDef("Predicate");
318       if (!Predicate2Index.contains(Pred))
319         Predicate2Index[Pred] = NumUniquePredicates++;
320 
321       RecVec Opcodes = EC->getValueAsListOfDefs("Opcodes");
322       for (const Record *Opcode : Opcodes) {
323         if (!Opcode2Index.contains(Opcode)) {
324           Opcode2Index[Opcode] = OpcodeMappings.size();
325           OpcodeMappings.emplace_back(Opcode, OpcodeInfo());
326         }
327       }
328     }
329   }
330 
331   // Initialize vector `OpcodeMasks` with default values.  We want to keep track
332   // of which processors "use" which opcodes.  We also want to be able to
333   // identify predicates that are used by different processors for a same
334   // opcode.
335   // This information is used later on by this algorithm to sort OpcodeMapping
336   // elements based on their processor and predicate sets.
337   OpcodeMasks.resize(OpcodeMappings.size());
338   APInt DefaultProcMask(ProcModelMap.size(), 0);
339   APInt DefaultPredMask(NumUniquePredicates, 0);
340   for (std::pair<APInt, APInt> &MaskPair : OpcodeMasks)
341     MaskPair = std::pair(DefaultProcMask, DefaultPredMask);
342 
343   // Construct a OpcodeInfo object for every unique opcode declared by an
344   // InstructionEquivalenceClass definition.
345   for (const Record *Def : Fn.getDefinitions()) {
346     RecVec Classes = Def->getValueAsListOfDefs("Classes");
347     const Record *SchedModel = Def->getValueAsDef("SchedModel");
348     unsigned ProcIndex = ProcModelMap.find(SchedModel)->second;
349     APInt ProcMask(ProcModelMap.size(), 0);
350     ProcMask.setBit(ProcIndex);
351 
352     for (const Record *EC : Classes) {
353       RecVec Opcodes = EC->getValueAsListOfDefs("Opcodes");
354 
355       std::vector<int64_t> OpIndices =
356           EC->getValueAsListOfInts("OperandIndices");
357       APInt OperandMask = constructOperandMask(OpIndices);
358 
359       const Record *Pred = EC->getValueAsDef("Predicate");
360       APInt PredMask(NumUniquePredicates, 0);
361       PredMask.setBit(Predicate2Index[Pred]);
362 
363       for (const Record *Opcode : Opcodes) {
364         unsigned OpcodeIdx = Opcode2Index[Opcode];
365         if (OpcodeMasks[OpcodeIdx].first[ProcIndex]) {
366           std::string Message =
367               "Opcode " + Opcode->getName().str() +
368               " used by multiple InstructionEquivalenceClass definitions.";
369           PrintFatalError(EC->getLoc(), Message);
370         }
371         OpcodeMasks[OpcodeIdx].first |= ProcMask;
372         OpcodeMasks[OpcodeIdx].second |= PredMask;
373         OpcodeInfo &OI = OpcodeMappings[OpcodeIdx].second;
374 
375         OI.addPredicateForProcModel(ProcMask, OperandMask, Pred);
376       }
377     }
378   }
379 
380   // Sort OpcodeMappings elements based on their CPU and predicate masks.
381   // As a last resort, order elements by opcode identifier.
382   llvm::sort(
383       OpcodeMappings, [&](const OpcodeMapPair &Lhs, const OpcodeMapPair &Rhs) {
384         unsigned LhsIdx = Opcode2Index[Lhs.first];
385         unsigned RhsIdx = Opcode2Index[Rhs.first];
386         const std::pair<APInt, APInt> &LhsMasks = OpcodeMasks[LhsIdx];
387         const std::pair<APInt, APInt> &RhsMasks = OpcodeMasks[RhsIdx];
388 
389         auto PopulationCountAndLeftBit =
390             [](const APInt &Other) -> std::pair<int, int> {
391           return std::pair<int, int>(Other.popcount(), -Other.countl_zero());
392         };
393         auto lhsmask_first = PopulationCountAndLeftBit(LhsMasks.first);
394         auto rhsmask_first = PopulationCountAndLeftBit(RhsMasks.first);
395         if (lhsmask_first != rhsmask_first)
396           return lhsmask_first < rhsmask_first;
397 
398         auto lhsmask_second = PopulationCountAndLeftBit(LhsMasks.second);
399         auto rhsmask_second = PopulationCountAndLeftBit(RhsMasks.second);
400         if (lhsmask_second != rhsmask_second)
401           return lhsmask_second < rhsmask_second;
402 
403         return LhsIdx < RhsIdx;
404       });
405 
406   // Now construct opcode groups. Groups are used by the SubtargetEmitter when
407   // expanding the body of a STIPredicate function. In particular, each opcode
408   // group is expanded into a sequence of labels in a switch statement.
409   // It identifies opcodes for which different processors define same predicates
410   // and same opcode masks.
411   for (OpcodeMapPair &Info : OpcodeMappings)
412     Fn.addOpcode(Info.first, std::move(Info.second));
413 }
414 
collectSTIPredicates()415 void CodeGenSchedModels::collectSTIPredicates() {
416   // Map STIPredicateDecl records to elements of vector
417   // CodeGenSchedModels::STIPredicates.
418   DenseMap<const Record *, unsigned> Decl2Index;
419 
420   RecVec RV = Records.getAllDerivedDefinitions("STIPredicate");
421   for (const Record *R : RV) {
422     const Record *Decl = R->getValueAsDef("Declaration");
423 
424     const auto It = Decl2Index.find(Decl);
425     if (It == Decl2Index.end()) {
426       Decl2Index[Decl] = STIPredicates.size();
427       STIPredicateFunction Predicate(Decl);
428       Predicate.addDefinition(R);
429       STIPredicates.emplace_back(std::move(Predicate));
430       continue;
431     }
432 
433     STIPredicateFunction &PreviousDef = STIPredicates[It->second];
434     PreviousDef.addDefinition(R);
435   }
436 
437   for (STIPredicateFunction &Fn : STIPredicates)
438     processSTIPredicate(Fn, ProcModelMap);
439 }
440 
addPredicateForProcModel(const llvm::APInt & CpuMask,const llvm::APInt & OperandMask,const Record * Predicate)441 void OpcodeInfo::addPredicateForProcModel(const llvm::APInt &CpuMask,
442                                           const llvm::APInt &OperandMask,
443                                           const Record *Predicate) {
444   auto It = llvm::find_if(
445       Predicates, [&OperandMask, &Predicate](const PredicateInfo &P) {
446         return P.Predicate == Predicate && P.OperandMask == OperandMask;
447       });
448   if (It == Predicates.end()) {
449     Predicates.emplace_back(CpuMask, OperandMask, Predicate);
450     return;
451   }
452   It->ProcModelMask |= CpuMask;
453 }
454 
checkMCInstPredicates() const455 void CodeGenSchedModels::checkMCInstPredicates() const {
456   RecVec MCPredicates = Records.getAllDerivedDefinitions("TIIPredicate");
457   if (MCPredicates.empty())
458     return;
459 
460   // A target cannot have multiple TIIPredicate definitions with a same name.
461   llvm::StringMap<const Record *> TIIPredicates(MCPredicates.size());
462   for (const Record *TIIPred : MCPredicates) {
463     StringRef Name = TIIPred->getValueAsString("FunctionName");
464     StringMap<const Record *>::const_iterator It = TIIPredicates.find(Name);
465     if (It == TIIPredicates.end()) {
466       TIIPredicates[Name] = TIIPred;
467       continue;
468     }
469 
470     PrintError(TIIPred->getLoc(),
471                "TIIPredicate " + Name + " is multiply defined.");
472     PrintFatalNote(It->second->getLoc(),
473                    " Previous definition of " + Name + " was here.");
474   }
475 }
476 
collectRetireControlUnits()477 void CodeGenSchedModels::collectRetireControlUnits() {
478   RecVec Units = Records.getAllDerivedDefinitions("RetireControlUnit");
479 
480   for (Record *RCU : Units) {
481     CodeGenProcModel &PM = getProcModel(RCU->getValueAsDef("SchedModel"));
482     if (PM.RetireControlUnit) {
483       PrintError(RCU->getLoc(),
484                  "Expected a single RetireControlUnit definition");
485       PrintNote(PM.RetireControlUnit->getLoc(),
486                 "Previous definition of RetireControlUnit was here");
487     }
488     PM.RetireControlUnit = RCU;
489   }
490 }
491 
collectLoadStoreQueueInfo()492 void CodeGenSchedModels::collectLoadStoreQueueInfo() {
493   RecVec Queues = Records.getAllDerivedDefinitions("MemoryQueue");
494 
495   for (Record *Queue : Queues) {
496     CodeGenProcModel &PM = getProcModel(Queue->getValueAsDef("SchedModel"));
497     if (Queue->isSubClassOf("LoadQueue")) {
498       if (PM.LoadQueue) {
499         PrintError(Queue->getLoc(), "Expected a single LoadQueue definition");
500         PrintNote(PM.LoadQueue->getLoc(),
501                   "Previous definition of LoadQueue was here");
502       }
503 
504       PM.LoadQueue = Queue;
505     }
506 
507     if (Queue->isSubClassOf("StoreQueue")) {
508       if (PM.StoreQueue) {
509         PrintError(Queue->getLoc(), "Expected a single StoreQueue definition");
510         PrintNote(PM.StoreQueue->getLoc(),
511                   "Previous definition of StoreQueue was here");
512       }
513 
514       PM.StoreQueue = Queue;
515     }
516   }
517 }
518 
519 /// Collect optional processor information.
collectOptionalProcessorInfo()520 void CodeGenSchedModels::collectOptionalProcessorInfo() {
521   // Find register file definitions for each processor.
522   collectRegisterFiles();
523 
524   // Collect processor RetireControlUnit descriptors if available.
525   collectRetireControlUnits();
526 
527   // Collect information about load/store queues.
528   collectLoadStoreQueueInfo();
529 
530   checkCompleteness();
531 }
532 
533 /// Gather all processor models.
collectProcModels()534 void CodeGenSchedModels::collectProcModels() {
535   RecVec ProcRecords = Records.getAllDerivedDefinitions("Processor");
536   llvm::sort(ProcRecords, LessRecordFieldName());
537 
538   // Check for duplicated names.
539   auto I = std::adjacent_find(ProcRecords.begin(), ProcRecords.end(),
540                               [](const Record *Rec1, const Record *Rec2) {
541                                 return Rec1->getValueAsString("Name") ==
542                                        Rec2->getValueAsString("Name");
543                               });
544   if (I != ProcRecords.end())
545     PrintFatalError((*I)->getLoc(), "Duplicate processor name " +
546                                         (*I)->getValueAsString("Name"));
547 
548   // Reserve space because we can. Reallocation would be ok.
549   ProcModels.reserve(ProcRecords.size() + 1);
550 
551   // Use idx=0 for NoModel/NoItineraries.
552   Record *NoModelDef = Records.getDef("NoSchedModel");
553   Record *NoItinsDef = Records.getDef("NoItineraries");
554   ProcModels.emplace_back(0, "NoSchedModel", NoModelDef, NoItinsDef);
555   ProcModelMap[NoModelDef] = 0;
556 
557   // For each processor, find a unique machine model.
558   LLVM_DEBUG(dbgs() << "+++ PROCESSOR MODELs (addProcModel) +++\n");
559   for (Record *ProcRecord : ProcRecords)
560     addProcModel(ProcRecord);
561 }
562 
563 /// Get a unique processor model based on the defined MachineModel and
564 /// ProcessorItineraries.
addProcModel(Record * ProcDef)565 void CodeGenSchedModels::addProcModel(Record *ProcDef) {
566   Record *ModelKey = getModelOrItinDef(ProcDef);
567   if (!ProcModelMap.insert(std::pair(ModelKey, ProcModels.size())).second)
568     return;
569 
570   std::string Name = std::string(ModelKey->getName());
571   if (ModelKey->isSubClassOf("SchedMachineModel")) {
572     Record *ItinsDef = ModelKey->getValueAsDef("Itineraries");
573     ProcModels.emplace_back(ProcModels.size(), Name, ModelKey, ItinsDef);
574   } else {
575     // An itinerary is defined without a machine model. Infer a new model.
576     if (!ModelKey->getValueAsListOfDefs("IID").empty())
577       Name = Name + "Model";
578     ProcModels.emplace_back(ProcModels.size(), Name,
579                             ProcDef->getValueAsDef("SchedModel"), ModelKey);
580   }
581   LLVM_DEBUG(ProcModels.back().dump());
582 }
583 
584 // Recursively find all reachable SchedReadWrite records.
scanSchedRW(Record * RWDef,RecVec & RWDefs,SmallPtrSet<Record *,16> & RWSet)585 static void scanSchedRW(Record *RWDef, RecVec &RWDefs,
586                         SmallPtrSet<Record *, 16> &RWSet) {
587   if (!RWSet.insert(RWDef).second)
588     return;
589   RWDefs.push_back(RWDef);
590   // Reads don't currently have sequence records, but it can be added later.
591   if (RWDef->isSubClassOf("WriteSequence")) {
592     RecVec Seq = RWDef->getValueAsListOfDefs("Writes");
593     for (Record *WSRec : Seq)
594       scanSchedRW(WSRec, RWDefs, RWSet);
595   } else if (RWDef->isSubClassOf("SchedVariant")) {
596     // Visit each variant (guarded by a different predicate).
597     RecVec Vars = RWDef->getValueAsListOfDefs("Variants");
598     for (Record *Variant : Vars) {
599       // Visit each RW in the sequence selected by the current variant.
600       RecVec Selected = Variant->getValueAsListOfDefs("Selected");
601       for (Record *SelDef : Selected)
602         scanSchedRW(SelDef, RWDefs, RWSet);
603     }
604   }
605 }
606 
607 // Collect and sort all SchedReadWrites reachable via tablegen records.
608 // More may be inferred later when inferring new SchedClasses from variants.
collectSchedRW()609 void CodeGenSchedModels::collectSchedRW() {
610   // Reserve idx=0 for invalid writes/reads.
611   SchedWrites.resize(1);
612   SchedReads.resize(1);
613 
614   SmallPtrSet<Record *, 16> RWSet;
615 
616   // Find all SchedReadWrites referenced by instruction defs.
617   RecVec SWDefs, SRDefs;
618   for (const CodeGenInstruction *Inst : Target.getInstructionsByEnumValue()) {
619     Record *SchedDef = Inst->TheDef;
620     if (SchedDef->isValueUnset("SchedRW"))
621       continue;
622     RecVec RWs = SchedDef->getValueAsListOfDefs("SchedRW");
623     for (Record *RW : RWs) {
624       if (RW->isSubClassOf("SchedWrite"))
625         scanSchedRW(RW, SWDefs, RWSet);
626       else {
627         assert(RW->isSubClassOf("SchedRead") && "Unknown SchedReadWrite");
628         scanSchedRW(RW, SRDefs, RWSet);
629       }
630     }
631   }
632   // Find all ReadWrites referenced by InstRW.
633   RecVec InstRWDefs = Records.getAllDerivedDefinitions("InstRW");
634   for (Record *InstRWDef : InstRWDefs) {
635     // For all OperandReadWrites.
636     RecVec RWDefs = InstRWDef->getValueAsListOfDefs("OperandReadWrites");
637     for (Record *RWDef : RWDefs) {
638       if (RWDef->isSubClassOf("SchedWrite"))
639         scanSchedRW(RWDef, SWDefs, RWSet);
640       else {
641         assert(RWDef->isSubClassOf("SchedRead") && "Unknown SchedReadWrite");
642         scanSchedRW(RWDef, SRDefs, RWSet);
643       }
644     }
645   }
646   // Find all ReadWrites referenced by ItinRW.
647   RecVec ItinRWDefs = Records.getAllDerivedDefinitions("ItinRW");
648   for (Record *ItinRWDef : ItinRWDefs) {
649     // For all OperandReadWrites.
650     RecVec RWDefs = ItinRWDef->getValueAsListOfDefs("OperandReadWrites");
651     for (Record *RWDef : RWDefs) {
652       if (RWDef->isSubClassOf("SchedWrite"))
653         scanSchedRW(RWDef, SWDefs, RWSet);
654       else {
655         assert(RWDef->isSubClassOf("SchedRead") && "Unknown SchedReadWrite");
656         scanSchedRW(RWDef, SRDefs, RWSet);
657       }
658     }
659   }
660   // Find all ReadWrites referenced by SchedAlias. AliasDefs needs to be sorted
661   // for the loop below that initializes Alias vectors.
662   RecVec AliasDefs = Records.getAllDerivedDefinitions("SchedAlias");
663   llvm::sort(AliasDefs, LessRecord());
664   for (Record *ADef : AliasDefs) {
665     Record *MatchDef = ADef->getValueAsDef("MatchRW");
666     Record *AliasDef = ADef->getValueAsDef("AliasRW");
667     if (MatchDef->isSubClassOf("SchedWrite")) {
668       if (!AliasDef->isSubClassOf("SchedWrite"))
669         PrintFatalError(ADef->getLoc(), "SchedWrite Alias must be SchedWrite");
670       scanSchedRW(AliasDef, SWDefs, RWSet);
671     } else {
672       assert(MatchDef->isSubClassOf("SchedRead") && "Unknown SchedReadWrite");
673       if (!AliasDef->isSubClassOf("SchedRead"))
674         PrintFatalError(ADef->getLoc(), "SchedRead Alias must be SchedRead");
675       scanSchedRW(AliasDef, SRDefs, RWSet);
676     }
677   }
678   // Sort and add the SchedReadWrites directly referenced by instructions or
679   // itinerary resources. Index reads and writes in separate domains.
680   llvm::sort(SWDefs, LessRecord());
681   for (Record *SWDef : SWDefs) {
682     assert(!getSchedRWIdx(SWDef, /*IsRead=*/false) && "duplicate SchedWrite");
683     SchedWrites.emplace_back(SchedWrites.size(), SWDef);
684   }
685   llvm::sort(SRDefs, LessRecord());
686   for (Record *SRDef : SRDefs) {
687     assert(!getSchedRWIdx(SRDef, /*IsRead-*/ true) && "duplicate SchedWrite");
688     SchedReads.emplace_back(SchedReads.size(), SRDef);
689   }
690   // Initialize WriteSequence vectors.
691   for (CodeGenSchedRW &CGRW : SchedWrites) {
692     if (!CGRW.IsSequence)
693       continue;
694     findRWs(CGRW.TheDef->getValueAsListOfDefs("Writes"), CGRW.Sequence,
695             /*IsRead=*/false);
696   }
697   // Initialize Aliases vectors.
698   for (Record *ADef : AliasDefs) {
699     Record *AliasDef = ADef->getValueAsDef("AliasRW");
700     getSchedRW(AliasDef).IsAlias = true;
701     Record *MatchDef = ADef->getValueAsDef("MatchRW");
702     CodeGenSchedRW &RW = getSchedRW(MatchDef);
703     if (RW.IsAlias)
704       PrintFatalError(ADef->getLoc(), "Cannot Alias an Alias");
705     RW.Aliases.push_back(ADef);
706   }
707   LLVM_DEBUG(
708       dbgs() << "\n+++ SCHED READS and WRITES (collectSchedRW) +++\n";
709       for (unsigned WIdx = 0, WEnd = SchedWrites.size(); WIdx != WEnd; ++WIdx) {
710         dbgs() << WIdx << ": ";
711         SchedWrites[WIdx].dump();
712         dbgs() << '\n';
713       } for (unsigned RIdx = 0, REnd = SchedReads.size(); RIdx != REnd;
714              ++RIdx) {
715         dbgs() << RIdx << ": ";
716         SchedReads[RIdx].dump();
717         dbgs() << '\n';
718       } RecVec RWDefs = Records.getAllDerivedDefinitions("SchedReadWrite");
719       for (Record *RWDef
720            : RWDefs) {
721         if (!getSchedRWIdx(RWDef, RWDef->isSubClassOf("SchedRead"))) {
722           StringRef Name = RWDef->getName();
723           if (Name != "NoWrite" && Name != "ReadDefault")
724             dbgs() << "Unused SchedReadWrite " << Name << '\n';
725         }
726       });
727 }
728 
729 /// Compute a SchedWrite name from a sequence of writes.
genRWName(ArrayRef<unsigned> Seq,bool IsRead)730 std::string CodeGenSchedModels::genRWName(ArrayRef<unsigned> Seq, bool IsRead) {
731   std::string Name("(");
732   ListSeparator LS("_");
733   for (unsigned I : Seq) {
734     Name += LS;
735     Name += getSchedRW(I, IsRead).Name;
736   }
737   Name += ')';
738   return Name;
739 }
740 
getSchedRWIdx(const Record * Def,bool IsRead) const741 unsigned CodeGenSchedModels::getSchedRWIdx(const Record *Def,
742                                            bool IsRead) const {
743   const std::vector<CodeGenSchedRW> &RWVec = IsRead ? SchedReads : SchedWrites;
744   const auto I = find_if(
745       RWVec, [Def](const CodeGenSchedRW &RW) { return RW.TheDef == Def; });
746   return I == RWVec.end() ? 0 : std::distance(RWVec.begin(), I);
747 }
748 
splitSchedReadWrites(const RecVec & RWDefs,RecVec & WriteDefs,RecVec & ReadDefs)749 static void splitSchedReadWrites(const RecVec &RWDefs, RecVec &WriteDefs,
750                                  RecVec &ReadDefs) {
751   for (Record *RWDef : RWDefs) {
752     if (RWDef->isSubClassOf("SchedWrite"))
753       WriteDefs.push_back(RWDef);
754     else {
755       assert(RWDef->isSubClassOf("SchedRead") && "unknown SchedReadWrite");
756       ReadDefs.push_back(RWDef);
757     }
758   }
759 }
760 
761 // Split the SchedReadWrites defs and call findRWs for each list.
findRWs(const RecVec & RWDefs,IdxVec & Writes,IdxVec & Reads) const762 void CodeGenSchedModels::findRWs(const RecVec &RWDefs, IdxVec &Writes,
763                                  IdxVec &Reads) const {
764   RecVec WriteDefs;
765   RecVec ReadDefs;
766   splitSchedReadWrites(RWDefs, WriteDefs, ReadDefs);
767   findRWs(WriteDefs, Writes, false);
768   findRWs(ReadDefs, Reads, true);
769 }
770 
771 // Call getSchedRWIdx for all elements in a sequence of SchedRW defs.
findRWs(const RecVec & RWDefs,IdxVec & RWs,bool IsRead) const772 void CodeGenSchedModels::findRWs(const RecVec &RWDefs, IdxVec &RWs,
773                                  bool IsRead) const {
774   for (Record *RWDef : RWDefs) {
775     unsigned Idx = getSchedRWIdx(RWDef, IsRead);
776     assert(Idx && "failed to collect SchedReadWrite");
777     RWs.push_back(Idx);
778   }
779 }
780 
expandRWSequence(unsigned RWIdx,IdxVec & RWSeq,bool IsRead) const781 void CodeGenSchedModels::expandRWSequence(unsigned RWIdx, IdxVec &RWSeq,
782                                           bool IsRead) const {
783   const CodeGenSchedRW &SchedRW = getSchedRW(RWIdx, IsRead);
784   if (!SchedRW.IsSequence) {
785     RWSeq.push_back(RWIdx);
786     return;
787   }
788   int Repeat = SchedRW.TheDef ? SchedRW.TheDef->getValueAsInt("Repeat") : 1;
789   for (int i = 0; i < Repeat; ++i) {
790     for (unsigned I : SchedRW.Sequence) {
791       expandRWSequence(I, RWSeq, IsRead);
792     }
793   }
794 }
795 
796 // Expand a SchedWrite as a sequence following any aliases that coincide with
797 // the given processor model.
expandRWSeqForProc(unsigned RWIdx,IdxVec & RWSeq,bool IsRead,const CodeGenProcModel & ProcModel) const798 void CodeGenSchedModels::expandRWSeqForProc(
799     unsigned RWIdx, IdxVec &RWSeq, bool IsRead,
800     const CodeGenProcModel &ProcModel) const {
801 
802   const CodeGenSchedRW &SchedWrite = getSchedRW(RWIdx, IsRead);
803   Record *AliasDef = nullptr;
804   for (const Record *Rec : SchedWrite.Aliases) {
805     const CodeGenSchedRW &AliasRW = getSchedRW(Rec->getValueAsDef("AliasRW"));
806     if (Rec->getValueInit("SchedModel")->isComplete()) {
807       Record *ModelDef = Rec->getValueAsDef("SchedModel");
808       if (&getProcModel(ModelDef) != &ProcModel)
809         continue;
810     }
811     if (AliasDef)
812       PrintFatalError(AliasRW.TheDef->getLoc(),
813                       "Multiple aliases "
814                       "defined for processor " +
815                           ProcModel.ModelName +
816                           " Ensure only one SchedAlias exists per RW.");
817     AliasDef = AliasRW.TheDef;
818   }
819   if (AliasDef) {
820     expandRWSeqForProc(getSchedRWIdx(AliasDef, IsRead), RWSeq, IsRead,
821                        ProcModel);
822     return;
823   }
824   if (!SchedWrite.IsSequence) {
825     RWSeq.push_back(RWIdx);
826     return;
827   }
828   int Repeat =
829       SchedWrite.TheDef ? SchedWrite.TheDef->getValueAsInt("Repeat") : 1;
830   for (int I = 0, E = Repeat; I < E; ++I) {
831     for (unsigned Idx : SchedWrite.Sequence) {
832       expandRWSeqForProc(Idx, RWSeq, IsRead, ProcModel);
833     }
834   }
835 }
836 
837 // Find the existing SchedWrite that models this sequence of writes.
findRWForSequence(ArrayRef<unsigned> Seq,bool IsRead)838 unsigned CodeGenSchedModels::findRWForSequence(ArrayRef<unsigned> Seq,
839                                                bool IsRead) {
840   std::vector<CodeGenSchedRW> &RWVec = IsRead ? SchedReads : SchedWrites;
841 
842   auto I = find_if(RWVec, [Seq](CodeGenSchedRW &RW) {
843     return ArrayRef(RW.Sequence) == Seq;
844   });
845   // Index zero reserved for invalid RW.
846   return I == RWVec.end() ? 0 : std::distance(RWVec.begin(), I);
847 }
848 
849 /// Add this ReadWrite if it doesn't already exist.
findOrInsertRW(ArrayRef<unsigned> Seq,bool IsRead)850 unsigned CodeGenSchedModels::findOrInsertRW(ArrayRef<unsigned> Seq,
851                                             bool IsRead) {
852   assert(!Seq.empty() && "cannot insert empty sequence");
853   if (Seq.size() == 1)
854     return Seq.back();
855 
856   unsigned Idx = findRWForSequence(Seq, IsRead);
857   if (Idx)
858     return Idx;
859 
860   std::vector<CodeGenSchedRW> &RWVec = IsRead ? SchedReads : SchedWrites;
861   unsigned RWIdx = RWVec.size();
862   CodeGenSchedRW SchedRW(RWIdx, IsRead, Seq, genRWName(Seq, IsRead));
863   RWVec.push_back(SchedRW);
864   return RWIdx;
865 }
866 
867 /// Visit all the instruction definitions for this target to gather and
868 /// enumerate the itinerary classes. These are the explicitly specified
869 /// SchedClasses. More SchedClasses may be inferred.
collectSchedClasses()870 void CodeGenSchedModels::collectSchedClasses() {
871 
872   // NoItinerary is always the first class at Idx=0
873   assert(SchedClasses.empty() && "Expected empty sched class");
874   SchedClasses.emplace_back(0, "NoInstrModel", Records.getDef("NoItinerary"));
875   SchedClasses.back().ProcIndices.push_back(0);
876 
877   // Create a SchedClass for each unique combination of itinerary class and
878   // SchedRW list.
879   for (const CodeGenInstruction *Inst : Target.getInstructionsByEnumValue()) {
880     Record *ItinDef = Inst->TheDef->getValueAsDef("Itinerary");
881     IdxVec Writes, Reads;
882     if (!Inst->TheDef->isValueUnset("SchedRW"))
883       findRWs(Inst->TheDef->getValueAsListOfDefs("SchedRW"), Writes, Reads);
884 
885     // ProcIdx == 0 indicates the class applies to all processors.
886     unsigned SCIdx = addSchedClass(ItinDef, Writes, Reads, /*ProcIndices*/ {0});
887     InstrClassMap[Inst->TheDef] = SCIdx;
888   }
889   // Create classes for InstRW defs.
890   RecVec InstRWDefs = Records.getAllDerivedDefinitions("InstRW");
891   llvm::sort(InstRWDefs, LessRecord());
892   LLVM_DEBUG(dbgs() << "\n+++ SCHED CLASSES (createInstRWClass) +++\n");
893   for (Record *RWDef : InstRWDefs)
894     createInstRWClass(RWDef);
895 
896   NumInstrSchedClasses = SchedClasses.size();
897 
898   bool EnableDump = false;
899   LLVM_DEBUG(EnableDump = true);
900   if (!EnableDump)
901     return;
902 
903   LLVM_DEBUG(
904       dbgs()
905       << "\n+++ ITINERARIES and/or MACHINE MODELS (collectSchedClasses) +++\n");
906   for (const CodeGenInstruction *Inst : Target.getInstructionsByEnumValue()) {
907     StringRef InstName = Inst->TheDef->getName();
908     unsigned SCIdx = getSchedClassIdx(*Inst);
909     if (!SCIdx) {
910       LLVM_DEBUG({
911         if (!Inst->hasNoSchedulingInfo)
912           dbgs() << "No machine model for " << Inst->TheDef->getName() << '\n';
913       });
914       continue;
915     }
916     CodeGenSchedClass &SC = getSchedClass(SCIdx);
917     if (SC.ProcIndices[0] != 0)
918       PrintFatalError(Inst->TheDef->getLoc(),
919                       "Instruction's sched class "
920                       "must not be subtarget specific.");
921 
922     IdxVec ProcIndices;
923     if (SC.ItinClassDef->getName() != "NoItinerary") {
924       ProcIndices.push_back(0);
925       dbgs() << "Itinerary for " << InstName << ": "
926              << SC.ItinClassDef->getName() << '\n';
927     }
928     if (!SC.Writes.empty()) {
929       ProcIndices.push_back(0);
930       LLVM_DEBUG({
931         dbgs() << "SchedRW machine model for " << InstName;
932         for (unsigned int Write : SC.Writes)
933           dbgs() << " " << SchedWrites[Write].Name;
934         for (unsigned int Read : SC.Reads)
935           dbgs() << " " << SchedReads[Read].Name;
936         dbgs() << '\n';
937       });
938     }
939     const RecVec &RWDefs = SchedClasses[SCIdx].InstRWs;
940     for (Record *RWDef : RWDefs) {
941       const CodeGenProcModel &ProcModel =
942           getProcModel(RWDef->getValueAsDef("SchedModel"));
943       ProcIndices.push_back(ProcModel.Index);
944       LLVM_DEBUG(dbgs() << "InstRW on " << ProcModel.ModelName << " for "
945                         << InstName);
946       IdxVec Writes;
947       IdxVec Reads;
948       findRWs(RWDef->getValueAsListOfDefs("OperandReadWrites"), Writes, Reads);
949       LLVM_DEBUG({
950         for (unsigned WIdx : Writes)
951           dbgs() << " " << SchedWrites[WIdx].Name;
952         for (unsigned RIdx : Reads)
953           dbgs() << " " << SchedReads[RIdx].Name;
954         dbgs() << '\n';
955       });
956     }
957     // If ProcIndices contains zero, the class applies to all processors.
958     LLVM_DEBUG({
959       if (!llvm::is_contained(ProcIndices, 0)) {
960         for (const CodeGenProcModel &PM : ProcModels) {
961           if (!llvm::is_contained(ProcIndices, PM.Index))
962             dbgs() << "No machine model for " << Inst->TheDef->getName()
963                    << " on processor " << PM.ModelName << '\n';
964         }
965       }
966     });
967   }
968 }
969 
970 // Get the SchedClass index for an instruction.
971 unsigned
getSchedClassIdx(const CodeGenInstruction & Inst) const972 CodeGenSchedModels::getSchedClassIdx(const CodeGenInstruction &Inst) const {
973   return InstrClassMap.lookup(Inst.TheDef);
974 }
975 
976 std::string
createSchedClassName(Record * ItinClassDef,ArrayRef<unsigned> OperWrites,ArrayRef<unsigned> OperReads)977 CodeGenSchedModels::createSchedClassName(Record *ItinClassDef,
978                                          ArrayRef<unsigned> OperWrites,
979                                          ArrayRef<unsigned> OperReads) {
980 
981   std::string Name;
982   if (ItinClassDef && ItinClassDef->getName() != "NoItinerary")
983     Name = std::string(ItinClassDef->getName());
984   for (unsigned Idx : OperWrites) {
985     if (!Name.empty())
986       Name += '_';
987     Name += SchedWrites[Idx].Name;
988   }
989   for (unsigned Idx : OperReads) {
990     Name += '_';
991     Name += SchedReads[Idx].Name;
992   }
993   return Name;
994 }
995 
createSchedClassName(const RecVec & InstDefs)996 std::string CodeGenSchedModels::createSchedClassName(const RecVec &InstDefs) {
997 
998   std::string Name;
999   ListSeparator LS("_");
1000   for (const Record *InstDef : InstDefs) {
1001     Name += LS;
1002     Name += InstDef->getName();
1003   }
1004   return Name;
1005 }
1006 
1007 /// Add an inferred sched class from an itinerary class and per-operand list of
1008 /// SchedWrites and SchedReads. ProcIndices contains the set of IDs of
1009 /// processors that may utilize this class.
addSchedClass(Record * ItinClassDef,ArrayRef<unsigned> OperWrites,ArrayRef<unsigned> OperReads,ArrayRef<unsigned> ProcIndices)1010 unsigned CodeGenSchedModels::addSchedClass(Record *ItinClassDef,
1011                                            ArrayRef<unsigned> OperWrites,
1012                                            ArrayRef<unsigned> OperReads,
1013                                            ArrayRef<unsigned> ProcIndices) {
1014   assert(!ProcIndices.empty() && "expect at least one ProcIdx");
1015 
1016   auto IsKeyEqual = [=](const CodeGenSchedClass &SC) {
1017     return SC.isKeyEqual(ItinClassDef, OperWrites, OperReads);
1018   };
1019 
1020   auto I = find_if(make_range(schedClassBegin(), schedClassEnd()), IsKeyEqual);
1021   unsigned Idx = I == schedClassEnd() ? 0 : std::distance(schedClassBegin(), I);
1022   if (Idx || SchedClasses[0].isKeyEqual(ItinClassDef, OperWrites, OperReads)) {
1023     IdxVec PI;
1024     std::set_union(SchedClasses[Idx].ProcIndices.begin(),
1025                    SchedClasses[Idx].ProcIndices.end(), ProcIndices.begin(),
1026                    ProcIndices.end(), std::back_inserter(PI));
1027     SchedClasses[Idx].ProcIndices = std::move(PI);
1028     return Idx;
1029   }
1030   Idx = SchedClasses.size();
1031   SchedClasses.emplace_back(
1032       Idx, createSchedClassName(ItinClassDef, OperWrites, OperReads),
1033       ItinClassDef);
1034   CodeGenSchedClass &SC = SchedClasses.back();
1035   SC.Writes = OperWrites;
1036   SC.Reads = OperReads;
1037   SC.ProcIndices = ProcIndices;
1038 
1039   return Idx;
1040 }
1041 
1042 // Create classes for each set of opcodes that are in the same InstReadWrite
1043 // definition across all processors.
createInstRWClass(Record * InstRWDef)1044 void CodeGenSchedModels::createInstRWClass(Record *InstRWDef) {
1045   // ClassInstrs will hold an entry for each subset of Instrs in InstRWDef that
1046   // intersects with an existing class via a previous InstRWDef. Instrs that do
1047   // not intersect with an existing class refer back to their former class as
1048   // determined from ItinDef or SchedRW.
1049   SmallMapVector<unsigned, SmallVector<Record *, 8>, 4> ClassInstrs;
1050   // Sort Instrs into sets.
1051   const RecVec *InstDefs = Sets.expand(InstRWDef);
1052   if (InstDefs->empty())
1053     PrintFatalError(InstRWDef->getLoc(), "No matching instruction opcodes");
1054 
1055   for (Record *InstDef : *InstDefs) {
1056     InstClassMapTy::const_iterator Pos = InstrClassMap.find(InstDef);
1057     if (Pos == InstrClassMap.end())
1058       PrintFatalError(InstDef->getLoc(), "No sched class for instruction.");
1059     unsigned SCIdx = Pos->second;
1060     ClassInstrs[SCIdx].push_back(InstDef);
1061   }
1062   // For each set of Instrs, create a new class if necessary, and map or remap
1063   // the Instrs to it.
1064   for (auto &Entry : ClassInstrs) {
1065     unsigned OldSCIdx = Entry.first;
1066     ArrayRef<Record *> InstDefs = Entry.second;
1067     // If the all instrs in the current class are accounted for, then leave
1068     // them mapped to their old class.
1069     if (OldSCIdx) {
1070       const RecVec &RWDefs = SchedClasses[OldSCIdx].InstRWs;
1071       if (!RWDefs.empty()) {
1072         const RecVec *OrigInstDefs = Sets.expand(RWDefs[0]);
1073         unsigned OrigNumInstrs = count_if(*OrigInstDefs, [&](Record *OIDef) {
1074           return InstrClassMap[OIDef] == OldSCIdx;
1075         });
1076         if (OrigNumInstrs == InstDefs.size()) {
1077           assert(SchedClasses[OldSCIdx].ProcIndices[0] == 0 &&
1078                  "expected a generic SchedClass");
1079           Record *RWModelDef = InstRWDef->getValueAsDef("SchedModel");
1080           // Make sure we didn't already have a InstRW containing this
1081           // instruction on this model.
1082           for (Record *RWD : RWDefs) {
1083             if (RWD->getValueAsDef("SchedModel") == RWModelDef &&
1084                 RWModelDef->getValueAsBit("FullInstRWOverlapCheck")) {
1085               assert(!InstDefs.empty()); // Checked at function start.
1086               PrintError(
1087                   InstRWDef->getLoc(),
1088                   "Overlapping InstRW definition for \"" +
1089                       InstDefs.front()->getName() +
1090                       "\" also matches previous \"" +
1091                       RWD->getValue("Instrs")->getValue()->getAsString() +
1092                       "\".");
1093               PrintFatalNote(RWD->getLoc(), "Previous match was here.");
1094             }
1095           }
1096           LLVM_DEBUG(dbgs() << "InstRW: Reuse SC " << OldSCIdx << ":"
1097                             << SchedClasses[OldSCIdx].Name << " on "
1098                             << RWModelDef->getName() << "\n");
1099           SchedClasses[OldSCIdx].InstRWs.push_back(InstRWDef);
1100           continue;
1101         }
1102       }
1103     }
1104     unsigned SCIdx = SchedClasses.size();
1105     SchedClasses.emplace_back(SCIdx, createSchedClassName(InstDefs), nullptr);
1106     CodeGenSchedClass &SC = SchedClasses.back();
1107     LLVM_DEBUG(dbgs() << "InstRW: New SC " << SCIdx << ":" << SC.Name << " on "
1108                       << InstRWDef->getValueAsDef("SchedModel")->getName()
1109                       << "\n");
1110 
1111     // Preserve ItinDef and Writes/Reads for processors without an InstRW entry.
1112     SC.ItinClassDef = SchedClasses[OldSCIdx].ItinClassDef;
1113     SC.Writes = SchedClasses[OldSCIdx].Writes;
1114     SC.Reads = SchedClasses[OldSCIdx].Reads;
1115     SC.ProcIndices.push_back(0);
1116     // If we had an old class, copy it's InstRWs to this new class.
1117     if (OldSCIdx) {
1118       Record *RWModelDef = InstRWDef->getValueAsDef("SchedModel");
1119       for (Record *OldRWDef : SchedClasses[OldSCIdx].InstRWs) {
1120         if (OldRWDef->getValueAsDef("SchedModel") == RWModelDef) {
1121           assert(!InstDefs.empty()); // Checked at function start.
1122           PrintError(
1123               InstRWDef->getLoc(),
1124               "Overlapping InstRW definition for \"" +
1125                   InstDefs.front()->getName() + "\" also matches previous \"" +
1126                   OldRWDef->getValue("Instrs")->getValue()->getAsString() +
1127                   "\".");
1128           PrintFatalNote(OldRWDef->getLoc(), "Previous match was here.");
1129         }
1130         assert(OldRWDef != InstRWDef && "SchedClass has duplicate InstRW def");
1131         SC.InstRWs.push_back(OldRWDef);
1132       }
1133     }
1134     // Map each Instr to this new class.
1135     for (Record *InstDef : InstDefs)
1136       InstrClassMap[InstDef] = SCIdx;
1137     SC.InstRWs.push_back(InstRWDef);
1138   }
1139 }
1140 
1141 // True if collectProcItins found anything.
hasItineraries() const1142 bool CodeGenSchedModels::hasItineraries() const {
1143   for (const CodeGenProcModel &PM :
1144        make_range(procModelBegin(), procModelEnd()))
1145     if (PM.hasItineraries())
1146       return true;
1147   return false;
1148 }
1149 
1150 // Gather the processor itineraries.
collectProcItins()1151 void CodeGenSchedModels::collectProcItins() {
1152   LLVM_DEBUG(dbgs() << "\n+++ PROBLEM ITINERARIES (collectProcItins) +++\n");
1153   for (CodeGenProcModel &ProcModel : ProcModels) {
1154     if (!ProcModel.hasItineraries())
1155       continue;
1156 
1157     RecVec ItinRecords = ProcModel.ItinsDef->getValueAsListOfDefs("IID");
1158     assert(!ItinRecords.empty() && "ProcModel.hasItineraries is incorrect");
1159 
1160     // Populate ItinDefList with Itinerary records.
1161     ProcModel.ItinDefList.resize(NumInstrSchedClasses);
1162 
1163     // Insert each itinerary data record in the correct position within
1164     // the processor model's ItinDefList.
1165     for (Record *ItinData : ItinRecords) {
1166       const Record *ItinDef = ItinData->getValueAsDef("TheClass");
1167       bool FoundClass = false;
1168 
1169       for (const CodeGenSchedClass &SC :
1170            make_range(schedClassBegin(), schedClassEnd())) {
1171         // Multiple SchedClasses may share an itinerary. Update all of them.
1172         if (SC.ItinClassDef == ItinDef) {
1173           ProcModel.ItinDefList[SC.Index] = ItinData;
1174           FoundClass = true;
1175         }
1176       }
1177       if (!FoundClass) {
1178         LLVM_DEBUG(dbgs() << ProcModel.ItinsDef->getName()
1179                           << " missing class for itinerary "
1180                           << ItinDef->getName() << '\n');
1181       }
1182     }
1183     // Check for missing itinerary entries.
1184     assert(!ProcModel.ItinDefList[0] && "NoItinerary class can't have rec");
1185     LLVM_DEBUG(
1186         for (unsigned i = 1, N = ProcModel.ItinDefList.size(); i < N; ++i) {
1187           if (!ProcModel.ItinDefList[i])
1188             dbgs() << ProcModel.ItinsDef->getName()
1189                    << " missing itinerary for class " << SchedClasses[i].Name
1190                    << '\n';
1191         });
1192   }
1193 }
1194 
1195 // Gather the read/write types for each itinerary class.
collectProcItinRW()1196 void CodeGenSchedModels::collectProcItinRW() {
1197   RecVec ItinRWDefs = Records.getAllDerivedDefinitions("ItinRW");
1198   llvm::sort(ItinRWDefs, LessRecord());
1199   for (Record *RWDef : ItinRWDefs) {
1200     if (!RWDef->getValueInit("SchedModel")->isComplete())
1201       PrintFatalError(RWDef->getLoc(), "SchedModel is undefined");
1202     Record *ModelDef = RWDef->getValueAsDef("SchedModel");
1203     ProcModelMapTy::const_iterator I = ProcModelMap.find(ModelDef);
1204     if (I == ProcModelMap.end()) {
1205       PrintFatalError(RWDef->getLoc(),
1206                       "Undefined SchedMachineModel " + ModelDef->getName());
1207     }
1208     ProcModels[I->second].ItinRWDefs.push_back(RWDef);
1209   }
1210 }
1211 
1212 // Gather the unsupported features for processor models.
collectProcUnsupportedFeatures()1213 void CodeGenSchedModels::collectProcUnsupportedFeatures() {
1214   for (CodeGenProcModel &ProcModel : ProcModels)
1215     append_range(
1216         ProcModel.UnsupportedFeaturesDefs,
1217         ProcModel.ModelDef->getValueAsListOfDefs("UnsupportedFeatures"));
1218 }
1219 
1220 /// Infer new classes from existing classes. In the process, this may create new
1221 /// SchedWrites from sequences of existing SchedWrites.
inferSchedClasses()1222 void CodeGenSchedModels::inferSchedClasses() {
1223   LLVM_DEBUG(
1224       dbgs() << "\n+++ INFERRING SCHED CLASSES (inferSchedClasses) +++\n");
1225   LLVM_DEBUG(dbgs() << NumInstrSchedClasses << " instr sched classes.\n");
1226 
1227   // Visit all existing classes and newly created classes.
1228   for (unsigned Idx = 0; Idx != SchedClasses.size(); ++Idx) {
1229     assert(SchedClasses[Idx].Index == Idx && "bad SCIdx");
1230 
1231     if (SchedClasses[Idx].ItinClassDef)
1232       inferFromItinClass(SchedClasses[Idx].ItinClassDef, Idx);
1233     if (!SchedClasses[Idx].InstRWs.empty())
1234       inferFromInstRWs(Idx);
1235     if (!SchedClasses[Idx].Writes.empty()) {
1236       inferFromRW(SchedClasses[Idx].Writes, SchedClasses[Idx].Reads, Idx,
1237                   SchedClasses[Idx].ProcIndices);
1238     }
1239     assert(SchedClasses.size() < (NumInstrSchedClasses * 6) &&
1240            "too many SchedVariants");
1241   }
1242 }
1243 
1244 /// Infer classes from per-processor itinerary resources.
inferFromItinClass(Record * ItinClassDef,unsigned FromClassIdx)1245 void CodeGenSchedModels::inferFromItinClass(Record *ItinClassDef,
1246                                             unsigned FromClassIdx) {
1247   for (unsigned PIdx = 0, PEnd = ProcModels.size(); PIdx != PEnd; ++PIdx) {
1248     const CodeGenProcModel &PM = ProcModels[PIdx];
1249     // For all ItinRW entries.
1250     bool HasMatch = false;
1251     for (const Record *Rec : PM.ItinRWDefs) {
1252       RecVec Matched = Rec->getValueAsListOfDefs("MatchedItinClasses");
1253       if (!llvm::is_contained(Matched, ItinClassDef))
1254         continue;
1255       if (HasMatch)
1256         PrintFatalError(Rec->getLoc(),
1257                         "Duplicate itinerary class " + ItinClassDef->getName() +
1258                             " in ItinResources for " + PM.ModelName);
1259       HasMatch = true;
1260       IdxVec Writes, Reads;
1261       findRWs(Rec->getValueAsListOfDefs("OperandReadWrites"), Writes, Reads);
1262       inferFromRW(Writes, Reads, FromClassIdx, PIdx);
1263     }
1264   }
1265 }
1266 
1267 /// Infer classes from per-processor InstReadWrite definitions.
inferFromInstRWs(unsigned SCIdx)1268 void CodeGenSchedModels::inferFromInstRWs(unsigned SCIdx) {
1269   for (unsigned I = 0, E = SchedClasses[SCIdx].InstRWs.size(); I != E; ++I) {
1270     assert(SchedClasses[SCIdx].InstRWs.size() == E && "InstrRWs was mutated!");
1271     Record *Rec = SchedClasses[SCIdx].InstRWs[I];
1272     const RecVec *InstDefs = Sets.expand(Rec);
1273     RecIter II = InstDefs->begin(), IE = InstDefs->end();
1274     for (; II != IE; ++II) {
1275       if (InstrClassMap[*II] == SCIdx)
1276         break;
1277     }
1278     // If this class no longer has any instructions mapped to it, it has become
1279     // irrelevant.
1280     if (II == IE)
1281       continue;
1282     IdxVec Writes, Reads;
1283     findRWs(Rec->getValueAsListOfDefs("OperandReadWrites"), Writes, Reads);
1284     unsigned PIdx = getProcModel(Rec->getValueAsDef("SchedModel")).Index;
1285     inferFromRW(Writes, Reads, SCIdx, PIdx); // May mutate SchedClasses.
1286     SchedClasses[SCIdx].InstRWProcIndices.insert(PIdx);
1287   }
1288 }
1289 
1290 namespace {
1291 
1292 // Helper for substituteVariantOperand.
1293 struct TransVariant {
1294   Record *VarOrSeqDef;  // Variant or sequence.
1295   unsigned RWIdx;       // Index of this variant or sequence's matched type.
1296   unsigned ProcIdx;     // Processor model index or zero for any.
1297   unsigned TransVecIdx; // Index into PredTransitions::TransVec.
1298 
TransVariant__anona30522b50b11::TransVariant1299   TransVariant(Record *def, unsigned rwi, unsigned pi, unsigned ti)
1300       : VarOrSeqDef(def), RWIdx(rwi), ProcIdx(pi), TransVecIdx(ti) {}
1301 };
1302 
1303 // Associate a predicate with the SchedReadWrite that it guards.
1304 // RWIdx is the index of the read/write variant.
1305 struct PredCheck {
1306   bool IsRead;
1307   unsigned RWIdx;
1308   Record *Predicate;
1309 
PredCheck__anona30522b50b11::PredCheck1310   PredCheck(bool r, unsigned w, Record *p)
1311       : IsRead(r), RWIdx(w), Predicate(p) {}
1312 };
1313 
1314 // A Predicate transition is a list of RW sequences guarded by a PredTerm.
1315 struct PredTransition {
1316   // A predicate term is a conjunction of PredChecks.
1317   SmallVector<PredCheck, 4> PredTerm;
1318   SmallVector<SmallVector<unsigned, 4>, 16> WriteSequences;
1319   SmallVector<SmallVector<unsigned, 4>, 16> ReadSequences;
1320   unsigned ProcIndex = 0;
1321 
1322   PredTransition() = default;
PredTransition__anona30522b50b11::PredTransition1323   PredTransition(ArrayRef<PredCheck> PT, unsigned ProcId) {
1324     PredTerm.assign(PT.begin(), PT.end());
1325     ProcIndex = ProcId;
1326   }
1327 };
1328 
1329 // Encapsulate a set of partially constructed transitions.
1330 // The results are built by repeated calls to substituteVariants.
1331 class PredTransitions {
1332   CodeGenSchedModels &SchedModels;
1333 
1334 public:
1335   std::vector<PredTransition> TransVec;
1336 
PredTransitions(CodeGenSchedModels & sm)1337   PredTransitions(CodeGenSchedModels &sm) : SchedModels(sm) {}
1338 
1339   bool substituteVariantOperand(const SmallVectorImpl<unsigned> &RWSeq,
1340                                 bool IsRead, unsigned StartIdx);
1341 
1342   bool substituteVariants(const PredTransition &Trans);
1343 
1344 #ifndef NDEBUG
1345   void dump() const;
1346 #endif
1347 
1348 private:
1349   bool mutuallyExclusive(Record *PredDef, ArrayRef<Record *> Preds,
1350                          ArrayRef<PredCheck> Term);
1351   void getIntersectingVariants(const CodeGenSchedRW &SchedRW, unsigned TransIdx,
1352                                std::vector<TransVariant> &IntersectingVariants);
1353   void pushVariant(const TransVariant &VInfo, bool IsRead);
1354 };
1355 
1356 } // end anonymous namespace
1357 
1358 // Return true if this predicate is mutually exclusive with a PredTerm. This
1359 // degenerates into checking if the predicate is mutually exclusive with any
1360 // predicate in the Term's conjunction.
1361 //
1362 // All predicates associated with a given SchedRW are considered mutually
1363 // exclusive. This should work even if the conditions expressed by the
1364 // predicates are not exclusive because the predicates for a given SchedWrite
1365 // are always checked in the order they are defined in the .td file. Later
1366 // conditions implicitly negate any prior condition.
mutuallyExclusive(Record * PredDef,ArrayRef<Record * > Preds,ArrayRef<PredCheck> Term)1367 bool PredTransitions::mutuallyExclusive(Record *PredDef,
1368                                         ArrayRef<Record *> Preds,
1369                                         ArrayRef<PredCheck> Term) {
1370   for (const PredCheck &PC : Term) {
1371     if (PC.Predicate == PredDef)
1372       return false;
1373 
1374     const CodeGenSchedRW &SchedRW = SchedModels.getSchedRW(PC.RWIdx, PC.IsRead);
1375     assert(SchedRW.HasVariants && "PredCheck must refer to a SchedVariant");
1376     RecVec Variants = SchedRW.TheDef->getValueAsListOfDefs("Variants");
1377     if (any_of(Variants, [PredDef](const Record *R) {
1378           return R->getValueAsDef("Predicate") == PredDef;
1379         })) {
1380       // To check if PredDef is mutually exclusive with PC we also need to
1381       // check that PC.Predicate is exclusive with all predicates from variant
1382       // we're expanding. Consider following RW sequence with two variants
1383       // (1 & 2), where A, B and C are predicates from corresponding SchedVars:
1384       //
1385       // 1:A/B - 2:C/B
1386       //
1387       // Here C is not mutually exclusive with variant (1), because A doesn't
1388       // exist in variant (2). This means we have possible transitions from A
1389       // to C and from A to B, and fully expanded sequence would look like:
1390       //
1391       // if (A & C) return ...;
1392       // if (A & B) return ...;
1393       // if (B) return ...;
1394       //
1395       // Now let's consider another sequence:
1396       //
1397       // 1:A/B - 2:A/B
1398       //
1399       // Here A in variant (2) is mutually exclusive with variant (1), because
1400       // A also exists in (2). This means A->B transition is impossible and
1401       // expanded sequence would look like:
1402       //
1403       // if (A) return ...;
1404       // if (B) return ...;
1405       if (!llvm::is_contained(Preds, PC.Predicate))
1406         continue;
1407       return true;
1408     }
1409   }
1410   return false;
1411 }
1412 
getAllPredicates(ArrayRef<TransVariant> Variants,unsigned ProcId)1413 static std::vector<Record *> getAllPredicates(ArrayRef<TransVariant> Variants,
1414                                               unsigned ProcId) {
1415   std::vector<Record *> Preds;
1416   for (auto &Variant : Variants) {
1417     if (!Variant.VarOrSeqDef->isSubClassOf("SchedVar"))
1418       continue;
1419     Preds.push_back(Variant.VarOrSeqDef->getValueAsDef("Predicate"));
1420   }
1421   return Preds;
1422 }
1423 
1424 // Populate IntersectingVariants with any variants or aliased sequences of the
1425 // given SchedRW whose processor indices and predicates are not mutually
1426 // exclusive with the given transition.
getIntersectingVariants(const CodeGenSchedRW & SchedRW,unsigned TransIdx,std::vector<TransVariant> & IntersectingVariants)1427 void PredTransitions::getIntersectingVariants(
1428     const CodeGenSchedRW &SchedRW, unsigned TransIdx,
1429     std::vector<TransVariant> &IntersectingVariants) {
1430 
1431   bool GenericRW = false;
1432 
1433   std::vector<TransVariant> Variants;
1434   if (SchedRW.HasVariants) {
1435     unsigned VarProcIdx = 0;
1436     if (SchedRW.TheDef->getValueInit("SchedModel")->isComplete()) {
1437       Record *ModelDef = SchedRW.TheDef->getValueAsDef("SchedModel");
1438       VarProcIdx = SchedModels.getProcModel(ModelDef).Index;
1439     }
1440     if (VarProcIdx == 0 || VarProcIdx == TransVec[TransIdx].ProcIndex) {
1441       // Push each variant. Assign TransVecIdx later.
1442       const RecVec VarDefs = SchedRW.TheDef->getValueAsListOfDefs("Variants");
1443       for (Record *VarDef : VarDefs)
1444         Variants.emplace_back(VarDef, SchedRW.Index, VarProcIdx, 0);
1445       if (VarProcIdx == 0)
1446         GenericRW = true;
1447     }
1448   }
1449   for (RecIter AI = SchedRW.Aliases.begin(), AE = SchedRW.Aliases.end();
1450        AI != AE; ++AI) {
1451     // If either the SchedAlias itself or the SchedReadWrite that it aliases
1452     // to is defined within a processor model, constrain all variants to
1453     // that processor.
1454     unsigned AliasProcIdx = 0;
1455     if ((*AI)->getValueInit("SchedModel")->isComplete()) {
1456       Record *ModelDef = (*AI)->getValueAsDef("SchedModel");
1457       AliasProcIdx = SchedModels.getProcModel(ModelDef).Index;
1458     }
1459     if (AliasProcIdx && AliasProcIdx != TransVec[TransIdx].ProcIndex)
1460       continue;
1461     if (!Variants.empty()) {
1462       const CodeGenProcModel &PM =
1463           *(SchedModels.procModelBegin() + AliasProcIdx);
1464       PrintFatalError((*AI)->getLoc(),
1465                       "Multiple variants defined for processor " +
1466                           PM.ModelName +
1467                           " Ensure only one SchedAlias exists per RW.");
1468     }
1469 
1470     const CodeGenSchedRW &AliasRW =
1471         SchedModels.getSchedRW((*AI)->getValueAsDef("AliasRW"));
1472 
1473     if (AliasRW.HasVariants) {
1474       const RecVec VarDefs = AliasRW.TheDef->getValueAsListOfDefs("Variants");
1475       for (Record *VD : VarDefs)
1476         Variants.emplace_back(VD, AliasRW.Index, AliasProcIdx, 0);
1477     }
1478     if (AliasRW.IsSequence)
1479       Variants.emplace_back(AliasRW.TheDef, SchedRW.Index, AliasProcIdx, 0);
1480     if (AliasProcIdx == 0)
1481       GenericRW = true;
1482   }
1483   std::vector<Record *> AllPreds =
1484       getAllPredicates(Variants, TransVec[TransIdx].ProcIndex);
1485   for (TransVariant &Variant : Variants) {
1486     // Don't expand variants if the processor models don't intersect.
1487     // A zero processor index means any processor.
1488     if (Variant.VarOrSeqDef->isSubClassOf("SchedVar")) {
1489       Record *PredDef = Variant.VarOrSeqDef->getValueAsDef("Predicate");
1490       if (mutuallyExclusive(PredDef, AllPreds, TransVec[TransIdx].PredTerm))
1491         continue;
1492     }
1493 
1494     if (IntersectingVariants.empty()) {
1495       // The first variant builds on the existing transition.
1496       Variant.TransVecIdx = TransIdx;
1497       IntersectingVariants.push_back(Variant);
1498     } else {
1499       // Push another copy of the current transition for more variants.
1500       Variant.TransVecIdx = TransVec.size();
1501       IntersectingVariants.push_back(Variant);
1502       TransVec.push_back(TransVec[TransIdx]);
1503     }
1504   }
1505   if (GenericRW && IntersectingVariants.empty()) {
1506     PrintFatalError(SchedRW.TheDef->getLoc(),
1507                     "No variant of this type has "
1508                     "a matching predicate on any processor");
1509   }
1510 }
1511 
1512 // Push the Reads/Writes selected by this variant onto the PredTransition
1513 // specified by VInfo.
pushVariant(const TransVariant & VInfo,bool IsRead)1514 void PredTransitions::pushVariant(const TransVariant &VInfo, bool IsRead) {
1515   PredTransition &Trans = TransVec[VInfo.TransVecIdx];
1516 
1517   // If this operand transition is reached through a processor-specific alias,
1518   // then the whole transition is specific to this processor.
1519   IdxVec SelectedRWs;
1520   if (VInfo.VarOrSeqDef->isSubClassOf("SchedVar")) {
1521     Record *PredDef = VInfo.VarOrSeqDef->getValueAsDef("Predicate");
1522     Trans.PredTerm.emplace_back(IsRead, VInfo.RWIdx, PredDef);
1523     RecVec SelectedDefs = VInfo.VarOrSeqDef->getValueAsListOfDefs("Selected");
1524     SchedModels.findRWs(SelectedDefs, SelectedRWs, IsRead);
1525   } else {
1526     assert(VInfo.VarOrSeqDef->isSubClassOf("WriteSequence") &&
1527            "variant must be a SchedVariant or aliased WriteSequence");
1528     SelectedRWs.push_back(SchedModels.getSchedRWIdx(VInfo.VarOrSeqDef, IsRead));
1529   }
1530 
1531   const CodeGenSchedRW &SchedRW = SchedModels.getSchedRW(VInfo.RWIdx, IsRead);
1532 
1533   SmallVectorImpl<SmallVector<unsigned, 4>> &RWSequences =
1534       IsRead ? Trans.ReadSequences : Trans.WriteSequences;
1535   if (SchedRW.IsVariadic) {
1536     unsigned OperIdx = RWSequences.size() - 1;
1537     // Make N-1 copies of this transition's last sequence.
1538     RWSequences.reserve(RWSequences.size() + SelectedRWs.size() - 1);
1539     RWSequences.insert(RWSequences.end(), SelectedRWs.size() - 1,
1540                        RWSequences[OperIdx]);
1541     // Push each of the N elements of the SelectedRWs onto a copy of the last
1542     // sequence (split the current operand into N operands).
1543     // Note that write sequences should be expanded within this loop--the entire
1544     // sequence belongs to a single operand.
1545     for (IdxIter RWI = SelectedRWs.begin(), RWE = SelectedRWs.end(); RWI != RWE;
1546          ++RWI, ++OperIdx) {
1547       IdxVec ExpandedRWs;
1548       if (IsRead)
1549         ExpandedRWs.push_back(*RWI);
1550       else
1551         SchedModels.expandRWSequence(*RWI, ExpandedRWs, IsRead);
1552       llvm::append_range(RWSequences[OperIdx], ExpandedRWs);
1553     }
1554     assert(OperIdx == RWSequences.size() && "missed a sequence");
1555   } else {
1556     // Push this transition's expanded sequence onto this transition's last
1557     // sequence (add to the current operand's sequence).
1558     SmallVectorImpl<unsigned> &Seq = RWSequences.back();
1559     IdxVec ExpandedRWs;
1560     for (unsigned int SelectedRW : SelectedRWs) {
1561       if (IsRead)
1562         ExpandedRWs.push_back(SelectedRW);
1563       else
1564         SchedModels.expandRWSequence(SelectedRW, ExpandedRWs, IsRead);
1565     }
1566     llvm::append_range(Seq, ExpandedRWs);
1567   }
1568 }
1569 
1570 // RWSeq is a sequence of all Reads or all Writes for the next read or write
1571 // operand. StartIdx is an index into TransVec where partial results
1572 // starts. RWSeq must be applied to all transitions between StartIdx and the end
1573 // of TransVec.
substituteVariantOperand(const SmallVectorImpl<unsigned> & RWSeq,bool IsRead,unsigned StartIdx)1574 bool PredTransitions::substituteVariantOperand(
1575     const SmallVectorImpl<unsigned> &RWSeq, bool IsRead, unsigned StartIdx) {
1576   bool Subst = false;
1577   // Visit each original RW within the current sequence.
1578   for (unsigned int RWI : RWSeq) {
1579     const CodeGenSchedRW &SchedRW = SchedModels.getSchedRW(RWI, IsRead);
1580     // Push this RW on all partial PredTransitions or distribute variants.
1581     // New PredTransitions may be pushed within this loop which should not be
1582     // revisited (TransEnd must be loop invariant).
1583     for (unsigned TransIdx = StartIdx, TransEnd = TransVec.size();
1584          TransIdx != TransEnd; ++TransIdx) {
1585       // Distribute this partial PredTransition across intersecting variants.
1586       // This will push a copies of TransVec[TransIdx] on the back of TransVec.
1587       std::vector<TransVariant> IntersectingVariants;
1588       getIntersectingVariants(SchedRW, TransIdx, IntersectingVariants);
1589       // Now expand each variant on top of its copy of the transition.
1590       for (const TransVariant &IV : IntersectingVariants)
1591         pushVariant(IV, IsRead);
1592       if (IntersectingVariants.empty()) {
1593         if (IsRead)
1594           TransVec[TransIdx].ReadSequences.back().push_back(RWI);
1595         else
1596           TransVec[TransIdx].WriteSequences.back().push_back(RWI);
1597         continue;
1598       } else {
1599         Subst = true;
1600       }
1601     }
1602   }
1603   return Subst;
1604 }
1605 
1606 // For each variant of a Read/Write in Trans, substitute the sequence of
1607 // Read/Writes guarded by the variant. This is exponential in the number of
1608 // variant Read/Writes, but in practice detection of mutually exclusive
1609 // predicates should result in linear growth in the total number variants.
1610 //
1611 // This is one step in a breadth-first search of nested variants.
substituteVariants(const PredTransition & Trans)1612 bool PredTransitions::substituteVariants(const PredTransition &Trans) {
1613   // Build up a set of partial results starting at the back of
1614   // PredTransitions. Remember the first new transition.
1615   unsigned StartIdx = TransVec.size();
1616   bool Subst = false;
1617   assert(Trans.ProcIndex != 0);
1618   TransVec.emplace_back(Trans.PredTerm, Trans.ProcIndex);
1619 
1620   // Visit each original write sequence.
1621   for (const auto &WriteSequence : Trans.WriteSequences) {
1622     // Push a new (empty) write sequence onto all partial Transitions.
1623     for (std::vector<PredTransition>::iterator I = TransVec.begin() + StartIdx,
1624                                                E = TransVec.end();
1625          I != E; ++I) {
1626       I->WriteSequences.emplace_back();
1627     }
1628     Subst |=
1629         substituteVariantOperand(WriteSequence, /*IsRead=*/false, StartIdx);
1630   }
1631   // Visit each original read sequence.
1632   for (const auto &ReadSequence : Trans.ReadSequences) {
1633     // Push a new (empty) read sequence onto all partial Transitions.
1634     for (std::vector<PredTransition>::iterator I = TransVec.begin() + StartIdx,
1635                                                E = TransVec.end();
1636          I != E; ++I) {
1637       I->ReadSequences.emplace_back();
1638     }
1639     Subst |= substituteVariantOperand(ReadSequence, /*IsRead=*/true, StartIdx);
1640   }
1641   return Subst;
1642 }
1643 
addSequences(CodeGenSchedModels & SchedModels,const SmallVectorImpl<SmallVector<unsigned,4>> & Seqs,IdxVec & Result,bool IsRead)1644 static void addSequences(CodeGenSchedModels &SchedModels,
1645                          const SmallVectorImpl<SmallVector<unsigned, 4>> &Seqs,
1646                          IdxVec &Result, bool IsRead) {
1647   for (const auto &S : Seqs)
1648     if (!S.empty())
1649       Result.push_back(SchedModels.findOrInsertRW(S, IsRead));
1650 }
1651 
1652 #ifndef NDEBUG
dumpRecVec(const RecVec & RV)1653 static void dumpRecVec(const RecVec &RV) {
1654   for (const Record *R : RV)
1655     dbgs() << R->getName() << ", ";
1656 }
1657 #endif
1658 
dumpTransition(const CodeGenSchedModels & SchedModels,const CodeGenSchedClass & FromSC,const CodeGenSchedTransition & SCTrans,const RecVec & Preds)1659 static void dumpTransition(const CodeGenSchedModels &SchedModels,
1660                            const CodeGenSchedClass &FromSC,
1661                            const CodeGenSchedTransition &SCTrans,
1662                            const RecVec &Preds) {
1663   LLVM_DEBUG(dbgs() << "Adding transition from " << FromSC.Name << "("
1664                     << FromSC.Index << ") to "
1665                     << SchedModels.getSchedClass(SCTrans.ToClassIdx).Name << "("
1666                     << SCTrans.ToClassIdx << ") on pred term: (";
1667              dumpRecVec(Preds);
1668              dbgs() << ") on processor (" << SCTrans.ProcIndex << ")\n");
1669 }
1670 // Create a new SchedClass for each variant found by inferFromRW. Pass
inferFromTransitions(ArrayRef<PredTransition> LastTransitions,unsigned FromClassIdx,CodeGenSchedModels & SchedModels)1671 static void inferFromTransitions(ArrayRef<PredTransition> LastTransitions,
1672                                  unsigned FromClassIdx,
1673                                  CodeGenSchedModels &SchedModels) {
1674   // For each PredTransition, create a new CodeGenSchedTransition, which usually
1675   // requires creating a new SchedClass.
1676   for (const auto &LastTransition : LastTransitions) {
1677     // Variant expansion (substituteVariants) may create unconditional
1678     // transitions. We don't need to build sched classes for them.
1679     if (LastTransition.PredTerm.empty())
1680       continue;
1681     IdxVec OperWritesVariant, OperReadsVariant;
1682     addSequences(SchedModels, LastTransition.WriteSequences, OperWritesVariant,
1683                  false);
1684     addSequences(SchedModels, LastTransition.ReadSequences, OperReadsVariant,
1685                  true);
1686     CodeGenSchedTransition SCTrans;
1687 
1688     // Transition should not contain processor indices already assigned to
1689     // InstRWs in this scheduling class.
1690     const CodeGenSchedClass &FromSC = SchedModels.getSchedClass(FromClassIdx);
1691     if (FromSC.InstRWProcIndices.count(LastTransition.ProcIndex))
1692       continue;
1693     SCTrans.ProcIndex = LastTransition.ProcIndex;
1694     SCTrans.ToClassIdx =
1695         SchedModels.addSchedClass(/*ItinClassDef=*/nullptr, OperWritesVariant,
1696                                   OperReadsVariant, LastTransition.ProcIndex);
1697 
1698     // The final PredTerm is unique set of predicates guarding the transition.
1699     RecVec Preds;
1700     transform(LastTransition.PredTerm, std::back_inserter(Preds),
1701               [](const PredCheck &P) { return P.Predicate; });
1702     Preds.erase(llvm::unique(Preds), Preds.end());
1703     dumpTransition(SchedModels, FromSC, SCTrans, Preds);
1704     SCTrans.PredTerm = std::move(Preds);
1705     SchedModels.getSchedClass(FromClassIdx)
1706         .Transitions.push_back(std::move(SCTrans));
1707   }
1708 }
1709 
getAllProcIndices() const1710 std::vector<unsigned> CodeGenSchedModels::getAllProcIndices() const {
1711   std::vector<unsigned> ProcIdVec;
1712   for (const auto &PM : ProcModelMap)
1713     if (PM.second != 0)
1714       ProcIdVec.push_back(PM.second);
1715   // The order of the keys (Record pointers) of ProcModelMap are not stable.
1716   // Sort to stabalize the values.
1717   llvm::sort(ProcIdVec);
1718   return ProcIdVec;
1719 }
1720 
1721 static std::vector<PredTransition>
makePerProcessorTransitions(const PredTransition & Trans,ArrayRef<unsigned> ProcIndices)1722 makePerProcessorTransitions(const PredTransition &Trans,
1723                             ArrayRef<unsigned> ProcIndices) {
1724   std::vector<PredTransition> PerCpuTransVec;
1725   for (unsigned ProcId : ProcIndices) {
1726     assert(ProcId != 0);
1727     PerCpuTransVec.push_back(Trans);
1728     PerCpuTransVec.back().ProcIndex = ProcId;
1729   }
1730   return PerCpuTransVec;
1731 }
1732 
1733 // Create new SchedClasses for the given ReadWrite list. If any of the
1734 // ReadWrites refers to a SchedVariant, create a new SchedClass for each variant
1735 // of the ReadWrite list, following Aliases if necessary.
inferFromRW(ArrayRef<unsigned> OperWrites,ArrayRef<unsigned> OperReads,unsigned FromClassIdx,ArrayRef<unsigned> ProcIndices)1736 void CodeGenSchedModels::inferFromRW(ArrayRef<unsigned> OperWrites,
1737                                      ArrayRef<unsigned> OperReads,
1738                                      unsigned FromClassIdx,
1739                                      ArrayRef<unsigned> ProcIndices) {
1740   LLVM_DEBUG(dbgs() << "INFER RW proc("; dumpIdxVec(ProcIndices);
1741              dbgs() << ") ");
1742   // Create a seed transition with an empty PredTerm and the expanded sequences
1743   // of SchedWrites for the current SchedClass.
1744   std::vector<PredTransition> LastTransitions;
1745   LastTransitions.emplace_back();
1746 
1747   for (unsigned WriteIdx : OperWrites) {
1748     IdxVec WriteSeq;
1749     expandRWSequence(WriteIdx, WriteSeq, /*IsRead=*/false);
1750     LastTransitions[0].WriteSequences.emplace_back();
1751     SmallVectorImpl<unsigned> &Seq = LastTransitions[0].WriteSequences.back();
1752     Seq.append(WriteSeq.begin(), WriteSeq.end());
1753     LLVM_DEBUG(dbgs() << "("; dumpIdxVec(Seq); dbgs() << ") ");
1754   }
1755   LLVM_DEBUG(dbgs() << " Reads: ");
1756   for (unsigned ReadIdx : OperReads) {
1757     IdxVec ReadSeq;
1758     expandRWSequence(ReadIdx, ReadSeq, /*IsRead=*/true);
1759     LastTransitions[0].ReadSequences.emplace_back();
1760     SmallVectorImpl<unsigned> &Seq = LastTransitions[0].ReadSequences.back();
1761     Seq.append(ReadSeq.begin(), ReadSeq.end());
1762     LLVM_DEBUG(dbgs() << "("; dumpIdxVec(Seq); dbgs() << ") ");
1763   }
1764   LLVM_DEBUG(dbgs() << '\n');
1765 
1766   LastTransitions = makePerProcessorTransitions(
1767       LastTransitions[0], llvm::is_contained(ProcIndices, 0)
1768                               ? ArrayRef<unsigned>(getAllProcIndices())
1769                               : ProcIndices);
1770   // Collect all PredTransitions for individual operands.
1771   // Iterate until no variant writes remain.
1772   bool SubstitutedAny;
1773   do {
1774     SubstitutedAny = false;
1775     PredTransitions Transitions(*this);
1776     for (const PredTransition &Trans : LastTransitions)
1777       SubstitutedAny |= Transitions.substituteVariants(Trans);
1778     LLVM_DEBUG(Transitions.dump());
1779     LastTransitions = std::move(Transitions.TransVec);
1780   } while (SubstitutedAny);
1781 
1782   // WARNING: We are about to mutate the SchedClasses vector. Do not refer to
1783   // OperWrites, OperReads, or ProcIndices after calling inferFromTransitions.
1784   inferFromTransitions(LastTransitions, FromClassIdx, *this);
1785 }
1786 
1787 // Check if any processor resource group contains all resource records in
1788 // SubUnits.
hasSuperGroup(RecVec & SubUnits,CodeGenProcModel & PM)1789 bool CodeGenSchedModels::hasSuperGroup(RecVec &SubUnits, CodeGenProcModel &PM) {
1790   for (Record *ProcResourceDef : PM.ProcResourceDefs) {
1791     if (!ProcResourceDef->isSubClassOf("ProcResGroup"))
1792       continue;
1793     RecVec SuperUnits = ProcResourceDef->getValueAsListOfDefs("Resources");
1794     RecIter RI = SubUnits.begin(), RE = SubUnits.end();
1795     for (; RI != RE; ++RI) {
1796       if (!is_contained(SuperUnits, *RI)) {
1797         break;
1798       }
1799     }
1800     if (RI == RE)
1801       return true;
1802   }
1803   return false;
1804 }
1805 
1806 // Verify that overlapping groups have a common supergroup.
verifyProcResourceGroups(CodeGenProcModel & PM)1807 void CodeGenSchedModels::verifyProcResourceGroups(CodeGenProcModel &PM) {
1808   for (unsigned i = 0, e = PM.ProcResourceDefs.size(); i < e; ++i) {
1809     if (!PM.ProcResourceDefs[i]->isSubClassOf("ProcResGroup"))
1810       continue;
1811     RecVec CheckUnits =
1812         PM.ProcResourceDefs[i]->getValueAsListOfDefs("Resources");
1813     for (unsigned j = i + 1; j < e; ++j) {
1814       if (!PM.ProcResourceDefs[j]->isSubClassOf("ProcResGroup"))
1815         continue;
1816       RecVec OtherUnits =
1817           PM.ProcResourceDefs[j]->getValueAsListOfDefs("Resources");
1818       if (std::find_first_of(CheckUnits.begin(), CheckUnits.end(),
1819                              OtherUnits.begin(),
1820                              OtherUnits.end()) != CheckUnits.end()) {
1821         // CheckUnits and OtherUnits overlap
1822         llvm::append_range(OtherUnits, CheckUnits);
1823         if (!hasSuperGroup(OtherUnits, PM)) {
1824           PrintFatalError((PM.ProcResourceDefs[i])->getLoc(),
1825                           "proc resource group overlaps with " +
1826                               PM.ProcResourceDefs[j]->getName() +
1827                               " but no supergroup contains both.");
1828         }
1829       }
1830     }
1831   }
1832 }
1833 
1834 // Collect all the RegisterFile definitions available in this target.
collectRegisterFiles()1835 void CodeGenSchedModels::collectRegisterFiles() {
1836   RecVec RegisterFileDefs = Records.getAllDerivedDefinitions("RegisterFile");
1837 
1838   // RegisterFiles is the vector of CodeGenRegisterFile.
1839   for (Record *RF : RegisterFileDefs) {
1840     // For each register file definition, construct a CodeGenRegisterFile object
1841     // and add it to the appropriate scheduling model.
1842     CodeGenProcModel &PM = getProcModel(RF->getValueAsDef("SchedModel"));
1843     PM.RegisterFiles.emplace_back(CodeGenRegisterFile(RF->getName(), RF));
1844     CodeGenRegisterFile &CGRF = PM.RegisterFiles.back();
1845     CGRF.MaxMovesEliminatedPerCycle =
1846         RF->getValueAsInt("MaxMovesEliminatedPerCycle");
1847     CGRF.AllowZeroMoveEliminationOnly =
1848         RF->getValueAsBit("AllowZeroMoveEliminationOnly");
1849 
1850     // Now set the number of physical registers as well as the cost of registers
1851     // in each register class.
1852     CGRF.NumPhysRegs = RF->getValueAsInt("NumPhysRegs");
1853     if (!CGRF.NumPhysRegs) {
1854       PrintFatalError(RF->getLoc(),
1855                       "Invalid RegisterFile with zero physical registers");
1856     }
1857 
1858     RecVec RegisterClasses = RF->getValueAsListOfDefs("RegClasses");
1859     std::vector<int64_t> RegisterCosts = RF->getValueAsListOfInts("RegCosts");
1860     ListInit *MoveElimInfo = RF->getValueAsListInit("AllowMoveElimination");
1861     for (unsigned I = 0, E = RegisterClasses.size(); I < E; ++I) {
1862       int Cost = RegisterCosts.size() > I ? RegisterCosts[I] : 1;
1863 
1864       bool AllowMoveElim = false;
1865       if (MoveElimInfo->size() > I) {
1866         BitInit *Val = cast<BitInit>(MoveElimInfo->getElement(I));
1867         AllowMoveElim = Val->getValue();
1868       }
1869 
1870       CGRF.Costs.emplace_back(RegisterClasses[I], Cost, AllowMoveElim);
1871     }
1872   }
1873 }
1874 
1875 // Collect and sort WriteRes, ReadAdvance, and ProcResources.
collectProcResources()1876 void CodeGenSchedModels::collectProcResources() {
1877   ProcResourceDefs = Records.getAllDerivedDefinitions("ProcResourceUnits");
1878   ProcResGroups = Records.getAllDerivedDefinitions("ProcResGroup");
1879 
1880   // Add any subtarget-specific SchedReadWrites that are directly associated
1881   // with processor resources. Refer to the parent SchedClass's ProcIndices to
1882   // determine which processors they apply to.
1883   for (const CodeGenSchedClass &SC :
1884        make_range(schedClassBegin(), schedClassEnd())) {
1885     if (SC.ItinClassDef) {
1886       collectItinProcResources(SC.ItinClassDef);
1887       continue;
1888     }
1889 
1890     // This class may have a default ReadWrite list which can be overriden by
1891     // InstRW definitions.
1892     for (Record *RW : SC.InstRWs) {
1893       Record *RWModelDef = RW->getValueAsDef("SchedModel");
1894       unsigned PIdx = getProcModel(RWModelDef).Index;
1895       IdxVec Writes, Reads;
1896       findRWs(RW->getValueAsListOfDefs("OperandReadWrites"), Writes, Reads);
1897       collectRWResources(Writes, Reads, PIdx);
1898     }
1899 
1900     collectRWResources(SC.Writes, SC.Reads, SC.ProcIndices);
1901   }
1902   // Add resources separately defined by each subtarget.
1903   RecVec WRDefs = Records.getAllDerivedDefinitions("WriteRes");
1904   for (Record *WR : WRDefs) {
1905     Record *ModelDef = WR->getValueAsDef("SchedModel");
1906     addWriteRes(WR, getProcModel(ModelDef).Index);
1907   }
1908   RecVec SWRDefs = Records.getAllDerivedDefinitions("SchedWriteRes");
1909   for (Record *SWR : SWRDefs) {
1910     Record *ModelDef = SWR->getValueAsDef("SchedModel");
1911     addWriteRes(SWR, getProcModel(ModelDef).Index);
1912   }
1913   RecVec RADefs = Records.getAllDerivedDefinitions("ReadAdvance");
1914   for (Record *RA : RADefs) {
1915     Record *ModelDef = RA->getValueAsDef("SchedModel");
1916     addReadAdvance(RA, getProcModel(ModelDef).Index);
1917   }
1918   RecVec SRADefs = Records.getAllDerivedDefinitions("SchedReadAdvance");
1919   for (Record *SRA : SRADefs) {
1920     if (SRA->getValueInit("SchedModel")->isComplete()) {
1921       Record *ModelDef = SRA->getValueAsDef("SchedModel");
1922       addReadAdvance(SRA, getProcModel(ModelDef).Index);
1923     }
1924   }
1925   // Add ProcResGroups that are defined within this processor model, which may
1926   // not be directly referenced but may directly specify a buffer size.
1927   RecVec ProcResGroups = Records.getAllDerivedDefinitions("ProcResGroup");
1928   for (Record *PRG : ProcResGroups) {
1929     if (!PRG->getValueInit("SchedModel")->isComplete())
1930       continue;
1931     CodeGenProcModel &PM = getProcModel(PRG->getValueAsDef("SchedModel"));
1932     if (!is_contained(PM.ProcResourceDefs, PRG))
1933       PM.ProcResourceDefs.push_back(PRG);
1934   }
1935   // Add ProcResourceUnits unconditionally.
1936   for (Record *PRU : Records.getAllDerivedDefinitions("ProcResourceUnits")) {
1937     if (!PRU->getValueInit("SchedModel")->isComplete())
1938       continue;
1939     CodeGenProcModel &PM = getProcModel(PRU->getValueAsDef("SchedModel"));
1940     if (!is_contained(PM.ProcResourceDefs, PRU))
1941       PM.ProcResourceDefs.push_back(PRU);
1942   }
1943   // Finalize each ProcModel by sorting the record arrays.
1944   for (CodeGenProcModel &PM : ProcModels) {
1945     llvm::sort(PM.WriteResDefs, LessRecord());
1946     llvm::sort(PM.ReadAdvanceDefs, LessRecord());
1947     llvm::sort(PM.ProcResourceDefs, LessRecord());
1948     LLVM_DEBUG(
1949         PM.dump(); dbgs() << "WriteResDefs: "; for (auto WriteResDef
1950                                                     : PM.WriteResDefs) {
1951           if (WriteResDef->isSubClassOf("WriteRes"))
1952             dbgs() << WriteResDef->getValueAsDef("WriteType")->getName() << " ";
1953           else
1954             dbgs() << WriteResDef->getName() << " ";
1955         } dbgs() << "\nReadAdvanceDefs: ";
1956         for (Record *ReadAdvanceDef
1957              : PM.ReadAdvanceDefs) {
1958           if (ReadAdvanceDef->isSubClassOf("ReadAdvance"))
1959             dbgs() << ReadAdvanceDef->getValueAsDef("ReadType")->getName()
1960                    << " ";
1961           else
1962             dbgs() << ReadAdvanceDef->getName() << " ";
1963         } dbgs()
1964         << "\nProcResourceDefs: ";
1965         for (Record *ProcResourceDef
1966              : PM.ProcResourceDefs) {
1967           dbgs() << ProcResourceDef->getName() << " ";
1968         } dbgs()
1969         << '\n');
1970     verifyProcResourceGroups(PM);
1971   }
1972 
1973   ProcResourceDefs.clear();
1974   ProcResGroups.clear();
1975 }
1976 
checkCompleteness()1977 void CodeGenSchedModels::checkCompleteness() {
1978   bool Complete = true;
1979   for (const CodeGenProcModel &ProcModel : procModels()) {
1980     const bool HasItineraries = ProcModel.hasItineraries();
1981     if (!ProcModel.ModelDef->getValueAsBit("CompleteModel"))
1982       continue;
1983     for (const CodeGenInstruction *Inst : Target.getInstructionsByEnumValue()) {
1984       if (Inst->hasNoSchedulingInfo)
1985         continue;
1986       if (ProcModel.isUnsupported(*Inst))
1987         continue;
1988       unsigned SCIdx = getSchedClassIdx(*Inst);
1989       if (!SCIdx) {
1990         if (Inst->TheDef->isValueUnset("SchedRW")) {
1991           PrintError(Inst->TheDef->getLoc(),
1992                      "No schedule information for instruction '" +
1993                          Inst->TheDef->getName() + "' in SchedMachineModel '" +
1994                          ProcModel.ModelDef->getName() + "'");
1995           Complete = false;
1996         }
1997         continue;
1998       }
1999 
2000       const CodeGenSchedClass &SC = getSchedClass(SCIdx);
2001       if (!SC.Writes.empty())
2002         continue;
2003       if (HasItineraries && SC.ItinClassDef != nullptr &&
2004           SC.ItinClassDef->getName() != "NoItinerary")
2005         continue;
2006 
2007       const RecVec &InstRWs = SC.InstRWs;
2008       auto I = find_if(InstRWs, [&ProcModel](const Record *R) {
2009         return R->getValueAsDef("SchedModel") == ProcModel.ModelDef;
2010       });
2011       if (I == InstRWs.end()) {
2012         PrintError(Inst->TheDef->getLoc(), "'" + ProcModel.ModelName +
2013                                                "' lacks information for '" +
2014                                                Inst->TheDef->getName() + "'");
2015         Complete = false;
2016       }
2017     }
2018   }
2019   if (!Complete) {
2020     errs()
2021         << "\n\nIncomplete schedule models found.\n"
2022         << "- Consider setting 'CompleteModel = 0' while developing new "
2023            "models.\n"
2024         << "- Pseudo instructions can be marked with 'hasNoSchedulingInfo = "
2025            "1'.\n"
2026         << "- Instructions should usually have Sched<[...]> as a superclass, "
2027            "you may temporarily use an empty list.\n"
2028         << "- Instructions related to unsupported features can be excluded "
2029            "with "
2030            "list<Predicate> UnsupportedFeatures = [HasA,..,HasY]; in the "
2031            "processor model.\n\n";
2032     PrintFatalError("Incomplete schedule model");
2033   }
2034 }
2035 
2036 // Collect itinerary class resources for each processor.
collectItinProcResources(Record * ItinClassDef)2037 void CodeGenSchedModels::collectItinProcResources(Record *ItinClassDef) {
2038   for (unsigned PIdx = 0, PEnd = ProcModels.size(); PIdx != PEnd; ++PIdx) {
2039     const CodeGenProcModel &PM = ProcModels[PIdx];
2040     // For all ItinRW entries.
2041     bool HasMatch = false;
2042     for (RecIter II = PM.ItinRWDefs.begin(), IE = PM.ItinRWDefs.end(); II != IE;
2043          ++II) {
2044       RecVec Matched = (*II)->getValueAsListOfDefs("MatchedItinClasses");
2045       if (!llvm::is_contained(Matched, ItinClassDef))
2046         continue;
2047       if (HasMatch)
2048         PrintFatalError((*II)->getLoc(),
2049                         "Duplicate itinerary class " + ItinClassDef->getName() +
2050                             " in ItinResources for " + PM.ModelName);
2051       HasMatch = true;
2052       IdxVec Writes, Reads;
2053       findRWs((*II)->getValueAsListOfDefs("OperandReadWrites"), Writes, Reads);
2054       collectRWResources(Writes, Reads, PIdx);
2055     }
2056   }
2057 }
2058 
collectRWResources(unsigned RWIdx,bool IsRead,ArrayRef<unsigned> ProcIndices)2059 void CodeGenSchedModels::collectRWResources(unsigned RWIdx, bool IsRead,
2060                                             ArrayRef<unsigned> ProcIndices) {
2061   const CodeGenSchedRW &SchedRW = getSchedRW(RWIdx, IsRead);
2062   if (SchedRW.TheDef) {
2063     if (!IsRead && SchedRW.TheDef->isSubClassOf("SchedWriteRes")) {
2064       for (unsigned Idx : ProcIndices)
2065         addWriteRes(SchedRW.TheDef, Idx);
2066     } else if (IsRead && SchedRW.TheDef->isSubClassOf("SchedReadAdvance")) {
2067       for (unsigned Idx : ProcIndices)
2068         addReadAdvance(SchedRW.TheDef, Idx);
2069     }
2070   }
2071   for (auto *Alias : SchedRW.Aliases) {
2072     IdxVec AliasProcIndices;
2073     if (Alias->getValueInit("SchedModel")->isComplete()) {
2074       AliasProcIndices.push_back(
2075           getProcModel(Alias->getValueAsDef("SchedModel")).Index);
2076     } else
2077       AliasProcIndices = ProcIndices;
2078     const CodeGenSchedRW &AliasRW = getSchedRW(Alias->getValueAsDef("AliasRW"));
2079     assert(AliasRW.IsRead == IsRead && "cannot alias reads to writes");
2080 
2081     IdxVec ExpandedRWs;
2082     expandRWSequence(AliasRW.Index, ExpandedRWs, IsRead);
2083     for (unsigned int ExpandedRW : ExpandedRWs) {
2084       collectRWResources(ExpandedRW, IsRead, AliasProcIndices);
2085     }
2086   }
2087 }
2088 
2089 // Collect resources for a set of read/write types and processor indices.
collectRWResources(ArrayRef<unsigned> Writes,ArrayRef<unsigned> Reads,ArrayRef<unsigned> ProcIndices)2090 void CodeGenSchedModels::collectRWResources(ArrayRef<unsigned> Writes,
2091                                             ArrayRef<unsigned> Reads,
2092                                             ArrayRef<unsigned> ProcIndices) {
2093   for (unsigned Idx : Writes)
2094     collectRWResources(Idx, /*IsRead=*/false, ProcIndices);
2095 
2096   for (unsigned Idx : Reads)
2097     collectRWResources(Idx, /*IsRead=*/true, ProcIndices);
2098 }
2099 
2100 // Find the processor's resource units for this kind of resource.
findProcResUnits(Record * ProcResKind,const CodeGenProcModel & PM,ArrayRef<SMLoc> Loc) const2101 Record *CodeGenSchedModels::findProcResUnits(Record *ProcResKind,
2102                                              const CodeGenProcModel &PM,
2103                                              ArrayRef<SMLoc> Loc) const {
2104   if (ProcResKind->isSubClassOf("ProcResourceUnits"))
2105     return ProcResKind;
2106 
2107   Record *ProcUnitDef = nullptr;
2108   assert(!ProcResourceDefs.empty());
2109   assert(!ProcResGroups.empty());
2110 
2111   for (Record *ProcResDef : ProcResourceDefs) {
2112     if (ProcResDef->getValueAsDef("Kind") == ProcResKind &&
2113         ProcResDef->getValueAsDef("SchedModel") == PM.ModelDef) {
2114       if (ProcUnitDef) {
2115         PrintFatalError(Loc,
2116                         "Multiple ProcessorResourceUnits associated with " +
2117                             ProcResKind->getName());
2118       }
2119       ProcUnitDef = ProcResDef;
2120     }
2121   }
2122   for (Record *ProcResGroup : ProcResGroups) {
2123     if (ProcResGroup == ProcResKind &&
2124         ProcResGroup->getValueAsDef("SchedModel") == PM.ModelDef) {
2125       if (ProcUnitDef) {
2126         PrintFatalError(Loc,
2127                         "Multiple ProcessorResourceUnits associated with " +
2128                             ProcResKind->getName());
2129       }
2130       ProcUnitDef = ProcResGroup;
2131     }
2132   }
2133   if (!ProcUnitDef) {
2134     PrintFatalError(Loc, "No ProcessorResources associated with " +
2135                              ProcResKind->getName());
2136   }
2137   return ProcUnitDef;
2138 }
2139 
2140 // Iteratively add a resource and its super resources.
addProcResource(Record * ProcResKind,CodeGenProcModel & PM,ArrayRef<SMLoc> Loc)2141 void CodeGenSchedModels::addProcResource(Record *ProcResKind,
2142                                          CodeGenProcModel &PM,
2143                                          ArrayRef<SMLoc> Loc) {
2144   while (true) {
2145     Record *ProcResUnits = findProcResUnits(ProcResKind, PM, Loc);
2146 
2147     // See if this ProcResource is already associated with this processor.
2148     if (is_contained(PM.ProcResourceDefs, ProcResUnits))
2149       return;
2150 
2151     PM.ProcResourceDefs.push_back(ProcResUnits);
2152     if (ProcResUnits->isSubClassOf("ProcResGroup"))
2153       return;
2154 
2155     if (!ProcResUnits->getValueInit("Super")->isComplete())
2156       return;
2157 
2158     ProcResKind = ProcResUnits->getValueAsDef("Super");
2159   }
2160 }
2161 
2162 // Add resources for a SchedWrite to this processor if they don't exist.
addWriteRes(Record * ProcWriteResDef,unsigned PIdx)2163 void CodeGenSchedModels::addWriteRes(Record *ProcWriteResDef, unsigned PIdx) {
2164   assert(PIdx && "don't add resources to an invalid Processor model");
2165 
2166   RecVec &WRDefs = ProcModels[PIdx].WriteResDefs;
2167   if (is_contained(WRDefs, ProcWriteResDef))
2168     return;
2169   WRDefs.push_back(ProcWriteResDef);
2170 
2171   // Visit ProcResourceKinds referenced by the newly discovered WriteRes.
2172   RecVec ProcResDefs = ProcWriteResDef->getValueAsListOfDefs("ProcResources");
2173   for (auto *ProcResDef : ProcResDefs) {
2174     addProcResource(ProcResDef, ProcModels[PIdx], ProcWriteResDef->getLoc());
2175   }
2176 }
2177 
2178 // Add resources for a ReadAdvance to this processor if they don't exist.
addReadAdvance(Record * ProcReadAdvanceDef,unsigned PIdx)2179 void CodeGenSchedModels::addReadAdvance(Record *ProcReadAdvanceDef,
2180                                         unsigned PIdx) {
2181   for (const Record *ValidWrite :
2182        ProcReadAdvanceDef->getValueAsListOfDefs("ValidWrites"))
2183     if (getSchedRWIdx(ValidWrite, /*IsRead=*/false) == 0)
2184       PrintFatalError(
2185           ProcReadAdvanceDef->getLoc(),
2186           "ReadAdvance referencing a ValidWrite that is not used by "
2187           "any instruction (" +
2188               ValidWrite->getName() + ")");
2189 
2190   RecVec &RADefs = ProcModels[PIdx].ReadAdvanceDefs;
2191   if (is_contained(RADefs, ProcReadAdvanceDef))
2192     return;
2193   RADefs.push_back(ProcReadAdvanceDef);
2194 }
2195 
getProcResourceIdx(Record * PRDef) const2196 unsigned CodeGenProcModel::getProcResourceIdx(Record *PRDef) const {
2197   RecIter PRPos = find(ProcResourceDefs, PRDef);
2198   if (PRPos == ProcResourceDefs.end())
2199     PrintFatalError(PRDef->getLoc(), "ProcResource def is not included in "
2200                                      "the ProcResources list for " +
2201                                          ModelName);
2202   // Idx=0 is reserved for invalid.
2203   return 1 + (PRPos - ProcResourceDefs.begin());
2204 }
2205 
isUnsupported(const CodeGenInstruction & Inst) const2206 bool CodeGenProcModel::isUnsupported(const CodeGenInstruction &Inst) const {
2207   for (const Record *TheDef : UnsupportedFeaturesDefs) {
2208     for (const Record *PredDef :
2209          Inst.TheDef->getValueAsListOfDefs("Predicates")) {
2210       if (TheDef->getName() == PredDef->getName())
2211         return true;
2212     }
2213   }
2214   return false;
2215 }
2216 
hasReadOfWrite(Record * WriteDef) const2217 bool CodeGenProcModel::hasReadOfWrite(Record *WriteDef) const {
2218   for (auto &RADef : ReadAdvanceDefs) {
2219     RecVec ValidWrites = RADef->getValueAsListOfDefs("ValidWrites");
2220     if (is_contained(ValidWrites, WriteDef))
2221       return true;
2222   }
2223   return false;
2224 }
2225 
2226 #ifndef NDEBUG
dump() const2227 void CodeGenProcModel::dump() const {
2228   dbgs() << Index << ": " << ModelName << " "
2229          << (ModelDef ? ModelDef->getName() : "inferred") << " "
2230          << (ItinsDef ? ItinsDef->getName() : "no itinerary") << '\n';
2231 }
2232 
dump() const2233 void CodeGenSchedRW::dump() const {
2234   dbgs() << Name << (IsVariadic ? " (V) " : " ");
2235   if (IsSequence) {
2236     dbgs() << "(";
2237     dumpIdxVec(Sequence);
2238     dbgs() << ")";
2239   }
2240 }
2241 
dump(const CodeGenSchedModels * SchedModels) const2242 void CodeGenSchedClass::dump(const CodeGenSchedModels *SchedModels) const {
2243   dbgs() << "SCHEDCLASS " << Index << ":" << Name << '\n' << "  Writes: ";
2244   for (unsigned i = 0, N = Writes.size(); i < N; ++i) {
2245     SchedModels->getSchedWrite(Writes[i]).dump();
2246     if (i < N - 1) {
2247       dbgs() << '\n';
2248       dbgs().indent(10);
2249     }
2250   }
2251   dbgs() << "\n  Reads: ";
2252   for (unsigned i = 0, N = Reads.size(); i < N; ++i) {
2253     SchedModels->getSchedRead(Reads[i]).dump();
2254     if (i < N - 1) {
2255       dbgs() << '\n';
2256       dbgs().indent(10);
2257     }
2258   }
2259   dbgs() << "\n  ProcIdx: ";
2260   dumpIdxVec(ProcIndices);
2261   if (!Transitions.empty()) {
2262     dbgs() << "\n Transitions for Proc ";
2263     for (const CodeGenSchedTransition &Transition : Transitions) {
2264       dbgs() << Transition.ProcIndex << ", ";
2265     }
2266   }
2267   dbgs() << '\n';
2268 }
2269 
dump() const2270 void PredTransitions::dump() const {
2271   dbgs() << "Expanded Variants:\n";
2272   for (const auto &TI : TransVec) {
2273     dbgs() << "{";
2274     ListSeparator LS;
2275     for (const PredCheck &PC : TI.PredTerm)
2276       dbgs() << LS << SchedModels.getSchedRW(PC.RWIdx, PC.IsRead).Name << ":"
2277              << PC.Predicate->getName();
2278     dbgs() << "},\n  => {";
2279     for (SmallVectorImpl<SmallVector<unsigned, 4>>::const_iterator
2280              WSI = TI.WriteSequences.begin(),
2281              WSE = TI.WriteSequences.end();
2282          WSI != WSE; ++WSI) {
2283       dbgs() << "(";
2284       ListSeparator LS;
2285       for (unsigned N : *WSI)
2286         dbgs() << LS << SchedModels.getSchedWrite(N).Name;
2287       dbgs() << "),";
2288     }
2289     dbgs() << "}\n";
2290   }
2291 }
2292 #endif // NDEBUG
2293