1 //===- MLIRContext.cpp - MLIR Type Classes --------------------------------===//
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 #include "mlir/IR/MLIRContext.h"
10 #include "AffineExprDetail.h"
11 #include "AffineMapDetail.h"
12 #include "AttributeDetail.h"
13 #include "IntegerSetDetail.h"
14 #include "LocationDetail.h"
15 #include "TypeDetail.h"
16 #include "mlir/IR/AffineExpr.h"
17 #include "mlir/IR/AffineMap.h"
18 #include "mlir/IR/Attributes.h"
19 #include "mlir/IR/BuiltinDialect.h"
20 #include "mlir/IR/Diagnostics.h"
21 #include "mlir/IR/Dialect.h"
22 #include "mlir/IR/Identifier.h"
23 #include "mlir/IR/IntegerSet.h"
24 #include "mlir/IR/Location.h"
25 #include "mlir/IR/OpImplementation.h"
26 #include "mlir/IR/Types.h"
27 #include "mlir/Support/ThreadLocalCache.h"
28 #include "llvm/ADT/DenseMap.h"
29 #include "llvm/ADT/DenseSet.h"
30 #include "llvm/ADT/SetVector.h"
31 #include "llvm/ADT/StringSet.h"
32 #include "llvm/ADT/Twine.h"
33 #include "llvm/Support/Allocator.h"
34 #include "llvm/Support/CommandLine.h"
35 #include "llvm/Support/Debug.h"
36 #include "llvm/Support/RWMutex.h"
37 #include "llvm/Support/raw_ostream.h"
38 #include <memory>
39
40 #define DEBUG_TYPE "mlircontext"
41
42 using namespace mlir;
43 using namespace mlir::detail;
44
45 using llvm::hash_combine;
46 using llvm::hash_combine_range;
47
48 //===----------------------------------------------------------------------===//
49 // MLIRContext CommandLine Options
50 //===----------------------------------------------------------------------===//
51
52 namespace {
53 /// This struct contains command line options that can be used to initialize
54 /// various bits of an MLIRContext. This uses a struct wrapper to avoid the need
55 /// for global command line options.
56 struct MLIRContextOptions {
57 llvm::cl::opt<bool> disableThreading{
58 "mlir-disable-threading",
59 llvm::cl::desc("Disabling multi-threading within MLIR")};
60
61 llvm::cl::opt<bool> printOpOnDiagnostic{
62 "mlir-print-op-on-diagnostic",
63 llvm::cl::desc("When a diagnostic is emitted on an operation, also print "
64 "the operation as an attached note"),
65 llvm::cl::init(true)};
66
67 llvm::cl::opt<bool> printStackTraceOnDiagnostic{
68 "mlir-print-stacktrace-on-diagnostic",
69 llvm::cl::desc("When a diagnostic is emitted, also print the stack trace "
70 "as an attached note")};
71 };
72 } // end anonymous namespace
73
74 static llvm::ManagedStatic<MLIRContextOptions> clOptions;
75
76 /// Register a set of useful command-line options that can be used to configure
77 /// various flags within the MLIRContext. These flags are used when constructing
78 /// an MLIR context for initialization.
registerMLIRContextCLOptions()79 void mlir::registerMLIRContextCLOptions() {
80 // Make sure that the options struct has been initialized.
81 *clOptions;
82 }
83
84 //===----------------------------------------------------------------------===//
85 // Locking Utilities
86 //===----------------------------------------------------------------------===//
87
88 namespace {
89 /// Utility reader lock that takes a runtime flag that specifies if we really
90 /// need to lock.
91 struct ScopedReaderLock {
ScopedReaderLock__anonf91fc13c0211::ScopedReaderLock92 ScopedReaderLock(llvm::sys::SmartRWMutex<true> &mutexParam, bool shouldLock)
93 : mutex(shouldLock ? &mutexParam : nullptr) {
94 if (mutex)
95 mutex->lock_shared();
96 }
~ScopedReaderLock__anonf91fc13c0211::ScopedReaderLock97 ~ScopedReaderLock() {
98 if (mutex)
99 mutex->unlock_shared();
100 }
101 llvm::sys::SmartRWMutex<true> *mutex;
102 };
103 /// Utility writer lock that takes a runtime flag that specifies if we really
104 /// need to lock.
105 struct ScopedWriterLock {
ScopedWriterLock__anonf91fc13c0211::ScopedWriterLock106 ScopedWriterLock(llvm::sys::SmartRWMutex<true> &mutexParam, bool shouldLock)
107 : mutex(shouldLock ? &mutexParam : nullptr) {
108 if (mutex)
109 mutex->lock();
110 }
~ScopedWriterLock__anonf91fc13c0211::ScopedWriterLock111 ~ScopedWriterLock() {
112 if (mutex)
113 mutex->unlock();
114 }
115 llvm::sys::SmartRWMutex<true> *mutex;
116 };
117 } // end anonymous namespace.
118
119 //===----------------------------------------------------------------------===//
120 // AffineMap and IntegerSet hashing
121 //===----------------------------------------------------------------------===//
122
123 /// A utility function to safely get or create a uniqued instance within the
124 /// given set container.
125 template <typename ValueT, typename DenseInfoT, typename KeyT,
126 typename ConstructorFn>
safeGetOrCreate(DenseSet<ValueT,DenseInfoT> & container,KeyT && key,llvm::sys::SmartRWMutex<true> & mutex,bool threadingIsEnabled,ConstructorFn && constructorFn)127 static ValueT safeGetOrCreate(DenseSet<ValueT, DenseInfoT> &container,
128 KeyT &&key, llvm::sys::SmartRWMutex<true> &mutex,
129 bool threadingIsEnabled,
130 ConstructorFn &&constructorFn) {
131 // Check for an existing instance in read-only mode.
132 if (threadingIsEnabled) {
133 llvm::sys::SmartScopedReader<true> instanceLock(mutex);
134 auto it = container.find_as(key);
135 if (it != container.end())
136 return *it;
137 }
138
139 // Acquire a writer-lock so that we can safely create the new instance.
140 ScopedWriterLock instanceLock(mutex, threadingIsEnabled);
141
142 // Check for an existing instance again here, because another writer thread
143 // may have already created one. Otherwise, construct a new instance.
144 auto existing = container.insert_as(ValueT(), key);
145 if (existing.second)
146 return *existing.first = constructorFn();
147 return *existing.first;
148 }
149
150 namespace {
151 struct AffineMapKeyInfo : DenseMapInfo<AffineMap> {
152 // Affine maps are uniqued based on their dim/symbol counts and affine
153 // expressions.
154 using KeyTy = std::tuple<unsigned, unsigned, ArrayRef<AffineExpr>>;
155 using DenseMapInfo<AffineMap>::isEqual;
156
getHashValue__anonf91fc13c0311::AffineMapKeyInfo157 static unsigned getHashValue(const AffineMap &key) {
158 return getHashValue(
159 KeyTy(key.getNumDims(), key.getNumSymbols(), key.getResults()));
160 }
161
getHashValue__anonf91fc13c0311::AffineMapKeyInfo162 static unsigned getHashValue(KeyTy key) {
163 return hash_combine(
164 std::get<0>(key), std::get<1>(key),
165 hash_combine_range(std::get<2>(key).begin(), std::get<2>(key).end()));
166 }
167
isEqual__anonf91fc13c0311::AffineMapKeyInfo168 static bool isEqual(const KeyTy &lhs, AffineMap rhs) {
169 if (rhs == getEmptyKey() || rhs == getTombstoneKey())
170 return false;
171 return lhs == std::make_tuple(rhs.getNumDims(), rhs.getNumSymbols(),
172 rhs.getResults());
173 }
174 };
175
176 struct IntegerSetKeyInfo : DenseMapInfo<IntegerSet> {
177 // Integer sets are uniqued based on their dim/symbol counts, affine
178 // expressions appearing in the LHS of constraints, and eqFlags.
179 using KeyTy =
180 std::tuple<unsigned, unsigned, ArrayRef<AffineExpr>, ArrayRef<bool>>;
181 using DenseMapInfo<IntegerSet>::isEqual;
182
getHashValue__anonf91fc13c0311::IntegerSetKeyInfo183 static unsigned getHashValue(const IntegerSet &key) {
184 return getHashValue(KeyTy(key.getNumDims(), key.getNumSymbols(),
185 key.getConstraints(), key.getEqFlags()));
186 }
187
getHashValue__anonf91fc13c0311::IntegerSetKeyInfo188 static unsigned getHashValue(KeyTy key) {
189 return hash_combine(
190 std::get<0>(key), std::get<1>(key),
191 hash_combine_range(std::get<2>(key).begin(), std::get<2>(key).end()),
192 hash_combine_range(std::get<3>(key).begin(), std::get<3>(key).end()));
193 }
194
isEqual__anonf91fc13c0311::IntegerSetKeyInfo195 static bool isEqual(const KeyTy &lhs, IntegerSet rhs) {
196 if (rhs == getEmptyKey() || rhs == getTombstoneKey())
197 return false;
198 return lhs == std::make_tuple(rhs.getNumDims(), rhs.getNumSymbols(),
199 rhs.getConstraints(), rhs.getEqFlags());
200 }
201 };
202 } // end anonymous namespace.
203
204 //===----------------------------------------------------------------------===//
205 // MLIRContextImpl
206 //===----------------------------------------------------------------------===//
207
208 namespace mlir {
209 /// This is the implementation of the MLIRContext class, using the pImpl idiom.
210 /// This class is completely private to this file, so everything is public.
211 class MLIRContextImpl {
212 public:
213 //===--------------------------------------------------------------------===//
214 // Identifier uniquing
215 //===--------------------------------------------------------------------===//
216
217 // Identifier allocator and mutex for thread safety.
218 llvm::BumpPtrAllocator identifierAllocator;
219 llvm::sys::SmartRWMutex<true> identifierMutex;
220
221 //===--------------------------------------------------------------------===//
222 // Diagnostics
223 //===--------------------------------------------------------------------===//
224 DiagnosticEngine diagEngine;
225
226 //===--------------------------------------------------------------------===//
227 // Options
228 //===--------------------------------------------------------------------===//
229
230 /// In most cases, creating operation in unregistered dialect is not desired
231 /// and indicate a misconfiguration of the compiler. This option enables to
232 /// detect such use cases
233 bool allowUnregisteredDialects = false;
234
235 /// Enable support for multi-threading within MLIR.
236 bool threadingIsEnabled = true;
237
238 /// Track if we are currently executing in a threaded execution environment
239 /// (like the pass-manager): this is only a debugging feature to help reducing
240 /// the chances of data races one some context APIs.
241 #ifndef NDEBUG
242 std::atomic<int> multiThreadedExecutionContext{0};
243 #endif
244
245 /// If the operation should be attached to diagnostics printed via the
246 /// Operation::emit methods.
247 bool printOpOnDiagnostic = true;
248
249 /// If the current stack trace should be attached when emitting diagnostics.
250 bool printStackTraceOnDiagnostic = false;
251
252 //===--------------------------------------------------------------------===//
253 // Other
254 //===--------------------------------------------------------------------===//
255
256 /// This is a list of dialects that are created referring to this context.
257 /// The MLIRContext owns the objects.
258 DenseMap<StringRef, std::unique_ptr<Dialect>> loadedDialects;
259 DialectRegistry dialectsRegistry;
260
261 /// This is a mapping from operation name to AbstractOperation for registered
262 /// operations.
263 llvm::StringMap<AbstractOperation> registeredOperations;
264
265 /// Identifiers are uniqued by string value and use the internal string set
266 /// for storage.
267 llvm::StringSet<llvm::BumpPtrAllocator &> identifiers;
268 /// A thread local cache of identifiers to reduce lock contention.
269 ThreadLocalCache<llvm::StringMap<llvm::StringMapEntry<llvm::NoneType> *>>
270 localIdentifierCache;
271
272 /// An allocator used for AbstractAttribute and AbstractType objects.
273 llvm::BumpPtrAllocator abstractDialectSymbolAllocator;
274
275 //===--------------------------------------------------------------------===//
276 // Affine uniquing
277 //===--------------------------------------------------------------------===//
278
279 // Affine allocator and mutex for thread safety.
280 llvm::BumpPtrAllocator affineAllocator;
281 llvm::sys::SmartRWMutex<true> affineMutex;
282
283 // Affine map uniquing.
284 using AffineMapSet = DenseSet<AffineMap, AffineMapKeyInfo>;
285 AffineMapSet affineMaps;
286
287 // Integer set uniquing.
288 using IntegerSets = DenseSet<IntegerSet, IntegerSetKeyInfo>;
289 IntegerSets integerSets;
290
291 // Affine expression uniquing.
292 StorageUniquer affineUniquer;
293
294 //===--------------------------------------------------------------------===//
295 // Type uniquing
296 //===--------------------------------------------------------------------===//
297
298 DenseMap<TypeID, const AbstractType *> registeredTypes;
299 StorageUniquer typeUniquer;
300
301 /// Cached Type Instances.
302 BFloat16Type bf16Ty;
303 Float16Type f16Ty;
304 Float32Type f32Ty;
305 Float64Type f64Ty;
306 IndexType indexTy;
307 IntegerType int1Ty, int8Ty, int16Ty, int32Ty, int64Ty, int128Ty;
308 NoneType noneType;
309
310 //===--------------------------------------------------------------------===//
311 // Attribute uniquing
312 //===--------------------------------------------------------------------===//
313
314 DenseMap<TypeID, const AbstractAttribute *> registeredAttributes;
315 StorageUniquer attributeUniquer;
316
317 /// Cached Attribute Instances.
318 BoolAttr falseAttr, trueAttr;
319 UnitAttr unitAttr;
320 UnknownLoc unknownLocAttr;
321 DictionaryAttr emptyDictionaryAttr;
322
323 public:
MLIRContextImpl()324 MLIRContextImpl() : identifiers(identifierAllocator) {}
~MLIRContextImpl()325 ~MLIRContextImpl() {
326 for (auto typeMapping : registeredTypes)
327 typeMapping.second->~AbstractType();
328 for (auto attrMapping : registeredAttributes)
329 attrMapping.second->~AbstractAttribute();
330 }
331 };
332 } // end namespace mlir
333
MLIRContext()334 MLIRContext::MLIRContext() : impl(new MLIRContextImpl()) {
335 // Initialize values based on the command line flags if they were provided.
336 if (clOptions.isConstructed()) {
337 disableMultithreading(clOptions->disableThreading);
338 printOpOnDiagnostic(clOptions->printOpOnDiagnostic);
339 printStackTraceOnDiagnostic(clOptions->printStackTraceOnDiagnostic);
340 }
341
342 // Ensure the builtin dialect is always pre-loaded.
343 getOrLoadDialect<BuiltinDialect>();
344
345 // Initialize several common attributes and types to avoid the need to lock
346 // the context when accessing them.
347
348 //// Types.
349 /// Floating-point Types.
350 impl->bf16Ty = TypeUniquer::get<BFloat16Type>(this);
351 impl->f16Ty = TypeUniquer::get<Float16Type>(this);
352 impl->f32Ty = TypeUniquer::get<Float32Type>(this);
353 impl->f64Ty = TypeUniquer::get<Float64Type>(this);
354 /// Index Type.
355 impl->indexTy = TypeUniquer::get<IndexType>(this);
356 /// Integer Types.
357 impl->int1Ty = TypeUniquer::get<IntegerType>(this, 1, IntegerType::Signless);
358 impl->int8Ty = TypeUniquer::get<IntegerType>(this, 8, IntegerType::Signless);
359 impl->int16Ty =
360 TypeUniquer::get<IntegerType>(this, 16, IntegerType::Signless);
361 impl->int32Ty =
362 TypeUniquer::get<IntegerType>(this, 32, IntegerType::Signless);
363 impl->int64Ty =
364 TypeUniquer::get<IntegerType>(this, 64, IntegerType::Signless);
365 impl->int128Ty =
366 TypeUniquer::get<IntegerType>(this, 128, IntegerType::Signless);
367 /// None Type.
368 impl->noneType = TypeUniquer::get<NoneType>(this);
369
370 //// Attributes.
371 //// Note: These must be registered after the types as they may generate one
372 //// of the above types internally.
373 /// Bool Attributes.
374 impl->falseAttr = AttributeUniquer::get<IntegerAttr>(
375 this, impl->int1Ty, APInt(/*numBits=*/1, false))
376 .cast<BoolAttr>();
377 impl->trueAttr = AttributeUniquer::get<IntegerAttr>(
378 this, impl->int1Ty, APInt(/*numBits=*/1, true))
379 .cast<BoolAttr>();
380 /// Unit Attribute.
381 impl->unitAttr = AttributeUniquer::get<UnitAttr>(this);
382 /// Unknown Location Attribute.
383 impl->unknownLocAttr = AttributeUniquer::get<UnknownLoc>(this);
384 /// The empty dictionary attribute.
385 impl->emptyDictionaryAttr =
386 AttributeUniquer::get<DictionaryAttr>(this, ArrayRef<NamedAttribute>());
387
388 // Register the affine storage objects with the uniquer.
389 impl->affineUniquer
390 .registerParametricStorageType<AffineBinaryOpExprStorage>();
391 impl->affineUniquer
392 .registerParametricStorageType<AffineConstantExprStorage>();
393 impl->affineUniquer.registerParametricStorageType<AffineDimExprStorage>();
394 }
395
~MLIRContext()396 MLIRContext::~MLIRContext() {}
397
398 /// Copy the specified array of elements into memory managed by the provided
399 /// bump pointer allocator. This assumes the elements are all PODs.
400 template <typename T>
copyArrayRefInto(llvm::BumpPtrAllocator & allocator,ArrayRef<T> elements)401 static ArrayRef<T> copyArrayRefInto(llvm::BumpPtrAllocator &allocator,
402 ArrayRef<T> elements) {
403 auto result = allocator.Allocate<T>(elements.size());
404 std::uninitialized_copy(elements.begin(), elements.end(), result);
405 return ArrayRef<T>(result, elements.size());
406 }
407
408 //===----------------------------------------------------------------------===//
409 // Diagnostic Handlers
410 //===----------------------------------------------------------------------===//
411
412 /// Returns the diagnostic engine for this context.
getDiagEngine()413 DiagnosticEngine &MLIRContext::getDiagEngine() { return getImpl().diagEngine; }
414
415 //===----------------------------------------------------------------------===//
416 // Dialect and Operation Registration
417 //===----------------------------------------------------------------------===//
418
getDialectRegistry()419 DialectRegistry &MLIRContext::getDialectRegistry() {
420 return impl->dialectsRegistry;
421 }
422
423 /// Return information about all registered IR dialects.
getLoadedDialects()424 std::vector<Dialect *> MLIRContext::getLoadedDialects() {
425 std::vector<Dialect *> result;
426 result.reserve(impl->loadedDialects.size());
427 for (auto &dialect : impl->loadedDialects)
428 result.push_back(dialect.second.get());
429 llvm::array_pod_sort(result.begin(), result.end(),
430 [](Dialect *const *lhs, Dialect *const *rhs) -> int {
431 return (*lhs)->getNamespace() < (*rhs)->getNamespace();
432 });
433 return result;
434 }
getAvailableDialects()435 std::vector<StringRef> MLIRContext::getAvailableDialects() {
436 std::vector<StringRef> result;
437 for (auto &dialect : impl->dialectsRegistry)
438 result.push_back(dialect.first);
439 return result;
440 }
441
442 /// Get a registered IR dialect with the given namespace. If none is found,
443 /// then return nullptr.
getLoadedDialect(StringRef name)444 Dialect *MLIRContext::getLoadedDialect(StringRef name) {
445 // Dialects are sorted by name, so we can use binary search for lookup.
446 auto it = impl->loadedDialects.find(name);
447 return (it != impl->loadedDialects.end()) ? it->second.get() : nullptr;
448 }
449
getOrLoadDialect(StringRef name)450 Dialect *MLIRContext::getOrLoadDialect(StringRef name) {
451 Dialect *dialect = getLoadedDialect(name);
452 if (dialect)
453 return dialect;
454 return impl->dialectsRegistry.loadByName(name, this);
455 }
456
457 /// Get a dialect for the provided namespace and TypeID: abort the program if a
458 /// dialect exist for this namespace with different TypeID. Returns a pointer to
459 /// the dialect owned by the context.
460 Dialect *
getOrLoadDialect(StringRef dialectNamespace,TypeID dialectID,function_ref<std::unique_ptr<Dialect> ()> ctor)461 MLIRContext::getOrLoadDialect(StringRef dialectNamespace, TypeID dialectID,
462 function_ref<std::unique_ptr<Dialect>()> ctor) {
463 auto &impl = getImpl();
464 // Get the correct insertion position sorted by namespace.
465 std::unique_ptr<Dialect> &dialect = impl.loadedDialects[dialectNamespace];
466
467 if (!dialect) {
468 LLVM_DEBUG(llvm::dbgs()
469 << "Load new dialect in Context " << dialectNamespace << "\n");
470 #ifndef NDEBUG
471 if (impl.multiThreadedExecutionContext != 0)
472 llvm::report_fatal_error(
473 "Loading a dialect (" + dialectNamespace +
474 ") while in a multi-threaded execution context (maybe "
475 "the PassManager): this can indicate a "
476 "missing `dependentDialects` in a pass for example.");
477 #endif
478 dialect = ctor();
479 assert(dialect && "dialect ctor failed");
480 return dialect.get();
481 }
482
483 // Abort if dialect with namespace has already been registered.
484 if (dialect->getTypeID() != dialectID)
485 llvm::report_fatal_error("a dialect with namespace '" + dialectNamespace +
486 "' has already been registered");
487
488 return dialect.get();
489 }
490
allowsUnregisteredDialects()491 bool MLIRContext::allowsUnregisteredDialects() {
492 return impl->allowUnregisteredDialects;
493 }
494
allowUnregisteredDialects(bool allowing)495 void MLIRContext::allowUnregisteredDialects(bool allowing) {
496 impl->allowUnregisteredDialects = allowing;
497 }
498
499 /// Return true if multi-threading is disabled by the context.
isMultithreadingEnabled()500 bool MLIRContext::isMultithreadingEnabled() {
501 return impl->threadingIsEnabled && llvm::llvm_is_multithreaded();
502 }
503
504 /// Set the flag specifying if multi-threading is disabled by the context.
disableMultithreading(bool disable)505 void MLIRContext::disableMultithreading(bool disable) {
506 impl->threadingIsEnabled = !disable;
507
508 // Update the threading mode for each of the uniquers.
509 impl->affineUniquer.disableMultithreading(disable);
510 impl->attributeUniquer.disableMultithreading(disable);
511 impl->typeUniquer.disableMultithreading(disable);
512 }
513
enterMultiThreadedExecution()514 void MLIRContext::enterMultiThreadedExecution() {
515 #ifndef NDEBUG
516 ++impl->multiThreadedExecutionContext;
517 #endif
518 }
exitMultiThreadedExecution()519 void MLIRContext::exitMultiThreadedExecution() {
520 #ifndef NDEBUG
521 --impl->multiThreadedExecutionContext;
522 #endif
523 }
524
525 /// Return true if we should attach the operation to diagnostics emitted via
526 /// Operation::emit.
shouldPrintOpOnDiagnostic()527 bool MLIRContext::shouldPrintOpOnDiagnostic() {
528 return impl->printOpOnDiagnostic;
529 }
530
531 /// Set the flag specifying if we should attach the operation to diagnostics
532 /// emitted via Operation::emit.
printOpOnDiagnostic(bool enable)533 void MLIRContext::printOpOnDiagnostic(bool enable) {
534 impl->printOpOnDiagnostic = enable;
535 }
536
537 /// Return true if we should attach the current stacktrace to diagnostics when
538 /// emitted.
shouldPrintStackTraceOnDiagnostic()539 bool MLIRContext::shouldPrintStackTraceOnDiagnostic() {
540 return impl->printStackTraceOnDiagnostic;
541 }
542
543 /// Set the flag specifying if we should attach the current stacktrace when
544 /// emitting diagnostics.
printStackTraceOnDiagnostic(bool enable)545 void MLIRContext::printStackTraceOnDiagnostic(bool enable) {
546 impl->printStackTraceOnDiagnostic = enable;
547 }
548
549 /// Return information about all registered operations. This isn't very
550 /// efficient, typically you should ask the operations about their properties
551 /// directly.
getRegisteredOperations()552 std::vector<AbstractOperation *> MLIRContext::getRegisteredOperations() {
553 // We just have the operations in a non-deterministic hash table order. Dump
554 // into a temporary array, then sort it by operation name to get a stable
555 // ordering.
556 llvm::StringMap<AbstractOperation> ®isteredOps =
557 impl->registeredOperations;
558
559 std::vector<AbstractOperation *> result;
560 result.reserve(registeredOps.size());
561 for (auto &elt : registeredOps)
562 result.push_back(&elt.second);
563 llvm::array_pod_sort(
564 result.begin(), result.end(),
565 [](AbstractOperation *const *lhs, AbstractOperation *const *rhs) {
566 return (*lhs)->name.compare((*rhs)->name);
567 });
568
569 return result;
570 }
571
isOperationRegistered(StringRef name)572 bool MLIRContext::isOperationRegistered(StringRef name) {
573 return impl->registeredOperations.count(name);
574 }
575
addType(TypeID typeID,AbstractType && typeInfo)576 void Dialect::addType(TypeID typeID, AbstractType &&typeInfo) {
577 auto &impl = context->getImpl();
578 assert(impl.multiThreadedExecutionContext == 0 &&
579 "Registering a new type kind while in a multi-threaded execution "
580 "context");
581 auto *newInfo =
582 new (impl.abstractDialectSymbolAllocator.Allocate<AbstractType>())
583 AbstractType(std::move(typeInfo));
584 if (!impl.registeredTypes.insert({typeID, newInfo}).second)
585 llvm::report_fatal_error("Dialect Type already registered.");
586 }
587
addAttribute(TypeID typeID,AbstractAttribute && attrInfo)588 void Dialect::addAttribute(TypeID typeID, AbstractAttribute &&attrInfo) {
589 auto &impl = context->getImpl();
590 assert(impl.multiThreadedExecutionContext == 0 &&
591 "Registering a new attribute kind while in a multi-threaded execution "
592 "context");
593 auto *newInfo =
594 new (impl.abstractDialectSymbolAllocator.Allocate<AbstractAttribute>())
595 AbstractAttribute(std::move(attrInfo));
596 if (!impl.registeredAttributes.insert({typeID, newInfo}).second)
597 llvm::report_fatal_error("Dialect Attribute already registered.");
598 }
599
600 //===----------------------------------------------------------------------===//
601 // AbstractAttribute
602 //===----------------------------------------------------------------------===//
603
604 /// Get the dialect that registered the attribute with the provided typeid.
lookup(TypeID typeID,MLIRContext * context)605 const AbstractAttribute &AbstractAttribute::lookup(TypeID typeID,
606 MLIRContext *context) {
607 auto &impl = context->getImpl();
608 auto it = impl.registeredAttributes.find(typeID);
609 if (it == impl.registeredAttributes.end())
610 llvm::report_fatal_error("Trying to create an Attribute that was not "
611 "registered in this MLIRContext.");
612 return *it->second;
613 }
614
615 //===----------------------------------------------------------------------===//
616 // AbstractOperation
617 //===----------------------------------------------------------------------===//
618
parseAssembly(OpAsmParser & parser,OperationState & result) const619 ParseResult AbstractOperation::parseAssembly(OpAsmParser &parser,
620 OperationState &result) const {
621 return parseAssemblyFn(parser, result);
622 }
623
624 /// Look up the specified operation in the operation set and return a pointer
625 /// to it if present. Otherwise, return a null pointer.
lookup(StringRef opName,MLIRContext * context)626 const AbstractOperation *AbstractOperation::lookup(StringRef opName,
627 MLIRContext *context) {
628 auto &impl = context->getImpl();
629 auto it = impl.registeredOperations.find(opName);
630 if (it != impl.registeredOperations.end())
631 return &it->second;
632 return nullptr;
633 }
634
insert(StringRef name,Dialect & dialect,OperationProperties opProperties,TypeID typeID,ParseAssemblyFn parseAssembly,PrintAssemblyFn printAssembly,VerifyInvariantsFn verifyInvariants,FoldHookFn foldHook,GetCanonicalizationPatternsFn getCanonicalizationPatterns,detail::InterfaceMap && interfaceMap,HasTraitFn hasTrait)635 void AbstractOperation::insert(
636 StringRef name, Dialect &dialect, OperationProperties opProperties,
637 TypeID typeID, ParseAssemblyFn parseAssembly, PrintAssemblyFn printAssembly,
638 VerifyInvariantsFn verifyInvariants, FoldHookFn foldHook,
639 GetCanonicalizationPatternsFn getCanonicalizationPatterns,
640 detail::InterfaceMap &&interfaceMap, HasTraitFn hasTrait) {
641 AbstractOperation opInfo(name, dialect, opProperties, typeID, parseAssembly,
642 printAssembly, verifyInvariants, foldHook,
643 getCanonicalizationPatterns, std::move(interfaceMap),
644 hasTrait);
645
646 auto &impl = dialect.getContext()->getImpl();
647 assert(impl.multiThreadedExecutionContext == 0 &&
648 "Registering a new operation kind while in a multi-threaded execution "
649 "context");
650 if (!impl.registeredOperations.insert({name, std::move(opInfo)}).second) {
651 llvm::errs() << "error: operation named '" << name
652 << "' is already registered.\n";
653 abort();
654 }
655 }
656
AbstractOperation(StringRef name,Dialect & dialect,OperationProperties opProperties,TypeID typeID,ParseAssemblyFn parseAssembly,PrintAssemblyFn printAssembly,VerifyInvariantsFn verifyInvariants,FoldHookFn foldHook,GetCanonicalizationPatternsFn getCanonicalizationPatterns,detail::InterfaceMap && interfaceMap,HasTraitFn hasTrait)657 AbstractOperation::AbstractOperation(
658 StringRef name, Dialect &dialect, OperationProperties opProperties,
659 TypeID typeID, ParseAssemblyFn parseAssembly, PrintAssemblyFn printAssembly,
660 VerifyInvariantsFn verifyInvariants, FoldHookFn foldHook,
661 GetCanonicalizationPatternsFn getCanonicalizationPatterns,
662 detail::InterfaceMap &&interfaceMap, HasTraitFn hasTrait)
663 : name(Identifier::get(name, dialect.getContext())), dialect(dialect),
664 typeID(typeID), opProperties(opProperties),
665 interfaceMap(std::move(interfaceMap)), foldHookFn(foldHook),
666 getCanonicalizationPatternsFn(getCanonicalizationPatterns),
667 hasTraitFn(hasTrait), parseAssemblyFn(parseAssembly),
668 printAssemblyFn(printAssembly), verifyInvariantsFn(verifyInvariants) {}
669
670 //===----------------------------------------------------------------------===//
671 // AbstractType
672 //===----------------------------------------------------------------------===//
673
lookup(TypeID typeID,MLIRContext * context)674 const AbstractType &AbstractType::lookup(TypeID typeID, MLIRContext *context) {
675 auto &impl = context->getImpl();
676 auto it = impl.registeredTypes.find(typeID);
677 if (it == impl.registeredTypes.end())
678 llvm::report_fatal_error(
679 "Trying to create a Type that was not registered in this MLIRContext.");
680 return *it->second;
681 }
682
683 //===----------------------------------------------------------------------===//
684 // Identifier uniquing
685 //===----------------------------------------------------------------------===//
686
687 /// Return an identifier for the specified string.
get(StringRef str,MLIRContext * context)688 Identifier Identifier::get(StringRef str, MLIRContext *context) {
689 // Check invariants after seeing if we already have something in the
690 // identifier table - if we already had it in the table, then it already
691 // passed invariant checks.
692 assert(!str.empty() && "Cannot create an empty identifier");
693 assert(str.find('\0') == StringRef::npos &&
694 "Cannot create an identifier with a nul character");
695
696 auto &impl = context->getImpl();
697 if (!context->isMultithreadingEnabled())
698 return Identifier(&*impl.identifiers.insert(str).first);
699
700 // Check for an existing instance in the local cache.
701 auto *&localEntry = (*impl.localIdentifierCache)[str];
702 if (localEntry)
703 return Identifier(localEntry);
704
705 // Check for an existing identifier in read-only mode.
706 {
707 llvm::sys::SmartScopedReader<true> contextLock(impl.identifierMutex);
708 auto it = impl.identifiers.find(str);
709 if (it != impl.identifiers.end()) {
710 localEntry = &*it;
711 return Identifier(localEntry);
712 }
713 }
714
715 // Acquire a writer-lock so that we can safely create the new instance.
716 llvm::sys::SmartScopedWriter<true> contextLock(impl.identifierMutex);
717 auto it = impl.identifiers.insert(str).first;
718 localEntry = &*it;
719 return Identifier(localEntry);
720 }
721
722 //===----------------------------------------------------------------------===//
723 // Type uniquing
724 //===----------------------------------------------------------------------===//
725
726 /// Returns the storage uniquer used for constructing type storage instances.
727 /// This should not be used directly.
getTypeUniquer()728 StorageUniquer &MLIRContext::getTypeUniquer() { return getImpl().typeUniquer; }
729
get(MLIRContext * context)730 BFloat16Type BFloat16Type::get(MLIRContext *context) {
731 return context->getImpl().bf16Ty;
732 }
get(MLIRContext * context)733 Float16Type Float16Type::get(MLIRContext *context) {
734 return context->getImpl().f16Ty;
735 }
get(MLIRContext * context)736 Float32Type Float32Type::get(MLIRContext *context) {
737 return context->getImpl().f32Ty;
738 }
get(MLIRContext * context)739 Float64Type Float64Type::get(MLIRContext *context) {
740 return context->getImpl().f64Ty;
741 }
742
743 /// Get an instance of the IndexType.
get(MLIRContext * context)744 IndexType IndexType::get(MLIRContext *context) {
745 return context->getImpl().indexTy;
746 }
747
748 /// Return an existing integer type instance if one is cached within the
749 /// context.
750 static IntegerType
getCachedIntegerType(unsigned width,IntegerType::SignednessSemantics signedness,MLIRContext * context)751 getCachedIntegerType(unsigned width,
752 IntegerType::SignednessSemantics signedness,
753 MLIRContext *context) {
754 if (signedness != IntegerType::Signless)
755 return IntegerType();
756
757 switch (width) {
758 case 1:
759 return context->getImpl().int1Ty;
760 case 8:
761 return context->getImpl().int8Ty;
762 case 16:
763 return context->getImpl().int16Ty;
764 case 32:
765 return context->getImpl().int32Ty;
766 case 64:
767 return context->getImpl().int64Ty;
768 case 128:
769 return context->getImpl().int128Ty;
770 default:
771 return IntegerType();
772 }
773 }
774
get(unsigned width,MLIRContext * context)775 IntegerType IntegerType::get(unsigned width, MLIRContext *context) {
776 return get(width, IntegerType::Signless, context);
777 }
778
get(unsigned width,IntegerType::SignednessSemantics signedness,MLIRContext * context)779 IntegerType IntegerType::get(unsigned width,
780 IntegerType::SignednessSemantics signedness,
781 MLIRContext *context) {
782 if (auto cached = getCachedIntegerType(width, signedness, context))
783 return cached;
784 return Base::get(context, width, signedness);
785 }
786
getChecked(unsigned width,Location location)787 IntegerType IntegerType::getChecked(unsigned width, Location location) {
788 return getChecked(width, IntegerType::Signless, location);
789 }
790
getChecked(unsigned width,SignednessSemantics signedness,Location location)791 IntegerType IntegerType::getChecked(unsigned width,
792 SignednessSemantics signedness,
793 Location location) {
794 if (auto cached =
795 getCachedIntegerType(width, signedness, location->getContext()))
796 return cached;
797 return Base::getChecked(location, width, signedness);
798 }
799
800 /// Get an instance of the NoneType.
get(MLIRContext * context)801 NoneType NoneType::get(MLIRContext *context) {
802 if (NoneType cachedInst = context->getImpl().noneType)
803 return cachedInst;
804 // Note: May happen when initializing the singleton attributes of the builtin
805 // dialect.
806 return Base::get(context);
807 }
808
809 //===----------------------------------------------------------------------===//
810 // Attribute uniquing
811 //===----------------------------------------------------------------------===//
812
813 /// Returns the storage uniquer used for constructing attribute storage
814 /// instances. This should not be used directly.
getAttributeUniquer()815 StorageUniquer &MLIRContext::getAttributeUniquer() {
816 return getImpl().attributeUniquer;
817 }
818
819 /// Initialize the given attribute storage instance.
initializeAttributeStorage(AttributeStorage * storage,MLIRContext * ctx,TypeID attrID)820 void AttributeUniquer::initializeAttributeStorage(AttributeStorage *storage,
821 MLIRContext *ctx,
822 TypeID attrID) {
823 storage->initialize(AbstractAttribute::lookup(attrID, ctx));
824
825 // If the attribute did not provide a type, then default to NoneType.
826 if (!storage->getType())
827 storage->setType(NoneType::get(ctx));
828 }
829
get(bool value,MLIRContext * context)830 BoolAttr BoolAttr::get(bool value, MLIRContext *context) {
831 return value ? context->getImpl().trueAttr : context->getImpl().falseAttr;
832 }
833
get(MLIRContext * context)834 UnitAttr UnitAttr::get(MLIRContext *context) {
835 return context->getImpl().unitAttr;
836 }
837
get(MLIRContext * context)838 Location UnknownLoc::get(MLIRContext *context) {
839 return context->getImpl().unknownLocAttr;
840 }
841
842 /// Return empty dictionary.
getEmpty(MLIRContext * context)843 DictionaryAttr DictionaryAttr::getEmpty(MLIRContext *context) {
844 return context->getImpl().emptyDictionaryAttr;
845 }
846
847 //===----------------------------------------------------------------------===//
848 // AffineMap uniquing
849 //===----------------------------------------------------------------------===//
850
getAffineUniquer()851 StorageUniquer &MLIRContext::getAffineUniquer() {
852 return getImpl().affineUniquer;
853 }
854
getImpl(unsigned dimCount,unsigned symbolCount,ArrayRef<AffineExpr> results,MLIRContext * context)855 AffineMap AffineMap::getImpl(unsigned dimCount, unsigned symbolCount,
856 ArrayRef<AffineExpr> results,
857 MLIRContext *context) {
858 auto &impl = context->getImpl();
859 auto key = std::make_tuple(dimCount, symbolCount, results);
860
861 // Safely get or create an AffineMap instance.
862 return safeGetOrCreate(
863 impl.affineMaps, key, impl.affineMutex, impl.threadingIsEnabled, [&] {
864 auto *res = impl.affineAllocator.Allocate<detail::AffineMapStorage>();
865
866 // Copy the results into the bump pointer.
867 results = copyArrayRefInto(impl.affineAllocator, results);
868
869 // Initialize the memory using placement new.
870 new (res)
871 detail::AffineMapStorage{dimCount, symbolCount, results, context};
872 return AffineMap(res);
873 });
874 }
875
get(MLIRContext * context)876 AffineMap AffineMap::get(MLIRContext *context) {
877 return getImpl(/*dimCount=*/0, /*symbolCount=*/0, /*results=*/{}, context);
878 }
879
get(unsigned dimCount,unsigned symbolCount,MLIRContext * context)880 AffineMap AffineMap::get(unsigned dimCount, unsigned symbolCount,
881 MLIRContext *context) {
882 return getImpl(dimCount, symbolCount, /*results=*/{}, context);
883 }
884
get(unsigned dimCount,unsigned symbolCount,AffineExpr result)885 AffineMap AffineMap::get(unsigned dimCount, unsigned symbolCount,
886 AffineExpr result) {
887 return getImpl(dimCount, symbolCount, {result}, result.getContext());
888 }
889
get(unsigned dimCount,unsigned symbolCount,ArrayRef<AffineExpr> results,MLIRContext * context)890 AffineMap AffineMap::get(unsigned dimCount, unsigned symbolCount,
891 ArrayRef<AffineExpr> results, MLIRContext *context) {
892 return getImpl(dimCount, symbolCount, results, context);
893 }
894
895 //===----------------------------------------------------------------------===//
896 // Integer Sets: these are allocated into the bump pointer, and are immutable.
897 // Unlike AffineMap's, these are uniqued only if they are small.
898 //===----------------------------------------------------------------------===//
899
get(unsigned dimCount,unsigned symbolCount,ArrayRef<AffineExpr> constraints,ArrayRef<bool> eqFlags)900 IntegerSet IntegerSet::get(unsigned dimCount, unsigned symbolCount,
901 ArrayRef<AffineExpr> constraints,
902 ArrayRef<bool> eqFlags) {
903 // The number of constraints can't be zero.
904 assert(!constraints.empty());
905 assert(constraints.size() == eqFlags.size());
906
907 auto &impl = constraints[0].getContext()->getImpl();
908
909 // A utility function to construct a new IntegerSetStorage instance.
910 auto constructorFn = [&] {
911 auto *res = impl.affineAllocator.Allocate<detail::IntegerSetStorage>();
912
913 // Copy the results and equality flags into the bump pointer.
914 constraints = copyArrayRefInto(impl.affineAllocator, constraints);
915 eqFlags = copyArrayRefInto(impl.affineAllocator, eqFlags);
916
917 // Initialize the memory using placement new.
918 new (res)
919 detail::IntegerSetStorage{dimCount, symbolCount, constraints, eqFlags};
920 return IntegerSet(res);
921 };
922
923 // If this instance is uniqued, then we handle it separately so that multiple
924 // threads may simultaneously access existing instances.
925 if (constraints.size() < IntegerSet::kUniquingThreshold) {
926 auto key = std::make_tuple(dimCount, symbolCount, constraints, eqFlags);
927 return safeGetOrCreate(impl.integerSets, key, impl.affineMutex,
928 impl.threadingIsEnabled, constructorFn);
929 }
930
931 // Otherwise, acquire a writer-lock so that we can safely create the new
932 // instance.
933 ScopedWriterLock affineLock(impl.affineMutex, impl.threadingIsEnabled);
934 return constructorFn();
935 }
936
937 //===----------------------------------------------------------------------===//
938 // StorageUniquerSupport
939 //===----------------------------------------------------------------------===//
940
941 /// Utility method to generate a default location for use when checking the
942 /// construction invariants of a storage object. This is defined out-of-line to
943 /// avoid the need to include Location.h.
944 const AttributeStorage *
generateUnknownStorageLocation(MLIRContext * ctx)945 mlir::detail::generateUnknownStorageLocation(MLIRContext *ctx) {
946 return reinterpret_cast<const AttributeStorage *>(
947 ctx->getImpl().unknownLocAttr.getAsOpaquePointer());
948 }
949