Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
14 | pmbaty | 1 | //===- MemorySSA.h - Build Memory SSA ---------------------------*- C++ -*-===// |
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 | /// \file |
||
10 | /// This file exposes an interface to building/using memory SSA to |
||
11 | /// walk memory instructions using a use/def graph. |
||
12 | /// |
||
13 | /// Memory SSA class builds an SSA form that links together memory access |
||
14 | /// instructions such as loads, stores, atomics, and calls. Additionally, it |
||
15 | /// does a trivial form of "heap versioning" Every time the memory state changes |
||
16 | /// in the program, we generate a new heap version. It generates |
||
17 | /// MemoryDef/Uses/Phis that are overlayed on top of the existing instructions. |
||
18 | /// |
||
19 | /// As a trivial example, |
||
20 | /// define i32 @main() #0 { |
||
21 | /// entry: |
||
22 | /// %call = call noalias i8* @_Znwm(i64 4) #2 |
||
23 | /// %0 = bitcast i8* %call to i32* |
||
24 | /// %call1 = call noalias i8* @_Znwm(i64 4) #2 |
||
25 | /// %1 = bitcast i8* %call1 to i32* |
||
26 | /// store i32 5, i32* %0, align 4 |
||
27 | /// store i32 7, i32* %1, align 4 |
||
28 | /// %2 = load i32* %0, align 4 |
||
29 | /// %3 = load i32* %1, align 4 |
||
30 | /// %add = add nsw i32 %2, %3 |
||
31 | /// ret i32 %add |
||
32 | /// } |
||
33 | /// |
||
34 | /// Will become |
||
35 | /// define i32 @main() #0 { |
||
36 | /// entry: |
||
37 | /// ; 1 = MemoryDef(0) |
||
38 | /// %call = call noalias i8* @_Znwm(i64 4) #3 |
||
39 | /// %2 = bitcast i8* %call to i32* |
||
40 | /// ; 2 = MemoryDef(1) |
||
41 | /// %call1 = call noalias i8* @_Znwm(i64 4) #3 |
||
42 | /// %4 = bitcast i8* %call1 to i32* |
||
43 | /// ; 3 = MemoryDef(2) |
||
44 | /// store i32 5, i32* %2, align 4 |
||
45 | /// ; 4 = MemoryDef(3) |
||
46 | /// store i32 7, i32* %4, align 4 |
||
47 | /// ; MemoryUse(3) |
||
48 | /// %7 = load i32* %2, align 4 |
||
49 | /// ; MemoryUse(4) |
||
50 | /// %8 = load i32* %4, align 4 |
||
51 | /// %add = add nsw i32 %7, %8 |
||
52 | /// ret i32 %add |
||
53 | /// } |
||
54 | /// |
||
55 | /// Given this form, all the stores that could ever effect the load at %8 can be |
||
56 | /// gotten by using the MemoryUse associated with it, and walking from use to |
||
57 | /// def until you hit the top of the function. |
||
58 | /// |
||
59 | /// Each def also has a list of users associated with it, so you can walk from |
||
60 | /// both def to users, and users to defs. Note that we disambiguate MemoryUses, |
||
61 | /// but not the RHS of MemoryDefs. You can see this above at %7, which would |
||
62 | /// otherwise be a MemoryUse(4). Being disambiguated means that for a given |
||
63 | /// store, all the MemoryUses on its use lists are may-aliases of that store |
||
64 | /// (but the MemoryDefs on its use list may not be). |
||
65 | /// |
||
66 | /// MemoryDefs are not disambiguated because it would require multiple reaching |
||
67 | /// definitions, which would require multiple phis, and multiple memoryaccesses |
||
68 | /// per instruction. |
||
69 | /// |
||
70 | /// In addition to the def/use graph described above, MemoryDefs also contain |
||
71 | /// an "optimized" definition use. The "optimized" use points to some def |
||
72 | /// reachable through the memory def chain. The optimized def *may* (but is |
||
73 | /// not required to) alias the original MemoryDef, but no def *closer* to the |
||
74 | /// source def may alias it. As the name implies, the purpose of the optimized |
||
75 | /// use is to allow caching of clobber searches for memory defs. The optimized |
||
76 | /// def may be nullptr, in which case clients must walk the defining access |
||
77 | /// chain. |
||
78 | /// |
||
79 | /// When iterating the uses of a MemoryDef, both defining uses and optimized |
||
80 | /// uses will be encountered. If only one type is needed, the client must |
||
81 | /// filter the use walk. |
||
82 | // |
||
83 | //===----------------------------------------------------------------------===// |
||
84 | |||
85 | #ifndef LLVM_ANALYSIS_MEMORYSSA_H |
||
86 | #define LLVM_ANALYSIS_MEMORYSSA_H |
||
87 | |||
88 | #include "llvm/ADT/DenseMap.h" |
||
89 | #include "llvm/ADT/SmallPtrSet.h" |
||
90 | #include "llvm/ADT/SmallVector.h" |
||
91 | #include "llvm/ADT/ilist_node.h" |
||
92 | #include "llvm/ADT/iterator_range.h" |
||
93 | #include "llvm/Analysis/AliasAnalysis.h" |
||
94 | #include "llvm/Analysis/MemoryLocation.h" |
||
95 | #include "llvm/Analysis/PHITransAddr.h" |
||
96 | #include "llvm/IR/DerivedUser.h" |
||
97 | #include "llvm/IR/Dominators.h" |
||
98 | #include "llvm/IR/Type.h" |
||
99 | #include "llvm/IR/User.h" |
||
100 | #include "llvm/Pass.h" |
||
101 | #include <algorithm> |
||
102 | #include <cassert> |
||
103 | #include <cstddef> |
||
104 | #include <iterator> |
||
105 | #include <memory> |
||
106 | #include <utility> |
||
107 | |||
108 | namespace llvm { |
||
109 | |||
110 | template <class GraphType> struct GraphTraits; |
||
111 | class BasicBlock; |
||
112 | class Function; |
||
113 | class Instruction; |
||
114 | class LLVMContext; |
||
115 | class MemoryAccess; |
||
116 | class MemorySSAWalker; |
||
117 | class Module; |
||
118 | class Use; |
||
119 | class Value; |
||
120 | class raw_ostream; |
||
121 | |||
122 | namespace MSSAHelpers { |
||
123 | |||
124 | struct AllAccessTag {}; |
||
125 | struct DefsOnlyTag {}; |
||
126 | |||
127 | } // end namespace MSSAHelpers |
||
128 | |||
129 | enum : unsigned { |
||
130 | // Used to signify what the default invalid ID is for MemoryAccess's |
||
131 | // getID() |
||
132 | INVALID_MEMORYACCESS_ID = -1U |
||
133 | }; |
||
134 | |||
135 | template <class T> class memoryaccess_def_iterator_base; |
||
136 | using memoryaccess_def_iterator = memoryaccess_def_iterator_base<MemoryAccess>; |
||
137 | using const_memoryaccess_def_iterator = |
||
138 | memoryaccess_def_iterator_base<const MemoryAccess>; |
||
139 | |||
140 | // The base for all memory accesses. All memory accesses in a block are |
||
141 | // linked together using an intrusive list. |
||
142 | class MemoryAccess |
||
143 | : public DerivedUser, |
||
144 | public ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>, |
||
145 | public ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>> { |
||
146 | public: |
||
147 | using AllAccessType = |
||
148 | ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>; |
||
149 | using DefsOnlyType = |
||
150 | ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>>; |
||
151 | |||
152 | MemoryAccess(const MemoryAccess &) = delete; |
||
153 | MemoryAccess &operator=(const MemoryAccess &) = delete; |
||
154 | |||
155 | void *operator new(size_t) = delete; |
||
156 | |||
157 | // Methods for support type inquiry through isa, cast, and |
||
158 | // dyn_cast |
||
159 | static bool classof(const Value *V) { |
||
160 | unsigned ID = V->getValueID(); |
||
161 | return ID == MemoryUseVal || ID == MemoryPhiVal || ID == MemoryDefVal; |
||
162 | } |
||
163 | |||
164 | BasicBlock *getBlock() const { return Block; } |
||
165 | |||
166 | void print(raw_ostream &OS) const; |
||
167 | void dump() const; |
||
168 | |||
169 | /// The user iterators for a memory access |
||
170 | using iterator = user_iterator; |
||
171 | using const_iterator = const_user_iterator; |
||
172 | |||
173 | /// This iterator walks over all of the defs in a given |
||
174 | /// MemoryAccess. For MemoryPhi nodes, this walks arguments. For |
||
175 | /// MemoryUse/MemoryDef, this walks the defining access. |
||
176 | memoryaccess_def_iterator defs_begin(); |
||
177 | const_memoryaccess_def_iterator defs_begin() const; |
||
178 | memoryaccess_def_iterator defs_end(); |
||
179 | const_memoryaccess_def_iterator defs_end() const; |
||
180 | |||
181 | /// Get the iterators for the all access list and the defs only list |
||
182 | /// We default to the all access list. |
||
183 | AllAccessType::self_iterator getIterator() { |
||
184 | return this->AllAccessType::getIterator(); |
||
185 | } |
||
186 | AllAccessType::const_self_iterator getIterator() const { |
||
187 | return this->AllAccessType::getIterator(); |
||
188 | } |
||
189 | AllAccessType::reverse_self_iterator getReverseIterator() { |
||
190 | return this->AllAccessType::getReverseIterator(); |
||
191 | } |
||
192 | AllAccessType::const_reverse_self_iterator getReverseIterator() const { |
||
193 | return this->AllAccessType::getReverseIterator(); |
||
194 | } |
||
195 | DefsOnlyType::self_iterator getDefsIterator() { |
||
196 | return this->DefsOnlyType::getIterator(); |
||
197 | } |
||
198 | DefsOnlyType::const_self_iterator getDefsIterator() const { |
||
199 | return this->DefsOnlyType::getIterator(); |
||
200 | } |
||
201 | DefsOnlyType::reverse_self_iterator getReverseDefsIterator() { |
||
202 | return this->DefsOnlyType::getReverseIterator(); |
||
203 | } |
||
204 | DefsOnlyType::const_reverse_self_iterator getReverseDefsIterator() const { |
||
205 | return this->DefsOnlyType::getReverseIterator(); |
||
206 | } |
||
207 | |||
208 | protected: |
||
209 | friend class MemoryDef; |
||
210 | friend class MemoryPhi; |
||
211 | friend class MemorySSA; |
||
212 | friend class MemoryUse; |
||
213 | friend class MemoryUseOrDef; |
||
214 | |||
215 | /// Used by MemorySSA to change the block of a MemoryAccess when it is |
||
216 | /// moved. |
||
217 | void setBlock(BasicBlock *BB) { Block = BB; } |
||
218 | |||
219 | /// Used for debugging and tracking things about MemoryAccesses. |
||
220 | /// Guaranteed unique among MemoryAccesses, no guarantees otherwise. |
||
221 | inline unsigned getID() const; |
||
222 | |||
223 | MemoryAccess(LLVMContext &C, unsigned Vty, DeleteValueTy DeleteValue, |
||
224 | BasicBlock *BB, unsigned NumOperands) |
||
225 | : DerivedUser(Type::getVoidTy(C), Vty, nullptr, NumOperands, DeleteValue), |
||
226 | Block(BB) {} |
||
227 | |||
228 | // Use deleteValue() to delete a generic MemoryAccess. |
||
229 | ~MemoryAccess() = default; |
||
230 | |||
231 | private: |
||
232 | BasicBlock *Block; |
||
233 | }; |
||
234 | |||
235 | template <> |
||
236 | struct ilist_alloc_traits<MemoryAccess> { |
||
237 | static void deleteNode(MemoryAccess *MA) { MA->deleteValue(); } |
||
238 | }; |
||
239 | |||
240 | inline raw_ostream &operator<<(raw_ostream &OS, const MemoryAccess &MA) { |
||
241 | MA.print(OS); |
||
242 | return OS; |
||
243 | } |
||
244 | |||
245 | /// Class that has the common methods + fields of memory uses/defs. It's |
||
246 | /// a little awkward to have, but there are many cases where we want either a |
||
247 | /// use or def, and there are many cases where uses are needed (defs aren't |
||
248 | /// acceptable), and vice-versa. |
||
249 | /// |
||
250 | /// This class should never be instantiated directly; make a MemoryUse or |
||
251 | /// MemoryDef instead. |
||
252 | class MemoryUseOrDef : public MemoryAccess { |
||
253 | public: |
||
254 | void *operator new(size_t) = delete; |
||
255 | |||
256 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); |
||
257 | |||
258 | /// Get the instruction that this MemoryUse represents. |
||
259 | Instruction *getMemoryInst() const { return MemoryInstruction; } |
||
260 | |||
261 | /// Get the access that produces the memory state used by this Use. |
||
262 | MemoryAccess *getDefiningAccess() const { return getOperand(0); } |
||
263 | |||
264 | static bool classof(const Value *MA) { |
||
265 | return MA->getValueID() == MemoryUseVal || MA->getValueID() == MemoryDefVal; |
||
266 | } |
||
267 | |||
268 | /// Do we have an optimized use? |
||
269 | inline bool isOptimized() const; |
||
270 | /// Return the MemoryAccess associated with the optimized use, or nullptr. |
||
271 | inline MemoryAccess *getOptimized() const; |
||
272 | /// Sets the optimized use for a MemoryDef. |
||
273 | inline void setOptimized(MemoryAccess *); |
||
274 | |||
275 | /// Reset the ID of what this MemoryUse was optimized to, causing it to |
||
276 | /// be rewalked by the walker if necessary. |
||
277 | /// This really should only be called by tests. |
||
278 | inline void resetOptimized(); |
||
279 | |||
280 | protected: |
||
281 | friend class MemorySSA; |
||
282 | friend class MemorySSAUpdater; |
||
283 | |||
284 | MemoryUseOrDef(LLVMContext &C, MemoryAccess *DMA, unsigned Vty, |
||
285 | DeleteValueTy DeleteValue, Instruction *MI, BasicBlock *BB, |
||
286 | unsigned NumOperands) |
||
287 | : MemoryAccess(C, Vty, DeleteValue, BB, NumOperands), |
||
288 | MemoryInstruction(MI) { |
||
289 | setDefiningAccess(DMA); |
||
290 | } |
||
291 | |||
292 | // Use deleteValue() to delete a generic MemoryUseOrDef. |
||
293 | ~MemoryUseOrDef() = default; |
||
294 | |||
295 | void setDefiningAccess(MemoryAccess *DMA, bool Optimized = false) { |
||
296 | if (!Optimized) { |
||
297 | setOperand(0, DMA); |
||
298 | return; |
||
299 | } |
||
300 | setOptimized(DMA); |
||
301 | } |
||
302 | |||
303 | private: |
||
304 | Instruction *MemoryInstruction; |
||
305 | }; |
||
306 | |||
307 | /// Represents read-only accesses to memory |
||
308 | /// |
||
309 | /// In particular, the set of Instructions that will be represented by |
||
310 | /// MemoryUse's is exactly the set of Instructions for which |
||
311 | /// AliasAnalysis::getModRefInfo returns "Ref". |
||
312 | class MemoryUse final : public MemoryUseOrDef { |
||
313 | public: |
||
314 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); |
||
315 | |||
316 | MemoryUse(LLVMContext &C, MemoryAccess *DMA, Instruction *MI, BasicBlock *BB) |
||
317 | : MemoryUseOrDef(C, DMA, MemoryUseVal, deleteMe, MI, BB, |
||
318 | /*NumOperands=*/1) {} |
||
319 | |||
320 | // allocate space for exactly one operand |
||
321 | void *operator new(size_t S) { return User::operator new(S, 1); } |
||
322 | void operator delete(void *Ptr) { User::operator delete(Ptr); } |
||
323 | |||
324 | static bool classof(const Value *MA) { |
||
325 | return MA->getValueID() == MemoryUseVal; |
||
326 | } |
||
327 | |||
328 | void print(raw_ostream &OS) const; |
||
329 | |||
330 | void setOptimized(MemoryAccess *DMA) { |
||
331 | OptimizedID = DMA->getID(); |
||
332 | setOperand(0, DMA); |
||
333 | } |
||
334 | |||
335 | /// Whether the MemoryUse is optimized. If ensureOptimizedUses() was called, |
||
336 | /// uses will usually be optimized, but this is not guaranteed (e.g. due to |
||
337 | /// invalidation and optimization limits.) |
||
338 | bool isOptimized() const { |
||
339 | return getDefiningAccess() && OptimizedID == getDefiningAccess()->getID(); |
||
340 | } |
||
341 | |||
342 | MemoryAccess *getOptimized() const { |
||
343 | return getDefiningAccess(); |
||
344 | } |
||
345 | |||
346 | void resetOptimized() { |
||
347 | OptimizedID = INVALID_MEMORYACCESS_ID; |
||
348 | } |
||
349 | |||
350 | protected: |
||
351 | friend class MemorySSA; |
||
352 | |||
353 | private: |
||
354 | static void deleteMe(DerivedUser *Self); |
||
355 | |||
356 | unsigned OptimizedID = INVALID_MEMORYACCESS_ID; |
||
357 | }; |
||
358 | |||
359 | template <> |
||
360 | struct OperandTraits<MemoryUse> : public FixedNumOperandTraits<MemoryUse, 1> {}; |
||
361 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryUse, MemoryAccess) |
||
362 | |||
363 | /// Represents a read-write access to memory, whether it is a must-alias, |
||
364 | /// or a may-alias. |
||
365 | /// |
||
366 | /// In particular, the set of Instructions that will be represented by |
||
367 | /// MemoryDef's is exactly the set of Instructions for which |
||
368 | /// AliasAnalysis::getModRefInfo returns "Mod" or "ModRef". |
||
369 | /// Note that, in order to provide def-def chains, all defs also have a use |
||
370 | /// associated with them. This use points to the nearest reaching |
||
371 | /// MemoryDef/MemoryPhi. |
||
372 | class MemoryDef final : public MemoryUseOrDef { |
||
373 | public: |
||
374 | friend class MemorySSA; |
||
375 | |||
376 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); |
||
377 | |||
378 | MemoryDef(LLVMContext &C, MemoryAccess *DMA, Instruction *MI, BasicBlock *BB, |
||
379 | unsigned Ver) |
||
380 | : MemoryUseOrDef(C, DMA, MemoryDefVal, deleteMe, MI, BB, |
||
381 | /*NumOperands=*/2), |
||
382 | ID(Ver) {} |
||
383 | |||
384 | // allocate space for exactly two operands |
||
385 | void *operator new(size_t S) { return User::operator new(S, 2); } |
||
386 | void operator delete(void *Ptr) { User::operator delete(Ptr); } |
||
387 | |||
388 | static bool classof(const Value *MA) { |
||
389 | return MA->getValueID() == MemoryDefVal; |
||
390 | } |
||
391 | |||
392 | void setOptimized(MemoryAccess *MA) { |
||
393 | setOperand(1, MA); |
||
394 | OptimizedID = MA->getID(); |
||
395 | } |
||
396 | |||
397 | MemoryAccess *getOptimized() const { |
||
398 | return cast_or_null<MemoryAccess>(getOperand(1)); |
||
399 | } |
||
400 | |||
401 | bool isOptimized() const { |
||
402 | return getOptimized() && OptimizedID == getOptimized()->getID(); |
||
403 | } |
||
404 | |||
405 | void resetOptimized() { |
||
406 | OptimizedID = INVALID_MEMORYACCESS_ID; |
||
407 | setOperand(1, nullptr); |
||
408 | } |
||
409 | |||
410 | void print(raw_ostream &OS) const; |
||
411 | |||
412 | unsigned getID() const { return ID; } |
||
413 | |||
414 | private: |
||
415 | static void deleteMe(DerivedUser *Self); |
||
416 | |||
417 | const unsigned ID; |
||
418 | unsigned OptimizedID = INVALID_MEMORYACCESS_ID; |
||
419 | }; |
||
420 | |||
421 | template <> |
||
422 | struct OperandTraits<MemoryDef> : public FixedNumOperandTraits<MemoryDef, 2> {}; |
||
423 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryDef, MemoryAccess) |
||
424 | |||
425 | template <> |
||
426 | struct OperandTraits<MemoryUseOrDef> { |
||
427 | static Use *op_begin(MemoryUseOrDef *MUD) { |
||
428 | if (auto *MU = dyn_cast<MemoryUse>(MUD)) |
||
429 | return OperandTraits<MemoryUse>::op_begin(MU); |
||
430 | return OperandTraits<MemoryDef>::op_begin(cast<MemoryDef>(MUD)); |
||
431 | } |
||
432 | |||
433 | static Use *op_end(MemoryUseOrDef *MUD) { |
||
434 | if (auto *MU = dyn_cast<MemoryUse>(MUD)) |
||
435 | return OperandTraits<MemoryUse>::op_end(MU); |
||
436 | return OperandTraits<MemoryDef>::op_end(cast<MemoryDef>(MUD)); |
||
437 | } |
||
438 | |||
439 | static unsigned operands(const MemoryUseOrDef *MUD) { |
||
440 | if (const auto *MU = dyn_cast<MemoryUse>(MUD)) |
||
441 | return OperandTraits<MemoryUse>::operands(MU); |
||
442 | return OperandTraits<MemoryDef>::operands(cast<MemoryDef>(MUD)); |
||
443 | } |
||
444 | }; |
||
445 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryUseOrDef, MemoryAccess) |
||
446 | |||
447 | /// Represents phi nodes for memory accesses. |
||
448 | /// |
||
449 | /// These have the same semantic as regular phi nodes, with the exception that |
||
450 | /// only one phi will ever exist in a given basic block. |
||
451 | /// Guaranteeing one phi per block means guaranteeing there is only ever one |
||
452 | /// valid reaching MemoryDef/MemoryPHI along each path to the phi node. |
||
453 | /// This is ensured by not allowing disambiguation of the RHS of a MemoryDef or |
||
454 | /// a MemoryPhi's operands. |
||
455 | /// That is, given |
||
456 | /// if (a) { |
||
457 | /// store %a |
||
458 | /// store %b |
||
459 | /// } |
||
460 | /// it *must* be transformed into |
||
461 | /// if (a) { |
||
462 | /// 1 = MemoryDef(liveOnEntry) |
||
463 | /// store %a |
||
464 | /// 2 = MemoryDef(1) |
||
465 | /// store %b |
||
466 | /// } |
||
467 | /// and *not* |
||
468 | /// if (a) { |
||
469 | /// 1 = MemoryDef(liveOnEntry) |
||
470 | /// store %a |
||
471 | /// 2 = MemoryDef(liveOnEntry) |
||
472 | /// store %b |
||
473 | /// } |
||
474 | /// even if the two stores do not conflict. Otherwise, both 1 and 2 reach the |
||
475 | /// end of the branch, and if there are not two phi nodes, one will be |
||
476 | /// disconnected completely from the SSA graph below that point. |
||
477 | /// Because MemoryUse's do not generate new definitions, they do not have this |
||
478 | /// issue. |
||
479 | class MemoryPhi final : public MemoryAccess { |
||
480 | // allocate space for exactly zero operands |
||
481 | void *operator new(size_t S) { return User::operator new(S); } |
||
482 | |||
483 | public: |
||
484 | void operator delete(void *Ptr) { User::operator delete(Ptr); } |
||
485 | |||
486 | /// Provide fast operand accessors |
||
487 | DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); |
||
488 | |||
489 | MemoryPhi(LLVMContext &C, BasicBlock *BB, unsigned Ver, unsigned NumPreds = 0) |
||
490 | : MemoryAccess(C, MemoryPhiVal, deleteMe, BB, 0), ID(Ver), |
||
491 | ReservedSpace(NumPreds) { |
||
492 | allocHungoffUses(ReservedSpace); |
||
493 | } |
||
494 | |||
495 | // Block iterator interface. This provides access to the list of incoming |
||
496 | // basic blocks, which parallels the list of incoming values. |
||
497 | using block_iterator = BasicBlock **; |
||
498 | using const_block_iterator = BasicBlock *const *; |
||
499 | |||
500 | block_iterator block_begin() { |
||
501 | return reinterpret_cast<block_iterator>(op_begin() + ReservedSpace); |
||
502 | } |
||
503 | |||
504 | const_block_iterator block_begin() const { |
||
505 | return reinterpret_cast<const_block_iterator>(op_begin() + ReservedSpace); |
||
506 | } |
||
507 | |||
508 | block_iterator block_end() { return block_begin() + getNumOperands(); } |
||
509 | |||
510 | const_block_iterator block_end() const { |
||
511 | return block_begin() + getNumOperands(); |
||
512 | } |
||
513 | |||
514 | iterator_range<block_iterator> blocks() { |
||
515 | return make_range(block_begin(), block_end()); |
||
516 | } |
||
517 | |||
518 | iterator_range<const_block_iterator> blocks() const { |
||
519 | return make_range(block_begin(), block_end()); |
||
520 | } |
||
521 | |||
522 | op_range incoming_values() { return operands(); } |
||
523 | |||
524 | const_op_range incoming_values() const { return operands(); } |
||
525 | |||
526 | /// Return the number of incoming edges |
||
527 | unsigned getNumIncomingValues() const { return getNumOperands(); } |
||
528 | |||
529 | /// Return incoming value number x |
||
530 | MemoryAccess *getIncomingValue(unsigned I) const { return getOperand(I); } |
||
531 | void setIncomingValue(unsigned I, MemoryAccess *V) { |
||
532 | assert(V && "PHI node got a null value!"); |
||
533 | setOperand(I, V); |
||
534 | } |
||
535 | |||
536 | static unsigned getOperandNumForIncomingValue(unsigned I) { return I; } |
||
537 | static unsigned getIncomingValueNumForOperand(unsigned I) { return I; } |
||
538 | |||
539 | /// Return incoming basic block number @p i. |
||
540 | BasicBlock *getIncomingBlock(unsigned I) const { return block_begin()[I]; } |
||
541 | |||
542 | /// Return incoming basic block corresponding |
||
543 | /// to an operand of the PHI. |
||
544 | BasicBlock *getIncomingBlock(const Use &U) const { |
||
545 | assert(this == U.getUser() && "Iterator doesn't point to PHI's Uses?"); |
||
546 | return getIncomingBlock(unsigned(&U - op_begin())); |
||
547 | } |
||
548 | |||
549 | /// Return incoming basic block corresponding |
||
550 | /// to value use iterator. |
||
551 | BasicBlock *getIncomingBlock(MemoryAccess::const_user_iterator I) const { |
||
552 | return getIncomingBlock(I.getUse()); |
||
553 | } |
||
554 | |||
555 | void setIncomingBlock(unsigned I, BasicBlock *BB) { |
||
556 | assert(BB && "PHI node got a null basic block!"); |
||
557 | block_begin()[I] = BB; |
||
558 | } |
||
559 | |||
560 | /// Add an incoming value to the end of the PHI list |
||
561 | void addIncoming(MemoryAccess *V, BasicBlock *BB) { |
||
562 | if (getNumOperands() == ReservedSpace) |
||
563 | growOperands(); // Get more space! |
||
564 | // Initialize some new operands. |
||
565 | setNumHungOffUseOperands(getNumOperands() + 1); |
||
566 | setIncomingValue(getNumOperands() - 1, V); |
||
567 | setIncomingBlock(getNumOperands() - 1, BB); |
||
568 | } |
||
569 | |||
570 | /// Return the first index of the specified basic |
||
571 | /// block in the value list for this PHI. Returns -1 if no instance. |
||
572 | int getBasicBlockIndex(const BasicBlock *BB) const { |
||
573 | for (unsigned I = 0, E = getNumOperands(); I != E; ++I) |
||
574 | if (block_begin()[I] == BB) |
||
575 | return I; |
||
576 | return -1; |
||
577 | } |
||
578 | |||
579 | MemoryAccess *getIncomingValueForBlock(const BasicBlock *BB) const { |
||
580 | int Idx = getBasicBlockIndex(BB); |
||
581 | assert(Idx >= 0 && "Invalid basic block argument!"); |
||
582 | return getIncomingValue(Idx); |
||
583 | } |
||
584 | |||
585 | // After deleting incoming position I, the order of incoming may be changed. |
||
586 | void unorderedDeleteIncoming(unsigned I) { |
||
587 | unsigned E = getNumOperands(); |
||
588 | assert(I < E && "Cannot remove out of bounds Phi entry."); |
||
589 | // MemoryPhi must have at least two incoming values, otherwise the MemoryPhi |
||
590 | // itself should be deleted. |
||
591 | assert(E >= 2 && "Cannot only remove incoming values in MemoryPhis with " |
||
592 | "at least 2 values."); |
||
593 | setIncomingValue(I, getIncomingValue(E - 1)); |
||
594 | setIncomingBlock(I, block_begin()[E - 1]); |
||
595 | setOperand(E - 1, nullptr); |
||
596 | block_begin()[E - 1] = nullptr; |
||
597 | setNumHungOffUseOperands(getNumOperands() - 1); |
||
598 | } |
||
599 | |||
600 | // After deleting entries that satisfy Pred, remaining entries may have |
||
601 | // changed order. |
||
602 | template <typename Fn> void unorderedDeleteIncomingIf(Fn &&Pred) { |
||
603 | for (unsigned I = 0, E = getNumOperands(); I != E; ++I) |
||
604 | if (Pred(getIncomingValue(I), getIncomingBlock(I))) { |
||
605 | unorderedDeleteIncoming(I); |
||
606 | E = getNumOperands(); |
||
607 | --I; |
||
608 | } |
||
609 | assert(getNumOperands() >= 1 && |
||
610 | "Cannot remove all incoming blocks in a MemoryPhi."); |
||
611 | } |
||
612 | |||
613 | // After deleting incoming block BB, the incoming blocks order may be changed. |
||
614 | void unorderedDeleteIncomingBlock(const BasicBlock *BB) { |
||
615 | unorderedDeleteIncomingIf( |
||
616 | [&](const MemoryAccess *, const BasicBlock *B) { return BB == B; }); |
||
617 | } |
||
618 | |||
619 | // After deleting incoming memory access MA, the incoming accesses order may |
||
620 | // be changed. |
||
621 | void unorderedDeleteIncomingValue(const MemoryAccess *MA) { |
||
622 | unorderedDeleteIncomingIf( |
||
623 | [&](const MemoryAccess *M, const BasicBlock *) { return MA == M; }); |
||
624 | } |
||
625 | |||
626 | static bool classof(const Value *V) { |
||
627 | return V->getValueID() == MemoryPhiVal; |
||
628 | } |
||
629 | |||
630 | void print(raw_ostream &OS) const; |
||
631 | |||
632 | unsigned getID() const { return ID; } |
||
633 | |||
634 | protected: |
||
635 | friend class MemorySSA; |
||
636 | |||
637 | /// this is more complicated than the generic |
||
638 | /// User::allocHungoffUses, because we have to allocate Uses for the incoming |
||
639 | /// values and pointers to the incoming blocks, all in one allocation. |
||
640 | void allocHungoffUses(unsigned N) { |
||
641 | User::allocHungoffUses(N, /* IsPhi */ true); |
||
642 | } |
||
643 | |||
644 | private: |
||
645 | // For debugging only |
||
646 | const unsigned ID; |
||
647 | unsigned ReservedSpace; |
||
648 | |||
649 | /// This grows the operand list in response to a push_back style of |
||
650 | /// operation. This grows the number of ops by 1.5 times. |
||
651 | void growOperands() { |
||
652 | unsigned E = getNumOperands(); |
||
653 | // 2 op PHI nodes are VERY common, so reserve at least enough for that. |
||
654 | ReservedSpace = std::max(E + E / 2, 2u); |
||
655 | growHungoffUses(ReservedSpace, /* IsPhi */ true); |
||
656 | } |
||
657 | |||
658 | static void deleteMe(DerivedUser *Self); |
||
659 | }; |
||
660 | |||
661 | inline unsigned MemoryAccess::getID() const { |
||
662 | assert((isa<MemoryDef>(this) || isa<MemoryPhi>(this)) && |
||
663 | "only memory defs and phis have ids"); |
||
664 | if (const auto *MD = dyn_cast<MemoryDef>(this)) |
||
665 | return MD->getID(); |
||
666 | return cast<MemoryPhi>(this)->getID(); |
||
667 | } |
||
668 | |||
669 | inline bool MemoryUseOrDef::isOptimized() const { |
||
670 | if (const auto *MD = dyn_cast<MemoryDef>(this)) |
||
671 | return MD->isOptimized(); |
||
672 | return cast<MemoryUse>(this)->isOptimized(); |
||
673 | } |
||
674 | |||
675 | inline MemoryAccess *MemoryUseOrDef::getOptimized() const { |
||
676 | if (const auto *MD = dyn_cast<MemoryDef>(this)) |
||
677 | return MD->getOptimized(); |
||
678 | return cast<MemoryUse>(this)->getOptimized(); |
||
679 | } |
||
680 | |||
681 | inline void MemoryUseOrDef::setOptimized(MemoryAccess *MA) { |
||
682 | if (auto *MD = dyn_cast<MemoryDef>(this)) |
||
683 | MD->setOptimized(MA); |
||
684 | else |
||
685 | cast<MemoryUse>(this)->setOptimized(MA); |
||
686 | } |
||
687 | |||
688 | inline void MemoryUseOrDef::resetOptimized() { |
||
689 | if (auto *MD = dyn_cast<MemoryDef>(this)) |
||
690 | MD->resetOptimized(); |
||
691 | else |
||
692 | cast<MemoryUse>(this)->resetOptimized(); |
||
693 | } |
||
694 | |||
695 | template <> struct OperandTraits<MemoryPhi> : public HungoffOperandTraits<2> {}; |
||
696 | DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryPhi, MemoryAccess) |
||
697 | |||
698 | /// Encapsulates MemorySSA, including all data associated with memory |
||
699 | /// accesses. |
||
700 | class MemorySSA { |
||
701 | public: |
||
702 | MemorySSA(Function &, AliasAnalysis *, DominatorTree *); |
||
703 | |||
704 | // MemorySSA must remain where it's constructed; Walkers it creates store |
||
705 | // pointers to it. |
||
706 | MemorySSA(MemorySSA &&) = delete; |
||
707 | |||
708 | ~MemorySSA(); |
||
709 | |||
710 | MemorySSAWalker *getWalker(); |
||
711 | MemorySSAWalker *getSkipSelfWalker(); |
||
712 | |||
713 | /// Given a memory Mod/Ref'ing instruction, get the MemorySSA |
||
714 | /// access associated with it. If passed a basic block gets the memory phi |
||
715 | /// node that exists for that block, if there is one. Otherwise, this will get |
||
716 | /// a MemoryUseOrDef. |
||
717 | MemoryUseOrDef *getMemoryAccess(const Instruction *I) const { |
||
718 | return cast_or_null<MemoryUseOrDef>(ValueToMemoryAccess.lookup(I)); |
||
719 | } |
||
720 | |||
721 | MemoryPhi *getMemoryAccess(const BasicBlock *BB) const { |
||
722 | return cast_or_null<MemoryPhi>(ValueToMemoryAccess.lookup(cast<Value>(BB))); |
||
723 | } |
||
724 | |||
725 | DominatorTree &getDomTree() const { return *DT; } |
||
726 | |||
727 | void dump() const; |
||
728 | void print(raw_ostream &) const; |
||
729 | |||
730 | /// Return true if \p MA represents the live on entry value |
||
731 | /// |
||
732 | /// Loads and stores from pointer arguments and other global values may be |
||
733 | /// defined by memory operations that do not occur in the current function, so |
||
734 | /// they may be live on entry to the function. MemorySSA represents such |
||
735 | /// memory state by the live on entry definition, which is guaranteed to occur |
||
736 | /// before any other memory access in the function. |
||
737 | inline bool isLiveOnEntryDef(const MemoryAccess *MA) const { |
||
738 | return MA == LiveOnEntryDef.get(); |
||
739 | } |
||
740 | |||
741 | inline MemoryAccess *getLiveOnEntryDef() const { |
||
742 | return LiveOnEntryDef.get(); |
||
743 | } |
||
744 | |||
745 | // Sadly, iplists, by default, owns and deletes pointers added to the |
||
746 | // list. It's not currently possible to have two iplists for the same type, |
||
747 | // where one owns the pointers, and one does not. This is because the traits |
||
748 | // are per-type, not per-tag. If this ever changes, we should make the |
||
749 | // DefList an iplist. |
||
750 | using AccessList = iplist<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>; |
||
751 | using DefsList = |
||
752 | simple_ilist<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>>; |
||
753 | |||
754 | /// Return the list of MemoryAccess's for a given basic block. |
||
755 | /// |
||
756 | /// This list is not modifiable by the user. |
||
757 | const AccessList *getBlockAccesses(const BasicBlock *BB) const { |
||
758 | return getWritableBlockAccesses(BB); |
||
759 | } |
||
760 | |||
761 | /// Return the list of MemoryDef's and MemoryPhi's for a given basic |
||
762 | /// block. |
||
763 | /// |
||
764 | /// This list is not modifiable by the user. |
||
765 | const DefsList *getBlockDefs(const BasicBlock *BB) const { |
||
766 | return getWritableBlockDefs(BB); |
||
767 | } |
||
768 | |||
769 | /// Given two memory accesses in the same basic block, determine |
||
770 | /// whether MemoryAccess \p A dominates MemoryAccess \p B. |
||
771 | bool locallyDominates(const MemoryAccess *A, const MemoryAccess *B) const; |
||
772 | |||
773 | /// Given two memory accesses in potentially different blocks, |
||
774 | /// determine whether MemoryAccess \p A dominates MemoryAccess \p B. |
||
775 | bool dominates(const MemoryAccess *A, const MemoryAccess *B) const; |
||
776 | |||
777 | /// Given a MemoryAccess and a Use, determine whether MemoryAccess \p A |
||
778 | /// dominates Use \p B. |
||
779 | bool dominates(const MemoryAccess *A, const Use &B) const; |
||
780 | |||
781 | enum class VerificationLevel { Fast, Full }; |
||
782 | /// Verify that MemorySSA is self consistent (IE definitions dominate |
||
783 | /// all uses, uses appear in the right places). This is used by unit tests. |
||
784 | void verifyMemorySSA(VerificationLevel = VerificationLevel::Fast) const; |
||
785 | |||
786 | /// Used in various insertion functions to specify whether we are talking |
||
787 | /// about the beginning or end of a block. |
||
788 | enum InsertionPlace { Beginning, End, BeforeTerminator }; |
||
789 | |||
790 | /// By default, uses are *not* optimized during MemorySSA construction. |
||
791 | /// Calling this method will attempt to optimize all MemoryUses, if this has |
||
792 | /// not happened yet for this MemorySSA instance. This should be done if you |
||
793 | /// plan to query the clobbering access for most uses, or if you walk the |
||
794 | /// def-use chain of uses. |
||
795 | void ensureOptimizedUses(); |
||
796 | |||
797 | AliasAnalysis &getAA() { return *AA; } |
||
798 | |||
799 | protected: |
||
800 | // Used by Memory SSA dumpers and wrapper pass |
||
801 | friend class MemorySSAPrinterLegacyPass; |
||
802 | friend class MemorySSAUpdater; |
||
803 | |||
804 | void verifyOrderingDominationAndDefUses( |
||
805 | Function &F, VerificationLevel = VerificationLevel::Fast) const; |
||
806 | void verifyDominationNumbers(const Function &F) const; |
||
807 | void verifyPrevDefInPhis(Function &F) const; |
||
808 | |||
809 | // This is used by the use optimizer and updater. |
||
810 | AccessList *getWritableBlockAccesses(const BasicBlock *BB) const { |
||
811 | auto It = PerBlockAccesses.find(BB); |
||
812 | return It == PerBlockAccesses.end() ? nullptr : It->second.get(); |
||
813 | } |
||
814 | |||
815 | // This is used by the use optimizer and updater. |
||
816 | DefsList *getWritableBlockDefs(const BasicBlock *BB) const { |
||
817 | auto It = PerBlockDefs.find(BB); |
||
818 | return It == PerBlockDefs.end() ? nullptr : It->second.get(); |
||
819 | } |
||
820 | |||
821 | // These is used by the updater to perform various internal MemorySSA |
||
822 | // machinsations. They do not always leave the IR in a correct state, and |
||
823 | // relies on the updater to fixup what it breaks, so it is not public. |
||
824 | |||
825 | void moveTo(MemoryUseOrDef *What, BasicBlock *BB, AccessList::iterator Where); |
||
826 | void moveTo(MemoryAccess *What, BasicBlock *BB, InsertionPlace Point); |
||
827 | |||
828 | // Rename the dominator tree branch rooted at BB. |
||
829 | void renamePass(BasicBlock *BB, MemoryAccess *IncomingVal, |
||
830 | SmallPtrSetImpl<BasicBlock *> &Visited) { |
||
831 | renamePass(DT->getNode(BB), IncomingVal, Visited, true, true); |
||
832 | } |
||
833 | |||
834 | void removeFromLookups(MemoryAccess *); |
||
835 | void removeFromLists(MemoryAccess *, bool ShouldDelete = true); |
||
836 | void insertIntoListsForBlock(MemoryAccess *, const BasicBlock *, |
||
837 | InsertionPlace); |
||
838 | void insertIntoListsBefore(MemoryAccess *, const BasicBlock *, |
||
839 | AccessList::iterator); |
||
840 | MemoryUseOrDef *createDefinedAccess(Instruction *, MemoryAccess *, |
||
841 | const MemoryUseOrDef *Template = nullptr, |
||
842 | bool CreationMustSucceed = true); |
||
843 | |||
844 | private: |
||
845 | class ClobberWalkerBase; |
||
846 | class CachingWalker; |
||
847 | class SkipSelfWalker; |
||
848 | class OptimizeUses; |
||
849 | |||
850 | CachingWalker *getWalkerImpl(); |
||
851 | void buildMemorySSA(BatchAAResults &BAA); |
||
852 | |||
853 | void prepareForMoveTo(MemoryAccess *, BasicBlock *); |
||
854 | void verifyUseInDefs(MemoryAccess *, MemoryAccess *) const; |
||
855 | |||
856 | using AccessMap = DenseMap<const BasicBlock *, std::unique_ptr<AccessList>>; |
||
857 | using DefsMap = DenseMap<const BasicBlock *, std::unique_ptr<DefsList>>; |
||
858 | |||
859 | void markUnreachableAsLiveOnEntry(BasicBlock *BB); |
||
860 | MemoryPhi *createMemoryPhi(BasicBlock *BB); |
||
861 | template <typename AliasAnalysisType> |
||
862 | MemoryUseOrDef *createNewAccess(Instruction *, AliasAnalysisType *, |
||
863 | const MemoryUseOrDef *Template = nullptr); |
||
864 | void placePHINodes(const SmallPtrSetImpl<BasicBlock *> &); |
||
865 | MemoryAccess *renameBlock(BasicBlock *, MemoryAccess *, bool); |
||
866 | void renameSuccessorPhis(BasicBlock *, MemoryAccess *, bool); |
||
867 | void renamePass(DomTreeNode *, MemoryAccess *IncomingVal, |
||
868 | SmallPtrSetImpl<BasicBlock *> &Visited, |
||
869 | bool SkipVisited = false, bool RenameAllUses = false); |
||
870 | AccessList *getOrCreateAccessList(const BasicBlock *); |
||
871 | DefsList *getOrCreateDefsList(const BasicBlock *); |
||
872 | void renumberBlock(const BasicBlock *) const; |
||
873 | AliasAnalysis *AA = nullptr; |
||
874 | DominatorTree *DT; |
||
875 | Function &F; |
||
876 | |||
877 | // Memory SSA mappings |
||
878 | DenseMap<const Value *, MemoryAccess *> ValueToMemoryAccess; |
||
879 | |||
880 | // These two mappings contain the main block to access/def mappings for |
||
881 | // MemorySSA. The list contained in PerBlockAccesses really owns all the |
||
882 | // MemoryAccesses. |
||
883 | // Both maps maintain the invariant that if a block is found in them, the |
||
884 | // corresponding list is not empty, and if a block is not found in them, the |
||
885 | // corresponding list is empty. |
||
886 | AccessMap PerBlockAccesses; |
||
887 | DefsMap PerBlockDefs; |
||
888 | std::unique_ptr<MemoryAccess, ValueDeleter> LiveOnEntryDef; |
||
889 | |||
890 | // Domination mappings |
||
891 | // Note that the numbering is local to a block, even though the map is |
||
892 | // global. |
||
893 | mutable SmallPtrSet<const BasicBlock *, 16> BlockNumberingValid; |
||
894 | mutable DenseMap<const MemoryAccess *, unsigned long> BlockNumbering; |
||
895 | |||
896 | // Memory SSA building info |
||
897 | std::unique_ptr<ClobberWalkerBase> WalkerBase; |
||
898 | std::unique_ptr<CachingWalker> Walker; |
||
899 | std::unique_ptr<SkipSelfWalker> SkipWalker; |
||
900 | unsigned NextID = 0; |
||
901 | bool IsOptimized = false; |
||
902 | }; |
||
903 | |||
904 | /// Enables verification of MemorySSA. |
||
905 | /// |
||
906 | /// The checks which this flag enables is exensive and disabled by default |
||
907 | /// unless `EXPENSIVE_CHECKS` is defined. The flag `-verify-memoryssa` can be |
||
908 | /// used to selectively enable the verification without re-compilation. |
||
909 | extern bool VerifyMemorySSA; |
||
910 | |||
911 | // Internal MemorySSA utils, for use by MemorySSA classes and walkers |
||
912 | class MemorySSAUtil { |
||
913 | protected: |
||
914 | friend class GVNHoist; |
||
915 | friend class MemorySSAWalker; |
||
916 | |||
917 | // This function should not be used by new passes. |
||
918 | static bool defClobbersUseOrDef(MemoryDef *MD, const MemoryUseOrDef *MU, |
||
919 | AliasAnalysis &AA); |
||
920 | }; |
||
921 | |||
922 | // This pass does eager building and then printing of MemorySSA. It is used by |
||
923 | // the tests to be able to build, dump, and verify Memory SSA. |
||
924 | class MemorySSAPrinterLegacyPass : public FunctionPass { |
||
925 | public: |
||
926 | MemorySSAPrinterLegacyPass(); |
||
927 | |||
928 | bool runOnFunction(Function &) override; |
||
929 | void getAnalysisUsage(AnalysisUsage &AU) const override; |
||
930 | |||
931 | static char ID; |
||
932 | }; |
||
933 | |||
934 | /// An analysis that produces \c MemorySSA for a function. |
||
935 | /// |
||
936 | class MemorySSAAnalysis : public AnalysisInfoMixin<MemorySSAAnalysis> { |
||
937 | friend AnalysisInfoMixin<MemorySSAAnalysis>; |
||
938 | |||
939 | static AnalysisKey Key; |
||
940 | |||
941 | public: |
||
942 | // Wrap MemorySSA result to ensure address stability of internal MemorySSA |
||
943 | // pointers after construction. Use a wrapper class instead of plain |
||
944 | // unique_ptr<MemorySSA> to avoid build breakage on MSVC. |
||
945 | struct Result { |
||
946 | Result(std::unique_ptr<MemorySSA> &&MSSA) : MSSA(std::move(MSSA)) {} |
||
947 | |||
948 | MemorySSA &getMSSA() { return *MSSA.get(); } |
||
949 | |||
950 | std::unique_ptr<MemorySSA> MSSA; |
||
951 | |||
952 | bool invalidate(Function &F, const PreservedAnalyses &PA, |
||
953 | FunctionAnalysisManager::Invalidator &Inv); |
||
954 | }; |
||
955 | |||
956 | Result run(Function &F, FunctionAnalysisManager &AM); |
||
957 | }; |
||
958 | |||
959 | /// Printer pass for \c MemorySSA. |
||
960 | class MemorySSAPrinterPass : public PassInfoMixin<MemorySSAPrinterPass> { |
||
961 | raw_ostream &OS; |
||
962 | |||
963 | public: |
||
964 | explicit MemorySSAPrinterPass(raw_ostream &OS) : OS(OS) {} |
||
965 | |||
966 | PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM); |
||
967 | }; |
||
968 | |||
969 | /// Printer pass for \c MemorySSA via the walker. |
||
970 | class MemorySSAWalkerPrinterPass |
||
971 | : public PassInfoMixin<MemorySSAWalkerPrinterPass> { |
||
972 | raw_ostream &OS; |
||
973 | |||
974 | public: |
||
975 | explicit MemorySSAWalkerPrinterPass(raw_ostream &OS) : OS(OS) {} |
||
976 | |||
977 | PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM); |
||
978 | }; |
||
979 | |||
980 | /// Verifier pass for \c MemorySSA. |
||
981 | struct MemorySSAVerifierPass : PassInfoMixin<MemorySSAVerifierPass> { |
||
982 | PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM); |
||
983 | }; |
||
984 | |||
985 | /// Legacy analysis pass which computes \c MemorySSA. |
||
986 | class MemorySSAWrapperPass : public FunctionPass { |
||
987 | public: |
||
988 | MemorySSAWrapperPass(); |
||
989 | |||
990 | static char ID; |
||
991 | |||
992 | bool runOnFunction(Function &) override; |
||
993 | void releaseMemory() override; |
||
994 | MemorySSA &getMSSA() { return *MSSA; } |
||
995 | const MemorySSA &getMSSA() const { return *MSSA; } |
||
996 | |||
997 | void getAnalysisUsage(AnalysisUsage &AU) const override; |
||
998 | |||
999 | void verifyAnalysis() const override; |
||
1000 | void print(raw_ostream &OS, const Module *M = nullptr) const override; |
||
1001 | |||
1002 | private: |
||
1003 | std::unique_ptr<MemorySSA> MSSA; |
||
1004 | }; |
||
1005 | |||
1006 | /// This is the generic walker interface for walkers of MemorySSA. |
||
1007 | /// Walkers are used to be able to further disambiguate the def-use chains |
||
1008 | /// MemorySSA gives you, or otherwise produce better info than MemorySSA gives |
||
1009 | /// you. |
||
1010 | /// In particular, while the def-use chains provide basic information, and are |
||
1011 | /// guaranteed to give, for example, the nearest may-aliasing MemoryDef for a |
||
1012 | /// MemoryUse as AliasAnalysis considers it, a user mant want better or other |
||
1013 | /// information. In particular, they may want to use SCEV info to further |
||
1014 | /// disambiguate memory accesses, or they may want the nearest dominating |
||
1015 | /// may-aliasing MemoryDef for a call or a store. This API enables a |
||
1016 | /// standardized interface to getting and using that info. |
||
1017 | class MemorySSAWalker { |
||
1018 | public: |
||
1019 | MemorySSAWalker(MemorySSA *); |
||
1020 | virtual ~MemorySSAWalker() = default; |
||
1021 | |||
1022 | using MemoryAccessSet = SmallVector<MemoryAccess *, 8>; |
||
1023 | |||
1024 | /// Given a memory Mod/Ref/ModRef'ing instruction, calling this |
||
1025 | /// will give you the nearest dominating MemoryAccess that Mod's the location |
||
1026 | /// the instruction accesses (by skipping any def which AA can prove does not |
||
1027 | /// alias the location(s) accessed by the instruction given). |
||
1028 | /// |
||
1029 | /// Note that this will return a single access, and it must dominate the |
||
1030 | /// Instruction, so if an operand of a MemoryPhi node Mod's the instruction, |
||
1031 | /// this will return the MemoryPhi, not the operand. This means that |
||
1032 | /// given: |
||
1033 | /// if (a) { |
||
1034 | /// 1 = MemoryDef(liveOnEntry) |
||
1035 | /// store %a |
||
1036 | /// } else { |
||
1037 | /// 2 = MemoryDef(liveOnEntry) |
||
1038 | /// store %b |
||
1039 | /// } |
||
1040 | /// 3 = MemoryPhi(2, 1) |
||
1041 | /// MemoryUse(3) |
||
1042 | /// load %a |
||
1043 | /// |
||
1044 | /// calling this API on load(%a) will return the MemoryPhi, not the MemoryDef |
||
1045 | /// in the if (a) branch. |
||
1046 | MemoryAccess *getClobberingMemoryAccess(const Instruction *I, |
||
1047 | BatchAAResults &AA) { |
||
1048 | MemoryAccess *MA = MSSA->getMemoryAccess(I); |
||
1049 | assert(MA && "Handed an instruction that MemorySSA doesn't recognize?"); |
||
1050 | return getClobberingMemoryAccess(MA, AA); |
||
1051 | } |
||
1052 | |||
1053 | /// Does the same thing as getClobberingMemoryAccess(const Instruction *I), |
||
1054 | /// but takes a MemoryAccess instead of an Instruction. |
||
1055 | virtual MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, |
||
1056 | BatchAAResults &AA) = 0; |
||
1057 | |||
1058 | /// Given a potentially clobbering memory access and a new location, |
||
1059 | /// calling this will give you the nearest dominating clobbering MemoryAccess |
||
1060 | /// (by skipping non-aliasing def links). |
||
1061 | /// |
||
1062 | /// This version of the function is mainly used to disambiguate phi translated |
||
1063 | /// pointers, where the value of a pointer may have changed from the initial |
||
1064 | /// memory access. Note that this expects to be handed either a MemoryUse, |
||
1065 | /// or an already potentially clobbering access. Unlike the above API, if |
||
1066 | /// given a MemoryDef that clobbers the pointer as the starting access, it |
||
1067 | /// will return that MemoryDef, whereas the above would return the clobber |
||
1068 | /// starting from the use side of the memory def. |
||
1069 | virtual MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, |
||
1070 | const MemoryLocation &, |
||
1071 | BatchAAResults &AA) = 0; |
||
1072 | |||
1073 | MemoryAccess *getClobberingMemoryAccess(const Instruction *I) { |
||
1074 | BatchAAResults BAA(MSSA->getAA()); |
||
1075 | return getClobberingMemoryAccess(I, BAA); |
||
1076 | } |
||
1077 | |||
1078 | MemoryAccess *getClobberingMemoryAccess(MemoryAccess *MA) { |
||
1079 | BatchAAResults BAA(MSSA->getAA()); |
||
1080 | return getClobberingMemoryAccess(MA, BAA); |
||
1081 | } |
||
1082 | |||
1083 | MemoryAccess *getClobberingMemoryAccess(MemoryAccess *MA, |
||
1084 | const MemoryLocation &Loc) { |
||
1085 | BatchAAResults BAA(MSSA->getAA()); |
||
1086 | return getClobberingMemoryAccess(MA, Loc, BAA); |
||
1087 | } |
||
1088 | |||
1089 | /// Given a memory access, invalidate anything this walker knows about |
||
1090 | /// that access. |
||
1091 | /// This API is used by walkers that store information to perform basic cache |
||
1092 | /// invalidation. This will be called by MemorySSA at appropriate times for |
||
1093 | /// the walker it uses or returns. |
||
1094 | virtual void invalidateInfo(MemoryAccess *) {} |
||
1095 | |||
1096 | protected: |
||
1097 | friend class MemorySSA; // For updating MSSA pointer in MemorySSA move |
||
1098 | // constructor. |
||
1099 | MemorySSA *MSSA; |
||
1100 | }; |
||
1101 | |||
1102 | /// A MemorySSAWalker that does no alias queries, or anything else. It |
||
1103 | /// simply returns the links as they were constructed by the builder. |
||
1104 | class DoNothingMemorySSAWalker final : public MemorySSAWalker { |
||
1105 | public: |
||
1106 | // Keep the overrides below from hiding the Instruction overload of |
||
1107 | // getClobberingMemoryAccess. |
||
1108 | using MemorySSAWalker::getClobberingMemoryAccess; |
||
1109 | |||
1110 | MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, |
||
1111 | BatchAAResults &) override; |
||
1112 | MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, |
||
1113 | const MemoryLocation &, |
||
1114 | BatchAAResults &) override; |
||
1115 | }; |
||
1116 | |||
1117 | using MemoryAccessPair = std::pair<MemoryAccess *, MemoryLocation>; |
||
1118 | using ConstMemoryAccessPair = std::pair<const MemoryAccess *, MemoryLocation>; |
||
1119 | |||
1120 | /// Iterator base class used to implement const and non-const iterators |
||
1121 | /// over the defining accesses of a MemoryAccess. |
||
1122 | template <class T> |
||
1123 | class memoryaccess_def_iterator_base |
||
1124 | : public iterator_facade_base<memoryaccess_def_iterator_base<T>, |
||
1125 | std::forward_iterator_tag, T, ptrdiff_t, T *, |
||
1126 | T *> { |
||
1127 | using BaseT = typename memoryaccess_def_iterator_base::iterator_facade_base; |
||
1128 | |||
1129 | public: |
||
1130 | memoryaccess_def_iterator_base(T *Start) : Access(Start) {} |
||
1131 | memoryaccess_def_iterator_base() = default; |
||
1132 | |||
1133 | bool operator==(const memoryaccess_def_iterator_base &Other) const { |
||
1134 | return Access == Other.Access && (!Access || ArgNo == Other.ArgNo); |
||
1135 | } |
||
1136 | |||
1137 | // This is a bit ugly, but for MemoryPHI's, unlike PHINodes, you can't get the |
||
1138 | // block from the operand in constant time (In a PHINode, the uselist has |
||
1139 | // both, so it's just subtraction). We provide it as part of the |
||
1140 | // iterator to avoid callers having to linear walk to get the block. |
||
1141 | // If the operation becomes constant time on MemoryPHI's, this bit of |
||
1142 | // abstraction breaking should be removed. |
||
1143 | BasicBlock *getPhiArgBlock() const { |
||
1144 | MemoryPhi *MP = dyn_cast<MemoryPhi>(Access); |
||
1145 | assert(MP && "Tried to get phi arg block when not iterating over a PHI"); |
||
1146 | return MP->getIncomingBlock(ArgNo); |
||
1147 | } |
||
1148 | |||
1149 | typename std::iterator_traits<BaseT>::pointer operator*() const { |
||
1150 | assert(Access && "Tried to access past the end of our iterator"); |
||
1151 | // Go to the first argument for phis, and the defining access for everything |
||
1152 | // else. |
||
1153 | if (const MemoryPhi *MP = dyn_cast<MemoryPhi>(Access)) |
||
1154 | return MP->getIncomingValue(ArgNo); |
||
1155 | return cast<MemoryUseOrDef>(Access)->getDefiningAccess(); |
||
1156 | } |
||
1157 | |||
1158 | using BaseT::operator++; |
||
1159 | memoryaccess_def_iterator_base &operator++() { |
||
1160 | assert(Access && "Hit end of iterator"); |
||
1161 | if (const MemoryPhi *MP = dyn_cast<MemoryPhi>(Access)) { |
||
1162 | if (++ArgNo >= MP->getNumIncomingValues()) { |
||
1163 | ArgNo = 0; |
||
1164 | Access = nullptr; |
||
1165 | } |
||
1166 | } else { |
||
1167 | Access = nullptr; |
||
1168 | } |
||
1169 | return *this; |
||
1170 | } |
||
1171 | |||
1172 | private: |
||
1173 | T *Access = nullptr; |
||
1174 | unsigned ArgNo = 0; |
||
1175 | }; |
||
1176 | |||
1177 | inline memoryaccess_def_iterator MemoryAccess::defs_begin() { |
||
1178 | return memoryaccess_def_iterator(this); |
||
1179 | } |
||
1180 | |||
1181 | inline const_memoryaccess_def_iterator MemoryAccess::defs_begin() const { |
||
1182 | return const_memoryaccess_def_iterator(this); |
||
1183 | } |
||
1184 | |||
1185 | inline memoryaccess_def_iterator MemoryAccess::defs_end() { |
||
1186 | return memoryaccess_def_iterator(); |
||
1187 | } |
||
1188 | |||
1189 | inline const_memoryaccess_def_iterator MemoryAccess::defs_end() const { |
||
1190 | return const_memoryaccess_def_iterator(); |
||
1191 | } |
||
1192 | |||
1193 | /// GraphTraits for a MemoryAccess, which walks defs in the normal case, |
||
1194 | /// and uses in the inverse case. |
||
1195 | template <> struct GraphTraits<MemoryAccess *> { |
||
1196 | using NodeRef = MemoryAccess *; |
||
1197 | using ChildIteratorType = memoryaccess_def_iterator; |
||
1198 | |||
1199 | static NodeRef getEntryNode(NodeRef N) { return N; } |
||
1200 | static ChildIteratorType child_begin(NodeRef N) { return N->defs_begin(); } |
||
1201 | static ChildIteratorType child_end(NodeRef N) { return N->defs_end(); } |
||
1202 | }; |
||
1203 | |||
1204 | template <> struct GraphTraits<Inverse<MemoryAccess *>> { |
||
1205 | using NodeRef = MemoryAccess *; |
||
1206 | using ChildIteratorType = MemoryAccess::iterator; |
||
1207 | |||
1208 | static NodeRef getEntryNode(NodeRef N) { return N; } |
||
1209 | static ChildIteratorType child_begin(NodeRef N) { return N->user_begin(); } |
||
1210 | static ChildIteratorType child_end(NodeRef N) { return N->user_end(); } |
||
1211 | }; |
||
1212 | |||
1213 | /// Provide an iterator that walks defs, giving both the memory access, |
||
1214 | /// and the current pointer location, updating the pointer location as it |
||
1215 | /// changes due to phi node translation. |
||
1216 | /// |
||
1217 | /// This iterator, while somewhat specialized, is what most clients actually |
||
1218 | /// want when walking upwards through MemorySSA def chains. It takes a pair of |
||
1219 | /// <MemoryAccess,MemoryLocation>, and walks defs, properly translating the |
||
1220 | /// memory location through phi nodes for the user. |
||
1221 | class upward_defs_iterator |
||
1222 | : public iterator_facade_base<upward_defs_iterator, |
||
1223 | std::forward_iterator_tag, |
||
1224 | const MemoryAccessPair> { |
||
1225 | using BaseT = upward_defs_iterator::iterator_facade_base; |
||
1226 | |||
1227 | public: |
||
1228 | upward_defs_iterator(const MemoryAccessPair &Info, DominatorTree *DT) |
||
1229 | : DefIterator(Info.first), Location(Info.second), |
||
1230 | OriginalAccess(Info.first), DT(DT) { |
||
1231 | CurrentPair.first = nullptr; |
||
1232 | |||
1233 | WalkingPhi = Info.first && isa<MemoryPhi>(Info.first); |
||
1234 | fillInCurrentPair(); |
||
1235 | } |
||
1236 | |||
1237 | upward_defs_iterator() { CurrentPair.first = nullptr; } |
||
1238 | |||
1239 | bool operator==(const upward_defs_iterator &Other) const { |
||
1240 | return DefIterator == Other.DefIterator; |
||
1241 | } |
||
1242 | |||
1243 | typename std::iterator_traits<BaseT>::reference operator*() const { |
||
1244 | assert(DefIterator != OriginalAccess->defs_end() && |
||
1245 | "Tried to access past the end of our iterator"); |
||
1246 | return CurrentPair; |
||
1247 | } |
||
1248 | |||
1249 | using BaseT::operator++; |
||
1250 | upward_defs_iterator &operator++() { |
||
1251 | assert(DefIterator != OriginalAccess->defs_end() && |
||
1252 | "Tried to access past the end of the iterator"); |
||
1253 | ++DefIterator; |
||
1254 | if (DefIterator != OriginalAccess->defs_end()) |
||
1255 | fillInCurrentPair(); |
||
1256 | return *this; |
||
1257 | } |
||
1258 | |||
1259 | BasicBlock *getPhiArgBlock() const { return DefIterator.getPhiArgBlock(); } |
||
1260 | |||
1261 | private: |
||
1262 | /// Returns true if \p Ptr is guaranteed to be loop invariant for any possible |
||
1263 | /// loop. In particular, this guarantees that it only references a single |
||
1264 | /// MemoryLocation during execution of the containing function. |
||
1265 | bool IsGuaranteedLoopInvariant(const Value *Ptr) const; |
||
1266 | |||
1267 | void fillInCurrentPair() { |
||
1268 | CurrentPair.first = *DefIterator; |
||
1269 | CurrentPair.second = Location; |
||
1270 | if (WalkingPhi && Location.Ptr) { |
||
1271 | PHITransAddr Translator( |
||
1272 | const_cast<Value *>(Location.Ptr), |
||
1273 | OriginalAccess->getBlock()->getModule()->getDataLayout(), nullptr); |
||
1274 | |||
1275 | if (!Translator.PHITranslateValue(OriginalAccess->getBlock(), |
||
1276 | DefIterator.getPhiArgBlock(), DT, true)) |
||
1277 | if (Translator.getAddr() != CurrentPair.second.Ptr) |
||
1278 | CurrentPair.second = |
||
1279 | CurrentPair.second.getWithNewPtr(Translator.getAddr()); |
||
1280 | |||
1281 | // Mark size as unknown, if the location is not guaranteed to be |
||
1282 | // loop-invariant for any possible loop in the function. Setting the size |
||
1283 | // to unknown guarantees that any memory accesses that access locations |
||
1284 | // after the pointer are considered as clobbers, which is important to |
||
1285 | // catch loop carried dependences. |
||
1286 | if (!IsGuaranteedLoopInvariant(CurrentPair.second.Ptr)) |
||
1287 | CurrentPair.second = CurrentPair.second.getWithNewSize( |
||
1288 | LocationSize::beforeOrAfterPointer()); |
||
1289 | } |
||
1290 | } |
||
1291 | |||
1292 | MemoryAccessPair CurrentPair; |
||
1293 | memoryaccess_def_iterator DefIterator; |
||
1294 | MemoryLocation Location; |
||
1295 | MemoryAccess *OriginalAccess = nullptr; |
||
1296 | DominatorTree *DT = nullptr; |
||
1297 | bool WalkingPhi = false; |
||
1298 | }; |
||
1299 | |||
1300 | inline upward_defs_iterator |
||
1301 | upward_defs_begin(const MemoryAccessPair &Pair, DominatorTree &DT) { |
||
1302 | return upward_defs_iterator(Pair, &DT); |
||
1303 | } |
||
1304 | |||
1305 | inline upward_defs_iterator upward_defs_end() { return upward_defs_iterator(); } |
||
1306 | |||
1307 | inline iterator_range<upward_defs_iterator> |
||
1308 | upward_defs(const MemoryAccessPair &Pair, DominatorTree &DT) { |
||
1309 | return make_range(upward_defs_begin(Pair, DT), upward_defs_end()); |
||
1310 | } |
||
1311 | |||
1312 | /// Walks the defining accesses of MemoryDefs. Stops after we hit something that |
||
1313 | /// has no defining use (e.g. a MemoryPhi or liveOnEntry). Note that, when |
||
1314 | /// comparing against a null def_chain_iterator, this will compare equal only |
||
1315 | /// after walking said Phi/liveOnEntry. |
||
1316 | /// |
||
1317 | /// The UseOptimizedChain flag specifies whether to walk the clobbering |
||
1318 | /// access chain, or all the accesses. |
||
1319 | /// |
||
1320 | /// Normally, MemoryDef are all just def/use linked together, so a def_chain on |
||
1321 | /// a MemoryDef will walk all MemoryDefs above it in the program until it hits |
||
1322 | /// a phi node. The optimized chain walks the clobbering access of a store. |
||
1323 | /// So if you are just trying to find, given a store, what the next |
||
1324 | /// thing that would clobber the same memory is, you want the optimized chain. |
||
1325 | template <class T, bool UseOptimizedChain = false> |
||
1326 | struct def_chain_iterator |
||
1327 | : public iterator_facade_base<def_chain_iterator<T, UseOptimizedChain>, |
||
1328 | std::forward_iterator_tag, MemoryAccess *> { |
||
1329 | def_chain_iterator() : MA(nullptr) {} |
||
1330 | def_chain_iterator(T MA) : MA(MA) {} |
||
1331 | |||
1332 | T operator*() const { return MA; } |
||
1333 | |||
1334 | def_chain_iterator &operator++() { |
||
1335 | // N.B. liveOnEntry has a null defining access. |
||
1336 | if (auto *MUD = dyn_cast<MemoryUseOrDef>(MA)) { |
||
1337 | if (UseOptimizedChain && MUD->isOptimized()) |
||
1338 | MA = MUD->getOptimized(); |
||
1339 | else |
||
1340 | MA = MUD->getDefiningAccess(); |
||
1341 | } else { |
||
1342 | MA = nullptr; |
||
1343 | } |
||
1344 | |||
1345 | return *this; |
||
1346 | } |
||
1347 | |||
1348 | bool operator==(const def_chain_iterator &O) const { return MA == O.MA; } |
||
1349 | |||
1350 | private: |
||
1351 | T MA; |
||
1352 | }; |
||
1353 | |||
1354 | template <class T> |
||
1355 | inline iterator_range<def_chain_iterator<T>> |
||
1356 | def_chain(T MA, MemoryAccess *UpTo = nullptr) { |
||
1357 | #ifdef EXPENSIVE_CHECKS |
||
1358 | assert((!UpTo || find(def_chain(MA), UpTo) != def_chain_iterator<T>()) && |
||
1359 | "UpTo isn't in the def chain!"); |
||
1360 | #endif |
||
1361 | return make_range(def_chain_iterator<T>(MA), def_chain_iterator<T>(UpTo)); |
||
1362 | } |
||
1363 | |||
1364 | template <class T> |
||
1365 | inline iterator_range<def_chain_iterator<T, true>> optimized_def_chain(T MA) { |
||
1366 | return make_range(def_chain_iterator<T, true>(MA), |
||
1367 | def_chain_iterator<T, true>(nullptr)); |
||
1368 | } |
||
1369 | |||
1370 | } // end namespace llvm |
||
1371 | |||
1372 | #endif // LLVM_ANALYSIS_MEMORYSSA_H |