File: | llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp |
Warning: | line 2516, column 19 Value stored to 'Current' during its initialization is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | //===- DeadStoreElimination.cpp - Fast Dead Store Elimination -------------===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | // |
9 | // This file implements a trivial dead store elimination that only considers |
10 | // basic-block local redundant stores. |
11 | // |
12 | // FIXME: This should eventually be extended to be a post-dominator tree |
13 | // traversal. Doing so would be pretty trivial. |
14 | // |
15 | //===----------------------------------------------------------------------===// |
16 | |
17 | #include "llvm/Transforms/Scalar/DeadStoreElimination.h" |
18 | #include "llvm/ADT/APInt.h" |
19 | #include "llvm/ADT/DenseMap.h" |
20 | #include "llvm/ADT/MapVector.h" |
21 | #include "llvm/ADT/PostOrderIterator.h" |
22 | #include "llvm/ADT/SetVector.h" |
23 | #include "llvm/ADT/SmallPtrSet.h" |
24 | #include "llvm/ADT/SmallVector.h" |
25 | #include "llvm/ADT/Statistic.h" |
26 | #include "llvm/ADT/StringRef.h" |
27 | #include "llvm/Analysis/AliasAnalysis.h" |
28 | #include "llvm/Analysis/CaptureTracking.h" |
29 | #include "llvm/Analysis/GlobalsModRef.h" |
30 | #include "llvm/Analysis/MemoryBuiltins.h" |
31 | #include "llvm/Analysis/MemoryDependenceAnalysis.h" |
32 | #include "llvm/Analysis/MemoryLocation.h" |
33 | #include "llvm/Analysis/MemorySSA.h" |
34 | #include "llvm/Analysis/MemorySSAUpdater.h" |
35 | #include "llvm/Analysis/PostDominators.h" |
36 | #include "llvm/Analysis/TargetLibraryInfo.h" |
37 | #include "llvm/Analysis/ValueTracking.h" |
38 | #include "llvm/IR/Argument.h" |
39 | #include "llvm/IR/BasicBlock.h" |
40 | #include "llvm/IR/Constant.h" |
41 | #include "llvm/IR/Constants.h" |
42 | #include "llvm/IR/DataLayout.h" |
43 | #include "llvm/IR/Dominators.h" |
44 | #include "llvm/IR/Function.h" |
45 | #include "llvm/IR/InstIterator.h" |
46 | #include "llvm/IR/InstrTypes.h" |
47 | #include "llvm/IR/Instruction.h" |
48 | #include "llvm/IR/Instructions.h" |
49 | #include "llvm/IR/IntrinsicInst.h" |
50 | #include "llvm/IR/Intrinsics.h" |
51 | #include "llvm/IR/LLVMContext.h" |
52 | #include "llvm/IR/Module.h" |
53 | #include "llvm/IR/PassManager.h" |
54 | #include "llvm/IR/PatternMatch.h" |
55 | #include "llvm/IR/Value.h" |
56 | #include "llvm/InitializePasses.h" |
57 | #include "llvm/Pass.h" |
58 | #include "llvm/Support/Casting.h" |
59 | #include "llvm/Support/CommandLine.h" |
60 | #include "llvm/Support/Debug.h" |
61 | #include "llvm/Support/DebugCounter.h" |
62 | #include "llvm/Support/ErrorHandling.h" |
63 | #include "llvm/Support/MathExtras.h" |
64 | #include "llvm/Support/raw_ostream.h" |
65 | #include "llvm/Transforms/Scalar.h" |
66 | #include "llvm/Transforms/Utils/AssumeBundleBuilder.h" |
67 | #include "llvm/Transforms/Utils/Local.h" |
68 | #include <algorithm> |
69 | #include <cassert> |
70 | #include <cstddef> |
71 | #include <cstdint> |
72 | #include <iterator> |
73 | #include <map> |
74 | #include <utility> |
75 | |
76 | using namespace llvm; |
77 | using namespace PatternMatch; |
78 | |
79 | #define DEBUG_TYPE"dse" "dse" |
80 | |
81 | STATISTIC(NumRemainingStores, "Number of stores remaining after DSE")static llvm::Statistic NumRemainingStores = {"dse", "NumRemainingStores" , "Number of stores remaining after DSE"}; |
82 | STATISTIC(NumRedundantStores, "Number of redundant stores deleted")static llvm::Statistic NumRedundantStores = {"dse", "NumRedundantStores" , "Number of redundant stores deleted"}; |
83 | STATISTIC(NumFastStores, "Number of stores deleted")static llvm::Statistic NumFastStores = {"dse", "NumFastStores" , "Number of stores deleted"}; |
84 | STATISTIC(NumFastOther, "Number of other instrs removed")static llvm::Statistic NumFastOther = {"dse", "NumFastOther", "Number of other instrs removed"}; |
85 | STATISTIC(NumCompletePartials, "Number of stores dead by later partials")static llvm::Statistic NumCompletePartials = {"dse", "NumCompletePartials" , "Number of stores dead by later partials"}; |
86 | STATISTIC(NumModifiedStores, "Number of stores modified")static llvm::Statistic NumModifiedStores = {"dse", "NumModifiedStores" , "Number of stores modified"}; |
87 | STATISTIC(NumCFGChecks, "Number of stores modified")static llvm::Statistic NumCFGChecks = {"dse", "NumCFGChecks", "Number of stores modified"}; |
88 | STATISTIC(NumCFGTries, "Number of stores modified")static llvm::Statistic NumCFGTries = {"dse", "NumCFGTries", "Number of stores modified" }; |
89 | STATISTIC(NumCFGSuccess, "Number of stores modified")static llvm::Statistic NumCFGSuccess = {"dse", "NumCFGSuccess" , "Number of stores modified"}; |
90 | STATISTIC(NumGetDomMemoryDefPassed,static llvm::Statistic NumGetDomMemoryDefPassed = {"dse", "NumGetDomMemoryDefPassed" , "Number of times a valid candidate is returned from getDomMemoryDef" } |
91 | "Number of times a valid candidate is returned from getDomMemoryDef")static llvm::Statistic NumGetDomMemoryDefPassed = {"dse", "NumGetDomMemoryDefPassed" , "Number of times a valid candidate is returned from getDomMemoryDef" }; |
92 | STATISTIC(NumDomMemDefChecks,static llvm::Statistic NumDomMemDefChecks = {"dse", "NumDomMemDefChecks" , "Number iterations check for reads in getDomMemoryDef"} |
93 | "Number iterations check for reads in getDomMemoryDef")static llvm::Statistic NumDomMemDefChecks = {"dse", "NumDomMemDefChecks" , "Number iterations check for reads in getDomMemoryDef"}; |
94 | |
95 | DEBUG_COUNTER(MemorySSACounter, "dse-memoryssa",static const unsigned MemorySSACounter = DebugCounter::registerCounter ("dse-memoryssa", "Controls which MemoryDefs are eliminated." ) |
96 | "Controls which MemoryDefs are eliminated.")static const unsigned MemorySSACounter = DebugCounter::registerCounter ("dse-memoryssa", "Controls which MemoryDefs are eliminated." ); |
97 | |
98 | static cl::opt<bool> |
99 | EnablePartialOverwriteTracking("enable-dse-partial-overwrite-tracking", |
100 | cl::init(true), cl::Hidden, |
101 | cl::desc("Enable partial-overwrite tracking in DSE")); |
102 | |
103 | static cl::opt<bool> |
104 | EnablePartialStoreMerging("enable-dse-partial-store-merging", |
105 | cl::init(true), cl::Hidden, |
106 | cl::desc("Enable partial store merging in DSE")); |
107 | |
108 | static cl::opt<bool> |
109 | EnableMemorySSA("enable-dse-memoryssa", cl::init(true), cl::Hidden, |
110 | cl::desc("Use the new MemorySSA-backed DSE.")); |
111 | |
112 | static cl::opt<unsigned> |
113 | MemorySSAScanLimit("dse-memoryssa-scanlimit", cl::init(150), cl::Hidden, |
114 | cl::desc("The number of memory instructions to scan for " |
115 | "dead store elimination (default = 100)")); |
116 | static cl::opt<unsigned> MemorySSAUpwardsStepLimit( |
117 | "dse-memoryssa-walklimit", cl::init(90), cl::Hidden, |
118 | cl::desc("The maximum number of steps while walking upwards to find " |
119 | "MemoryDefs that may be killed (default = 90)")); |
120 | |
121 | static cl::opt<unsigned> MemorySSAPartialStoreLimit( |
122 | "dse-memoryssa-partial-store-limit", cl::init(5), cl::Hidden, |
123 | cl::desc("The maximum number candidates that only partially overwrite the " |
124 | "killing MemoryDef to consider" |
125 | " (default = 5)")); |
126 | |
127 | static cl::opt<unsigned> MemorySSADefsPerBlockLimit( |
128 | "dse-memoryssa-defs-per-block-limit", cl::init(5000), cl::Hidden, |
129 | cl::desc("The number of MemoryDefs we consider as candidates to eliminated " |
130 | "other stores per basic block (default = 5000)")); |
131 | |
132 | static cl::opt<unsigned> MemorySSASameBBStepCost( |
133 | "dse-memoryssa-samebb-cost", cl::init(1), cl::Hidden, |
134 | cl::desc( |
135 | "The cost of a step in the same basic block as the killing MemoryDef" |
136 | "(default = 1)")); |
137 | |
138 | static cl::opt<unsigned> |
139 | MemorySSAOtherBBStepCost("dse-memoryssa-otherbb-cost", cl::init(5), |
140 | cl::Hidden, |
141 | cl::desc("The cost of a step in a different basic " |
142 | "block than the killing MemoryDef" |
143 | "(default = 5)")); |
144 | |
145 | static cl::opt<unsigned> MemorySSAPathCheckLimit( |
146 | "dse-memoryssa-path-check-limit", cl::init(50), cl::Hidden, |
147 | cl::desc("The maximum number of blocks to check when trying to prove that " |
148 | "all paths to an exit go through a killing block (default = 50)")); |
149 | |
150 | //===----------------------------------------------------------------------===// |
151 | // Helper functions |
152 | //===----------------------------------------------------------------------===// |
153 | using OverlapIntervalsTy = std::map<int64_t, int64_t>; |
154 | using InstOverlapIntervalsTy = DenseMap<Instruction *, OverlapIntervalsTy>; |
155 | |
156 | /// Delete this instruction. Before we do, go through and zero out all the |
157 | /// operands of this instruction. If any of them become dead, delete them and |
158 | /// the computation tree that feeds them. |
159 | /// If ValueSet is non-null, remove any deleted instructions from it as well. |
160 | static void |
161 | deleteDeadInstruction(Instruction *I, BasicBlock::iterator *BBI, |
162 | MemoryDependenceResults &MD, const TargetLibraryInfo &TLI, |
163 | InstOverlapIntervalsTy &IOL, |
164 | MapVector<Instruction *, bool> &ThrowableInst, |
165 | SmallSetVector<const Value *, 16> *ValueSet = nullptr) { |
166 | SmallVector<Instruction*, 32> NowDeadInsts; |
167 | |
168 | NowDeadInsts.push_back(I); |
169 | --NumFastOther; |
170 | |
171 | // Keeping the iterator straight is a pain, so we let this routine tell the |
172 | // caller what the next instruction is after we're done mucking about. |
173 | BasicBlock::iterator NewIter = *BBI; |
174 | |
175 | // Before we touch this instruction, remove it from memdep! |
176 | do { |
177 | Instruction *DeadInst = NowDeadInsts.pop_back_val(); |
178 | // Mark the DeadInst as dead in the list of throwable instructions. |
179 | auto It = ThrowableInst.find(DeadInst); |
180 | if (It != ThrowableInst.end()) |
181 | ThrowableInst[It->first] = false; |
182 | ++NumFastOther; |
183 | |
184 | // Try to preserve debug information attached to the dead instruction. |
185 | salvageDebugInfo(*DeadInst); |
186 | salvageKnowledge(DeadInst); |
187 | |
188 | // This instruction is dead, zap it, in stages. Start by removing it from |
189 | // MemDep, which needs to know the operands and needs it to be in the |
190 | // function. |
191 | MD.removeInstruction(DeadInst); |
192 | |
193 | for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) { |
194 | Value *Op = DeadInst->getOperand(op); |
195 | DeadInst->setOperand(op, nullptr); |
196 | |
197 | // If this operand just became dead, add it to the NowDeadInsts list. |
198 | if (!Op->use_empty()) continue; |
199 | |
200 | if (Instruction *OpI = dyn_cast<Instruction>(Op)) |
201 | if (isInstructionTriviallyDead(OpI, &TLI)) |
202 | NowDeadInsts.push_back(OpI); |
203 | } |
204 | |
205 | if (ValueSet) ValueSet->remove(DeadInst); |
206 | IOL.erase(DeadInst); |
207 | |
208 | if (NewIter == DeadInst->getIterator()) |
209 | NewIter = DeadInst->eraseFromParent(); |
210 | else |
211 | DeadInst->eraseFromParent(); |
212 | } while (!NowDeadInsts.empty()); |
213 | *BBI = NewIter; |
214 | // Pop dead entries from back of ThrowableInst till we find an alive entry. |
215 | while (!ThrowableInst.empty() && !ThrowableInst.back().second) |
216 | ThrowableInst.pop_back(); |
217 | } |
218 | |
219 | /// Does this instruction write some memory? This only returns true for things |
220 | /// that we can analyze with other helpers below. |
221 | static bool hasAnalyzableMemoryWrite(Instruction *I, |
222 | const TargetLibraryInfo &TLI) { |
223 | if (isa<StoreInst>(I)) |
224 | return true; |
225 | if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) { |
226 | switch (II->getIntrinsicID()) { |
227 | default: |
228 | return false; |
229 | case Intrinsic::memset: |
230 | case Intrinsic::memmove: |
231 | case Intrinsic::memcpy: |
232 | case Intrinsic::memcpy_inline: |
233 | case Intrinsic::memcpy_element_unordered_atomic: |
234 | case Intrinsic::memmove_element_unordered_atomic: |
235 | case Intrinsic::memset_element_unordered_atomic: |
236 | case Intrinsic::init_trampoline: |
237 | case Intrinsic::lifetime_end: |
238 | case Intrinsic::masked_store: |
239 | return true; |
240 | } |
241 | } |
242 | if (auto *CB = dyn_cast<CallBase>(I)) { |
243 | LibFunc LF; |
244 | if (TLI.getLibFunc(*CB, LF) && TLI.has(LF)) { |
245 | switch (LF) { |
246 | case LibFunc_strcpy: |
247 | case LibFunc_strncpy: |
248 | case LibFunc_strcat: |
249 | case LibFunc_strncat: |
250 | return true; |
251 | default: |
252 | return false; |
253 | } |
254 | } |
255 | } |
256 | return false; |
257 | } |
258 | |
259 | /// Return a Location stored to by the specified instruction. If isRemovable |
260 | /// returns true, this function and getLocForRead completely describe the memory |
261 | /// operations for this instruction. |
262 | static MemoryLocation getLocForWrite(Instruction *Inst, |
263 | const TargetLibraryInfo &TLI) { |
264 | if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) |
265 | return MemoryLocation::get(SI); |
266 | |
267 | if (auto *MI = dyn_cast<AnyMemIntrinsic>(Inst)) { |
268 | // memcpy/memmove/memset. |
269 | MemoryLocation Loc = MemoryLocation::getForDest(MI); |
270 | return Loc; |
271 | } |
272 | |
273 | if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) { |
274 | switch (II->getIntrinsicID()) { |
275 | default: |
276 | return MemoryLocation(); // Unhandled intrinsic. |
277 | case Intrinsic::init_trampoline: |
278 | return MemoryLocation(II->getArgOperand(0)); |
279 | case Intrinsic::masked_store: |
280 | return MemoryLocation::getForArgument(II, 1, TLI); |
281 | case Intrinsic::lifetime_end: { |
282 | uint64_t Len = cast<ConstantInt>(II->getArgOperand(0))->getZExtValue(); |
283 | return MemoryLocation(II->getArgOperand(1), Len); |
284 | } |
285 | } |
286 | } |
287 | if (auto *CB = dyn_cast<CallBase>(Inst)) |
288 | // All the supported TLI functions so far happen to have dest as their |
289 | // first argument. |
290 | return MemoryLocation(CB->getArgOperand(0)); |
291 | return MemoryLocation(); |
292 | } |
293 | |
294 | /// Return the location read by the specified "hasAnalyzableMemoryWrite" |
295 | /// instruction if any. |
296 | static MemoryLocation getLocForRead(Instruction *Inst, |
297 | const TargetLibraryInfo &TLI) { |
298 | assert(hasAnalyzableMemoryWrite(Inst, TLI) && "Unknown instruction case")((hasAnalyzableMemoryWrite(Inst, TLI) && "Unknown instruction case" ) ? static_cast<void> (0) : __assert_fail ("hasAnalyzableMemoryWrite(Inst, TLI) && \"Unknown instruction case\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 298, __PRETTY_FUNCTION__)); |
299 | |
300 | // The only instructions that both read and write are the mem transfer |
301 | // instructions (memcpy/memmove). |
302 | if (auto *MTI = dyn_cast<AnyMemTransferInst>(Inst)) |
303 | return MemoryLocation::getForSource(MTI); |
304 | return MemoryLocation(); |
305 | } |
306 | |
307 | /// If the value of this instruction and the memory it writes to is unused, may |
308 | /// we delete this instruction? |
309 | static bool isRemovable(Instruction *I) { |
310 | // Don't remove volatile/atomic stores. |
311 | if (StoreInst *SI = dyn_cast<StoreInst>(I)) |
312 | return SI->isUnordered(); |
313 | |
314 | if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) { |
315 | switch (II->getIntrinsicID()) { |
316 | default: llvm_unreachable("doesn't pass 'hasAnalyzableMemoryWrite' predicate")::llvm::llvm_unreachable_internal("doesn't pass 'hasAnalyzableMemoryWrite' predicate" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 316); |
317 | case Intrinsic::lifetime_end: |
318 | // Never remove dead lifetime_end's, e.g. because it is followed by a |
319 | // free. |
320 | return false; |
321 | case Intrinsic::init_trampoline: |
322 | // Always safe to remove init_trampoline. |
323 | return true; |
324 | case Intrinsic::memset: |
325 | case Intrinsic::memmove: |
326 | case Intrinsic::memcpy: |
327 | case Intrinsic::memcpy_inline: |
328 | // Don't remove volatile memory intrinsics. |
329 | return !cast<MemIntrinsic>(II)->isVolatile(); |
330 | case Intrinsic::memcpy_element_unordered_atomic: |
331 | case Intrinsic::memmove_element_unordered_atomic: |
332 | case Intrinsic::memset_element_unordered_atomic: |
333 | case Intrinsic::masked_store: |
334 | return true; |
335 | } |
336 | } |
337 | |
338 | // note: only get here for calls with analyzable writes - i.e. libcalls |
339 | if (auto *CB = dyn_cast<CallBase>(I)) |
340 | return CB->use_empty(); |
341 | |
342 | return false; |
343 | } |
344 | |
345 | /// Returns true if the end of this instruction can be safely shortened in |
346 | /// length. |
347 | static bool isShortenableAtTheEnd(Instruction *I) { |
348 | // Don't shorten stores for now |
349 | if (isa<StoreInst>(I)) |
350 | return false; |
351 | |
352 | if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) { |
353 | switch (II->getIntrinsicID()) { |
354 | default: return false; |
355 | case Intrinsic::memset: |
356 | case Intrinsic::memcpy: |
357 | case Intrinsic::memcpy_element_unordered_atomic: |
358 | case Intrinsic::memset_element_unordered_atomic: |
359 | // Do shorten memory intrinsics. |
360 | // FIXME: Add memmove if it's also safe to transform. |
361 | return true; |
362 | } |
363 | } |
364 | |
365 | // Don't shorten libcalls calls for now. |
366 | |
367 | return false; |
368 | } |
369 | |
370 | /// Returns true if the beginning of this instruction can be safely shortened |
371 | /// in length. |
372 | static bool isShortenableAtTheBeginning(Instruction *I) { |
373 | // FIXME: Handle only memset for now. Supporting memcpy/memmove should be |
374 | // easily done by offsetting the source address. |
375 | return isa<AnyMemSetInst>(I); |
376 | } |
377 | |
378 | /// Return the pointer that is being written to. |
379 | static Value *getStoredPointerOperand(Instruction *I, |
380 | const TargetLibraryInfo &TLI) { |
381 | //TODO: factor this to reuse getLocForWrite |
382 | MemoryLocation Loc = getLocForWrite(I, TLI); |
383 | assert(Loc.Ptr &&((Loc.Ptr && "unable to find pointer written for analyzable instruction?" ) ? static_cast<void> (0) : __assert_fail ("Loc.Ptr && \"unable to find pointer written for analyzable instruction?\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 384, __PRETTY_FUNCTION__)) |
384 | "unable to find pointer written for analyzable instruction?")((Loc.Ptr && "unable to find pointer written for analyzable instruction?" ) ? static_cast<void> (0) : __assert_fail ("Loc.Ptr && \"unable to find pointer written for analyzable instruction?\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 384, __PRETTY_FUNCTION__)); |
385 | // TODO: most APIs don't expect const Value * |
386 | return const_cast<Value*>(Loc.Ptr); |
387 | } |
388 | |
389 | static uint64_t getPointerSize(const Value *V, const DataLayout &DL, |
390 | const TargetLibraryInfo &TLI, |
391 | const Function *F) { |
392 | uint64_t Size; |
393 | ObjectSizeOpts Opts; |
394 | Opts.NullIsUnknownSize = NullPointerIsDefined(F); |
395 | |
396 | if (getObjectSize(V, Size, DL, &TLI, Opts)) |
397 | return Size; |
398 | return MemoryLocation::UnknownSize; |
399 | } |
400 | |
401 | namespace { |
402 | |
403 | enum OverwriteResult { |
404 | OW_Begin, |
405 | OW_Complete, |
406 | OW_End, |
407 | OW_PartialEarlierWithFullLater, |
408 | OW_MaybePartial, |
409 | OW_Unknown |
410 | }; |
411 | |
412 | } // end anonymous namespace |
413 | |
414 | /// Check if two instruction are masked stores that completely |
415 | /// overwrite one another. More specifically, \p Later has to |
416 | /// overwrite \p Earlier. |
417 | template <typename AATy> |
418 | static OverwriteResult isMaskedStoreOverwrite(const Instruction *Later, |
419 | const Instruction *Earlier, |
420 | AATy &AA) { |
421 | const auto *IIL = dyn_cast<IntrinsicInst>(Later); |
422 | const auto *IIE = dyn_cast<IntrinsicInst>(Earlier); |
423 | if (IIL == nullptr || IIE == nullptr) |
424 | return OW_Unknown; |
425 | if (IIL->getIntrinsicID() != Intrinsic::masked_store || |
426 | IIE->getIntrinsicID() != Intrinsic::masked_store) |
427 | return OW_Unknown; |
428 | // Pointers. |
429 | Value *LP = IIL->getArgOperand(1)->stripPointerCasts(); |
430 | Value *EP = IIE->getArgOperand(1)->stripPointerCasts(); |
431 | if (LP != EP && !AA.isMustAlias(LP, EP)) |
432 | return OW_Unknown; |
433 | // Masks. |
434 | // TODO: check that Later's mask is a superset of the Earlier's mask. |
435 | if (IIL->getArgOperand(3) != IIE->getArgOperand(3)) |
436 | return OW_Unknown; |
437 | return OW_Complete; |
438 | } |
439 | |
440 | /// Return 'OW_Complete' if a store to the 'Later' location (by \p LaterI |
441 | /// instruction) completely overwrites a store to the 'Earlier' location. |
442 | /// (by \p EarlierI instruction). |
443 | /// Return OW_MaybePartial if \p Later does not completely overwrite |
444 | /// \p Earlier, but they both write to the same underlying object. In that |
445 | /// case, use isPartialOverwrite to check if \p Later partially overwrites |
446 | /// \p Earlier. Returns 'OW_Unknown' if nothing can be determined. |
447 | template <typename AATy> |
448 | static OverwriteResult |
449 | isOverwrite(const Instruction *LaterI, const Instruction *EarlierI, |
450 | const MemoryLocation &Later, const MemoryLocation &Earlier, |
451 | const DataLayout &DL, const TargetLibraryInfo &TLI, |
452 | int64_t &EarlierOff, int64_t &LaterOff, AATy &AA, |
453 | const Function *F) { |
454 | // FIXME: Vet that this works for size upper-bounds. Seems unlikely that we'll |
455 | // get imprecise values here, though (except for unknown sizes). |
456 | if (!Later.Size.isPrecise() || !Earlier.Size.isPrecise()) { |
457 | // Masked stores have imprecise locations, but we can reason about them |
458 | // to some extent. |
459 | return isMaskedStoreOverwrite(LaterI, EarlierI, AA); |
460 | } |
461 | |
462 | const uint64_t LaterSize = Later.Size.getValue(); |
463 | const uint64_t EarlierSize = Earlier.Size.getValue(); |
464 | |
465 | const Value *P1 = Earlier.Ptr->stripPointerCasts(); |
466 | const Value *P2 = Later.Ptr->stripPointerCasts(); |
467 | |
468 | // If the start pointers are the same, we just have to compare sizes to see if |
469 | // the later store was larger than the earlier store. |
470 | if (P1 == P2 || AA.isMustAlias(P1, P2)) { |
471 | // Make sure that the Later size is >= the Earlier size. |
472 | if (LaterSize >= EarlierSize) |
473 | return OW_Complete; |
474 | } |
475 | |
476 | // Check to see if the later store is to the entire object (either a global, |
477 | // an alloca, or a byval/inalloca argument). If so, then it clearly |
478 | // overwrites any other store to the same object. |
479 | const Value *UO1 = getUnderlyingObject(P1), *UO2 = getUnderlyingObject(P2); |
480 | |
481 | // If we can't resolve the same pointers to the same object, then we can't |
482 | // analyze them at all. |
483 | if (UO1 != UO2) |
484 | return OW_Unknown; |
485 | |
486 | // If the "Later" store is to a recognizable object, get its size. |
487 | uint64_t ObjectSize = getPointerSize(UO2, DL, TLI, F); |
488 | if (ObjectSize != MemoryLocation::UnknownSize) |
489 | if (ObjectSize == LaterSize && ObjectSize >= EarlierSize) |
490 | return OW_Complete; |
491 | |
492 | // Okay, we have stores to two completely different pointers. Try to |
493 | // decompose the pointer into a "base + constant_offset" form. If the base |
494 | // pointers are equal, then we can reason about the two stores. |
495 | EarlierOff = 0; |
496 | LaterOff = 0; |
497 | const Value *BP1 = GetPointerBaseWithConstantOffset(P1, EarlierOff, DL); |
498 | const Value *BP2 = GetPointerBaseWithConstantOffset(P2, LaterOff, DL); |
499 | |
500 | // If the base pointers still differ, we have two completely different stores. |
501 | if (BP1 != BP2) |
502 | return OW_Unknown; |
503 | |
504 | // The later store completely overlaps the earlier store if: |
505 | // |
506 | // 1. Both start at the same offset and the later one's size is greater than |
507 | // or equal to the earlier one's, or |
508 | // |
509 | // |--earlier--| |
510 | // |-- later --| |
511 | // |
512 | // 2. The earlier store has an offset greater than the later offset, but which |
513 | // still lies completely within the later store. |
514 | // |
515 | // |--earlier--| |
516 | // |----- later ------| |
517 | // |
518 | // We have to be careful here as *Off is signed while *.Size is unsigned. |
519 | if (EarlierOff >= LaterOff && |
520 | LaterSize >= EarlierSize && |
521 | uint64_t(EarlierOff - LaterOff) + EarlierSize <= LaterSize) |
522 | return OW_Complete; |
523 | |
524 | // Later may overwrite earlier completely with other partial writes. |
525 | return OW_MaybePartial; |
526 | } |
527 | |
528 | /// Return 'OW_Complete' if a store to the 'Later' location completely |
529 | /// overwrites a store to the 'Earlier' location, 'OW_End' if the end of the |
530 | /// 'Earlier' location is completely overwritten by 'Later', 'OW_Begin' if the |
531 | /// beginning of the 'Earlier' location is overwritten by 'Later'. |
532 | /// 'OW_PartialEarlierWithFullLater' means that an earlier (big) store was |
533 | /// overwritten by a latter (smaller) store which doesn't write outside the big |
534 | /// store's memory locations. Returns 'OW_Unknown' if nothing can be determined. |
535 | /// NOTE: This function must only be called if both \p Later and \p Earlier |
536 | /// write to the same underlying object with valid \p EarlierOff and \p |
537 | /// LaterOff. |
538 | static OverwriteResult isPartialOverwrite(const MemoryLocation &Later, |
539 | const MemoryLocation &Earlier, |
540 | int64_t EarlierOff, int64_t LaterOff, |
541 | Instruction *DepWrite, |
542 | InstOverlapIntervalsTy &IOL) { |
543 | const uint64_t LaterSize = Later.Size.getValue(); |
544 | const uint64_t EarlierSize = Earlier.Size.getValue(); |
545 | // We may now overlap, although the overlap is not complete. There might also |
546 | // be other incomplete overlaps, and together, they might cover the complete |
547 | // earlier write. |
548 | // Note: The correctness of this logic depends on the fact that this function |
549 | // is not even called providing DepWrite when there are any intervening reads. |
550 | if (EnablePartialOverwriteTracking && |
551 | LaterOff < int64_t(EarlierOff + EarlierSize) && |
552 | int64_t(LaterOff + LaterSize) >= EarlierOff) { |
553 | |
554 | // Insert our part of the overlap into the map. |
555 | auto &IM = IOL[DepWrite]; |
556 | LLVM_DEBUG(dbgs() << "DSE: Partial overwrite: Earlier [" << EarlierOffdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Later [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false) |
557 | << ", " << int64_t(EarlierOff + EarlierSize)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Later [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false) |
558 | << ") Later [" << LaterOff << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Later [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false) |
559 | << int64_t(LaterOff + LaterSize) << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Later [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false); |
560 | |
561 | // Make sure that we only insert non-overlapping intervals and combine |
562 | // adjacent intervals. The intervals are stored in the map with the ending |
563 | // offset as the key (in the half-open sense) and the starting offset as |
564 | // the value. |
565 | int64_t LaterIntStart = LaterOff, LaterIntEnd = LaterOff + LaterSize; |
566 | |
567 | // Find any intervals ending at, or after, LaterIntStart which start |
568 | // before LaterIntEnd. |
569 | auto ILI = IM.lower_bound(LaterIntStart); |
570 | if (ILI != IM.end() && ILI->second <= LaterIntEnd) { |
571 | // This existing interval is overlapped with the current store somewhere |
572 | // in [LaterIntStart, LaterIntEnd]. Merge them by erasing the existing |
573 | // intervals and adjusting our start and end. |
574 | LaterIntStart = std::min(LaterIntStart, ILI->second); |
575 | LaterIntEnd = std::max(LaterIntEnd, ILI->first); |
576 | ILI = IM.erase(ILI); |
577 | |
578 | // Continue erasing and adjusting our end in case other previous |
579 | // intervals are also overlapped with the current store. |
580 | // |
581 | // |--- ealier 1 ---| |--- ealier 2 ---| |
582 | // |------- later---------| |
583 | // |
584 | while (ILI != IM.end() && ILI->second <= LaterIntEnd) { |
585 | assert(ILI->second > LaterIntStart && "Unexpected interval")((ILI->second > LaterIntStart && "Unexpected interval" ) ? static_cast<void> (0) : __assert_fail ("ILI->second > LaterIntStart && \"Unexpected interval\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 585, __PRETTY_FUNCTION__)); |
586 | LaterIntEnd = std::max(LaterIntEnd, ILI->first); |
587 | ILI = IM.erase(ILI); |
588 | } |
589 | } |
590 | |
591 | IM[LaterIntEnd] = LaterIntStart; |
592 | |
593 | ILI = IM.begin(); |
594 | if (ILI->second <= EarlierOff && |
595 | ILI->first >= int64_t(EarlierOff + EarlierSize)) { |
596 | LLVM_DEBUG(dbgs() << "DSE: Full overwrite from partials: Earlier ["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Full overwrite from partials: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Composite Later [" << ILI-> second << ", " << ILI->first << ")\n"; } } while (false) |
597 | << EarlierOff << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Full overwrite from partials: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Composite Later [" << ILI-> second << ", " << ILI->first << ")\n"; } } while (false) |
598 | << int64_t(EarlierOff + EarlierSize)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Full overwrite from partials: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Composite Later [" << ILI-> second << ", " << ILI->first << ")\n"; } } while (false) |
599 | << ") Composite Later [" << ILI->second << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Full overwrite from partials: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Composite Later [" << ILI-> second << ", " << ILI->first << ")\n"; } } while (false) |
600 | << ILI->first << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Full overwrite from partials: Earlier [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") Composite Later [" << ILI-> second << ", " << ILI->first << ")\n"; } } while (false); |
601 | ++NumCompletePartials; |
602 | return OW_Complete; |
603 | } |
604 | } |
605 | |
606 | // Check for an earlier store which writes to all the memory locations that |
607 | // the later store writes to. |
608 | if (EnablePartialStoreMerging && LaterOff >= EarlierOff && |
609 | int64_t(EarlierOff + EarlierSize) > LaterOff && |
610 | uint64_t(LaterOff - EarlierOff) + LaterSize <= EarlierSize) { |
611 | LLVM_DEBUG(dbgs() << "DSE: Partial overwrite an earlier load ["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite an earlier load [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") by a later store [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false) |
612 | << EarlierOff << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite an earlier load [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") by a later store [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false) |
613 | << int64_t(EarlierOff + EarlierSize)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite an earlier load [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") by a later store [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false) |
614 | << ") by a later store [" << LaterOff << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite an earlier load [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") by a later store [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false) |
615 | << int64_t(LaterOff + LaterSize) << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Partial overwrite an earlier load [" << EarlierOff << ", " << int64_t(EarlierOff + EarlierSize) << ") by a later store [" << LaterOff << ", " << int64_t(LaterOff + LaterSize) << ")\n"; } } while (false); |
616 | // TODO: Maybe come up with a better name? |
617 | return OW_PartialEarlierWithFullLater; |
618 | } |
619 | |
620 | // Another interesting case is if the later store overwrites the end of the |
621 | // earlier store. |
622 | // |
623 | // |--earlier--| |
624 | // |-- later --| |
625 | // |
626 | // In this case we may want to trim the size of earlier to avoid generating |
627 | // writes to addresses which will definitely be overwritten later |
628 | if (!EnablePartialOverwriteTracking && |
629 | (LaterOff > EarlierOff && LaterOff < int64_t(EarlierOff + EarlierSize) && |
630 | int64_t(LaterOff + LaterSize) >= int64_t(EarlierOff + EarlierSize))) |
631 | return OW_End; |
632 | |
633 | // Finally, we also need to check if the later store overwrites the beginning |
634 | // of the earlier store. |
635 | // |
636 | // |--earlier--| |
637 | // |-- later --| |
638 | // |
639 | // In this case we may want to move the destination address and trim the size |
640 | // of earlier to avoid generating writes to addresses which will definitely |
641 | // be overwritten later. |
642 | if (!EnablePartialOverwriteTracking && |
643 | (LaterOff <= EarlierOff && int64_t(LaterOff + LaterSize) > EarlierOff)) { |
644 | assert(int64_t(LaterOff + LaterSize) < int64_t(EarlierOff + EarlierSize) &&((int64_t(LaterOff + LaterSize) < int64_t(EarlierOff + EarlierSize ) && "Expect to be handled as OW_Complete") ? static_cast <void> (0) : __assert_fail ("int64_t(LaterOff + LaterSize) < int64_t(EarlierOff + EarlierSize) && \"Expect to be handled as OW_Complete\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 645, __PRETTY_FUNCTION__)) |
645 | "Expect to be handled as OW_Complete")((int64_t(LaterOff + LaterSize) < int64_t(EarlierOff + EarlierSize ) && "Expect to be handled as OW_Complete") ? static_cast <void> (0) : __assert_fail ("int64_t(LaterOff + LaterSize) < int64_t(EarlierOff + EarlierSize) && \"Expect to be handled as OW_Complete\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 645, __PRETTY_FUNCTION__)); |
646 | return OW_Begin; |
647 | } |
648 | // Otherwise, they don't completely overlap. |
649 | return OW_Unknown; |
650 | } |
651 | |
652 | /// If 'Inst' might be a self read (i.e. a noop copy of a |
653 | /// memory region into an identical pointer) then it doesn't actually make its |
654 | /// input dead in the traditional sense. Consider this case: |
655 | /// |
656 | /// memmove(A <- B) |
657 | /// memmove(A <- A) |
658 | /// |
659 | /// In this case, the second store to A does not make the first store to A dead. |
660 | /// The usual situation isn't an explicit A<-A store like this (which can be |
661 | /// trivially removed) but a case where two pointers may alias. |
662 | /// |
663 | /// This function detects when it is unsafe to remove a dependent instruction |
664 | /// because the DSE inducing instruction may be a self-read. |
665 | static bool isPossibleSelfRead(Instruction *Inst, |
666 | const MemoryLocation &InstStoreLoc, |
667 | Instruction *DepWrite, |
668 | const TargetLibraryInfo &TLI, |
669 | AliasAnalysis &AA) { |
670 | // Self reads can only happen for instructions that read memory. Get the |
671 | // location read. |
672 | MemoryLocation InstReadLoc = getLocForRead(Inst, TLI); |
673 | if (!InstReadLoc.Ptr) |
674 | return false; // Not a reading instruction. |
675 | |
676 | // If the read and written loc obviously don't alias, it isn't a read. |
677 | if (AA.isNoAlias(InstReadLoc, InstStoreLoc)) |
678 | return false; |
679 | |
680 | if (isa<AnyMemCpyInst>(Inst)) { |
681 | // LLVM's memcpy overlap semantics are not fully fleshed out (see PR11763) |
682 | // but in practice memcpy(A <- B) either means that A and B are disjoint or |
683 | // are equal (i.e. there are not partial overlaps). Given that, if we have: |
684 | // |
685 | // memcpy/memmove(A <- B) // DepWrite |
686 | // memcpy(A <- B) // Inst |
687 | // |
688 | // with Inst reading/writing a >= size than DepWrite, we can reason as |
689 | // follows: |
690 | // |
691 | // - If A == B then both the copies are no-ops, so the DepWrite can be |
692 | // removed. |
693 | // - If A != B then A and B are disjoint locations in Inst. Since |
694 | // Inst.size >= DepWrite.size A and B are disjoint in DepWrite too. |
695 | // Therefore DepWrite can be removed. |
696 | MemoryLocation DepReadLoc = getLocForRead(DepWrite, TLI); |
697 | |
698 | if (DepReadLoc.Ptr && AA.isMustAlias(InstReadLoc.Ptr, DepReadLoc.Ptr)) |
699 | return false; |
700 | } |
701 | |
702 | // If DepWrite doesn't read memory or if we can't prove it is a must alias, |
703 | // then it can't be considered dead. |
704 | return true; |
705 | } |
706 | |
707 | /// Returns true if the memory which is accessed by the second instruction is not |
708 | /// modified between the first and the second instruction. |
709 | /// Precondition: Second instruction must be dominated by the first |
710 | /// instruction. |
711 | template <typename AATy> |
712 | static bool |
713 | memoryIsNotModifiedBetween(Instruction *FirstI, Instruction *SecondI, AATy &AA, |
714 | const DataLayout &DL, DominatorTree *DT) { |
715 | // Do a backwards scan through the CFG from SecondI to FirstI. Look for |
716 | // instructions which can modify the memory location accessed by SecondI. |
717 | // |
718 | // While doing the walk keep track of the address to check. It might be |
719 | // different in different basic blocks due to PHI translation. |
720 | using BlockAddressPair = std::pair<BasicBlock *, PHITransAddr>; |
721 | SmallVector<BlockAddressPair, 16> WorkList; |
722 | // Keep track of the address we visited each block with. Bail out if we |
723 | // visit a block with different addresses. |
724 | DenseMap<BasicBlock *, Value *> Visited; |
725 | |
726 | BasicBlock::iterator FirstBBI(FirstI); |
727 | ++FirstBBI; |
728 | BasicBlock::iterator SecondBBI(SecondI); |
729 | BasicBlock *FirstBB = FirstI->getParent(); |
730 | BasicBlock *SecondBB = SecondI->getParent(); |
731 | MemoryLocation MemLoc = MemoryLocation::get(SecondI); |
732 | auto *MemLocPtr = const_cast<Value *>(MemLoc.Ptr); |
733 | |
734 | // Start checking the SecondBB. |
735 | WorkList.push_back( |
736 | std::make_pair(SecondBB, PHITransAddr(MemLocPtr, DL, nullptr))); |
737 | bool isFirstBlock = true; |
738 | |
739 | // Check all blocks going backward until we reach the FirstBB. |
740 | while (!WorkList.empty()) { |
741 | BlockAddressPair Current = WorkList.pop_back_val(); |
742 | BasicBlock *B = Current.first; |
743 | PHITransAddr &Addr = Current.second; |
744 | Value *Ptr = Addr.getAddr(); |
745 | |
746 | // Ignore instructions before FirstI if this is the FirstBB. |
747 | BasicBlock::iterator BI = (B == FirstBB ? FirstBBI : B->begin()); |
748 | |
749 | BasicBlock::iterator EI; |
750 | if (isFirstBlock) { |
751 | // Ignore instructions after SecondI if this is the first visit of SecondBB. |
752 | assert(B == SecondBB && "first block is not the store block")((B == SecondBB && "first block is not the store block" ) ? static_cast<void> (0) : __assert_fail ("B == SecondBB && \"first block is not the store block\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 752, __PRETTY_FUNCTION__)); |
753 | EI = SecondBBI; |
754 | isFirstBlock = false; |
755 | } else { |
756 | // It's not SecondBB or (in case of a loop) the second visit of SecondBB. |
757 | // In this case we also have to look at instructions after SecondI. |
758 | EI = B->end(); |
759 | } |
760 | for (; BI != EI; ++BI) { |
761 | Instruction *I = &*BI; |
762 | if (I->mayWriteToMemory() && I != SecondI) |
763 | if (isModSet(AA.getModRefInfo(I, MemLoc.getWithNewPtr(Ptr)))) |
764 | return false; |
765 | } |
766 | if (B != FirstBB) { |
767 | assert(B != &FirstBB->getParent()->getEntryBlock() &&((B != &FirstBB->getParent()->getEntryBlock() && "Should not hit the entry block because SI must be dominated by LI" ) ? static_cast<void> (0) : __assert_fail ("B != &FirstBB->getParent()->getEntryBlock() && \"Should not hit the entry block because SI must be dominated by LI\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 768, __PRETTY_FUNCTION__)) |
768 | "Should not hit the entry block because SI must be dominated by LI")((B != &FirstBB->getParent()->getEntryBlock() && "Should not hit the entry block because SI must be dominated by LI" ) ? static_cast<void> (0) : __assert_fail ("B != &FirstBB->getParent()->getEntryBlock() && \"Should not hit the entry block because SI must be dominated by LI\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 768, __PRETTY_FUNCTION__)); |
769 | for (auto PredI = pred_begin(B), PE = pred_end(B); PredI != PE; ++PredI) { |
770 | PHITransAddr PredAddr = Addr; |
771 | if (PredAddr.NeedsPHITranslationFromBlock(B)) { |
772 | if (!PredAddr.IsPotentiallyPHITranslatable()) |
773 | return false; |
774 | if (PredAddr.PHITranslateValue(B, *PredI, DT, false)) |
775 | return false; |
776 | } |
777 | Value *TranslatedPtr = PredAddr.getAddr(); |
778 | auto Inserted = Visited.insert(std::make_pair(*PredI, TranslatedPtr)); |
779 | if (!Inserted.second) { |
780 | // We already visited this block before. If it was with a different |
781 | // address - bail out! |
782 | if (TranslatedPtr != Inserted.first->second) |
783 | return false; |
784 | // ... otherwise just skip it. |
785 | continue; |
786 | } |
787 | WorkList.push_back(std::make_pair(*PredI, PredAddr)); |
788 | } |
789 | } |
790 | } |
791 | return true; |
792 | } |
793 | |
794 | /// Find all blocks that will unconditionally lead to the block BB and append |
795 | /// them to F. |
796 | static void findUnconditionalPreds(SmallVectorImpl<BasicBlock *> &Blocks, |
797 | BasicBlock *BB, DominatorTree *DT) { |
798 | for (pred_iterator I = pred_begin(BB), E = pred_end(BB); I != E; ++I) { |
799 | BasicBlock *Pred = *I; |
800 | if (Pred == BB) continue; |
801 | Instruction *PredTI = Pred->getTerminator(); |
802 | if (PredTI->getNumSuccessors() != 1) |
803 | continue; |
804 | |
805 | if (DT->isReachableFromEntry(Pred)) |
806 | Blocks.push_back(Pred); |
807 | } |
808 | } |
809 | |
810 | /// Handle frees of entire structures whose dependency is a store |
811 | /// to a field of that structure. |
812 | static bool handleFree(CallInst *F, AliasAnalysis *AA, |
813 | MemoryDependenceResults *MD, DominatorTree *DT, |
814 | const TargetLibraryInfo *TLI, |
815 | InstOverlapIntervalsTy &IOL, |
816 | MapVector<Instruction *, bool> &ThrowableInst) { |
817 | bool MadeChange = false; |
818 | |
819 | MemoryLocation Loc = MemoryLocation(F->getOperand(0)); |
820 | SmallVector<BasicBlock *, 16> Blocks; |
821 | Blocks.push_back(F->getParent()); |
822 | |
823 | while (!Blocks.empty()) { |
824 | BasicBlock *BB = Blocks.pop_back_val(); |
825 | Instruction *InstPt = BB->getTerminator(); |
826 | if (BB == F->getParent()) InstPt = F; |
827 | |
828 | MemDepResult Dep = |
829 | MD->getPointerDependencyFrom(Loc, false, InstPt->getIterator(), BB); |
830 | while (Dep.isDef() || Dep.isClobber()) { |
831 | Instruction *Dependency = Dep.getInst(); |
832 | if (!hasAnalyzableMemoryWrite(Dependency, *TLI) || |
833 | !isRemovable(Dependency)) |
834 | break; |
835 | |
836 | Value *DepPointer = |
837 | getUnderlyingObject(getStoredPointerOperand(Dependency, *TLI)); |
838 | |
839 | // Check for aliasing. |
840 | if (!AA->isMustAlias(F->getArgOperand(0), DepPointer)) |
841 | break; |
842 | |
843 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store to soon to be freed memory:\n DEAD: " << *Dependency << '\n'; } } while (false) |
844 | dbgs() << "DSE: Dead Store to soon to be freed memory:\n DEAD: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store to soon to be freed memory:\n DEAD: " << *Dependency << '\n'; } } while (false) |
845 | << *Dependency << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store to soon to be freed memory:\n DEAD: " << *Dependency << '\n'; } } while (false); |
846 | |
847 | // DCE instructions only used to calculate that store. |
848 | BasicBlock::iterator BBI(Dependency); |
849 | deleteDeadInstruction(Dependency, &BBI, *MD, *TLI, IOL, |
850 | ThrowableInst); |
851 | ++NumFastStores; |
852 | MadeChange = true; |
853 | |
854 | // Inst's old Dependency is now deleted. Compute the next dependency, |
855 | // which may also be dead, as in |
856 | // s[0] = 0; |
857 | // s[1] = 0; // This has just been deleted. |
858 | // free(s); |
859 | Dep = MD->getPointerDependencyFrom(Loc, false, BBI, BB); |
860 | } |
861 | |
862 | if (Dep.isNonLocal()) |
863 | findUnconditionalPreds(Blocks, BB, DT); |
864 | } |
865 | |
866 | return MadeChange; |
867 | } |
868 | |
869 | /// Check to see if the specified location may alias any of the stack objects in |
870 | /// the DeadStackObjects set. If so, they become live because the location is |
871 | /// being loaded. |
872 | static void removeAccessedObjects(const MemoryLocation &LoadedLoc, |
873 | SmallSetVector<const Value *, 16> &DeadStackObjects, |
874 | const DataLayout &DL, AliasAnalysis *AA, |
875 | const TargetLibraryInfo *TLI, |
876 | const Function *F) { |
877 | const Value *UnderlyingPointer = getUnderlyingObject(LoadedLoc.Ptr); |
878 | |
879 | // A constant can't be in the dead pointer set. |
880 | if (isa<Constant>(UnderlyingPointer)) |
881 | return; |
882 | |
883 | // If the kill pointer can be easily reduced to an alloca, don't bother doing |
884 | // extraneous AA queries. |
885 | if (isa<AllocaInst>(UnderlyingPointer) || isa<Argument>(UnderlyingPointer)) { |
886 | DeadStackObjects.remove(UnderlyingPointer); |
887 | return; |
888 | } |
889 | |
890 | // Remove objects that could alias LoadedLoc. |
891 | DeadStackObjects.remove_if([&](const Value *I) { |
892 | // See if the loaded location could alias the stack location. |
893 | MemoryLocation StackLoc(I, getPointerSize(I, DL, *TLI, F)); |
894 | return !AA->isNoAlias(StackLoc, LoadedLoc); |
895 | }); |
896 | } |
897 | |
898 | /// Remove dead stores to stack-allocated locations in the function end block. |
899 | /// Ex: |
900 | /// %A = alloca i32 |
901 | /// ... |
902 | /// store i32 1, i32* %A |
903 | /// ret void |
904 | static bool handleEndBlock(BasicBlock &BB, AliasAnalysis *AA, |
905 | MemoryDependenceResults *MD, |
906 | const TargetLibraryInfo *TLI, |
907 | InstOverlapIntervalsTy &IOL, |
908 | MapVector<Instruction *, bool> &ThrowableInst) { |
909 | bool MadeChange = false; |
910 | |
911 | // Keep track of all of the stack objects that are dead at the end of the |
912 | // function. |
913 | SmallSetVector<const Value*, 16> DeadStackObjects; |
914 | |
915 | // Find all of the alloca'd pointers in the entry block. |
916 | BasicBlock &Entry = BB.getParent()->front(); |
917 | for (Instruction &I : Entry) { |
918 | if (isa<AllocaInst>(&I)) |
919 | DeadStackObjects.insert(&I); |
920 | |
921 | // Okay, so these are dead heap objects, but if the pointer never escapes |
922 | // then it's leaked by this function anyways. |
923 | else if (isAllocLikeFn(&I, TLI) && !PointerMayBeCaptured(&I, true, true)) |
924 | DeadStackObjects.insert(&I); |
925 | } |
926 | |
927 | // Treat byval or inalloca arguments the same, stores to them are dead at the |
928 | // end of the function. |
929 | for (Argument &AI : BB.getParent()->args()) |
930 | if (AI.hasPassPointeeByValueCopyAttr()) |
931 | DeadStackObjects.insert(&AI); |
932 | |
933 | const DataLayout &DL = BB.getModule()->getDataLayout(); |
934 | |
935 | // Scan the basic block backwards |
936 | for (BasicBlock::iterator BBI = BB.end(); BBI != BB.begin(); ){ |
937 | --BBI; |
938 | |
939 | // If we find a store, check to see if it points into a dead stack value. |
940 | if (hasAnalyzableMemoryWrite(&*BBI, *TLI) && isRemovable(&*BBI)) { |
941 | // See through pointer-to-pointer bitcasts |
942 | SmallVector<const Value *, 4> Pointers; |
943 | getUnderlyingObjects(getStoredPointerOperand(&*BBI, *TLI), Pointers); |
944 | |
945 | // Stores to stack values are valid candidates for removal. |
946 | bool AllDead = true; |
947 | for (const Value *Pointer : Pointers) |
948 | if (!DeadStackObjects.count(Pointer)) { |
949 | AllDead = false; |
950 | break; |
951 | } |
952 | |
953 | if (AllDead) { |
954 | Instruction *Dead = &*BBI; |
955 | |
956 | LLVM_DEBUG(dbgs() << "DSE: Dead Store at End of Block:\n DEAD: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
957 | << *Dead << "\n Objects: ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
958 | for (SmallVectorImpl<const Value *>::iterator I =do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
959 | Pointers.begin(),do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
960 | E = Pointers.end();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
961 | I != E; ++I) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
962 | dbgs() << **I;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
963 | if (std::next(I) != E)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
964 | dbgs() << ", ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
965 | } dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false) |
966 | << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " << *Dead << "\n Objects: "; for (SmallVectorImpl <const Value *>::iterator I = Pointers.begin(), E = Pointers .end(); I != E; ++I) { dbgs() << **I; if (std::next(I) != E) dbgs() << ", "; } dbgs() << '\n'; } } while ( false); |
967 | |
968 | // DCE instructions only used to calculate that store. |
969 | deleteDeadInstruction(Dead, &BBI, *MD, *TLI, IOL, ThrowableInst, |
970 | &DeadStackObjects); |
971 | ++NumFastStores; |
972 | MadeChange = true; |
973 | continue; |
974 | } |
975 | } |
976 | |
977 | // Remove any dead non-memory-mutating instructions. |
978 | if (isInstructionTriviallyDead(&*BBI, TLI)) { |
979 | LLVM_DEBUG(dbgs() << "DSE: Removing trivially dead instruction:\n DEAD: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Removing trivially dead instruction:\n DEAD: " << *&*BBI << '\n'; } } while (false) |
980 | << *&*BBI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Removing trivially dead instruction:\n DEAD: " << *&*BBI << '\n'; } } while (false); |
981 | deleteDeadInstruction(&*BBI, &BBI, *MD, *TLI, IOL, ThrowableInst, |
982 | &DeadStackObjects); |
983 | ++NumFastOther; |
984 | MadeChange = true; |
985 | continue; |
986 | } |
987 | |
988 | if (isa<AllocaInst>(BBI)) { |
989 | // Remove allocas from the list of dead stack objects; there can't be |
990 | // any references before the definition. |
991 | DeadStackObjects.remove(&*BBI); |
992 | continue; |
993 | } |
994 | |
995 | if (auto *Call = dyn_cast<CallBase>(&*BBI)) { |
996 | // Remove allocation function calls from the list of dead stack objects; |
997 | // there can't be any references before the definition. |
998 | if (isAllocLikeFn(&*BBI, TLI)) |
999 | DeadStackObjects.remove(&*BBI); |
1000 | |
1001 | // If this call does not access memory, it can't be loading any of our |
1002 | // pointers. |
1003 | if (AA->doesNotAccessMemory(Call)) |
1004 | continue; |
1005 | |
1006 | // If the call might load from any of our allocas, then any store above |
1007 | // the call is live. |
1008 | DeadStackObjects.remove_if([&](const Value *I) { |
1009 | // See if the call site touches the value. |
1010 | return isRefSet(AA->getModRefInfo( |
1011 | Call, I, getPointerSize(I, DL, *TLI, BB.getParent()))); |
1012 | }); |
1013 | |
1014 | // If all of the allocas were clobbered by the call then we're not going |
1015 | // to find anything else to process. |
1016 | if (DeadStackObjects.empty()) |
1017 | break; |
1018 | |
1019 | continue; |
1020 | } |
1021 | |
1022 | // We can remove the dead stores, irrespective of the fence and its ordering |
1023 | // (release/acquire/seq_cst). Fences only constraints the ordering of |
1024 | // already visible stores, it does not make a store visible to other |
1025 | // threads. So, skipping over a fence does not change a store from being |
1026 | // dead. |
1027 | if (isa<FenceInst>(*BBI)) |
1028 | continue; |
1029 | |
1030 | MemoryLocation LoadedLoc; |
1031 | |
1032 | // If we encounter a use of the pointer, it is no longer considered dead |
1033 | if (LoadInst *L = dyn_cast<LoadInst>(BBI)) { |
1034 | if (!L->isUnordered()) // Be conservative with atomic/volatile load |
1035 | break; |
1036 | LoadedLoc = MemoryLocation::get(L); |
1037 | } else if (VAArgInst *V = dyn_cast<VAArgInst>(BBI)) { |
1038 | LoadedLoc = MemoryLocation::get(V); |
1039 | } else if (!BBI->mayReadFromMemory()) { |
1040 | // Instruction doesn't read memory. Note that stores that weren't removed |
1041 | // above will hit this case. |
1042 | continue; |
1043 | } else { |
1044 | // Unknown inst; assume it clobbers everything. |
1045 | break; |
1046 | } |
1047 | |
1048 | // Remove any allocas from the DeadPointer set that are loaded, as this |
1049 | // makes any stores above the access live. |
1050 | removeAccessedObjects(LoadedLoc, DeadStackObjects, DL, AA, TLI, BB.getParent()); |
1051 | |
1052 | // If all of the allocas were clobbered by the access then we're not going |
1053 | // to find anything else to process. |
1054 | if (DeadStackObjects.empty()) |
1055 | break; |
1056 | } |
1057 | |
1058 | return MadeChange; |
1059 | } |
1060 | |
1061 | static bool tryToShorten(Instruction *EarlierWrite, int64_t &EarlierOffset, |
1062 | int64_t &EarlierSize, int64_t LaterOffset, |
1063 | int64_t LaterSize, bool IsOverwriteEnd) { |
1064 | // TODO: base this on the target vector size so that if the earlier |
1065 | // store was too small to get vector writes anyway then its likely |
1066 | // a good idea to shorten it |
1067 | // Power of 2 vector writes are probably always a bad idea to optimize |
1068 | // as any store/memset/memcpy is likely using vector instructions so |
1069 | // shortening it to not vector size is likely to be slower |
1070 | auto *EarlierIntrinsic = cast<AnyMemIntrinsic>(EarlierWrite); |
1071 | unsigned EarlierWriteAlign = EarlierIntrinsic->getDestAlignment(); |
1072 | if (!IsOverwriteEnd) |
1073 | LaterOffset = int64_t(LaterOffset + LaterSize); |
1074 | |
1075 | if (!(isPowerOf2_64(LaterOffset) && EarlierWriteAlign <= LaterOffset) && |
1076 | !((EarlierWriteAlign != 0) && LaterOffset % EarlierWriteAlign == 0)) |
1077 | return false; |
1078 | |
1079 | int64_t NewLength = IsOverwriteEnd |
1080 | ? LaterOffset - EarlierOffset |
1081 | : EarlierSize - (LaterOffset - EarlierOffset); |
1082 | |
1083 | if (auto *AMI = dyn_cast<AtomicMemIntrinsic>(EarlierWrite)) { |
1084 | // When shortening an atomic memory intrinsic, the newly shortened |
1085 | // length must remain an integer multiple of the element size. |
1086 | const uint32_t ElementSize = AMI->getElementSizeInBytes(); |
1087 | if (0 != NewLength % ElementSize) |
1088 | return false; |
1089 | } |
1090 | |
1091 | LLVM_DEBUG(dbgs() << "DSE: Remove Dead Store:\n OW "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n OW " << (IsOverwriteEnd ? "END" : "BEGIN") << ": " << *EarlierWrite << "\n KILLER (offset " << LaterOffset << ", " << EarlierSize << ")\n"; } } while (false) |
1092 | << (IsOverwriteEnd ? "END" : "BEGIN") << ": "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n OW " << (IsOverwriteEnd ? "END" : "BEGIN") << ": " << *EarlierWrite << "\n KILLER (offset " << LaterOffset << ", " << EarlierSize << ")\n"; } } while (false) |
1093 | << *EarlierWrite << "\n KILLER (offset " << LaterOffsetdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n OW " << (IsOverwriteEnd ? "END" : "BEGIN") << ": " << *EarlierWrite << "\n KILLER (offset " << LaterOffset << ", " << EarlierSize << ")\n"; } } while (false) |
1094 | << ", " << EarlierSize << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n OW " << (IsOverwriteEnd ? "END" : "BEGIN") << ": " << *EarlierWrite << "\n KILLER (offset " << LaterOffset << ", " << EarlierSize << ")\n"; } } while (false); |
1095 | |
1096 | Value *EarlierWriteLength = EarlierIntrinsic->getLength(); |
1097 | Value *TrimmedLength = |
1098 | ConstantInt::get(EarlierWriteLength->getType(), NewLength); |
1099 | EarlierIntrinsic->setLength(TrimmedLength); |
1100 | |
1101 | EarlierSize = NewLength; |
1102 | if (!IsOverwriteEnd) { |
1103 | int64_t OffsetMoved = (LaterOffset - EarlierOffset); |
1104 | Value *Indices[1] = { |
1105 | ConstantInt::get(EarlierWriteLength->getType(), OffsetMoved)}; |
1106 | GetElementPtrInst *NewDestGEP = GetElementPtrInst::CreateInBounds( |
1107 | EarlierIntrinsic->getRawDest()->getType()->getPointerElementType(), |
1108 | EarlierIntrinsic->getRawDest(), Indices, "", EarlierWrite); |
1109 | NewDestGEP->setDebugLoc(EarlierIntrinsic->getDebugLoc()); |
1110 | EarlierIntrinsic->setDest(NewDestGEP); |
1111 | EarlierOffset = EarlierOffset + OffsetMoved; |
1112 | } |
1113 | return true; |
1114 | } |
1115 | |
1116 | static bool tryToShortenEnd(Instruction *EarlierWrite, |
1117 | OverlapIntervalsTy &IntervalMap, |
1118 | int64_t &EarlierStart, int64_t &EarlierSize) { |
1119 | if (IntervalMap.empty() || !isShortenableAtTheEnd(EarlierWrite)) |
1120 | return false; |
1121 | |
1122 | OverlapIntervalsTy::iterator OII = --IntervalMap.end(); |
1123 | int64_t LaterStart = OII->second; |
1124 | int64_t LaterSize = OII->first - LaterStart; |
1125 | |
1126 | if (LaterStart > EarlierStart && LaterStart < EarlierStart + EarlierSize && |
1127 | LaterStart + LaterSize >= EarlierStart + EarlierSize) { |
1128 | if (tryToShorten(EarlierWrite, EarlierStart, EarlierSize, LaterStart, |
1129 | LaterSize, true)) { |
1130 | IntervalMap.erase(OII); |
1131 | return true; |
1132 | } |
1133 | } |
1134 | return false; |
1135 | } |
1136 | |
1137 | static bool tryToShortenBegin(Instruction *EarlierWrite, |
1138 | OverlapIntervalsTy &IntervalMap, |
1139 | int64_t &EarlierStart, int64_t &EarlierSize) { |
1140 | if (IntervalMap.empty() || !isShortenableAtTheBeginning(EarlierWrite)) |
1141 | return false; |
1142 | |
1143 | OverlapIntervalsTy::iterator OII = IntervalMap.begin(); |
1144 | int64_t LaterStart = OII->second; |
1145 | int64_t LaterSize = OII->first - LaterStart; |
1146 | |
1147 | if (LaterStart <= EarlierStart && LaterStart + LaterSize > EarlierStart) { |
1148 | assert(LaterStart + LaterSize < EarlierStart + EarlierSize &&((LaterStart + LaterSize < EarlierStart + EarlierSize && "Should have been handled as OW_Complete") ? static_cast< void> (0) : __assert_fail ("LaterStart + LaterSize < EarlierStart + EarlierSize && \"Should have been handled as OW_Complete\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1149, __PRETTY_FUNCTION__)) |
1149 | "Should have been handled as OW_Complete")((LaterStart + LaterSize < EarlierStart + EarlierSize && "Should have been handled as OW_Complete") ? static_cast< void> (0) : __assert_fail ("LaterStart + LaterSize < EarlierStart + EarlierSize && \"Should have been handled as OW_Complete\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1149, __PRETTY_FUNCTION__)); |
1150 | if (tryToShorten(EarlierWrite, EarlierStart, EarlierSize, LaterStart, |
1151 | LaterSize, false)) { |
1152 | IntervalMap.erase(OII); |
1153 | return true; |
1154 | } |
1155 | } |
1156 | return false; |
1157 | } |
1158 | |
1159 | static bool removePartiallyOverlappedStores(const DataLayout &DL, |
1160 | InstOverlapIntervalsTy &IOL, |
1161 | const TargetLibraryInfo &TLI) { |
1162 | bool Changed = false; |
1163 | for (auto OI : IOL) { |
1164 | Instruction *EarlierWrite = OI.first; |
1165 | MemoryLocation Loc = getLocForWrite(EarlierWrite, TLI); |
1166 | assert(isRemovable(EarlierWrite) && "Expect only removable instruction")((isRemovable(EarlierWrite) && "Expect only removable instruction" ) ? static_cast<void> (0) : __assert_fail ("isRemovable(EarlierWrite) && \"Expect only removable instruction\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1166, __PRETTY_FUNCTION__)); |
1167 | |
1168 | const Value *Ptr = Loc.Ptr->stripPointerCasts(); |
1169 | int64_t EarlierStart = 0; |
1170 | int64_t EarlierSize = int64_t(Loc.Size.getValue()); |
1171 | GetPointerBaseWithConstantOffset(Ptr, EarlierStart, DL); |
1172 | OverlapIntervalsTy &IntervalMap = OI.second; |
1173 | Changed |= |
1174 | tryToShortenEnd(EarlierWrite, IntervalMap, EarlierStart, EarlierSize); |
1175 | if (IntervalMap.empty()) |
1176 | continue; |
1177 | Changed |= |
1178 | tryToShortenBegin(EarlierWrite, IntervalMap, EarlierStart, EarlierSize); |
1179 | } |
1180 | return Changed; |
1181 | } |
1182 | |
1183 | static bool eliminateNoopStore(Instruction *Inst, BasicBlock::iterator &BBI, |
1184 | AliasAnalysis *AA, MemoryDependenceResults *MD, |
1185 | const DataLayout &DL, |
1186 | const TargetLibraryInfo *TLI, |
1187 | InstOverlapIntervalsTy &IOL, |
1188 | MapVector<Instruction *, bool> &ThrowableInst, |
1189 | DominatorTree *DT) { |
1190 | // Must be a store instruction. |
1191 | StoreInst *SI = dyn_cast<StoreInst>(Inst); |
1192 | if (!SI) |
1193 | return false; |
1194 | |
1195 | // If we're storing the same value back to a pointer that we just loaded from, |
1196 | // then the store can be removed. |
1197 | if (LoadInst *DepLoad = dyn_cast<LoadInst>(SI->getValueOperand())) { |
1198 | if (SI->getPointerOperand() == DepLoad->getPointerOperand() && |
1199 | isRemovable(SI) && |
1200 | memoryIsNotModifiedBetween(DepLoad, SI, *AA, DL, DT)) { |
1201 | |
1202 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Store Of Load from same pointer:\n LOAD: " << *DepLoad << "\n STORE: " << *SI << '\n'; } } while (false) |
1203 | dbgs() << "DSE: Remove Store Of Load from same pointer:\n LOAD: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Store Of Load from same pointer:\n LOAD: " << *DepLoad << "\n STORE: " << *SI << '\n'; } } while (false) |
1204 | << *DepLoad << "\n STORE: " << *SI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Store Of Load from same pointer:\n LOAD: " << *DepLoad << "\n STORE: " << *SI << '\n'; } } while (false); |
1205 | |
1206 | deleteDeadInstruction(SI, &BBI, *MD, *TLI, IOL, ThrowableInst); |
1207 | ++NumRedundantStores; |
1208 | return true; |
1209 | } |
1210 | } |
1211 | |
1212 | // Remove null stores into the calloc'ed objects |
1213 | Constant *StoredConstant = dyn_cast<Constant>(SI->getValueOperand()); |
1214 | if (StoredConstant && StoredConstant->isNullValue() && isRemovable(SI)) { |
1215 | Instruction *UnderlyingPointer = |
1216 | dyn_cast<Instruction>(getUnderlyingObject(SI->getPointerOperand())); |
1217 | |
1218 | if (UnderlyingPointer && isCallocLikeFn(UnderlyingPointer, TLI) && |
1219 | memoryIsNotModifiedBetween(UnderlyingPointer, SI, *AA, DL, DT)) { |
1220 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove null store to the calloc'ed object:\n DEAD: " << *Inst << "\n OBJECT: " << *UnderlyingPointer << '\n'; } } while (false) |
1221 | dbgs() << "DSE: Remove null store to the calloc'ed object:\n DEAD: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove null store to the calloc'ed object:\n DEAD: " << *Inst << "\n OBJECT: " << *UnderlyingPointer << '\n'; } } while (false) |
1222 | << *Inst << "\n OBJECT: " << *UnderlyingPointer << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove null store to the calloc'ed object:\n DEAD: " << *Inst << "\n OBJECT: " << *UnderlyingPointer << '\n'; } } while (false); |
1223 | |
1224 | deleteDeadInstruction(SI, &BBI, *MD, *TLI, IOL, ThrowableInst); |
1225 | ++NumRedundantStores; |
1226 | return true; |
1227 | } |
1228 | } |
1229 | return false; |
1230 | } |
1231 | |
1232 | template <typename AATy> |
1233 | static Constant *tryToMergePartialOverlappingStores( |
1234 | StoreInst *Earlier, StoreInst *Later, int64_t InstWriteOffset, |
1235 | int64_t DepWriteOffset, const DataLayout &DL, AATy &AA, DominatorTree *DT) { |
1236 | |
1237 | if (Earlier && isa<ConstantInt>(Earlier->getValueOperand()) && |
1238 | DL.typeSizeEqualsStoreSize(Earlier->getValueOperand()->getType()) && |
1239 | Later && isa<ConstantInt>(Later->getValueOperand()) && |
1240 | DL.typeSizeEqualsStoreSize(Later->getValueOperand()->getType()) && |
1241 | memoryIsNotModifiedBetween(Earlier, Later, AA, DL, DT)) { |
1242 | // If the store we find is: |
1243 | // a) partially overwritten by the store to 'Loc' |
1244 | // b) the later store is fully contained in the earlier one and |
1245 | // c) they both have a constant value |
1246 | // d) none of the two stores need padding |
1247 | // Merge the two stores, replacing the earlier store's value with a |
1248 | // merge of both values. |
1249 | // TODO: Deal with other constant types (vectors, etc), and probably |
1250 | // some mem intrinsics (if needed) |
1251 | |
1252 | APInt EarlierValue = |
1253 | cast<ConstantInt>(Earlier->getValueOperand())->getValue(); |
1254 | APInt LaterValue = cast<ConstantInt>(Later->getValueOperand())->getValue(); |
1255 | unsigned LaterBits = LaterValue.getBitWidth(); |
1256 | assert(EarlierValue.getBitWidth() > LaterValue.getBitWidth())((EarlierValue.getBitWidth() > LaterValue.getBitWidth()) ? static_cast<void> (0) : __assert_fail ("EarlierValue.getBitWidth() > LaterValue.getBitWidth()" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1256, __PRETTY_FUNCTION__)); |
1257 | LaterValue = LaterValue.zext(EarlierValue.getBitWidth()); |
1258 | |
1259 | // Offset of the smaller store inside the larger store |
1260 | unsigned BitOffsetDiff = (InstWriteOffset - DepWriteOffset) * 8; |
1261 | unsigned LShiftAmount = DL.isBigEndian() ? EarlierValue.getBitWidth() - |
1262 | BitOffsetDiff - LaterBits |
1263 | : BitOffsetDiff; |
1264 | APInt Mask = APInt::getBitsSet(EarlierValue.getBitWidth(), LShiftAmount, |
1265 | LShiftAmount + LaterBits); |
1266 | // Clear the bits we'll be replacing, then OR with the smaller |
1267 | // store, shifted appropriately. |
1268 | APInt Merged = (EarlierValue & ~Mask) | (LaterValue << LShiftAmount); |
1269 | LLVM_DEBUG(dbgs() << "DSE: Merge Stores:\n Earlier: " << *Earlierdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Merge Stores:\n Earlier: " << *Earlier << "\n Later: " << *Later << "\n Merged Value: " << Merged << '\n'; } } while (false) |
1270 | << "\n Later: " << *Laterdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Merge Stores:\n Earlier: " << *Earlier << "\n Later: " << *Later << "\n Merged Value: " << Merged << '\n'; } } while (false) |
1271 | << "\n Merged Value: " << Merged << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Merge Stores:\n Earlier: " << *Earlier << "\n Later: " << *Later << "\n Merged Value: " << Merged << '\n'; } } while (false); |
1272 | return ConstantInt::get(Earlier->getValueOperand()->getType(), Merged); |
1273 | } |
1274 | return nullptr; |
1275 | } |
1276 | |
1277 | static bool eliminateDeadStores(BasicBlock &BB, AliasAnalysis *AA, |
1278 | MemoryDependenceResults *MD, DominatorTree *DT, |
1279 | const TargetLibraryInfo *TLI) { |
1280 | const DataLayout &DL = BB.getModule()->getDataLayout(); |
1281 | bool MadeChange = false; |
1282 | |
1283 | MapVector<Instruction *, bool> ThrowableInst; |
1284 | |
1285 | // A map of interval maps representing partially-overwritten value parts. |
1286 | InstOverlapIntervalsTy IOL; |
1287 | |
1288 | // Do a top-down walk on the BB. |
1289 | for (BasicBlock::iterator BBI = BB.begin(), BBE = BB.end(); BBI != BBE; ) { |
1290 | // Handle 'free' calls specially. |
1291 | if (CallInst *F = isFreeCall(&*BBI, TLI)) { |
1292 | MadeChange |= handleFree(F, AA, MD, DT, TLI, IOL, ThrowableInst); |
1293 | // Increment BBI after handleFree has potentially deleted instructions. |
1294 | // This ensures we maintain a valid iterator. |
1295 | ++BBI; |
1296 | continue; |
1297 | } |
1298 | |
1299 | Instruction *Inst = &*BBI++; |
1300 | |
1301 | if (Inst->mayThrow()) { |
1302 | ThrowableInst[Inst] = true; |
1303 | continue; |
1304 | } |
1305 | |
1306 | // Check to see if Inst writes to memory. If not, continue. |
1307 | if (!hasAnalyzableMemoryWrite(Inst, *TLI)) |
1308 | continue; |
1309 | |
1310 | // eliminateNoopStore will update in iterator, if necessary. |
1311 | if (eliminateNoopStore(Inst, BBI, AA, MD, DL, TLI, IOL, |
1312 | ThrowableInst, DT)) { |
1313 | MadeChange = true; |
1314 | continue; |
1315 | } |
1316 | |
1317 | // If we find something that writes memory, get its memory dependence. |
1318 | MemDepResult InstDep = MD->getDependency(Inst); |
1319 | |
1320 | // Ignore any store where we can't find a local dependence. |
1321 | // FIXME: cross-block DSE would be fun. :) |
1322 | if (!InstDep.isDef() && !InstDep.isClobber()) |
1323 | continue; |
1324 | |
1325 | // Figure out what location is being stored to. |
1326 | MemoryLocation Loc = getLocForWrite(Inst, *TLI); |
1327 | |
1328 | // If we didn't get a useful location, fail. |
1329 | if (!Loc.Ptr) |
1330 | continue; |
1331 | |
1332 | // Loop until we find a store we can eliminate or a load that |
1333 | // invalidates the analysis. Without an upper bound on the number of |
1334 | // instructions examined, this analysis can become very time-consuming. |
1335 | // However, the potential gain diminishes as we process more instructions |
1336 | // without eliminating any of them. Therefore, we limit the number of |
1337 | // instructions we look at. |
1338 | auto Limit = MD->getDefaultBlockScanLimit(); |
1339 | while (InstDep.isDef() || InstDep.isClobber()) { |
1340 | // Get the memory clobbered by the instruction we depend on. MemDep will |
1341 | // skip any instructions that 'Loc' clearly doesn't interact with. If we |
1342 | // end up depending on a may- or must-aliased load, then we can't optimize |
1343 | // away the store and we bail out. However, if we depend on something |
1344 | // that overwrites the memory location we *can* potentially optimize it. |
1345 | // |
1346 | // Find out what memory location the dependent instruction stores. |
1347 | Instruction *DepWrite = InstDep.getInst(); |
1348 | if (!hasAnalyzableMemoryWrite(DepWrite, *TLI)) |
1349 | break; |
1350 | MemoryLocation DepLoc = getLocForWrite(DepWrite, *TLI); |
1351 | // If we didn't get a useful location, or if it isn't a size, bail out. |
1352 | if (!DepLoc.Ptr) |
1353 | break; |
1354 | |
1355 | // Find the last throwable instruction not removed by call to |
1356 | // deleteDeadInstruction. |
1357 | Instruction *LastThrowing = nullptr; |
1358 | if (!ThrowableInst.empty()) |
1359 | LastThrowing = ThrowableInst.back().first; |
1360 | |
1361 | // Make sure we don't look past a call which might throw. This is an |
1362 | // issue because MemoryDependenceAnalysis works in the wrong direction: |
1363 | // it finds instructions which dominate the current instruction, rather than |
1364 | // instructions which are post-dominated by the current instruction. |
1365 | // |
1366 | // If the underlying object is a non-escaping memory allocation, any store |
1367 | // to it is dead along the unwind edge. Otherwise, we need to preserve |
1368 | // the store. |
1369 | if (LastThrowing && DepWrite->comesBefore(LastThrowing)) { |
1370 | const Value *Underlying = getUnderlyingObject(DepLoc.Ptr); |
1371 | bool IsStoreDeadOnUnwind = isa<AllocaInst>(Underlying); |
1372 | if (!IsStoreDeadOnUnwind) { |
1373 | // We're looking for a call to an allocation function |
1374 | // where the allocation doesn't escape before the last |
1375 | // throwing instruction; PointerMayBeCaptured |
1376 | // reasonably fast approximation. |
1377 | IsStoreDeadOnUnwind = isAllocLikeFn(Underlying, TLI) && |
1378 | !PointerMayBeCaptured(Underlying, false, true); |
1379 | } |
1380 | if (!IsStoreDeadOnUnwind) |
1381 | break; |
1382 | } |
1383 | |
1384 | // If we find a write that is a) removable (i.e., non-volatile), b) is |
1385 | // completely obliterated by the store to 'Loc', and c) which we know that |
1386 | // 'Inst' doesn't load from, then we can remove it. |
1387 | // Also try to merge two stores if a later one only touches memory written |
1388 | // to by the earlier one. |
1389 | if (isRemovable(DepWrite) && |
1390 | !isPossibleSelfRead(Inst, Loc, DepWrite, *TLI, *AA)) { |
1391 | int64_t InstWriteOffset, DepWriteOffset; |
1392 | OverwriteResult OR = isOverwrite(Inst, DepWrite, Loc, DepLoc, DL, *TLI, |
1393 | DepWriteOffset, InstWriteOffset, *AA, |
1394 | BB.getParent()); |
1395 | if (OR == OW_MaybePartial) |
1396 | OR = isPartialOverwrite(Loc, DepLoc, DepWriteOffset, InstWriteOffset, |
1397 | DepWrite, IOL); |
1398 | |
1399 | if (OR == OW_Complete) { |
1400 | LLVM_DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *DepWritedo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *DepWrite << "\n KILLER: " << *Inst << '\n'; } } while (false) |
1401 | << "\n KILLER: " << *Inst << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *DepWrite << "\n KILLER: " << *Inst << '\n'; } } while (false); |
1402 | |
1403 | // Delete the store and now-dead instructions that feed it. |
1404 | deleteDeadInstruction(DepWrite, &BBI, *MD, *TLI, IOL, |
1405 | ThrowableInst); |
1406 | ++NumFastStores; |
1407 | MadeChange = true; |
1408 | |
1409 | // We erased DepWrite; start over. |
1410 | InstDep = MD->getDependency(Inst); |
1411 | continue; |
1412 | } else if ((OR == OW_End && isShortenableAtTheEnd(DepWrite)) || |
1413 | ((OR == OW_Begin && |
1414 | isShortenableAtTheBeginning(DepWrite)))) { |
1415 | assert(!EnablePartialOverwriteTracking && "Do not expect to perform "((!EnablePartialOverwriteTracking && "Do not expect to perform " "when partial-overwrite " "tracking is enabled") ? static_cast <void> (0) : __assert_fail ("!EnablePartialOverwriteTracking && \"Do not expect to perform \" \"when partial-overwrite \" \"tracking is enabled\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1417, __PRETTY_FUNCTION__)) |
1416 | "when partial-overwrite "((!EnablePartialOverwriteTracking && "Do not expect to perform " "when partial-overwrite " "tracking is enabled") ? static_cast <void> (0) : __assert_fail ("!EnablePartialOverwriteTracking && \"Do not expect to perform \" \"when partial-overwrite \" \"tracking is enabled\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1417, __PRETTY_FUNCTION__)) |
1417 | "tracking is enabled")((!EnablePartialOverwriteTracking && "Do not expect to perform " "when partial-overwrite " "tracking is enabled") ? static_cast <void> (0) : __assert_fail ("!EnablePartialOverwriteTracking && \"Do not expect to perform \" \"when partial-overwrite \" \"tracking is enabled\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1417, __PRETTY_FUNCTION__)); |
1418 | // The overwrite result is known, so these must be known, too. |
1419 | int64_t EarlierSize = DepLoc.Size.getValue(); |
1420 | int64_t LaterSize = Loc.Size.getValue(); |
1421 | bool IsOverwriteEnd = (OR == OW_End); |
1422 | MadeChange |= tryToShorten(DepWrite, DepWriteOffset, EarlierSize, |
1423 | InstWriteOffset, LaterSize, IsOverwriteEnd); |
1424 | } else if (EnablePartialStoreMerging && |
1425 | OR == OW_PartialEarlierWithFullLater) { |
1426 | auto *Earlier = dyn_cast<StoreInst>(DepWrite); |
1427 | auto *Later = dyn_cast<StoreInst>(Inst); |
1428 | if (Constant *C = tryToMergePartialOverlappingStores( |
1429 | Earlier, Later, InstWriteOffset, DepWriteOffset, DL, *AA, |
1430 | DT)) { |
1431 | auto *SI = new StoreInst( |
1432 | C, Earlier->getPointerOperand(), false, Earlier->getAlign(), |
1433 | Earlier->getOrdering(), Earlier->getSyncScopeID(), DepWrite); |
1434 | |
1435 | unsigned MDToKeep[] = {LLVMContext::MD_dbg, LLVMContext::MD_tbaa, |
1436 | LLVMContext::MD_alias_scope, |
1437 | LLVMContext::MD_noalias, |
1438 | LLVMContext::MD_nontemporal}; |
1439 | SI->copyMetadata(*DepWrite, MDToKeep); |
1440 | ++NumModifiedStores; |
1441 | |
1442 | // Delete the old stores and now-dead instructions that feed them. |
1443 | deleteDeadInstruction(Inst, &BBI, *MD, *TLI, IOL, |
1444 | ThrowableInst); |
1445 | deleteDeadInstruction(DepWrite, &BBI, *MD, *TLI, IOL, |
1446 | ThrowableInst); |
1447 | MadeChange = true; |
1448 | |
1449 | // We erased DepWrite and Inst (Loc); start over. |
1450 | break; |
1451 | } |
1452 | } |
1453 | } |
1454 | |
1455 | // If this is a may-aliased store that is clobbering the store value, we |
1456 | // can keep searching past it for another must-aliased pointer that stores |
1457 | // to the same location. For example, in: |
1458 | // store -> P |
1459 | // store -> Q |
1460 | // store -> P |
1461 | // we can remove the first store to P even though we don't know if P and Q |
1462 | // alias. |
1463 | if (DepWrite == &BB.front()) break; |
1464 | |
1465 | // Can't look past this instruction if it might read 'Loc'. |
1466 | if (isRefSet(AA->getModRefInfo(DepWrite, Loc))) |
1467 | break; |
1468 | |
1469 | InstDep = MD->getPointerDependencyFrom(Loc, /*isLoad=*/ false, |
1470 | DepWrite->getIterator(), &BB, |
1471 | /*QueryInst=*/ nullptr, &Limit); |
1472 | } |
1473 | } |
1474 | |
1475 | if (EnablePartialOverwriteTracking) |
1476 | MadeChange |= removePartiallyOverlappedStores(DL, IOL, *TLI); |
1477 | |
1478 | // If this block ends in a return, unwind, or unreachable, all allocas are |
1479 | // dead at its end, which means stores to them are also dead. |
1480 | if (BB.getTerminator()->getNumSuccessors() == 0) |
1481 | MadeChange |= handleEndBlock(BB, AA, MD, TLI, IOL, ThrowableInst); |
1482 | |
1483 | return MadeChange; |
1484 | } |
1485 | |
1486 | static bool eliminateDeadStores(Function &F, AliasAnalysis *AA, |
1487 | MemoryDependenceResults *MD, DominatorTree *DT, |
1488 | const TargetLibraryInfo *TLI) { |
1489 | bool MadeChange = false; |
1490 | for (BasicBlock &BB : F) |
1491 | // Only check non-dead blocks. Dead blocks may have strange pointer |
1492 | // cycles that will confuse alias analysis. |
1493 | if (DT->isReachableFromEntry(&BB)) |
1494 | MadeChange |= eliminateDeadStores(BB, AA, MD, DT, TLI); |
1495 | |
1496 | return MadeChange; |
1497 | } |
1498 | |
1499 | namespace { |
1500 | //============================================================================= |
1501 | // MemorySSA backed dead store elimination. |
1502 | // |
1503 | // The code below implements dead store elimination using MemorySSA. It uses |
1504 | // the following general approach: given a MemoryDef, walk upwards to find |
1505 | // clobbering MemoryDefs that may be killed by the starting def. Then check |
1506 | // that there are no uses that may read the location of the original MemoryDef |
1507 | // in between both MemoryDefs. A bit more concretely: |
1508 | // |
1509 | // For all MemoryDefs StartDef: |
1510 | // 1. Get the next dominating clobbering MemoryDef (EarlierAccess) by walking |
1511 | // upwards. |
1512 | // 2. Check that there are no reads between EarlierAccess and the StartDef by |
1513 | // checking all uses starting at EarlierAccess and walking until we see |
1514 | // StartDef. |
1515 | // 3. For each found CurrentDef, check that: |
1516 | // 1. There are no barrier instructions between CurrentDef and StartDef (like |
1517 | // throws or stores with ordering constraints). |
1518 | // 2. StartDef is executed whenever CurrentDef is executed. |
1519 | // 3. StartDef completely overwrites CurrentDef. |
1520 | // 4. Erase CurrentDef from the function and MemorySSA. |
1521 | |
1522 | // Returns true if \p I is an intrisnic that does not read or write memory. |
1523 | bool isNoopIntrinsic(Instruction *I) { |
1524 | if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) { |
1525 | switch (II->getIntrinsicID()) { |
1526 | case Intrinsic::lifetime_start: |
1527 | case Intrinsic::lifetime_end: |
1528 | case Intrinsic::invariant_end: |
1529 | case Intrinsic::launder_invariant_group: |
1530 | case Intrinsic::assume: |
1531 | return true; |
1532 | case Intrinsic::dbg_addr: |
1533 | case Intrinsic::dbg_declare: |
1534 | case Intrinsic::dbg_label: |
1535 | case Intrinsic::dbg_value: |
1536 | llvm_unreachable("Intrinsic should not be modeled in MemorySSA")::llvm::llvm_unreachable_internal("Intrinsic should not be modeled in MemorySSA" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1536); |
1537 | default: |
1538 | return false; |
1539 | } |
1540 | } |
1541 | return false; |
1542 | } |
1543 | |
1544 | // Check if we can ignore \p D for DSE. |
1545 | bool canSkipDef(MemoryDef *D, bool DefVisibleToCaller) { |
1546 | Instruction *DI = D->getMemoryInst(); |
1547 | // Calls that only access inaccessible memory cannot read or write any memory |
1548 | // locations we consider for elimination. |
1549 | if (auto *CB = dyn_cast<CallBase>(DI)) |
1550 | if (CB->onlyAccessesInaccessibleMemory()) |
1551 | return true; |
1552 | |
1553 | // We can eliminate stores to locations not visible to the caller across |
1554 | // throwing instructions. |
1555 | if (DI->mayThrow() && !DefVisibleToCaller) |
1556 | return true; |
1557 | |
1558 | // We can remove the dead stores, irrespective of the fence and its ordering |
1559 | // (release/acquire/seq_cst). Fences only constraints the ordering of |
1560 | // already visible stores, it does not make a store visible to other |
1561 | // threads. So, skipping over a fence does not change a store from being |
1562 | // dead. |
1563 | if (isa<FenceInst>(DI)) |
1564 | return true; |
1565 | |
1566 | // Skip intrinsics that do not really read or modify memory. |
1567 | if (isNoopIntrinsic(D->getMemoryInst())) |
1568 | return true; |
1569 | |
1570 | return false; |
1571 | } |
1572 | |
1573 | struct DSEState { |
1574 | Function &F; |
1575 | AliasAnalysis &AA; |
1576 | |
1577 | /// The single BatchAA instance that is used to cache AA queries. It will |
1578 | /// not be invalidated over the whole run. This is safe, because: |
1579 | /// 1. Only memory writes are removed, so the alias cache for memory |
1580 | /// locations remains valid. |
1581 | /// 2. No new instructions are added (only instructions removed), so cached |
1582 | /// information for a deleted value cannot be accessed by a re-used new |
1583 | /// value pointer. |
1584 | BatchAAResults BatchAA; |
1585 | |
1586 | MemorySSA &MSSA; |
1587 | DominatorTree &DT; |
1588 | PostDominatorTree &PDT; |
1589 | const TargetLibraryInfo &TLI; |
1590 | const DataLayout &DL; |
1591 | |
1592 | // All MemoryDefs that potentially could kill other MemDefs. |
1593 | SmallVector<MemoryDef *, 64> MemDefs; |
1594 | // Any that should be skipped as they are already deleted |
1595 | SmallPtrSet<MemoryAccess *, 4> SkipStores; |
1596 | // Keep track of all of the objects that are invisible to the caller before |
1597 | // the function returns. |
1598 | // SmallPtrSet<const Value *, 16> InvisibleToCallerBeforeRet; |
1599 | DenseMap<const Value *, bool> InvisibleToCallerBeforeRet; |
1600 | // Keep track of all of the objects that are invisible to the caller after |
1601 | // the function returns. |
1602 | DenseMap<const Value *, bool> InvisibleToCallerAfterRet; |
1603 | // Keep track of blocks with throwing instructions not modeled in MemorySSA. |
1604 | SmallPtrSet<BasicBlock *, 16> ThrowingBlocks; |
1605 | // Post-order numbers for each basic block. Used to figure out if memory |
1606 | // accesses are executed before another access. |
1607 | DenseMap<BasicBlock *, unsigned> PostOrderNumbers; |
1608 | |
1609 | /// Keep track of instructions (partly) overlapping with killing MemoryDefs per |
1610 | /// basic block. |
1611 | DenseMap<BasicBlock *, InstOverlapIntervalsTy> IOLs; |
1612 | |
1613 | struct CheckCache { |
1614 | SmallPtrSet<MemoryAccess *, 16> KnownNoReads; |
1615 | SmallPtrSet<MemoryAccess *, 16> KnownReads; |
1616 | |
1617 | bool isKnownNoRead(MemoryAccess *A) const { |
1618 | return KnownNoReads.find(A) != KnownNoReads.end(); |
1619 | } |
1620 | bool isKnownRead(MemoryAccess *A) const { |
1621 | return KnownReads.find(A) != KnownReads.end(); |
1622 | } |
1623 | }; |
1624 | |
1625 | DSEState(Function &F, AliasAnalysis &AA, MemorySSA &MSSA, DominatorTree &DT, |
1626 | PostDominatorTree &PDT, const TargetLibraryInfo &TLI) |
1627 | : F(F), AA(AA), BatchAA(AA), MSSA(MSSA), DT(DT), PDT(PDT), TLI(TLI), |
1628 | DL(F.getParent()->getDataLayout()) {} |
1629 | |
1630 | static DSEState get(Function &F, AliasAnalysis &AA, MemorySSA &MSSA, |
1631 | DominatorTree &DT, PostDominatorTree &PDT, |
1632 | const TargetLibraryInfo &TLI) { |
1633 | DSEState State(F, AA, MSSA, DT, PDT, TLI); |
1634 | // Collect blocks with throwing instructions not modeled in MemorySSA and |
1635 | // alloc-like objects. |
1636 | unsigned PO = 0; |
1637 | for (BasicBlock *BB : post_order(&F)) { |
1638 | State.PostOrderNumbers[BB] = PO++; |
1639 | for (Instruction &I : *BB) { |
1640 | MemoryAccess *MA = MSSA.getMemoryAccess(&I); |
1641 | if (I.mayThrow() && !MA) |
1642 | State.ThrowingBlocks.insert(I.getParent()); |
1643 | |
1644 | auto *MD = dyn_cast_or_null<MemoryDef>(MA); |
1645 | if (MD && State.MemDefs.size() < MemorySSADefsPerBlockLimit && |
1646 | (State.getLocForWriteEx(&I) || State.isMemTerminatorInst(&I))) |
1647 | State.MemDefs.push_back(MD); |
1648 | } |
1649 | } |
1650 | |
1651 | // Treat byval or inalloca arguments the same as Allocas, stores to them are |
1652 | // dead at the end of the function. |
1653 | for (Argument &AI : F.args()) |
1654 | if (AI.hasPassPointeeByValueCopyAttr()) { |
1655 | // For byval, the caller doesn't know the address of the allocation. |
1656 | if (AI.hasByValAttr()) |
1657 | State.InvisibleToCallerBeforeRet.insert({&AI, true}); |
1658 | State.InvisibleToCallerAfterRet.insert({&AI, true}); |
1659 | } |
1660 | |
1661 | return State; |
1662 | } |
1663 | |
1664 | bool isInvisibleToCallerAfterRet(const Value *V) { |
1665 | if (isa<AllocaInst>(V)) |
1666 | return true; |
1667 | auto I = InvisibleToCallerAfterRet.insert({V, false}); |
1668 | if (I.second) { |
1669 | if (!isInvisibleToCallerBeforeRet(V)) { |
1670 | I.first->second = false; |
1671 | } else { |
1672 | auto *Inst = dyn_cast<Instruction>(V); |
1673 | if (Inst && isAllocLikeFn(Inst, &TLI)) |
1674 | I.first->second = !PointerMayBeCaptured(V, true, false); |
1675 | } |
1676 | } |
1677 | return I.first->second; |
1678 | } |
1679 | |
1680 | bool isInvisibleToCallerBeforeRet(const Value *V) { |
1681 | if (isa<AllocaInst>(V)) |
1682 | return true; |
1683 | auto I = InvisibleToCallerBeforeRet.insert({V, false}); |
1684 | if (I.second) { |
1685 | auto *Inst = dyn_cast<Instruction>(V); |
1686 | if (Inst && isAllocLikeFn(Inst, &TLI)) |
1687 | // NOTE: This could be made more precise by PointerMayBeCapturedBefore |
1688 | // with the killing MemoryDef. But we refrain from doing so for now to |
1689 | // limit compile-time and this does not cause any changes to the number |
1690 | // of stores removed on a large test set in practice. |
1691 | I.first->second = !PointerMayBeCaptured(V, false, true); |
1692 | } |
1693 | return I.first->second; |
1694 | } |
1695 | |
1696 | Optional<MemoryLocation> getLocForWriteEx(Instruction *I) const { |
1697 | if (!I->mayWriteToMemory()) |
1698 | return None; |
1699 | |
1700 | if (auto *MTI = dyn_cast<AnyMemIntrinsic>(I)) |
1701 | return {MemoryLocation::getForDest(MTI)}; |
1702 | |
1703 | if (auto *CB = dyn_cast<CallBase>(I)) { |
1704 | // If the functions may write to memory we do not know about, bail out. |
1705 | if (!CB->onlyAccessesArgMemory() && |
1706 | !CB->onlyAccessesInaccessibleMemOrArgMem()) |
1707 | return None; |
1708 | |
1709 | LibFunc LF; |
1710 | if (TLI.getLibFunc(*CB, LF) && TLI.has(LF)) { |
1711 | switch (LF) { |
1712 | case LibFunc_strcpy: |
1713 | case LibFunc_strncpy: |
1714 | case LibFunc_strcat: |
1715 | case LibFunc_strncat: |
1716 | return {MemoryLocation(CB->getArgOperand(0))}; |
1717 | default: |
1718 | break; |
1719 | } |
1720 | } |
1721 | switch (CB->getIntrinsicID()) { |
1722 | case Intrinsic::init_trampoline: |
1723 | return {MemoryLocation(CB->getArgOperand(0))}; |
1724 | case Intrinsic::masked_store: |
1725 | return {MemoryLocation::getForArgument(CB, 1, TLI)}; |
1726 | default: |
1727 | break; |
1728 | } |
1729 | return None; |
1730 | } |
1731 | |
1732 | return MemoryLocation::getOrNone(I); |
1733 | } |
1734 | |
1735 | /// Returns true if \p UseInst completely overwrites \p DefLoc |
1736 | /// (stored by \p DefInst). |
1737 | bool isCompleteOverwrite(MemoryLocation DefLoc, Instruction *DefInst, |
1738 | Instruction *UseInst) { |
1739 | // UseInst has a MemoryDef associated in MemorySSA. It's possible for a |
1740 | // MemoryDef to not write to memory, e.g. a volatile load is modeled as a |
1741 | // MemoryDef. |
1742 | if (!UseInst->mayWriteToMemory()) |
1743 | return false; |
1744 | |
1745 | if (auto *CB = dyn_cast<CallBase>(UseInst)) |
1746 | if (CB->onlyAccessesInaccessibleMemory()) |
1747 | return false; |
1748 | |
1749 | int64_t InstWriteOffset, DepWriteOffset; |
1750 | if (auto CC = getLocForWriteEx(UseInst)) |
1751 | return isOverwrite(UseInst, DefInst, *CC, DefLoc, DL, TLI, DepWriteOffset, |
1752 | InstWriteOffset, BatchAA, &F) == OW_Complete; |
1753 | return false; |
1754 | } |
1755 | |
1756 | /// Returns true if \p Def is not read before returning from the function. |
1757 | bool isWriteAtEndOfFunction(MemoryDef *Def) { |
1758 | LLVM_DEBUG(dbgs() << " Check if def " << *Def << " ("do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " Check if def " << *Def << " (" << *Def->getMemoryInst() << ") is at the end the function \n" ; } } while (false) |
1759 | << *Def->getMemoryInst()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " Check if def " << *Def << " (" << *Def->getMemoryInst() << ") is at the end the function \n" ; } } while (false) |
1760 | << ") is at the end the function \n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " Check if def " << *Def << " (" << *Def->getMemoryInst() << ") is at the end the function \n" ; } } while (false); |
1761 | |
1762 | auto MaybeLoc = getLocForWriteEx(Def->getMemoryInst()); |
1763 | if (!MaybeLoc) { |
1764 | LLVM_DEBUG(dbgs() << " ... could not get location for write.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... could not get location for write.\n" ; } } while (false); |
1765 | return false; |
1766 | } |
1767 | |
1768 | SmallVector<MemoryAccess *, 4> WorkList; |
1769 | SmallPtrSet<MemoryAccess *, 8> Visited; |
1770 | auto PushMemUses = [&WorkList, &Visited](MemoryAccess *Acc) { |
1771 | if (!Visited.insert(Acc).second) |
1772 | return; |
1773 | for (Use &U : Acc->uses()) |
1774 | WorkList.push_back(cast<MemoryAccess>(U.getUser())); |
1775 | }; |
1776 | PushMemUses(Def); |
1777 | for (unsigned I = 0; I < WorkList.size(); I++) { |
1778 | if (WorkList.size() >= MemorySSAScanLimit) { |
1779 | LLVM_DEBUG(dbgs() << " ... hit exploration limit.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... hit exploration limit.\n"; } } while (false); |
1780 | return false; |
1781 | } |
1782 | |
1783 | MemoryAccess *UseAccess = WorkList[I]; |
1784 | // Simply adding the users of MemoryPhi to the worklist is not enough, |
1785 | // because we might miss read clobbers in different iterations of a loop, |
1786 | // for example. |
1787 | // TODO: Add support for phi translation to handle the loop case. |
1788 | if (isa<MemoryPhi>(UseAccess)) |
1789 | return false; |
1790 | |
1791 | // TODO: Checking for aliasing is expensive. Consider reducing the amount |
1792 | // of times this is called and/or caching it. |
1793 | Instruction *UseInst = cast<MemoryUseOrDef>(UseAccess)->getMemoryInst(); |
1794 | if (isReadClobber(*MaybeLoc, UseInst)) { |
1795 | LLVM_DEBUG(dbgs() << " ... hit read clobber " << *UseInst << ".\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... hit read clobber " << *UseInst << ".\n"; } } while (false); |
1796 | return false; |
1797 | } |
1798 | |
1799 | if (MemoryDef *UseDef = dyn_cast<MemoryDef>(UseAccess)) |
1800 | PushMemUses(UseDef); |
1801 | } |
1802 | return true; |
1803 | } |
1804 | |
1805 | /// If \p I is a memory terminator like llvm.lifetime.end or free, return a |
1806 | /// pair with the MemoryLocation terminated by \p I and a boolean flag |
1807 | /// indicating whether \p I is a free-like call. |
1808 | Optional<std::pair<MemoryLocation, bool>> |
1809 | getLocForTerminator(Instruction *I) const { |
1810 | uint64_t Len; |
1811 | Value *Ptr; |
1812 | if (match(I, m_Intrinsic<Intrinsic::lifetime_end>(m_ConstantInt(Len), |
1813 | m_Value(Ptr)))) |
1814 | return {std::make_pair(MemoryLocation(Ptr, Len), false)}; |
1815 | |
1816 | if (auto *CB = dyn_cast<CallBase>(I)) { |
1817 | if (isFreeCall(I, &TLI)) |
1818 | return {std::make_pair(MemoryLocation(CB->getArgOperand(0)), true)}; |
1819 | } |
1820 | |
1821 | return None; |
1822 | } |
1823 | |
1824 | /// Returns true if \p I is a memory terminator instruction like |
1825 | /// llvm.lifetime.end or free. |
1826 | bool isMemTerminatorInst(Instruction *I) const { |
1827 | IntrinsicInst *II = dyn_cast<IntrinsicInst>(I); |
1828 | return (II && II->getIntrinsicID() == Intrinsic::lifetime_end) || |
1829 | isFreeCall(I, &TLI); |
1830 | } |
1831 | |
1832 | /// Returns true if \p MaybeTerm is a memory terminator for \p Loc from |
1833 | /// instruction \p AccessI. |
1834 | bool isMemTerminator(MemoryLocation Loc, Instruction *AccessI, |
1835 | Instruction *MaybeTerm) { |
1836 | Optional<std::pair<MemoryLocation, bool>> MaybeTermLoc = |
1837 | getLocForTerminator(MaybeTerm); |
1838 | |
1839 | if (!MaybeTermLoc) |
1840 | return false; |
1841 | |
1842 | // If the terminator is a free-like call, all accesses to the underlying |
1843 | // object can be considered terminated. |
1844 | if (getUnderlyingObject(Loc.Ptr) != |
1845 | getUnderlyingObject(MaybeTermLoc->first.Ptr)) |
1846 | return false; |
1847 | |
1848 | int64_t InstWriteOffset, DepWriteOffset; |
1849 | return MaybeTermLoc->second || |
1850 | isOverwrite(MaybeTerm, AccessI, MaybeTermLoc->first, Loc, DL, TLI, |
1851 | DepWriteOffset, InstWriteOffset, BatchAA, |
1852 | &F) == OW_Complete; |
1853 | } |
1854 | |
1855 | // Returns true if \p Use may read from \p DefLoc. |
1856 | bool isReadClobber(MemoryLocation DefLoc, Instruction *UseInst) { |
1857 | if (isNoopIntrinsic(UseInst)) |
1858 | return false; |
1859 | |
1860 | // Monotonic or weaker atomic stores can be re-ordered and do not need to be |
1861 | // treated as read clobber. |
1862 | if (auto SI = dyn_cast<StoreInst>(UseInst)) |
1863 | return isStrongerThan(SI->getOrdering(), AtomicOrdering::Monotonic); |
1864 | |
1865 | if (!UseInst->mayReadFromMemory()) |
1866 | return false; |
1867 | |
1868 | if (auto *CB = dyn_cast<CallBase>(UseInst)) |
1869 | if (CB->onlyAccessesInaccessibleMemory()) |
1870 | return false; |
1871 | |
1872 | // NOTE: For calls, the number of stores removed could be slightly improved |
1873 | // by using AA.callCapturesBefore(UseInst, DefLoc, &DT), but that showed to |
1874 | // be expensive compared to the benefits in practice. For now, avoid more |
1875 | // expensive analysis to limit compile-time. |
1876 | return isRefSet(BatchAA.getModRefInfo(UseInst, DefLoc)); |
1877 | } |
1878 | |
1879 | /// Returns true if \p Ptr is guaranteed to be loop invariant for any possible |
1880 | /// loop. In particular, this guarantees that it only references a single |
1881 | /// MemoryLocation during execution of the containing function. |
1882 | bool IsGuaranteedLoopInvariant(Value *Ptr) { |
1883 | auto IsGuaranteedLoopInvariantBase = [this](Value *Ptr) { |
1884 | Ptr = Ptr->stripPointerCasts(); |
1885 | if (auto *I = dyn_cast<Instruction>(Ptr)) { |
1886 | if (isa<AllocaInst>(Ptr)) |
1887 | return true; |
1888 | |
1889 | if (isAllocLikeFn(I, &TLI)) |
1890 | return true; |
1891 | |
1892 | return false; |
1893 | } |
1894 | return true; |
1895 | }; |
1896 | |
1897 | Ptr = Ptr->stripPointerCasts(); |
1898 | if (auto *GEP = dyn_cast<GEPOperator>(Ptr)) { |
1899 | return IsGuaranteedLoopInvariantBase(GEP->getPointerOperand()) && |
1900 | GEP->hasAllConstantIndices(); |
1901 | } |
1902 | return IsGuaranteedLoopInvariantBase(Ptr); |
1903 | } |
1904 | |
1905 | // Find a MemoryDef writing to \p DefLoc and dominating \p StartAccess, with |
1906 | // no read access between them or on any other path to a function exit block |
1907 | // if \p DefLoc is not accessible after the function returns. If there is no |
1908 | // such MemoryDef, return None. The returned value may not (completely) |
1909 | // overwrite \p DefLoc. Currently we bail out when we encounter an aliasing |
1910 | // MemoryUse (read). |
1911 | Optional<MemoryAccess *> |
1912 | getDomMemoryDef(MemoryDef *KillingDef, MemoryAccess *StartAccess, |
1913 | MemoryLocation DefLoc, const Value *DefUO, CheckCache &Cache, |
1914 | unsigned &ScanLimit, unsigned &WalkerStepLimit, |
1915 | bool IsMemTerm, unsigned &PartialLimit) { |
1916 | if (ScanLimit == 0 || WalkerStepLimit == 0) { |
1917 | LLVM_DEBUG(dbgs() << "\n ... hit scan limit\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "\n ... hit scan limit\n"; } } while (false); |
1918 | return None; |
1919 | } |
1920 | |
1921 | MemoryAccess *Current = StartAccess; |
1922 | Instruction *KillingI = KillingDef->getMemoryInst(); |
1923 | bool StepAgain; |
1924 | LLVM_DEBUG(dbgs() << " trying to get dominating access\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " trying to get dominating access\n" ; } } while (false); |
1925 | |
1926 | // Find the next clobbering Mod access for DefLoc, starting at StartAccess. |
1927 | do { |
1928 | StepAgain = false; |
1929 | LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { { dbgs() << " visiting " << *Current ; if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef >(Current)) dbgs() << " (" << *cast<MemoryUseOrDef >(Current)->getMemoryInst() << ")"; dbgs() << "\n"; }; } } while (false) |
1930 | dbgs() << " visiting " << *Current;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { { dbgs() << " visiting " << *Current ; if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef >(Current)) dbgs() << " (" << *cast<MemoryUseOrDef >(Current)->getMemoryInst() << ")"; dbgs() << "\n"; }; } } while (false) |
1931 | if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef>(Current))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { { dbgs() << " visiting " << *Current ; if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef >(Current)) dbgs() << " (" << *cast<MemoryUseOrDef >(Current)->getMemoryInst() << ")"; dbgs() << "\n"; }; } } while (false) |
1932 | dbgs() << " (" << *cast<MemoryUseOrDef>(Current)->getMemoryInst()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { { dbgs() << " visiting " << *Current ; if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef >(Current)) dbgs() << " (" << *cast<MemoryUseOrDef >(Current)->getMemoryInst() << ")"; dbgs() << "\n"; }; } } while (false) |
1933 | << ")";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { { dbgs() << " visiting " << *Current ; if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef >(Current)) dbgs() << " (" << *cast<MemoryUseOrDef >(Current)->getMemoryInst() << ")"; dbgs() << "\n"; }; } } while (false) |
1934 | dbgs() << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { { dbgs() << " visiting " << *Current ; if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef >(Current)) dbgs() << " (" << *cast<MemoryUseOrDef >(Current)->getMemoryInst() << ")"; dbgs() << "\n"; }; } } while (false) |
1935 | })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { { dbgs() << " visiting " << *Current ; if (!MSSA.isLiveOnEntryDef(Current) && isa<MemoryUseOrDef >(Current)) dbgs() << " (" << *cast<MemoryUseOrDef >(Current)->getMemoryInst() << ")"; dbgs() << "\n"; }; } } while (false); |
1936 | |
1937 | // Reached TOP. |
1938 | if (MSSA.isLiveOnEntryDef(Current)) { |
1939 | LLVM_DEBUG(dbgs() << " ... found LiveOnEntryDef\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found LiveOnEntryDef\n"; } } while (false); |
1940 | return None; |
1941 | } |
1942 | |
1943 | // Cost of a step. Accesses in the same block are more likely to be valid |
1944 | // candidates for elimination, hence consider them cheaper. |
1945 | unsigned StepCost = KillingDef->getBlock() == Current->getBlock() |
1946 | ? MemorySSASameBBStepCost |
1947 | : MemorySSAOtherBBStepCost; |
1948 | if (WalkerStepLimit <= StepCost) { |
1949 | LLVM_DEBUG(dbgs() << " ... hit walker step limit\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... hit walker step limit\n"; } } while (false); |
1950 | return None; |
1951 | } |
1952 | WalkerStepLimit -= StepCost; |
1953 | |
1954 | // Return for MemoryPhis. They cannot be eliminated directly and the |
1955 | // caller is responsible for traversing them. |
1956 | if (isa<MemoryPhi>(Current)) { |
1957 | LLVM_DEBUG(dbgs() << " ... found MemoryPhi\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found MemoryPhi\n"; } } while (false); |
1958 | return Current; |
1959 | } |
1960 | |
1961 | // Below, check if CurrentDef is a valid candidate to be eliminated by |
1962 | // KillingDef. If it is not, check the next candidate. |
1963 | MemoryDef *CurrentDef = cast<MemoryDef>(Current); |
1964 | Instruction *CurrentI = CurrentDef->getMemoryInst(); |
1965 | |
1966 | if (canSkipDef(CurrentDef, !isInvisibleToCallerBeforeRet(DefUO))) { |
1967 | StepAgain = true; |
1968 | Current = CurrentDef->getDefiningAccess(); |
1969 | continue; |
1970 | } |
1971 | |
1972 | // Before we try to remove anything, check for any extra throwing |
1973 | // instructions that block us from DSEing |
1974 | if (mayThrowBetween(KillingI, CurrentI, DefUO)) { |
1975 | LLVM_DEBUG(dbgs() << " ... skip, may throw!\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, may throw!\n"; } } while (false); |
1976 | return None; |
1977 | } |
1978 | |
1979 | // Check for anything that looks like it will be a barrier to further |
1980 | // removal |
1981 | if (isDSEBarrier(DefUO, CurrentI)) { |
1982 | LLVM_DEBUG(dbgs() << " ... skip, barrier\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, barrier\n"; } } while (false); |
1983 | return None; |
1984 | } |
1985 | |
1986 | // If Current is known to be on path that reads DefLoc or is a read |
1987 | // clobber, bail out, as the path is not profitable. We skip this check |
1988 | // for intrinsic calls, because the code knows how to handle memcpy |
1989 | // intrinsics. |
1990 | if (!isa<IntrinsicInst>(CurrentI) && |
1991 | (Cache.KnownReads.contains(Current) || |
1992 | isReadClobber(DefLoc, CurrentI))) { |
1993 | Cache.KnownReads.insert(Current); |
1994 | return None; |
1995 | } |
1996 | |
1997 | // Quick check if there are direct uses that are read-clobbers. |
1998 | if (any_of(Current->uses(), [this, &DefLoc, StartAccess](Use &U) { |
1999 | if (auto *UseOrDef = dyn_cast<MemoryUseOrDef>(U.getUser())) |
2000 | return !MSSA.dominates(StartAccess, UseOrDef) && |
2001 | isReadClobber(DefLoc, UseOrDef->getMemoryInst()); |
2002 | return false; |
2003 | })) { |
2004 | Cache.KnownReads.insert(Current); |
2005 | LLVM_DEBUG(dbgs() << " ... found a read clobber\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found a read clobber\n"; } } while (false); |
2006 | return None; |
2007 | } |
2008 | |
2009 | // If Current cannot be analyzed or is not removable, check the next |
2010 | // candidate. |
2011 | if (!hasAnalyzableMemoryWrite(CurrentI, TLI) || !isRemovable(CurrentI)) { |
2012 | StepAgain = true; |
2013 | Current = CurrentDef->getDefiningAccess(); |
2014 | continue; |
2015 | } |
2016 | |
2017 | // If Current does not have an analyzable write location, skip it |
2018 | auto CurrentLoc = getLocForWriteEx(CurrentI); |
2019 | if (!CurrentLoc) { |
2020 | StepAgain = true; |
2021 | Current = CurrentDef->getDefiningAccess(); |
2022 | continue; |
2023 | } |
2024 | |
2025 | if (IsMemTerm) { |
2026 | // If the killing def is a memory terminator (e.g. lifetime.end), check |
2027 | // the next candidate if the current Current does not write the same |
2028 | // underlying object as the terminator. |
2029 | if (!isMemTerminator(*CurrentLoc, CurrentI, KillingI)) { |
2030 | StepAgain = true; |
2031 | Current = CurrentDef->getDefiningAccess(); |
2032 | } |
2033 | continue; |
2034 | } else { |
2035 | // AliasAnalysis does not account for loops. Limit elimination to |
2036 | // candidates for which we can guarantee they always store to the same |
2037 | // memory location and not multiple locations in a loop. |
2038 | if (Current->getBlock() != KillingDef->getBlock() && |
2039 | !IsGuaranteedLoopInvariant(const_cast<Value *>(CurrentLoc->Ptr))) { |
2040 | StepAgain = true; |
2041 | Current = CurrentDef->getDefiningAccess(); |
2042 | WalkerStepLimit -= 1; |
2043 | continue; |
2044 | } |
2045 | |
2046 | int64_t InstWriteOffset, DepWriteOffset; |
2047 | auto OR = isOverwrite(KillingI, CurrentI, DefLoc, *CurrentLoc, DL, TLI, |
2048 | DepWriteOffset, InstWriteOffset, BatchAA, &F); |
2049 | // If Current does not write to the same object as KillingDef, check |
2050 | // the next candidate. |
2051 | if (OR == OW_Unknown) { |
2052 | StepAgain = true; |
2053 | Current = CurrentDef->getDefiningAccess(); |
2054 | } else if (OR == OW_MaybePartial) { |
2055 | // If KillingDef only partially overwrites Current, check the next |
2056 | // candidate if the partial step limit is exceeded. This aggressively |
2057 | // limits the number of candidates for partial store elimination, |
2058 | // which are less likely to be removable in the end. |
2059 | if (PartialLimit <= 1) { |
2060 | StepAgain = true; |
2061 | Current = CurrentDef->getDefiningAccess(); |
2062 | WalkerStepLimit -= 1; |
2063 | continue; |
2064 | } |
2065 | PartialLimit -= 1; |
2066 | } |
2067 | } |
2068 | } while (StepAgain); |
2069 | |
2070 | // Accesses to objects accessible after the function returns can only be |
2071 | // eliminated if the access is killed along all paths to the exit. Collect |
2072 | // the blocks with killing (=completely overwriting MemoryDefs) and check if |
2073 | // they cover all paths from EarlierAccess to any function exit. |
2074 | SmallPtrSet<Instruction *, 16> KillingDefs; |
2075 | KillingDefs.insert(KillingDef->getMemoryInst()); |
2076 | MemoryAccess *EarlierAccess = Current; |
2077 | Instruction *EarlierMemInst = |
2078 | cast<MemoryDef>(EarlierAccess)->getMemoryInst(); |
2079 | LLVM_DEBUG(dbgs() << " Checking for reads of " << *EarlierAccess << " ("do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " Checking for reads of " << *EarlierAccess << " (" << *EarlierMemInst << ")\n"; } } while (false) |
2080 | << *EarlierMemInst << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " Checking for reads of " << *EarlierAccess << " (" << *EarlierMemInst << ")\n"; } } while (false); |
2081 | |
2082 | SmallSetVector<MemoryAccess *, 32> WorkList; |
2083 | auto PushMemUses = [&WorkList](MemoryAccess *Acc) { |
2084 | for (Use &U : Acc->uses()) |
2085 | WorkList.insert(cast<MemoryAccess>(U.getUser())); |
2086 | }; |
2087 | PushMemUses(EarlierAccess); |
2088 | |
2089 | // Optimistically collect all accesses for reads. If we do not find any |
2090 | // read clobbers, add them to the cache. |
2091 | SmallPtrSet<MemoryAccess *, 16> KnownNoReads; |
2092 | if (!EarlierMemInst->mayReadFromMemory()) |
2093 | KnownNoReads.insert(EarlierAccess); |
2094 | // Check if EarlierDef may be read. |
2095 | for (unsigned I = 0; I < WorkList.size(); I++) { |
2096 | MemoryAccess *UseAccess = WorkList[I]; |
2097 | |
2098 | LLVM_DEBUG(dbgs() << " " << *UseAccess)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " " << *UseAccess; } } while (false); |
2099 | // Bail out if the number of accesses to check exceeds the scan limit. |
2100 | if (ScanLimit < (WorkList.size() - I)) { |
2101 | LLVM_DEBUG(dbgs() << "\n ... hit scan limit\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "\n ... hit scan limit\n"; } } while (false); |
2102 | return None; |
2103 | } |
2104 | --ScanLimit; |
2105 | NumDomMemDefChecks++; |
2106 | |
2107 | // Check if we already visited this access. |
2108 | if (Cache.isKnownNoRead(UseAccess)) { |
2109 | LLVM_DEBUG(dbgs() << " ... skip, discovered that " << *UseAccessdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, discovered that " << *UseAccess << " is safe earlier.\n"; } } while (false) |
2110 | << " is safe earlier.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, discovered that " << *UseAccess << " is safe earlier.\n"; } } while (false); |
2111 | continue; |
2112 | } |
2113 | if (Cache.isKnownRead(UseAccess)) { |
2114 | LLVM_DEBUG(dbgs() << " ... bail out, discovered that " << *UseAccessdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... bail out, discovered that " << *UseAccess << " has a read-clobber earlier.\n"; } } while (false) |
2115 | << " has a read-clobber earlier.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... bail out, discovered that " << *UseAccess << " has a read-clobber earlier.\n"; } } while (false); |
2116 | return None; |
2117 | } |
2118 | KnownNoReads.insert(UseAccess); |
2119 | |
2120 | if (isa<MemoryPhi>(UseAccess)) { |
2121 | if (any_of(KillingDefs, [this, UseAccess](Instruction *KI) { |
2122 | return DT.properlyDominates(KI->getParent(), |
2123 | UseAccess->getBlock()); |
2124 | })) { |
2125 | LLVM_DEBUG(dbgs() << " ... skipping, dominated by killing block\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, dominated by killing block\n" ; } } while (false); |
2126 | continue; |
2127 | } |
2128 | LLVM_DEBUG(dbgs() << "\n ... adding PHI uses\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "\n ... adding PHI uses\n"; } } while (false); |
2129 | PushMemUses(UseAccess); |
2130 | continue; |
2131 | } |
2132 | |
2133 | Instruction *UseInst = cast<MemoryUseOrDef>(UseAccess)->getMemoryInst(); |
2134 | LLVM_DEBUG(dbgs() << " (" << *UseInst << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " (" << *UseInst << ")\n" ; } } while (false); |
2135 | |
2136 | if (any_of(KillingDefs, [this, UseInst](Instruction *KI) { |
2137 | return DT.dominates(KI, UseInst); |
2138 | })) { |
2139 | LLVM_DEBUG(dbgs() << " ... skipping, dominated by killing def\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, dominated by killing def\n" ; } } while (false); |
2140 | continue; |
2141 | } |
2142 | |
2143 | // A memory terminator kills all preceeding MemoryDefs and all succeeding |
2144 | // MemoryAccesses. We do not have to check it's users. |
2145 | if (isMemTerminator(DefLoc, KillingI, UseInst)) { |
2146 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, memterminator invalidates following accesses\n" ; } } while (false) |
2147 | dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, memterminator invalidates following accesses\n" ; } } while (false) |
2148 | << " ... skipping, memterminator invalidates following accesses\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, memterminator invalidates following accesses\n" ; } } while (false); |
2149 | continue; |
2150 | } |
2151 | |
2152 | if (isNoopIntrinsic(cast<MemoryUseOrDef>(UseAccess)->getMemoryInst())) { |
2153 | LLVM_DEBUG(dbgs() << " ... adding uses of intrinsic\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... adding uses of intrinsic\n" ; } } while (false); |
2154 | PushMemUses(UseAccess); |
2155 | continue; |
2156 | } |
2157 | |
2158 | if (UseInst->mayThrow() && !isInvisibleToCallerBeforeRet(DefUO)) { |
2159 | LLVM_DEBUG(dbgs() << " ... found throwing instruction\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found throwing instruction\n" ; } } while (false); |
2160 | Cache.KnownReads.insert(UseAccess); |
2161 | Cache.KnownReads.insert(StartAccess); |
2162 | Cache.KnownReads.insert(EarlierAccess); |
2163 | return None; |
2164 | } |
2165 | |
2166 | // Uses which may read the original MemoryDef mean we cannot eliminate the |
2167 | // original MD. Stop walk. |
2168 | if (isReadClobber(DefLoc, UseInst)) { |
2169 | LLVM_DEBUG(dbgs() << " ... found read clobber\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found read clobber\n"; } } while (false); |
2170 | Cache.KnownReads.insert(UseAccess); |
2171 | Cache.KnownReads.insert(StartAccess); |
2172 | Cache.KnownReads.insert(EarlierAccess); |
2173 | return None; |
2174 | } |
2175 | |
2176 | // For the KillingDef and EarlierAccess we only have to check if it reads |
2177 | // the memory location. |
2178 | // TODO: It would probably be better to check for self-reads before |
2179 | // calling the function. |
2180 | if (KillingDef == UseAccess || EarlierAccess == UseAccess) { |
2181 | LLVM_DEBUG(dbgs() << " ... skipping killing def/dom access\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping killing def/dom access\n" ; } } while (false); |
2182 | continue; |
2183 | } |
2184 | |
2185 | // Check all uses for MemoryDefs, except for defs completely overwriting |
2186 | // the original location. Otherwise we have to check uses of *all* |
2187 | // MemoryDefs we discover, including non-aliasing ones. Otherwise we might |
2188 | // miss cases like the following |
2189 | // 1 = Def(LoE) ; <----- EarlierDef stores [0,1] |
2190 | // 2 = Def(1) ; (2, 1) = NoAlias, stores [2,3] |
2191 | // Use(2) ; MayAlias 2 *and* 1, loads [0, 3]. |
2192 | // (The Use points to the *first* Def it may alias) |
2193 | // 3 = Def(1) ; <---- Current (3, 2) = NoAlias, (3,1) = MayAlias, |
2194 | // stores [0,1] |
2195 | if (MemoryDef *UseDef = dyn_cast<MemoryDef>(UseAccess)) { |
2196 | if (isCompleteOverwrite(DefLoc, KillingI, UseInst)) { |
2197 | if (!isInvisibleToCallerAfterRet(DefUO) && |
2198 | UseAccess != EarlierAccess) { |
2199 | BasicBlock *MaybeKillingBlock = UseInst->getParent(); |
2200 | if (PostOrderNumbers.find(MaybeKillingBlock)->second < |
2201 | PostOrderNumbers.find(EarlierAccess->getBlock())->second) { |
2202 | |
2203 | LLVM_DEBUG(dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found killing def " << *UseInst << "\n"; } } while (false) |
2204 | << " ... found killing def " << *UseInst << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found killing def " << *UseInst << "\n"; } } while (false); |
2205 | KillingDefs.insert(UseInst); |
2206 | } |
2207 | } |
2208 | } else |
2209 | PushMemUses(UseDef); |
2210 | } |
2211 | } |
2212 | |
2213 | // For accesses to locations visible after the function returns, make sure |
2214 | // that the location is killed (=overwritten) along all paths from |
2215 | // EarlierAccess to the exit. |
2216 | if (!isInvisibleToCallerAfterRet(DefUO)) { |
2217 | SmallPtrSet<BasicBlock *, 16> KillingBlocks; |
2218 | for (Instruction *KD : KillingDefs) |
2219 | KillingBlocks.insert(KD->getParent()); |
2220 | assert(!KillingBlocks.empty() &&((!KillingBlocks.empty() && "Expected at least a single killing block" ) ? static_cast<void> (0) : __assert_fail ("!KillingBlocks.empty() && \"Expected at least a single killing block\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2221, __PRETTY_FUNCTION__)) |
2221 | "Expected at least a single killing block")((!KillingBlocks.empty() && "Expected at least a single killing block" ) ? static_cast<void> (0) : __assert_fail ("!KillingBlocks.empty() && \"Expected at least a single killing block\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2221, __PRETTY_FUNCTION__)); |
2222 | |
2223 | // Find the common post-dominator of all killing blocks. |
2224 | BasicBlock *CommonPred = *KillingBlocks.begin(); |
2225 | for (auto I = std::next(KillingBlocks.begin()), E = KillingBlocks.end(); |
2226 | I != E; I++) { |
2227 | if (!CommonPred) |
2228 | break; |
2229 | CommonPred = PDT.findNearestCommonDominator(CommonPred, *I); |
2230 | } |
2231 | |
2232 | // If CommonPred is in the set of killing blocks, just check if it |
2233 | // post-dominates EarlierAccess. |
2234 | if (KillingBlocks.count(CommonPred)) { |
2235 | if (PDT.dominates(CommonPred, EarlierAccess->getBlock())) |
2236 | return {EarlierAccess}; |
2237 | return None; |
2238 | } |
2239 | |
2240 | // If the common post-dominator does not post-dominate EarlierAccess, |
2241 | // there is a path from EarlierAccess to an exit not going through a |
2242 | // killing block. |
2243 | if (PDT.dominates(CommonPred, EarlierAccess->getBlock())) { |
2244 | SetVector<BasicBlock *> WorkList; |
2245 | |
2246 | // If CommonPred is null, there are multiple exits from the function. |
2247 | // They all have to be added to the worklist. |
2248 | if (CommonPred) |
2249 | WorkList.insert(CommonPred); |
2250 | else |
2251 | for (BasicBlock *R : PDT.roots()) |
2252 | WorkList.insert(R); |
2253 | |
2254 | NumCFGTries++; |
2255 | // Check if all paths starting from an exit node go through one of the |
2256 | // killing blocks before reaching EarlierAccess. |
2257 | for (unsigned I = 0; I < WorkList.size(); I++) { |
2258 | NumCFGChecks++; |
2259 | BasicBlock *Current = WorkList[I]; |
2260 | if (KillingBlocks.count(Current)) |
2261 | continue; |
2262 | if (Current == EarlierAccess->getBlock()) |
2263 | return None; |
2264 | |
2265 | // EarlierAccess is reachable from the entry, so we don't have to |
2266 | // explore unreachable blocks further. |
2267 | if (!DT.isReachableFromEntry(Current)) |
2268 | continue; |
2269 | |
2270 | for (BasicBlock *Pred : predecessors(Current)) |
2271 | WorkList.insert(Pred); |
2272 | |
2273 | if (WorkList.size() >= MemorySSAPathCheckLimit) |
2274 | return None; |
2275 | } |
2276 | NumCFGSuccess++; |
2277 | return {EarlierAccess}; |
2278 | } |
2279 | return None; |
2280 | } |
2281 | |
2282 | // No aliasing MemoryUses of EarlierAccess found, EarlierAccess is |
2283 | // potentially dead. |
2284 | Cache.KnownNoReads.insert(KnownNoReads.begin(), KnownNoReads.end()); |
2285 | return {EarlierAccess}; |
2286 | } |
2287 | |
2288 | // Delete dead memory defs |
2289 | void deleteDeadInstruction(Instruction *SI) { |
2290 | MemorySSAUpdater Updater(&MSSA); |
2291 | SmallVector<Instruction *, 32> NowDeadInsts; |
2292 | NowDeadInsts.push_back(SI); |
2293 | --NumFastOther; |
2294 | |
2295 | while (!NowDeadInsts.empty()) { |
2296 | Instruction *DeadInst = NowDeadInsts.pop_back_val(); |
2297 | ++NumFastOther; |
2298 | |
2299 | // Try to preserve debug information attached to the dead instruction. |
2300 | salvageDebugInfo(*DeadInst); |
2301 | salvageKnowledge(DeadInst); |
2302 | |
2303 | // Remove the Instruction from MSSA. |
2304 | if (MemoryAccess *MA = MSSA.getMemoryAccess(DeadInst)) { |
2305 | if (MemoryDef *MD = dyn_cast<MemoryDef>(MA)) { |
2306 | SkipStores.insert(MD); |
2307 | } |
2308 | Updater.removeMemoryAccess(MA); |
2309 | } |
2310 | |
2311 | auto I = IOLs.find(DeadInst->getParent()); |
2312 | if (I != IOLs.end()) |
2313 | I->second.erase(DeadInst); |
2314 | // Remove its operands |
2315 | for (Use &O : DeadInst->operands()) |
2316 | if (Instruction *OpI = dyn_cast<Instruction>(O)) { |
2317 | O = nullptr; |
2318 | if (isInstructionTriviallyDead(OpI, &TLI)) |
2319 | NowDeadInsts.push_back(OpI); |
2320 | } |
2321 | |
2322 | DeadInst->eraseFromParent(); |
2323 | } |
2324 | } |
2325 | |
2326 | // Check for any extra throws between SI and NI that block DSE. This only |
2327 | // checks extra maythrows (those that aren't MemoryDef's). MemoryDef that may |
2328 | // throw are handled during the walk from one def to the next. |
2329 | bool mayThrowBetween(Instruction *SI, Instruction *NI, |
2330 | const Value *SILocUnd) { |
2331 | // First see if we can ignore it by using the fact that SI is an |
2332 | // alloca/alloca like object that is not visible to the caller during |
2333 | // execution of the function. |
2334 | if (SILocUnd && isInvisibleToCallerBeforeRet(SILocUnd)) |
2335 | return false; |
2336 | |
2337 | if (SI->getParent() == NI->getParent()) |
2338 | return ThrowingBlocks.count(SI->getParent()); |
2339 | return !ThrowingBlocks.empty(); |
2340 | } |
2341 | |
2342 | // Check if \p NI acts as a DSE barrier for \p SI. The following instructions |
2343 | // act as barriers: |
2344 | // * A memory instruction that may throw and \p SI accesses a non-stack |
2345 | // object. |
2346 | // * Atomic stores stronger that monotonic. |
2347 | bool isDSEBarrier(const Value *SILocUnd, Instruction *NI) { |
2348 | // If NI may throw it acts as a barrier, unless we are to an alloca/alloca |
2349 | // like object that does not escape. |
2350 | if (NI->mayThrow() && !isInvisibleToCallerBeforeRet(SILocUnd)) |
2351 | return true; |
2352 | |
2353 | // If NI is an atomic load/store stronger than monotonic, do not try to |
2354 | // eliminate/reorder it. |
2355 | if (NI->isAtomic()) { |
2356 | if (auto *LI = dyn_cast<LoadInst>(NI)) |
2357 | return isStrongerThanMonotonic(LI->getOrdering()); |
2358 | if (auto *SI = dyn_cast<StoreInst>(NI)) |
2359 | return isStrongerThanMonotonic(SI->getOrdering()); |
2360 | if (auto *ARMW = dyn_cast<AtomicRMWInst>(NI)) |
2361 | return isStrongerThanMonotonic(ARMW->getOrdering()); |
2362 | if (auto *CmpXchg = dyn_cast<AtomicCmpXchgInst>(NI)) |
2363 | return isStrongerThanMonotonic(CmpXchg->getSuccessOrdering()) || |
2364 | isStrongerThanMonotonic(CmpXchg->getFailureOrdering()); |
2365 | llvm_unreachable("other instructions should be skipped in MemorySSA")::llvm::llvm_unreachable_internal("other instructions should be skipped in MemorySSA" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2365); |
2366 | } |
2367 | return false; |
2368 | } |
2369 | |
2370 | /// Eliminate writes to objects that are not visible in the caller and are not |
2371 | /// accessed before returning from the function. |
2372 | bool eliminateDeadWritesAtEndOfFunction() { |
2373 | bool MadeChange = false; |
2374 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs at the end of the function\n" ; } } while (false) |
2375 | dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs at the end of the function\n" ; } } while (false) |
2376 | << "Trying to eliminate MemoryDefs at the end of the function\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs at the end of the function\n" ; } } while (false); |
2377 | for (int I = MemDefs.size() - 1; I >= 0; I--) { |
2378 | MemoryDef *Def = MemDefs[I]; |
2379 | if (SkipStores.find(Def) != SkipStores.end() || |
2380 | !isRemovable(Def->getMemoryInst())) |
2381 | continue; |
2382 | |
2383 | Instruction *DefI = Def->getMemoryInst(); |
2384 | SmallVector<const Value *, 4> Pointers; |
2385 | auto DefLoc = getLocForWriteEx(DefI); |
2386 | if (!DefLoc) |
2387 | continue; |
2388 | |
2389 | // NOTE: Currently eliminating writes at the end of a function is limited |
2390 | // to MemoryDefs with a single underlying object, to save compile-time. In |
2391 | // practice it appears the case with multiple underlying objects is very |
2392 | // uncommon. If it turns out to be important, we can use |
2393 | // getUnderlyingObjects here instead. |
2394 | const Value *UO = getUnderlyingObject(DefLoc->Ptr); |
2395 | if (!UO || !isInvisibleToCallerAfterRet(UO)) |
2396 | continue; |
2397 | |
2398 | if (isWriteAtEndOfFunction(Def)) { |
2399 | // See through pointer-to-pointer bitcasts |
2400 | LLVM_DEBUG(dbgs() << " ... MemoryDef is not accessed until the end "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... MemoryDef is not accessed until the end " "of the function\n"; } } while (false) |
2401 | "of the function\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... MemoryDef is not accessed until the end " "of the function\n"; } } while (false); |
2402 | deleteDeadInstruction(DefI); |
2403 | ++NumFastStores; |
2404 | MadeChange = true; |
2405 | } |
2406 | } |
2407 | return MadeChange; |
2408 | } |
2409 | |
2410 | /// \returns true if \p Def is a no-op store, either because it |
2411 | /// directly stores back a loaded value or stores zero to a calloced object. |
2412 | bool storeIsNoop(MemoryDef *Def, MemoryLocation DefLoc, const Value *DefUO) { |
2413 | StoreInst *Store = dyn_cast<StoreInst>(Def->getMemoryInst()); |
2414 | if (!Store) |
2415 | return false; |
2416 | |
2417 | if (auto *LoadI = dyn_cast<LoadInst>(Store->getOperand(0))) { |
2418 | if (LoadI->getPointerOperand() == Store->getOperand(1)) { |
2419 | // Get the defining access for the load. |
2420 | auto *LoadAccess = MSSA.getMemoryAccess(LoadI)->getDefiningAccess(); |
2421 | // Fast path: the defining accesses are the same. |
2422 | if (LoadAccess == Def->getDefiningAccess()) |
2423 | return true; |
2424 | |
2425 | // Look through phi accesses. Recursively scan all phi accesses by |
2426 | // adding them to a worklist. Bail when we run into a memory def that |
2427 | // does not match LoadAccess. |
2428 | SetVector<MemoryAccess *> ToCheck; |
2429 | MemoryAccess *Current = |
2430 | MSSA.getWalker()->getClobberingMemoryAccess(Def); |
2431 | // We don't want to bail when we run into the store memory def. But, |
2432 | // the phi access may point to it. So, pretend like we've already |
2433 | // checked it. |
2434 | ToCheck.insert(Def); |
2435 | ToCheck.insert(Current); |
2436 | // Start at current (1) to simulate already having checked Def. |
2437 | for (unsigned I = 1; I < ToCheck.size(); ++I) { |
2438 | Current = ToCheck[I]; |
2439 | if (auto PhiAccess = dyn_cast<MemoryPhi>(Current)) { |
2440 | // Check all the operands. |
2441 | for (auto &Use : PhiAccess->incoming_values()) |
2442 | ToCheck.insert(cast<MemoryAccess>(&Use)); |
2443 | continue; |
2444 | } |
2445 | |
2446 | // If we found a memory def, bail. This happens when we have an |
2447 | // unrelated write in between an otherwise noop store. |
2448 | assert(isa<MemoryDef>(Current) &&((isa<MemoryDef>(Current) && "Only MemoryDefs should reach here." ) ? static_cast<void> (0) : __assert_fail ("isa<MemoryDef>(Current) && \"Only MemoryDefs should reach here.\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2449, __PRETTY_FUNCTION__)) |
2449 | "Only MemoryDefs should reach here.")((isa<MemoryDef>(Current) && "Only MemoryDefs should reach here." ) ? static_cast<void> (0) : __assert_fail ("isa<MemoryDef>(Current) && \"Only MemoryDefs should reach here.\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2449, __PRETTY_FUNCTION__)); |
2450 | // TODO: Skip no alias MemoryDefs that have no aliasing reads. |
2451 | // We are searching for the definition of the store's destination. |
2452 | // So, if that is the same definition as the load, then this is a |
2453 | // noop. Otherwise, fail. |
2454 | if (LoadAccess != Current) |
2455 | return false; |
2456 | } |
2457 | return true; |
2458 | } |
2459 | } |
2460 | |
2461 | Constant *StoredConstant = dyn_cast<Constant>(Store->getOperand(0)); |
2462 | if (StoredConstant && StoredConstant->isNullValue()) { |
2463 | auto *DefUOInst = dyn_cast<Instruction>(DefUO); |
2464 | if (DefUOInst && isCallocLikeFn(DefUOInst, &TLI)) { |
2465 | auto *UnderlyingDef = cast<MemoryDef>(MSSA.getMemoryAccess(DefUOInst)); |
2466 | // If UnderlyingDef is the clobbering access of Def, no instructions |
2467 | // between them can modify the memory location. |
2468 | auto *ClobberDef = |
2469 | MSSA.getSkipSelfWalker()->getClobberingMemoryAccess(Def); |
2470 | return UnderlyingDef == ClobberDef; |
2471 | } |
2472 | } |
2473 | return false; |
2474 | } |
2475 | }; |
2476 | |
2477 | bool eliminateDeadStoresMemorySSA(Function &F, AliasAnalysis &AA, |
2478 | MemorySSA &MSSA, DominatorTree &DT, |
2479 | PostDominatorTree &PDT, |
2480 | const TargetLibraryInfo &TLI) { |
2481 | bool MadeChange = false; |
2482 | |
2483 | DSEState State = DSEState::get(F, AA, MSSA, DT, PDT, TLI); |
2484 | // For each store: |
2485 | for (unsigned I = 0; I < State.MemDefs.size(); I++) { |
2486 | MemoryDef *KillingDef = State.MemDefs[I]; |
2487 | if (State.SkipStores.count(KillingDef)) |
2488 | continue; |
2489 | Instruction *SI = KillingDef->getMemoryInst(); |
2490 | |
2491 | auto MaybeSILoc = State.getLocForWriteEx(SI); |
2492 | if (State.isMemTerminatorInst(SI)) |
2493 | MaybeSILoc = State.getLocForTerminator(SI).map( |
2494 | [](const std::pair<MemoryLocation, bool> &P) { return P.first; }); |
2495 | else |
2496 | MaybeSILoc = State.getLocForWriteEx(SI); |
2497 | |
2498 | if (!MaybeSILoc) { |
2499 | LLVM_DEBUG(dbgs() << "Failed to find analyzable write location for "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Failed to find analyzable write location for " << *SI << "\n"; } } while (false) |
2500 | << *SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Failed to find analyzable write location for " << *SI << "\n"; } } while (false); |
2501 | continue; |
2502 | } |
2503 | MemoryLocation SILoc = *MaybeSILoc; |
2504 | assert(SILoc.Ptr && "SILoc should not be null")((SILoc.Ptr && "SILoc should not be null") ? static_cast <void> (0) : __assert_fail ("SILoc.Ptr && \"SILoc should not be null\"" , "/build/llvm-toolchain-snapshot-12~++20201029100616+6c2ad4cf875/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2504, __PRETTY_FUNCTION__)); |
2505 | const Value *SILocUnd = getUnderlyingObject(SILoc.Ptr); |
2506 | |
2507 | // Check if the store is a no-op. |
2508 | if (isRemovable(SI) && State.storeIsNoop(KillingDef, SILoc, SILocUnd)) { |
2509 | LLVM_DEBUG(dbgs() << "DSE: Remove No-Op Store:\n DEAD: " << *SI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove No-Op Store:\n DEAD: " << *SI << '\n'; } } while (false); |
2510 | State.deleteDeadInstruction(SI); |
2511 | NumRedundantStores++; |
2512 | MadeChange = true; |
2513 | continue; |
2514 | } |
2515 | |
2516 | MemoryAccess *Current = KillingDef; |
Value stored to 'Current' during its initialization is never read | |
2517 | LLVM_DEBUG(dbgs() << "Trying to eliminate MemoryDefs killed by "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs killed by " << *KillingDef << " (" << *SI << ")\n" ; } } while (false) |
2518 | << *KillingDef << " (" << *SI << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs killed by " << *KillingDef << " (" << *SI << ")\n" ; } } while (false); |
2519 | |
2520 | unsigned ScanLimit = MemorySSAScanLimit; |
2521 | unsigned WalkerStepLimit = MemorySSAUpwardsStepLimit; |
2522 | unsigned PartialLimit = MemorySSAPartialStoreLimit; |
2523 | // Worklist of MemoryAccesses that may be killed by KillingDef. |
2524 | SetVector<MemoryAccess *> ToCheck; |
2525 | ToCheck.insert(KillingDef->getDefiningAccess()); |
2526 | |
2527 | if (!SILocUnd) |
2528 | continue; |
2529 | bool IsMemTerm = State.isMemTerminatorInst(SI); |
2530 | DSEState::CheckCache Cache; |
2531 | // Check if MemoryAccesses in the worklist are killed by KillingDef. |
2532 | for (unsigned I = 0; I < ToCheck.size(); I++) { |
2533 | Current = ToCheck[I]; |
2534 | if (State.SkipStores.count(Current)) |
2535 | continue; |
2536 | |
2537 | Optional<MemoryAccess *> Next = State.getDomMemoryDef( |
2538 | KillingDef, Current, SILoc, SILocUnd, Cache, ScanLimit, |
2539 | WalkerStepLimit, IsMemTerm, PartialLimit); |
2540 | |
2541 | if (!Next) { |
2542 | LLVM_DEBUG(dbgs() << " finished walk\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " finished walk\n"; } } while (false ); |
2543 | continue; |
2544 | } |
2545 | |
2546 | MemoryAccess *EarlierAccess = *Next; |
2547 | LLVM_DEBUG(dbgs() << " Checking if we can kill " << *EarlierAccess)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " Checking if we can kill " << *EarlierAccess; } } while (false); |
2548 | if (isa<MemoryPhi>(EarlierAccess)) { |
2549 | LLVM_DEBUG(dbgs() << "\n ... adding incoming values to worklist\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "\n ... adding incoming values to worklist\n" ; } } while (false); |
2550 | for (Value *V : cast<MemoryPhi>(EarlierAccess)->incoming_values()) { |
2551 | MemoryAccess *IncomingAccess = cast<MemoryAccess>(V); |
2552 | BasicBlock *IncomingBlock = IncomingAccess->getBlock(); |
2553 | BasicBlock *PhiBlock = EarlierAccess->getBlock(); |
2554 | |
2555 | // We only consider incoming MemoryAccesses that come before the |
2556 | // MemoryPhi. Otherwise we could discover candidates that do not |
2557 | // strictly dominate our starting def. |
2558 | if (State.PostOrderNumbers[IncomingBlock] > |
2559 | State.PostOrderNumbers[PhiBlock]) |
2560 | ToCheck.insert(IncomingAccess); |
2561 | } |
2562 | continue; |
2563 | } |
2564 | MemoryDef *NextDef = dyn_cast<MemoryDef>(EarlierAccess); |
2565 | Instruction *NI = NextDef->getMemoryInst(); |
2566 | LLVM_DEBUG(dbgs() << " (" << *NI << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " (" << *NI << ")\n"; } } while (false); |
2567 | ToCheck.insert(NextDef->getDefiningAccess()); |
2568 | NumGetDomMemoryDefPassed++; |
2569 | |
2570 | if (!DebugCounter::shouldExecute(MemorySSACounter)) |
2571 | continue; |
2572 | |
2573 | MemoryLocation NILoc = *State.getLocForWriteEx(NI); |
2574 | |
2575 | if (IsMemTerm) { |
2576 | const Value *NIUnd = getUnderlyingObject(NILoc.Ptr); |
2577 | if (SILocUnd != NIUnd) |
2578 | continue; |
2579 | LLVM_DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *NIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *NI << "\n KILLER: " << *SI << '\n' ; } } while (false) |
2580 | << "\n KILLER: " << *SI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *NI << "\n KILLER: " << *SI << '\n' ; } } while (false); |
2581 | State.deleteDeadInstruction(NI); |
2582 | ++NumFastStores; |
2583 | MadeChange = true; |
2584 | } else { |
2585 | // Check if NI overwrites SI. |
2586 | int64_t InstWriteOffset, DepWriteOffset; |
2587 | OverwriteResult OR = |
2588 | isOverwrite(SI, NI, SILoc, NILoc, State.DL, TLI, DepWriteOffset, |
2589 | InstWriteOffset, State.BatchAA, &F); |
2590 | if (OR == OW_MaybePartial) { |
2591 | auto Iter = State.IOLs.insert( |
2592 | std::make_pair<BasicBlock *, InstOverlapIntervalsTy>( |
2593 | NI->getParent(), InstOverlapIntervalsTy())); |
2594 | auto &IOL = Iter.first->second; |
2595 | OR = isPartialOverwrite(SILoc, NILoc, DepWriteOffset, InstWriteOffset, |
2596 | NI, IOL); |
2597 | } |
2598 | |
2599 | if (EnablePartialStoreMerging && OR == OW_PartialEarlierWithFullLater) { |
2600 | auto *Earlier = dyn_cast<StoreInst>(NI); |
2601 | auto *Later = dyn_cast<StoreInst>(SI); |
2602 | // We are re-using tryToMergePartialOverlappingStores, which requires |
2603 | // Earlier to domiante Later. |
2604 | // TODO: implement tryToMergeParialOverlappingStores using MemorySSA. |
2605 | if (Earlier && Later && DT.dominates(Earlier, Later)) { |
2606 | if (Constant *Merged = tryToMergePartialOverlappingStores( |
2607 | Earlier, Later, InstWriteOffset, DepWriteOffset, State.DL, |
2608 | State.BatchAA, &DT)) { |
2609 | |
2610 | // Update stored value of earlier store to merged constant. |
2611 | Earlier->setOperand(0, Merged); |
2612 | ++NumModifiedStores; |
2613 | MadeChange = true; |
2614 | |
2615 | // Remove later store and remove any outstanding overlap intervals |
2616 | // for the updated store. |
2617 | State.deleteDeadInstruction(Later); |
2618 | auto I = State.IOLs.find(Earlier->getParent()); |
2619 | if (I != State.IOLs.end()) |
2620 | I->second.erase(Earlier); |
2621 | break; |
2622 | } |
2623 | } |
2624 | } |
2625 | |
2626 | if (OR == OW_Complete) { |
2627 | LLVM_DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *NIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *NI << "\n KILLER: " << *SI << '\n' ; } } while (false) |
2628 | << "\n KILLER: " << *SI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Remove Dead Store:\n DEAD: " << *NI << "\n KILLER: " << *SI << '\n' ; } } while (false); |
2629 | State.deleteDeadInstruction(NI); |
2630 | ++NumFastStores; |
2631 | MadeChange = true; |
2632 | } |
2633 | } |
2634 | } |
2635 | } |
2636 | |
2637 | if (EnablePartialOverwriteTracking) |
2638 | for (auto &KV : State.IOLs) |
2639 | MadeChange |= removePartiallyOverlappedStores(State.DL, KV.second, TLI); |
2640 | |
2641 | MadeChange |= State.eliminateDeadWritesAtEndOfFunction(); |
2642 | return MadeChange; |
2643 | } |
2644 | } // end anonymous namespace |
2645 | |
2646 | //===----------------------------------------------------------------------===// |
2647 | // DSE Pass |
2648 | //===----------------------------------------------------------------------===// |
2649 | PreservedAnalyses DSEPass::run(Function &F, FunctionAnalysisManager &AM) { |
2650 | AliasAnalysis &AA = AM.getResult<AAManager>(F); |
2651 | const TargetLibraryInfo &TLI = AM.getResult<TargetLibraryAnalysis>(F); |
2652 | DominatorTree &DT = AM.getResult<DominatorTreeAnalysis>(F); |
2653 | |
2654 | bool Changed = false; |
2655 | if (EnableMemorySSA) { |
2656 | MemorySSA &MSSA = AM.getResult<MemorySSAAnalysis>(F).getMSSA(); |
2657 | PostDominatorTree &PDT = AM.getResult<PostDominatorTreeAnalysis>(F); |
2658 | |
2659 | Changed = eliminateDeadStoresMemorySSA(F, AA, MSSA, DT, PDT, TLI); |
2660 | } else { |
2661 | MemoryDependenceResults &MD = AM.getResult<MemoryDependenceAnalysis>(F); |
2662 | |
2663 | Changed = eliminateDeadStores(F, &AA, &MD, &DT, &TLI); |
2664 | } |
2665 | |
2666 | #ifdef LLVM_ENABLE_STATS1 |
2667 | if (AreStatisticsEnabled()) |
2668 | for (auto &I : instructions(F)) |
2669 | NumRemainingStores += isa<StoreInst>(&I); |
2670 | #endif |
2671 | |
2672 | if (!Changed) |
2673 | return PreservedAnalyses::all(); |
2674 | |
2675 | PreservedAnalyses PA; |
2676 | PA.preserveSet<CFGAnalyses>(); |
2677 | PA.preserve<GlobalsAA>(); |
2678 | if (EnableMemorySSA) |
2679 | PA.preserve<MemorySSAAnalysis>(); |
2680 | else |
2681 | PA.preserve<MemoryDependenceAnalysis>(); |
2682 | return PA; |
2683 | } |
2684 | |
2685 | namespace { |
2686 | |
2687 | /// A legacy pass for the legacy pass manager that wraps \c DSEPass. |
2688 | class DSELegacyPass : public FunctionPass { |
2689 | public: |
2690 | static char ID; // Pass identification, replacement for typeid |
2691 | |
2692 | DSELegacyPass() : FunctionPass(ID) { |
2693 | initializeDSELegacyPassPass(*PassRegistry::getPassRegistry()); |
2694 | } |
2695 | |
2696 | bool runOnFunction(Function &F) override { |
2697 | if (skipFunction(F)) |
2698 | return false; |
2699 | |
2700 | AliasAnalysis &AA = getAnalysis<AAResultsWrapperPass>().getAAResults(); |
2701 | DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree(); |
2702 | const TargetLibraryInfo &TLI = |
2703 | getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F); |
2704 | |
2705 | bool Changed = false; |
2706 | if (EnableMemorySSA) { |
2707 | MemorySSA &MSSA = getAnalysis<MemorySSAWrapperPass>().getMSSA(); |
2708 | PostDominatorTree &PDT = |
2709 | getAnalysis<PostDominatorTreeWrapperPass>().getPostDomTree(); |
2710 | |
2711 | Changed = eliminateDeadStoresMemorySSA(F, AA, MSSA, DT, PDT, TLI); |
2712 | } else { |
2713 | MemoryDependenceResults &MD = |
2714 | getAnalysis<MemoryDependenceWrapperPass>().getMemDep(); |
2715 | |
2716 | Changed = eliminateDeadStores(F, &AA, &MD, &DT, &TLI); |
2717 | } |
2718 | |
2719 | #ifdef LLVM_ENABLE_STATS1 |
2720 | if (AreStatisticsEnabled()) |
2721 | for (auto &I : instructions(F)) |
2722 | NumRemainingStores += isa<StoreInst>(&I); |
2723 | #endif |
2724 | |
2725 | return Changed; |
2726 | } |
2727 | |
2728 | void getAnalysisUsage(AnalysisUsage &AU) const override { |
2729 | AU.setPreservesCFG(); |
2730 | AU.addRequired<AAResultsWrapperPass>(); |
2731 | AU.addRequired<TargetLibraryInfoWrapperPass>(); |
2732 | AU.addPreserved<GlobalsAAWrapperPass>(); |
2733 | AU.addRequired<DominatorTreeWrapperPass>(); |
2734 | AU.addPreserved<DominatorTreeWrapperPass>(); |
2735 | |
2736 | if (EnableMemorySSA) { |
2737 | AU.addRequired<PostDominatorTreeWrapperPass>(); |
2738 | AU.addRequired<MemorySSAWrapperPass>(); |
2739 | AU.addPreserved<PostDominatorTreeWrapperPass>(); |
2740 | AU.addPreserved<MemorySSAWrapperPass>(); |
2741 | } else { |
2742 | AU.addRequired<MemoryDependenceWrapperPass>(); |
2743 | AU.addPreserved<MemoryDependenceWrapperPass>(); |
2744 | } |
2745 | } |
2746 | }; |
2747 | |
2748 | } // end anonymous namespace |
2749 | |
2750 | char DSELegacyPass::ID = 0; |
2751 | |
2752 | INITIALIZE_PASS_BEGIN(DSELegacyPass, "dse", "Dead Store Elimination", false,static void *initializeDSELegacyPassPassOnce(PassRegistry & Registry) { |
2753 | false)static void *initializeDSELegacyPassPassOnce(PassRegistry & Registry) { |
2754 | INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)initializeDominatorTreeWrapperPassPass(Registry); |
2755 | INITIALIZE_PASS_DEPENDENCY(PostDominatorTreeWrapperPass)initializePostDominatorTreeWrapperPassPass(Registry); |
2756 | INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)initializeAAResultsWrapperPassPass(Registry); |
2757 | INITIALIZE_PASS_DEPENDENCY(GlobalsAAWrapperPass)initializeGlobalsAAWrapperPassPass(Registry); |
2758 | INITIALIZE_PASS_DEPENDENCY(MemorySSAWrapperPass)initializeMemorySSAWrapperPassPass(Registry); |
2759 | INITIALIZE_PASS_DEPENDENCY(MemoryDependenceWrapperPass)initializeMemoryDependenceWrapperPassPass(Registry); |
2760 | INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)initializeTargetLibraryInfoWrapperPassPass(Registry); |
2761 | INITIALIZE_PASS_END(DSELegacyPass, "dse", "Dead Store Elimination", false,PassInfo *PI = new PassInfo( "Dead Store Elimination", "dse", &DSELegacyPass::ID, PassInfo::NormalCtor_t(callDefaultCtor <DSELegacyPass>), false, false); Registry.registerPass( *PI, true); return PI; } static llvm::once_flag InitializeDSELegacyPassPassFlag ; void llvm::initializeDSELegacyPassPass(PassRegistry &Registry ) { llvm::call_once(InitializeDSELegacyPassPassFlag, initializeDSELegacyPassPassOnce , std::ref(Registry)); } |
2762 | false)PassInfo *PI = new PassInfo( "Dead Store Elimination", "dse", &DSELegacyPass::ID, PassInfo::NormalCtor_t(callDefaultCtor <DSELegacyPass>), false, false); Registry.registerPass( *PI, true); return PI; } static llvm::once_flag InitializeDSELegacyPassPassFlag ; void llvm::initializeDSELegacyPassPass(PassRegistry &Registry ) { llvm::call_once(InitializeDSELegacyPassPassFlag, initializeDSELegacyPassPassOnce , std::ref(Registry)); } |
2763 | |
2764 | FunctionPass *llvm::createDeadStoreEliminationPass() { |
2765 | return new DSELegacyPass(); |
2766 | } |