File: | llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp |
Warning: | line 2573, column 25 Called C++ object pointer is null |
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-12.0.0~++20201102111116+1ed2ca68191/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-12.0.0~++20201102111116+1ed2ca68191/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-12.0.0~++20201102111116+1ed2ca68191/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-12.0.0~++20201102111116+1ed2ca68191/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 access completely overlaps the earlier store if and only if | |||
505 | // both start and end of the earlier one is "inside" the later one: | |||
506 | // |<->|--earlier--|<->| | |||
507 | // |-------later-------| | |||
508 | // Accesses may overlap if and only if start of one of them is "inside" | |||
509 | // another one: | |||
510 | // |<->|--earlier--|<----->| | |||
511 | // |-------later-------| | |||
512 | // OR | |||
513 | // |----- earlier -----| | |||
514 | // |<->|---later---|<----->| | |||
515 | // | |||
516 | // We have to be careful here as *Off is signed while *.Size is unsigned. | |||
517 | ||||
518 | // Check if the earlier access starts "not before" the later one. | |||
519 | if (EarlierOff >= LaterOff) { | |||
520 | // If the earlier access ends "not after" the later access then the earlier | |||
521 | // one is completely overwritten by the later one. | |||
522 | if (uint64_t(EarlierOff - LaterOff) + EarlierSize <= LaterSize) | |||
523 | return OW_Complete; | |||
524 | // If start of the earlier access is "before" end of the later access then | |||
525 | // accesses overlap. | |||
526 | else if ((uint64_t)(EarlierOff - LaterOff) < LaterSize) | |||
527 | return OW_MaybePartial; | |||
528 | } | |||
529 | // If start of the later access is "before" end of the earlier access then | |||
530 | // accesses overlap. | |||
531 | else if ((uint64_t)(LaterOff - EarlierOff) < EarlierSize) { | |||
532 | return OW_MaybePartial; | |||
533 | } | |||
534 | ||||
535 | // Can reach here only if accesses are known not to overlap. There is no | |||
536 | // dedicated code to indicate no overlap so signal "unknown". | |||
537 | return OW_Unknown; | |||
538 | } | |||
539 | ||||
540 | /// Return 'OW_Complete' if a store to the 'Later' location completely | |||
541 | /// overwrites a store to the 'Earlier' location, 'OW_End' if the end of the | |||
542 | /// 'Earlier' location is completely overwritten by 'Later', 'OW_Begin' if the | |||
543 | /// beginning of the 'Earlier' location is overwritten by 'Later'. | |||
544 | /// 'OW_PartialEarlierWithFullLater' means that an earlier (big) store was | |||
545 | /// overwritten by a latter (smaller) store which doesn't write outside the big | |||
546 | /// store's memory locations. Returns 'OW_Unknown' if nothing can be determined. | |||
547 | /// NOTE: This function must only be called if both \p Later and \p Earlier | |||
548 | /// write to the same underlying object with valid \p EarlierOff and \p | |||
549 | /// LaterOff. | |||
550 | static OverwriteResult isPartialOverwrite(const MemoryLocation &Later, | |||
551 | const MemoryLocation &Earlier, | |||
552 | int64_t EarlierOff, int64_t LaterOff, | |||
553 | Instruction *DepWrite, | |||
554 | InstOverlapIntervalsTy &IOL) { | |||
555 | const uint64_t LaterSize = Later.Size.getValue(); | |||
556 | const uint64_t EarlierSize = Earlier.Size.getValue(); | |||
557 | // We may now overlap, although the overlap is not complete. There might also | |||
558 | // be other incomplete overlaps, and together, they might cover the complete | |||
559 | // earlier write. | |||
560 | // Note: The correctness of this logic depends on the fact that this function | |||
561 | // is not even called providing DepWrite when there are any intervening reads. | |||
562 | if (EnablePartialOverwriteTracking && | |||
563 | LaterOff < int64_t(EarlierOff + EarlierSize) && | |||
564 | int64_t(LaterOff + LaterSize) >= EarlierOff) { | |||
565 | ||||
566 | // Insert our part of the overlap into the map. | |||
567 | auto &IM = IOL[DepWrite]; | |||
568 | 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) | |||
569 | << ", " << 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) | |||
570 | << ") 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) | |||
571 | << 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); | |||
572 | ||||
573 | // Make sure that we only insert non-overlapping intervals and combine | |||
574 | // adjacent intervals. The intervals are stored in the map with the ending | |||
575 | // offset as the key (in the half-open sense) and the starting offset as | |||
576 | // the value. | |||
577 | int64_t LaterIntStart = LaterOff, LaterIntEnd = LaterOff + LaterSize; | |||
578 | ||||
579 | // Find any intervals ending at, or after, LaterIntStart which start | |||
580 | // before LaterIntEnd. | |||
581 | auto ILI = IM.lower_bound(LaterIntStart); | |||
582 | if (ILI != IM.end() && ILI->second <= LaterIntEnd) { | |||
583 | // This existing interval is overlapped with the current store somewhere | |||
584 | // in [LaterIntStart, LaterIntEnd]. Merge them by erasing the existing | |||
585 | // intervals and adjusting our start and end. | |||
586 | LaterIntStart = std::min(LaterIntStart, ILI->second); | |||
587 | LaterIntEnd = std::max(LaterIntEnd, ILI->first); | |||
588 | ILI = IM.erase(ILI); | |||
589 | ||||
590 | // Continue erasing and adjusting our end in case other previous | |||
591 | // intervals are also overlapped with the current store. | |||
592 | // | |||
593 | // |--- ealier 1 ---| |--- ealier 2 ---| | |||
594 | // |------- later---------| | |||
595 | // | |||
596 | while (ILI != IM.end() && ILI->second <= LaterIntEnd) { | |||
597 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 597, __PRETTY_FUNCTION__)); | |||
598 | LaterIntEnd = std::max(LaterIntEnd, ILI->first); | |||
599 | ILI = IM.erase(ILI); | |||
600 | } | |||
601 | } | |||
602 | ||||
603 | IM[LaterIntEnd] = LaterIntStart; | |||
604 | ||||
605 | ILI = IM.begin(); | |||
606 | if (ILI->second <= EarlierOff && | |||
607 | ILI->first >= int64_t(EarlierOff + EarlierSize)) { | |||
608 | 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) | |||
609 | << 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) | |||
610 | << 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) | |||
611 | << ") 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) | |||
612 | << 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); | |||
613 | ++NumCompletePartials; | |||
614 | return OW_Complete; | |||
615 | } | |||
616 | } | |||
617 | ||||
618 | // Check for an earlier store which writes to all the memory locations that | |||
619 | // the later store writes to. | |||
620 | if (EnablePartialStoreMerging && LaterOff >= EarlierOff && | |||
621 | int64_t(EarlierOff + EarlierSize) > LaterOff && | |||
622 | uint64_t(LaterOff - EarlierOff) + LaterSize <= EarlierSize) { | |||
623 | 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) | |||
624 | << 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) | |||
625 | << 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) | |||
626 | << ") 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) | |||
627 | << 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); | |||
628 | // TODO: Maybe come up with a better name? | |||
629 | return OW_PartialEarlierWithFullLater; | |||
630 | } | |||
631 | ||||
632 | // Another interesting case is if the later store overwrites the end of the | |||
633 | // earlier store. | |||
634 | // | |||
635 | // |--earlier--| | |||
636 | // |-- later --| | |||
637 | // | |||
638 | // In this case we may want to trim the size of earlier to avoid generating | |||
639 | // writes to addresses which will definitely be overwritten later | |||
640 | if (!EnablePartialOverwriteTracking && | |||
641 | (LaterOff > EarlierOff && LaterOff < int64_t(EarlierOff + EarlierSize) && | |||
642 | int64_t(LaterOff + LaterSize) >= int64_t(EarlierOff + EarlierSize))) | |||
643 | return OW_End; | |||
644 | ||||
645 | // Finally, we also need to check if the later store overwrites the beginning | |||
646 | // of the earlier store. | |||
647 | // | |||
648 | // |--earlier--| | |||
649 | // |-- later --| | |||
650 | // | |||
651 | // In this case we may want to move the destination address and trim the size | |||
652 | // of earlier to avoid generating writes to addresses which will definitely | |||
653 | // be overwritten later. | |||
654 | if (!EnablePartialOverwriteTracking && | |||
655 | (LaterOff <= EarlierOff && int64_t(LaterOff + LaterSize) > EarlierOff)) { | |||
656 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 657, __PRETTY_FUNCTION__)) | |||
657 | "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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 657, __PRETTY_FUNCTION__)); | |||
658 | return OW_Begin; | |||
659 | } | |||
660 | // Otherwise, they don't completely overlap. | |||
661 | return OW_Unknown; | |||
662 | } | |||
663 | ||||
664 | /// If 'Inst' might be a self read (i.e. a noop copy of a | |||
665 | /// memory region into an identical pointer) then it doesn't actually make its | |||
666 | /// input dead in the traditional sense. Consider this case: | |||
667 | /// | |||
668 | /// memmove(A <- B) | |||
669 | /// memmove(A <- A) | |||
670 | /// | |||
671 | /// In this case, the second store to A does not make the first store to A dead. | |||
672 | /// The usual situation isn't an explicit A<-A store like this (which can be | |||
673 | /// trivially removed) but a case where two pointers may alias. | |||
674 | /// | |||
675 | /// This function detects when it is unsafe to remove a dependent instruction | |||
676 | /// because the DSE inducing instruction may be a self-read. | |||
677 | static bool isPossibleSelfRead(Instruction *Inst, | |||
678 | const MemoryLocation &InstStoreLoc, | |||
679 | Instruction *DepWrite, | |||
680 | const TargetLibraryInfo &TLI, | |||
681 | AliasAnalysis &AA) { | |||
682 | // Self reads can only happen for instructions that read memory. Get the | |||
683 | // location read. | |||
684 | MemoryLocation InstReadLoc = getLocForRead(Inst, TLI); | |||
685 | if (!InstReadLoc.Ptr) | |||
686 | return false; // Not a reading instruction. | |||
687 | ||||
688 | // If the read and written loc obviously don't alias, it isn't a read. | |||
689 | if (AA.isNoAlias(InstReadLoc, InstStoreLoc)) | |||
690 | return false; | |||
691 | ||||
692 | if (isa<AnyMemCpyInst>(Inst)) { | |||
693 | // LLVM's memcpy overlap semantics are not fully fleshed out (see PR11763) | |||
694 | // but in practice memcpy(A <- B) either means that A and B are disjoint or | |||
695 | // are equal (i.e. there are not partial overlaps). Given that, if we have: | |||
696 | // | |||
697 | // memcpy/memmove(A <- B) // DepWrite | |||
698 | // memcpy(A <- B) // Inst | |||
699 | // | |||
700 | // with Inst reading/writing a >= size than DepWrite, we can reason as | |||
701 | // follows: | |||
702 | // | |||
703 | // - If A == B then both the copies are no-ops, so the DepWrite can be | |||
704 | // removed. | |||
705 | // - If A != B then A and B are disjoint locations in Inst. Since | |||
706 | // Inst.size >= DepWrite.size A and B are disjoint in DepWrite too. | |||
707 | // Therefore DepWrite can be removed. | |||
708 | MemoryLocation DepReadLoc = getLocForRead(DepWrite, TLI); | |||
709 | ||||
710 | if (DepReadLoc.Ptr && AA.isMustAlias(InstReadLoc.Ptr, DepReadLoc.Ptr)) | |||
711 | return false; | |||
712 | } | |||
713 | ||||
714 | // If DepWrite doesn't read memory or if we can't prove it is a must alias, | |||
715 | // then it can't be considered dead. | |||
716 | return true; | |||
717 | } | |||
718 | ||||
719 | /// Returns true if the memory which is accessed by the second instruction is not | |||
720 | /// modified between the first and the second instruction. | |||
721 | /// Precondition: Second instruction must be dominated by the first | |||
722 | /// instruction. | |||
723 | template <typename AATy> | |||
724 | static bool | |||
725 | memoryIsNotModifiedBetween(Instruction *FirstI, Instruction *SecondI, AATy &AA, | |||
726 | const DataLayout &DL, DominatorTree *DT) { | |||
727 | // Do a backwards scan through the CFG from SecondI to FirstI. Look for | |||
728 | // instructions which can modify the memory location accessed by SecondI. | |||
729 | // | |||
730 | // While doing the walk keep track of the address to check. It might be | |||
731 | // different in different basic blocks due to PHI translation. | |||
732 | using BlockAddressPair = std::pair<BasicBlock *, PHITransAddr>; | |||
733 | SmallVector<BlockAddressPair, 16> WorkList; | |||
734 | // Keep track of the address we visited each block with. Bail out if we | |||
735 | // visit a block with different addresses. | |||
736 | DenseMap<BasicBlock *, Value *> Visited; | |||
737 | ||||
738 | BasicBlock::iterator FirstBBI(FirstI); | |||
739 | ++FirstBBI; | |||
740 | BasicBlock::iterator SecondBBI(SecondI); | |||
741 | BasicBlock *FirstBB = FirstI->getParent(); | |||
742 | BasicBlock *SecondBB = SecondI->getParent(); | |||
743 | MemoryLocation MemLoc = MemoryLocation::get(SecondI); | |||
744 | auto *MemLocPtr = const_cast<Value *>(MemLoc.Ptr); | |||
745 | ||||
746 | // Start checking the SecondBB. | |||
747 | WorkList.push_back( | |||
748 | std::make_pair(SecondBB, PHITransAddr(MemLocPtr, DL, nullptr))); | |||
749 | bool isFirstBlock = true; | |||
750 | ||||
751 | // Check all blocks going backward until we reach the FirstBB. | |||
752 | while (!WorkList.empty()) { | |||
753 | BlockAddressPair Current = WorkList.pop_back_val(); | |||
754 | BasicBlock *B = Current.first; | |||
755 | PHITransAddr &Addr = Current.second; | |||
756 | Value *Ptr = Addr.getAddr(); | |||
757 | ||||
758 | // Ignore instructions before FirstI if this is the FirstBB. | |||
759 | BasicBlock::iterator BI = (B == FirstBB ? FirstBBI : B->begin()); | |||
760 | ||||
761 | BasicBlock::iterator EI; | |||
762 | if (isFirstBlock) { | |||
763 | // Ignore instructions after SecondI if this is the first visit of SecondBB. | |||
764 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 764, __PRETTY_FUNCTION__)); | |||
765 | EI = SecondBBI; | |||
766 | isFirstBlock = false; | |||
767 | } else { | |||
768 | // It's not SecondBB or (in case of a loop) the second visit of SecondBB. | |||
769 | // In this case we also have to look at instructions after SecondI. | |||
770 | EI = B->end(); | |||
771 | } | |||
772 | for (; BI != EI; ++BI) { | |||
773 | Instruction *I = &*BI; | |||
774 | if (I->mayWriteToMemory() && I != SecondI) | |||
775 | if (isModSet(AA.getModRefInfo(I, MemLoc.getWithNewPtr(Ptr)))) | |||
776 | return false; | |||
777 | } | |||
778 | if (B != FirstBB) { | |||
779 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 780, __PRETTY_FUNCTION__)) | |||
780 | "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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 780, __PRETTY_FUNCTION__)); | |||
781 | for (auto PredI = pred_begin(B), PE = pred_end(B); PredI != PE; ++PredI) { | |||
782 | PHITransAddr PredAddr = Addr; | |||
783 | if (PredAddr.NeedsPHITranslationFromBlock(B)) { | |||
784 | if (!PredAddr.IsPotentiallyPHITranslatable()) | |||
785 | return false; | |||
786 | if (PredAddr.PHITranslateValue(B, *PredI, DT, false)) | |||
787 | return false; | |||
788 | } | |||
789 | Value *TranslatedPtr = PredAddr.getAddr(); | |||
790 | auto Inserted = Visited.insert(std::make_pair(*PredI, TranslatedPtr)); | |||
791 | if (!Inserted.second) { | |||
792 | // We already visited this block before. If it was with a different | |||
793 | // address - bail out! | |||
794 | if (TranslatedPtr != Inserted.first->second) | |||
795 | return false; | |||
796 | // ... otherwise just skip it. | |||
797 | continue; | |||
798 | } | |||
799 | WorkList.push_back(std::make_pair(*PredI, PredAddr)); | |||
800 | } | |||
801 | } | |||
802 | } | |||
803 | return true; | |||
804 | } | |||
805 | ||||
806 | /// Find all blocks that will unconditionally lead to the block BB and append | |||
807 | /// them to F. | |||
808 | static void findUnconditionalPreds(SmallVectorImpl<BasicBlock *> &Blocks, | |||
809 | BasicBlock *BB, DominatorTree *DT) { | |||
810 | for (pred_iterator I = pred_begin(BB), E = pred_end(BB); I != E; ++I) { | |||
811 | BasicBlock *Pred = *I; | |||
812 | if (Pred == BB) continue; | |||
813 | Instruction *PredTI = Pred->getTerminator(); | |||
814 | if (PredTI->getNumSuccessors() != 1) | |||
815 | continue; | |||
816 | ||||
817 | if (DT->isReachableFromEntry(Pred)) | |||
818 | Blocks.push_back(Pred); | |||
819 | } | |||
820 | } | |||
821 | ||||
822 | /// Handle frees of entire structures whose dependency is a store | |||
823 | /// to a field of that structure. | |||
824 | static bool handleFree(CallInst *F, AliasAnalysis *AA, | |||
825 | MemoryDependenceResults *MD, DominatorTree *DT, | |||
826 | const TargetLibraryInfo *TLI, | |||
827 | InstOverlapIntervalsTy &IOL, | |||
828 | MapVector<Instruction *, bool> &ThrowableInst) { | |||
829 | bool MadeChange = false; | |||
830 | ||||
831 | MemoryLocation Loc = MemoryLocation(F->getOperand(0)); | |||
832 | SmallVector<BasicBlock *, 16> Blocks; | |||
833 | Blocks.push_back(F->getParent()); | |||
834 | ||||
835 | while (!Blocks.empty()) { | |||
836 | BasicBlock *BB = Blocks.pop_back_val(); | |||
837 | Instruction *InstPt = BB->getTerminator(); | |||
838 | if (BB == F->getParent()) InstPt = F; | |||
839 | ||||
840 | MemDepResult Dep = | |||
841 | MD->getPointerDependencyFrom(Loc, false, InstPt->getIterator(), BB); | |||
842 | while (Dep.isDef() || Dep.isClobber()) { | |||
843 | Instruction *Dependency = Dep.getInst(); | |||
844 | if (!hasAnalyzableMemoryWrite(Dependency, *TLI) || | |||
845 | !isRemovable(Dependency)) | |||
846 | break; | |||
847 | ||||
848 | Value *DepPointer = | |||
849 | getUnderlyingObject(getStoredPointerOperand(Dependency, *TLI)); | |||
850 | ||||
851 | // Check for aliasing. | |||
852 | if (!AA->isMustAlias(F->getArgOperand(0), DepPointer)) | |||
853 | break; | |||
854 | ||||
855 | 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) | |||
856 | 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) | |||
857 | << *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); | |||
858 | ||||
859 | // DCE instructions only used to calculate that store. | |||
860 | BasicBlock::iterator BBI(Dependency); | |||
861 | deleteDeadInstruction(Dependency, &BBI, *MD, *TLI, IOL, | |||
862 | ThrowableInst); | |||
863 | ++NumFastStores; | |||
864 | MadeChange = true; | |||
865 | ||||
866 | // Inst's old Dependency is now deleted. Compute the next dependency, | |||
867 | // which may also be dead, as in | |||
868 | // s[0] = 0; | |||
869 | // s[1] = 0; // This has just been deleted. | |||
870 | // free(s); | |||
871 | Dep = MD->getPointerDependencyFrom(Loc, false, BBI, BB); | |||
872 | } | |||
873 | ||||
874 | if (Dep.isNonLocal()) | |||
875 | findUnconditionalPreds(Blocks, BB, DT); | |||
876 | } | |||
877 | ||||
878 | return MadeChange; | |||
879 | } | |||
880 | ||||
881 | /// Check to see if the specified location may alias any of the stack objects in | |||
882 | /// the DeadStackObjects set. If so, they become live because the location is | |||
883 | /// being loaded. | |||
884 | static void removeAccessedObjects(const MemoryLocation &LoadedLoc, | |||
885 | SmallSetVector<const Value *, 16> &DeadStackObjects, | |||
886 | const DataLayout &DL, AliasAnalysis *AA, | |||
887 | const TargetLibraryInfo *TLI, | |||
888 | const Function *F) { | |||
889 | const Value *UnderlyingPointer = getUnderlyingObject(LoadedLoc.Ptr); | |||
890 | ||||
891 | // A constant can't be in the dead pointer set. | |||
892 | if (isa<Constant>(UnderlyingPointer)) | |||
893 | return; | |||
894 | ||||
895 | // If the kill pointer can be easily reduced to an alloca, don't bother doing | |||
896 | // extraneous AA queries. | |||
897 | if (isa<AllocaInst>(UnderlyingPointer) || isa<Argument>(UnderlyingPointer)) { | |||
898 | DeadStackObjects.remove(UnderlyingPointer); | |||
899 | return; | |||
900 | } | |||
901 | ||||
902 | // Remove objects that could alias LoadedLoc. | |||
903 | DeadStackObjects.remove_if([&](const Value *I) { | |||
904 | // See if the loaded location could alias the stack location. | |||
905 | MemoryLocation StackLoc(I, getPointerSize(I, DL, *TLI, F)); | |||
906 | return !AA->isNoAlias(StackLoc, LoadedLoc); | |||
907 | }); | |||
908 | } | |||
909 | ||||
910 | /// Remove dead stores to stack-allocated locations in the function end block. | |||
911 | /// Ex: | |||
912 | /// %A = alloca i32 | |||
913 | /// ... | |||
914 | /// store i32 1, i32* %A | |||
915 | /// ret void | |||
916 | static bool handleEndBlock(BasicBlock &BB, AliasAnalysis *AA, | |||
917 | MemoryDependenceResults *MD, | |||
918 | const TargetLibraryInfo *TLI, | |||
919 | InstOverlapIntervalsTy &IOL, | |||
920 | MapVector<Instruction *, bool> &ThrowableInst) { | |||
921 | bool MadeChange = false; | |||
922 | ||||
923 | // Keep track of all of the stack objects that are dead at the end of the | |||
924 | // function. | |||
925 | SmallSetVector<const Value*, 16> DeadStackObjects; | |||
926 | ||||
927 | // Find all of the alloca'd pointers in the entry block. | |||
928 | BasicBlock &Entry = BB.getParent()->front(); | |||
929 | for (Instruction &I : Entry) { | |||
930 | if (isa<AllocaInst>(&I)) | |||
931 | DeadStackObjects.insert(&I); | |||
932 | ||||
933 | // Okay, so these are dead heap objects, but if the pointer never escapes | |||
934 | // then it's leaked by this function anyways. | |||
935 | else if (isAllocLikeFn(&I, TLI) && !PointerMayBeCaptured(&I, true, true)) | |||
936 | DeadStackObjects.insert(&I); | |||
937 | } | |||
938 | ||||
939 | // Treat byval or inalloca arguments the same, stores to them are dead at the | |||
940 | // end of the function. | |||
941 | for (Argument &AI : BB.getParent()->args()) | |||
942 | if (AI.hasPassPointeeByValueCopyAttr()) | |||
943 | DeadStackObjects.insert(&AI); | |||
944 | ||||
945 | const DataLayout &DL = BB.getModule()->getDataLayout(); | |||
946 | ||||
947 | // Scan the basic block backwards | |||
948 | for (BasicBlock::iterator BBI = BB.end(); BBI != BB.begin(); ){ | |||
949 | --BBI; | |||
950 | ||||
951 | // If we find a store, check to see if it points into a dead stack value. | |||
952 | if (hasAnalyzableMemoryWrite(&*BBI, *TLI) && isRemovable(&*BBI)) { | |||
953 | // See through pointer-to-pointer bitcasts | |||
954 | SmallVector<const Value *, 4> Pointers; | |||
955 | getUnderlyingObjects(getStoredPointerOperand(&*BBI, *TLI), Pointers); | |||
956 | ||||
957 | // Stores to stack values are valid candidates for removal. | |||
958 | bool AllDead = true; | |||
959 | for (const Value *Pointer : Pointers) | |||
960 | if (!DeadStackObjects.count(Pointer)) { | |||
961 | AllDead = false; | |||
962 | break; | |||
963 | } | |||
964 | ||||
965 | if (AllDead) { | |||
966 | Instruction *Dead = &*BBI; | |||
967 | ||||
968 | 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) | |||
969 | << *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) | |||
970 | 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) | |||
971 | 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) | |||
972 | 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) | |||
973 | 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) | |||
974 | 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) | |||
975 | 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) | |||
976 | 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) | |||
977 | } 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) | |||
978 | << '\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); | |||
979 | ||||
980 | // DCE instructions only used to calculate that store. | |||
981 | deleteDeadInstruction(Dead, &BBI, *MD, *TLI, IOL, ThrowableInst, | |||
982 | &DeadStackObjects); | |||
983 | ++NumFastStores; | |||
984 | MadeChange = true; | |||
985 | continue; | |||
986 | } | |||
987 | } | |||
988 | ||||
989 | // Remove any dead non-memory-mutating instructions. | |||
990 | if (isInstructionTriviallyDead(&*BBI, TLI)) { | |||
991 | 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) | |||
992 | << *&*BBI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "DSE: Removing trivially dead instruction:\n DEAD: " << *&*BBI << '\n'; } } while (false); | |||
993 | deleteDeadInstruction(&*BBI, &BBI, *MD, *TLI, IOL, ThrowableInst, | |||
994 | &DeadStackObjects); | |||
995 | ++NumFastOther; | |||
996 | MadeChange = true; | |||
997 | continue; | |||
998 | } | |||
999 | ||||
1000 | if (isa<AllocaInst>(BBI)) { | |||
1001 | // Remove allocas from the list of dead stack objects; there can't be | |||
1002 | // any references before the definition. | |||
1003 | DeadStackObjects.remove(&*BBI); | |||
1004 | continue; | |||
1005 | } | |||
1006 | ||||
1007 | if (auto *Call = dyn_cast<CallBase>(&*BBI)) { | |||
1008 | // Remove allocation function calls from the list of dead stack objects; | |||
1009 | // there can't be any references before the definition. | |||
1010 | if (isAllocLikeFn(&*BBI, TLI)) | |||
1011 | DeadStackObjects.remove(&*BBI); | |||
1012 | ||||
1013 | // If this call does not access memory, it can't be loading any of our | |||
1014 | // pointers. | |||
1015 | if (AA->doesNotAccessMemory(Call)) | |||
1016 | continue; | |||
1017 | ||||
1018 | // If the call might load from any of our allocas, then any store above | |||
1019 | // the call is live. | |||
1020 | DeadStackObjects.remove_if([&](const Value *I) { | |||
1021 | // See if the call site touches the value. | |||
1022 | return isRefSet(AA->getModRefInfo( | |||
1023 | Call, I, getPointerSize(I, DL, *TLI, BB.getParent()))); | |||
1024 | }); | |||
1025 | ||||
1026 | // If all of the allocas were clobbered by the call then we're not going | |||
1027 | // to find anything else to process. | |||
1028 | if (DeadStackObjects.empty()) | |||
1029 | break; | |||
1030 | ||||
1031 | continue; | |||
1032 | } | |||
1033 | ||||
1034 | // We can remove the dead stores, irrespective of the fence and its ordering | |||
1035 | // (release/acquire/seq_cst). Fences only constraints the ordering of | |||
1036 | // already visible stores, it does not make a store visible to other | |||
1037 | // threads. So, skipping over a fence does not change a store from being | |||
1038 | // dead. | |||
1039 | if (isa<FenceInst>(*BBI)) | |||
1040 | continue; | |||
1041 | ||||
1042 | MemoryLocation LoadedLoc; | |||
1043 | ||||
1044 | // If we encounter a use of the pointer, it is no longer considered dead | |||
1045 | if (LoadInst *L = dyn_cast<LoadInst>(BBI)) { | |||
1046 | if (!L->isUnordered()) // Be conservative with atomic/volatile load | |||
1047 | break; | |||
1048 | LoadedLoc = MemoryLocation::get(L); | |||
1049 | } else if (VAArgInst *V = dyn_cast<VAArgInst>(BBI)) { | |||
1050 | LoadedLoc = MemoryLocation::get(V); | |||
1051 | } else if (!BBI->mayReadFromMemory()) { | |||
1052 | // Instruction doesn't read memory. Note that stores that weren't removed | |||
1053 | // above will hit this case. | |||
1054 | continue; | |||
1055 | } else { | |||
1056 | // Unknown inst; assume it clobbers everything. | |||
1057 | break; | |||
1058 | } | |||
1059 | ||||
1060 | // Remove any allocas from the DeadPointer set that are loaded, as this | |||
1061 | // makes any stores above the access live. | |||
1062 | removeAccessedObjects(LoadedLoc, DeadStackObjects, DL, AA, TLI, BB.getParent()); | |||
1063 | ||||
1064 | // If all of the allocas were clobbered by the access then we're not going | |||
1065 | // to find anything else to process. | |||
1066 | if (DeadStackObjects.empty()) | |||
1067 | break; | |||
1068 | } | |||
1069 | ||||
1070 | return MadeChange; | |||
1071 | } | |||
1072 | ||||
1073 | static bool tryToShorten(Instruction *EarlierWrite, int64_t &EarlierOffset, | |||
1074 | int64_t &EarlierSize, int64_t LaterOffset, | |||
1075 | int64_t LaterSize, bool IsOverwriteEnd) { | |||
1076 | // TODO: base this on the target vector size so that if the earlier | |||
1077 | // store was too small to get vector writes anyway then its likely | |||
1078 | // a good idea to shorten it | |||
1079 | // Power of 2 vector writes are probably always a bad idea to optimize | |||
1080 | // as any store/memset/memcpy is likely using vector instructions so | |||
1081 | // shortening it to not vector size is likely to be slower | |||
1082 | auto *EarlierIntrinsic = cast<AnyMemIntrinsic>(EarlierWrite); | |||
1083 | unsigned EarlierWriteAlign = EarlierIntrinsic->getDestAlignment(); | |||
1084 | if (!IsOverwriteEnd) | |||
1085 | LaterOffset = int64_t(LaterOffset + LaterSize); | |||
1086 | ||||
1087 | if (!(isPowerOf2_64(LaterOffset) && EarlierWriteAlign <= LaterOffset) && | |||
1088 | !((EarlierWriteAlign != 0) && LaterOffset % EarlierWriteAlign == 0)) | |||
1089 | return false; | |||
1090 | ||||
1091 | int64_t NewLength = IsOverwriteEnd | |||
1092 | ? LaterOffset - EarlierOffset | |||
1093 | : EarlierSize - (LaterOffset - EarlierOffset); | |||
1094 | ||||
1095 | if (auto *AMI = dyn_cast<AtomicMemIntrinsic>(EarlierWrite)) { | |||
1096 | // When shortening an atomic memory intrinsic, the newly shortened | |||
1097 | // length must remain an integer multiple of the element size. | |||
1098 | const uint32_t ElementSize = AMI->getElementSizeInBytes(); | |||
1099 | if (0 != NewLength % ElementSize) | |||
1100 | return false; | |||
1101 | } | |||
1102 | ||||
1103 | 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) | |||
1104 | << (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) | |||
1105 | << *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) | |||
1106 | << ", " << 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); | |||
1107 | ||||
1108 | Value *EarlierWriteLength = EarlierIntrinsic->getLength(); | |||
1109 | Value *TrimmedLength = | |||
1110 | ConstantInt::get(EarlierWriteLength->getType(), NewLength); | |||
1111 | EarlierIntrinsic->setLength(TrimmedLength); | |||
1112 | ||||
1113 | EarlierSize = NewLength; | |||
1114 | if (!IsOverwriteEnd) { | |||
1115 | int64_t OffsetMoved = (LaterOffset - EarlierOffset); | |||
1116 | Value *Indices[1] = { | |||
1117 | ConstantInt::get(EarlierWriteLength->getType(), OffsetMoved)}; | |||
1118 | GetElementPtrInst *NewDestGEP = GetElementPtrInst::CreateInBounds( | |||
1119 | EarlierIntrinsic->getRawDest()->getType()->getPointerElementType(), | |||
1120 | EarlierIntrinsic->getRawDest(), Indices, "", EarlierWrite); | |||
1121 | NewDestGEP->setDebugLoc(EarlierIntrinsic->getDebugLoc()); | |||
1122 | EarlierIntrinsic->setDest(NewDestGEP); | |||
1123 | EarlierOffset = EarlierOffset + OffsetMoved; | |||
1124 | } | |||
1125 | return true; | |||
1126 | } | |||
1127 | ||||
1128 | static bool tryToShortenEnd(Instruction *EarlierWrite, | |||
1129 | OverlapIntervalsTy &IntervalMap, | |||
1130 | int64_t &EarlierStart, int64_t &EarlierSize) { | |||
1131 | if (IntervalMap.empty() || !isShortenableAtTheEnd(EarlierWrite)) | |||
1132 | return false; | |||
1133 | ||||
1134 | OverlapIntervalsTy::iterator OII = --IntervalMap.end(); | |||
1135 | int64_t LaterStart = OII->second; | |||
1136 | int64_t LaterSize = OII->first - LaterStart; | |||
1137 | ||||
1138 | if (LaterStart > EarlierStart && LaterStart < EarlierStart + EarlierSize && | |||
1139 | LaterStart + LaterSize >= EarlierStart + EarlierSize) { | |||
1140 | if (tryToShorten(EarlierWrite, EarlierStart, EarlierSize, LaterStart, | |||
1141 | LaterSize, true)) { | |||
1142 | IntervalMap.erase(OII); | |||
1143 | return true; | |||
1144 | } | |||
1145 | } | |||
1146 | return false; | |||
1147 | } | |||
1148 | ||||
1149 | static bool tryToShortenBegin(Instruction *EarlierWrite, | |||
1150 | OverlapIntervalsTy &IntervalMap, | |||
1151 | int64_t &EarlierStart, int64_t &EarlierSize) { | |||
1152 | if (IntervalMap.empty() || !isShortenableAtTheBeginning(EarlierWrite)) | |||
1153 | return false; | |||
1154 | ||||
1155 | OverlapIntervalsTy::iterator OII = IntervalMap.begin(); | |||
1156 | int64_t LaterStart = OII->second; | |||
1157 | int64_t LaterSize = OII->first - LaterStart; | |||
1158 | ||||
1159 | if (LaterStart <= EarlierStart && LaterStart + LaterSize > EarlierStart) { | |||
1160 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1161, __PRETTY_FUNCTION__)) | |||
1161 | "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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1161, __PRETTY_FUNCTION__)); | |||
1162 | if (tryToShorten(EarlierWrite, EarlierStart, EarlierSize, LaterStart, | |||
1163 | LaterSize, false)) { | |||
1164 | IntervalMap.erase(OII); | |||
1165 | return true; | |||
1166 | } | |||
1167 | } | |||
1168 | return false; | |||
1169 | } | |||
1170 | ||||
1171 | static bool removePartiallyOverlappedStores(const DataLayout &DL, | |||
1172 | InstOverlapIntervalsTy &IOL, | |||
1173 | const TargetLibraryInfo &TLI) { | |||
1174 | bool Changed = false; | |||
1175 | for (auto OI : IOL) { | |||
1176 | Instruction *EarlierWrite = OI.first; | |||
1177 | MemoryLocation Loc = getLocForWrite(EarlierWrite, TLI); | |||
1178 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1178, __PRETTY_FUNCTION__)); | |||
1179 | ||||
1180 | const Value *Ptr = Loc.Ptr->stripPointerCasts(); | |||
1181 | int64_t EarlierStart = 0; | |||
1182 | int64_t EarlierSize = int64_t(Loc.Size.getValue()); | |||
1183 | GetPointerBaseWithConstantOffset(Ptr, EarlierStart, DL); | |||
1184 | OverlapIntervalsTy &IntervalMap = OI.second; | |||
1185 | Changed |= | |||
1186 | tryToShortenEnd(EarlierWrite, IntervalMap, EarlierStart, EarlierSize); | |||
1187 | if (IntervalMap.empty()) | |||
1188 | continue; | |||
1189 | Changed |= | |||
1190 | tryToShortenBegin(EarlierWrite, IntervalMap, EarlierStart, EarlierSize); | |||
1191 | } | |||
1192 | return Changed; | |||
1193 | } | |||
1194 | ||||
1195 | static bool eliminateNoopStore(Instruction *Inst, BasicBlock::iterator &BBI, | |||
1196 | AliasAnalysis *AA, MemoryDependenceResults *MD, | |||
1197 | const DataLayout &DL, | |||
1198 | const TargetLibraryInfo *TLI, | |||
1199 | InstOverlapIntervalsTy &IOL, | |||
1200 | MapVector<Instruction *, bool> &ThrowableInst, | |||
1201 | DominatorTree *DT) { | |||
1202 | // Must be a store instruction. | |||
1203 | StoreInst *SI = dyn_cast<StoreInst>(Inst); | |||
1204 | if (!SI) | |||
1205 | return false; | |||
1206 | ||||
1207 | // If we're storing the same value back to a pointer that we just loaded from, | |||
1208 | // then the store can be removed. | |||
1209 | if (LoadInst *DepLoad = dyn_cast<LoadInst>(SI->getValueOperand())) { | |||
1210 | if (SI->getPointerOperand() == DepLoad->getPointerOperand() && | |||
1211 | isRemovable(SI) && | |||
1212 | memoryIsNotModifiedBetween(DepLoad, SI, *AA, DL, DT)) { | |||
1213 | ||||
1214 | 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) | |||
1215 | 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) | |||
1216 | << *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); | |||
1217 | ||||
1218 | deleteDeadInstruction(SI, &BBI, *MD, *TLI, IOL, ThrowableInst); | |||
1219 | ++NumRedundantStores; | |||
1220 | return true; | |||
1221 | } | |||
1222 | } | |||
1223 | ||||
1224 | // Remove null stores into the calloc'ed objects | |||
1225 | Constant *StoredConstant = dyn_cast<Constant>(SI->getValueOperand()); | |||
1226 | if (StoredConstant && StoredConstant->isNullValue() && isRemovable(SI)) { | |||
1227 | Instruction *UnderlyingPointer = | |||
1228 | dyn_cast<Instruction>(getUnderlyingObject(SI->getPointerOperand())); | |||
1229 | ||||
1230 | if (UnderlyingPointer && isCallocLikeFn(UnderlyingPointer, TLI) && | |||
1231 | memoryIsNotModifiedBetween(UnderlyingPointer, SI, *AA, DL, DT)) { | |||
1232 | 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) | |||
1233 | 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) | |||
1234 | << *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); | |||
1235 | ||||
1236 | deleteDeadInstruction(SI, &BBI, *MD, *TLI, IOL, ThrowableInst); | |||
1237 | ++NumRedundantStores; | |||
1238 | return true; | |||
1239 | } | |||
1240 | } | |||
1241 | return false; | |||
1242 | } | |||
1243 | ||||
1244 | template <typename AATy> | |||
1245 | static Constant *tryToMergePartialOverlappingStores( | |||
1246 | StoreInst *Earlier, StoreInst *Later, int64_t InstWriteOffset, | |||
1247 | int64_t DepWriteOffset, const DataLayout &DL, AATy &AA, DominatorTree *DT) { | |||
1248 | ||||
1249 | if (Earlier && isa<ConstantInt>(Earlier->getValueOperand()) && | |||
1250 | DL.typeSizeEqualsStoreSize(Earlier->getValueOperand()->getType()) && | |||
1251 | Later && isa<ConstantInt>(Later->getValueOperand()) && | |||
1252 | DL.typeSizeEqualsStoreSize(Later->getValueOperand()->getType()) && | |||
1253 | memoryIsNotModifiedBetween(Earlier, Later, AA, DL, DT)) { | |||
1254 | // If the store we find is: | |||
1255 | // a) partially overwritten by the store to 'Loc' | |||
1256 | // b) the later store is fully contained in the earlier one and | |||
1257 | // c) they both have a constant value | |||
1258 | // d) none of the two stores need padding | |||
1259 | // Merge the two stores, replacing the earlier store's value with a | |||
1260 | // merge of both values. | |||
1261 | // TODO: Deal with other constant types (vectors, etc), and probably | |||
1262 | // some mem intrinsics (if needed) | |||
1263 | ||||
1264 | APInt EarlierValue = | |||
1265 | cast<ConstantInt>(Earlier->getValueOperand())->getValue(); | |||
1266 | APInt LaterValue = cast<ConstantInt>(Later->getValueOperand())->getValue(); | |||
1267 | unsigned LaterBits = LaterValue.getBitWidth(); | |||
1268 | assert(EarlierValue.getBitWidth() > LaterValue.getBitWidth())((EarlierValue.getBitWidth() > LaterValue.getBitWidth()) ? static_cast<void> (0) : __assert_fail ("EarlierValue.getBitWidth() > LaterValue.getBitWidth()" , "/build/llvm-toolchain-snapshot-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1268, __PRETTY_FUNCTION__)); | |||
1269 | LaterValue = LaterValue.zext(EarlierValue.getBitWidth()); | |||
1270 | ||||
1271 | // Offset of the smaller store inside the larger store | |||
1272 | unsigned BitOffsetDiff = (InstWriteOffset - DepWriteOffset) * 8; | |||
1273 | unsigned LShiftAmount = DL.isBigEndian() ? EarlierValue.getBitWidth() - | |||
1274 | BitOffsetDiff - LaterBits | |||
1275 | : BitOffsetDiff; | |||
1276 | APInt Mask = APInt::getBitsSet(EarlierValue.getBitWidth(), LShiftAmount, | |||
1277 | LShiftAmount + LaterBits); | |||
1278 | // Clear the bits we'll be replacing, then OR with the smaller | |||
1279 | // store, shifted appropriately. | |||
1280 | APInt Merged = (EarlierValue & ~Mask) | (LaterValue << LShiftAmount); | |||
1281 | 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) | |||
1282 | << "\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) | |||
1283 | << "\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); | |||
1284 | return ConstantInt::get(Earlier->getValueOperand()->getType(), Merged); | |||
1285 | } | |||
1286 | return nullptr; | |||
1287 | } | |||
1288 | ||||
1289 | static bool eliminateDeadStores(BasicBlock &BB, AliasAnalysis *AA, | |||
1290 | MemoryDependenceResults *MD, DominatorTree *DT, | |||
1291 | const TargetLibraryInfo *TLI) { | |||
1292 | const DataLayout &DL = BB.getModule()->getDataLayout(); | |||
1293 | bool MadeChange = false; | |||
1294 | ||||
1295 | MapVector<Instruction *, bool> ThrowableInst; | |||
1296 | ||||
1297 | // A map of interval maps representing partially-overwritten value parts. | |||
1298 | InstOverlapIntervalsTy IOL; | |||
1299 | ||||
1300 | // Do a top-down walk on the BB. | |||
1301 | for (BasicBlock::iterator BBI = BB.begin(), BBE = BB.end(); BBI != BBE; ) { | |||
1302 | // Handle 'free' calls specially. | |||
1303 | if (CallInst *F = isFreeCall(&*BBI, TLI)) { | |||
1304 | MadeChange |= handleFree(F, AA, MD, DT, TLI, IOL, ThrowableInst); | |||
1305 | // Increment BBI after handleFree has potentially deleted instructions. | |||
1306 | // This ensures we maintain a valid iterator. | |||
1307 | ++BBI; | |||
1308 | continue; | |||
1309 | } | |||
1310 | ||||
1311 | Instruction *Inst = &*BBI++; | |||
1312 | ||||
1313 | if (Inst->mayThrow()) { | |||
1314 | ThrowableInst[Inst] = true; | |||
1315 | continue; | |||
1316 | } | |||
1317 | ||||
1318 | // Check to see if Inst writes to memory. If not, continue. | |||
1319 | if (!hasAnalyzableMemoryWrite(Inst, *TLI)) | |||
1320 | continue; | |||
1321 | ||||
1322 | // eliminateNoopStore will update in iterator, if necessary. | |||
1323 | if (eliminateNoopStore(Inst, BBI, AA, MD, DL, TLI, IOL, | |||
1324 | ThrowableInst, DT)) { | |||
1325 | MadeChange = true; | |||
1326 | continue; | |||
1327 | } | |||
1328 | ||||
1329 | // If we find something that writes memory, get its memory dependence. | |||
1330 | MemDepResult InstDep = MD->getDependency(Inst); | |||
1331 | ||||
1332 | // Ignore any store where we can't find a local dependence. | |||
1333 | // FIXME: cross-block DSE would be fun. :) | |||
1334 | if (!InstDep.isDef() && !InstDep.isClobber()) | |||
1335 | continue; | |||
1336 | ||||
1337 | // Figure out what location is being stored to. | |||
1338 | MemoryLocation Loc = getLocForWrite(Inst, *TLI); | |||
1339 | ||||
1340 | // If we didn't get a useful location, fail. | |||
1341 | if (!Loc.Ptr) | |||
1342 | continue; | |||
1343 | ||||
1344 | // Loop until we find a store we can eliminate or a load that | |||
1345 | // invalidates the analysis. Without an upper bound on the number of | |||
1346 | // instructions examined, this analysis can become very time-consuming. | |||
1347 | // However, the potential gain diminishes as we process more instructions | |||
1348 | // without eliminating any of them. Therefore, we limit the number of | |||
1349 | // instructions we look at. | |||
1350 | auto Limit = MD->getDefaultBlockScanLimit(); | |||
1351 | while (InstDep.isDef() || InstDep.isClobber()) { | |||
1352 | // Get the memory clobbered by the instruction we depend on. MemDep will | |||
1353 | // skip any instructions that 'Loc' clearly doesn't interact with. If we | |||
1354 | // end up depending on a may- or must-aliased load, then we can't optimize | |||
1355 | // away the store and we bail out. However, if we depend on something | |||
1356 | // that overwrites the memory location we *can* potentially optimize it. | |||
1357 | // | |||
1358 | // Find out what memory location the dependent instruction stores. | |||
1359 | Instruction *DepWrite = InstDep.getInst(); | |||
1360 | if (!hasAnalyzableMemoryWrite(DepWrite, *TLI)) | |||
1361 | break; | |||
1362 | MemoryLocation DepLoc = getLocForWrite(DepWrite, *TLI); | |||
1363 | // If we didn't get a useful location, or if it isn't a size, bail out. | |||
1364 | if (!DepLoc.Ptr) | |||
1365 | break; | |||
1366 | ||||
1367 | // Find the last throwable instruction not removed by call to | |||
1368 | // deleteDeadInstruction. | |||
1369 | Instruction *LastThrowing = nullptr; | |||
1370 | if (!ThrowableInst.empty()) | |||
1371 | LastThrowing = ThrowableInst.back().first; | |||
1372 | ||||
1373 | // Make sure we don't look past a call which might throw. This is an | |||
1374 | // issue because MemoryDependenceAnalysis works in the wrong direction: | |||
1375 | // it finds instructions which dominate the current instruction, rather than | |||
1376 | // instructions which are post-dominated by the current instruction. | |||
1377 | // | |||
1378 | // If the underlying object is a non-escaping memory allocation, any store | |||
1379 | // to it is dead along the unwind edge. Otherwise, we need to preserve | |||
1380 | // the store. | |||
1381 | if (LastThrowing && DepWrite->comesBefore(LastThrowing)) { | |||
1382 | const Value *Underlying = getUnderlyingObject(DepLoc.Ptr); | |||
1383 | bool IsStoreDeadOnUnwind = isa<AllocaInst>(Underlying); | |||
1384 | if (!IsStoreDeadOnUnwind) { | |||
1385 | // We're looking for a call to an allocation function | |||
1386 | // where the allocation doesn't escape before the last | |||
1387 | // throwing instruction; PointerMayBeCaptured | |||
1388 | // reasonably fast approximation. | |||
1389 | IsStoreDeadOnUnwind = isAllocLikeFn(Underlying, TLI) && | |||
1390 | !PointerMayBeCaptured(Underlying, false, true); | |||
1391 | } | |||
1392 | if (!IsStoreDeadOnUnwind) | |||
1393 | break; | |||
1394 | } | |||
1395 | ||||
1396 | // If we find a write that is a) removable (i.e., non-volatile), b) is | |||
1397 | // completely obliterated by the store to 'Loc', and c) which we know that | |||
1398 | // 'Inst' doesn't load from, then we can remove it. | |||
1399 | // Also try to merge two stores if a later one only touches memory written | |||
1400 | // to by the earlier one. | |||
1401 | if (isRemovable(DepWrite) && | |||
1402 | !isPossibleSelfRead(Inst, Loc, DepWrite, *TLI, *AA)) { | |||
1403 | int64_t InstWriteOffset, DepWriteOffset; | |||
1404 | OverwriteResult OR = isOverwrite(Inst, DepWrite, Loc, DepLoc, DL, *TLI, | |||
1405 | DepWriteOffset, InstWriteOffset, *AA, | |||
1406 | BB.getParent()); | |||
1407 | if (OR == OW_MaybePartial) | |||
1408 | OR = isPartialOverwrite(Loc, DepLoc, DepWriteOffset, InstWriteOffset, | |||
1409 | DepWrite, IOL); | |||
1410 | ||||
1411 | if (OR == OW_Complete) { | |||
1412 | 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) | |||
1413 | << "\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); | |||
1414 | ||||
1415 | // Delete the store and now-dead instructions that feed it. | |||
1416 | deleteDeadInstruction(DepWrite, &BBI, *MD, *TLI, IOL, | |||
1417 | ThrowableInst); | |||
1418 | ++NumFastStores; | |||
1419 | MadeChange = true; | |||
1420 | ||||
1421 | // We erased DepWrite; start over. | |||
1422 | InstDep = MD->getDependency(Inst); | |||
1423 | continue; | |||
1424 | } else if ((OR == OW_End && isShortenableAtTheEnd(DepWrite)) || | |||
1425 | ((OR == OW_Begin && | |||
1426 | isShortenableAtTheBeginning(DepWrite)))) { | |||
1427 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1429, __PRETTY_FUNCTION__)) | |||
1428 | "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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1429, __PRETTY_FUNCTION__)) | |||
1429 | "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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1429, __PRETTY_FUNCTION__)); | |||
1430 | // The overwrite result is known, so these must be known, too. | |||
1431 | int64_t EarlierSize = DepLoc.Size.getValue(); | |||
1432 | int64_t LaterSize = Loc.Size.getValue(); | |||
1433 | bool IsOverwriteEnd = (OR == OW_End); | |||
1434 | MadeChange |= tryToShorten(DepWrite, DepWriteOffset, EarlierSize, | |||
1435 | InstWriteOffset, LaterSize, IsOverwriteEnd); | |||
1436 | } else if (EnablePartialStoreMerging && | |||
1437 | OR == OW_PartialEarlierWithFullLater) { | |||
1438 | auto *Earlier = dyn_cast<StoreInst>(DepWrite); | |||
1439 | auto *Later = dyn_cast<StoreInst>(Inst); | |||
1440 | if (Constant *C = tryToMergePartialOverlappingStores( | |||
1441 | Earlier, Later, InstWriteOffset, DepWriteOffset, DL, *AA, | |||
1442 | DT)) { | |||
1443 | auto *SI = new StoreInst( | |||
1444 | C, Earlier->getPointerOperand(), false, Earlier->getAlign(), | |||
1445 | Earlier->getOrdering(), Earlier->getSyncScopeID(), DepWrite); | |||
1446 | ||||
1447 | unsigned MDToKeep[] = {LLVMContext::MD_dbg, LLVMContext::MD_tbaa, | |||
1448 | LLVMContext::MD_alias_scope, | |||
1449 | LLVMContext::MD_noalias, | |||
1450 | LLVMContext::MD_nontemporal}; | |||
1451 | SI->copyMetadata(*DepWrite, MDToKeep); | |||
1452 | ++NumModifiedStores; | |||
1453 | ||||
1454 | // Delete the old stores and now-dead instructions that feed them. | |||
1455 | deleteDeadInstruction(Inst, &BBI, *MD, *TLI, IOL, | |||
1456 | ThrowableInst); | |||
1457 | deleteDeadInstruction(DepWrite, &BBI, *MD, *TLI, IOL, | |||
1458 | ThrowableInst); | |||
1459 | MadeChange = true; | |||
1460 | ||||
1461 | // We erased DepWrite and Inst (Loc); start over. | |||
1462 | break; | |||
1463 | } | |||
1464 | } | |||
1465 | } | |||
1466 | ||||
1467 | // If this is a may-aliased store that is clobbering the store value, we | |||
1468 | // can keep searching past it for another must-aliased pointer that stores | |||
1469 | // to the same location. For example, in: | |||
1470 | // store -> P | |||
1471 | // store -> Q | |||
1472 | // store -> P | |||
1473 | // we can remove the first store to P even though we don't know if P and Q | |||
1474 | // alias. | |||
1475 | if (DepWrite == &BB.front()) break; | |||
1476 | ||||
1477 | // Can't look past this instruction if it might read 'Loc'. | |||
1478 | if (isRefSet(AA->getModRefInfo(DepWrite, Loc))) | |||
1479 | break; | |||
1480 | ||||
1481 | InstDep = MD->getPointerDependencyFrom(Loc, /*isLoad=*/ false, | |||
1482 | DepWrite->getIterator(), &BB, | |||
1483 | /*QueryInst=*/ nullptr, &Limit); | |||
1484 | } | |||
1485 | } | |||
1486 | ||||
1487 | if (EnablePartialOverwriteTracking) | |||
1488 | MadeChange |= removePartiallyOverlappedStores(DL, IOL, *TLI); | |||
1489 | ||||
1490 | // If this block ends in a return, unwind, or unreachable, all allocas are | |||
1491 | // dead at its end, which means stores to them are also dead. | |||
1492 | if (BB.getTerminator()->getNumSuccessors() == 0) | |||
1493 | MadeChange |= handleEndBlock(BB, AA, MD, TLI, IOL, ThrowableInst); | |||
1494 | ||||
1495 | return MadeChange; | |||
1496 | } | |||
1497 | ||||
1498 | static bool eliminateDeadStores(Function &F, AliasAnalysis *AA, | |||
1499 | MemoryDependenceResults *MD, DominatorTree *DT, | |||
1500 | const TargetLibraryInfo *TLI) { | |||
1501 | bool MadeChange = false; | |||
1502 | for (BasicBlock &BB : F) | |||
1503 | // Only check non-dead blocks. Dead blocks may have strange pointer | |||
1504 | // cycles that will confuse alias analysis. | |||
1505 | if (DT->isReachableFromEntry(&BB)) | |||
1506 | MadeChange |= eliminateDeadStores(BB, AA, MD, DT, TLI); | |||
1507 | ||||
1508 | return MadeChange; | |||
1509 | } | |||
1510 | ||||
1511 | namespace { | |||
1512 | //============================================================================= | |||
1513 | // MemorySSA backed dead store elimination. | |||
1514 | // | |||
1515 | // The code below implements dead store elimination using MemorySSA. It uses | |||
1516 | // the following general approach: given a MemoryDef, walk upwards to find | |||
1517 | // clobbering MemoryDefs that may be killed by the starting def. Then check | |||
1518 | // that there are no uses that may read the location of the original MemoryDef | |||
1519 | // in between both MemoryDefs. A bit more concretely: | |||
1520 | // | |||
1521 | // For all MemoryDefs StartDef: | |||
1522 | // 1. Get the next dominating clobbering MemoryDef (EarlierAccess) by walking | |||
1523 | // upwards. | |||
1524 | // 2. Check that there are no reads between EarlierAccess and the StartDef by | |||
1525 | // checking all uses starting at EarlierAccess and walking until we see | |||
1526 | // StartDef. | |||
1527 | // 3. For each found CurrentDef, check that: | |||
1528 | // 1. There are no barrier instructions between CurrentDef and StartDef (like | |||
1529 | // throws or stores with ordering constraints). | |||
1530 | // 2. StartDef is executed whenever CurrentDef is executed. | |||
1531 | // 3. StartDef completely overwrites CurrentDef. | |||
1532 | // 4. Erase CurrentDef from the function and MemorySSA. | |||
1533 | ||||
1534 | // Returns true if \p I is an intrisnic that does not read or write memory. | |||
1535 | bool isNoopIntrinsic(Instruction *I) { | |||
1536 | if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) { | |||
1537 | switch (II->getIntrinsicID()) { | |||
1538 | case Intrinsic::lifetime_start: | |||
1539 | case Intrinsic::lifetime_end: | |||
1540 | case Intrinsic::invariant_end: | |||
1541 | case Intrinsic::launder_invariant_group: | |||
1542 | case Intrinsic::assume: | |||
1543 | return true; | |||
1544 | case Intrinsic::dbg_addr: | |||
1545 | case Intrinsic::dbg_declare: | |||
1546 | case Intrinsic::dbg_label: | |||
1547 | case Intrinsic::dbg_value: | |||
1548 | llvm_unreachable("Intrinsic should not be modeled in MemorySSA")::llvm::llvm_unreachable_internal("Intrinsic should not be modeled in MemorySSA" , "/build/llvm-toolchain-snapshot-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 1548); | |||
1549 | default: | |||
1550 | return false; | |||
1551 | } | |||
1552 | } | |||
1553 | return false; | |||
1554 | } | |||
1555 | ||||
1556 | // Check if we can ignore \p D for DSE. | |||
1557 | bool canSkipDef(MemoryDef *D, bool DefVisibleToCaller) { | |||
1558 | Instruction *DI = D->getMemoryInst(); | |||
1559 | // Calls that only access inaccessible memory cannot read or write any memory | |||
1560 | // locations we consider for elimination. | |||
1561 | if (auto *CB = dyn_cast<CallBase>(DI)) | |||
1562 | if (CB->onlyAccessesInaccessibleMemory()) | |||
1563 | return true; | |||
1564 | ||||
1565 | // We can eliminate stores to locations not visible to the caller across | |||
1566 | // throwing instructions. | |||
1567 | if (DI->mayThrow() && !DefVisibleToCaller) | |||
1568 | return true; | |||
1569 | ||||
1570 | // We can remove the dead stores, irrespective of the fence and its ordering | |||
1571 | // (release/acquire/seq_cst). Fences only constraints the ordering of | |||
1572 | // already visible stores, it does not make a store visible to other | |||
1573 | // threads. So, skipping over a fence does not change a store from being | |||
1574 | // dead. | |||
1575 | if (isa<FenceInst>(DI)) | |||
1576 | return true; | |||
1577 | ||||
1578 | // Skip intrinsics that do not really read or modify memory. | |||
1579 | if (isNoopIntrinsic(D->getMemoryInst())) | |||
1580 | return true; | |||
1581 | ||||
1582 | return false; | |||
1583 | } | |||
1584 | ||||
1585 | struct DSEState { | |||
1586 | Function &F; | |||
1587 | AliasAnalysis &AA; | |||
1588 | ||||
1589 | /// The single BatchAA instance that is used to cache AA queries. It will | |||
1590 | /// not be invalidated over the whole run. This is safe, because: | |||
1591 | /// 1. Only memory writes are removed, so the alias cache for memory | |||
1592 | /// locations remains valid. | |||
1593 | /// 2. No new instructions are added (only instructions removed), so cached | |||
1594 | /// information for a deleted value cannot be accessed by a re-used new | |||
1595 | /// value pointer. | |||
1596 | BatchAAResults BatchAA; | |||
1597 | ||||
1598 | MemorySSA &MSSA; | |||
1599 | DominatorTree &DT; | |||
1600 | PostDominatorTree &PDT; | |||
1601 | const TargetLibraryInfo &TLI; | |||
1602 | const DataLayout &DL; | |||
1603 | ||||
1604 | // All MemoryDefs that potentially could kill other MemDefs. | |||
1605 | SmallVector<MemoryDef *, 64> MemDefs; | |||
1606 | // Any that should be skipped as they are already deleted | |||
1607 | SmallPtrSet<MemoryAccess *, 4> SkipStores; | |||
1608 | // Keep track of all of the objects that are invisible to the caller before | |||
1609 | // the function returns. | |||
1610 | // SmallPtrSet<const Value *, 16> InvisibleToCallerBeforeRet; | |||
1611 | DenseMap<const Value *, bool> InvisibleToCallerBeforeRet; | |||
1612 | // Keep track of all of the objects that are invisible to the caller after | |||
1613 | // the function returns. | |||
1614 | DenseMap<const Value *, bool> InvisibleToCallerAfterRet; | |||
1615 | // Keep track of blocks with throwing instructions not modeled in MemorySSA. | |||
1616 | SmallPtrSet<BasicBlock *, 16> ThrowingBlocks; | |||
1617 | // Post-order numbers for each basic block. Used to figure out if memory | |||
1618 | // accesses are executed before another access. | |||
1619 | DenseMap<BasicBlock *, unsigned> PostOrderNumbers; | |||
1620 | ||||
1621 | /// Keep track of instructions (partly) overlapping with killing MemoryDefs per | |||
1622 | /// basic block. | |||
1623 | DenseMap<BasicBlock *, InstOverlapIntervalsTy> IOLs; | |||
1624 | ||||
1625 | struct CheckCache { | |||
1626 | SmallPtrSet<MemoryAccess *, 16> KnownNoReads; | |||
1627 | SmallPtrSet<MemoryAccess *, 16> KnownReads; | |||
1628 | ||||
1629 | bool isKnownNoRead(MemoryAccess *A) const { | |||
1630 | return KnownNoReads.find(A) != KnownNoReads.end(); | |||
1631 | } | |||
1632 | bool isKnownRead(MemoryAccess *A) const { | |||
1633 | return KnownReads.find(A) != KnownReads.end(); | |||
1634 | } | |||
1635 | }; | |||
1636 | ||||
1637 | DSEState(Function &F, AliasAnalysis &AA, MemorySSA &MSSA, DominatorTree &DT, | |||
1638 | PostDominatorTree &PDT, const TargetLibraryInfo &TLI) | |||
1639 | : F(F), AA(AA), BatchAA(AA), MSSA(MSSA), DT(DT), PDT(PDT), TLI(TLI), | |||
1640 | DL(F.getParent()->getDataLayout()) {} | |||
1641 | ||||
1642 | static DSEState get(Function &F, AliasAnalysis &AA, MemorySSA &MSSA, | |||
1643 | DominatorTree &DT, PostDominatorTree &PDT, | |||
1644 | const TargetLibraryInfo &TLI) { | |||
1645 | DSEState State(F, AA, MSSA, DT, PDT, TLI); | |||
1646 | // Collect blocks with throwing instructions not modeled in MemorySSA and | |||
1647 | // alloc-like objects. | |||
1648 | unsigned PO = 0; | |||
1649 | for (BasicBlock *BB : post_order(&F)) { | |||
1650 | State.PostOrderNumbers[BB] = PO++; | |||
1651 | for (Instruction &I : *BB) { | |||
1652 | MemoryAccess *MA = MSSA.getMemoryAccess(&I); | |||
1653 | if (I.mayThrow() && !MA) | |||
1654 | State.ThrowingBlocks.insert(I.getParent()); | |||
1655 | ||||
1656 | auto *MD = dyn_cast_or_null<MemoryDef>(MA); | |||
1657 | if (MD && State.MemDefs.size() < MemorySSADefsPerBlockLimit && | |||
1658 | (State.getLocForWriteEx(&I) || State.isMemTerminatorInst(&I))) | |||
1659 | State.MemDefs.push_back(MD); | |||
1660 | } | |||
1661 | } | |||
1662 | ||||
1663 | // Treat byval or inalloca arguments the same as Allocas, stores to them are | |||
1664 | // dead at the end of the function. | |||
1665 | for (Argument &AI : F.args()) | |||
1666 | if (AI.hasPassPointeeByValueCopyAttr()) { | |||
1667 | // For byval, the caller doesn't know the address of the allocation. | |||
1668 | if (AI.hasByValAttr()) | |||
1669 | State.InvisibleToCallerBeforeRet.insert({&AI, true}); | |||
1670 | State.InvisibleToCallerAfterRet.insert({&AI, true}); | |||
1671 | } | |||
1672 | ||||
1673 | return State; | |||
1674 | } | |||
1675 | ||||
1676 | bool isInvisibleToCallerAfterRet(const Value *V) { | |||
1677 | if (isa<AllocaInst>(V)) | |||
1678 | return true; | |||
1679 | auto I = InvisibleToCallerAfterRet.insert({V, false}); | |||
1680 | if (I.second) { | |||
1681 | if (!isInvisibleToCallerBeforeRet(V)) { | |||
1682 | I.first->second = false; | |||
1683 | } else { | |||
1684 | auto *Inst = dyn_cast<Instruction>(V); | |||
1685 | if (Inst && isAllocLikeFn(Inst, &TLI)) | |||
1686 | I.first->second = !PointerMayBeCaptured(V, true, false); | |||
1687 | } | |||
1688 | } | |||
1689 | return I.first->second; | |||
1690 | } | |||
1691 | ||||
1692 | bool isInvisibleToCallerBeforeRet(const Value *V) { | |||
1693 | if (isa<AllocaInst>(V)) | |||
1694 | return true; | |||
1695 | auto I = InvisibleToCallerBeforeRet.insert({V, false}); | |||
1696 | if (I.second) { | |||
1697 | auto *Inst = dyn_cast<Instruction>(V); | |||
1698 | if (Inst && isAllocLikeFn(Inst, &TLI)) | |||
1699 | // NOTE: This could be made more precise by PointerMayBeCapturedBefore | |||
1700 | // with the killing MemoryDef. But we refrain from doing so for now to | |||
1701 | // limit compile-time and this does not cause any changes to the number | |||
1702 | // of stores removed on a large test set in practice. | |||
1703 | I.first->second = !PointerMayBeCaptured(V, false, true); | |||
1704 | } | |||
1705 | return I.first->second; | |||
1706 | } | |||
1707 | ||||
1708 | Optional<MemoryLocation> getLocForWriteEx(Instruction *I) const { | |||
1709 | if (!I->mayWriteToMemory()) | |||
1710 | return None; | |||
1711 | ||||
1712 | if (auto *MTI = dyn_cast<AnyMemIntrinsic>(I)) | |||
1713 | return {MemoryLocation::getForDest(MTI)}; | |||
1714 | ||||
1715 | if (auto *CB = dyn_cast<CallBase>(I)) { | |||
1716 | // If the functions may write to memory we do not know about, bail out. | |||
1717 | if (!CB->onlyAccessesArgMemory() && | |||
1718 | !CB->onlyAccessesInaccessibleMemOrArgMem()) | |||
1719 | return None; | |||
1720 | ||||
1721 | LibFunc LF; | |||
1722 | if (TLI.getLibFunc(*CB, LF) && TLI.has(LF)) { | |||
1723 | switch (LF) { | |||
1724 | case LibFunc_strcpy: | |||
1725 | case LibFunc_strncpy: | |||
1726 | case LibFunc_strcat: | |||
1727 | case LibFunc_strncat: | |||
1728 | return {MemoryLocation(CB->getArgOperand(0))}; | |||
1729 | default: | |||
1730 | break; | |||
1731 | } | |||
1732 | } | |||
1733 | switch (CB->getIntrinsicID()) { | |||
1734 | case Intrinsic::init_trampoline: | |||
1735 | return {MemoryLocation(CB->getArgOperand(0))}; | |||
1736 | case Intrinsic::masked_store: | |||
1737 | return {MemoryLocation::getForArgument(CB, 1, TLI)}; | |||
1738 | default: | |||
1739 | break; | |||
1740 | } | |||
1741 | return None; | |||
1742 | } | |||
1743 | ||||
1744 | return MemoryLocation::getOrNone(I); | |||
1745 | } | |||
1746 | ||||
1747 | /// Returns true if \p UseInst completely overwrites \p DefLoc | |||
1748 | /// (stored by \p DefInst). | |||
1749 | bool isCompleteOverwrite(MemoryLocation DefLoc, Instruction *DefInst, | |||
1750 | Instruction *UseInst) { | |||
1751 | // UseInst has a MemoryDef associated in MemorySSA. It's possible for a | |||
1752 | // MemoryDef to not write to memory, e.g. a volatile load is modeled as a | |||
1753 | // MemoryDef. | |||
1754 | if (!UseInst->mayWriteToMemory()) | |||
1755 | return false; | |||
1756 | ||||
1757 | if (auto *CB = dyn_cast<CallBase>(UseInst)) | |||
1758 | if (CB->onlyAccessesInaccessibleMemory()) | |||
1759 | return false; | |||
1760 | ||||
1761 | int64_t InstWriteOffset, DepWriteOffset; | |||
1762 | if (auto CC = getLocForWriteEx(UseInst)) | |||
1763 | return isOverwrite(UseInst, DefInst, *CC, DefLoc, DL, TLI, DepWriteOffset, | |||
1764 | InstWriteOffset, BatchAA, &F) == OW_Complete; | |||
1765 | return false; | |||
1766 | } | |||
1767 | ||||
1768 | /// Returns true if \p Def is not read before returning from the function. | |||
1769 | bool isWriteAtEndOfFunction(MemoryDef *Def) { | |||
1770 | 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) | |||
1771 | << *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) | |||
1772 | << ") 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); | |||
1773 | ||||
1774 | auto MaybeLoc = getLocForWriteEx(Def->getMemoryInst()); | |||
1775 | if (!MaybeLoc) { | |||
1776 | 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); | |||
1777 | return false; | |||
1778 | } | |||
1779 | ||||
1780 | SmallVector<MemoryAccess *, 4> WorkList; | |||
1781 | SmallPtrSet<MemoryAccess *, 8> Visited; | |||
1782 | auto PushMemUses = [&WorkList, &Visited](MemoryAccess *Acc) { | |||
1783 | if (!Visited.insert(Acc).second) | |||
1784 | return; | |||
1785 | for (Use &U : Acc->uses()) | |||
1786 | WorkList.push_back(cast<MemoryAccess>(U.getUser())); | |||
1787 | }; | |||
1788 | PushMemUses(Def); | |||
1789 | for (unsigned I = 0; I < WorkList.size(); I++) { | |||
1790 | if (WorkList.size() >= MemorySSAScanLimit) { | |||
1791 | LLVM_DEBUG(dbgs() << " ... hit exploration limit.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... hit exploration limit.\n"; } } while (false); | |||
1792 | return false; | |||
1793 | } | |||
1794 | ||||
1795 | MemoryAccess *UseAccess = WorkList[I]; | |||
1796 | // Simply adding the users of MemoryPhi to the worklist is not enough, | |||
1797 | // because we might miss read clobbers in different iterations of a loop, | |||
1798 | // for example. | |||
1799 | // TODO: Add support for phi translation to handle the loop case. | |||
1800 | if (isa<MemoryPhi>(UseAccess)) | |||
1801 | return false; | |||
1802 | ||||
1803 | // TODO: Checking for aliasing is expensive. Consider reducing the amount | |||
1804 | // of times this is called and/or caching it. | |||
1805 | Instruction *UseInst = cast<MemoryUseOrDef>(UseAccess)->getMemoryInst(); | |||
1806 | if (isReadClobber(*MaybeLoc, UseInst)) { | |||
1807 | LLVM_DEBUG(dbgs() << " ... hit read clobber " << *UseInst << ".\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... hit read clobber " << *UseInst << ".\n"; } } while (false); | |||
1808 | return false; | |||
1809 | } | |||
1810 | ||||
1811 | if (MemoryDef *UseDef = dyn_cast<MemoryDef>(UseAccess)) | |||
1812 | PushMemUses(UseDef); | |||
1813 | } | |||
1814 | return true; | |||
1815 | } | |||
1816 | ||||
1817 | /// If \p I is a memory terminator like llvm.lifetime.end or free, return a | |||
1818 | /// pair with the MemoryLocation terminated by \p I and a boolean flag | |||
1819 | /// indicating whether \p I is a free-like call. | |||
1820 | Optional<std::pair<MemoryLocation, bool>> | |||
1821 | getLocForTerminator(Instruction *I) const { | |||
1822 | uint64_t Len; | |||
1823 | Value *Ptr; | |||
1824 | if (match(I, m_Intrinsic<Intrinsic::lifetime_end>(m_ConstantInt(Len), | |||
1825 | m_Value(Ptr)))) | |||
1826 | return {std::make_pair(MemoryLocation(Ptr, Len), false)}; | |||
1827 | ||||
1828 | if (auto *CB = dyn_cast<CallBase>(I)) { | |||
1829 | if (isFreeCall(I, &TLI)) | |||
1830 | return {std::make_pair(MemoryLocation(CB->getArgOperand(0)), true)}; | |||
1831 | } | |||
1832 | ||||
1833 | return None; | |||
1834 | } | |||
1835 | ||||
1836 | /// Returns true if \p I is a memory terminator instruction like | |||
1837 | /// llvm.lifetime.end or free. | |||
1838 | bool isMemTerminatorInst(Instruction *I) const { | |||
1839 | IntrinsicInst *II = dyn_cast<IntrinsicInst>(I); | |||
1840 | return (II && II->getIntrinsicID() == Intrinsic::lifetime_end) || | |||
1841 | isFreeCall(I, &TLI); | |||
1842 | } | |||
1843 | ||||
1844 | /// Returns true if \p MaybeTerm is a memory terminator for \p Loc from | |||
1845 | /// instruction \p AccessI. | |||
1846 | bool isMemTerminator(MemoryLocation Loc, Instruction *AccessI, | |||
1847 | Instruction *MaybeTerm) { | |||
1848 | Optional<std::pair<MemoryLocation, bool>> MaybeTermLoc = | |||
1849 | getLocForTerminator(MaybeTerm); | |||
1850 | ||||
1851 | if (!MaybeTermLoc) | |||
1852 | return false; | |||
1853 | ||||
1854 | // If the terminator is a free-like call, all accesses to the underlying | |||
1855 | // object can be considered terminated. | |||
1856 | if (getUnderlyingObject(Loc.Ptr) != | |||
1857 | getUnderlyingObject(MaybeTermLoc->first.Ptr)) | |||
1858 | return false; | |||
1859 | ||||
1860 | auto TermLoc = MaybeTermLoc->first; | |||
1861 | if (MaybeTermLoc->second) { | |||
1862 | const Value *LocUO = getUnderlyingObject(Loc.Ptr); | |||
1863 | return BatchAA.isMustAlias(TermLoc.Ptr, LocUO); | |||
1864 | } | |||
1865 | int64_t InstWriteOffset, DepWriteOffset; | |||
1866 | return isOverwrite(MaybeTerm, AccessI, TermLoc, Loc, DL, TLI, | |||
1867 | DepWriteOffset, InstWriteOffset, BatchAA, | |||
1868 | &F) == OW_Complete; | |||
1869 | } | |||
1870 | ||||
1871 | // Returns true if \p Use may read from \p DefLoc. | |||
1872 | bool isReadClobber(MemoryLocation DefLoc, Instruction *UseInst) { | |||
1873 | if (isNoopIntrinsic(UseInst)) | |||
1874 | return false; | |||
1875 | ||||
1876 | // Monotonic or weaker atomic stores can be re-ordered and do not need to be | |||
1877 | // treated as read clobber. | |||
1878 | if (auto SI = dyn_cast<StoreInst>(UseInst)) | |||
1879 | return isStrongerThan(SI->getOrdering(), AtomicOrdering::Monotonic); | |||
1880 | ||||
1881 | if (!UseInst->mayReadFromMemory()) | |||
1882 | return false; | |||
1883 | ||||
1884 | if (auto *CB = dyn_cast<CallBase>(UseInst)) | |||
1885 | if (CB->onlyAccessesInaccessibleMemory()) | |||
1886 | return false; | |||
1887 | ||||
1888 | // NOTE: For calls, the number of stores removed could be slightly improved | |||
1889 | // by using AA.callCapturesBefore(UseInst, DefLoc, &DT), but that showed to | |||
1890 | // be expensive compared to the benefits in practice. For now, avoid more | |||
1891 | // expensive analysis to limit compile-time. | |||
1892 | return isRefSet(BatchAA.getModRefInfo(UseInst, DefLoc)); | |||
1893 | } | |||
1894 | ||||
1895 | /// Returns true if \p Ptr is guaranteed to be loop invariant for any possible | |||
1896 | /// loop. In particular, this guarantees that it only references a single | |||
1897 | /// MemoryLocation during execution of the containing function. | |||
1898 | bool IsGuaranteedLoopInvariant(Value *Ptr) { | |||
1899 | auto IsGuaranteedLoopInvariantBase = [this](Value *Ptr) { | |||
1900 | Ptr = Ptr->stripPointerCasts(); | |||
1901 | if (auto *I = dyn_cast<Instruction>(Ptr)) { | |||
1902 | if (isa<AllocaInst>(Ptr)) | |||
1903 | return true; | |||
1904 | ||||
1905 | if (isAllocLikeFn(I, &TLI)) | |||
1906 | return true; | |||
1907 | ||||
1908 | return false; | |||
1909 | } | |||
1910 | return true; | |||
1911 | }; | |||
1912 | ||||
1913 | Ptr = Ptr->stripPointerCasts(); | |||
1914 | if (auto *GEP = dyn_cast<GEPOperator>(Ptr)) { | |||
1915 | return IsGuaranteedLoopInvariantBase(GEP->getPointerOperand()) && | |||
1916 | GEP->hasAllConstantIndices(); | |||
1917 | } | |||
1918 | return IsGuaranteedLoopInvariantBase(Ptr); | |||
1919 | } | |||
1920 | ||||
1921 | // Find a MemoryDef writing to \p DefLoc and dominating \p StartAccess, with | |||
1922 | // no read access between them or on any other path to a function exit block | |||
1923 | // if \p DefLoc is not accessible after the function returns. If there is no | |||
1924 | // such MemoryDef, return None. The returned value may not (completely) | |||
1925 | // overwrite \p DefLoc. Currently we bail out when we encounter an aliasing | |||
1926 | // MemoryUse (read). | |||
1927 | Optional<MemoryAccess *> | |||
1928 | getDomMemoryDef(MemoryDef *KillingDef, MemoryAccess *StartAccess, | |||
1929 | MemoryLocation DefLoc, const Value *DefUO, CheckCache &Cache, | |||
1930 | unsigned &ScanLimit, unsigned &WalkerStepLimit, | |||
1931 | bool IsMemTerm, unsigned &PartialLimit) { | |||
1932 | if (ScanLimit == 0 || WalkerStepLimit == 0) { | |||
1933 | LLVM_DEBUG(dbgs() << "\n ... hit scan limit\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "\n ... hit scan limit\n"; } } while (false); | |||
1934 | return None; | |||
1935 | } | |||
1936 | ||||
1937 | MemoryAccess *Current = StartAccess; | |||
1938 | Instruction *KillingI = KillingDef->getMemoryInst(); | |||
1939 | bool StepAgain; | |||
1940 | 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); | |||
1941 | ||||
1942 | // Find the next clobbering Mod access for DefLoc, starting at StartAccess. | |||
1943 | do { | |||
1944 | StepAgain = false; | |||
1945 | 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) | |||
1946 | 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) | |||
1947 | 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) | |||
1948 | 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) | |||
1949 | << ")";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) | |||
1950 | 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) | |||
1951 | })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); | |||
1952 | ||||
1953 | // Reached TOP. | |||
1954 | if (MSSA.isLiveOnEntryDef(Current)) { | |||
1955 | LLVM_DEBUG(dbgs() << " ... found LiveOnEntryDef\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found LiveOnEntryDef\n"; } } while (false); | |||
1956 | return None; | |||
1957 | } | |||
1958 | ||||
1959 | // Cost of a step. Accesses in the same block are more likely to be valid | |||
1960 | // candidates for elimination, hence consider them cheaper. | |||
1961 | unsigned StepCost = KillingDef->getBlock() == Current->getBlock() | |||
1962 | ? MemorySSASameBBStepCost | |||
1963 | : MemorySSAOtherBBStepCost; | |||
1964 | if (WalkerStepLimit <= StepCost) { | |||
1965 | LLVM_DEBUG(dbgs() << " ... hit walker step limit\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... hit walker step limit\n"; } } while (false); | |||
1966 | return None; | |||
1967 | } | |||
1968 | WalkerStepLimit -= StepCost; | |||
1969 | ||||
1970 | // Return for MemoryPhis. They cannot be eliminated directly and the | |||
1971 | // caller is responsible for traversing them. | |||
1972 | if (isa<MemoryPhi>(Current)) { | |||
1973 | LLVM_DEBUG(dbgs() << " ... found MemoryPhi\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found MemoryPhi\n"; } } while (false); | |||
1974 | return Current; | |||
1975 | } | |||
1976 | ||||
1977 | // Below, check if CurrentDef is a valid candidate to be eliminated by | |||
1978 | // KillingDef. If it is not, check the next candidate. | |||
1979 | MemoryDef *CurrentDef = cast<MemoryDef>(Current); | |||
1980 | Instruction *CurrentI = CurrentDef->getMemoryInst(); | |||
1981 | ||||
1982 | if (canSkipDef(CurrentDef, !isInvisibleToCallerBeforeRet(DefUO))) { | |||
1983 | StepAgain = true; | |||
1984 | Current = CurrentDef->getDefiningAccess(); | |||
1985 | continue; | |||
1986 | } | |||
1987 | ||||
1988 | // Before we try to remove anything, check for any extra throwing | |||
1989 | // instructions that block us from DSEing | |||
1990 | if (mayThrowBetween(KillingI, CurrentI, DefUO)) { | |||
1991 | LLVM_DEBUG(dbgs() << " ... skip, may throw!\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, may throw!\n"; } } while (false); | |||
1992 | return None; | |||
1993 | } | |||
1994 | ||||
1995 | // Check for anything that looks like it will be a barrier to further | |||
1996 | // removal | |||
1997 | if (isDSEBarrier(DefUO, CurrentI)) { | |||
1998 | LLVM_DEBUG(dbgs() << " ... skip, barrier\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, barrier\n"; } } while (false); | |||
1999 | return None; | |||
2000 | } | |||
2001 | ||||
2002 | // If Current is known to be on path that reads DefLoc or is a read | |||
2003 | // clobber, bail out, as the path is not profitable. We skip this check | |||
2004 | // for intrinsic calls, because the code knows how to handle memcpy | |||
2005 | // intrinsics. | |||
2006 | if (!isa<IntrinsicInst>(CurrentI) && | |||
2007 | (Cache.KnownReads.contains(Current) || | |||
2008 | isReadClobber(DefLoc, CurrentI))) { | |||
2009 | Cache.KnownReads.insert(Current); | |||
2010 | return None; | |||
2011 | } | |||
2012 | ||||
2013 | // Quick check if there are direct uses that are read-clobbers. | |||
2014 | if (any_of(Current->uses(), [this, &DefLoc, StartAccess](Use &U) { | |||
2015 | if (auto *UseOrDef = dyn_cast<MemoryUseOrDef>(U.getUser())) | |||
2016 | return !MSSA.dominates(StartAccess, UseOrDef) && | |||
2017 | isReadClobber(DefLoc, UseOrDef->getMemoryInst()); | |||
2018 | return false; | |||
2019 | })) { | |||
2020 | Cache.KnownReads.insert(Current); | |||
2021 | LLVM_DEBUG(dbgs() << " ... found a read clobber\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found a read clobber\n"; } } while (false); | |||
2022 | return None; | |||
2023 | } | |||
2024 | ||||
2025 | // If Current cannot be analyzed or is not removable, check the next | |||
2026 | // candidate. | |||
2027 | if (!hasAnalyzableMemoryWrite(CurrentI, TLI) || !isRemovable(CurrentI)) { | |||
2028 | StepAgain = true; | |||
2029 | Current = CurrentDef->getDefiningAccess(); | |||
2030 | continue; | |||
2031 | } | |||
2032 | ||||
2033 | // If Current does not have an analyzable write location, skip it | |||
2034 | auto CurrentLoc = getLocForWriteEx(CurrentI); | |||
2035 | if (!CurrentLoc) { | |||
2036 | StepAgain = true; | |||
2037 | Current = CurrentDef->getDefiningAccess(); | |||
2038 | continue; | |||
2039 | } | |||
2040 | ||||
2041 | if (IsMemTerm) { | |||
2042 | // If the killing def is a memory terminator (e.g. lifetime.end), check | |||
2043 | // the next candidate if the current Current does not write the same | |||
2044 | // underlying object as the terminator. | |||
2045 | if (!isMemTerminator(*CurrentLoc, CurrentI, KillingI)) { | |||
2046 | StepAgain = true; | |||
2047 | Current = CurrentDef->getDefiningAccess(); | |||
2048 | } | |||
2049 | continue; | |||
2050 | } else { | |||
2051 | // AliasAnalysis does not account for loops. Limit elimination to | |||
2052 | // candidates for which we can guarantee they always store to the same | |||
2053 | // memory location and not multiple locations in a loop. | |||
2054 | if (Current->getBlock() != KillingDef->getBlock() && | |||
2055 | !IsGuaranteedLoopInvariant(const_cast<Value *>(CurrentLoc->Ptr))) { | |||
2056 | StepAgain = true; | |||
2057 | Current = CurrentDef->getDefiningAccess(); | |||
2058 | WalkerStepLimit -= 1; | |||
2059 | continue; | |||
2060 | } | |||
2061 | ||||
2062 | int64_t InstWriteOffset, DepWriteOffset; | |||
2063 | auto OR = isOverwrite(KillingI, CurrentI, DefLoc, *CurrentLoc, DL, TLI, | |||
2064 | DepWriteOffset, InstWriteOffset, BatchAA, &F); | |||
2065 | // If Current does not write to the same object as KillingDef, check | |||
2066 | // the next candidate. | |||
2067 | if (OR == OW_Unknown) { | |||
2068 | StepAgain = true; | |||
2069 | Current = CurrentDef->getDefiningAccess(); | |||
2070 | } else if (OR == OW_MaybePartial) { | |||
2071 | // If KillingDef only partially overwrites Current, check the next | |||
2072 | // candidate if the partial step limit is exceeded. This aggressively | |||
2073 | // limits the number of candidates for partial store elimination, | |||
2074 | // which are less likely to be removable in the end. | |||
2075 | if (PartialLimit <= 1) { | |||
2076 | StepAgain = true; | |||
2077 | Current = CurrentDef->getDefiningAccess(); | |||
2078 | WalkerStepLimit -= 1; | |||
2079 | continue; | |||
2080 | } | |||
2081 | PartialLimit -= 1; | |||
2082 | } | |||
2083 | } | |||
2084 | } while (StepAgain); | |||
2085 | ||||
2086 | // Accesses to objects accessible after the function returns can only be | |||
2087 | // eliminated if the access is killed along all paths to the exit. Collect | |||
2088 | // the blocks with killing (=completely overwriting MemoryDefs) and check if | |||
2089 | // they cover all paths from EarlierAccess to any function exit. | |||
2090 | SmallPtrSet<Instruction *, 16> KillingDefs; | |||
2091 | KillingDefs.insert(KillingDef->getMemoryInst()); | |||
2092 | MemoryAccess *EarlierAccess = Current; | |||
2093 | Instruction *EarlierMemInst = | |||
2094 | cast<MemoryDef>(EarlierAccess)->getMemoryInst(); | |||
2095 | 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) | |||
2096 | << *EarlierMemInst << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " Checking for reads of " << *EarlierAccess << " (" << *EarlierMemInst << ")\n"; } } while (false); | |||
2097 | ||||
2098 | SmallSetVector<MemoryAccess *, 32> WorkList; | |||
2099 | auto PushMemUses = [&WorkList](MemoryAccess *Acc) { | |||
2100 | for (Use &U : Acc->uses()) | |||
2101 | WorkList.insert(cast<MemoryAccess>(U.getUser())); | |||
2102 | }; | |||
2103 | PushMemUses(EarlierAccess); | |||
2104 | ||||
2105 | // Optimistically collect all accesses for reads. If we do not find any | |||
2106 | // read clobbers, add them to the cache. | |||
2107 | SmallPtrSet<MemoryAccess *, 16> KnownNoReads; | |||
2108 | if (!EarlierMemInst->mayReadFromMemory()) | |||
2109 | KnownNoReads.insert(EarlierAccess); | |||
2110 | // Check if EarlierDef may be read. | |||
2111 | for (unsigned I = 0; I < WorkList.size(); I++) { | |||
2112 | MemoryAccess *UseAccess = WorkList[I]; | |||
2113 | ||||
2114 | LLVM_DEBUG(dbgs() << " " << *UseAccess)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " " << *UseAccess; } } while (false); | |||
2115 | // Bail out if the number of accesses to check exceeds the scan limit. | |||
2116 | if (ScanLimit < (WorkList.size() - I)) { | |||
2117 | LLVM_DEBUG(dbgs() << "\n ... hit scan limit\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "\n ... hit scan limit\n"; } } while (false); | |||
2118 | return None; | |||
2119 | } | |||
2120 | --ScanLimit; | |||
2121 | NumDomMemDefChecks++; | |||
2122 | ||||
2123 | // Check if we already visited this access. | |||
2124 | if (Cache.isKnownNoRead(UseAccess)) { | |||
2125 | LLVM_DEBUG(dbgs() << " ... skip, discovered that " << *UseAccessdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, discovered that " << *UseAccess << " is safe earlier.\n"; } } while (false) | |||
2126 | << " is safe earlier.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skip, discovered that " << *UseAccess << " is safe earlier.\n"; } } while (false); | |||
2127 | continue; | |||
2128 | } | |||
2129 | if (Cache.isKnownRead(UseAccess)) { | |||
2130 | 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) | |||
2131 | << " 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); | |||
2132 | return None; | |||
2133 | } | |||
2134 | KnownNoReads.insert(UseAccess); | |||
2135 | ||||
2136 | if (isa<MemoryPhi>(UseAccess)) { | |||
2137 | if (any_of(KillingDefs, [this, UseAccess](Instruction *KI) { | |||
2138 | return DT.properlyDominates(KI->getParent(), | |||
2139 | UseAccess->getBlock()); | |||
2140 | })) { | |||
2141 | 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); | |||
2142 | continue; | |||
2143 | } | |||
2144 | LLVM_DEBUG(dbgs() << "\n ... adding PHI uses\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "\n ... adding PHI uses\n"; } } while (false); | |||
2145 | PushMemUses(UseAccess); | |||
2146 | continue; | |||
2147 | } | |||
2148 | ||||
2149 | Instruction *UseInst = cast<MemoryUseOrDef>(UseAccess)->getMemoryInst(); | |||
2150 | LLVM_DEBUG(dbgs() << " (" << *UseInst << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " (" << *UseInst << ")\n" ; } } while (false); | |||
2151 | ||||
2152 | if (any_of(KillingDefs, [this, UseInst](Instruction *KI) { | |||
2153 | return DT.dominates(KI, UseInst); | |||
2154 | })) { | |||
2155 | 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); | |||
2156 | continue; | |||
2157 | } | |||
2158 | ||||
2159 | // A memory terminator kills all preceeding MemoryDefs and all succeeding | |||
2160 | // MemoryAccesses. We do not have to check it's users. | |||
2161 | if (isMemTerminator(DefLoc, KillingI, UseInst)) { | |||
2162 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, memterminator invalidates following accesses\n" ; } } while (false) | |||
2163 | dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, memterminator invalidates following accesses\n" ; } } while (false) | |||
2164 | << " ... skipping, memterminator invalidates following accesses\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... skipping, memterminator invalidates following accesses\n" ; } } while (false); | |||
2165 | continue; | |||
2166 | } | |||
2167 | ||||
2168 | if (isNoopIntrinsic(cast<MemoryUseOrDef>(UseAccess)->getMemoryInst())) { | |||
2169 | LLVM_DEBUG(dbgs() << " ... adding uses of intrinsic\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... adding uses of intrinsic\n" ; } } while (false); | |||
2170 | PushMemUses(UseAccess); | |||
2171 | continue; | |||
2172 | } | |||
2173 | ||||
2174 | if (UseInst->mayThrow() && !isInvisibleToCallerBeforeRet(DefUO)) { | |||
2175 | LLVM_DEBUG(dbgs() << " ... found throwing instruction\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found throwing instruction\n" ; } } while (false); | |||
2176 | Cache.KnownReads.insert(UseAccess); | |||
2177 | Cache.KnownReads.insert(StartAccess); | |||
2178 | Cache.KnownReads.insert(EarlierAccess); | |||
2179 | return None; | |||
2180 | } | |||
2181 | ||||
2182 | // Uses which may read the original MemoryDef mean we cannot eliminate the | |||
2183 | // original MD. Stop walk. | |||
2184 | if (isReadClobber(DefLoc, UseInst)) { | |||
2185 | LLVM_DEBUG(dbgs() << " ... found read clobber\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found read clobber\n"; } } while (false); | |||
2186 | Cache.KnownReads.insert(UseAccess); | |||
2187 | Cache.KnownReads.insert(StartAccess); | |||
2188 | Cache.KnownReads.insert(EarlierAccess); | |||
2189 | return None; | |||
2190 | } | |||
2191 | ||||
2192 | // For the KillingDef and EarlierAccess we only have to check if it reads | |||
2193 | // the memory location. | |||
2194 | // TODO: It would probably be better to check for self-reads before | |||
2195 | // calling the function. | |||
2196 | if (KillingDef == UseAccess || EarlierAccess == UseAccess) { | |||
2197 | 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); | |||
2198 | continue; | |||
2199 | } | |||
2200 | ||||
2201 | // Check all uses for MemoryDefs, except for defs completely overwriting | |||
2202 | // the original location. Otherwise we have to check uses of *all* | |||
2203 | // MemoryDefs we discover, including non-aliasing ones. Otherwise we might | |||
2204 | // miss cases like the following | |||
2205 | // 1 = Def(LoE) ; <----- EarlierDef stores [0,1] | |||
2206 | // 2 = Def(1) ; (2, 1) = NoAlias, stores [2,3] | |||
2207 | // Use(2) ; MayAlias 2 *and* 1, loads [0, 3]. | |||
2208 | // (The Use points to the *first* Def it may alias) | |||
2209 | // 3 = Def(1) ; <---- Current (3, 2) = NoAlias, (3,1) = MayAlias, | |||
2210 | // stores [0,1] | |||
2211 | if (MemoryDef *UseDef = dyn_cast<MemoryDef>(UseAccess)) { | |||
2212 | if (isCompleteOverwrite(DefLoc, KillingI, UseInst)) { | |||
2213 | if (!isInvisibleToCallerAfterRet(DefUO) && | |||
2214 | UseAccess != EarlierAccess) { | |||
2215 | BasicBlock *MaybeKillingBlock = UseInst->getParent(); | |||
2216 | if (PostOrderNumbers.find(MaybeKillingBlock)->second < | |||
2217 | PostOrderNumbers.find(EarlierAccess->getBlock())->second) { | |||
2218 | ||||
2219 | LLVM_DEBUG(dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found killing def " << *UseInst << "\n"; } } while (false) | |||
2220 | << " ... found killing def " << *UseInst << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " ... found killing def " << *UseInst << "\n"; } } while (false); | |||
2221 | KillingDefs.insert(UseInst); | |||
2222 | } | |||
2223 | } | |||
2224 | } else | |||
2225 | PushMemUses(UseDef); | |||
2226 | } | |||
2227 | } | |||
2228 | ||||
2229 | // For accesses to locations visible after the function returns, make sure | |||
2230 | // that the location is killed (=overwritten) along all paths from | |||
2231 | // EarlierAccess to the exit. | |||
2232 | if (!isInvisibleToCallerAfterRet(DefUO)) { | |||
2233 | SmallPtrSet<BasicBlock *, 16> KillingBlocks; | |||
2234 | for (Instruction *KD : KillingDefs) | |||
2235 | KillingBlocks.insert(KD->getParent()); | |||
2236 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2237, __PRETTY_FUNCTION__)) | |||
2237 | "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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2237, __PRETTY_FUNCTION__)); | |||
2238 | ||||
2239 | // Find the common post-dominator of all killing blocks. | |||
2240 | BasicBlock *CommonPred = *KillingBlocks.begin(); | |||
2241 | for (auto I = std::next(KillingBlocks.begin()), E = KillingBlocks.end(); | |||
2242 | I != E; I++) { | |||
2243 | if (!CommonPred) | |||
2244 | break; | |||
2245 | CommonPred = PDT.findNearestCommonDominator(CommonPred, *I); | |||
2246 | } | |||
2247 | ||||
2248 | // If CommonPred is in the set of killing blocks, just check if it | |||
2249 | // post-dominates EarlierAccess. | |||
2250 | if (KillingBlocks.count(CommonPred)) { | |||
2251 | if (PDT.dominates(CommonPred, EarlierAccess->getBlock())) | |||
2252 | return {EarlierAccess}; | |||
2253 | return None; | |||
2254 | } | |||
2255 | ||||
2256 | // If the common post-dominator does not post-dominate EarlierAccess, | |||
2257 | // there is a path from EarlierAccess to an exit not going through a | |||
2258 | // killing block. | |||
2259 | if (PDT.dominates(CommonPred, EarlierAccess->getBlock())) { | |||
2260 | SetVector<BasicBlock *> WorkList; | |||
2261 | ||||
2262 | // If CommonPred is null, there are multiple exits from the function. | |||
2263 | // They all have to be added to the worklist. | |||
2264 | if (CommonPred) | |||
2265 | WorkList.insert(CommonPred); | |||
2266 | else | |||
2267 | for (BasicBlock *R : PDT.roots()) | |||
2268 | WorkList.insert(R); | |||
2269 | ||||
2270 | NumCFGTries++; | |||
2271 | // Check if all paths starting from an exit node go through one of the | |||
2272 | // killing blocks before reaching EarlierAccess. | |||
2273 | for (unsigned I = 0; I < WorkList.size(); I++) { | |||
2274 | NumCFGChecks++; | |||
2275 | BasicBlock *Current = WorkList[I]; | |||
2276 | if (KillingBlocks.count(Current)) | |||
2277 | continue; | |||
2278 | if (Current == EarlierAccess->getBlock()) | |||
2279 | return None; | |||
2280 | ||||
2281 | // EarlierAccess is reachable from the entry, so we don't have to | |||
2282 | // explore unreachable blocks further. | |||
2283 | if (!DT.isReachableFromEntry(Current)) | |||
2284 | continue; | |||
2285 | ||||
2286 | for (BasicBlock *Pred : predecessors(Current)) | |||
2287 | WorkList.insert(Pred); | |||
2288 | ||||
2289 | if (WorkList.size() >= MemorySSAPathCheckLimit) | |||
2290 | return None; | |||
2291 | } | |||
2292 | NumCFGSuccess++; | |||
2293 | return {EarlierAccess}; | |||
2294 | } | |||
2295 | return None; | |||
2296 | } | |||
2297 | ||||
2298 | // No aliasing MemoryUses of EarlierAccess found, EarlierAccess is | |||
2299 | // potentially dead. | |||
2300 | Cache.KnownNoReads.insert(KnownNoReads.begin(), KnownNoReads.end()); | |||
2301 | return {EarlierAccess}; | |||
2302 | } | |||
2303 | ||||
2304 | // Delete dead memory defs | |||
2305 | void deleteDeadInstruction(Instruction *SI) { | |||
2306 | MemorySSAUpdater Updater(&MSSA); | |||
2307 | SmallVector<Instruction *, 32> NowDeadInsts; | |||
2308 | NowDeadInsts.push_back(SI); | |||
2309 | --NumFastOther; | |||
2310 | ||||
2311 | while (!NowDeadInsts.empty()) { | |||
2312 | Instruction *DeadInst = NowDeadInsts.pop_back_val(); | |||
2313 | ++NumFastOther; | |||
2314 | ||||
2315 | // Try to preserve debug information attached to the dead instruction. | |||
2316 | salvageDebugInfo(*DeadInst); | |||
2317 | salvageKnowledge(DeadInst); | |||
2318 | ||||
2319 | // Remove the Instruction from MSSA. | |||
2320 | if (MemoryAccess *MA = MSSA.getMemoryAccess(DeadInst)) { | |||
2321 | if (MemoryDef *MD = dyn_cast<MemoryDef>(MA)) { | |||
2322 | SkipStores.insert(MD); | |||
2323 | } | |||
2324 | Updater.removeMemoryAccess(MA); | |||
2325 | } | |||
2326 | ||||
2327 | auto I = IOLs.find(DeadInst->getParent()); | |||
2328 | if (I != IOLs.end()) | |||
2329 | I->second.erase(DeadInst); | |||
2330 | // Remove its operands | |||
2331 | for (Use &O : DeadInst->operands()) | |||
2332 | if (Instruction *OpI = dyn_cast<Instruction>(O)) { | |||
2333 | O = nullptr; | |||
2334 | if (isInstructionTriviallyDead(OpI, &TLI)) | |||
2335 | NowDeadInsts.push_back(OpI); | |||
2336 | } | |||
2337 | ||||
2338 | DeadInst->eraseFromParent(); | |||
2339 | } | |||
2340 | } | |||
2341 | ||||
2342 | // Check for any extra throws between SI and NI that block DSE. This only | |||
2343 | // checks extra maythrows (those that aren't MemoryDef's). MemoryDef that may | |||
2344 | // throw are handled during the walk from one def to the next. | |||
2345 | bool mayThrowBetween(Instruction *SI, Instruction *NI, | |||
2346 | const Value *SILocUnd) { | |||
2347 | // First see if we can ignore it by using the fact that SI is an | |||
2348 | // alloca/alloca like object that is not visible to the caller during | |||
2349 | // execution of the function. | |||
2350 | if (SILocUnd && isInvisibleToCallerBeforeRet(SILocUnd)) | |||
2351 | return false; | |||
2352 | ||||
2353 | if (SI->getParent() == NI->getParent()) | |||
2354 | return ThrowingBlocks.count(SI->getParent()); | |||
2355 | return !ThrowingBlocks.empty(); | |||
2356 | } | |||
2357 | ||||
2358 | // Check if \p NI acts as a DSE barrier for \p SI. The following instructions | |||
2359 | // act as barriers: | |||
2360 | // * A memory instruction that may throw and \p SI accesses a non-stack | |||
2361 | // object. | |||
2362 | // * Atomic stores stronger that monotonic. | |||
2363 | bool isDSEBarrier(const Value *SILocUnd, Instruction *NI) { | |||
2364 | // If NI may throw it acts as a barrier, unless we are to an alloca/alloca | |||
2365 | // like object that does not escape. | |||
2366 | if (NI->mayThrow() && !isInvisibleToCallerBeforeRet(SILocUnd)) | |||
2367 | return true; | |||
2368 | ||||
2369 | // If NI is an atomic load/store stronger than monotonic, do not try to | |||
2370 | // eliminate/reorder it. | |||
2371 | if (NI->isAtomic()) { | |||
2372 | if (auto *LI = dyn_cast<LoadInst>(NI)) | |||
2373 | return isStrongerThanMonotonic(LI->getOrdering()); | |||
2374 | if (auto *SI = dyn_cast<StoreInst>(NI)) | |||
2375 | return isStrongerThanMonotonic(SI->getOrdering()); | |||
2376 | if (auto *ARMW = dyn_cast<AtomicRMWInst>(NI)) | |||
2377 | return isStrongerThanMonotonic(ARMW->getOrdering()); | |||
2378 | if (auto *CmpXchg = dyn_cast<AtomicCmpXchgInst>(NI)) | |||
2379 | return isStrongerThanMonotonic(CmpXchg->getSuccessOrdering()) || | |||
2380 | isStrongerThanMonotonic(CmpXchg->getFailureOrdering()); | |||
2381 | llvm_unreachable("other instructions should be skipped in MemorySSA")::llvm::llvm_unreachable_internal("other instructions should be skipped in MemorySSA" , "/build/llvm-toolchain-snapshot-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2381); | |||
2382 | } | |||
2383 | return false; | |||
2384 | } | |||
2385 | ||||
2386 | /// Eliminate writes to objects that are not visible in the caller and are not | |||
2387 | /// accessed before returning from the function. | |||
2388 | bool eliminateDeadWritesAtEndOfFunction() { | |||
2389 | bool MadeChange = false; | |||
2390 | LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs at the end of the function\n" ; } } while (false) | |||
2391 | dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs at the end of the function\n" ; } } while (false) | |||
2392 | << "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); | |||
2393 | for (int I = MemDefs.size() - 1; I >= 0; I--) { | |||
2394 | MemoryDef *Def = MemDefs[I]; | |||
2395 | if (SkipStores.find(Def) != SkipStores.end() || | |||
2396 | !isRemovable(Def->getMemoryInst())) | |||
2397 | continue; | |||
2398 | ||||
2399 | Instruction *DefI = Def->getMemoryInst(); | |||
2400 | SmallVector<const Value *, 4> Pointers; | |||
2401 | auto DefLoc = getLocForWriteEx(DefI); | |||
2402 | if (!DefLoc) | |||
2403 | continue; | |||
2404 | ||||
2405 | // NOTE: Currently eliminating writes at the end of a function is limited | |||
2406 | // to MemoryDefs with a single underlying object, to save compile-time. In | |||
2407 | // practice it appears the case with multiple underlying objects is very | |||
2408 | // uncommon. If it turns out to be important, we can use | |||
2409 | // getUnderlyingObjects here instead. | |||
2410 | const Value *UO = getUnderlyingObject(DefLoc->Ptr); | |||
2411 | if (!UO || !isInvisibleToCallerAfterRet(UO)) | |||
2412 | continue; | |||
2413 | ||||
2414 | if (isWriteAtEndOfFunction(Def)) { | |||
2415 | // See through pointer-to-pointer bitcasts | |||
2416 | 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) | |||
2417 | "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); | |||
2418 | deleteDeadInstruction(DefI); | |||
2419 | ++NumFastStores; | |||
2420 | MadeChange = true; | |||
2421 | } | |||
2422 | } | |||
2423 | return MadeChange; | |||
2424 | } | |||
2425 | ||||
2426 | /// \returns true if \p Def is a no-op store, either because it | |||
2427 | /// directly stores back a loaded value or stores zero to a calloced object. | |||
2428 | bool storeIsNoop(MemoryDef *Def, MemoryLocation DefLoc, const Value *DefUO) { | |||
2429 | StoreInst *Store = dyn_cast<StoreInst>(Def->getMemoryInst()); | |||
2430 | if (!Store) | |||
2431 | return false; | |||
2432 | ||||
2433 | if (auto *LoadI = dyn_cast<LoadInst>(Store->getOperand(0))) { | |||
2434 | if (LoadI->getPointerOperand() == Store->getOperand(1)) { | |||
2435 | // Get the defining access for the load. | |||
2436 | auto *LoadAccess = MSSA.getMemoryAccess(LoadI)->getDefiningAccess(); | |||
2437 | // Fast path: the defining accesses are the same. | |||
2438 | if (LoadAccess == Def->getDefiningAccess()) | |||
2439 | return true; | |||
2440 | ||||
2441 | // Look through phi accesses. Recursively scan all phi accesses by | |||
2442 | // adding them to a worklist. Bail when we run into a memory def that | |||
2443 | // does not match LoadAccess. | |||
2444 | SetVector<MemoryAccess *> ToCheck; | |||
2445 | MemoryAccess *Current = | |||
2446 | MSSA.getWalker()->getClobberingMemoryAccess(Def); | |||
2447 | // We don't want to bail when we run into the store memory def. But, | |||
2448 | // the phi access may point to it. So, pretend like we've already | |||
2449 | // checked it. | |||
2450 | ToCheck.insert(Def); | |||
2451 | ToCheck.insert(Current); | |||
2452 | // Start at current (1) to simulate already having checked Def. | |||
2453 | for (unsigned I = 1; I < ToCheck.size(); ++I) { | |||
2454 | Current = ToCheck[I]; | |||
2455 | if (auto PhiAccess = dyn_cast<MemoryPhi>(Current)) { | |||
2456 | // Check all the operands. | |||
2457 | for (auto &Use : PhiAccess->incoming_values()) | |||
2458 | ToCheck.insert(cast<MemoryAccess>(&Use)); | |||
2459 | continue; | |||
2460 | } | |||
2461 | ||||
2462 | // If we found a memory def, bail. This happens when we have an | |||
2463 | // unrelated write in between an otherwise noop store. | |||
2464 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2465, __PRETTY_FUNCTION__)) | |||
2465 | "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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2465, __PRETTY_FUNCTION__)); | |||
2466 | // TODO: Skip no alias MemoryDefs that have no aliasing reads. | |||
2467 | // We are searching for the definition of the store's destination. | |||
2468 | // So, if that is the same definition as the load, then this is a | |||
2469 | // noop. Otherwise, fail. | |||
2470 | if (LoadAccess != Current) | |||
2471 | return false; | |||
2472 | } | |||
2473 | return true; | |||
2474 | } | |||
2475 | } | |||
2476 | ||||
2477 | Constant *StoredConstant = dyn_cast<Constant>(Store->getOperand(0)); | |||
2478 | if (StoredConstant && StoredConstant->isNullValue()) { | |||
2479 | auto *DefUOInst = dyn_cast<Instruction>(DefUO); | |||
2480 | if (DefUOInst && isCallocLikeFn(DefUOInst, &TLI)) { | |||
2481 | auto *UnderlyingDef = cast<MemoryDef>(MSSA.getMemoryAccess(DefUOInst)); | |||
2482 | // If UnderlyingDef is the clobbering access of Def, no instructions | |||
2483 | // between them can modify the memory location. | |||
2484 | auto *ClobberDef = | |||
2485 | MSSA.getSkipSelfWalker()->getClobberingMemoryAccess(Def); | |||
2486 | return UnderlyingDef == ClobberDef; | |||
2487 | } | |||
2488 | } | |||
2489 | return false; | |||
2490 | } | |||
2491 | }; | |||
2492 | ||||
2493 | bool eliminateDeadStoresMemorySSA(Function &F, AliasAnalysis &AA, | |||
2494 | MemorySSA &MSSA, DominatorTree &DT, | |||
2495 | PostDominatorTree &PDT, | |||
2496 | const TargetLibraryInfo &TLI) { | |||
2497 | bool MadeChange = false; | |||
2498 | ||||
2499 | DSEState State = DSEState::get(F, AA, MSSA, DT, PDT, TLI); | |||
2500 | // For each store: | |||
2501 | for (unsigned I = 0; I < State.MemDefs.size(); I++) { | |||
| ||||
2502 | MemoryDef *KillingDef = State.MemDefs[I]; | |||
2503 | if (State.SkipStores.count(KillingDef)) | |||
2504 | continue; | |||
2505 | Instruction *SI = KillingDef->getMemoryInst(); | |||
2506 | ||||
2507 | auto MaybeSILoc = State.getLocForWriteEx(SI); | |||
2508 | if (State.isMemTerminatorInst(SI)) | |||
2509 | MaybeSILoc = State.getLocForTerminator(SI).map( | |||
2510 | [](const std::pair<MemoryLocation, bool> &P) { return P.first; }); | |||
2511 | else | |||
2512 | MaybeSILoc = State.getLocForWriteEx(SI); | |||
2513 | ||||
2514 | if (!MaybeSILoc) { | |||
2515 | 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) | |||
2516 | << *SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Failed to find analyzable write location for " << *SI << "\n"; } } while (false); | |||
2517 | continue; | |||
2518 | } | |||
2519 | MemoryLocation SILoc = *MaybeSILoc; | |||
2520 | 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-12.0.0~++20201102111116+1ed2ca68191/llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp" , 2520, __PRETTY_FUNCTION__)); | |||
2521 | const Value *SILocUnd = getUnderlyingObject(SILoc.Ptr); | |||
2522 | ||||
2523 | MemoryAccess *Current = KillingDef; | |||
2524 | 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) | |||
2525 | << *KillingDef << " (" << *SI << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << "Trying to eliminate MemoryDefs killed by " << *KillingDef << " (" << *SI << ")\n" ; } } while (false); | |||
2526 | ||||
2527 | unsigned ScanLimit = MemorySSAScanLimit; | |||
2528 | unsigned WalkerStepLimit = MemorySSAUpwardsStepLimit; | |||
2529 | unsigned PartialLimit = MemorySSAPartialStoreLimit; | |||
2530 | // Worklist of MemoryAccesses that may be killed by KillingDef. | |||
2531 | SetVector<MemoryAccess *> ToCheck; | |||
2532 | ||||
2533 | if (SILocUnd) | |||
2534 | ToCheck.insert(KillingDef->getDefiningAccess()); | |||
2535 | ||||
2536 | bool Shortend = false; | |||
2537 | bool IsMemTerm = State.isMemTerminatorInst(SI); | |||
2538 | DSEState::CheckCache Cache; | |||
2539 | // Check if MemoryAccesses in the worklist are killed by KillingDef. | |||
2540 | for (unsigned I = 0; I < ToCheck.size(); I++) { | |||
2541 | Current = ToCheck[I]; | |||
2542 | if (State.SkipStores.count(Current)) | |||
2543 | continue; | |||
2544 | ||||
2545 | Optional<MemoryAccess *> Next = State.getDomMemoryDef( | |||
2546 | KillingDef, Current, SILoc, SILocUnd, Cache, ScanLimit, | |||
2547 | WalkerStepLimit, IsMemTerm, PartialLimit); | |||
2548 | ||||
2549 | if (!Next) { | |||
2550 | LLVM_DEBUG(dbgs() << " finished walk\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " finished walk\n"; } } while (false ); | |||
2551 | continue; | |||
2552 | } | |||
2553 | ||||
2554 | MemoryAccess *EarlierAccess = *Next; | |||
2555 | 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); | |||
2556 | if (isa<MemoryPhi>(EarlierAccess)) { | |||
2557 | 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); | |||
2558 | for (Value *V : cast<MemoryPhi>(EarlierAccess)->incoming_values()) { | |||
2559 | MemoryAccess *IncomingAccess = cast<MemoryAccess>(V); | |||
2560 | BasicBlock *IncomingBlock = IncomingAccess->getBlock(); | |||
2561 | BasicBlock *PhiBlock = EarlierAccess->getBlock(); | |||
2562 | ||||
2563 | // We only consider incoming MemoryAccesses that come before the | |||
2564 | // MemoryPhi. Otherwise we could discover candidates that do not | |||
2565 | // strictly dominate our starting def. | |||
2566 | if (State.PostOrderNumbers[IncomingBlock] > | |||
2567 | State.PostOrderNumbers[PhiBlock]) | |||
2568 | ToCheck.insert(IncomingAccess); | |||
2569 | } | |||
2570 | continue; | |||
2571 | } | |||
2572 | MemoryDef *NextDef = dyn_cast<MemoryDef>(EarlierAccess); | |||
2573 | Instruction *NI = NextDef->getMemoryInst(); | |||
| ||||
2574 | LLVM_DEBUG(dbgs() << " (" << *NI << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType ("dse")) { dbgs() << " (" << *NI << ")\n"; } } while (false); | |||
2575 | ToCheck.insert(NextDef->getDefiningAccess()); | |||
2576 | NumGetDomMemoryDefPassed++; | |||
2577 | ||||
2578 | if (!DebugCounter::shouldExecute(MemorySSACounter)) | |||
2579 | continue; | |||
2580 | ||||
2581 | MemoryLocation NILoc = *State.getLocForWriteEx(NI); | |||
2582 | ||||
2583 | if (IsMemTerm) { | |||
2584 | const Value *NIUnd = getUnderlyingObject(NILoc.Ptr); | |||
2585 | if (SILocUnd != NIUnd) | |||
2586 | continue; | |||
2587 | 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) | |||
2588 | << "\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); | |||
2589 | State.deleteDeadInstruction(NI); | |||
2590 | ++NumFastStores; | |||
2591 | MadeChange = true; | |||
2592 | } else { | |||
2593 | // Check if NI overwrites SI. | |||
2594 | int64_t InstWriteOffset, DepWriteOffset; | |||
2595 | OverwriteResult OR = | |||
2596 | isOverwrite(SI, NI, SILoc, NILoc, State.DL, TLI, DepWriteOffset, | |||
2597 | InstWriteOffset, State.BatchAA, &F); | |||
2598 | if (OR == OW_MaybePartial) { | |||
2599 | auto Iter = State.IOLs.insert( | |||
2600 | std::make_pair<BasicBlock *, InstOverlapIntervalsTy>( | |||
2601 | NI->getParent(), InstOverlapIntervalsTy())); | |||
2602 | auto &IOL = Iter.first->second; | |||
2603 | OR = isPartialOverwrite(SILoc, NILoc, DepWriteOffset, InstWriteOffset, | |||
2604 | NI, IOL); | |||
2605 | } | |||
2606 | ||||
2607 | if (EnablePartialStoreMerging && OR == OW_PartialEarlierWithFullLater) { | |||
2608 | auto *Earlier = dyn_cast<StoreInst>(NI); | |||
2609 | auto *Later = dyn_cast<StoreInst>(SI); | |||
2610 | // We are re-using tryToMergePartialOverlappingStores, which requires | |||
2611 | // Earlier to domiante Later. | |||
2612 | // TODO: implement tryToMergeParialOverlappingStores using MemorySSA. | |||
2613 | if (Earlier && Later && DT.dominates(Earlier, Later)) { | |||
2614 | if (Constant *Merged = tryToMergePartialOverlappingStores( | |||
2615 | Earlier, Later, InstWriteOffset, DepWriteOffset, State.DL, | |||
2616 | State.BatchAA, &DT)) { | |||
2617 | ||||
2618 | // Update stored value of earlier store to merged constant. | |||
2619 | Earlier->setOperand(0, Merged); | |||
2620 | ++NumModifiedStores; | |||
2621 | MadeChange = true; | |||
2622 | ||||
2623 | Shortend = true; | |||
2624 | // Remove later store and remove any outstanding overlap intervals | |||
2625 | // for the updated store. | |||
2626 | State.deleteDeadInstruction(Later); | |||
2627 | auto I = State.IOLs.find(Earlier->getParent()); | |||
2628 | if (I != State.IOLs.end()) | |||
2629 | I->second.erase(Earlier); | |||
2630 | break; | |||
2631 | } | |||
2632 | } | |||
2633 | } | |||
2634 | ||||
2635 | if (OR == OW_Complete) { | |||
2636 | 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) | |||
2637 | << "\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); | |||
2638 | State.deleteDeadInstruction(NI); | |||
2639 | ++NumFastStores; | |||
2640 | MadeChange = true; | |||
2641 | } | |||
2642 | } | |||
2643 | } | |||
2644 | ||||
2645 | // Check if the store is a no-op. | |||
2646 | if (!Shortend && isRemovable(SI) && | |||
2647 | State.storeIsNoop(KillingDef, SILoc, SILocUnd)) { | |||
2648 | 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); | |||
2649 | State.deleteDeadInstruction(SI); | |||
2650 | NumRedundantStores++; | |||
2651 | MadeChange = true; | |||
2652 | continue; | |||
2653 | } | |||
2654 | } | |||
2655 | ||||
2656 | if (EnablePartialOverwriteTracking) | |||
2657 | for (auto &KV : State.IOLs) | |||
2658 | MadeChange |= removePartiallyOverlappedStores(State.DL, KV.second, TLI); | |||
2659 | ||||
2660 | MadeChange |= State.eliminateDeadWritesAtEndOfFunction(); | |||
2661 | return MadeChange; | |||
2662 | } | |||
2663 | } // end anonymous namespace | |||
2664 | ||||
2665 | //===----------------------------------------------------------------------===// | |||
2666 | // DSE Pass | |||
2667 | //===----------------------------------------------------------------------===// | |||
2668 | PreservedAnalyses DSEPass::run(Function &F, FunctionAnalysisManager &AM) { | |||
2669 | AliasAnalysis &AA = AM.getResult<AAManager>(F); | |||
2670 | const TargetLibraryInfo &TLI = AM.getResult<TargetLibraryAnalysis>(F); | |||
2671 | DominatorTree &DT = AM.getResult<DominatorTreeAnalysis>(F); | |||
2672 | ||||
2673 | bool Changed = false; | |||
2674 | if (EnableMemorySSA) { | |||
2675 | MemorySSA &MSSA = AM.getResult<MemorySSAAnalysis>(F).getMSSA(); | |||
2676 | PostDominatorTree &PDT = AM.getResult<PostDominatorTreeAnalysis>(F); | |||
2677 | ||||
2678 | Changed = eliminateDeadStoresMemorySSA(F, AA, MSSA, DT, PDT, TLI); | |||
2679 | } else { | |||
2680 | MemoryDependenceResults &MD = AM.getResult<MemoryDependenceAnalysis>(F); | |||
2681 | ||||
2682 | Changed = eliminateDeadStores(F, &AA, &MD, &DT, &TLI); | |||
2683 | } | |||
2684 | ||||
2685 | #ifdef LLVM_ENABLE_STATS1 | |||
2686 | if (AreStatisticsEnabled()) | |||
2687 | for (auto &I : instructions(F)) | |||
2688 | NumRemainingStores += isa<StoreInst>(&I); | |||
2689 | #endif | |||
2690 | ||||
2691 | if (!Changed) | |||
2692 | return PreservedAnalyses::all(); | |||
2693 | ||||
2694 | PreservedAnalyses PA; | |||
2695 | PA.preserveSet<CFGAnalyses>(); | |||
2696 | PA.preserve<GlobalsAA>(); | |||
2697 | if (EnableMemorySSA) | |||
2698 | PA.preserve<MemorySSAAnalysis>(); | |||
2699 | else | |||
2700 | PA.preserve<MemoryDependenceAnalysis>(); | |||
2701 | return PA; | |||
2702 | } | |||
2703 | ||||
2704 | namespace { | |||
2705 | ||||
2706 | /// A legacy pass for the legacy pass manager that wraps \c DSEPass. | |||
2707 | class DSELegacyPass : public FunctionPass { | |||
2708 | public: | |||
2709 | static char ID; // Pass identification, replacement for typeid | |||
2710 | ||||
2711 | DSELegacyPass() : FunctionPass(ID) { | |||
2712 | initializeDSELegacyPassPass(*PassRegistry::getPassRegistry()); | |||
2713 | } | |||
2714 | ||||
2715 | bool runOnFunction(Function &F) override { | |||
2716 | if (skipFunction(F)) | |||
2717 | return false; | |||
2718 | ||||
2719 | AliasAnalysis &AA = getAnalysis<AAResultsWrapperPass>().getAAResults(); | |||
2720 | DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree(); | |||
2721 | const TargetLibraryInfo &TLI = | |||
2722 | getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F); | |||
2723 | ||||
2724 | bool Changed = false; | |||
2725 | if (EnableMemorySSA) { | |||
2726 | MemorySSA &MSSA = getAnalysis<MemorySSAWrapperPass>().getMSSA(); | |||
2727 | PostDominatorTree &PDT = | |||
2728 | getAnalysis<PostDominatorTreeWrapperPass>().getPostDomTree(); | |||
2729 | ||||
2730 | Changed = eliminateDeadStoresMemorySSA(F, AA, MSSA, DT, PDT, TLI); | |||
2731 | } else { | |||
2732 | MemoryDependenceResults &MD = | |||
2733 | getAnalysis<MemoryDependenceWrapperPass>().getMemDep(); | |||
2734 | ||||
2735 | Changed = eliminateDeadStores(F, &AA, &MD, &DT, &TLI); | |||
2736 | } | |||
2737 | ||||
2738 | #ifdef LLVM_ENABLE_STATS1 | |||
2739 | if (AreStatisticsEnabled()) | |||
2740 | for (auto &I : instructions(F)) | |||
2741 | NumRemainingStores += isa<StoreInst>(&I); | |||
2742 | #endif | |||
2743 | ||||
2744 | return Changed; | |||
2745 | } | |||
2746 | ||||
2747 | void getAnalysisUsage(AnalysisUsage &AU) const override { | |||
2748 | AU.setPreservesCFG(); | |||
2749 | AU.addRequired<AAResultsWrapperPass>(); | |||
2750 | AU.addRequired<TargetLibraryInfoWrapperPass>(); | |||
2751 | AU.addPreserved<GlobalsAAWrapperPass>(); | |||
2752 | AU.addRequired<DominatorTreeWrapperPass>(); | |||
2753 | AU.addPreserved<DominatorTreeWrapperPass>(); | |||
2754 | ||||
2755 | if (EnableMemorySSA) { | |||
2756 | AU.addRequired<PostDominatorTreeWrapperPass>(); | |||
2757 | AU.addRequired<MemorySSAWrapperPass>(); | |||
2758 | AU.addPreserved<PostDominatorTreeWrapperPass>(); | |||
2759 | AU.addPreserved<MemorySSAWrapperPass>(); | |||
2760 | } else { | |||
2761 | AU.addRequired<MemoryDependenceWrapperPass>(); | |||
2762 | AU.addPreserved<MemoryDependenceWrapperPass>(); | |||
2763 | } | |||
2764 | } | |||
2765 | }; | |||
2766 | ||||
2767 | } // end anonymous namespace | |||
2768 | ||||
2769 | char DSELegacyPass::ID = 0; | |||
2770 | ||||
2771 | INITIALIZE_PASS_BEGIN(DSELegacyPass, "dse", "Dead Store Elimination", false,static void *initializeDSELegacyPassPassOnce(PassRegistry & Registry) { | |||
2772 | false)static void *initializeDSELegacyPassPassOnce(PassRegistry & Registry) { | |||
2773 | INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)initializeDominatorTreeWrapperPassPass(Registry); | |||
2774 | INITIALIZE_PASS_DEPENDENCY(PostDominatorTreeWrapperPass)initializePostDominatorTreeWrapperPassPass(Registry); | |||
2775 | INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)initializeAAResultsWrapperPassPass(Registry); | |||
2776 | INITIALIZE_PASS_DEPENDENCY(GlobalsAAWrapperPass)initializeGlobalsAAWrapperPassPass(Registry); | |||
2777 | INITIALIZE_PASS_DEPENDENCY(MemorySSAWrapperPass)initializeMemorySSAWrapperPassPass(Registry); | |||
2778 | INITIALIZE_PASS_DEPENDENCY(MemoryDependenceWrapperPass)initializeMemoryDependenceWrapperPassPass(Registry); | |||
2779 | INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)initializeTargetLibraryInfoWrapperPassPass(Registry); | |||
2780 | 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)); } | |||
2781 | 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)); } | |||
2782 | ||||
2783 | FunctionPass *llvm::createDeadStoreEliminationPass() { | |||
2784 | return new DSELegacyPass(); | |||
2785 | } |