Bug Summary

File:llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
Warning:line 1504, column 48
Called C++ object pointer is null

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name LoopIdiomRecognize.cpp -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -analyzer-config-compatibility-mode=true -mrelocation-model pic -pic-level 2 -mthread-model posix -mframe-pointer=none -fmath-errno -fno-rounding-math -masm-verbose -mconstructor-aliases -munwind-tables -fuse-init-array -target-cpu x86-64 -dwarf-column-info -debugger-tuning=gdb -ffunction-sections -fdata-sections -resource-dir /usr/lib/llvm-10/lib/clang/10.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/build-llvm/lib/Transforms/Scalar -I /build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar -I /build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/build-llvm/include -I /build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/include -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/x86_64-linux-gnu/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/x86_64-linux-gnu/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/c++/6.3.0/backward -internal-isystem /usr/local/include -internal-isystem /usr/lib/llvm-10/lib/clang/10.0.0/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-comment -std=c++14 -fdeprecated-macro -fdebug-compilation-dir /build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/build-llvm/lib/Transforms/Scalar -fdebug-prefix-map=/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809=. -ferror-limit 19 -fmessage-length 0 -fvisibility-inlines-hidden -stack-protector 2 -fgnuc-version=4.2.1 -fobjc-runtime=gcc -fdiagnostics-show-option -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -faddrsig -o /tmp/scan-build-2019-12-07-102640-14763-1 -x c++ /build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp

/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp

1//===- LoopIdiomRecognize.cpp - Loop idiom recognition --------------------===//
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 pass implements an idiom recognizer that transforms simple loops into a
10// non-loop form. In cases that this kicks in, it can be a significant
11// performance win.
12//
13// If compiling for code size we avoid idiom recognition if the resulting
14// code could be larger than the code for the original loop. One way this could
15// happen is if the loop is not removable after idiom recognition due to the
16// presence of non-idiom instructions. The initial implementation of the
17// heuristics applies to idioms in multi-block loops.
18//
19//===----------------------------------------------------------------------===//
20//
21// TODO List:
22//
23// Future loop memory idioms to recognize:
24// memcmp, memmove, strlen, etc.
25// Future floating point idioms to recognize in -ffast-math mode:
26// fpowi
27// Future integer operation idioms to recognize:
28// ctpop
29//
30// Beware that isel's default lowering for ctpop is highly inefficient for
31// i64 and larger types when i64 is legal and the value has few bits set. It
32// would be good to enhance isel to emit a loop for ctpop in this case.
33//
34// This could recognize common matrix multiplies and dot product idioms and
35// replace them with calls to BLAS (if linked in??).
36//
37//===----------------------------------------------------------------------===//
38
39#include "llvm/Transforms/Scalar/LoopIdiomRecognize.h"
40#include "llvm/ADT/APInt.h"
41#include "llvm/ADT/ArrayRef.h"
42#include "llvm/ADT/DenseMap.h"
43#include "llvm/ADT/MapVector.h"
44#include "llvm/ADT/SetVector.h"
45#include "llvm/ADT/SmallPtrSet.h"
46#include "llvm/ADT/SmallVector.h"
47#include "llvm/ADT/Statistic.h"
48#include "llvm/ADT/StringRef.h"
49#include "llvm/Analysis/AliasAnalysis.h"
50#include "llvm/Analysis/LoopAccessAnalysis.h"
51#include "llvm/Analysis/LoopInfo.h"
52#include "llvm/Analysis/LoopPass.h"
53#include "llvm/Analysis/MemoryLocation.h"
54#include "llvm/Analysis/OptimizationRemarkEmitter.h"
55#include "llvm/Analysis/ScalarEvolution.h"
56#include "llvm/Analysis/ScalarEvolutionExpander.h"
57#include "llvm/Analysis/ScalarEvolutionExpressions.h"
58#include "llvm/Analysis/TargetLibraryInfo.h"
59#include "llvm/Analysis/TargetTransformInfo.h"
60#include "llvm/Analysis/ValueTracking.h"
61#include "llvm/IR/Attributes.h"
62#include "llvm/IR/BasicBlock.h"
63#include "llvm/IR/Constant.h"
64#include "llvm/IR/Constants.h"
65#include "llvm/IR/DataLayout.h"
66#include "llvm/IR/DebugLoc.h"
67#include "llvm/IR/DerivedTypes.h"
68#include "llvm/IR/Dominators.h"
69#include "llvm/IR/GlobalValue.h"
70#include "llvm/IR/GlobalVariable.h"
71#include "llvm/IR/IRBuilder.h"
72#include "llvm/IR/InstrTypes.h"
73#include "llvm/IR/Instruction.h"
74#include "llvm/IR/Instructions.h"
75#include "llvm/IR/IntrinsicInst.h"
76#include "llvm/IR/Intrinsics.h"
77#include "llvm/IR/LLVMContext.h"
78#include "llvm/IR/Module.h"
79#include "llvm/IR/PassManager.h"
80#include "llvm/IR/Type.h"
81#include "llvm/IR/User.h"
82#include "llvm/IR/Value.h"
83#include "llvm/IR/ValueHandle.h"
84#include "llvm/Pass.h"
85#include "llvm/Support/Casting.h"
86#include "llvm/Support/CommandLine.h"
87#include "llvm/Support/Debug.h"
88#include "llvm/Support/raw_ostream.h"
89#include "llvm/Transforms/Scalar.h"
90#include "llvm/Transforms/Utils/BuildLibCalls.h"
91#include "llvm/Transforms/Utils/Local.h"
92#include "llvm/Transforms/Utils/LoopUtils.h"
93#include <algorithm>
94#include <cassert>
95#include <cstdint>
96#include <utility>
97#include <vector>
98
99using namespace llvm;
100
101#define DEBUG_TYPE"loop-idiom" "loop-idiom"
102
103STATISTIC(NumMemSet, "Number of memset's formed from loop stores")static llvm::Statistic NumMemSet = {"loop-idiom", "NumMemSet"
, "Number of memset's formed from loop stores"}
;
104STATISTIC(NumMemCpy, "Number of memcpy's formed from loop load+stores")static llvm::Statistic NumMemCpy = {"loop-idiom", "NumMemCpy"
, "Number of memcpy's formed from loop load+stores"}
;
105
106static cl::opt<bool> UseLIRCodeSizeHeurs(
107 "use-lir-code-size-heurs",
108 cl::desc("Use loop idiom recognition code size heuristics when compiling"
109 "with -Os/-Oz"),
110 cl::init(true), cl::Hidden);
111
112namespace {
113
114class LoopIdiomRecognize {
115 Loop *CurLoop = nullptr;
116 AliasAnalysis *AA;
117 DominatorTree *DT;
118 LoopInfo *LI;
119 ScalarEvolution *SE;
120 TargetLibraryInfo *TLI;
121 const TargetTransformInfo *TTI;
122 const DataLayout *DL;
123 OptimizationRemarkEmitter &ORE;
124 bool ApplyCodeSizeHeuristics;
125
126public:
127 explicit LoopIdiomRecognize(AliasAnalysis *AA, DominatorTree *DT,
128 LoopInfo *LI, ScalarEvolution *SE,
129 TargetLibraryInfo *TLI,
130 const TargetTransformInfo *TTI,
131 const DataLayout *DL,
132 OptimizationRemarkEmitter &ORE)
133 : AA(AA), DT(DT), LI(LI), SE(SE), TLI(TLI), TTI(TTI), DL(DL), ORE(ORE) {}
134
135 bool runOnLoop(Loop *L);
136
137private:
138 using StoreList = SmallVector<StoreInst *, 8>;
139 using StoreListMap = MapVector<Value *, StoreList>;
140
141 StoreListMap StoreRefsForMemset;
142 StoreListMap StoreRefsForMemsetPattern;
143 StoreList StoreRefsForMemcpy;
144 bool HasMemset;
145 bool HasMemsetPattern;
146 bool HasMemcpy;
147
148 /// Return code for isLegalStore()
149 enum LegalStoreKind {
150 None = 0,
151 Memset,
152 MemsetPattern,
153 Memcpy,
154 UnorderedAtomicMemcpy,
155 DontUse // Dummy retval never to be used. Allows catching errors in retval
156 // handling.
157 };
158
159 /// \name Countable Loop Idiom Handling
160 /// @{
161
162 bool runOnCountableLoop();
163 bool runOnLoopBlock(BasicBlock *BB, const SCEV *BECount,
164 SmallVectorImpl<BasicBlock *> &ExitBlocks);
165
166 void collectStores(BasicBlock *BB);
167 LegalStoreKind isLegalStore(StoreInst *SI);
168 enum class ForMemset { No, Yes };
169 bool processLoopStores(SmallVectorImpl<StoreInst *> &SL, const SCEV *BECount,
170 ForMemset For);
171 bool processLoopMemSet(MemSetInst *MSI, const SCEV *BECount);
172
173 bool processLoopStridedStore(Value *DestPtr, unsigned StoreSize,
174 unsigned StoreAlignment, Value *StoredVal,
175 Instruction *TheStore,
176 SmallPtrSetImpl<Instruction *> &Stores,
177 const SCEVAddRecExpr *Ev, const SCEV *BECount,
178 bool NegStride, bool IsLoopMemset = false);
179 bool processLoopStoreOfLoopLoad(StoreInst *SI, const SCEV *BECount);
180 bool avoidLIRForMultiBlockLoop(bool IsMemset = false,
181 bool IsLoopMemset = false);
182
183 /// @}
184 /// \name Noncountable Loop Idiom Handling
185 /// @{
186
187 bool runOnNoncountableLoop();
188
189 bool recognizePopcount();
190 void transformLoopToPopcount(BasicBlock *PreCondBB, Instruction *CntInst,
191 PHINode *CntPhi, Value *Var);
192 bool recognizeAndInsertFFS(); /// Find First Set: ctlz or cttz
193 void transformLoopToCountable(Intrinsic::ID IntrinID, BasicBlock *PreCondBB,
194 Instruction *CntInst, PHINode *CntPhi,
195 Value *Var, Instruction *DefX,
196 const DebugLoc &DL, bool ZeroCheck,
197 bool IsCntPhiUsedOutsideLoop);
198
199 /// @}
200};
201
202class LoopIdiomRecognizeLegacyPass : public LoopPass {
203public:
204 static char ID;
205
206 explicit LoopIdiomRecognizeLegacyPass() : LoopPass(ID) {
207 initializeLoopIdiomRecognizeLegacyPassPass(
208 *PassRegistry::getPassRegistry());
209 }
210
211 bool runOnLoop(Loop *L, LPPassManager &LPM) override {
212 if (skipLoop(L))
213 return false;
214
215 AliasAnalysis *AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
216 DominatorTree *DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
217 LoopInfo *LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
218 ScalarEvolution *SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE();
219 TargetLibraryInfo *TLI =
220 &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(
221 *L->getHeader()->getParent());
222 const TargetTransformInfo *TTI =
223 &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(
224 *L->getHeader()->getParent());
225 const DataLayout *DL = &L->getHeader()->getModule()->getDataLayout();
226
227 // For the old PM, we can't use OptimizationRemarkEmitter as an analysis
228 // pass. Function analyses need to be preserved across loop transformations
229 // but ORE cannot be preserved (see comment before the pass definition).
230 OptimizationRemarkEmitter ORE(L->getHeader()->getParent());
231
232 LoopIdiomRecognize LIR(AA, DT, LI, SE, TLI, TTI, DL, ORE);
233 return LIR.runOnLoop(L);
234 }
235
236 /// This transformation requires natural loop information & requires that
237 /// loop preheaders be inserted into the CFG.
238 void getAnalysisUsage(AnalysisUsage &AU) const override {
239 AU.addRequired<TargetLibraryInfoWrapperPass>();
240 AU.addRequired<TargetTransformInfoWrapperPass>();
241 getLoopAnalysisUsage(AU);
242 }
243};
244
245} // end anonymous namespace
246
247char LoopIdiomRecognizeLegacyPass::ID = 0;
248
249PreservedAnalyses LoopIdiomRecognizePass::run(Loop &L, LoopAnalysisManager &AM,
250 LoopStandardAnalysisResults &AR,
251 LPMUpdater &) {
252 const auto *DL = &L.getHeader()->getModule()->getDataLayout();
253
254 const auto &FAM =
255 AM.getResult<FunctionAnalysisManagerLoopProxy>(L, AR).getManager();
256 Function *F = L.getHeader()->getParent();
257
258 auto *ORE = FAM.getCachedResult<OptimizationRemarkEmitterAnalysis>(*F);
259 // FIXME: This should probably be optional rather than required.
260 if (!ORE
0.1
'ORE' is non-null
0.1
'ORE' is non-null
)
1
Taking false branch
261 report_fatal_error(
262 "LoopIdiomRecognizePass: OptimizationRemarkEmitterAnalysis not cached "
263 "at a higher level");
264
265 LoopIdiomRecognize LIR(&AR.AA, &AR.DT, &AR.LI, &AR.SE, &AR.TLI, &AR.TTI, DL,
266 *ORE);
267 if (!LIR.runOnLoop(&L))
2
Calling 'LoopIdiomRecognize::runOnLoop'
268 return PreservedAnalyses::all();
269
270 return getLoopPassPreservedAnalyses();
271}
272
273INITIALIZE_PASS_BEGIN(LoopIdiomRecognizeLegacyPass, "loop-idiom",static void *initializeLoopIdiomRecognizeLegacyPassPassOnce(PassRegistry
&Registry) {
274 "Recognize loop idioms", false, false)static void *initializeLoopIdiomRecognizeLegacyPassPassOnce(PassRegistry
&Registry) {
275INITIALIZE_PASS_DEPENDENCY(LoopPass)initializeLoopPassPass(Registry);
276INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)initializeTargetLibraryInfoWrapperPassPass(Registry);
277INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)initializeTargetTransformInfoWrapperPassPass(Registry);
278INITIALIZE_PASS_END(LoopIdiomRecognizeLegacyPass, "loop-idiom",PassInfo *PI = new PassInfo( "Recognize loop idioms", "loop-idiom"
, &LoopIdiomRecognizeLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<LoopIdiomRecognizeLegacyPass>), false,
false); Registry.registerPass(*PI, true); return PI; } static
llvm::once_flag InitializeLoopIdiomRecognizeLegacyPassPassFlag
; void llvm::initializeLoopIdiomRecognizeLegacyPassPass(PassRegistry
&Registry) { llvm::call_once(InitializeLoopIdiomRecognizeLegacyPassPassFlag
, initializeLoopIdiomRecognizeLegacyPassPassOnce, std::ref(Registry
)); }
279 "Recognize loop idioms", false, false)PassInfo *PI = new PassInfo( "Recognize loop idioms", "loop-idiom"
, &LoopIdiomRecognizeLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<LoopIdiomRecognizeLegacyPass>), false,
false); Registry.registerPass(*PI, true); return PI; } static
llvm::once_flag InitializeLoopIdiomRecognizeLegacyPassPassFlag
; void llvm::initializeLoopIdiomRecognizeLegacyPassPass(PassRegistry
&Registry) { llvm::call_once(InitializeLoopIdiomRecognizeLegacyPassPassFlag
, initializeLoopIdiomRecognizeLegacyPassPassOnce, std::ref(Registry
)); }
280
281Pass *llvm::createLoopIdiomPass() { return new LoopIdiomRecognizeLegacyPass(); }
282
283static void deleteDeadInstruction(Instruction *I) {
284 I->replaceAllUsesWith(UndefValue::get(I->getType()));
285 I->eraseFromParent();
286}
287
288//===----------------------------------------------------------------------===//
289//
290// Implementation of LoopIdiomRecognize
291//
292//===----------------------------------------------------------------------===//
293
294bool LoopIdiomRecognize::runOnLoop(Loop *L) {
295 CurLoop = L;
296 // If the loop could not be converted to canonical form, it must have an
297 // indirectbr in it, just give up.
298 if (!L->getLoopPreheader())
3
Assuming the condition is false
4
Taking false branch
299 return false;
300
301 // Disable loop idiom recognition if the function's name is a common idiom.
302 StringRef Name = L->getHeader()->getParent()->getName();
303 if (Name == "memset" || Name == "memcpy")
5
Assuming the condition is false
6
Assuming the condition is false
7
Taking false branch
304 return false;
305
306 // Determine if code size heuristics need to be applied.
307 ApplyCodeSizeHeuristics =
308 L->getHeader()->getParent()->hasOptSize() && UseLIRCodeSizeHeurs;
8
Assuming the condition is false
309
310 HasMemset = TLI->has(LibFunc_memset);
311 HasMemsetPattern = TLI->has(LibFunc_memset_pattern16);
312 HasMemcpy = TLI->has(LibFunc_memcpy);
313
314 if (HasMemset
8.1
Field 'HasMemset' is false
8.1
Field 'HasMemset' is false
|| HasMemsetPattern
8.2
Field 'HasMemsetPattern' is false
8.2
Field 'HasMemsetPattern' is false
|| HasMemcpy
8.3
Field 'HasMemcpy' is false
8.3
Field 'HasMemcpy' is false
)
9
Taking false branch
315 if (SE->hasLoopInvariantBackedgeTakenCount(L))
316 return runOnCountableLoop();
317
318 return runOnNoncountableLoop();
10
Calling 'LoopIdiomRecognize::runOnNoncountableLoop'
319}
320
321bool LoopIdiomRecognize::runOnCountableLoop() {
322 const SCEV *BECount = SE->getBackedgeTakenCount(CurLoop);
323 assert(!isa<SCEVCouldNotCompute>(BECount) &&((!isa<SCEVCouldNotCompute>(BECount) && "runOnCountableLoop() called on a loop without a predictable"
"backedge-taken count") ? static_cast<void> (0) : __assert_fail
("!isa<SCEVCouldNotCompute>(BECount) && \"runOnCountableLoop() called on a loop without a predictable\" \"backedge-taken count\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 325, __PRETTY_FUNCTION__))
324 "runOnCountableLoop() called on a loop without a predictable"((!isa<SCEVCouldNotCompute>(BECount) && "runOnCountableLoop() called on a loop without a predictable"
"backedge-taken count") ? static_cast<void> (0) : __assert_fail
("!isa<SCEVCouldNotCompute>(BECount) && \"runOnCountableLoop() called on a loop without a predictable\" \"backedge-taken count\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 325, __PRETTY_FUNCTION__))
325 "backedge-taken count")((!isa<SCEVCouldNotCompute>(BECount) && "runOnCountableLoop() called on a loop without a predictable"
"backedge-taken count") ? static_cast<void> (0) : __assert_fail
("!isa<SCEVCouldNotCompute>(BECount) && \"runOnCountableLoop() called on a loop without a predictable\" \"backedge-taken count\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 325, __PRETTY_FUNCTION__))
;
326
327 // If this loop executes exactly one time, then it should be peeled, not
328 // optimized by this pass.
329 if (const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount))
330 if (BECst->getAPInt() == 0)
331 return false;
332
333 SmallVector<BasicBlock *, 8> ExitBlocks;
334 CurLoop->getUniqueExitBlocks(ExitBlocks);
335
336 LLVM_DEBUG(dbgs() << DEBUG_TYPE " Scanning: F["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Countable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
337 << CurLoop->getHeader()->getParent()->getName()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Countable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
338 << "] Countable Loop %" << CurLoop->getHeader()->getName()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Countable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
339 << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Countable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
;
340
341 bool MadeChange = false;
342
343 // The following transforms hoist stores/memsets into the loop pre-header.
344 // Give up if the loop has instructions may throw.
345 SimpleLoopSafetyInfo SafetyInfo;
346 SafetyInfo.computeLoopSafetyInfo(CurLoop);
347 if (SafetyInfo.anyBlockMayThrow())
348 return MadeChange;
349
350 // Scan all the blocks in the loop that are not in subloops.
351 for (auto *BB : CurLoop->getBlocks()) {
352 // Ignore blocks in subloops.
353 if (LI->getLoopFor(BB) != CurLoop)
354 continue;
355
356 MadeChange |= runOnLoopBlock(BB, BECount, ExitBlocks);
357 }
358 return MadeChange;
359}
360
361static APInt getStoreStride(const SCEVAddRecExpr *StoreEv) {
362 const SCEVConstant *ConstStride = cast<SCEVConstant>(StoreEv->getOperand(1));
363 return ConstStride->getAPInt();
364}
365
366/// getMemSetPatternValue - If a strided store of the specified value is safe to
367/// turn into a memset_pattern16, return a ConstantArray of 16 bytes that should
368/// be passed in. Otherwise, return null.
369///
370/// Note that we don't ever attempt to use memset_pattern8 or 4, because these
371/// just replicate their input array and then pass on to memset_pattern16.
372static Constant *getMemSetPatternValue(Value *V, const DataLayout *DL) {
373 // FIXME: This could check for UndefValue because it can be merged into any
374 // other valid pattern.
375
376 // If the value isn't a constant, we can't promote it to being in a constant
377 // array. We could theoretically do a store to an alloca or something, but
378 // that doesn't seem worthwhile.
379 Constant *C = dyn_cast<Constant>(V);
380 if (!C)
381 return nullptr;
382
383 // Only handle simple values that are a power of two bytes in size.
384 uint64_t Size = DL->getTypeSizeInBits(V->getType());
385 if (Size == 0 || (Size & 7) || (Size & (Size - 1)))
386 return nullptr;
387
388 // Don't care enough about darwin/ppc to implement this.
389 if (DL->isBigEndian())
390 return nullptr;
391
392 // Convert to size in bytes.
393 Size /= 8;
394
395 // TODO: If CI is larger than 16-bytes, we can try slicing it in half to see
396 // if the top and bottom are the same (e.g. for vectors and large integers).
397 if (Size > 16)
398 return nullptr;
399
400 // If the constant is exactly 16 bytes, just use it.
401 if (Size == 16)
402 return C;
403
404 // Otherwise, we'll use an array of the constants.
405 unsigned ArraySize = 16 / Size;
406 ArrayType *AT = ArrayType::get(V->getType(), ArraySize);
407 return ConstantArray::get(AT, std::vector<Constant *>(ArraySize, C));
408}
409
410LoopIdiomRecognize::LegalStoreKind
411LoopIdiomRecognize::isLegalStore(StoreInst *SI) {
412 // Don't touch volatile stores.
413 if (SI->isVolatile())
414 return LegalStoreKind::None;
415 // We only want simple or unordered-atomic stores.
416 if (!SI->isUnordered())
417 return LegalStoreKind::None;
418
419 // Don't convert stores of non-integral pointer types to memsets (which stores
420 // integers).
421 if (DL->isNonIntegralPointerType(SI->getValueOperand()->getType()))
422 return LegalStoreKind::None;
423
424 // Avoid merging nontemporal stores.
425 if (SI->getMetadata(LLVMContext::MD_nontemporal))
426 return LegalStoreKind::None;
427
428 Value *StoredVal = SI->getValueOperand();
429 Value *StorePtr = SI->getPointerOperand();
430
431 // Reject stores that are so large that they overflow an unsigned.
432 uint64_t SizeInBits = DL->getTypeSizeInBits(StoredVal->getType());
433 if ((SizeInBits & 7) || (SizeInBits >> 32) != 0)
434 return LegalStoreKind::None;
435
436 // See if the pointer expression is an AddRec like {base,+,1} on the current
437 // loop, which indicates a strided store. If we have something else, it's a
438 // random store we can't handle.
439 const SCEVAddRecExpr *StoreEv =
440 dyn_cast<SCEVAddRecExpr>(SE->getSCEV(StorePtr));
441 if (!StoreEv || StoreEv->getLoop() != CurLoop || !StoreEv->isAffine())
442 return LegalStoreKind::None;
443
444 // Check to see if we have a constant stride.
445 if (!isa<SCEVConstant>(StoreEv->getOperand(1)))
446 return LegalStoreKind::None;
447
448 // See if the store can be turned into a memset.
449
450 // If the stored value is a byte-wise value (like i32 -1), then it may be
451 // turned into a memset of i8 -1, assuming that all the consecutive bytes
452 // are stored. A store of i32 0x01020304 can never be turned into a memset,
453 // but it can be turned into memset_pattern if the target supports it.
454 Value *SplatValue = isBytewiseValue(StoredVal, *DL);
455 Constant *PatternValue = nullptr;
456
457 // Note: memset and memset_pattern on unordered-atomic is yet not supported
458 bool UnorderedAtomic = SI->isUnordered() && !SI->isSimple();
459
460 // If we're allowed to form a memset, and the stored value would be
461 // acceptable for memset, use it.
462 if (!UnorderedAtomic && HasMemset && SplatValue &&
463 // Verify that the stored value is loop invariant. If not, we can't
464 // promote the memset.
465 CurLoop->isLoopInvariant(SplatValue)) {
466 // It looks like we can use SplatValue.
467 return LegalStoreKind::Memset;
468 } else if (!UnorderedAtomic && HasMemsetPattern &&
469 // Don't create memset_pattern16s with address spaces.
470 StorePtr->getType()->getPointerAddressSpace() == 0 &&
471 (PatternValue = getMemSetPatternValue(StoredVal, DL))) {
472 // It looks like we can use PatternValue!
473 return LegalStoreKind::MemsetPattern;
474 }
475
476 // Otherwise, see if the store can be turned into a memcpy.
477 if (HasMemcpy) {
478 // Check to see if the stride matches the size of the store. If so, then we
479 // know that every byte is touched in the loop.
480 APInt Stride = getStoreStride(StoreEv);
481 unsigned StoreSize = DL->getTypeStoreSize(SI->getValueOperand()->getType());
482 if (StoreSize != Stride && StoreSize != -Stride)
483 return LegalStoreKind::None;
484
485 // The store must be feeding a non-volatile load.
486 LoadInst *LI = dyn_cast<LoadInst>(SI->getValueOperand());
487
488 // Only allow non-volatile loads
489 if (!LI || LI->isVolatile())
490 return LegalStoreKind::None;
491 // Only allow simple or unordered-atomic loads
492 if (!LI->isUnordered())
493 return LegalStoreKind::None;
494
495 // See if the pointer expression is an AddRec like {base,+,1} on the current
496 // loop, which indicates a strided load. If we have something else, it's a
497 // random load we can't handle.
498 const SCEVAddRecExpr *LoadEv =
499 dyn_cast<SCEVAddRecExpr>(SE->getSCEV(LI->getPointerOperand()));
500 if (!LoadEv || LoadEv->getLoop() != CurLoop || !LoadEv->isAffine())
501 return LegalStoreKind::None;
502
503 // The store and load must share the same stride.
504 if (StoreEv->getOperand(1) != LoadEv->getOperand(1))
505 return LegalStoreKind::None;
506
507 // Success. This store can be converted into a memcpy.
508 UnorderedAtomic = UnorderedAtomic || LI->isAtomic();
509 return UnorderedAtomic ? LegalStoreKind::UnorderedAtomicMemcpy
510 : LegalStoreKind::Memcpy;
511 }
512 // This store can't be transformed into a memset/memcpy.
513 return LegalStoreKind::None;
514}
515
516void LoopIdiomRecognize::collectStores(BasicBlock *BB) {
517 StoreRefsForMemset.clear();
518 StoreRefsForMemsetPattern.clear();
519 StoreRefsForMemcpy.clear();
520 for (Instruction &I : *BB) {
521 StoreInst *SI = dyn_cast<StoreInst>(&I);
522 if (!SI)
523 continue;
524
525 // Make sure this is a strided store with a constant stride.
526 switch (isLegalStore(SI)) {
527 case LegalStoreKind::None:
528 // Nothing to do
529 break;
530 case LegalStoreKind::Memset: {
531 // Find the base pointer.
532 Value *Ptr = GetUnderlyingObject(SI->getPointerOperand(), *DL);
533 StoreRefsForMemset[Ptr].push_back(SI);
534 } break;
535 case LegalStoreKind::MemsetPattern: {
536 // Find the base pointer.
537 Value *Ptr = GetUnderlyingObject(SI->getPointerOperand(), *DL);
538 StoreRefsForMemsetPattern[Ptr].push_back(SI);
539 } break;
540 case LegalStoreKind::Memcpy:
541 case LegalStoreKind::UnorderedAtomicMemcpy:
542 StoreRefsForMemcpy.push_back(SI);
543 break;
544 default:
545 assert(false && "unhandled return value")((false && "unhandled return value") ? static_cast<
void> (0) : __assert_fail ("false && \"unhandled return value\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 545, __PRETTY_FUNCTION__))
;
546 break;
547 }
548 }
549}
550
551/// runOnLoopBlock - Process the specified block, which lives in a counted loop
552/// with the specified backedge count. This block is known to be in the current
553/// loop and not in any subloops.
554bool LoopIdiomRecognize::runOnLoopBlock(
555 BasicBlock *BB, const SCEV *BECount,
556 SmallVectorImpl<BasicBlock *> &ExitBlocks) {
557 // We can only promote stores in this block if they are unconditionally
558 // executed in the loop. For a block to be unconditionally executed, it has
559 // to dominate all the exit blocks of the loop. Verify this now.
560 for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i)
561 if (!DT->dominates(BB, ExitBlocks[i]))
562 return false;
563
564 bool MadeChange = false;
565 // Look for store instructions, which may be optimized to memset/memcpy.
566 collectStores(BB);
567
568 // Look for a single store or sets of stores with a common base, which can be
569 // optimized into a memset (memset_pattern). The latter most commonly happens
570 // with structs and handunrolled loops.
571 for (auto &SL : StoreRefsForMemset)
572 MadeChange |= processLoopStores(SL.second, BECount, ForMemset::Yes);
573
574 for (auto &SL : StoreRefsForMemsetPattern)
575 MadeChange |= processLoopStores(SL.second, BECount, ForMemset::No);
576
577 // Optimize the store into a memcpy, if it feeds an similarly strided load.
578 for (auto &SI : StoreRefsForMemcpy)
579 MadeChange |= processLoopStoreOfLoopLoad(SI, BECount);
580
581 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;) {
582 Instruction *Inst = &*I++;
583 // Look for memset instructions, which may be optimized to a larger memset.
584 if (MemSetInst *MSI = dyn_cast<MemSetInst>(Inst)) {
585 WeakTrackingVH InstPtr(&*I);
586 if (!processLoopMemSet(MSI, BECount))
587 continue;
588 MadeChange = true;
589
590 // If processing the memset invalidated our iterator, start over from the
591 // top of the block.
592 if (!InstPtr)
593 I = BB->begin();
594 continue;
595 }
596 }
597
598 return MadeChange;
599}
600
601/// See if this store(s) can be promoted to a memset.
602bool LoopIdiomRecognize::processLoopStores(SmallVectorImpl<StoreInst *> &SL,
603 const SCEV *BECount, ForMemset For) {
604 // Try to find consecutive stores that can be transformed into memsets.
605 SetVector<StoreInst *> Heads, Tails;
606 SmallDenseMap<StoreInst *, StoreInst *> ConsecutiveChain;
607
608 // Do a quadratic search on all of the given stores and find
609 // all of the pairs of stores that follow each other.
610 SmallVector<unsigned, 16> IndexQueue;
611 for (unsigned i = 0, e = SL.size(); i < e; ++i) {
612 assert(SL[i]->isSimple() && "Expected only non-volatile stores.")((SL[i]->isSimple() && "Expected only non-volatile stores."
) ? static_cast<void> (0) : __assert_fail ("SL[i]->isSimple() && \"Expected only non-volatile stores.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 612, __PRETTY_FUNCTION__))
;
613
614 Value *FirstStoredVal = SL[i]->getValueOperand();
615 Value *FirstStorePtr = SL[i]->getPointerOperand();
616 const SCEVAddRecExpr *FirstStoreEv =
617 cast<SCEVAddRecExpr>(SE->getSCEV(FirstStorePtr));
618 APInt FirstStride = getStoreStride(FirstStoreEv);
619 unsigned FirstStoreSize = DL->getTypeStoreSize(SL[i]->getValueOperand()->getType());
620
621 // See if we can optimize just this store in isolation.
622 if (FirstStride == FirstStoreSize || -FirstStride == FirstStoreSize) {
623 Heads.insert(SL[i]);
624 continue;
625 }
626
627 Value *FirstSplatValue = nullptr;
628 Constant *FirstPatternValue = nullptr;
629
630 if (For == ForMemset::Yes)
631 FirstSplatValue = isBytewiseValue(FirstStoredVal, *DL);
632 else
633 FirstPatternValue = getMemSetPatternValue(FirstStoredVal, DL);
634
635 assert((FirstSplatValue || FirstPatternValue) &&(((FirstSplatValue || FirstPatternValue) && "Expected either splat value or pattern value."
) ? static_cast<void> (0) : __assert_fail ("(FirstSplatValue || FirstPatternValue) && \"Expected either splat value or pattern value.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 636, __PRETTY_FUNCTION__))
636 "Expected either splat value or pattern value.")(((FirstSplatValue || FirstPatternValue) && "Expected either splat value or pattern value."
) ? static_cast<void> (0) : __assert_fail ("(FirstSplatValue || FirstPatternValue) && \"Expected either splat value or pattern value.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 636, __PRETTY_FUNCTION__))
;
637
638 IndexQueue.clear();
639 // If a store has multiple consecutive store candidates, search Stores
640 // array according to the sequence: from i+1 to e, then from i-1 to 0.
641 // This is because usually pairing with immediate succeeding or preceding
642 // candidate create the best chance to find memset opportunity.
643 unsigned j = 0;
644 for (j = i + 1; j < e; ++j)
645 IndexQueue.push_back(j);
646 for (j = i; j > 0; --j)
647 IndexQueue.push_back(j - 1);
648
649 for (auto &k : IndexQueue) {
650 assert(SL[k]->isSimple() && "Expected only non-volatile stores.")((SL[k]->isSimple() && "Expected only non-volatile stores."
) ? static_cast<void> (0) : __assert_fail ("SL[k]->isSimple() && \"Expected only non-volatile stores.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 650, __PRETTY_FUNCTION__))
;
651 Value *SecondStorePtr = SL[k]->getPointerOperand();
652 const SCEVAddRecExpr *SecondStoreEv =
653 cast<SCEVAddRecExpr>(SE->getSCEV(SecondStorePtr));
654 APInt SecondStride = getStoreStride(SecondStoreEv);
655
656 if (FirstStride != SecondStride)
657 continue;
658
659 Value *SecondStoredVal = SL[k]->getValueOperand();
660 Value *SecondSplatValue = nullptr;
661 Constant *SecondPatternValue = nullptr;
662
663 if (For == ForMemset::Yes)
664 SecondSplatValue = isBytewiseValue(SecondStoredVal, *DL);
665 else
666 SecondPatternValue = getMemSetPatternValue(SecondStoredVal, DL);
667
668 assert((SecondSplatValue || SecondPatternValue) &&(((SecondSplatValue || SecondPatternValue) && "Expected either splat value or pattern value."
) ? static_cast<void> (0) : __assert_fail ("(SecondSplatValue || SecondPatternValue) && \"Expected either splat value or pattern value.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 669, __PRETTY_FUNCTION__))
669 "Expected either splat value or pattern value.")(((SecondSplatValue || SecondPatternValue) && "Expected either splat value or pattern value."
) ? static_cast<void> (0) : __assert_fail ("(SecondSplatValue || SecondPatternValue) && \"Expected either splat value or pattern value.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 669, __PRETTY_FUNCTION__))
;
670
671 if (isConsecutiveAccess(SL[i], SL[k], *DL, *SE, false)) {
672 if (For == ForMemset::Yes) {
673 if (isa<UndefValue>(FirstSplatValue))
674 FirstSplatValue = SecondSplatValue;
675 if (FirstSplatValue != SecondSplatValue)
676 continue;
677 } else {
678 if (isa<UndefValue>(FirstPatternValue))
679 FirstPatternValue = SecondPatternValue;
680 if (FirstPatternValue != SecondPatternValue)
681 continue;
682 }
683 Tails.insert(SL[k]);
684 Heads.insert(SL[i]);
685 ConsecutiveChain[SL[i]] = SL[k];
686 break;
687 }
688 }
689 }
690
691 // We may run into multiple chains that merge into a single chain. We mark the
692 // stores that we transformed so that we don't visit the same store twice.
693 SmallPtrSet<Value *, 16> TransformedStores;
694 bool Changed = false;
695
696 // For stores that start but don't end a link in the chain:
697 for (SetVector<StoreInst *>::iterator it = Heads.begin(), e = Heads.end();
698 it != e; ++it) {
699 if (Tails.count(*it))
700 continue;
701
702 // We found a store instr that starts a chain. Now follow the chain and try
703 // to transform it.
704 SmallPtrSet<Instruction *, 8> AdjacentStores;
705 StoreInst *I = *it;
706
707 StoreInst *HeadStore = I;
708 unsigned StoreSize = 0;
709
710 // Collect the chain into a list.
711 while (Tails.count(I) || Heads.count(I)) {
712 if (TransformedStores.count(I))
713 break;
714 AdjacentStores.insert(I);
715
716 StoreSize += DL->getTypeStoreSize(I->getValueOperand()->getType());
717 // Move to the next value in the chain.
718 I = ConsecutiveChain[I];
719 }
720
721 Value *StoredVal = HeadStore->getValueOperand();
722 Value *StorePtr = HeadStore->getPointerOperand();
723 const SCEVAddRecExpr *StoreEv = cast<SCEVAddRecExpr>(SE->getSCEV(StorePtr));
724 APInt Stride = getStoreStride(StoreEv);
725
726 // Check to see if the stride matches the size of the stores. If so, then
727 // we know that every byte is touched in the loop.
728 if (StoreSize != Stride && StoreSize != -Stride)
729 continue;
730
731 bool NegStride = StoreSize == -Stride;
732
733 if (processLoopStridedStore(StorePtr, StoreSize, HeadStore->getAlignment(),
734 StoredVal, HeadStore, AdjacentStores, StoreEv,
735 BECount, NegStride)) {
736 TransformedStores.insert(AdjacentStores.begin(), AdjacentStores.end());
737 Changed = true;
738 }
739 }
740
741 return Changed;
742}
743
744/// processLoopMemSet - See if this memset can be promoted to a large memset.
745bool LoopIdiomRecognize::processLoopMemSet(MemSetInst *MSI,
746 const SCEV *BECount) {
747 // We can only handle non-volatile memsets with a constant size.
748 if (MSI->isVolatile() || !isa<ConstantInt>(MSI->getLength()))
749 return false;
750
751 // If we're not allowed to hack on memset, we fail.
752 if (!HasMemset)
753 return false;
754
755 Value *Pointer = MSI->getDest();
756
757 // See if the pointer expression is an AddRec like {base,+,1} on the current
758 // loop, which indicates a strided store. If we have something else, it's a
759 // random store we can't handle.
760 const SCEVAddRecExpr *Ev = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(Pointer));
761 if (!Ev || Ev->getLoop() != CurLoop || !Ev->isAffine())
762 return false;
763
764 // Reject memsets that are so large that they overflow an unsigned.
765 uint64_t SizeInBytes = cast<ConstantInt>(MSI->getLength())->getZExtValue();
766 if ((SizeInBytes >> 32) != 0)
767 return false;
768
769 // Check to see if the stride matches the size of the memset. If so, then we
770 // know that every byte is touched in the loop.
771 const SCEVConstant *ConstStride = dyn_cast<SCEVConstant>(Ev->getOperand(1));
772 if (!ConstStride)
773 return false;
774
775 APInt Stride = ConstStride->getAPInt();
776 if (SizeInBytes != Stride && SizeInBytes != -Stride)
777 return false;
778
779 // Verify that the memset value is loop invariant. If not, we can't promote
780 // the memset.
781 Value *SplatValue = MSI->getValue();
782 if (!SplatValue || !CurLoop->isLoopInvariant(SplatValue))
783 return false;
784
785 SmallPtrSet<Instruction *, 1> MSIs;
786 MSIs.insert(MSI);
787 bool NegStride = SizeInBytes == -Stride;
788 return processLoopStridedStore(Pointer, (unsigned)SizeInBytes,
789 MSI->getDestAlignment(), SplatValue, MSI, MSIs,
790 Ev, BECount, NegStride, /*IsLoopMemset=*/true);
791}
792
793/// mayLoopAccessLocation - Return true if the specified loop might access the
794/// specified pointer location, which is a loop-strided access. The 'Access'
795/// argument specifies what the verboten forms of access are (read or write).
796static bool
797mayLoopAccessLocation(Value *Ptr, ModRefInfo Access, Loop *L,
798 const SCEV *BECount, unsigned StoreSize,
799 AliasAnalysis &AA,
800 SmallPtrSetImpl<Instruction *> &IgnoredStores) {
801 // Get the location that may be stored across the loop. Since the access is
802 // strided positively through memory, we say that the modified location starts
803 // at the pointer and has infinite size.
804 LocationSize AccessSize = LocationSize::unknown();
805
806 // If the loop iterates a fixed number of times, we can refine the access size
807 // to be exactly the size of the memset, which is (BECount+1)*StoreSize
808 if (const SCEVConstant *BECst = dyn_cast<SCEVConstant>(BECount))
809 AccessSize = LocationSize::precise((BECst->getValue()->getZExtValue() + 1) *
810 StoreSize);
811
812 // TODO: For this to be really effective, we have to dive into the pointer
813 // operand in the store. Store to &A[i] of 100 will always return may alias
814 // with store of &A[100], we need to StoreLoc to be "A" with size of 100,
815 // which will then no-alias a store to &A[100].
816 MemoryLocation StoreLoc(Ptr, AccessSize);
817
818 for (Loop::block_iterator BI = L->block_begin(), E = L->block_end(); BI != E;
819 ++BI)
820 for (Instruction &I : **BI)
821 if (IgnoredStores.count(&I) == 0 &&
822 isModOrRefSet(
823 intersectModRef(AA.getModRefInfo(&I, StoreLoc), Access)))
824 return true;
825
826 return false;
827}
828
829// If we have a negative stride, Start refers to the end of the memory location
830// we're trying to memset. Therefore, we need to recompute the base pointer,
831// which is just Start - BECount*Size.
832static const SCEV *getStartForNegStride(const SCEV *Start, const SCEV *BECount,
833 Type *IntPtr, unsigned StoreSize,
834 ScalarEvolution *SE) {
835 const SCEV *Index = SE->getTruncateOrZeroExtend(BECount, IntPtr);
836 if (StoreSize != 1)
837 Index = SE->getMulExpr(Index, SE->getConstant(IntPtr, StoreSize),
838 SCEV::FlagNUW);
839 return SE->getMinusSCEV(Start, Index);
840}
841
842/// Compute the number of bytes as a SCEV from the backedge taken count.
843///
844/// This also maps the SCEV into the provided type and tries to handle the
845/// computation in a way that will fold cleanly.
846static const SCEV *getNumBytes(const SCEV *BECount, Type *IntPtr,
847 unsigned StoreSize, Loop *CurLoop,
848 const DataLayout *DL, ScalarEvolution *SE) {
849 const SCEV *NumBytesS;
850 // The # stored bytes is (BECount+1)*Size. Expand the trip count out to
851 // pointer size if it isn't already.
852 //
853 // If we're going to need to zero extend the BE count, check if we can add
854 // one to it prior to zero extending without overflow. Provided this is safe,
855 // it allows better simplification of the +1.
856 if (DL->getTypeSizeInBits(BECount->getType()) <
857 DL->getTypeSizeInBits(IntPtr) &&
858 SE->isLoopEntryGuardedByCond(
859 CurLoop, ICmpInst::ICMP_NE, BECount,
860 SE->getNegativeSCEV(SE->getOne(BECount->getType())))) {
861 NumBytesS = SE->getZeroExtendExpr(
862 SE->getAddExpr(BECount, SE->getOne(BECount->getType()), SCEV::FlagNUW),
863 IntPtr);
864 } else {
865 NumBytesS = SE->getAddExpr(SE->getTruncateOrZeroExtend(BECount, IntPtr),
866 SE->getOne(IntPtr), SCEV::FlagNUW);
867 }
868
869 // And scale it based on the store size.
870 if (StoreSize != 1) {
871 NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize),
872 SCEV::FlagNUW);
873 }
874 return NumBytesS;
875}
876
877/// processLoopStridedStore - We see a strided store of some value. If we can
878/// transform this into a memset or memset_pattern in the loop preheader, do so.
879bool LoopIdiomRecognize::processLoopStridedStore(
880 Value *DestPtr, unsigned StoreSize, unsigned StoreAlignment,
881 Value *StoredVal, Instruction *TheStore,
882 SmallPtrSetImpl<Instruction *> &Stores, const SCEVAddRecExpr *Ev,
883 const SCEV *BECount, bool NegStride, bool IsLoopMemset) {
884 Value *SplatValue = isBytewiseValue(StoredVal, *DL);
885 Constant *PatternValue = nullptr;
886
887 if (!SplatValue)
888 PatternValue = getMemSetPatternValue(StoredVal, DL);
889
890 assert((SplatValue || PatternValue) &&(((SplatValue || PatternValue) && "Expected either splat value or pattern value."
) ? static_cast<void> (0) : __assert_fail ("(SplatValue || PatternValue) && \"Expected either splat value or pattern value.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 891, __PRETTY_FUNCTION__))
891 "Expected either splat value or pattern value.")(((SplatValue || PatternValue) && "Expected either splat value or pattern value."
) ? static_cast<void> (0) : __assert_fail ("(SplatValue || PatternValue) && \"Expected either splat value or pattern value.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 891, __PRETTY_FUNCTION__))
;
892
893 // The trip count of the loop and the base pointer of the addrec SCEV is
894 // guaranteed to be loop invariant, which means that it should dominate the
895 // header. This allows us to insert code for it in the preheader.
896 unsigned DestAS = DestPtr->getType()->getPointerAddressSpace();
897 BasicBlock *Preheader = CurLoop->getLoopPreheader();
898 IRBuilder<> Builder(Preheader->getTerminator());
899 SCEVExpander Expander(*SE, *DL, "loop-idiom");
900
901 Type *DestInt8PtrTy = Builder.getInt8PtrTy(DestAS);
902 Type *IntPtr = Builder.getIntPtrTy(*DL, DestAS);
903
904 const SCEV *Start = Ev->getStart();
905 // Handle negative strided loops.
906 if (NegStride)
907 Start = getStartForNegStride(Start, BECount, IntPtr, StoreSize, SE);
908
909 // TODO: ideally we should still be able to generate memset if SCEV expander
910 // is taught to generate the dependencies at the latest point.
911 if (!isSafeToExpand(Start, *SE))
912 return false;
913
914 // Okay, we have a strided store "p[i]" of a splattable value. We can turn
915 // this into a memset in the loop preheader now if we want. However, this
916 // would be unsafe to do if there is anything else in the loop that may read
917 // or write to the aliased location. Check for any overlap by generating the
918 // base pointer and checking the region.
919 Value *BasePtr =
920 Expander.expandCodeFor(Start, DestInt8PtrTy, Preheader->getTerminator());
921 if (mayLoopAccessLocation(BasePtr, ModRefInfo::ModRef, CurLoop, BECount,
922 StoreSize, *AA, Stores)) {
923 Expander.clear();
924 // If we generated new code for the base pointer, clean up.
925 RecursivelyDeleteTriviallyDeadInstructions(BasePtr, TLI);
926 return false;
927 }
928
929 if (avoidLIRForMultiBlockLoop(/*IsMemset=*/true, IsLoopMemset))
930 return false;
931
932 // Okay, everything looks good, insert the memset.
933
934 const SCEV *NumBytesS =
935 getNumBytes(BECount, IntPtr, StoreSize, CurLoop, DL, SE);
936
937 // TODO: ideally we should still be able to generate memset if SCEV expander
938 // is taught to generate the dependencies at the latest point.
939 if (!isSafeToExpand(NumBytesS, *SE))
940 return false;
941
942 Value *NumBytes =
943 Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator());
944
945 CallInst *NewCall;
946 if (SplatValue) {
947 NewCall =
948 Builder.CreateMemSet(BasePtr, SplatValue, NumBytes, StoreAlignment);
949 } else {
950 // Everything is emitted in default address space
951 Type *Int8PtrTy = DestInt8PtrTy;
952
953 Module *M = TheStore->getModule();
954 StringRef FuncName = "memset_pattern16";
955 FunctionCallee MSP = M->getOrInsertFunction(FuncName, Builder.getVoidTy(),
956 Int8PtrTy, Int8PtrTy, IntPtr);
957 inferLibFuncAttributes(M, FuncName, *TLI);
958
959 // Otherwise we should form a memset_pattern16. PatternValue is known to be
960 // an constant array of 16-bytes. Plop the value into a mergable global.
961 GlobalVariable *GV = new GlobalVariable(*M, PatternValue->getType(), true,
962 GlobalValue::PrivateLinkage,
963 PatternValue, ".memset_pattern");
964 GV->setUnnamedAddr(GlobalValue::UnnamedAddr::Global); // Ok to merge these.
965 GV->setAlignment(Align(16));
966 Value *PatternPtr = ConstantExpr::getBitCast(GV, Int8PtrTy);
967 NewCall = Builder.CreateCall(MSP, {BasePtr, PatternPtr, NumBytes});
968 }
969
970 LLVM_DEBUG(dbgs() << " Formed memset: " << *NewCall << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " Formed memset: " <<
*NewCall << "\n" << " from store to: " <<
*Ev << " at: " << *TheStore << "\n"; } } while
(false)
971 << " from store to: " << *Ev << " at: " << *TheStoredo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " Formed memset: " <<
*NewCall << "\n" << " from store to: " <<
*Ev << " at: " << *TheStore << "\n"; } } while
(false)
972 << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " Formed memset: " <<
*NewCall << "\n" << " from store to: " <<
*Ev << " at: " << *TheStore << "\n"; } } while
(false)
;
973 NewCall->setDebugLoc(TheStore->getDebugLoc());
974
975 ORE.emit([&]() {
976 return OptimizationRemark(DEBUG_TYPE"loop-idiom", "ProcessLoopStridedStore",
977 NewCall->getDebugLoc(), Preheader)
978 << "Transformed loop-strided store into a call to "
979 << ore::NV("NewFunction", NewCall->getCalledFunction())
980 << "() function";
981 });
982
983 // Okay, the memset has been formed. Zap the original store and anything that
984 // feeds into it.
985 for (auto *I : Stores)
986 deleteDeadInstruction(I);
987 ++NumMemSet;
988 return true;
989}
990
991/// If the stored value is a strided load in the same loop with the same stride
992/// this may be transformable into a memcpy. This kicks in for stuff like
993/// for (i) A[i] = B[i];
994bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(StoreInst *SI,
995 const SCEV *BECount) {
996 assert(SI->isUnordered() && "Expected only non-volatile non-ordered stores.")((SI->isUnordered() && "Expected only non-volatile non-ordered stores."
) ? static_cast<void> (0) : __assert_fail ("SI->isUnordered() && \"Expected only non-volatile non-ordered stores.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 996, __PRETTY_FUNCTION__))
;
997
998 Value *StorePtr = SI->getPointerOperand();
999 const SCEVAddRecExpr *StoreEv = cast<SCEVAddRecExpr>(SE->getSCEV(StorePtr));
1000 APInt Stride = getStoreStride(StoreEv);
1001 unsigned StoreSize = DL->getTypeStoreSize(SI->getValueOperand()->getType());
1002 bool NegStride = StoreSize == -Stride;
1003
1004 // The store must be feeding a non-volatile load.
1005 LoadInst *LI = cast<LoadInst>(SI->getValueOperand());
1006 assert(LI->isUnordered() && "Expected only non-volatile non-ordered loads.")((LI->isUnordered() && "Expected only non-volatile non-ordered loads."
) ? static_cast<void> (0) : __assert_fail ("LI->isUnordered() && \"Expected only non-volatile non-ordered loads.\""
, "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 1006, __PRETTY_FUNCTION__))
;
1007
1008 // See if the pointer expression is an AddRec like {base,+,1} on the current
1009 // loop, which indicates a strided load. If we have something else, it's a
1010 // random load we can't handle.
1011 const SCEVAddRecExpr *LoadEv =
1012 cast<SCEVAddRecExpr>(SE->getSCEV(LI->getPointerOperand()));
1013
1014 // The trip count of the loop and the base pointer of the addrec SCEV is
1015 // guaranteed to be loop invariant, which means that it should dominate the
1016 // header. This allows us to insert code for it in the preheader.
1017 BasicBlock *Preheader = CurLoop->getLoopPreheader();
1018 IRBuilder<> Builder(Preheader->getTerminator());
1019 SCEVExpander Expander(*SE, *DL, "loop-idiom");
1020
1021 const SCEV *StrStart = StoreEv->getStart();
1022 unsigned StrAS = SI->getPointerAddressSpace();
1023 Type *IntPtrTy = Builder.getIntPtrTy(*DL, StrAS);
1024
1025 // Handle negative strided loops.
1026 if (NegStride)
1027 StrStart = getStartForNegStride(StrStart, BECount, IntPtrTy, StoreSize, SE);
1028
1029 // Okay, we have a strided store "p[i]" of a loaded value. We can turn
1030 // this into a memcpy in the loop preheader now if we want. However, this
1031 // would be unsafe to do if there is anything else in the loop that may read
1032 // or write the memory region we're storing to. This includes the load that
1033 // feeds the stores. Check for an alias by generating the base address and
1034 // checking everything.
1035 Value *StoreBasePtr = Expander.expandCodeFor(
1036 StrStart, Builder.getInt8PtrTy(StrAS), Preheader->getTerminator());
1037
1038 SmallPtrSet<Instruction *, 1> Stores;
1039 Stores.insert(SI);
1040 if (mayLoopAccessLocation(StoreBasePtr, ModRefInfo::ModRef, CurLoop, BECount,
1041 StoreSize, *AA, Stores)) {
1042 Expander.clear();
1043 // If we generated new code for the base pointer, clean up.
1044 RecursivelyDeleteTriviallyDeadInstructions(StoreBasePtr, TLI);
1045 return false;
1046 }
1047
1048 const SCEV *LdStart = LoadEv->getStart();
1049 unsigned LdAS = LI->getPointerAddressSpace();
1050
1051 // Handle negative strided loops.
1052 if (NegStride)
1053 LdStart = getStartForNegStride(LdStart, BECount, IntPtrTy, StoreSize, SE);
1054
1055 // For a memcpy, we have to make sure that the input array is not being
1056 // mutated by the loop.
1057 Value *LoadBasePtr = Expander.expandCodeFor(
1058 LdStart, Builder.getInt8PtrTy(LdAS), Preheader->getTerminator());
1059
1060 if (mayLoopAccessLocation(LoadBasePtr, ModRefInfo::Mod, CurLoop, BECount,
1061 StoreSize, *AA, Stores)) {
1062 Expander.clear();
1063 // If we generated new code for the base pointer, clean up.
1064 RecursivelyDeleteTriviallyDeadInstructions(LoadBasePtr, TLI);
1065 RecursivelyDeleteTriviallyDeadInstructions(StoreBasePtr, TLI);
1066 return false;
1067 }
1068
1069 if (avoidLIRForMultiBlockLoop())
1070 return false;
1071
1072 // Okay, everything is safe, we can transform this!
1073
1074 const SCEV *NumBytesS =
1075 getNumBytes(BECount, IntPtrTy, StoreSize, CurLoop, DL, SE);
1076
1077 Value *NumBytes =
1078 Expander.expandCodeFor(NumBytesS, IntPtrTy, Preheader->getTerminator());
1079
1080 CallInst *NewCall = nullptr;
1081 // Check whether to generate an unordered atomic memcpy:
1082 // If the load or store are atomic, then they must necessarily be unordered
1083 // by previous checks.
1084 if (!SI->isAtomic() && !LI->isAtomic())
1085 NewCall = Builder.CreateMemCpy(StoreBasePtr, SI->getAlignment(),
1086 LoadBasePtr, LI->getAlignment(), NumBytes);
1087 else {
1088 // We cannot allow unaligned ops for unordered load/store, so reject
1089 // anything where the alignment isn't at least the element size.
1090 unsigned Align = std::min(SI->getAlignment(), LI->getAlignment());
1091 if (Align < StoreSize)
1092 return false;
1093
1094 // If the element.atomic memcpy is not lowered into explicit
1095 // loads/stores later, then it will be lowered into an element-size
1096 // specific lib call. If the lib call doesn't exist for our store size, then
1097 // we shouldn't generate the memcpy.
1098 if (StoreSize > TTI->getAtomicMemIntrinsicMaxElementSize())
1099 return false;
1100
1101 // Create the call.
1102 // Note that unordered atomic loads/stores are *required* by the spec to
1103 // have an alignment but non-atomic loads/stores may not.
1104 NewCall = Builder.CreateElementUnorderedAtomicMemCpy(
1105 StoreBasePtr, SI->getAlignment(), LoadBasePtr, LI->getAlignment(),
1106 NumBytes, StoreSize);
1107 }
1108 NewCall->setDebugLoc(SI->getDebugLoc());
1109
1110 LLVM_DEBUG(dbgs() << " Formed memcpy: " << *NewCall << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " Formed memcpy: " <<
*NewCall << "\n" << " from load ptr=" <<
*LoadEv << " at: " << *LI << "\n" <<
" from store ptr=" << *StoreEv << " at: " <<
*SI << "\n"; } } while (false)
1111 << " from load ptr=" << *LoadEv << " at: " << *LI << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " Formed memcpy: " <<
*NewCall << "\n" << " from load ptr=" <<
*LoadEv << " at: " << *LI << "\n" <<
" from store ptr=" << *StoreEv << " at: " <<
*SI << "\n"; } } while (false)
1112 << " from store ptr=" << *StoreEv << " at: " << *SIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " Formed memcpy: " <<
*NewCall << "\n" << " from load ptr=" <<
*LoadEv << " at: " << *LI << "\n" <<
" from store ptr=" << *StoreEv << " at: " <<
*SI << "\n"; } } while (false)
1113 << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " Formed memcpy: " <<
*NewCall << "\n" << " from load ptr=" <<
*LoadEv << " at: " << *LI << "\n" <<
" from store ptr=" << *StoreEv << " at: " <<
*SI << "\n"; } } while (false)
;
1114
1115 ORE.emit([&]() {
1116 return OptimizationRemark(DEBUG_TYPE"loop-idiom", "ProcessLoopStoreOfLoopLoad",
1117 NewCall->getDebugLoc(), Preheader)
1118 << "Formed a call to "
1119 << ore::NV("NewFunction", NewCall->getCalledFunction())
1120 << "() function";
1121 });
1122
1123 // Okay, the memcpy has been formed. Zap the original store and anything that
1124 // feeds into it.
1125 deleteDeadInstruction(SI);
1126 ++NumMemCpy;
1127 return true;
1128}
1129
1130// When compiling for codesize we avoid idiom recognition for a multi-block loop
1131// unless it is a loop_memset idiom or a memset/memcpy idiom in a nested loop.
1132//
1133bool LoopIdiomRecognize::avoidLIRForMultiBlockLoop(bool IsMemset,
1134 bool IsLoopMemset) {
1135 if (ApplyCodeSizeHeuristics && CurLoop->getNumBlocks() > 1) {
1136 if (!CurLoop->getParentLoop() && (!IsMemset || !IsLoopMemset)) {
1137 LLVM_DEBUG(dbgs() << " " << CurLoop->getHeader()->getParent()->getName()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " " << CurLoop->getHeader
()->getParent()->getName() << " : LIR " << (
IsMemset ? "Memset" : "Memcpy") << " avoided: multi-block top-level loop\n"
; } } while (false)
1138 << " : LIR " << (IsMemset ? "Memset" : "Memcpy")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " " << CurLoop->getHeader
()->getParent()->getName() << " : LIR " << (
IsMemset ? "Memset" : "Memcpy") << " avoided: multi-block top-level loop\n"
; } } while (false)
1139 << " avoided: multi-block top-level loop\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << " " << CurLoop->getHeader
()->getParent()->getName() << " : LIR " << (
IsMemset ? "Memset" : "Memcpy") << " avoided: multi-block top-level loop\n"
; } } while (false)
;
1140 return true;
1141 }
1142 }
1143
1144 return false;
1145}
1146
1147bool LoopIdiomRecognize::runOnNoncountableLoop() {
1148 LLVM_DEBUG(dbgs() << DEBUG_TYPE " Scanning: F["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Noncountable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
11
Assuming 'DebugFlag' is false
12
Loop condition is false. Exiting loop
1149 << CurLoop->getHeader()->getParent()->getName()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Noncountable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
1150 << "] Noncountable Loop %"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Noncountable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
1151 << CurLoop->getHeader()->getName() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-idiom")) { dbgs() << "loop-idiom" " Scanning: F["
<< CurLoop->getHeader()->getParent()->getName
() << "] Noncountable Loop %" << CurLoop->getHeader
()->getName() << "\n"; } } while (false)
;
1152
1153 return recognizePopcount() || recognizeAndInsertFFS();
13
Calling 'LoopIdiomRecognize::recognizeAndInsertFFS'
1154}
1155
1156/// Check if the given conditional branch is based on the comparison between
1157/// a variable and zero, and if the variable is non-zero or zero (JmpOnZero is
1158/// true), the control yields to the loop entry. If the branch matches the
1159/// behavior, the variable involved in the comparison is returned. This function
1160/// will be called to see if the precondition and postcondition of the loop are
1161/// in desirable form.
1162static Value *matchCondition(BranchInst *BI, BasicBlock *LoopEntry,
1163 bool JmpOnZero = false) {
1164 if (!BI || !BI->isConditional())
1165 return nullptr;
1166
1167 ICmpInst *Cond = dyn_cast<ICmpInst>(BI->getCondition());
1168 if (!Cond)
1169 return nullptr;
1170
1171 ConstantInt *CmpZero = dyn_cast<ConstantInt>(Cond->getOperand(1));
1172 if (!CmpZero || !CmpZero->isZero())
1173 return nullptr;
1174
1175 BasicBlock *TrueSucc = BI->getSuccessor(0);
1176 BasicBlock *FalseSucc = BI->getSuccessor(1);
1177 if (JmpOnZero)
1178 std::swap(TrueSucc, FalseSucc);
1179
1180 ICmpInst::Predicate Pred = Cond->getPredicate();
1181 if ((Pred == ICmpInst::ICMP_NE && TrueSucc == LoopEntry) ||
1182 (Pred == ICmpInst::ICMP_EQ && FalseSucc == LoopEntry))
1183 return Cond->getOperand(0);
1184
1185 return nullptr;
1186}
1187
1188// Check if the recurrence variable `VarX` is in the right form to create
1189// the idiom. Returns the value coerced to a PHINode if so.
1190static PHINode *getRecurrenceVar(Value *VarX, Instruction *DefX,
1191 BasicBlock *LoopEntry) {
1192 auto *PhiX = dyn_cast<PHINode>(VarX);
1193 if (PhiX && PhiX->getParent() == LoopEntry &&
1194 (PhiX->getOperand(0) == DefX || PhiX->getOperand(1) == DefX))
1195 return PhiX;
1196 return nullptr;
1197}
1198
1199/// Return true iff the idiom is detected in the loop.
1200///
1201/// Additionally:
1202/// 1) \p CntInst is set to the instruction counting the population bit.
1203/// 2) \p CntPhi is set to the corresponding phi node.
1204/// 3) \p Var is set to the value whose population bits are being counted.
1205///
1206/// The core idiom we are trying to detect is:
1207/// \code
1208/// if (x0 != 0)
1209/// goto loop-exit // the precondition of the loop
1210/// cnt0 = init-val;
1211/// do {
1212/// x1 = phi (x0, x2);
1213/// cnt1 = phi(cnt0, cnt2);
1214///
1215/// cnt2 = cnt1 + 1;
1216/// ...
1217/// x2 = x1 & (x1 - 1);
1218/// ...
1219/// } while(x != 0);
1220///
1221/// loop-exit:
1222/// \endcode
1223static bool detectPopcountIdiom(Loop *CurLoop, BasicBlock *PreCondBB,
1224 Instruction *&CntInst, PHINode *&CntPhi,
1225 Value *&Var) {
1226 // step 1: Check to see if the look-back branch match this pattern:
1227 // "if (a!=0) goto loop-entry".
1228 BasicBlock *LoopEntry;
1229 Instruction *DefX2, *CountInst;
1230 Value *VarX1, *VarX0;
1231 PHINode *PhiX, *CountPhi;
1232
1233 DefX2 = CountInst = nullptr;
1234 VarX1 = VarX0 = nullptr;
1235 PhiX = CountPhi = nullptr;
1236 LoopEntry = *(CurLoop->block_begin());
1237
1238 // step 1: Check if the loop-back branch is in desirable form.
1239 {
1240 if (Value *T = matchCondition(
1241 dyn_cast<BranchInst>(LoopEntry->getTerminator()), LoopEntry))
1242 DefX2 = dyn_cast<Instruction>(T);
1243 else
1244 return false;
1245 }
1246
1247 // step 2: detect instructions corresponding to "x2 = x1 & (x1 - 1)"
1248 {
1249 if (!DefX2 || DefX2->getOpcode() != Instruction::And)
1250 return false;
1251
1252 BinaryOperator *SubOneOp;
1253
1254 if ((SubOneOp = dyn_cast<BinaryOperator>(DefX2->getOperand(0))))
1255 VarX1 = DefX2->getOperand(1);
1256 else {
1257 VarX1 = DefX2->getOperand(0);
1258 SubOneOp = dyn_cast<BinaryOperator>(DefX2->getOperand(1));
1259 }
1260 if (!SubOneOp || SubOneOp->getOperand(0) != VarX1)
1261 return false;
1262
1263 ConstantInt *Dec = dyn_cast<ConstantInt>(SubOneOp->getOperand(1));
1264 if (!Dec ||
1265 !((SubOneOp->getOpcode() == Instruction::Sub && Dec->isOne()) ||
1266 (SubOneOp->getOpcode() == Instruction::Add &&
1267 Dec->isMinusOne()))) {
1268 return false;
1269 }
1270 }
1271
1272 // step 3: Check the recurrence of variable X
1273 PhiX = getRecurrenceVar(VarX1, DefX2, LoopEntry);
1274 if (!PhiX)
1275 return false;
1276
1277 // step 4: Find the instruction which count the population: cnt2 = cnt1 + 1
1278 {
1279 CountInst = nullptr;
1280 for (BasicBlock::iterator Iter = LoopEntry->getFirstNonPHI()->getIterator(),
1281 IterE = LoopEntry->end();
1282 Iter != IterE; Iter++) {
1283 Instruction *Inst = &*Iter;
1284 if (Inst->getOpcode() != Instruction::Add)
1285 continue;
1286
1287 ConstantInt *Inc = dyn_cast<ConstantInt>(Inst->getOperand(1));
1288 if (!Inc || !Inc->isOne())
1289 continue;
1290
1291 PHINode *Phi = getRecurrenceVar(Inst->getOperand(0), Inst, LoopEntry);
1292 if (!Phi)
1293 continue;
1294
1295 // Check if the result of the instruction is live of the loop.
1296 bool LiveOutLoop = false;
1297 for (User *U : Inst->users()) {
1298 if ((cast<Instruction>(U))->getParent() != LoopEntry) {
1299 LiveOutLoop = true;
1300 break;
1301 }
1302 }
1303
1304 if (LiveOutLoop) {
1305 CountInst = Inst;
1306 CountPhi = Phi;
1307 break;
1308 }
1309 }
1310
1311 if (!CountInst)
1312 return false;
1313 }
1314
1315 // step 5: check if the precondition is in this form:
1316 // "if (x != 0) goto loop-head ; else goto somewhere-we-don't-care;"
1317 {
1318 auto *PreCondBr = dyn_cast<BranchInst>(PreCondBB->getTerminator());
1319 Value *T = matchCondition(PreCondBr, CurLoop->getLoopPreheader());
1320 if (T != PhiX->getOperand(0) && T != PhiX->getOperand(1))
1321 return false;
1322
1323 CntInst = CountInst;
1324 CntPhi = CountPhi;
1325 Var = T;
1326 }
1327
1328 return true;
1329}
1330
1331/// Return true if the idiom is detected in the loop.
1332///
1333/// Additionally:
1334/// 1) \p CntInst is set to the instruction Counting Leading Zeros (CTLZ)
1335/// or nullptr if there is no such.
1336/// 2) \p CntPhi is set to the corresponding phi node
1337/// or nullptr if there is no such.
1338/// 3) \p Var is set to the value whose CTLZ could be used.
1339/// 4) \p DefX is set to the instruction calculating Loop exit condition.
1340///
1341/// The core idiom we are trying to detect is:
1342/// \code
1343/// if (x0 == 0)
1344/// goto loop-exit // the precondition of the loop
1345/// cnt0 = init-val;
1346/// do {
1347/// x = phi (x0, x.next); //PhiX
1348/// cnt = phi(cnt0, cnt.next);
1349///
1350/// cnt.next = cnt + 1;
1351/// ...
1352/// x.next = x >> 1; // DefX
1353/// ...
1354/// } while(x.next != 0);
1355///
1356/// loop-exit:
1357/// \endcode
1358static bool detectShiftUntilZeroIdiom(Loop *CurLoop, const DataLayout &DL,
1359 Intrinsic::ID &IntrinID, Value *&InitX,
1360 Instruction *&CntInst, PHINode *&CntPhi,
1361 Instruction *&DefX) {
1362 BasicBlock *LoopEntry;
1363 Value *VarX = nullptr;
1364
1365 DefX = nullptr;
1366 CntInst = nullptr;
1367 CntPhi = nullptr;
1368 LoopEntry = *(CurLoop->block_begin());
1369
1370 // step 1: Check if the loop-back branch is in desirable form.
1371 if (Value *T = matchCondition(
19
Assuming 'T' is non-null, which participates in a condition later
20
Taking true branch
1372 dyn_cast<BranchInst>(LoopEntry->getTerminator()), LoopEntry))
18
Assuming the object is not a 'BranchInst'
1373 DefX = dyn_cast<Instruction>(T);
21
Assuming 'T' is a 'Instruction'
1374 else
1375 return false;
1376
1377 // step 2: detect instructions corresponding to "x.next = x >> 1 or x << 1"
1378 if (!DefX
21.1
'DefX' is non-null
21.1
'DefX' is non-null
|| !DefX->isShift())
22
Assuming the condition is false
23
Taking false branch
1379 return false;
1380 IntrinID = DefX->getOpcode() == Instruction::Shl ? Intrinsic::cttz :
24
Assuming the condition is false
25
'?' condition is false
1381 Intrinsic::ctlz;
1382 ConstantInt *Shft = dyn_cast<ConstantInt>(DefX->getOperand(1));
26
Assuming the object is a 'ConstantInt'
1383 if (!Shft
26.1
'Shft' is non-null, which participates in a condition later
26.1
'Shft' is non-null, which participates in a condition later
|| !Shft->isOne())
27
Assuming the condition is false
28
Taking false branch
1384 return false;
1385 VarX = DefX->getOperand(0);
1386
1387 // step 3: Check the recurrence of variable X
1388 PHINode *PhiX = getRecurrenceVar(VarX, DefX, LoopEntry);
1389 if (!PhiX)
29
Assuming 'PhiX' is non-null, which participates in a condition later
30
Taking false branch
1390 return false;
1391
1392 InitX = PhiX->getIncomingValueForBlock(CurLoop->getLoopPreheader());
31
Value assigned to 'InitX'
1393
1394 // Make sure the initial value can't be negative otherwise the ashr in the
1395 // loop might never reach zero which would make the loop infinite.
1396 if (DefX->getOpcode() == Instruction::AShr && !isKnownNonNegative(InitX, DL))
32
Assuming the condition is false
1397 return false;
1398
1399 // step 4: Find the instruction which count the CTLZ: cnt.next = cnt + 1
1400 // TODO: We can skip the step. If loop trip count is known (CTLZ),
1401 // then all uses of "cnt.next" could be optimized to the trip count
1402 // plus "cnt0". Currently it is not optimized.
1403 // This step could be used to detect POPCNT instruction:
1404 // cnt.next = cnt + (x.next & 1)
1405 for (BasicBlock::iterator Iter = LoopEntry->getFirstNonPHI()->getIterator(),
37
Loop condition is true. Entering loop body
1406 IterE = LoopEntry->end();
1407 Iter != IterE; Iter++) {
33
Calling 'operator!='
36
Returning from 'operator!='
1408 Instruction *Inst = &*Iter;
1409 if (Inst->getOpcode() != Instruction::Add)
38
Assuming the condition is false
39
Taking false branch
1410 continue;
1411
1412 ConstantInt *Inc = dyn_cast<ConstantInt>(Inst->getOperand(1));
40
Assuming the object is a 'ConstantInt'
1413 if (!Inc
40.1
'Inc' is non-null, which participates in a condition later
40.1
'Inc' is non-null, which participates in a condition later
|| !Inc->isOne())
41
Assuming the condition is false
42
Taking false branch
1414 continue;
1415
1416 PHINode *Phi = getRecurrenceVar(Inst->getOperand(0), Inst, LoopEntry);
1417 if (!Phi)
43
Assuming 'Phi' is non-null, which participates in a condition later
44
Taking false branch
1418 continue;
1419
1420 CntInst = Inst;
1421 CntPhi = Phi;
1422 break;
45
Execution continues on line 1424
1423 }
1424 if (!CntInst
45.1
'CntInst' is non-null
45.1
'CntInst' is non-null
)
46
Taking false branch
1425 return false;
1426
1427 return true;
47
Returning the value 1, which participates in a condition later
1428}
1429
1430/// Recognize CTLZ or CTTZ idiom in a non-countable loop and convert the loop
1431/// to countable (with CTLZ / CTTZ trip count). If CTLZ / CTTZ inserted as a new
1432/// trip count returns true; otherwise, returns false.
1433bool LoopIdiomRecognize::recognizeAndInsertFFS() {
1434 // Give up if the loop has multiple blocks or multiple backedges.
1435 if (CurLoop->getNumBackEdges() != 1 || CurLoop->getNumBlocks() != 1)
14
Assuming the condition is false
15
Assuming the condition is false
16
Taking false branch
1436 return false;
1437
1438 Intrinsic::ID IntrinID;
1439 Value *InitX;
1440 Instruction *DefX = nullptr;
1441 PHINode *CntPhi = nullptr;
1442 Instruction *CntInst = nullptr;
1443 // Help decide if transformation is profitable. For ShiftUntilZero idiom,
1444 // this is always 6.
1445 size_t IdiomCanonicalSize = 6;
1446
1447 if (!detectShiftUntilZeroIdiom(CurLoop, *DL, IntrinID, InitX,
17
Calling 'detectShiftUntilZeroIdiom'
48
Returning from 'detectShiftUntilZeroIdiom'
49
Taking false branch
1448 CntInst, CntPhi, DefX))
1449 return false;
1450
1451 bool IsCntPhiUsedOutsideLoop = false;
1452 for (User *U : CntPhi->users())
1453 if (!CurLoop->contains(cast<Instruction>(U))) {
1454 IsCntPhiUsedOutsideLoop = true;
1455 break;
1456 }
1457 bool IsCntInstUsedOutsideLoop = false;
1458 for (User *U : CntInst->users())
1459 if (!CurLoop->contains(cast<Instruction>(U))) {
1460 IsCntInstUsedOutsideLoop = true;
1461 break;
1462 }
1463 // If both CntInst and CntPhi are used outside the loop the profitability
1464 // is questionable.
1465 if (IsCntInstUsedOutsideLoop
49.1
'IsCntInstUsedOutsideLoop' is false
49.1
'IsCntInstUsedOutsideLoop' is false
&& IsCntPhiUsedOutsideLoop)
1466 return false;
1467
1468 // For some CPUs result of CTLZ(X) intrinsic is undefined
1469 // when X is 0. If we can not guarantee X != 0, we need to check this
1470 // when expand.
1471 bool ZeroCheck = false;
1472 // It is safe to assume Preheader exist as it was checked in
1473 // parent function RunOnLoop.
1474 BasicBlock *PH = CurLoop->getLoopPreheader();
1475
1476 // If we are using the count instruction outside the loop, make sure we
1477 // have a zero check as a precondition. Without the check the loop would run
1478 // one iteration for before any check of the input value. This means 0 and 1
1479 // would have identical behavior in the original loop and thus
1480 if (!IsCntPhiUsedOutsideLoop
49.2
'IsCntPhiUsedOutsideLoop' is false
49.2
'IsCntPhiUsedOutsideLoop' is false
) {
50
Taking true branch
1481 auto *PreCondBB = PH->getSinglePredecessor();
1482 if (!PreCondBB)
51
Assuming 'PreCondBB' is non-null
52
Taking false branch
1483 return false;
1484 auto *PreCondBI = dyn_cast<BranchInst>(PreCondBB->getTerminator());
53
Assuming the object is a 'BranchInst'
1485 if (!PreCondBI
53.1
'PreCondBI' is non-null
53.1
'PreCondBI' is non-null
)
54
Taking false branch
1486 return false;
1487 if (matchCondition(PreCondBI, PH) != InitX)
55
Assuming pointer value is null
56
Taking false branch
1488 return false;
1489 ZeroCheck = true;
1490 }
1491
1492 // Check if CTLZ / CTTZ intrinsic is profitable. Assume it is always
1493 // profitable if we delete the loop.
1494
1495 // the loop has only 6 instructions:
1496 // %n.addr.0 = phi [ %n, %entry ], [ %shr, %while.cond ]
1497 // %i.0 = phi [ %i0, %entry ], [ %inc, %while.cond ]
1498 // %shr = ashr %n.addr.0, 1
1499 // %tobool = icmp eq %shr, 0
1500 // %inc = add nsw %i.0, 1
1501 // br i1 %tobool
1502
1503 const Value *Args[] =
1504 {InitX, ZeroCheck
56.1
'ZeroCheck' is true
56.1
'ZeroCheck' is true
? ConstantInt::getTrue(InitX->getContext())
57
'?' condition is true
58
Called C++ object pointer is null
1505 : ConstantInt::getFalse(InitX->getContext())};
1506
1507 // @llvm.dbg doesn't count as they have no semantic effect.
1508 auto InstWithoutDebugIt = CurLoop->getHeader()->instructionsWithoutDebug();
1509 uint32_t HeaderSize =
1510 std::distance(InstWithoutDebugIt.begin(), InstWithoutDebugIt.end());
1511
1512 if (HeaderSize != IdiomCanonicalSize &&
1513 TTI->getIntrinsicCost(IntrinID, InitX->getType(), Args) >
1514 TargetTransformInfo::TCC_Basic)
1515 return false;
1516
1517 transformLoopToCountable(IntrinID, PH, CntInst, CntPhi, InitX, DefX,
1518 DefX->getDebugLoc(), ZeroCheck,
1519 IsCntPhiUsedOutsideLoop);
1520 return true;
1521}
1522
1523/// Recognizes a population count idiom in a non-countable loop.
1524///
1525/// If detected, transforms the relevant code to issue the popcount intrinsic
1526/// function call, and returns true; otherwise, returns false.
1527bool LoopIdiomRecognize::recognizePopcount() {
1528 if (TTI->getPopcntSupport(32) != TargetTransformInfo::PSK_FastHardware)
1529 return false;
1530
1531 // Counting population are usually conducted by few arithmetic instructions.
1532 // Such instructions can be easily "absorbed" by vacant slots in a
1533 // non-compact loop. Therefore, recognizing popcount idiom only makes sense
1534 // in a compact loop.
1535
1536 // Give up if the loop has multiple blocks or multiple backedges.
1537 if (CurLoop->getNumBackEdges() != 1 || CurLoop->getNumBlocks() != 1)
1538 return false;
1539
1540 BasicBlock *LoopBody = *(CurLoop->block_begin());
1541 if (LoopBody->size() >= 20) {
1542 // The loop is too big, bail out.
1543 return false;
1544 }
1545
1546 // It should have a preheader containing nothing but an unconditional branch.
1547 BasicBlock *PH = CurLoop->getLoopPreheader();
1548 if (!PH || &PH->front() != PH->getTerminator())
1549 return false;
1550 auto *EntryBI = dyn_cast<BranchInst>(PH->getTerminator());
1551 if (!EntryBI || EntryBI->isConditional())
1552 return false;
1553
1554 // It should have a precondition block where the generated popcount intrinsic
1555 // function can be inserted.
1556 auto *PreCondBB = PH->getSinglePredecessor();
1557 if (!PreCondBB)
1558 return false;
1559 auto *PreCondBI = dyn_cast<BranchInst>(PreCondBB->getTerminator());
1560 if (!PreCondBI || PreCondBI->isUnconditional())
1561 return false;
1562
1563 Instruction *CntInst;
1564 PHINode *CntPhi;
1565 Value *Val;
1566 if (!detectPopcountIdiom(CurLoop, PreCondBB, CntInst, CntPhi, Val))
1567 return false;
1568
1569 transformLoopToPopcount(PreCondBB, CntInst, CntPhi, Val);
1570 return true;
1571}
1572
1573static CallInst *createPopcntIntrinsic(IRBuilder<> &IRBuilder, Value *Val,
1574 const DebugLoc &DL) {
1575 Value *Ops[] = {Val};
1576 Type *Tys[] = {Val->getType()};
1577
1578 Module *M = IRBuilder.GetInsertBlock()->getParent()->getParent();
1579 Function *Func = Intrinsic::getDeclaration(M, Intrinsic::ctpop, Tys);
1580 CallInst *CI = IRBuilder.CreateCall(Func, Ops);
1581 CI->setDebugLoc(DL);
1582
1583 return CI;
1584}
1585
1586static CallInst *createFFSIntrinsic(IRBuilder<> &IRBuilder, Value *Val,
1587 const DebugLoc &DL, bool ZeroCheck,
1588 Intrinsic::ID IID) {
1589 Value *Ops[] = {Val, ZeroCheck ? IRBuilder.getTrue() : IRBuilder.getFalse()};
1590 Type *Tys[] = {Val->getType()};
1591
1592 Module *M = IRBuilder.GetInsertBlock()->getParent()->getParent();
1593 Function *Func = Intrinsic::getDeclaration(M, IID, Tys);
1594 CallInst *CI = IRBuilder.CreateCall(Func, Ops);
1595 CI->setDebugLoc(DL);
1596
1597 return CI;
1598}
1599
1600/// Transform the following loop (Using CTLZ, CTTZ is similar):
1601/// loop:
1602/// CntPhi = PHI [Cnt0, CntInst]
1603/// PhiX = PHI [InitX, DefX]
1604/// CntInst = CntPhi + 1
1605/// DefX = PhiX >> 1
1606/// LOOP_BODY
1607/// Br: loop if (DefX != 0)
1608/// Use(CntPhi) or Use(CntInst)
1609///
1610/// Into:
1611/// If CntPhi used outside the loop:
1612/// CountPrev = BitWidth(InitX) - CTLZ(InitX >> 1)
1613/// Count = CountPrev + 1
1614/// else
1615/// Count = BitWidth(InitX) - CTLZ(InitX)
1616/// loop:
1617/// CntPhi = PHI [Cnt0, CntInst]
1618/// PhiX = PHI [InitX, DefX]
1619/// PhiCount = PHI [Count, Dec]
1620/// CntInst = CntPhi + 1
1621/// DefX = PhiX >> 1
1622/// Dec = PhiCount - 1
1623/// LOOP_BODY
1624/// Br: loop if (Dec != 0)
1625/// Use(CountPrev + Cnt0) // Use(CntPhi)
1626/// or
1627/// Use(Count + Cnt0) // Use(CntInst)
1628///
1629/// If LOOP_BODY is empty the loop will be deleted.
1630/// If CntInst and DefX are not used in LOOP_BODY they will be removed.
1631void LoopIdiomRecognize::transformLoopToCountable(
1632 Intrinsic::ID IntrinID, BasicBlock *Preheader, Instruction *CntInst,
1633 PHINode *CntPhi, Value *InitX, Instruction *DefX, const DebugLoc &DL,
1634 bool ZeroCheck, bool IsCntPhiUsedOutsideLoop) {
1635 BranchInst *PreheaderBr = cast<BranchInst>(Preheader->getTerminator());
1636
1637 // Step 1: Insert the CTLZ/CTTZ instruction at the end of the preheader block
1638 IRBuilder<> Builder(PreheaderBr);
1639 Builder.SetCurrentDebugLocation(DL);
1640 Value *FFS, *Count, *CountPrev, *NewCount, *InitXNext;
1641
1642 // Count = BitWidth - CTLZ(InitX);
1643 // If there are uses of CntPhi create:
1644 // CountPrev = BitWidth - CTLZ(InitX >> 1);
1645 if (IsCntPhiUsedOutsideLoop) {
1646 if (DefX->getOpcode() == Instruction::AShr)
1647 InitXNext =
1648 Builder.CreateAShr(InitX, ConstantInt::get(InitX->getType(), 1));
1649 else if (DefX->getOpcode() == Instruction::LShr)
1650 InitXNext =
1651 Builder.CreateLShr(InitX, ConstantInt::get(InitX->getType(), 1));
1652 else if (DefX->getOpcode() == Instruction::Shl) // cttz
1653 InitXNext =
1654 Builder.CreateShl(InitX, ConstantInt::get(InitX->getType(), 1));
1655 else
1656 llvm_unreachable("Unexpected opcode!")::llvm::llvm_unreachable_internal("Unexpected opcode!", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp"
, 1656)
;
1657 } else
1658 InitXNext = InitX;
1659 FFS = createFFSIntrinsic(Builder, InitXNext, DL, ZeroCheck, IntrinID);
1660 Count = Builder.CreateSub(
1661 ConstantInt::get(FFS->getType(),
1662 FFS->getType()->getIntegerBitWidth()),
1663 FFS);
1664 if (IsCntPhiUsedOutsideLoop) {
1665 CountPrev = Count;
1666 Count = Builder.CreateAdd(
1667 CountPrev,
1668 ConstantInt::get(CountPrev->getType(), 1));
1669 }
1670
1671 NewCount = Builder.CreateZExtOrTrunc(
1672 IsCntPhiUsedOutsideLoop ? CountPrev : Count,
1673 cast<IntegerType>(CntInst->getType()));
1674
1675 // If the counter's initial value is not zero, insert Add Inst.
1676 Value *CntInitVal = CntPhi->getIncomingValueForBlock(Preheader);
1677 ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
1678 if (!InitConst || !InitConst->isZero())
1679 NewCount = Builder.CreateAdd(NewCount, CntInitVal);
1680
1681 // Step 2: Insert new IV and loop condition:
1682 // loop:
1683 // ...
1684 // PhiCount = PHI [Count, Dec]
1685 // ...
1686 // Dec = PhiCount - 1
1687 // ...
1688 // Br: loop if (Dec != 0)
1689 BasicBlock *Body = *(CurLoop->block_begin());
1690 auto *LbBr = cast<BranchInst>(Body->getTerminator());
1691 ICmpInst *LbCond = cast<ICmpInst>(LbBr->getCondition());
1692 Type *Ty = Count->getType();
1693
1694 PHINode *TcPhi = PHINode::Create(Ty, 2, "tcphi", &Body->front());
1695
1696 Builder.SetInsertPoint(LbCond);
1697 Instruction *TcDec = cast<Instruction>(
1698 Builder.CreateSub(TcPhi, ConstantInt::get(Ty, 1),
1699 "tcdec", false, true));
1700
1701 TcPhi->addIncoming(Count, Preheader);
1702 TcPhi->addIncoming(TcDec, Body);
1703
1704 CmpInst::Predicate Pred =
1705 (LbBr->getSuccessor(0) == Body) ? CmpInst::ICMP_NE : CmpInst::ICMP_EQ;
1706 LbCond->setPredicate(Pred);
1707 LbCond->setOperand(0, TcDec);
1708 LbCond->setOperand(1, ConstantInt::get(Ty, 0));
1709
1710 // Step 3: All the references to the original counter outside
1711 // the loop are replaced with the NewCount
1712 if (IsCntPhiUsedOutsideLoop)
1713 CntPhi->replaceUsesOutsideBlock(NewCount, Body);
1714 else
1715 CntInst->replaceUsesOutsideBlock(NewCount, Body);
1716
1717 // step 4: Forget the "non-computable" trip-count SCEV associated with the
1718 // loop. The loop would otherwise not be deleted even if it becomes empty.
1719 SE->forgetLoop(CurLoop);
1720}
1721
1722void LoopIdiomRecognize::transformLoopToPopcount(BasicBlock *PreCondBB,
1723 Instruction *CntInst,
1724 PHINode *CntPhi, Value *Var) {
1725 BasicBlock *PreHead = CurLoop->getLoopPreheader();
1726 auto *PreCondBr = cast<BranchInst>(PreCondBB->getTerminator());
1727 const DebugLoc &DL = CntInst->getDebugLoc();
1728
1729 // Assuming before transformation, the loop is following:
1730 // if (x) // the precondition
1731 // do { cnt++; x &= x - 1; } while(x);
1732
1733 // Step 1: Insert the ctpop instruction at the end of the precondition block
1734 IRBuilder<> Builder(PreCondBr);
1735 Value *PopCnt, *PopCntZext, *NewCount, *TripCnt;
1736 {
1737 PopCnt = createPopcntIntrinsic(Builder, Var, DL);
1738 NewCount = PopCntZext =
1739 Builder.CreateZExtOrTrunc(PopCnt, cast<IntegerType>(CntPhi->getType()));
1740
1741 if (NewCount != PopCnt)
1742 (cast<Instruction>(NewCount))->setDebugLoc(DL);
1743
1744 // TripCnt is exactly the number of iterations the loop has
1745 TripCnt = NewCount;
1746
1747 // If the population counter's initial value is not zero, insert Add Inst.
1748 Value *CntInitVal = CntPhi->getIncomingValueForBlock(PreHead);
1749 ConstantInt *InitConst = dyn_cast<ConstantInt>(CntInitVal);
1750 if (!InitConst || !InitConst->isZero()) {
1751 NewCount = Builder.CreateAdd(NewCount, CntInitVal);
1752 (cast<Instruction>(NewCount))->setDebugLoc(DL);
1753 }
1754 }
1755
1756 // Step 2: Replace the precondition from "if (x == 0) goto loop-exit" to
1757 // "if (NewCount == 0) loop-exit". Without this change, the intrinsic
1758 // function would be partial dead code, and downstream passes will drag
1759 // it back from the precondition block to the preheader.
1760 {
1761 ICmpInst *PreCond = cast<ICmpInst>(PreCondBr->getCondition());
1762
1763 Value *Opnd0 = PopCntZext;
1764 Value *Opnd1 = ConstantInt::get(PopCntZext->getType(), 0);
1765 if (PreCond->getOperand(0) != Var)
1766 std::swap(Opnd0, Opnd1);
1767
1768 ICmpInst *NewPreCond = cast<ICmpInst>(
1769 Builder.CreateICmp(PreCond->getPredicate(), Opnd0, Opnd1));
1770 PreCondBr->setCondition(NewPreCond);
1771
1772 RecursivelyDeleteTriviallyDeadInstructions(PreCond, TLI);
1773 }
1774
1775 // Step 3: Note that the population count is exactly the trip count of the
1776 // loop in question, which enable us to convert the loop from noncountable
1777 // loop into a countable one. The benefit is twofold:
1778 //
1779 // - If the loop only counts population, the entire loop becomes dead after
1780 // the transformation. It is a lot easier to prove a countable loop dead
1781 // than to prove a noncountable one. (In some C dialects, an infinite loop
1782 // isn't dead even if it computes nothing useful. In general, DCE needs
1783 // to prove a noncountable loop finite before safely delete it.)
1784 //
1785 // - If the loop also performs something else, it remains alive.
1786 // Since it is transformed to countable form, it can be aggressively
1787 // optimized by some optimizations which are in general not applicable
1788 // to a noncountable loop.
1789 //
1790 // After this step, this loop (conceptually) would look like following:
1791 // newcnt = __builtin_ctpop(x);
1792 // t = newcnt;
1793 // if (x)
1794 // do { cnt++; x &= x-1; t--) } while (t > 0);
1795 BasicBlock *Body = *(CurLoop->block_begin());
1796 {
1797 auto *LbBr = cast<BranchInst>(Body->getTerminator());
1798 ICmpInst *LbCond = cast<ICmpInst>(LbBr->getCondition());
1799 Type *Ty = TripCnt->getType();
1800
1801 PHINode *TcPhi = PHINode::Create(Ty, 2, "tcphi", &Body->front());
1802
1803 Builder.SetInsertPoint(LbCond);
1804 Instruction *TcDec = cast<Instruction>(
1805 Builder.CreateSub(TcPhi, ConstantInt::get(Ty, 1),
1806 "tcdec", false, true));
1807
1808 TcPhi->addIncoming(TripCnt, PreHead);
1809 TcPhi->addIncoming(TcDec, Body);
1810
1811 CmpInst::Predicate Pred =
1812 (LbBr->getSuccessor(0) == Body) ? CmpInst::ICMP_UGT : CmpInst::ICMP_SLE;
1813 LbCond->setPredicate(Pred);
1814 LbCond->setOperand(0, TcDec);
1815 LbCond->setOperand(1, ConstantInt::get(Ty, 0));
1816 }
1817
1818 // Step 4: All the references to the original population counter outside
1819 // the loop are replaced with the NewCount -- the value returned from
1820 // __builtin_ctpop().
1821 CntInst->replaceUsesOutsideBlock(NewCount, Body);
1822
1823 // step 5: Forget the "non-computable" trip-count SCEV associated with the
1824 // loop. The loop would otherwise not be deleted even if it becomes empty.
1825 SE->forgetLoop(CurLoop);
1826}

/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/include/llvm/ADT/ilist_iterator.h

1//===- llvm/ADT/ilist_iterator.h - Intrusive List Iterator ------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8
9#ifndef LLVM_ADT_ILIST_ITERATOR_H
10#define LLVM_ADT_ILIST_ITERATOR_H
11
12#include "llvm/ADT/ilist_node.h"
13#include <cassert>
14#include <cstddef>
15#include <iterator>
16#include <type_traits>
17
18namespace llvm {
19
20namespace ilist_detail {
21
22/// Find const-correct node types.
23template <class OptionsT, bool IsConst> struct IteratorTraits;
24template <class OptionsT> struct IteratorTraits<OptionsT, false> {
25 using value_type = typename OptionsT::value_type;
26 using pointer = typename OptionsT::pointer;
27 using reference = typename OptionsT::reference;
28 using node_pointer = ilist_node_impl<OptionsT> *;
29 using node_reference = ilist_node_impl<OptionsT> &;
30};
31template <class OptionsT> struct IteratorTraits<OptionsT, true> {
32 using value_type = const typename OptionsT::value_type;
33 using pointer = typename OptionsT::const_pointer;
34 using reference = typename OptionsT::const_reference;
35 using node_pointer = const ilist_node_impl<OptionsT> *;
36 using node_reference = const ilist_node_impl<OptionsT> &;
37};
38
39template <bool IsReverse> struct IteratorHelper;
40template <> struct IteratorHelper<false> : ilist_detail::NodeAccess {
41 using Access = ilist_detail::NodeAccess;
42
43 template <class T> static void increment(T *&I) { I = Access::getNext(*I); }
44 template <class T> static void decrement(T *&I) { I = Access::getPrev(*I); }
45};
46template <> struct IteratorHelper<true> : ilist_detail::NodeAccess {
47 using Access = ilist_detail::NodeAccess;
48
49 template <class T> static void increment(T *&I) { I = Access::getPrev(*I); }
50 template <class T> static void decrement(T *&I) { I = Access::getNext(*I); }
51};
52
53} // end namespace ilist_detail
54
55/// Iterator for intrusive lists based on ilist_node.
56template <class OptionsT, bool IsReverse, bool IsConst>
57class ilist_iterator : ilist_detail::SpecificNodeAccess<OptionsT> {
58 friend ilist_iterator<OptionsT, IsReverse, !IsConst>;
59 friend ilist_iterator<OptionsT, !IsReverse, IsConst>;
60 friend ilist_iterator<OptionsT, !IsReverse, !IsConst>;
61
62 using Traits = ilist_detail::IteratorTraits<OptionsT, IsConst>;
63 using Access = ilist_detail::SpecificNodeAccess<OptionsT>;
64
65public:
66 using value_type = typename Traits::value_type;
67 using pointer = typename Traits::pointer;
68 using reference = typename Traits::reference;
69 using difference_type = ptrdiff_t;
70 using iterator_category = std::bidirectional_iterator_tag;
71 using const_pointer = typename OptionsT::const_pointer;
72 using const_reference = typename OptionsT::const_reference;
73
74private:
75 using node_pointer = typename Traits::node_pointer;
76 using node_reference = typename Traits::node_reference;
77
78 node_pointer NodePtr = nullptr;
79
80public:
81 /// Create from an ilist_node.
82 explicit ilist_iterator(node_reference N) : NodePtr(&N) {}
83
84 explicit ilist_iterator(pointer NP) : NodePtr(Access::getNodePtr(NP)) {}
85 explicit ilist_iterator(reference NR) : NodePtr(Access::getNodePtr(&NR)) {}
86 ilist_iterator() = default;
87
88 // This is templated so that we can allow constructing a const iterator from
89 // a nonconst iterator...
90 template <bool RHSIsConst>
91 ilist_iterator(
92 const ilist_iterator<OptionsT, IsReverse, RHSIsConst> &RHS,
93 typename std::enable_if<IsConst || !RHSIsConst, void *>::type = nullptr)
94 : NodePtr(RHS.NodePtr) {}
95
96 // This is templated so that we can allow assigning to a const iterator from
97 // a nonconst iterator...
98 template <bool RHSIsConst>
99 typename std::enable_if<IsConst || !RHSIsConst, ilist_iterator &>::type
100 operator=(const ilist_iterator<OptionsT, IsReverse, RHSIsConst> &RHS) {
101 NodePtr = RHS.NodePtr;
102 return *this;
103 }
104
105 /// Explicit conversion between forward/reverse iterators.
106 ///
107 /// Translate between forward and reverse iterators without changing range
108 /// boundaries. The resulting iterator will dereference (and have a handle)
109 /// to the previous node, which is somewhat unexpected; but converting the
110 /// two endpoints in a range will give the same range in reverse.
111 ///
112 /// This matches std::reverse_iterator conversions.
113 explicit ilist_iterator(
114 const ilist_iterator<OptionsT, !IsReverse, IsConst> &RHS)
115 : ilist_iterator(++RHS.getReverse()) {}
116
117 /// Get a reverse iterator to the same node.
118 ///
119 /// Gives a reverse iterator that will dereference (and have a handle) to the
120 /// same node. Converting the endpoint iterators in a range will give a
121 /// different range; for range operations, use the explicit conversions.
122 ilist_iterator<OptionsT, !IsReverse, IsConst> getReverse() const {
123 if (NodePtr)
124 return ilist_iterator<OptionsT, !IsReverse, IsConst>(*NodePtr);
125 return ilist_iterator<OptionsT, !IsReverse, IsConst>();
126 }
127
128 /// Const-cast.
129 ilist_iterator<OptionsT, IsReverse, false> getNonConst() const {
130 if (NodePtr)
131 return ilist_iterator<OptionsT, IsReverse, false>(
132 const_cast<typename ilist_iterator<OptionsT, IsReverse,
133 false>::node_reference>(*NodePtr));
134 return ilist_iterator<OptionsT, IsReverse, false>();
135 }
136
137 // Accessors...
138 reference operator*() const {
139 assert(!NodePtr->isKnownSentinel())((!NodePtr->isKnownSentinel()) ? static_cast<void> (
0) : __assert_fail ("!NodePtr->isKnownSentinel()", "/build/llvm-toolchain-snapshot-10~+201911111502510600c19528f1809/llvm/include/llvm/ADT/ilist_iterator.h"
, 139, __PRETTY_FUNCTION__))
;
140 return *Access::getValuePtr(NodePtr);
141 }
142 pointer operator->() const { return &operator*(); }
143
144 // Comparison operators
145 friend bool operator==(const ilist_iterator &LHS, const ilist_iterator &RHS) {
146 return LHS.NodePtr == RHS.NodePtr;
147 }
148 friend bool operator!=(const ilist_iterator &LHS, const ilist_iterator &RHS) {
149 return LHS.NodePtr != RHS.NodePtr;
34
Assuming 'LHS.NodePtr' is not equal to 'RHS.NodePtr'
35
Returning the value 1, which participates in a condition later
150 }
151
152 // Increment and decrement operators...
153 ilist_iterator &operator--() {
154 NodePtr = IsReverse ? NodePtr->getNext() : NodePtr->getPrev();
155 return *this;
156 }
157 ilist_iterator &operator++() {
158 NodePtr = IsReverse ? NodePtr->getPrev() : NodePtr->getNext();
159 return *this;
160 }
161 ilist_iterator operator--(int) {
162 ilist_iterator tmp = *this;
163 --*this;
164 return tmp;
165 }
166 ilist_iterator operator++(int) {
167 ilist_iterator tmp = *this;
168 ++*this;
169 return tmp;
170 }
171
172 /// Get the underlying ilist_node.
173 node_pointer getNodePtr() const { return static_cast<node_pointer>(NodePtr); }
174
175 /// Check for end. Only valid if ilist_sentinel_tracking<true>.
176 bool isEnd() const { return NodePtr ? NodePtr->isSentinel() : false; }
177};
178
179template <typename From> struct simplify_type;
180
181/// Allow ilist_iterators to convert into pointers to a node automatically when
182/// used by the dyn_cast, cast, isa mechanisms...
183///
184/// FIXME: remove this, since there is no implicit conversion to NodeTy.
185template <class OptionsT, bool IsConst>
186struct simplify_type<ilist_iterator<OptionsT, false, IsConst>> {
187 using iterator = ilist_iterator<OptionsT, false, IsConst>;
188 using SimpleType = typename iterator::pointer;
189
190 static SimpleType getSimplifiedValue(const iterator &Node) { return &*Node; }
191};
192template <class OptionsT, bool IsConst>
193struct simplify_type<const ilist_iterator<OptionsT, false, IsConst>>
194 : simplify_type<ilist_iterator<OptionsT, false, IsConst>> {};
195
196} // end namespace llvm
197
198#endif // LLVM_ADT_ILIST_ITERATOR_H