1 //===- LoopUnrollAndJam.cpp - Loop unroll and jam pass --------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This pass implements an unroll and jam pass. Most of the work is done by
11 // Utils/UnrollLoopAndJam.cpp.
12 //===----------------------------------------------------------------------===//
13
14 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h"
15 #include "llvm/ADT/None.h"
16 #include "llvm/ADT/STLExtras.h"
17 #include "llvm/ADT/SmallPtrSet.h"
18 #include "llvm/ADT/StringRef.h"
19 #include "llvm/Analysis/AssumptionCache.h"
20 #include "llvm/Analysis/CodeMetrics.h"
21 #include "llvm/Analysis/DependenceAnalysis.h"
22 #include "llvm/Analysis/LoopAnalysisManager.h"
23 #include "llvm/Analysis/LoopInfo.h"
24 #include "llvm/Analysis/LoopPass.h"
25 #include "llvm/Analysis/OptimizationRemarkEmitter.h"
26 #include "llvm/Analysis/ScalarEvolution.h"
27 #include "llvm/Analysis/TargetTransformInfo.h"
28 #include "llvm/IR/BasicBlock.h"
29 #include "llvm/IR/CFG.h"
30 #include "llvm/IR/Constant.h"
31 #include "llvm/IR/Constants.h"
32 #include "llvm/IR/Dominators.h"
33 #include "llvm/IR/Function.h"
34 #include "llvm/IR/Instruction.h"
35 #include "llvm/IR/Instructions.h"
36 #include "llvm/IR/IntrinsicInst.h"
37 #include "llvm/IR/Metadata.h"
38 #include "llvm/IR/PassManager.h"
39 #include "llvm/Pass.h"
40 #include "llvm/Support/Casting.h"
41 #include "llvm/Support/CommandLine.h"
42 #include "llvm/Support/Debug.h"
43 #include "llvm/Support/ErrorHandling.h"
44 #include "llvm/Support/raw_ostream.h"
45 #include "llvm/Transforms/Scalar.h"
46 #include "llvm/Transforms/Scalar/LoopPassManager.h"
47 #include "llvm/Transforms/Utils.h"
48 #include "llvm/Transforms/Utils/LoopUtils.h"
49 #include "llvm/Transforms/Utils/UnrollLoop.h"
50 #include <algorithm>
51 #include <cassert>
52 #include <cstdint>
53 #include <string>
54
55 using namespace llvm;
56
57 #define DEBUG_TYPE "loop-unroll-and-jam"
58
59 /// @{
60 /// Metadata attribute names
61 static const char *const LLVMLoopUnrollAndJamFollowupAll =
62 "llvm.loop.unroll_and_jam.followup_all";
63 static const char *const LLVMLoopUnrollAndJamFollowupInner =
64 "llvm.loop.unroll_and_jam.followup_inner";
65 static const char *const LLVMLoopUnrollAndJamFollowupOuter =
66 "llvm.loop.unroll_and_jam.followup_outer";
67 static const char *const LLVMLoopUnrollAndJamFollowupRemainderInner =
68 "llvm.loop.unroll_and_jam.followup_remainder_inner";
69 static const char *const LLVMLoopUnrollAndJamFollowupRemainderOuter =
70 "llvm.loop.unroll_and_jam.followup_remainder_outer";
71 /// @}
72
73 static cl::opt<bool>
74 AllowUnrollAndJam("allow-unroll-and-jam", cl::Hidden,
75 cl::desc("Allows loops to be unroll-and-jammed."));
76
77 static cl::opt<unsigned> UnrollAndJamCount(
78 "unroll-and-jam-count", cl::Hidden,
79 cl::desc("Use this unroll count for all loops including those with "
80 "unroll_and_jam_count pragma values, for testing purposes"));
81
82 static cl::opt<unsigned> UnrollAndJamThreshold(
83 "unroll-and-jam-threshold", cl::init(60), cl::Hidden,
84 cl::desc("Threshold to use for inner loop when doing unroll and jam."));
85
86 static cl::opt<unsigned> PragmaUnrollAndJamThreshold(
87 "pragma-unroll-and-jam-threshold", cl::init(1024), cl::Hidden,
88 cl::desc("Unrolled size limit for loops with an unroll_and_jam(full) or "
89 "unroll_count pragma."));
90
91 // Returns the loop hint metadata node with the given name (for example,
92 // "llvm.loop.unroll.count"). If no such metadata node exists, then nullptr is
93 // returned.
GetUnrollMetadataForLoop(const Loop * L,StringRef Name)94 static MDNode *GetUnrollMetadataForLoop(const Loop *L, StringRef Name) {
95 if (MDNode *LoopID = L->getLoopID())
96 return GetUnrollMetadata(LoopID, Name);
97 return nullptr;
98 }
99
100 // Returns true if the loop has any metadata starting with Prefix. For example a
101 // Prefix of "llvm.loop.unroll." returns true if we have any unroll metadata.
HasAnyUnrollPragma(const Loop * L,StringRef Prefix)102 static bool HasAnyUnrollPragma(const Loop *L, StringRef Prefix) {
103 if (MDNode *LoopID = L->getLoopID()) {
104 // First operand should refer to the loop id itself.
105 assert(LoopID->getNumOperands() > 0 && "requires at least one operand");
106 assert(LoopID->getOperand(0) == LoopID && "invalid loop id");
107
108 for (unsigned i = 1, e = LoopID->getNumOperands(); i < e; ++i) {
109 MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i));
110 if (!MD)
111 continue;
112
113 MDString *S = dyn_cast<MDString>(MD->getOperand(0));
114 if (!S)
115 continue;
116
117 if (S->getString().startswith(Prefix))
118 return true;
119 }
120 }
121 return false;
122 }
123
124 // Returns true if the loop has an unroll_and_jam(enable) pragma.
HasUnrollAndJamEnablePragma(const Loop * L)125 static bool HasUnrollAndJamEnablePragma(const Loop *L) {
126 return GetUnrollMetadataForLoop(L, "llvm.loop.unroll_and_jam.enable");
127 }
128
129 // If loop has an unroll_and_jam_count pragma return the (necessarily
130 // positive) value from the pragma. Otherwise return 0.
UnrollAndJamCountPragmaValue(const Loop * L)131 static unsigned UnrollAndJamCountPragmaValue(const Loop *L) {
132 MDNode *MD = GetUnrollMetadataForLoop(L, "llvm.loop.unroll_and_jam.count");
133 if (MD) {
134 assert(MD->getNumOperands() == 2 &&
135 "Unroll count hint metadata should have two operands.");
136 unsigned Count =
137 mdconst::extract<ConstantInt>(MD->getOperand(1))->getZExtValue();
138 assert(Count >= 1 && "Unroll count must be positive.");
139 return Count;
140 }
141 return 0;
142 }
143
144 // Returns loop size estimation for unrolled loop.
145 static uint64_t
getUnrollAndJammedLoopSize(unsigned LoopSize,TargetTransformInfo::UnrollingPreferences & UP)146 getUnrollAndJammedLoopSize(unsigned LoopSize,
147 TargetTransformInfo::UnrollingPreferences &UP) {
148 assert(LoopSize >= UP.BEInsns && "LoopSize should not be less than BEInsns!");
149 return static_cast<uint64_t>(LoopSize - UP.BEInsns) * UP.Count + UP.BEInsns;
150 }
151
152 // Calculates unroll and jam count and writes it to UP.Count. Returns true if
153 // unroll count was set explicitly.
computeUnrollAndJamCount(Loop * L,Loop * SubLoop,const TargetTransformInfo & TTI,DominatorTree & DT,LoopInfo * LI,ScalarEvolution & SE,const SmallPtrSetImpl<const Value * > & EphValues,OptimizationRemarkEmitter * ORE,unsigned OuterTripCount,unsigned OuterTripMultiple,unsigned OuterLoopSize,unsigned InnerTripCount,unsigned InnerLoopSize,TargetTransformInfo::UnrollingPreferences & UP)154 static bool computeUnrollAndJamCount(
155 Loop *L, Loop *SubLoop, const TargetTransformInfo &TTI, DominatorTree &DT,
156 LoopInfo *LI, ScalarEvolution &SE,
157 const SmallPtrSetImpl<const Value *> &EphValues,
158 OptimizationRemarkEmitter *ORE, unsigned OuterTripCount,
159 unsigned OuterTripMultiple, unsigned OuterLoopSize, unsigned InnerTripCount,
160 unsigned InnerLoopSize, TargetTransformInfo::UnrollingPreferences &UP) {
161 // First up use computeUnrollCount from the loop unroller to get a count
162 // for unrolling the outer loop, plus any loops requiring explicit
163 // unrolling we leave to the unroller. This uses UP.Threshold /
164 // UP.PartialThreshold / UP.MaxCount to come up with sensible loop values.
165 // We have already checked that the loop has no unroll.* pragmas.
166 unsigned MaxTripCount = 0;
167 bool UseUpperBound = false;
168 bool ExplicitUnroll = computeUnrollCount(
169 L, TTI, DT, LI, SE, EphValues, ORE, OuterTripCount, MaxTripCount,
170 OuterTripMultiple, OuterLoopSize, UP, UseUpperBound);
171 if (ExplicitUnroll || UseUpperBound) {
172 // If the user explicitly set the loop as unrolled, dont UnJ it. Leave it
173 // for the unroller instead.
174 LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; explicit count set by "
175 "computeUnrollCount\n");
176 UP.Count = 0;
177 return false;
178 }
179
180 // Override with any explicit Count from the "unroll-and-jam-count" option.
181 bool UserUnrollCount = UnrollAndJamCount.getNumOccurrences() > 0;
182 if (UserUnrollCount) {
183 UP.Count = UnrollAndJamCount;
184 UP.Force = true;
185 if (UP.AllowRemainder &&
186 getUnrollAndJammedLoopSize(OuterLoopSize, UP) < UP.Threshold &&
187 getUnrollAndJammedLoopSize(InnerLoopSize, UP) <
188 UP.UnrollAndJamInnerLoopThreshold)
189 return true;
190 }
191
192 // Check for unroll_and_jam pragmas
193 unsigned PragmaCount = UnrollAndJamCountPragmaValue(L);
194 if (PragmaCount > 0) {
195 UP.Count = PragmaCount;
196 UP.Runtime = true;
197 UP.Force = true;
198 if ((UP.AllowRemainder || (OuterTripMultiple % PragmaCount == 0)) &&
199 getUnrollAndJammedLoopSize(OuterLoopSize, UP) < UP.Threshold &&
200 getUnrollAndJammedLoopSize(InnerLoopSize, UP) <
201 UP.UnrollAndJamInnerLoopThreshold)
202 return true;
203 }
204
205 bool PragmaEnableUnroll = HasUnrollAndJamEnablePragma(L);
206 bool ExplicitUnrollAndJamCount = PragmaCount > 0 || UserUnrollCount;
207 bool ExplicitUnrollAndJam = PragmaEnableUnroll || ExplicitUnrollAndJamCount;
208
209 // If the loop has an unrolling pragma, we want to be more aggressive with
210 // unrolling limits.
211 if (ExplicitUnrollAndJam)
212 UP.UnrollAndJamInnerLoopThreshold = PragmaUnrollAndJamThreshold;
213
214 if (!UP.AllowRemainder && getUnrollAndJammedLoopSize(InnerLoopSize, UP) >=
215 UP.UnrollAndJamInnerLoopThreshold) {
216 LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; can't create remainder and "
217 "inner loop too large\n");
218 UP.Count = 0;
219 return false;
220 }
221
222 // We have a sensible limit for the outer loop, now adjust it for the inner
223 // loop and UP.UnrollAndJamInnerLoopThreshold. If the outer limit was set
224 // explicitly, we want to stick to it.
225 if (!ExplicitUnrollAndJamCount && UP.AllowRemainder) {
226 while (UP.Count != 0 && getUnrollAndJammedLoopSize(InnerLoopSize, UP) >=
227 UP.UnrollAndJamInnerLoopThreshold)
228 UP.Count--;
229 }
230
231 // If we are explicitly unroll and jamming, we are done. Otherwise there are a
232 // number of extra performance heuristics to check.
233 if (ExplicitUnrollAndJam)
234 return true;
235
236 // If the inner loop count is known and small, leave the entire loop nest to
237 // be the unroller
238 if (InnerTripCount && InnerLoopSize * InnerTripCount < UP.Threshold) {
239 LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; small inner loop count is "
240 "being left for the unroller\n");
241 UP.Count = 0;
242 return false;
243 }
244
245 // Check for situations where UnJ is likely to be unprofitable. Including
246 // subloops with more than 1 block.
247 if (SubLoop->getBlocks().size() != 1) {
248 LLVM_DEBUG(
249 dbgs() << "Won't unroll-and-jam; More than one inner loop block\n");
250 UP.Count = 0;
251 return false;
252 }
253
254 // Limit to loops where there is something to gain from unrolling and
255 // jamming the loop. In this case, look for loads that are invariant in the
256 // outer loop and can become shared.
257 unsigned NumInvariant = 0;
258 for (BasicBlock *BB : SubLoop->getBlocks()) {
259 for (Instruction &I : *BB) {
260 if (auto *Ld = dyn_cast<LoadInst>(&I)) {
261 Value *V = Ld->getPointerOperand();
262 const SCEV *LSCEV = SE.getSCEVAtScope(V, L);
263 if (SE.isLoopInvariant(LSCEV, L))
264 NumInvariant++;
265 }
266 }
267 }
268 if (NumInvariant == 0) {
269 LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; No loop invariant loads\n");
270 UP.Count = 0;
271 return false;
272 }
273
274 return false;
275 }
276
277 static LoopUnrollResult
tryToUnrollAndJamLoop(Loop * L,DominatorTree & DT,LoopInfo * LI,ScalarEvolution & SE,const TargetTransformInfo & TTI,AssumptionCache & AC,DependenceInfo & DI,OptimizationRemarkEmitter & ORE,int OptLevel)278 tryToUnrollAndJamLoop(Loop *L, DominatorTree &DT, LoopInfo *LI,
279 ScalarEvolution &SE, const TargetTransformInfo &TTI,
280 AssumptionCache &AC, DependenceInfo &DI,
281 OptimizationRemarkEmitter &ORE, int OptLevel) {
282 // Quick checks of the correct loop form
283 if (!L->isLoopSimplifyForm() || L->getSubLoops().size() != 1)
284 return LoopUnrollResult::Unmodified;
285 Loop *SubLoop = L->getSubLoops()[0];
286 if (!SubLoop->isLoopSimplifyForm())
287 return LoopUnrollResult::Unmodified;
288
289 BasicBlock *Latch = L->getLoopLatch();
290 BasicBlock *Exit = L->getExitingBlock();
291 BasicBlock *SubLoopLatch = SubLoop->getLoopLatch();
292 BasicBlock *SubLoopExit = SubLoop->getExitingBlock();
293
294 if (Latch != Exit || SubLoopLatch != SubLoopExit)
295 return LoopUnrollResult::Unmodified;
296
297 TargetTransformInfo::UnrollingPreferences UP = gatherUnrollingPreferences(
298 L, SE, TTI, OptLevel, None, None, None, None, None, None);
299 if (AllowUnrollAndJam.getNumOccurrences() > 0)
300 UP.UnrollAndJam = AllowUnrollAndJam;
301 if (UnrollAndJamThreshold.getNumOccurrences() > 0)
302 UP.UnrollAndJamInnerLoopThreshold = UnrollAndJamThreshold;
303 // Exit early if unrolling is disabled.
304 if (!UP.UnrollAndJam || UP.UnrollAndJamInnerLoopThreshold == 0)
305 return LoopUnrollResult::Unmodified;
306
307 LLVM_DEBUG(dbgs() << "Loop Unroll and Jam: F["
308 << L->getHeader()->getParent()->getName() << "] Loop %"
309 << L->getHeader()->getName() << "\n");
310
311 TransformationMode EnableMode = hasUnrollAndJamTransformation(L);
312 if (EnableMode & TM_Disable)
313 return LoopUnrollResult::Unmodified;
314
315 // A loop with any unroll pragma (enabling/disabling/count/etc) is left for
316 // the unroller, so long as it does not explicitly have unroll_and_jam
317 // metadata. This means #pragma nounroll will disable unroll and jam as well
318 // as unrolling
319 if (HasAnyUnrollPragma(L, "llvm.loop.unroll.") &&
320 !HasAnyUnrollPragma(L, "llvm.loop.unroll_and_jam.")) {
321 LLVM_DEBUG(dbgs() << " Disabled due to pragma.\n");
322 return LoopUnrollResult::Unmodified;
323 }
324
325 if (!isSafeToUnrollAndJam(L, SE, DT, DI)) {
326 LLVM_DEBUG(dbgs() << " Disabled due to not being safe.\n");
327 return LoopUnrollResult::Unmodified;
328 }
329
330 // Approximate the loop size and collect useful info
331 unsigned NumInlineCandidates;
332 bool NotDuplicatable;
333 bool Convergent;
334 SmallPtrSet<const Value *, 32> EphValues;
335 CodeMetrics::collectEphemeralValues(L, &AC, EphValues);
336 unsigned InnerLoopSize =
337 ApproximateLoopSize(SubLoop, NumInlineCandidates, NotDuplicatable,
338 Convergent, TTI, EphValues, UP.BEInsns);
339 unsigned OuterLoopSize =
340 ApproximateLoopSize(L, NumInlineCandidates, NotDuplicatable, Convergent,
341 TTI, EphValues, UP.BEInsns);
342 LLVM_DEBUG(dbgs() << " Outer Loop Size: " << OuterLoopSize << "\n");
343 LLVM_DEBUG(dbgs() << " Inner Loop Size: " << InnerLoopSize << "\n");
344 if (NotDuplicatable) {
345 LLVM_DEBUG(dbgs() << " Not unrolling loop which contains non-duplicatable "
346 "instructions.\n");
347 return LoopUnrollResult::Unmodified;
348 }
349 if (NumInlineCandidates != 0) {
350 LLVM_DEBUG(dbgs() << " Not unrolling loop with inlinable calls.\n");
351 return LoopUnrollResult::Unmodified;
352 }
353 if (Convergent) {
354 LLVM_DEBUG(
355 dbgs() << " Not unrolling loop with convergent instructions.\n");
356 return LoopUnrollResult::Unmodified;
357 }
358
359 // Save original loop IDs for after the transformation.
360 MDNode *OrigOuterLoopID = L->getLoopID();
361 MDNode *OrigSubLoopID = SubLoop->getLoopID();
362
363 // To assign the loop id of the epilogue, assign it before unrolling it so it
364 // is applied to every inner loop of the epilogue. We later apply the loop ID
365 // for the jammed inner loop.
366 Optional<MDNode *> NewInnerEpilogueLoopID = makeFollowupLoopID(
367 OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,
368 LLVMLoopUnrollAndJamFollowupRemainderInner});
369 if (NewInnerEpilogueLoopID.hasValue())
370 SubLoop->setLoopID(NewInnerEpilogueLoopID.getValue());
371
372 // Find trip count and trip multiple
373 unsigned OuterTripCount = SE.getSmallConstantTripCount(L, Latch);
374 unsigned OuterTripMultiple = SE.getSmallConstantTripMultiple(L, Latch);
375 unsigned InnerTripCount = SE.getSmallConstantTripCount(SubLoop, SubLoopLatch);
376
377 // Decide if, and by how much, to unroll
378 bool IsCountSetExplicitly = computeUnrollAndJamCount(
379 L, SubLoop, TTI, DT, LI, SE, EphValues, &ORE, OuterTripCount,
380 OuterTripMultiple, OuterLoopSize, InnerTripCount, InnerLoopSize, UP);
381 if (UP.Count <= 1)
382 return LoopUnrollResult::Unmodified;
383 // Unroll factor (Count) must be less or equal to TripCount.
384 if (OuterTripCount && UP.Count > OuterTripCount)
385 UP.Count = OuterTripCount;
386
387 Loop *EpilogueOuterLoop = nullptr;
388 LoopUnrollResult UnrollResult = UnrollAndJamLoop(
389 L, UP.Count, OuterTripCount, OuterTripMultiple, UP.UnrollRemainder, LI,
390 &SE, &DT, &AC, &ORE, &EpilogueOuterLoop);
391
392 // Assign new loop attributes.
393 if (EpilogueOuterLoop) {
394 Optional<MDNode *> NewOuterEpilogueLoopID = makeFollowupLoopID(
395 OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,
396 LLVMLoopUnrollAndJamFollowupRemainderOuter});
397 if (NewOuterEpilogueLoopID.hasValue())
398 EpilogueOuterLoop->setLoopID(NewOuterEpilogueLoopID.getValue());
399 }
400
401 Optional<MDNode *> NewInnerLoopID =
402 makeFollowupLoopID(OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,
403 LLVMLoopUnrollAndJamFollowupInner});
404 if (NewInnerLoopID.hasValue())
405 SubLoop->setLoopID(NewInnerLoopID.getValue());
406 else
407 SubLoop->setLoopID(OrigSubLoopID);
408
409 if (UnrollResult == LoopUnrollResult::PartiallyUnrolled) {
410 Optional<MDNode *> NewOuterLoopID = makeFollowupLoopID(
411 OrigOuterLoopID,
412 {LLVMLoopUnrollAndJamFollowupAll, LLVMLoopUnrollAndJamFollowupOuter});
413 if (NewOuterLoopID.hasValue()) {
414 L->setLoopID(NewOuterLoopID.getValue());
415
416 // Do not setLoopAlreadyUnrolled if a followup was given.
417 return UnrollResult;
418 }
419 }
420
421 // If loop has an unroll count pragma or unrolled by explicitly set count
422 // mark loop as unrolled to prevent unrolling beyond that requested.
423 if (UnrollResult != LoopUnrollResult::FullyUnrolled && IsCountSetExplicitly)
424 L->setLoopAlreadyUnrolled();
425
426 return UnrollResult;
427 }
428
429 namespace {
430
431 class LoopUnrollAndJam : public LoopPass {
432 public:
433 static char ID; // Pass ID, replacement for typeid
434 unsigned OptLevel;
435
LoopUnrollAndJam(int OptLevel=2)436 LoopUnrollAndJam(int OptLevel = 2) : LoopPass(ID), OptLevel(OptLevel) {
437 initializeLoopUnrollAndJamPass(*PassRegistry::getPassRegistry());
438 }
439
runOnLoop(Loop * L,LPPassManager & LPM)440 bool runOnLoop(Loop *L, LPPassManager &LPM) override {
441 if (skipLoop(L))
442 return false;
443
444 Function &F = *L->getHeader()->getParent();
445
446 auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
447 LoopInfo *LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
448 ScalarEvolution &SE = getAnalysis<ScalarEvolutionWrapperPass>().getSE();
449 const TargetTransformInfo &TTI =
450 getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
451 auto &AC = getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
452 auto &DI = getAnalysis<DependenceAnalysisWrapperPass>().getDI();
453 // For the old PM, we can't use OptimizationRemarkEmitter as an analysis
454 // pass. Function analyses need to be preserved across loop transformations
455 // but ORE cannot be preserved (see comment before the pass definition).
456 OptimizationRemarkEmitter ORE(&F);
457
458 LoopUnrollResult Result =
459 tryToUnrollAndJamLoop(L, DT, LI, SE, TTI, AC, DI, ORE, OptLevel);
460
461 if (Result == LoopUnrollResult::FullyUnrolled)
462 LPM.markLoopAsDeleted(*L);
463
464 return Result != LoopUnrollResult::Unmodified;
465 }
466
467 /// This transformation requires natural loop information & requires that
468 /// loop preheaders be inserted into the CFG...
getAnalysisUsage(AnalysisUsage & AU) const469 void getAnalysisUsage(AnalysisUsage &AU) const override {
470 AU.addRequired<AssumptionCacheTracker>();
471 AU.addRequired<TargetTransformInfoWrapperPass>();
472 AU.addRequired<DependenceAnalysisWrapperPass>();
473 getLoopAnalysisUsage(AU);
474 }
475 };
476
477 } // end anonymous namespace
478
479 char LoopUnrollAndJam::ID = 0;
480
481 INITIALIZE_PASS_BEGIN(LoopUnrollAndJam, "loop-unroll-and-jam",
482 "Unroll and Jam loops", false, false)
INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)483 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
484 INITIALIZE_PASS_DEPENDENCY(LoopPass)
485 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
486 INITIALIZE_PASS_DEPENDENCY(DependenceAnalysisWrapperPass)
487 INITIALIZE_PASS_END(LoopUnrollAndJam, "loop-unroll-and-jam",
488 "Unroll and Jam loops", false, false)
489
490 Pass *llvm::createLoopUnrollAndJamPass(int OptLevel) {
491 return new LoopUnrollAndJam(OptLevel);
492 }
493
run(Loop & L,LoopAnalysisManager & AM,LoopStandardAnalysisResults & AR,LPMUpdater &)494 PreservedAnalyses LoopUnrollAndJamPass::run(Loop &L, LoopAnalysisManager &AM,
495 LoopStandardAnalysisResults &AR,
496 LPMUpdater &) {
497 const auto &FAM =
498 AM.getResult<FunctionAnalysisManagerLoopProxy>(L, AR).getManager();
499 Function *F = L.getHeader()->getParent();
500
501 auto *ORE = FAM.getCachedResult<OptimizationRemarkEmitterAnalysis>(*F);
502 // FIXME: This should probably be optional rather than required.
503 if (!ORE)
504 report_fatal_error(
505 "LoopUnrollAndJamPass: OptimizationRemarkEmitterAnalysis not cached at "
506 "a higher level");
507
508 DependenceInfo DI(F, &AR.AA, &AR.SE, &AR.LI);
509
510 LoopUnrollResult Result = tryToUnrollAndJamLoop(
511 &L, AR.DT, &AR.LI, AR.SE, AR.TTI, AR.AC, DI, *ORE, OptLevel);
512
513 if (Result == LoopUnrollResult::Unmodified)
514 return PreservedAnalyses::all();
515
516 return getLoopPassPreservedAnalyses();
517 }
518