Bug Summary

File:lib/Transforms/Scalar/LoopStrengthReduce.cpp
Warning:line 3045, column 3
Forming reference to null pointer

Annotated Source Code

1//===- LoopStrengthReduce.cpp - Strength Reduce IVs in Loops --------------===//
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 transformation analyzes and transforms the induction variables (and
11// computations derived from them) into forms suitable for efficient execution
12// on the target.
13//
14// This pass performs a strength reduction on array references inside loops that
15// have as one or more of their components the loop induction variable, it
16// rewrites expressions to take advantage of scaled-index addressing modes
17// available on the target, and it performs a variety of other optimizations
18// related to loop induction variables.
19//
20// Terminology note: this code has a lot of handling for "post-increment" or
21// "post-inc" users. This is not talking about post-increment addressing modes;
22// it is instead talking about code like this:
23//
24// %i = phi [ 0, %entry ], [ %i.next, %latch ]
25// ...
26// %i.next = add %i, 1
27// %c = icmp eq %i.next, %n
28//
29// The SCEV for %i is {0,+,1}<%L>. The SCEV for %i.next is {1,+,1}<%L>, however
30// it's useful to think about these as the same register, with some uses using
31// the value of the register before the add and some using it after. In this
32// example, the icmp is a post-increment user, since it uses %i.next, which is
33// the value of the induction variable after the increment. The other common
34// case of post-increment users is users outside the loop.
35//
36// TODO: More sophistication in the way Formulae are generated and filtered.
37//
38// TODO: Handle multiple loops at a time.
39//
40// TODO: Should the addressing mode BaseGV be changed to a ConstantExpr instead
41// of a GlobalValue?
42//
43// TODO: When truncation is free, truncate ICmp users' operands to make it a
44// smaller encoding (on x86 at least).
45//
46// TODO: When a negated register is used by an add (such as in a list of
47// multiple base registers, or as the increment expression in an addrec),
48// we may not actually need both reg and (-1 * reg) in registers; the
49// negation can be implemented by using a sub instead of an add. The
50// lack of support for taking this into consideration when making
51// register pressure decisions is partly worked around by the "Special"
52// use kind.
53//
54//===----------------------------------------------------------------------===//
55
56#include "llvm/Transforms/Scalar/LoopStrengthReduce.h"
57#include "llvm/ADT/APInt.h"
58#include "llvm/ADT/DenseMap.h"
59#include "llvm/ADT/DenseSet.h"
60#include "llvm/ADT/Hashing.h"
61#include "llvm/ADT/PointerIntPair.h"
62#include "llvm/ADT/STLExtras.h"
63#include "llvm/ADT/SetVector.h"
64#include "llvm/ADT/SmallBitVector.h"
65#include "llvm/ADT/SmallPtrSet.h"
66#include "llvm/ADT/SmallSet.h"
67#include "llvm/ADT/SmallVector.h"
68#include "llvm/Analysis/IVUsers.h"
69#include "llvm/Analysis/LoopInfo.h"
70#include "llvm/Analysis/LoopPass.h"
71#include "llvm/Analysis/ScalarEvolution.h"
72#include "llvm/Analysis/ScalarEvolutionExpander.h"
73#include "llvm/Analysis/ScalarEvolutionExpressions.h"
74#include "llvm/Analysis/ScalarEvolutionNormalization.h"
75#include "llvm/Analysis/TargetTransformInfo.h"
76#include "llvm/IR/BasicBlock.h"
77#include "llvm/IR/Constant.h"
78#include "llvm/IR/Constants.h"
79#include "llvm/IR/DerivedTypes.h"
80#include "llvm/IR/Dominators.h"
81#include "llvm/IR/GlobalValue.h"
82#include "llvm/IR/IRBuilder.h"
83#include "llvm/IR/Instruction.h"
84#include "llvm/IR/Instructions.h"
85#include "llvm/IR/IntrinsicInst.h"
86#include "llvm/IR/Module.h"
87#include "llvm/IR/OperandTraits.h"
88#include "llvm/IR/Operator.h"
89#include "llvm/IR/Type.h"
90#include "llvm/IR/Value.h"
91#include "llvm/IR/ValueHandle.h"
92#include "llvm/Pass.h"
93#include "llvm/Support/Casting.h"
94#include "llvm/Support/CommandLine.h"
95#include "llvm/Support/Compiler.h"
96#include "llvm/Support/Debug.h"
97#include "llvm/Support/ErrorHandling.h"
98#include "llvm/Support/MathExtras.h"
99#include "llvm/Support/raw_ostream.h"
100#include "llvm/Transforms/Scalar.h"
101#include "llvm/Transforms/Scalar/LoopPassManager.h"
102#include "llvm/Transforms/Utils/BasicBlockUtils.h"
103#include "llvm/Transforms/Utils/Local.h"
104#include <algorithm>
105#include <cassert>
106#include <cstddef>
107#include <cstdint>
108#include <cstdlib>
109#include <iterator>
110#include <map>
111#include <tuple>
112#include <utility>
113
114using namespace llvm;
115
116#define DEBUG_TYPE"loop-reduce" "loop-reduce"
117
118/// MaxIVUsers is an arbitrary threshold that provides an early opportunitiy for
119/// bail out. This threshold is far beyond the number of users that LSR can
120/// conceivably solve, so it should not affect generated code, but catches the
121/// worst cases before LSR burns too much compile time and stack space.
122static const unsigned MaxIVUsers = 200;
123
124// Temporary flag to cleanup congruent phis after LSR phi expansion.
125// It's currently disabled until we can determine whether it's truly useful or
126// not. The flag should be removed after the v3.0 release.
127// This is now needed for ivchains.
128static cl::opt<bool> EnablePhiElim(
129 "enable-lsr-phielim", cl::Hidden, cl::init(true),
130 cl::desc("Enable LSR phi elimination"));
131
132// The flag adds instruction count to solutions cost comparision.
133static cl::opt<bool> InsnsCost(
134 "lsr-insns-cost", cl::Hidden, cl::init(false),
135 cl::desc("Add instruction count to a LSR cost model"));
136
137// Flag to choose how to narrow complex lsr solution
138static cl::opt<bool> LSRExpNarrow(
139 "lsr-exp-narrow", cl::Hidden, cl::init(false),
140 cl::desc("Narrow LSR complex solution using"
141 " expectation of registers number"));
142
143#ifndef NDEBUG
144// Stress test IV chain generation.
145static cl::opt<bool> StressIVChain(
146 "stress-ivchain", cl::Hidden, cl::init(false),
147 cl::desc("Stress test LSR IV chains"));
148#else
149static bool StressIVChain = false;
150#endif
151
152namespace {
153
154struct MemAccessTy {
155 /// Used in situations where the accessed memory type is unknown.
156 static const unsigned UnknownAddressSpace = ~0u;
157
158 Type *MemTy;
159 unsigned AddrSpace;
160
161 MemAccessTy() : MemTy(nullptr), AddrSpace(UnknownAddressSpace) {}
162
163 MemAccessTy(Type *Ty, unsigned AS) :
164 MemTy(Ty), AddrSpace(AS) {}
165
166 bool operator==(MemAccessTy Other) const {
167 return MemTy == Other.MemTy && AddrSpace == Other.AddrSpace;
168 }
169
170 bool operator!=(MemAccessTy Other) const { return !(*this == Other); }
171
172 static MemAccessTy getUnknown(LLVMContext &Ctx,
173 unsigned AS = UnknownAddressSpace) {
174 return MemAccessTy(Type::getVoidTy(Ctx), AS);
175 }
176};
177
178/// This class holds data which is used to order reuse candidates.
179class RegSortData {
180public:
181 /// This represents the set of LSRUse indices which reference
182 /// a particular register.
183 SmallBitVector UsedByIndices;
184
185 void print(raw_ostream &OS) const;
186 void dump() const;
187};
188
189} // end anonymous namespace
190
191void RegSortData::print(raw_ostream &OS) const {
192 OS << "[NumUses=" << UsedByIndices.count() << ']';
193}
194
195#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
196LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void RegSortData::dump() const {
197 print(errs()); errs() << '\n';
198}
199#endif
200
201namespace {
202
203/// Map register candidates to information about how they are used.
204class RegUseTracker {
205 typedef DenseMap<const SCEV *, RegSortData> RegUsesTy;
206
207 RegUsesTy RegUsesMap;
208 SmallVector<const SCEV *, 16> RegSequence;
209
210public:
211 void countRegister(const SCEV *Reg, size_t LUIdx);
212 void dropRegister(const SCEV *Reg, size_t LUIdx);
213 void swapAndDropUse(size_t LUIdx, size_t LastLUIdx);
214
215 bool isRegUsedByUsesOtherThan(const SCEV *Reg, size_t LUIdx) const;
216
217 const SmallBitVector &getUsedByIndices(const SCEV *Reg) const;
218
219 void clear();
220
221 typedef SmallVectorImpl<const SCEV *>::iterator iterator;
222 typedef SmallVectorImpl<const SCEV *>::const_iterator const_iterator;
223 iterator begin() { return RegSequence.begin(); }
224 iterator end() { return RegSequence.end(); }
225 const_iterator begin() const { return RegSequence.begin(); }
226 const_iterator end() const { return RegSequence.end(); }
227};
228
229} // end anonymous namespace
230
231void
232RegUseTracker::countRegister(const SCEV *Reg, size_t LUIdx) {
233 std::pair<RegUsesTy::iterator, bool> Pair =
234 RegUsesMap.insert(std::make_pair(Reg, RegSortData()));
235 RegSortData &RSD = Pair.first->second;
236 if (Pair.second)
237 RegSequence.push_back(Reg);
238 RSD.UsedByIndices.resize(std::max(RSD.UsedByIndices.size(), LUIdx + 1));
239 RSD.UsedByIndices.set(LUIdx);
240}
241
242void
243RegUseTracker::dropRegister(const SCEV *Reg, size_t LUIdx) {
244 RegUsesTy::iterator It = RegUsesMap.find(Reg);
245 assert(It != RegUsesMap.end())((It != RegUsesMap.end()) ? static_cast<void> (0) : __assert_fail
("It != RegUsesMap.end()", "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 245, __PRETTY_FUNCTION__))
;
246 RegSortData &RSD = It->second;
247 assert(RSD.UsedByIndices.size() > LUIdx)((RSD.UsedByIndices.size() > LUIdx) ? static_cast<void>
(0) : __assert_fail ("RSD.UsedByIndices.size() > LUIdx", "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 247, __PRETTY_FUNCTION__))
;
248 RSD.UsedByIndices.reset(LUIdx);
249}
250
251void
252RegUseTracker::swapAndDropUse(size_t LUIdx, size_t LastLUIdx) {
253 assert(LUIdx <= LastLUIdx)((LUIdx <= LastLUIdx) ? static_cast<void> (0) : __assert_fail
("LUIdx <= LastLUIdx", "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 253, __PRETTY_FUNCTION__))
;
254
255 // Update RegUses. The data structure is not optimized for this purpose;
256 // we must iterate through it and update each of the bit vectors.
257 for (auto &Pair : RegUsesMap) {
258 SmallBitVector &UsedByIndices = Pair.second.UsedByIndices;
259 if (LUIdx < UsedByIndices.size())
260 UsedByIndices[LUIdx] =
261 LastLUIdx < UsedByIndices.size() ? UsedByIndices[LastLUIdx] : false;
262 UsedByIndices.resize(std::min(UsedByIndices.size(), LastLUIdx));
263 }
264}
265
266bool
267RegUseTracker::isRegUsedByUsesOtherThan(const SCEV *Reg, size_t LUIdx) const {
268 RegUsesTy::const_iterator I = RegUsesMap.find(Reg);
269 if (I == RegUsesMap.end())
270 return false;
271 const SmallBitVector &UsedByIndices = I->second.UsedByIndices;
272 int i = UsedByIndices.find_first();
273 if (i == -1) return false;
274 if ((size_t)i != LUIdx) return true;
275 return UsedByIndices.find_next(i) != -1;
276}
277
278const SmallBitVector &RegUseTracker::getUsedByIndices(const SCEV *Reg) const {
279 RegUsesTy::const_iterator I = RegUsesMap.find(Reg);
280 assert(I != RegUsesMap.end() && "Unknown register!")((I != RegUsesMap.end() && "Unknown register!") ? static_cast
<void> (0) : __assert_fail ("I != RegUsesMap.end() && \"Unknown register!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 280, __PRETTY_FUNCTION__))
;
281 return I->second.UsedByIndices;
282}
283
284void RegUseTracker::clear() {
285 RegUsesMap.clear();
286 RegSequence.clear();
287}
288
289namespace {
290
291/// This class holds information that describes a formula for computing
292/// satisfying a use. It may include broken-out immediates and scaled registers.
293struct Formula {
294 /// Global base address used for complex addressing.
295 GlobalValue *BaseGV;
296
297 /// Base offset for complex addressing.
298 int64_t BaseOffset;
299
300 /// Whether any complex addressing has a base register.
301 bool HasBaseReg;
302
303 /// The scale of any complex addressing.
304 int64_t Scale;
305
306 /// The list of "base" registers for this use. When this is non-empty. The
307 /// canonical representation of a formula is
308 /// 1. BaseRegs.size > 1 implies ScaledReg != NULL and
309 /// 2. ScaledReg != NULL implies Scale != 1 || !BaseRegs.empty().
310 /// 3. The reg containing recurrent expr related with currect loop in the
311 /// formula should be put in the ScaledReg.
312 /// #1 enforces that the scaled register is always used when at least two
313 /// registers are needed by the formula: e.g., reg1 + reg2 is reg1 + 1 * reg2.
314 /// #2 enforces that 1 * reg is reg.
315 /// #3 ensures invariant regs with respect to current loop can be combined
316 /// together in LSR codegen.
317 /// This invariant can be temporarly broken while building a formula.
318 /// However, every formula inserted into the LSRInstance must be in canonical
319 /// form.
320 SmallVector<const SCEV *, 4> BaseRegs;
321
322 /// The 'scaled' register for this use. This should be non-null when Scale is
323 /// not zero.
324 const SCEV *ScaledReg;
325
326 /// An additional constant offset which added near the use. This requires a
327 /// temporary register, but the offset itself can live in an add immediate
328 /// field rather than a register.
329 int64_t UnfoldedOffset;
330
331 Formula()
332 : BaseGV(nullptr), BaseOffset(0), HasBaseReg(false), Scale(0),
333 ScaledReg(nullptr), UnfoldedOffset(0) {}
334
335 void initialMatch(const SCEV *S, Loop *L, ScalarEvolution &SE);
336
337 bool isCanonical(const Loop &L) const;
338
339 void canonicalize(const Loop &L);
340
341 bool unscale();
342
343 bool hasZeroEnd() const;
344
345 size_t getNumRegs() const;
346 Type *getType() const;
347
348 void deleteBaseReg(const SCEV *&S);
349
350 bool referencesReg(const SCEV *S) const;
351 bool hasRegsUsedByUsesOtherThan(size_t LUIdx,
352 const RegUseTracker &RegUses) const;
353
354 void print(raw_ostream &OS) const;
355 void dump() const;
356};
357
358} // end anonymous namespace
359
360/// Recursion helper for initialMatch.
361static void DoInitialMatch(const SCEV *S, Loop *L,
362 SmallVectorImpl<const SCEV *> &Good,
363 SmallVectorImpl<const SCEV *> &Bad,
364 ScalarEvolution &SE) {
365 // Collect expressions which properly dominate the loop header.
366 if (SE.properlyDominates(S, L->getHeader())) {
367 Good.push_back(S);
368 return;
369 }
370
371 // Look at add operands.
372 if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
373 for (const SCEV *S : Add->operands())
374 DoInitialMatch(S, L, Good, Bad, SE);
375 return;
376 }
377
378 // Look at addrec operands.
379 if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S))
380 if (!AR->getStart()->isZero() && AR->isAffine()) {
381 DoInitialMatch(AR->getStart(), L, Good, Bad, SE);
382 DoInitialMatch(SE.getAddRecExpr(SE.getConstant(AR->getType(), 0),
383 AR->getStepRecurrence(SE),
384 // FIXME: AR->getNoWrapFlags()
385 AR->getLoop(), SCEV::FlagAnyWrap),
386 L, Good, Bad, SE);
387 return;
388 }
389
390 // Handle a multiplication by -1 (negation) if it didn't fold.
391 if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(S))
392 if (Mul->getOperand(0)->isAllOnesValue()) {
393 SmallVector<const SCEV *, 4> Ops(Mul->op_begin()+1, Mul->op_end());
394 const SCEV *NewMul = SE.getMulExpr(Ops);
395
396 SmallVector<const SCEV *, 4> MyGood;
397 SmallVector<const SCEV *, 4> MyBad;
398 DoInitialMatch(NewMul, L, MyGood, MyBad, SE);
399 const SCEV *NegOne = SE.getSCEV(ConstantInt::getAllOnesValue(
400 SE.getEffectiveSCEVType(NewMul->getType())));
401 for (const SCEV *S : MyGood)
402 Good.push_back(SE.getMulExpr(NegOne, S));
403 for (const SCEV *S : MyBad)
404 Bad.push_back(SE.getMulExpr(NegOne, S));
405 return;
406 }
407
408 // Ok, we can't do anything interesting. Just stuff the whole thing into a
409 // register and hope for the best.
410 Bad.push_back(S);
411}
412
413/// Incorporate loop-variant parts of S into this Formula, attempting to keep
414/// all loop-invariant and loop-computable values in a single base register.
415void Formula::initialMatch(const SCEV *S, Loop *L, ScalarEvolution &SE) {
416 SmallVector<const SCEV *, 4> Good;
417 SmallVector<const SCEV *, 4> Bad;
418 DoInitialMatch(S, L, Good, Bad, SE);
419 if (!Good.empty()) {
420 const SCEV *Sum = SE.getAddExpr(Good);
421 if (!Sum->isZero())
422 BaseRegs.push_back(Sum);
423 HasBaseReg = true;
424 }
425 if (!Bad.empty()) {
426 const SCEV *Sum = SE.getAddExpr(Bad);
427 if (!Sum->isZero())
428 BaseRegs.push_back(Sum);
429 HasBaseReg = true;
430 }
431 canonicalize(*L);
432}
433
434/// \brief Check whether or not this formula statisfies the canonical
435/// representation.
436/// \see Formula::BaseRegs.
437bool Formula::isCanonical(const Loop &L) const {
438 if (!ScaledReg)
439 return BaseRegs.size() <= 1;
440
441 if (Scale != 1)
442 return true;
443
444 if (Scale == 1 && BaseRegs.empty())
445 return false;
446
447 const SCEVAddRecExpr *SAR = dyn_cast<const SCEVAddRecExpr>(ScaledReg);
448 if (SAR && SAR->getLoop() == &L)
449 return true;
450
451 // If ScaledReg is not a recurrent expr, or it is but its loop is not current
452 // loop, meanwhile BaseRegs contains a recurrent expr reg related with current
453 // loop, we want to swap the reg in BaseRegs with ScaledReg.
454 auto I =
455 find_if(make_range(BaseRegs.begin(), BaseRegs.end()), [&](const SCEV *S) {
456 return isa<const SCEVAddRecExpr>(S) &&
457 (cast<SCEVAddRecExpr>(S)->getLoop() == &L);
458 });
459 return I == BaseRegs.end();
460}
461
462/// \brief Helper method to morph a formula into its canonical representation.
463/// \see Formula::BaseRegs.
464/// Every formula having more than one base register, must use the ScaledReg
465/// field. Otherwise, we would have to do special cases everywhere in LSR
466/// to treat reg1 + reg2 + ... the same way as reg1 + 1*reg2 + ...
467/// On the other hand, 1*reg should be canonicalized into reg.
468void Formula::canonicalize(const Loop &L) {
469 if (isCanonical(L))
470 return;
471 // So far we did not need this case. This is easy to implement but it is
472 // useless to maintain dead code. Beside it could hurt compile time.
473 assert(!BaseRegs.empty() && "1*reg => reg, should not be needed.")((!BaseRegs.empty() && "1*reg => reg, should not be needed."
) ? static_cast<void> (0) : __assert_fail ("!BaseRegs.empty() && \"1*reg => reg, should not be needed.\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 473, __PRETTY_FUNCTION__))
;
474
475 // Keep the invariant sum in BaseRegs and one of the variant sum in ScaledReg.
476 if (!ScaledReg) {
477 ScaledReg = BaseRegs.back();
478 BaseRegs.pop_back();
479 Scale = 1;
480 }
481
482 // If ScaledReg is an invariant with respect to L, find the reg from
483 // BaseRegs containing the recurrent expr related with Loop L. Swap the
484 // reg with ScaledReg.
485 const SCEVAddRecExpr *SAR = dyn_cast<const SCEVAddRecExpr>(ScaledReg);
486 if (!SAR || SAR->getLoop() != &L) {
487 auto I = find_if(make_range(BaseRegs.begin(), BaseRegs.end()),
488 [&](const SCEV *S) {
489 return isa<const SCEVAddRecExpr>(S) &&
490 (cast<SCEVAddRecExpr>(S)->getLoop() == &L);
491 });
492 if (I != BaseRegs.end())
493 std::swap(ScaledReg, *I);
494 }
495}
496
497/// \brief Get rid of the scale in the formula.
498/// In other words, this method morphes reg1 + 1*reg2 into reg1 + reg2.
499/// \return true if it was possible to get rid of the scale, false otherwise.
500/// \note After this operation the formula may not be in the canonical form.
501bool Formula::unscale() {
502 if (Scale != 1)
503 return false;
504 Scale = 0;
505 BaseRegs.push_back(ScaledReg);
506 ScaledReg = nullptr;
507 return true;
508}
509
510bool Formula::hasZeroEnd() const {
511 if (UnfoldedOffset || BaseOffset)
512 return false;
513 if (BaseRegs.size() != 1 || ScaledReg)
514 return false;
515 return true;
516}
517
518/// Return the total number of register operands used by this formula. This does
519/// not include register uses implied by non-constant addrec strides.
520size_t Formula::getNumRegs() const {
521 return !!ScaledReg + BaseRegs.size();
522}
523
524/// Return the type of this formula, if it has one, or null otherwise. This type
525/// is meaningless except for the bit size.
526Type *Formula::getType() const {
527 return !BaseRegs.empty() ? BaseRegs.front()->getType() :
528 ScaledReg ? ScaledReg->getType() :
529 BaseGV ? BaseGV->getType() :
530 nullptr;
531}
532
533/// Delete the given base reg from the BaseRegs list.
534void Formula::deleteBaseReg(const SCEV *&S) {
535 if (&S != &BaseRegs.back())
536 std::swap(S, BaseRegs.back());
537 BaseRegs.pop_back();
538}
539
540/// Test if this formula references the given register.
541bool Formula::referencesReg(const SCEV *S) const {
542 return S == ScaledReg || is_contained(BaseRegs, S);
543}
544
545/// Test whether this formula uses registers which are used by uses other than
546/// the use with the given index.
547bool Formula::hasRegsUsedByUsesOtherThan(size_t LUIdx,
548 const RegUseTracker &RegUses) const {
549 if (ScaledReg)
550 if (RegUses.isRegUsedByUsesOtherThan(ScaledReg, LUIdx))
551 return true;
552 for (const SCEV *BaseReg : BaseRegs)
553 if (RegUses.isRegUsedByUsesOtherThan(BaseReg, LUIdx))
554 return true;
555 return false;
556}
557
558void Formula::print(raw_ostream &OS) const {
559 bool First = true;
560 if (BaseGV) {
561 if (!First) OS << " + "; else First = false;
562 BaseGV->printAsOperand(OS, /*PrintType=*/false);
563 }
564 if (BaseOffset != 0) {
565 if (!First) OS << " + "; else First = false;
566 OS << BaseOffset;
567 }
568 for (const SCEV *BaseReg : BaseRegs) {
569 if (!First) OS << " + "; else First = false;
570 OS << "reg(" << *BaseReg << ')';
571 }
572 if (HasBaseReg && BaseRegs.empty()) {
573 if (!First) OS << " + "; else First = false;
574 OS << "**error: HasBaseReg**";
575 } else if (!HasBaseReg && !BaseRegs.empty()) {
576 if (!First) OS << " + "; else First = false;
577 OS << "**error: !HasBaseReg**";
578 }
579 if (Scale != 0) {
580 if (!First) OS << " + "; else First = false;
581 OS << Scale << "*reg(";
582 if (ScaledReg)
583 OS << *ScaledReg;
584 else
585 OS << "<unknown>";
586 OS << ')';
587 }
588 if (UnfoldedOffset != 0) {
589 if (!First) OS << " + ";
590 OS << "imm(" << UnfoldedOffset << ')';
591 }
592}
593
594#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
595LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void Formula::dump() const {
596 print(errs()); errs() << '\n';
597}
598#endif
599
600/// Return true if the given addrec can be sign-extended without changing its
601/// value.
602static bool isAddRecSExtable(const SCEVAddRecExpr *AR, ScalarEvolution &SE) {
603 Type *WideTy =
604 IntegerType::get(SE.getContext(), SE.getTypeSizeInBits(AR->getType()) + 1);
605 return isa<SCEVAddRecExpr>(SE.getSignExtendExpr(AR, WideTy));
606}
607
608/// Return true if the given add can be sign-extended without changing its
609/// value.
610static bool isAddSExtable(const SCEVAddExpr *A, ScalarEvolution &SE) {
611 Type *WideTy =
612 IntegerType::get(SE.getContext(), SE.getTypeSizeInBits(A->getType()) + 1);
613 return isa<SCEVAddExpr>(SE.getSignExtendExpr(A, WideTy));
614}
615
616/// Return true if the given mul can be sign-extended without changing its
617/// value.
618static bool isMulSExtable(const SCEVMulExpr *M, ScalarEvolution &SE) {
619 Type *WideTy =
620 IntegerType::get(SE.getContext(),
621 SE.getTypeSizeInBits(M->getType()) * M->getNumOperands());
622 return isa<SCEVMulExpr>(SE.getSignExtendExpr(M, WideTy));
623}
624
625/// Return an expression for LHS /s RHS, if it can be determined and if the
626/// remainder is known to be zero, or null otherwise. If IgnoreSignificantBits
627/// is true, expressions like (X * Y) /s Y are simplified to Y, ignoring that
628/// the multiplication may overflow, which is useful when the result will be
629/// used in a context where the most significant bits are ignored.
630static const SCEV *getExactSDiv(const SCEV *LHS, const SCEV *RHS,
631 ScalarEvolution &SE,
632 bool IgnoreSignificantBits = false) {
633 // Handle the trivial case, which works for any SCEV type.
634 if (LHS == RHS)
635 return SE.getConstant(LHS->getType(), 1);
636
637 // Handle a few RHS special cases.
638 const SCEVConstant *RC = dyn_cast<SCEVConstant>(RHS);
639 if (RC) {
640 const APInt &RA = RC->getAPInt();
641 // Handle x /s -1 as x * -1, to give ScalarEvolution a chance to do
642 // some folding.
643 if (RA.isAllOnesValue())
644 return SE.getMulExpr(LHS, RC);
645 // Handle x /s 1 as x.
646 if (RA == 1)
647 return LHS;
648 }
649
650 // Check for a division of a constant by a constant.
651 if (const SCEVConstant *C = dyn_cast<SCEVConstant>(LHS)) {
652 if (!RC)
653 return nullptr;
654 const APInt &LA = C->getAPInt();
655 const APInt &RA = RC->getAPInt();
656 if (LA.srem(RA) != 0)
657 return nullptr;
658 return SE.getConstant(LA.sdiv(RA));
659 }
660
661 // Distribute the sdiv over addrec operands, if the addrec doesn't overflow.
662 if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(LHS)) {
663 if ((IgnoreSignificantBits || isAddRecSExtable(AR, SE)) && AR->isAffine()) {
664 const SCEV *Step = getExactSDiv(AR->getStepRecurrence(SE), RHS, SE,
665 IgnoreSignificantBits);
666 if (!Step) return nullptr;
667 const SCEV *Start = getExactSDiv(AR->getStart(), RHS, SE,
668 IgnoreSignificantBits);
669 if (!Start) return nullptr;
670 // FlagNW is independent of the start value, step direction, and is
671 // preserved with smaller magnitude steps.
672 // FIXME: AR->getNoWrapFlags(SCEV::FlagNW)
673 return SE.getAddRecExpr(Start, Step, AR->getLoop(), SCEV::FlagAnyWrap);
674 }
675 return nullptr;
676 }
677
678 // Distribute the sdiv over add operands, if the add doesn't overflow.
679 if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(LHS)) {
680 if (IgnoreSignificantBits || isAddSExtable(Add, SE)) {
681 SmallVector<const SCEV *, 8> Ops;
682 for (const SCEV *S : Add->operands()) {
683 const SCEV *Op = getExactSDiv(S, RHS, SE, IgnoreSignificantBits);
684 if (!Op) return nullptr;
685 Ops.push_back(Op);
686 }
687 return SE.getAddExpr(Ops);
688 }
689 return nullptr;
690 }
691
692 // Check for a multiply operand that we can pull RHS out of.
693 if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(LHS)) {
694 if (IgnoreSignificantBits || isMulSExtable(Mul, SE)) {
695 SmallVector<const SCEV *, 4> Ops;
696 bool Found = false;
697 for (const SCEV *S : Mul->operands()) {
698 if (!Found)
699 if (const SCEV *Q = getExactSDiv(S, RHS, SE,
700 IgnoreSignificantBits)) {
701 S = Q;
702 Found = true;
703 }
704 Ops.push_back(S);
705 }
706 return Found ? SE.getMulExpr(Ops) : nullptr;
707 }
708 return nullptr;
709 }
710
711 // Otherwise we don't know.
712 return nullptr;
713}
714
715/// If S involves the addition of a constant integer value, return that integer
716/// value, and mutate S to point to a new SCEV with that value excluded.
717static int64_t ExtractImmediate(const SCEV *&S, ScalarEvolution &SE) {
718 if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S)) {
719 if (C->getAPInt().getMinSignedBits() <= 64) {
720 S = SE.getConstant(C->getType(), 0);
721 return C->getValue()->getSExtValue();
722 }
723 } else if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
724 SmallVector<const SCEV *, 8> NewOps(Add->op_begin(), Add->op_end());
725 int64_t Result = ExtractImmediate(NewOps.front(), SE);
726 if (Result != 0)
727 S = SE.getAddExpr(NewOps);
728 return Result;
729 } else if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
730 SmallVector<const SCEV *, 8> NewOps(AR->op_begin(), AR->op_end());
731 int64_t Result = ExtractImmediate(NewOps.front(), SE);
732 if (Result != 0)
733 S = SE.getAddRecExpr(NewOps, AR->getLoop(),
734 // FIXME: AR->getNoWrapFlags(SCEV::FlagNW)
735 SCEV::FlagAnyWrap);
736 return Result;
737 }
738 return 0;
739}
740
741/// If S involves the addition of a GlobalValue address, return that symbol, and
742/// mutate S to point to a new SCEV with that value excluded.
743static GlobalValue *ExtractSymbol(const SCEV *&S, ScalarEvolution &SE) {
744 if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
745 if (GlobalValue *GV = dyn_cast<GlobalValue>(U->getValue())) {
746 S = SE.getConstant(GV->getType(), 0);
747 return GV;
748 }
749 } else if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
750 SmallVector<const SCEV *, 8> NewOps(Add->op_begin(), Add->op_end());
751 GlobalValue *Result = ExtractSymbol(NewOps.back(), SE);
752 if (Result)
753 S = SE.getAddExpr(NewOps);
754 return Result;
755 } else if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
756 SmallVector<const SCEV *, 8> NewOps(AR->op_begin(), AR->op_end());
757 GlobalValue *Result = ExtractSymbol(NewOps.front(), SE);
758 if (Result)
759 S = SE.getAddRecExpr(NewOps, AR->getLoop(),
760 // FIXME: AR->getNoWrapFlags(SCEV::FlagNW)
761 SCEV::FlagAnyWrap);
762 return Result;
763 }
764 return nullptr;
765}
766
767/// Returns true if the specified instruction is using the specified value as an
768/// address.
769static bool isAddressUse(Instruction *Inst, Value *OperandVal) {
770 bool isAddress = isa<LoadInst>(Inst);
771 if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
772 if (SI->getPointerOperand() == OperandVal)
773 isAddress = true;
774 } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
775 // Addressing modes can also be folded into prefetches and a variety
776 // of intrinsics.
777 switch (II->getIntrinsicID()) {
778 default: break;
779 case Intrinsic::prefetch:
780 if (II->getArgOperand(0) == OperandVal)
781 isAddress = true;
782 break;
783 }
784 } else if (AtomicRMWInst *RMW = dyn_cast<AtomicRMWInst>(Inst)) {
785 if (RMW->getPointerOperand() == OperandVal)
786 isAddress = true;
787 } else if (AtomicCmpXchgInst *CmpX = dyn_cast<AtomicCmpXchgInst>(Inst)) {
788 if (CmpX->getPointerOperand() == OperandVal)
789 isAddress = true;
790 }
791 return isAddress;
792}
793
794/// Return the type of the memory being accessed.
795static MemAccessTy getAccessType(const Instruction *Inst) {
796 MemAccessTy AccessTy(Inst->getType(), MemAccessTy::UnknownAddressSpace);
797 if (const StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
798 AccessTy.MemTy = SI->getOperand(0)->getType();
799 AccessTy.AddrSpace = SI->getPointerAddressSpace();
800 } else if (const LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
801 AccessTy.AddrSpace = LI->getPointerAddressSpace();
802 } else if (const AtomicRMWInst *RMW = dyn_cast<AtomicRMWInst>(Inst)) {
803 AccessTy.AddrSpace = RMW->getPointerAddressSpace();
804 } else if (const AtomicCmpXchgInst *CmpX = dyn_cast<AtomicCmpXchgInst>(Inst)) {
805 AccessTy.AddrSpace = CmpX->getPointerAddressSpace();
806 }
807
808 // All pointers have the same requirements, so canonicalize them to an
809 // arbitrary pointer type to minimize variation.
810 if (PointerType *PTy = dyn_cast<PointerType>(AccessTy.MemTy))
811 AccessTy.MemTy = PointerType::get(IntegerType::get(PTy->getContext(), 1),
812 PTy->getAddressSpace());
813
814 return AccessTy;
815}
816
817/// Return true if this AddRec is already a phi in its loop.
818static bool isExistingPhi(const SCEVAddRecExpr *AR, ScalarEvolution &SE) {
819 for (BasicBlock::iterator I = AR->getLoop()->getHeader()->begin();
820 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
821 if (SE.isSCEVable(PN->getType()) &&
822 (SE.getEffectiveSCEVType(PN->getType()) ==
823 SE.getEffectiveSCEVType(AR->getType())) &&
824 SE.getSCEV(PN) == AR)
825 return true;
826 }
827 return false;
828}
829
830/// Check if expanding this expression is likely to incur significant cost. This
831/// is tricky because SCEV doesn't track which expressions are actually computed
832/// by the current IR.
833///
834/// We currently allow expansion of IV increments that involve adds,
835/// multiplication by constants, and AddRecs from existing phis.
836///
837/// TODO: Allow UDivExpr if we can find an existing IV increment that is an
838/// obvious multiple of the UDivExpr.
839static bool isHighCostExpansion(const SCEV *S,
840 SmallPtrSetImpl<const SCEV*> &Processed,
841 ScalarEvolution &SE) {
842 // Zero/One operand expressions
843 switch (S->getSCEVType()) {
844 case scUnknown:
845 case scConstant:
846 return false;
847 case scTruncate:
848 return isHighCostExpansion(cast<SCEVTruncateExpr>(S)->getOperand(),
849 Processed, SE);
850 case scZeroExtend:
851 return isHighCostExpansion(cast<SCEVZeroExtendExpr>(S)->getOperand(),
852 Processed, SE);
853 case scSignExtend:
854 return isHighCostExpansion(cast<SCEVSignExtendExpr>(S)->getOperand(),
855 Processed, SE);
856 }
857
858 if (!Processed.insert(S).second)
859 return false;
860
861 if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
862 for (const SCEV *S : Add->operands()) {
863 if (isHighCostExpansion(S, Processed, SE))
864 return true;
865 }
866 return false;
867 }
868
869 if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(S)) {
870 if (Mul->getNumOperands() == 2) {
871 // Multiplication by a constant is ok
872 if (isa<SCEVConstant>(Mul->getOperand(0)))
873 return isHighCostExpansion(Mul->getOperand(1), Processed, SE);
874
875 // If we have the value of one operand, check if an existing
876 // multiplication already generates this expression.
877 if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(Mul->getOperand(1))) {
878 Value *UVal = U->getValue();
879 for (User *UR : UVal->users()) {
880 // If U is a constant, it may be used by a ConstantExpr.
881 Instruction *UI = dyn_cast<Instruction>(UR);
882 if (UI && UI->getOpcode() == Instruction::Mul &&
883 SE.isSCEVable(UI->getType())) {
884 return SE.getSCEV(UI) == Mul;
885 }
886 }
887 }
888 }
889 }
890
891 if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
892 if (isExistingPhi(AR, SE))
893 return false;
894 }
895
896 // Fow now, consider any other type of expression (div/mul/min/max) high cost.
897 return true;
898}
899
900/// If any of the instructions is the specified set are trivially dead, delete
901/// them and see if this makes any of their operands subsequently dead.
902static bool
903DeleteTriviallyDeadInstructions(SmallVectorImpl<WeakTrackingVH> &DeadInsts) {
904 bool Changed = false;
905
906 while (!DeadInsts.empty()) {
907 Value *V = DeadInsts.pop_back_val();
908 Instruction *I = dyn_cast_or_null<Instruction>(V);
909
910 if (!I || !isInstructionTriviallyDead(I))
911 continue;
912
913 for (Use &O : I->operands())
914 if (Instruction *U = dyn_cast<Instruction>(O)) {
915 O = nullptr;
916 if (U->use_empty())
917 DeadInsts.emplace_back(U);
918 }
919
920 I->eraseFromParent();
921 Changed = true;
922 }
923
924 return Changed;
925}
926
927namespace {
928
929class LSRUse;
930
931} // end anonymous namespace
932
933/// \brief Check if the addressing mode defined by \p F is completely
934/// folded in \p LU at isel time.
935/// This includes address-mode folding and special icmp tricks.
936/// This function returns true if \p LU can accommodate what \p F
937/// defines and up to 1 base + 1 scaled + offset.
938/// In other words, if \p F has several base registers, this function may
939/// still return true. Therefore, users still need to account for
940/// additional base registers and/or unfolded offsets to derive an
941/// accurate cost model.
942static bool isAMCompletelyFolded(const TargetTransformInfo &TTI,
943 const LSRUse &LU, const Formula &F);
944// Get the cost of the scaling factor used in F for LU.
945static unsigned getScalingFactorCost(const TargetTransformInfo &TTI,
946 const LSRUse &LU, const Formula &F,
947 const Loop &L);
948
949namespace {
950
951/// This class is used to measure and compare candidate formulae.
952class Cost {
953 /// TODO: Some of these could be merged. Also, a lexical ordering
954 /// isn't always optimal.
955 unsigned Insns;
956 unsigned NumRegs;
957 unsigned AddRecCost;
958 unsigned NumIVMuls;
959 unsigned NumBaseAdds;
960 unsigned ImmCost;
961 unsigned SetupCost;
962 unsigned ScaleCost;
963
964public:
965 Cost()
966 : Insns(0), NumRegs(0), AddRecCost(0), NumIVMuls(0), NumBaseAdds(0),
967 ImmCost(0), SetupCost(0), ScaleCost(0) {}
968
969 bool operator<(const Cost &Other) const;
970
971 void Lose();
972
973#ifndef NDEBUG
974 // Once any of the metrics loses, they must all remain losers.
975 bool isValid() {
976 return ((Insns | NumRegs | AddRecCost | NumIVMuls | NumBaseAdds
977 | ImmCost | SetupCost | ScaleCost) != ~0u)
978 || ((Insns & NumRegs & AddRecCost & NumIVMuls & NumBaseAdds
979 & ImmCost & SetupCost & ScaleCost) == ~0u);
980 }
981#endif
982
983 bool isLoser() {
984 assert(isValid() && "invalid cost")((isValid() && "invalid cost") ? static_cast<void>
(0) : __assert_fail ("isValid() && \"invalid cost\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 984, __PRETTY_FUNCTION__))
;
985 return NumRegs == ~0u;
986 }
987
988 void RateFormula(const TargetTransformInfo &TTI,
989 const Formula &F,
990 SmallPtrSetImpl<const SCEV *> &Regs,
991 const DenseSet<const SCEV *> &VisitedRegs,
992 const Loop *L,
993 ScalarEvolution &SE, DominatorTree &DT,
994 const LSRUse &LU,
995 SmallPtrSetImpl<const SCEV *> *LoserRegs = nullptr);
996
997 void print(raw_ostream &OS) const;
998 void dump() const;
999
1000private:
1001 void RateRegister(const SCEV *Reg,
1002 SmallPtrSetImpl<const SCEV *> &Regs,
1003 const Loop *L,
1004 ScalarEvolution &SE, DominatorTree &DT);
1005 void RatePrimaryRegister(const SCEV *Reg,
1006 SmallPtrSetImpl<const SCEV *> &Regs,
1007 const Loop *L,
1008 ScalarEvolution &SE, DominatorTree &DT,
1009 SmallPtrSetImpl<const SCEV *> *LoserRegs);
1010};
1011
1012/// An operand value in an instruction which is to be replaced with some
1013/// equivalent, possibly strength-reduced, replacement.
1014struct LSRFixup {
1015 /// The instruction which will be updated.
1016 Instruction *UserInst;
1017
1018 /// The operand of the instruction which will be replaced. The operand may be
1019 /// used more than once; every instance will be replaced.
1020 Value *OperandValToReplace;
1021
1022 /// If this user is to use the post-incremented value of an induction
1023 /// variable, this variable is non-null and holds the loop associated with the
1024 /// induction variable.
1025 PostIncLoopSet PostIncLoops;
1026
1027 /// A constant offset to be added to the LSRUse expression. This allows
1028 /// multiple fixups to share the same LSRUse with different offsets, for
1029 /// example in an unrolled loop.
1030 int64_t Offset;
1031
1032 bool isUseFullyOutsideLoop(const Loop *L) const;
1033
1034 LSRFixup();
1035
1036 void print(raw_ostream &OS) const;
1037 void dump() const;
1038};
1039
1040/// A DenseMapInfo implementation for holding DenseMaps and DenseSets of sorted
1041/// SmallVectors of const SCEV*.
1042struct UniquifierDenseMapInfo {
1043 static SmallVector<const SCEV *, 4> getEmptyKey() {
1044 SmallVector<const SCEV *, 4> V;
1045 V.push_back(reinterpret_cast<const SCEV *>(-1));
1046 return V;
1047 }
1048
1049 static SmallVector<const SCEV *, 4> getTombstoneKey() {
1050 SmallVector<const SCEV *, 4> V;
1051 V.push_back(reinterpret_cast<const SCEV *>(-2));
1052 return V;
1053 }
1054
1055 static unsigned getHashValue(const SmallVector<const SCEV *, 4> &V) {
1056 return static_cast<unsigned>(hash_combine_range(V.begin(), V.end()));
1057 }
1058
1059 static bool isEqual(const SmallVector<const SCEV *, 4> &LHS,
1060 const SmallVector<const SCEV *, 4> &RHS) {
1061 return LHS == RHS;
1062 }
1063};
1064
1065/// This class holds the state that LSR keeps for each use in IVUsers, as well
1066/// as uses invented by LSR itself. It includes information about what kinds of
1067/// things can be folded into the user, information about the user itself, and
1068/// information about how the use may be satisfied. TODO: Represent multiple
1069/// users of the same expression in common?
1070class LSRUse {
1071 DenseSet<SmallVector<const SCEV *, 4>, UniquifierDenseMapInfo> Uniquifier;
1072
1073public:
1074 /// An enum for a kind of use, indicating what types of scaled and immediate
1075 /// operands it might support.
1076 enum KindType {
1077 Basic, ///< A normal use, with no folding.
1078 Special, ///< A special case of basic, allowing -1 scales.
1079 Address, ///< An address use; folding according to TargetLowering
1080 ICmpZero ///< An equality icmp with both operands folded into one.
1081 // TODO: Add a generic icmp too?
1082 };
1083
1084 typedef PointerIntPair<const SCEV *, 2, KindType> SCEVUseKindPair;
1085
1086 KindType Kind;
1087 MemAccessTy AccessTy;
1088
1089 /// The list of operands which are to be replaced.
1090 SmallVector<LSRFixup, 8> Fixups;
1091
1092 /// Keep track of the min and max offsets of the fixups.
1093 int64_t MinOffset;
1094 int64_t MaxOffset;
1095
1096 /// This records whether all of the fixups using this LSRUse are outside of
1097 /// the loop, in which case some special-case heuristics may be used.
1098 bool AllFixupsOutsideLoop;
1099
1100 /// RigidFormula is set to true to guarantee that this use will be associated
1101 /// with a single formula--the one that initially matched. Some SCEV
1102 /// expressions cannot be expanded. This allows LSR to consider the registers
1103 /// used by those expressions without the need to expand them later after
1104 /// changing the formula.
1105 bool RigidFormula;
1106
1107 /// This records the widest use type for any fixup using this
1108 /// LSRUse. FindUseWithSimilarFormula can't consider uses with different max
1109 /// fixup widths to be equivalent, because the narrower one may be relying on
1110 /// the implicit truncation to truncate away bogus bits.
1111 Type *WidestFixupType;
1112
1113 /// A list of ways to build a value that can satisfy this user. After the
1114 /// list is populated, one of these is selected heuristically and used to
1115 /// formulate a replacement for OperandValToReplace in UserInst.
1116 SmallVector<Formula, 12> Formulae;
1117
1118 /// The set of register candidates used by all formulae in this LSRUse.
1119 SmallPtrSet<const SCEV *, 4> Regs;
1120
1121 LSRUse(KindType K, MemAccessTy AT)
1122 : Kind(K), AccessTy(AT), MinOffset(INT64_MAX(9223372036854775807L)), MaxOffset(INT64_MIN(-9223372036854775807L -1)),
1123 AllFixupsOutsideLoop(true), RigidFormula(false),
1124 WidestFixupType(nullptr) {}
1125
1126 LSRFixup &getNewFixup() {
1127 Fixups.push_back(LSRFixup());
1128 return Fixups.back();
1129 }
1130
1131 void pushFixup(LSRFixup &f) {
1132 Fixups.push_back(f);
1133 if (f.Offset > MaxOffset)
1134 MaxOffset = f.Offset;
1135 if (f.Offset < MinOffset)
1136 MinOffset = f.Offset;
1137 }
1138
1139 bool HasFormulaWithSameRegs(const Formula &F) const;
1140 float getNotSelectedProbability(const SCEV *Reg) const;
1141 bool InsertFormula(const Formula &F, const Loop &L);
1142 void DeleteFormula(Formula &F);
1143 void RecomputeRegs(size_t LUIdx, RegUseTracker &Reguses);
1144
1145 void print(raw_ostream &OS) const;
1146 void dump() const;
1147};
1148
1149} // end anonymous namespace
1150
1151/// Tally up interesting quantities from the given register.
1152void Cost::RateRegister(const SCEV *Reg,
1153 SmallPtrSetImpl<const SCEV *> &Regs,
1154 const Loop *L,
1155 ScalarEvolution &SE, DominatorTree &DT) {
1156 if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Reg)) {
1157 // If this is an addrec for another loop, it should be an invariant
1158 // with respect to L since L is the innermost loop (at least
1159 // for now LSR only handles innermost loops).
1160 if (AR->getLoop() != L) {
1161 // If the AddRec exists, consider it's register free and leave it alone.
1162 if (isExistingPhi(AR, SE))
1163 return;
1164
1165 // It is bad to allow LSR for current loop to add induction variables
1166 // for its sibling loops.
1167 if (!AR->getLoop()->contains(L)) {
1168 Lose();
1169 return;
1170 }
1171
1172 // Otherwise, it will be an invariant with respect to Loop L.
1173 ++NumRegs;
1174 return;
1175 }
1176 AddRecCost += 1; /// TODO: This should be a function of the stride.
1177
1178 // Add the step value register, if it needs one.
1179 // TODO: The non-affine case isn't precisely modeled here.
1180 if (!AR->isAffine() || !isa<SCEVConstant>(AR->getOperand(1))) {
1181 if (!Regs.count(AR->getOperand(1))) {
1182 RateRegister(AR->getOperand(1), Regs, L, SE, DT);
1183 if (isLoser())
1184 return;
1185 }
1186 }
1187 }
1188 ++NumRegs;
1189
1190 // Rough heuristic; favor registers which don't require extra setup
1191 // instructions in the preheader.
1192 if (!isa<SCEVUnknown>(Reg) &&
1193 !isa<SCEVConstant>(Reg) &&
1194 !(isa<SCEVAddRecExpr>(Reg) &&
1195 (isa<SCEVUnknown>(cast<SCEVAddRecExpr>(Reg)->getStart()) ||
1196 isa<SCEVConstant>(cast<SCEVAddRecExpr>(Reg)->getStart()))))
1197 ++SetupCost;
1198
1199 NumIVMuls += isa<SCEVMulExpr>(Reg) &&
1200 SE.hasComputableLoopEvolution(Reg, L);
1201}
1202
1203/// Record this register in the set. If we haven't seen it before, rate
1204/// it. Optional LoserRegs provides a way to declare any formula that refers to
1205/// one of those regs an instant loser.
1206void Cost::RatePrimaryRegister(const SCEV *Reg,
1207 SmallPtrSetImpl<const SCEV *> &Regs,
1208 const Loop *L,
1209 ScalarEvolution &SE, DominatorTree &DT,
1210 SmallPtrSetImpl<const SCEV *> *LoserRegs) {
1211 if (LoserRegs && LoserRegs->count(Reg)) {
1212 Lose();
1213 return;
1214 }
1215 if (Regs.insert(Reg).second) {
1216 RateRegister(Reg, Regs, L, SE, DT);
1217 if (LoserRegs && isLoser())
1218 LoserRegs->insert(Reg);
1219 }
1220}
1221
1222void Cost::RateFormula(const TargetTransformInfo &TTI,
1223 const Formula &F,
1224 SmallPtrSetImpl<const SCEV *> &Regs,
1225 const DenseSet<const SCEV *> &VisitedRegs,
1226 const Loop *L,
1227 ScalarEvolution &SE, DominatorTree &DT,
1228 const LSRUse &LU,
1229 SmallPtrSetImpl<const SCEV *> *LoserRegs) {
1230 assert(F.isCanonical(*L) && "Cost is accurate only for canonical formula")((F.isCanonical(*L) && "Cost is accurate only for canonical formula"
) ? static_cast<void> (0) : __assert_fail ("F.isCanonical(*L) && \"Cost is accurate only for canonical formula\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1230, __PRETTY_FUNCTION__))
;
1231 // Tally up the registers.
1232 unsigned PrevAddRecCost = AddRecCost;
1233 unsigned PrevNumRegs = NumRegs;
1234 unsigned PrevNumBaseAdds = NumBaseAdds;
1235 if (const SCEV *ScaledReg = F.ScaledReg) {
1236 if (VisitedRegs.count(ScaledReg)) {
1237 Lose();
1238 return;
1239 }
1240 RatePrimaryRegister(ScaledReg, Regs, L, SE, DT, LoserRegs);
1241 if (isLoser())
1242 return;
1243 }
1244 for (const SCEV *BaseReg : F.BaseRegs) {
1245 if (VisitedRegs.count(BaseReg)) {
1246 Lose();
1247 return;
1248 }
1249 RatePrimaryRegister(BaseReg, Regs, L, SE, DT, LoserRegs);
1250 if (isLoser())
1251 return;
1252 }
1253
1254 // Treat every new register that exceeds TTI.getNumberOfRegisters() - 1 as
1255 // additional instruction (at least fill).
1256 unsigned TTIRegNum = TTI.getNumberOfRegisters(false) - 1;
1257 if (NumRegs > TTIRegNum) {
1258 // Cost already exceeded TTIRegNum, then only newly added register can add
1259 // new instructions.
1260 if (PrevNumRegs > TTIRegNum)
1261 Insns += (NumRegs - PrevNumRegs);
1262 else
1263 Insns += (NumRegs - TTIRegNum);
1264 }
1265
1266 // Determine how many (unfolded) adds we'll need inside the loop.
1267 size_t NumBaseParts = F.getNumRegs();
1268 if (NumBaseParts > 1)
1269 // Do not count the base and a possible second register if the target
1270 // allows to fold 2 registers.
1271 NumBaseAdds +=
1272 NumBaseParts - (1 + (F.Scale && isAMCompletelyFolded(TTI, LU, F)));
1273 NumBaseAdds += (F.UnfoldedOffset != 0);
1274
1275 // Accumulate non-free scaling amounts.
1276 ScaleCost += getScalingFactorCost(TTI, LU, F, *L);
1277
1278 // Tally up the non-zero immediates.
1279 for (const LSRFixup &Fixup : LU.Fixups) {
1280 int64_t O = Fixup.Offset;
1281 int64_t Offset = (uint64_t)O + F.BaseOffset;
1282 if (F.BaseGV)
1283 ImmCost += 64; // Handle symbolic values conservatively.
1284 // TODO: This should probably be the pointer size.
1285 else if (Offset != 0)
1286 ImmCost += APInt(64, Offset, true).getMinSignedBits();
1287
1288 // Check with target if this offset with this instruction is
1289 // specifically not supported.
1290 if ((isa<LoadInst>(Fixup.UserInst) || isa<StoreInst>(Fixup.UserInst)) &&
1291 !TTI.isFoldableMemAccessOffset(Fixup.UserInst, Offset))
1292 NumBaseAdds++;
1293 }
1294
1295 // If ICmpZero formula ends with not 0, it could not be replaced by
1296 // just add or sub. We'll need to compare final result of AddRec.
1297 // That means we'll need an additional instruction.
1298 // For -10 + {0, +, 1}:
1299 // i = i + 1;
1300 // cmp i, 10
1301 //
1302 // For {-10, +, 1}:
1303 // i = i + 1;
1304 if (LU.Kind == LSRUse::ICmpZero && !F.hasZeroEnd())
1305 Insns++;
1306 // Each new AddRec adds 1 instruction to calculation.
1307 Insns += (AddRecCost - PrevAddRecCost);
1308
1309 // BaseAdds adds instructions for unfolded registers.
1310 if (LU.Kind != LSRUse::ICmpZero)
1311 Insns += NumBaseAdds - PrevNumBaseAdds;
1312 assert(isValid() && "invalid cost")((isValid() && "invalid cost") ? static_cast<void>
(0) : __assert_fail ("isValid() && \"invalid cost\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1312, __PRETTY_FUNCTION__))
;
1313}
1314
1315/// Set this cost to a losing value.
1316void Cost::Lose() {
1317 Insns = ~0u;
1318 NumRegs = ~0u;
1319 AddRecCost = ~0u;
1320 NumIVMuls = ~0u;
1321 NumBaseAdds = ~0u;
1322 ImmCost = ~0u;
1323 SetupCost = ~0u;
1324 ScaleCost = ~0u;
1325}
1326
1327/// Choose the lower cost.
1328bool Cost::operator<(const Cost &Other) const {
1329 if (InsnsCost && Insns != Other.Insns)
1330 return Insns < Other.Insns;
1331 return std::tie(NumRegs, AddRecCost, NumIVMuls, NumBaseAdds, ScaleCost,
1332 ImmCost, SetupCost) <
1333 std::tie(Other.NumRegs, Other.AddRecCost, Other.NumIVMuls,
1334 Other.NumBaseAdds, Other.ScaleCost, Other.ImmCost,
1335 Other.SetupCost);
1336}
1337
1338void Cost::print(raw_ostream &OS) const {
1339 OS << Insns << " instruction" << (Insns == 1 ? " " : "s ");
1340 OS << NumRegs << " reg" << (NumRegs == 1 ? "" : "s");
1341 if (AddRecCost != 0)
1342 OS << ", with addrec cost " << AddRecCost;
1343 if (NumIVMuls != 0)
1344 OS << ", plus " << NumIVMuls << " IV mul" << (NumIVMuls == 1 ? "" : "s");
1345 if (NumBaseAdds != 0)
1346 OS << ", plus " << NumBaseAdds << " base add"
1347 << (NumBaseAdds == 1 ? "" : "s");
1348 if (ScaleCost != 0)
1349 OS << ", plus " << ScaleCost << " scale cost";
1350 if (ImmCost != 0)
1351 OS << ", plus " << ImmCost << " imm cost";
1352 if (SetupCost != 0)
1353 OS << ", plus " << SetupCost << " setup cost";
1354}
1355
1356#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
1357LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void Cost::dump() const {
1358 print(errs()); errs() << '\n';
1359}
1360#endif
1361
1362LSRFixup::LSRFixup()
1363 : UserInst(nullptr), OperandValToReplace(nullptr),
1364 Offset(0) {}
1365
1366/// Test whether this fixup always uses its value outside of the given loop.
1367bool LSRFixup::isUseFullyOutsideLoop(const Loop *L) const {
1368 // PHI nodes use their value in their incoming blocks.
1369 if (const PHINode *PN = dyn_cast<PHINode>(UserInst)) {
1370 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
1371 if (PN->getIncomingValue(i) == OperandValToReplace &&
1372 L->contains(PN->getIncomingBlock(i)))
1373 return false;
1374 return true;
1375 }
1376
1377 return !L->contains(UserInst);
1378}
1379
1380void LSRFixup::print(raw_ostream &OS) const {
1381 OS << "UserInst=";
1382 // Store is common and interesting enough to be worth special-casing.
1383 if (StoreInst *Store = dyn_cast<StoreInst>(UserInst)) {
1384 OS << "store ";
1385 Store->getOperand(0)->printAsOperand(OS, /*PrintType=*/false);
1386 } else if (UserInst->getType()->isVoidTy())
1387 OS << UserInst->getOpcodeName();
1388 else
1389 UserInst->printAsOperand(OS, /*PrintType=*/false);
1390
1391 OS << ", OperandValToReplace=";
1392 OperandValToReplace->printAsOperand(OS, /*PrintType=*/false);
1393
1394 for (const Loop *PIL : PostIncLoops) {
1395 OS << ", PostIncLoop=";
1396 PIL->getHeader()->printAsOperand(OS, /*PrintType=*/false);
1397 }
1398
1399 if (Offset != 0)
1400 OS << ", Offset=" << Offset;
1401}
1402
1403#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
1404LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void LSRFixup::dump() const {
1405 print(errs()); errs() << '\n';
1406}
1407#endif
1408
1409/// Test whether this use as a formula which has the same registers as the given
1410/// formula.
1411bool LSRUse::HasFormulaWithSameRegs(const Formula &F) const {
1412 SmallVector<const SCEV *, 4> Key = F.BaseRegs;
1413 if (F.ScaledReg) Key.push_back(F.ScaledReg);
1414 // Unstable sort by host order ok, because this is only used for uniquifying.
1415 std::sort(Key.begin(), Key.end());
1416 return Uniquifier.count(Key);
1417}
1418
1419/// The function returns a probability of selecting formula without Reg.
1420float LSRUse::getNotSelectedProbability(const SCEV *Reg) const {
1421 unsigned FNum = 0;
1422 for (const Formula &F : Formulae)
1423 if (F.referencesReg(Reg))
1424 FNum++;
1425 return ((float)(Formulae.size() - FNum)) / Formulae.size();
1426}
1427
1428/// If the given formula has not yet been inserted, add it to the list, and
1429/// return true. Return false otherwise. The formula must be in canonical form.
1430bool LSRUse::InsertFormula(const Formula &F, const Loop &L) {
1431 assert(F.isCanonical(L) && "Invalid canonical representation")((F.isCanonical(L) && "Invalid canonical representation"
) ? static_cast<void> (0) : __assert_fail ("F.isCanonical(L) && \"Invalid canonical representation\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1431, __PRETTY_FUNCTION__))
;
1432
1433 if (!Formulae.empty() && RigidFormula)
1434 return false;
1435
1436 SmallVector<const SCEV *, 4> Key = F.BaseRegs;
1437 if (F.ScaledReg) Key.push_back(F.ScaledReg);
1438 // Unstable sort by host order ok, because this is only used for uniquifying.
1439 std::sort(Key.begin(), Key.end());
1440
1441 if (!Uniquifier.insert(Key).second)
1442 return false;
1443
1444 // Using a register to hold the value of 0 is not profitable.
1445 assert((!F.ScaledReg || !F.ScaledReg->isZero()) &&(((!F.ScaledReg || !F.ScaledReg->isZero()) && "Zero allocated in a scaled register!"
) ? static_cast<void> (0) : __assert_fail ("(!F.ScaledReg || !F.ScaledReg->isZero()) && \"Zero allocated in a scaled register!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1446, __PRETTY_FUNCTION__))
1446 "Zero allocated in a scaled register!")(((!F.ScaledReg || !F.ScaledReg->isZero()) && "Zero allocated in a scaled register!"
) ? static_cast<void> (0) : __assert_fail ("(!F.ScaledReg || !F.ScaledReg->isZero()) && \"Zero allocated in a scaled register!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1446, __PRETTY_FUNCTION__))
;
1447#ifndef NDEBUG
1448 for (const SCEV *BaseReg : F.BaseRegs)
1449 assert(!BaseReg->isZero() && "Zero allocated in a base register!")((!BaseReg->isZero() && "Zero allocated in a base register!"
) ? static_cast<void> (0) : __assert_fail ("!BaseReg->isZero() && \"Zero allocated in a base register!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1449, __PRETTY_FUNCTION__))
;
1450#endif
1451
1452 // Add the formula to the list.
1453 Formulae.push_back(F);
1454
1455 // Record registers now being used by this use.
1456 Regs.insert(F.BaseRegs.begin(), F.BaseRegs.end());
1457 if (F.ScaledReg)
1458 Regs.insert(F.ScaledReg);
1459
1460 return true;
1461}
1462
1463/// Remove the given formula from this use's list.
1464void LSRUse::DeleteFormula(Formula &F) {
1465 if (&F != &Formulae.back())
1466 std::swap(F, Formulae.back());
1467 Formulae.pop_back();
1468}
1469
1470/// Recompute the Regs field, and update RegUses.
1471void LSRUse::RecomputeRegs(size_t LUIdx, RegUseTracker &RegUses) {
1472 // Now that we've filtered out some formulae, recompute the Regs set.
1473 SmallPtrSet<const SCEV *, 4> OldRegs = std::move(Regs);
1474 Regs.clear();
1475 for (const Formula &F : Formulae) {
1476 if (F.ScaledReg) Regs.insert(F.ScaledReg);
1477 Regs.insert(F.BaseRegs.begin(), F.BaseRegs.end());
1478 }
1479
1480 // Update the RegTracker.
1481 for (const SCEV *S : OldRegs)
1482 if (!Regs.count(S))
1483 RegUses.dropRegister(S, LUIdx);
1484}
1485
1486void LSRUse::print(raw_ostream &OS) const {
1487 OS << "LSR Use: Kind=";
1488 switch (Kind) {
1489 case Basic: OS << "Basic"; break;
1490 case Special: OS << "Special"; break;
1491 case ICmpZero: OS << "ICmpZero"; break;
1492 case Address:
1493 OS << "Address of ";
1494 if (AccessTy.MemTy->isPointerTy())
1495 OS << "pointer"; // the full pointer type could be really verbose
1496 else {
1497 OS << *AccessTy.MemTy;
1498 }
1499
1500 OS << " in addrspace(" << AccessTy.AddrSpace << ')';
1501 }
1502
1503 OS << ", Offsets={";
1504 bool NeedComma = false;
1505 for (const LSRFixup &Fixup : Fixups) {
1506 if (NeedComma) OS << ',';
1507 OS << Fixup.Offset;
1508 NeedComma = true;
1509 }
1510 OS << '}';
1511
1512 if (AllFixupsOutsideLoop)
1513 OS << ", all-fixups-outside-loop";
1514
1515 if (WidestFixupType)
1516 OS << ", widest fixup type: " << *WidestFixupType;
1517}
1518
1519#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
1520LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void LSRUse::dump() const {
1521 print(errs()); errs() << '\n';
1522}
1523#endif
1524
1525static bool isAMCompletelyFolded(const TargetTransformInfo &TTI,
1526 LSRUse::KindType Kind, MemAccessTy AccessTy,
1527 GlobalValue *BaseGV, int64_t BaseOffset,
1528 bool HasBaseReg, int64_t Scale) {
1529 switch (Kind) {
1530 case LSRUse::Address:
1531 return TTI.isLegalAddressingMode(AccessTy.MemTy, BaseGV, BaseOffset,
1532 HasBaseReg, Scale, AccessTy.AddrSpace);
1533
1534 case LSRUse::ICmpZero:
1535 // There's not even a target hook for querying whether it would be legal to
1536 // fold a GV into an ICmp.
1537 if (BaseGV)
1538 return false;
1539
1540 // ICmp only has two operands; don't allow more than two non-trivial parts.
1541 if (Scale != 0 && HasBaseReg && BaseOffset != 0)
1542 return false;
1543
1544 // ICmp only supports no scale or a -1 scale, as we can "fold" a -1 scale by
1545 // putting the scaled register in the other operand of the icmp.
1546 if (Scale != 0 && Scale != -1)
1547 return false;
1548
1549 // If we have low-level target information, ask the target if it can fold an
1550 // integer immediate on an icmp.
1551 if (BaseOffset != 0) {
1552 // We have one of:
1553 // ICmpZero BaseReg + BaseOffset => ICmp BaseReg, -BaseOffset
1554 // ICmpZero -1*ScaleReg + BaseOffset => ICmp ScaleReg, BaseOffset
1555 // Offs is the ICmp immediate.
1556 if (Scale == 0)
1557 // The cast does the right thing with INT64_MIN.
1558 BaseOffset = -(uint64_t)BaseOffset;
1559 return TTI.isLegalICmpImmediate(BaseOffset);
1560 }
1561
1562 // ICmpZero BaseReg + -1*ScaleReg => ICmp BaseReg, ScaleReg
1563 return true;
1564
1565 case LSRUse::Basic:
1566 // Only handle single-register values.
1567 return !BaseGV && Scale == 0 && BaseOffset == 0;
1568
1569 case LSRUse::Special:
1570 // Special case Basic to handle -1 scales.
1571 return !BaseGV && (Scale == 0 || Scale == -1) && BaseOffset == 0;
1572 }
1573
1574 llvm_unreachable("Invalid LSRUse Kind!")::llvm::llvm_unreachable_internal("Invalid LSRUse Kind!", "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1574)
;
1575}
1576
1577static bool isAMCompletelyFolded(const TargetTransformInfo &TTI,
1578 int64_t MinOffset, int64_t MaxOffset,
1579 LSRUse::KindType Kind, MemAccessTy AccessTy,
1580 GlobalValue *BaseGV, int64_t BaseOffset,
1581 bool HasBaseReg, int64_t Scale) {
1582 // Check for overflow.
1583 if (((int64_t)((uint64_t)BaseOffset + MinOffset) > BaseOffset) !=
1584 (MinOffset > 0))
1585 return false;
1586 MinOffset = (uint64_t)BaseOffset + MinOffset;
1587 if (((int64_t)((uint64_t)BaseOffset + MaxOffset) > BaseOffset) !=
1588 (MaxOffset > 0))
1589 return false;
1590 MaxOffset = (uint64_t)BaseOffset + MaxOffset;
1591
1592 return isAMCompletelyFolded(TTI, Kind, AccessTy, BaseGV, MinOffset,
1593 HasBaseReg, Scale) &&
1594 isAMCompletelyFolded(TTI, Kind, AccessTy, BaseGV, MaxOffset,
1595 HasBaseReg, Scale);
1596}
1597
1598static bool isAMCompletelyFolded(const TargetTransformInfo &TTI,
1599 int64_t MinOffset, int64_t MaxOffset,
1600 LSRUse::KindType Kind, MemAccessTy AccessTy,
1601 const Formula &F, const Loop &L) {
1602 // For the purpose of isAMCompletelyFolded either having a canonical formula
1603 // or a scale not equal to zero is correct.
1604 // Problems may arise from non canonical formulae having a scale == 0.
1605 // Strictly speaking it would best to just rely on canonical formulae.
1606 // However, when we generate the scaled formulae, we first check that the
1607 // scaling factor is profitable before computing the actual ScaledReg for
1608 // compile time sake.
1609 assert((F.isCanonical(L) || F.Scale != 0))(((F.isCanonical(L) || F.Scale != 0)) ? static_cast<void>
(0) : __assert_fail ("(F.isCanonical(L) || F.Scale != 0)", "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1609, __PRETTY_FUNCTION__))
;
1610 return isAMCompletelyFolded(TTI, MinOffset, MaxOffset, Kind, AccessTy,
1611 F.BaseGV, F.BaseOffset, F.HasBaseReg, F.Scale);
1612}
1613
1614/// Test whether we know how to expand the current formula.
1615static bool isLegalUse(const TargetTransformInfo &TTI, int64_t MinOffset,
1616 int64_t MaxOffset, LSRUse::KindType Kind,
1617 MemAccessTy AccessTy, GlobalValue *BaseGV,
1618 int64_t BaseOffset, bool HasBaseReg, int64_t Scale) {
1619 // We know how to expand completely foldable formulae.
1620 return isAMCompletelyFolded(TTI, MinOffset, MaxOffset, Kind, AccessTy, BaseGV,
1621 BaseOffset, HasBaseReg, Scale) ||
1622 // Or formulae that use a base register produced by a sum of base
1623 // registers.
1624 (Scale == 1 &&
1625 isAMCompletelyFolded(TTI, MinOffset, MaxOffset, Kind, AccessTy,
1626 BaseGV, BaseOffset, true, 0));
1627}
1628
1629static bool isLegalUse(const TargetTransformInfo &TTI, int64_t MinOffset,
1630 int64_t MaxOffset, LSRUse::KindType Kind,
1631 MemAccessTy AccessTy, const Formula &F) {
1632 return isLegalUse(TTI, MinOffset, MaxOffset, Kind, AccessTy, F.BaseGV,
1633 F.BaseOffset, F.HasBaseReg, F.Scale);
1634}
1635
1636static bool isAMCompletelyFolded(const TargetTransformInfo &TTI,
1637 const LSRUse &LU, const Formula &F) {
1638 return isAMCompletelyFolded(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind,
1639 LU.AccessTy, F.BaseGV, F.BaseOffset, F.HasBaseReg,
1640 F.Scale);
1641}
1642
1643static unsigned getScalingFactorCost(const TargetTransformInfo &TTI,
1644 const LSRUse &LU, const Formula &F,
1645 const Loop &L) {
1646 if (!F.Scale)
1647 return 0;
1648
1649 // If the use is not completely folded in that instruction, we will have to
1650 // pay an extra cost only for scale != 1.
1651 if (!isAMCompletelyFolded(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind,
1652 LU.AccessTy, F, L))
1653 return F.Scale != 1;
1654
1655 switch (LU.Kind) {
1656 case LSRUse::Address: {
1657 // Check the scaling factor cost with both the min and max offsets.
1658 int ScaleCostMinOffset = TTI.getScalingFactorCost(
1659 LU.AccessTy.MemTy, F.BaseGV, F.BaseOffset + LU.MinOffset, F.HasBaseReg,
1660 F.Scale, LU.AccessTy.AddrSpace);
1661 int ScaleCostMaxOffset = TTI.getScalingFactorCost(
1662 LU.AccessTy.MemTy, F.BaseGV, F.BaseOffset + LU.MaxOffset, F.HasBaseReg,
1663 F.Scale, LU.AccessTy.AddrSpace);
1664
1665 assert(ScaleCostMinOffset >= 0 && ScaleCostMaxOffset >= 0 &&((ScaleCostMinOffset >= 0 && ScaleCostMaxOffset >=
0 && "Legal addressing mode has an illegal cost!") ?
static_cast<void> (0) : __assert_fail ("ScaleCostMinOffset >= 0 && ScaleCostMaxOffset >= 0 && \"Legal addressing mode has an illegal cost!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1666, __PRETTY_FUNCTION__))
1666 "Legal addressing mode has an illegal cost!")((ScaleCostMinOffset >= 0 && ScaleCostMaxOffset >=
0 && "Legal addressing mode has an illegal cost!") ?
static_cast<void> (0) : __assert_fail ("ScaleCostMinOffset >= 0 && ScaleCostMaxOffset >= 0 && \"Legal addressing mode has an illegal cost!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1666, __PRETTY_FUNCTION__))
;
1667 return std::max(ScaleCostMinOffset, ScaleCostMaxOffset);
1668 }
1669 case LSRUse::ICmpZero:
1670 case LSRUse::Basic:
1671 case LSRUse::Special:
1672 // The use is completely folded, i.e., everything is folded into the
1673 // instruction.
1674 return 0;
1675 }
1676
1677 llvm_unreachable("Invalid LSRUse Kind!")::llvm::llvm_unreachable_internal("Invalid LSRUse Kind!", "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1677)
;
1678}
1679
1680static bool isAlwaysFoldable(const TargetTransformInfo &TTI,
1681 LSRUse::KindType Kind, MemAccessTy AccessTy,
1682 GlobalValue *BaseGV, int64_t BaseOffset,
1683 bool HasBaseReg) {
1684 // Fast-path: zero is always foldable.
1685 if (BaseOffset == 0 && !BaseGV) return true;
1686
1687 // Conservatively, create an address with an immediate and a
1688 // base and a scale.
1689 int64_t Scale = Kind == LSRUse::ICmpZero ? -1 : 1;
1690
1691 // Canonicalize a scale of 1 to a base register if the formula doesn't
1692 // already have a base register.
1693 if (!HasBaseReg && Scale == 1) {
1694 Scale = 0;
1695 HasBaseReg = true;
1696 }
1697
1698 return isAMCompletelyFolded(TTI, Kind, AccessTy, BaseGV, BaseOffset,
1699 HasBaseReg, Scale);
1700}
1701
1702static bool isAlwaysFoldable(const TargetTransformInfo &TTI,
1703 ScalarEvolution &SE, int64_t MinOffset,
1704 int64_t MaxOffset, LSRUse::KindType Kind,
1705 MemAccessTy AccessTy, const SCEV *S,
1706 bool HasBaseReg) {
1707 // Fast-path: zero is always foldable.
1708 if (S->isZero()) return true;
1709
1710 // Conservatively, create an address with an immediate and a
1711 // base and a scale.
1712 int64_t BaseOffset = ExtractImmediate(S, SE);
1713 GlobalValue *BaseGV = ExtractSymbol(S, SE);
1714
1715 // If there's anything else involved, it's not foldable.
1716 if (!S->isZero()) return false;
1717
1718 // Fast-path: zero is always foldable.
1719 if (BaseOffset == 0 && !BaseGV) return true;
1720
1721 // Conservatively, create an address with an immediate and a
1722 // base and a scale.
1723 int64_t Scale = Kind == LSRUse::ICmpZero ? -1 : 1;
1724
1725 return isAMCompletelyFolded(TTI, MinOffset, MaxOffset, Kind, AccessTy, BaseGV,
1726 BaseOffset, HasBaseReg, Scale);
1727}
1728
1729namespace {
1730
1731/// An individual increment in a Chain of IV increments. Relate an IV user to
1732/// an expression that computes the IV it uses from the IV used by the previous
1733/// link in the Chain.
1734///
1735/// For the head of a chain, IncExpr holds the absolute SCEV expression for the
1736/// original IVOperand. The head of the chain's IVOperand is only valid during
1737/// chain collection, before LSR replaces IV users. During chain generation,
1738/// IncExpr can be used to find the new IVOperand that computes the same
1739/// expression.
1740struct IVInc {
1741 Instruction *UserInst;
1742 Value* IVOperand;
1743 const SCEV *IncExpr;
1744
1745 IVInc(Instruction *U, Value *O, const SCEV *E):
1746 UserInst(U), IVOperand(O), IncExpr(E) {}
1747};
1748
1749// The list of IV increments in program order. We typically add the head of a
1750// chain without finding subsequent links.
1751struct IVChain {
1752 SmallVector<IVInc,1> Incs;
1753 const SCEV *ExprBase;
1754
1755 IVChain() : ExprBase(nullptr) {}
1756
1757 IVChain(const IVInc &Head, const SCEV *Base)
1758 : Incs(1, Head), ExprBase(Base) {}
1759
1760 typedef SmallVectorImpl<IVInc>::const_iterator const_iterator;
1761
1762 // Return the first increment in the chain.
1763 const_iterator begin() const {
1764 assert(!Incs.empty())((!Incs.empty()) ? static_cast<void> (0) : __assert_fail
("!Incs.empty()", "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 1764, __PRETTY_FUNCTION__))
;
1765 return std::next(Incs.begin());
1766 }
1767 const_iterator end() const {
1768 return Incs.end();
1769 }
1770
1771 // Returns true if this chain contains any increments.
1772 bool hasIncs() const { return Incs.size() >= 2; }
1773
1774 // Add an IVInc to the end of this chain.
1775 void add(const IVInc &X) { Incs.push_back(X); }
1776
1777 // Returns the last UserInst in the chain.
1778 Instruction *tailUserInst() const { return Incs.back().UserInst; }
1779
1780 // Returns true if IncExpr can be profitably added to this chain.
1781 bool isProfitableIncrement(const SCEV *OperExpr,
1782 const SCEV *IncExpr,
1783 ScalarEvolution&);
1784};
1785
1786/// Helper for CollectChains to track multiple IV increment uses. Distinguish
1787/// between FarUsers that definitely cross IV increments and NearUsers that may
1788/// be used between IV increments.
1789struct ChainUsers {
1790 SmallPtrSet<Instruction*, 4> FarUsers;
1791 SmallPtrSet<Instruction*, 4> NearUsers;
1792};
1793
1794/// This class holds state for the main loop strength reduction logic.
1795class LSRInstance {
1796 IVUsers &IU;
1797 ScalarEvolution &SE;
1798 DominatorTree &DT;
1799 LoopInfo &LI;
1800 const TargetTransformInfo &TTI;
1801 Loop *const L;
1802 bool Changed;
1803
1804 /// This is the insert position that the current loop's induction variable
1805 /// increment should be placed. In simple loops, this is the latch block's
1806 /// terminator. But in more complicated cases, this is a position which will
1807 /// dominate all the in-loop post-increment users.
1808 Instruction *IVIncInsertPos;
1809
1810 /// Interesting factors between use strides.
1811 ///
1812 /// We explicitly use a SetVector which contains a SmallSet, instead of the
1813 /// default, a SmallDenseSet, because we need to use the full range of
1814 /// int64_ts, and there's currently no good way of doing that with
1815 /// SmallDenseSet.
1816 SetVector<int64_t, SmallVector<int64_t, 8>, SmallSet<int64_t, 8>> Factors;
1817
1818 /// Interesting use types, to facilitate truncation reuse.
1819 SmallSetVector<Type *, 4> Types;
1820
1821 /// The list of interesting uses.
1822 SmallVector<LSRUse, 16> Uses;
1823
1824 /// Track which uses use which register candidates.
1825 RegUseTracker RegUses;
1826
1827 // Limit the number of chains to avoid quadratic behavior. We don't expect to
1828 // have more than a few IV increment chains in a loop. Missing a Chain falls
1829 // back to normal LSR behavior for those uses.
1830 static const unsigned MaxChains = 8;
1831
1832 /// IV users can form a chain of IV increments.
1833 SmallVector<IVChain, MaxChains> IVChainVec;
1834
1835 /// IV users that belong to profitable IVChains.
1836 SmallPtrSet<Use*, MaxChains> IVIncSet;
1837
1838 void OptimizeShadowIV();
1839 bool FindIVUserForCond(ICmpInst *Cond, IVStrideUse *&CondUse);
1840 ICmpInst *OptimizeMax(ICmpInst *Cond, IVStrideUse* &CondUse);
1841 void OptimizeLoopTermCond();
1842
1843 void ChainInstruction(Instruction *UserInst, Instruction *IVOper,
1844 SmallVectorImpl<ChainUsers> &ChainUsersVec);
1845 void FinalizeChain(IVChain &Chain);
1846 void CollectChains();
1847 void GenerateIVChain(const IVChain &Chain, SCEVExpander &Rewriter,
1848 SmallVectorImpl<WeakTrackingVH> &DeadInsts);
1849
1850 void CollectInterestingTypesAndFactors();
1851 void CollectFixupsAndInitialFormulae();
1852
1853 // Support for sharing of LSRUses between LSRFixups.
1854 typedef DenseMap<LSRUse::SCEVUseKindPair, size_t> UseMapTy;
1855 UseMapTy UseMap;
1856
1857 bool reconcileNewOffset(LSRUse &LU, int64_t NewOffset, bool HasBaseReg,
1858 LSRUse::KindType Kind, MemAccessTy AccessTy);
1859
1860 std::pair<size_t, int64_t> getUse(const SCEV *&Expr, LSRUse::KindType Kind,
1861 MemAccessTy AccessTy);
1862
1863 void DeleteUse(LSRUse &LU, size_t LUIdx);
1864
1865 LSRUse *FindUseWithSimilarFormula(const Formula &F, const LSRUse &OrigLU);
1866
1867 void InsertInitialFormula(const SCEV *S, LSRUse &LU, size_t LUIdx);
1868 void InsertSupplementalFormula(const SCEV *S, LSRUse &LU, size_t LUIdx);
1869 void CountRegisters(const Formula &F, size_t LUIdx);
1870 bool InsertFormula(LSRUse &LU, unsigned LUIdx, const Formula &F);
1871
1872 void CollectLoopInvariantFixupsAndFormulae();
1873
1874 void GenerateReassociations(LSRUse &LU, unsigned LUIdx, Formula Base,
1875 unsigned Depth = 0);
1876
1877 void GenerateReassociationsImpl(LSRUse &LU, unsigned LUIdx,
1878 const Formula &Base, unsigned Depth,
1879 size_t Idx, bool IsScaledReg = false);
1880 void GenerateCombinations(LSRUse &LU, unsigned LUIdx, Formula Base);
1881 void GenerateSymbolicOffsetsImpl(LSRUse &LU, unsigned LUIdx,
1882 const Formula &Base, size_t Idx,
1883 bool IsScaledReg = false);
1884 void GenerateSymbolicOffsets(LSRUse &LU, unsigned LUIdx, Formula Base);
1885 void GenerateConstantOffsetsImpl(LSRUse &LU, unsigned LUIdx,
1886 const Formula &Base,
1887 const SmallVectorImpl<int64_t> &Worklist,
1888 size_t Idx, bool IsScaledReg = false);
1889 void GenerateConstantOffsets(LSRUse &LU, unsigned LUIdx, Formula Base);
1890 void GenerateICmpZeroScales(LSRUse &LU, unsigned LUIdx, Formula Base);
1891 void GenerateScales(LSRUse &LU, unsigned LUIdx, Formula Base);
1892 void GenerateTruncates(LSRUse &LU, unsigned LUIdx, Formula Base);
1893 void GenerateCrossUseConstantOffsets();
1894 void GenerateAllReuseFormulae();
1895
1896 void FilterOutUndesirableDedicatedRegisters();
1897
1898 size_t EstimateSearchSpaceComplexity() const;
1899 void NarrowSearchSpaceByDetectingSupersets();
1900 void NarrowSearchSpaceByCollapsingUnrolledCode();
1901 void NarrowSearchSpaceByRefilteringUndesirableDedicatedRegisters();
1902 void NarrowSearchSpaceByDeletingCostlyFormulas();
1903 void NarrowSearchSpaceByPickingWinnerRegs();
1904 void NarrowSearchSpaceUsingHeuristics();
1905
1906 void SolveRecurse(SmallVectorImpl<const Formula *> &Solution,
1907 Cost &SolutionCost,
1908 SmallVectorImpl<const Formula *> &Workspace,
1909 const Cost &CurCost,
1910 const SmallPtrSet<const SCEV *, 16> &CurRegs,
1911 DenseSet<const SCEV *> &VisitedRegs) const;
1912 void Solve(SmallVectorImpl<const Formula *> &Solution) const;
1913
1914 BasicBlock::iterator
1915 HoistInsertPosition(BasicBlock::iterator IP,
1916 const SmallVectorImpl<Instruction *> &Inputs) const;
1917 BasicBlock::iterator
1918 AdjustInsertPositionForExpand(BasicBlock::iterator IP,
1919 const LSRFixup &LF,
1920 const LSRUse &LU,
1921 SCEVExpander &Rewriter) const;
1922
1923 Value *Expand(const LSRUse &LU, const LSRFixup &LF, const Formula &F,
1924 BasicBlock::iterator IP, SCEVExpander &Rewriter,
1925 SmallVectorImpl<WeakTrackingVH> &DeadInsts) const;
1926 void RewriteForPHI(PHINode *PN, const LSRUse &LU, const LSRFixup &LF,
1927 const Formula &F, SCEVExpander &Rewriter,
1928 SmallVectorImpl<WeakTrackingVH> &DeadInsts) const;
1929 void Rewrite(const LSRUse &LU, const LSRFixup &LF, const Formula &F,
1930 SCEVExpander &Rewriter,
1931 SmallVectorImpl<WeakTrackingVH> &DeadInsts) const;
1932 void ImplementSolution(const SmallVectorImpl<const Formula *> &Solution);
1933
1934public:
1935 LSRInstance(Loop *L, IVUsers &IU, ScalarEvolution &SE, DominatorTree &DT,
1936 LoopInfo &LI, const TargetTransformInfo &TTI);
1937
1938 bool getChanged() const { return Changed; }
1939
1940 void print_factors_and_types(raw_ostream &OS) const;
1941 void print_fixups(raw_ostream &OS) const;
1942 void print_uses(raw_ostream &OS) const;
1943 void print(raw_ostream &OS) const;
1944 void dump() const;
1945};
1946
1947} // end anonymous namespace
1948
1949/// If IV is used in a int-to-float cast inside the loop then try to eliminate
1950/// the cast operation.
1951void LSRInstance::OptimizeShadowIV() {
1952 const SCEV *BackedgeTakenCount = SE.getBackedgeTakenCount(L);
1953 if (isa<SCEVCouldNotCompute>(BackedgeTakenCount))
1954 return;
1955
1956 for (IVUsers::const_iterator UI = IU.begin(), E = IU.end();
1957 UI != E; /* empty */) {
1958 IVUsers::const_iterator CandidateUI = UI;
1959 ++UI;
1960 Instruction *ShadowUse = CandidateUI->getUser();
1961 Type *DestTy = nullptr;
1962 bool IsSigned = false;
1963
1964 /* If shadow use is a int->float cast then insert a second IV
1965 to eliminate this cast.
1966
1967 for (unsigned i = 0; i < n; ++i)
1968 foo((double)i);
1969
1970 is transformed into
1971
1972 double d = 0.0;
1973 for (unsigned i = 0; i < n; ++i, ++d)
1974 foo(d);
1975 */
1976 if (UIToFPInst *UCast = dyn_cast<UIToFPInst>(CandidateUI->getUser())) {
1977 IsSigned = false;
1978 DestTy = UCast->getDestTy();
1979 }
1980 else if (SIToFPInst *SCast = dyn_cast<SIToFPInst>(CandidateUI->getUser())) {
1981 IsSigned = true;
1982 DestTy = SCast->getDestTy();
1983 }
1984 if (!DestTy) continue;
1985
1986 // If target does not support DestTy natively then do not apply
1987 // this transformation.
1988 if (!TTI.isTypeLegal(DestTy)) continue;
1989
1990 PHINode *PH = dyn_cast<PHINode>(ShadowUse->getOperand(0));
1991 if (!PH) continue;
1992 if (PH->getNumIncomingValues() != 2) continue;
1993
1994 Type *SrcTy = PH->getType();
1995 int Mantissa = DestTy->getFPMantissaWidth();
1996 if (Mantissa == -1) continue;
1997 if ((int)SE.getTypeSizeInBits(SrcTy) > Mantissa)
1998 continue;
1999
2000 unsigned Entry, Latch;
2001 if (PH->getIncomingBlock(0) == L->getLoopPreheader()) {
2002 Entry = 0;
2003 Latch = 1;
2004 } else {
2005 Entry = 1;
2006 Latch = 0;
2007 }
2008
2009 ConstantInt *Init = dyn_cast<ConstantInt>(PH->getIncomingValue(Entry));
2010 if (!Init) continue;
2011 Constant *NewInit = ConstantFP::get(DestTy, IsSigned ?
2012 (double)Init->getSExtValue() :
2013 (double)Init->getZExtValue());
2014
2015 BinaryOperator *Incr =
2016 dyn_cast<BinaryOperator>(PH->getIncomingValue(Latch));
2017 if (!Incr) continue;
2018 if (Incr->getOpcode() != Instruction::Add
2019 && Incr->getOpcode() != Instruction::Sub)
2020 continue;
2021
2022 /* Initialize new IV, double d = 0.0 in above example. */
2023 ConstantInt *C = nullptr;
2024 if (Incr->getOperand(0) == PH)
2025 C = dyn_cast<ConstantInt>(Incr->getOperand(1));
2026 else if (Incr->getOperand(1) == PH)
2027 C = dyn_cast<ConstantInt>(Incr->getOperand(0));
2028 else
2029 continue;
2030
2031 if (!C) continue;
2032
2033 // Ignore negative constants, as the code below doesn't handle them
2034 // correctly. TODO: Remove this restriction.
2035 if (!C->getValue().isStrictlyPositive()) continue;
2036
2037 /* Add new PHINode. */
2038 PHINode *NewPH = PHINode::Create(DestTy, 2, "IV.S.", PH);
2039
2040 /* create new increment. '++d' in above example. */
2041 Constant *CFP = ConstantFP::get(DestTy, C->getZExtValue());
2042 BinaryOperator *NewIncr =
2043 BinaryOperator::Create(Incr->getOpcode() == Instruction::Add ?
2044 Instruction::FAdd : Instruction::FSub,
2045 NewPH, CFP, "IV.S.next.", Incr);
2046
2047 NewPH->addIncoming(NewInit, PH->getIncomingBlock(Entry));
2048 NewPH->addIncoming(NewIncr, PH->getIncomingBlock(Latch));
2049
2050 /* Remove cast operation */
2051 ShadowUse->replaceAllUsesWith(NewPH);
2052 ShadowUse->eraseFromParent();
2053 Changed = true;
2054 break;
2055 }
2056}
2057
2058/// If Cond has an operand that is an expression of an IV, set the IV user and
2059/// stride information and return true, otherwise return false.
2060bool LSRInstance::FindIVUserForCond(ICmpInst *Cond, IVStrideUse *&CondUse) {
2061 for (IVStrideUse &U : IU)
2062 if (U.getUser() == Cond) {
2063 // NOTE: we could handle setcc instructions with multiple uses here, but
2064 // InstCombine does it as well for simple uses, it's not clear that it
2065 // occurs enough in real life to handle.
2066 CondUse = &U;
2067 return true;
2068 }
2069 return false;
2070}
2071
2072/// Rewrite the loop's terminating condition if it uses a max computation.
2073///
2074/// This is a narrow solution to a specific, but acute, problem. For loops
2075/// like this:
2076///
2077/// i = 0;
2078/// do {
2079/// p[i] = 0.0;
2080/// } while (++i < n);
2081///
2082/// the trip count isn't just 'n', because 'n' might not be positive. And
2083/// unfortunately this can come up even for loops where the user didn't use
2084/// a C do-while loop. For example, seemingly well-behaved top-test loops
2085/// will commonly be lowered like this:
2086//
2087/// if (n > 0) {
2088/// i = 0;
2089/// do {
2090/// p[i] = 0.0;
2091/// } while (++i < n);
2092/// }
2093///
2094/// and then it's possible for subsequent optimization to obscure the if
2095/// test in such a way that indvars can't find it.
2096///
2097/// When indvars can't find the if test in loops like this, it creates a
2098/// max expression, which allows it to give the loop a canonical
2099/// induction variable:
2100///
2101/// i = 0;
2102/// max = n < 1 ? 1 : n;
2103/// do {
2104/// p[i] = 0.0;
2105/// } while (++i != max);
2106///
2107/// Canonical induction variables are necessary because the loop passes
2108/// are designed around them. The most obvious example of this is the
2109/// LoopInfo analysis, which doesn't remember trip count values. It
2110/// expects to be able to rediscover the trip count each time it is
2111/// needed, and it does this using a simple analysis that only succeeds if
2112/// the loop has a canonical induction variable.
2113///
2114/// However, when it comes time to generate code, the maximum operation
2115/// can be quite costly, especially if it's inside of an outer loop.
2116///
2117/// This function solves this problem by detecting this type of loop and
2118/// rewriting their conditions from ICMP_NE back to ICMP_SLT, and deleting
2119/// the instructions for the maximum computation.
2120///
2121ICmpInst *LSRInstance::OptimizeMax(ICmpInst *Cond, IVStrideUse* &CondUse) {
2122 // Check that the loop matches the pattern we're looking for.
2123 if (Cond->getPredicate() != CmpInst::ICMP_EQ &&
2124 Cond->getPredicate() != CmpInst::ICMP_NE)
2125 return Cond;
2126
2127 SelectInst *Sel = dyn_cast<SelectInst>(Cond->getOperand(1));
2128 if (!Sel || !Sel->hasOneUse()) return Cond;
2129
2130 const SCEV *BackedgeTakenCount = SE.getBackedgeTakenCount(L);
2131 if (isa<SCEVCouldNotCompute>(BackedgeTakenCount))
2132 return Cond;
2133 const SCEV *One = SE.getConstant(BackedgeTakenCount->getType(), 1);
2134
2135 // Add one to the backedge-taken count to get the trip count.
2136 const SCEV *IterationCount = SE.getAddExpr(One, BackedgeTakenCount);
2137 if (IterationCount != SE.getSCEV(Sel)) return Cond;
2138
2139 // Check for a max calculation that matches the pattern. There's no check
2140 // for ICMP_ULE here because the comparison would be with zero, which
2141 // isn't interesting.
2142 CmpInst::Predicate Pred = ICmpInst::BAD_ICMP_PREDICATE;
2143 const SCEVNAryExpr *Max = nullptr;
2144 if (const SCEVSMaxExpr *S = dyn_cast<SCEVSMaxExpr>(BackedgeTakenCount)) {
2145 Pred = ICmpInst::ICMP_SLE;
2146 Max = S;
2147 } else if (const SCEVSMaxExpr *S = dyn_cast<SCEVSMaxExpr>(IterationCount)) {
2148 Pred = ICmpInst::ICMP_SLT;
2149 Max = S;
2150 } else if (const SCEVUMaxExpr *U = dyn_cast<SCEVUMaxExpr>(IterationCount)) {
2151 Pred = ICmpInst::ICMP_ULT;
2152 Max = U;
2153 } else {
2154 // No match; bail.
2155 return Cond;
2156 }
2157
2158 // To handle a max with more than two operands, this optimization would
2159 // require additional checking and setup.
2160 if (Max->getNumOperands() != 2)
2161 return Cond;
2162
2163 const SCEV *MaxLHS = Max->getOperand(0);
2164 const SCEV *MaxRHS = Max->getOperand(1);
2165
2166 // ScalarEvolution canonicalizes constants to the left. For < and >, look
2167 // for a comparison with 1. For <= and >=, a comparison with zero.
2168 if (!MaxLHS ||
2169 (ICmpInst::isTrueWhenEqual(Pred) ? !MaxLHS->isZero() : (MaxLHS != One)))
2170 return Cond;
2171
2172 // Check the relevant induction variable for conformance to
2173 // the pattern.
2174 const SCEV *IV = SE.getSCEV(Cond->getOperand(0));
2175 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(IV);
2176 if (!AR || !AR->isAffine() ||
2177 AR->getStart() != One ||
2178 AR->getStepRecurrence(SE) != One)
2179 return Cond;
2180
2181 assert(AR->getLoop() == L &&((AR->getLoop() == L && "Loop condition operand is an addrec in a different loop!"
) ? static_cast<void> (0) : __assert_fail ("AR->getLoop() == L && \"Loop condition operand is an addrec in a different loop!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 2182, __PRETTY_FUNCTION__))
2182 "Loop condition operand is an addrec in a different loop!")((AR->getLoop() == L && "Loop condition operand is an addrec in a different loop!"
) ? static_cast<void> (0) : __assert_fail ("AR->getLoop() == L && \"Loop condition operand is an addrec in a different loop!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 2182, __PRETTY_FUNCTION__))
;
2183
2184 // Check the right operand of the select, and remember it, as it will
2185 // be used in the new comparison instruction.
2186 Value *NewRHS = nullptr;
2187 if (ICmpInst::isTrueWhenEqual(Pred)) {
2188 // Look for n+1, and grab n.
2189 if (AddOperator *BO = dyn_cast<AddOperator>(Sel->getOperand(1)))
2190 if (ConstantInt *BO1 = dyn_cast<ConstantInt>(BO->getOperand(1)))
2191 if (BO1->isOne() && SE.getSCEV(BO->getOperand(0)) == MaxRHS)
2192 NewRHS = BO->getOperand(0);
2193 if (AddOperator *BO = dyn_cast<AddOperator>(Sel->getOperand(2)))
2194 if (ConstantInt *BO1 = dyn_cast<ConstantInt>(BO->getOperand(1)))
2195 if (BO1->isOne() && SE.getSCEV(BO->getOperand(0)) == MaxRHS)
2196 NewRHS = BO->getOperand(0);
2197 if (!NewRHS)
2198 return Cond;
2199 } else if (SE.getSCEV(Sel->getOperand(1)) == MaxRHS)
2200 NewRHS = Sel->getOperand(1);
2201 else if (SE.getSCEV(Sel->getOperand(2)) == MaxRHS)
2202 NewRHS = Sel->getOperand(2);
2203 else if (const SCEVUnknown *SU = dyn_cast<SCEVUnknown>(MaxRHS))
2204 NewRHS = SU->getValue();
2205 else
2206 // Max doesn't match expected pattern.
2207 return Cond;
2208
2209 // Determine the new comparison opcode. It may be signed or unsigned,
2210 // and the original comparison may be either equality or inequality.
2211 if (Cond->getPredicate() == CmpInst::ICMP_EQ)
2212 Pred = CmpInst::getInversePredicate(Pred);
2213
2214 // Ok, everything looks ok to change the condition into an SLT or SGE and
2215 // delete the max calculation.
2216 ICmpInst *NewCond =
2217 new ICmpInst(Cond, Pred, Cond->getOperand(0), NewRHS, "scmp");
2218
2219 // Delete the max calculation instructions.
2220 Cond->replaceAllUsesWith(NewCond);
2221 CondUse->setUser(NewCond);
2222 Instruction *Cmp = cast<Instruction>(Sel->getOperand(0));
2223 Cond->eraseFromParent();
2224 Sel->eraseFromParent();
2225 if (Cmp->use_empty())
2226 Cmp->eraseFromParent();
2227 return NewCond;
2228}
2229
2230/// Change loop terminating condition to use the postinc iv when possible.
2231void
2232LSRInstance::OptimizeLoopTermCond() {
2233 SmallPtrSet<Instruction *, 4> PostIncs;
2234
2235 // We need a different set of heuristics for rotated and non-rotated loops.
2236 // If a loop is rotated then the latch is also the backedge, so inserting
2237 // post-inc expressions just before the latch is ideal. To reduce live ranges
2238 // it also makes sense to rewrite terminating conditions to use post-inc
2239 // expressions.
2240 //
2241 // If the loop is not rotated then the latch is not a backedge; the latch
2242 // check is done in the loop head. Adding post-inc expressions before the
2243 // latch will cause overlapping live-ranges of pre-inc and post-inc expressions
2244 // in the loop body. In this case we do *not* want to use post-inc expressions
2245 // in the latch check, and we want to insert post-inc expressions before
2246 // the backedge.
2247 BasicBlock *LatchBlock = L->getLoopLatch();
2248 SmallVector<BasicBlock*, 8> ExitingBlocks;
2249 L->getExitingBlocks(ExitingBlocks);
2250 if (llvm::all_of(ExitingBlocks, [&LatchBlock](const BasicBlock *BB) {
2251 return LatchBlock != BB;
2252 })) {
2253 // The backedge doesn't exit the loop; treat this as a head-tested loop.
2254 IVIncInsertPos = LatchBlock->getTerminator();
2255 return;
2256 }
2257
2258 // Otherwise treat this as a rotated loop.
2259 for (BasicBlock *ExitingBlock : ExitingBlocks) {
2260
2261 // Get the terminating condition for the loop if possible. If we
2262 // can, we want to change it to use a post-incremented version of its
2263 // induction variable, to allow coalescing the live ranges for the IV into
2264 // one register value.
2265
2266 BranchInst *TermBr = dyn_cast<BranchInst>(ExitingBlock->getTerminator());
2267 if (!TermBr)
2268 continue;
2269 // FIXME: Overly conservative, termination condition could be an 'or' etc..
2270 if (TermBr->isUnconditional() || !isa<ICmpInst>(TermBr->getCondition()))
2271 continue;
2272
2273 // Search IVUsesByStride to find Cond's IVUse if there is one.
2274 IVStrideUse *CondUse = nullptr;
2275 ICmpInst *Cond = cast<ICmpInst>(TermBr->getCondition());
2276 if (!FindIVUserForCond(Cond, CondUse))
2277 continue;
2278
2279 // If the trip count is computed in terms of a max (due to ScalarEvolution
2280 // being unable to find a sufficient guard, for example), change the loop
2281 // comparison to use SLT or ULT instead of NE.
2282 // One consequence of doing this now is that it disrupts the count-down
2283 // optimization. That's not always a bad thing though, because in such
2284 // cases it may still be worthwhile to avoid a max.
2285 Cond = OptimizeMax(Cond, CondUse);
2286
2287 // If this exiting block dominates the latch block, it may also use
2288 // the post-inc value if it won't be shared with other uses.
2289 // Check for dominance.
2290 if (!DT.dominates(ExitingBlock, LatchBlock))
2291 continue;
2292
2293 // Conservatively avoid trying to use the post-inc value in non-latch
2294 // exits if there may be pre-inc users in intervening blocks.
2295 if (LatchBlock != ExitingBlock)
2296 for (IVUsers::const_iterator UI = IU.begin(), E = IU.end(); UI != E; ++UI)
2297 // Test if the use is reachable from the exiting block. This dominator
2298 // query is a conservative approximation of reachability.
2299 if (&*UI != CondUse &&
2300 !DT.properlyDominates(UI->getUser()->getParent(), ExitingBlock)) {
2301 // Conservatively assume there may be reuse if the quotient of their
2302 // strides could be a legal scale.
2303 const SCEV *A = IU.getStride(*CondUse, L);
2304 const SCEV *B = IU.getStride(*UI, L);
2305 if (!A || !B) continue;
2306 if (SE.getTypeSizeInBits(A->getType()) !=
2307 SE.getTypeSizeInBits(B->getType())) {
2308 if (SE.getTypeSizeInBits(A->getType()) >
2309 SE.getTypeSizeInBits(B->getType()))
2310 B = SE.getSignExtendExpr(B, A->getType());
2311 else
2312 A = SE.getSignExtendExpr(A, B->getType());
2313 }
2314 if (const SCEVConstant *D =
2315 dyn_cast_or_null<SCEVConstant>(getExactSDiv(B, A, SE))) {
2316 const ConstantInt *C = D->getValue();
2317 // Stride of one or negative one can have reuse with non-addresses.
2318 if (C->isOne() || C->isAllOnesValue())
2319 goto decline_post_inc;
2320 // Avoid weird situations.
2321 if (C->getValue().getMinSignedBits() >= 64 ||
2322 C->getValue().isMinSignedValue())
2323 goto decline_post_inc;
2324 // Check for possible scaled-address reuse.
2325 MemAccessTy AccessTy = getAccessType(UI->getUser());
2326 int64_t Scale = C->getSExtValue();
2327 if (TTI.isLegalAddressingMode(AccessTy.MemTy, /*BaseGV=*/nullptr,
2328 /*BaseOffset=*/0,
2329 /*HasBaseReg=*/false, Scale,
2330 AccessTy.AddrSpace))
2331 goto decline_post_inc;
2332 Scale = -Scale;
2333 if (TTI.isLegalAddressingMode(AccessTy.MemTy, /*BaseGV=*/nullptr,
2334 /*BaseOffset=*/0,
2335 /*HasBaseReg=*/false, Scale,
2336 AccessTy.AddrSpace))
2337 goto decline_post_inc;
2338 }
2339 }
2340
2341 DEBUG(dbgs() << " Change loop exiting icmp to use postinc iv: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Change loop exiting icmp to use postinc iv: "
<< *Cond << '\n'; } } while (false)
2342 << *Cond << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Change loop exiting icmp to use postinc iv: "
<< *Cond << '\n'; } } while (false)
;
2343
2344 // It's possible for the setcc instruction to be anywhere in the loop, and
2345 // possible for it to have multiple users. If it is not immediately before
2346 // the exiting block branch, move it.
2347 if (&*++BasicBlock::iterator(Cond) != TermBr) {
2348 if (Cond->hasOneUse()) {
2349 Cond->moveBefore(TermBr);
2350 } else {
2351 // Clone the terminating condition and insert into the loopend.
2352 ICmpInst *OldCond = Cond;
2353 Cond = cast<ICmpInst>(Cond->clone());
2354 Cond->setName(L->getHeader()->getName() + ".termcond");
2355 ExitingBlock->getInstList().insert(TermBr->getIterator(), Cond);
2356
2357 // Clone the IVUse, as the old use still exists!
2358 CondUse = &IU.AddUser(Cond, CondUse->getOperandValToReplace());
2359 TermBr->replaceUsesOfWith(OldCond, Cond);
2360 }
2361 }
2362
2363 // If we get to here, we know that we can transform the setcc instruction to
2364 // use the post-incremented version of the IV, allowing us to coalesce the
2365 // live ranges for the IV correctly.
2366 CondUse->transformToPostInc(L);
2367 Changed = true;
2368
2369 PostIncs.insert(Cond);
2370 decline_post_inc:;
2371 }
2372
2373 // Determine an insertion point for the loop induction variable increment. It
2374 // must dominate all the post-inc comparisons we just set up, and it must
2375 // dominate the loop latch edge.
2376 IVIncInsertPos = L->getLoopLatch()->getTerminator();
2377 for (Instruction *Inst : PostIncs) {
2378 BasicBlock *BB =
2379 DT.findNearestCommonDominator(IVIncInsertPos->getParent(),
2380 Inst->getParent());
2381 if (BB == Inst->getParent())
2382 IVIncInsertPos = Inst;
2383 else if (BB != IVIncInsertPos->getParent())
2384 IVIncInsertPos = BB->getTerminator();
2385 }
2386}
2387
2388/// Determine if the given use can accommodate a fixup at the given offset and
2389/// other details. If so, update the use and return true.
2390bool LSRInstance::reconcileNewOffset(LSRUse &LU, int64_t NewOffset,
2391 bool HasBaseReg, LSRUse::KindType Kind,
2392 MemAccessTy AccessTy) {
2393 int64_t NewMinOffset = LU.MinOffset;
2394 int64_t NewMaxOffset = LU.MaxOffset;
2395 MemAccessTy NewAccessTy = AccessTy;
2396
2397 // Check for a mismatched kind. It's tempting to collapse mismatched kinds to
2398 // something conservative, however this can pessimize in the case that one of
2399 // the uses will have all its uses outside the loop, for example.
2400 if (LU.Kind != Kind)
2401 return false;
2402
2403 // Check for a mismatched access type, and fall back conservatively as needed.
2404 // TODO: Be less conservative when the type is similar and can use the same
2405 // addressing modes.
2406 if (Kind == LSRUse::Address) {
2407 if (AccessTy.MemTy != LU.AccessTy.MemTy) {
2408 NewAccessTy = MemAccessTy::getUnknown(AccessTy.MemTy->getContext(),
2409 AccessTy.AddrSpace);
2410 }
2411 }
2412
2413 // Conservatively assume HasBaseReg is true for now.
2414 if (NewOffset < LU.MinOffset) {
2415 if (!isAlwaysFoldable(TTI, Kind, NewAccessTy, /*BaseGV=*/nullptr,
2416 LU.MaxOffset - NewOffset, HasBaseReg))
2417 return false;
2418 NewMinOffset = NewOffset;
2419 } else if (NewOffset > LU.MaxOffset) {
2420 if (!isAlwaysFoldable(TTI, Kind, NewAccessTy, /*BaseGV=*/nullptr,
2421 NewOffset - LU.MinOffset, HasBaseReg))
2422 return false;
2423 NewMaxOffset = NewOffset;
2424 }
2425
2426 // Update the use.
2427 LU.MinOffset = NewMinOffset;
2428 LU.MaxOffset = NewMaxOffset;
2429 LU.AccessTy = NewAccessTy;
2430 return true;
2431}
2432
2433/// Return an LSRUse index and an offset value for a fixup which needs the given
2434/// expression, with the given kind and optional access type. Either reuse an
2435/// existing use or create a new one, as needed.
2436std::pair<size_t, int64_t> LSRInstance::getUse(const SCEV *&Expr,
2437 LSRUse::KindType Kind,
2438 MemAccessTy AccessTy) {
2439 const SCEV *Copy = Expr;
2440 int64_t Offset = ExtractImmediate(Expr, SE);
2441
2442 // Basic uses can't accept any offset, for example.
2443 if (!isAlwaysFoldable(TTI, Kind, AccessTy, /*BaseGV=*/ nullptr,
2444 Offset, /*HasBaseReg=*/ true)) {
2445 Expr = Copy;
2446 Offset = 0;
2447 }
2448
2449 std::pair<UseMapTy::iterator, bool> P =
2450 UseMap.insert(std::make_pair(LSRUse::SCEVUseKindPair(Expr, Kind), 0));
2451 if (!P.second) {
2452 // A use already existed with this base.
2453 size_t LUIdx = P.first->second;
2454 LSRUse &LU = Uses[LUIdx];
2455 if (reconcileNewOffset(LU, Offset, /*HasBaseReg=*/true, Kind, AccessTy))
2456 // Reuse this use.
2457 return std::make_pair(LUIdx, Offset);
2458 }
2459
2460 // Create a new use.
2461 size_t LUIdx = Uses.size();
2462 P.first->second = LUIdx;
2463 Uses.push_back(LSRUse(Kind, AccessTy));
2464 LSRUse &LU = Uses[LUIdx];
2465
2466 LU.MinOffset = Offset;
2467 LU.MaxOffset = Offset;
2468 return std::make_pair(LUIdx, Offset);
2469}
2470
2471/// Delete the given use from the Uses list.
2472void LSRInstance::DeleteUse(LSRUse &LU, size_t LUIdx) {
2473 if (&LU != &Uses.back())
2474 std::swap(LU, Uses.back());
2475 Uses.pop_back();
2476
2477 // Update RegUses.
2478 RegUses.swapAndDropUse(LUIdx, Uses.size());
2479}
2480
2481/// Look for a use distinct from OrigLU which is has a formula that has the same
2482/// registers as the given formula.
2483LSRUse *
2484LSRInstance::FindUseWithSimilarFormula(const Formula &OrigF,
2485 const LSRUse &OrigLU) {
2486 // Search all uses for the formula. This could be more clever.
2487 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
2488 LSRUse &LU = Uses[LUIdx];
2489 // Check whether this use is close enough to OrigLU, to see whether it's
2490 // worthwhile looking through its formulae.
2491 // Ignore ICmpZero uses because they may contain formulae generated by
2492 // GenerateICmpZeroScales, in which case adding fixup offsets may
2493 // be invalid.
2494 if (&LU != &OrigLU &&
2495 LU.Kind != LSRUse::ICmpZero &&
2496 LU.Kind == OrigLU.Kind && OrigLU.AccessTy == LU.AccessTy &&
2497 LU.WidestFixupType == OrigLU.WidestFixupType &&
2498 LU.HasFormulaWithSameRegs(OrigF)) {
2499 // Scan through this use's formulae.
2500 for (const Formula &F : LU.Formulae) {
2501 // Check to see if this formula has the same registers and symbols
2502 // as OrigF.
2503 if (F.BaseRegs == OrigF.BaseRegs &&
2504 F.ScaledReg == OrigF.ScaledReg &&
2505 F.BaseGV == OrigF.BaseGV &&
2506 F.Scale == OrigF.Scale &&
2507 F.UnfoldedOffset == OrigF.UnfoldedOffset) {
2508 if (F.BaseOffset == 0)
2509 return &LU;
2510 // This is the formula where all the registers and symbols matched;
2511 // there aren't going to be any others. Since we declined it, we
2512 // can skip the rest of the formulae and proceed to the next LSRUse.
2513 break;
2514 }
2515 }
2516 }
2517 }
2518
2519 // Nothing looked good.
2520 return nullptr;
2521}
2522
2523void LSRInstance::CollectInterestingTypesAndFactors() {
2524 SmallSetVector<const SCEV *, 4> Strides;
2525
2526 // Collect interesting types and strides.
2527 SmallVector<const SCEV *, 4> Worklist;
2528 for (const IVStrideUse &U : IU) {
2529 const SCEV *Expr = IU.getExpr(U);
2530
2531 // Collect interesting types.
2532 Types.insert(SE.getEffectiveSCEVType(Expr->getType()));
2533
2534 // Add strides for mentioned loops.
2535 Worklist.push_back(Expr);
2536 do {
2537 const SCEV *S = Worklist.pop_back_val();
2538 if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
2539 if (AR->getLoop() == L)
2540 Strides.insert(AR->getStepRecurrence(SE));
2541 Worklist.push_back(AR->getStart());
2542 } else if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
2543 Worklist.append(Add->op_begin(), Add->op_end());
2544 }
2545 } while (!Worklist.empty());
2546 }
2547
2548 // Compute interesting factors from the set of interesting strides.
2549 for (SmallSetVector<const SCEV *, 4>::const_iterator
2550 I = Strides.begin(), E = Strides.end(); I != E; ++I)
2551 for (SmallSetVector<const SCEV *, 4>::const_iterator NewStrideIter =
2552 std::next(I); NewStrideIter != E; ++NewStrideIter) {
2553 const SCEV *OldStride = *I;
2554 const SCEV *NewStride = *NewStrideIter;
2555
2556 if (SE.getTypeSizeInBits(OldStride->getType()) !=
2557 SE.getTypeSizeInBits(NewStride->getType())) {
2558 if (SE.getTypeSizeInBits(OldStride->getType()) >
2559 SE.getTypeSizeInBits(NewStride->getType()))
2560 NewStride = SE.getSignExtendExpr(NewStride, OldStride->getType());
2561 else
2562 OldStride = SE.getSignExtendExpr(OldStride, NewStride->getType());
2563 }
2564 if (const SCEVConstant *Factor =
2565 dyn_cast_or_null<SCEVConstant>(getExactSDiv(NewStride, OldStride,
2566 SE, true))) {
2567 if (Factor->getAPInt().getMinSignedBits() <= 64)
2568 Factors.insert(Factor->getAPInt().getSExtValue());
2569 } else if (const SCEVConstant *Factor =
2570 dyn_cast_or_null<SCEVConstant>(getExactSDiv(OldStride,
2571 NewStride,
2572 SE, true))) {
2573 if (Factor->getAPInt().getMinSignedBits() <= 64)
2574 Factors.insert(Factor->getAPInt().getSExtValue());
2575 }
2576 }
2577
2578 // If all uses use the same type, don't bother looking for truncation-based
2579 // reuse.
2580 if (Types.size() == 1)
2581 Types.clear();
2582
2583 DEBUG(print_factors_and_types(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { print_factors_and_types(dbgs()); } } while
(false)
;
2584}
2585
2586/// Helper for CollectChains that finds an IV operand (computed by an AddRec in
2587/// this loop) within [OI,OE) or returns OE. If IVUsers mapped Instructions to
2588/// IVStrideUses, we could partially skip this.
2589static User::op_iterator
2590findIVOperand(User::op_iterator OI, User::op_iterator OE,
2591 Loop *L, ScalarEvolution &SE) {
2592 for(; OI != OE; ++OI) {
2593 if (Instruction *Oper = dyn_cast<Instruction>(*OI)) {
2594 if (!SE.isSCEVable(Oper->getType()))
2595 continue;
2596
2597 if (const SCEVAddRecExpr *AR =
2598 dyn_cast<SCEVAddRecExpr>(SE.getSCEV(Oper))) {
2599 if (AR->getLoop() == L)
2600 break;
2601 }
2602 }
2603 }
2604 return OI;
2605}
2606
2607/// IVChain logic must consistenctly peek base TruncInst operands, so wrap it in
2608/// a convenient helper.
2609static Value *getWideOperand(Value *Oper) {
2610 if (TruncInst *Trunc = dyn_cast<TruncInst>(Oper))
2611 return Trunc->getOperand(0);
2612 return Oper;
2613}
2614
2615/// Return true if we allow an IV chain to include both types.
2616static bool isCompatibleIVType(Value *LVal, Value *RVal) {
2617 Type *LType = LVal->getType();
2618 Type *RType = RVal->getType();
2619 return (LType == RType) || (LType->isPointerTy() && RType->isPointerTy() &&
2620 // Different address spaces means (possibly)
2621 // different types of the pointer implementation,
2622 // e.g. i16 vs i32 so disallow that.
2623 (LType->getPointerAddressSpace() ==
2624 RType->getPointerAddressSpace()));
2625}
2626
2627/// Return an approximation of this SCEV expression's "base", or NULL for any
2628/// constant. Returning the expression itself is conservative. Returning a
2629/// deeper subexpression is more precise and valid as long as it isn't less
2630/// complex than another subexpression. For expressions involving multiple
2631/// unscaled values, we need to return the pointer-type SCEVUnknown. This avoids
2632/// forming chains across objects, such as: PrevOper==a[i], IVOper==b[i],
2633/// IVInc==b-a.
2634///
2635/// Since SCEVUnknown is the rightmost type, and pointers are the rightmost
2636/// SCEVUnknown, we simply return the rightmost SCEV operand.
2637static const SCEV *getExprBase(const SCEV *S) {
2638 switch (S->getSCEVType()) {
2639 default: // uncluding scUnknown.
2640 return S;
2641 case scConstant:
2642 return nullptr;
2643 case scTruncate:
2644 return getExprBase(cast<SCEVTruncateExpr>(S)->getOperand());
2645 case scZeroExtend:
2646 return getExprBase(cast<SCEVZeroExtendExpr>(S)->getOperand());
2647 case scSignExtend:
2648 return getExprBase(cast<SCEVSignExtendExpr>(S)->getOperand());
2649 case scAddExpr: {
2650 // Skip over scaled operands (scMulExpr) to follow add operands as long as
2651 // there's nothing more complex.
2652 // FIXME: not sure if we want to recognize negation.
2653 const SCEVAddExpr *Add = cast<SCEVAddExpr>(S);
2654 for (std::reverse_iterator<SCEVAddExpr::op_iterator> I(Add->op_end()),
2655 E(Add->op_begin()); I != E; ++I) {
2656 const SCEV *SubExpr = *I;
2657 if (SubExpr->getSCEVType() == scAddExpr)
2658 return getExprBase(SubExpr);
2659
2660 if (SubExpr->getSCEVType() != scMulExpr)
2661 return SubExpr;
2662 }
2663 return S; // all operands are scaled, be conservative.
2664 }
2665 case scAddRecExpr:
2666 return getExprBase(cast<SCEVAddRecExpr>(S)->getStart());
2667 }
2668}
2669
2670/// Return true if the chain increment is profitable to expand into a loop
2671/// invariant value, which may require its own register. A profitable chain
2672/// increment will be an offset relative to the same base. We allow such offsets
2673/// to potentially be used as chain increment as long as it's not obviously
2674/// expensive to expand using real instructions.
2675bool IVChain::isProfitableIncrement(const SCEV *OperExpr,
2676 const SCEV *IncExpr,
2677 ScalarEvolution &SE) {
2678 // Aggressively form chains when -stress-ivchain.
2679 if (StressIVChain)
2680 return true;
2681
2682 // Do not replace a constant offset from IV head with a nonconstant IV
2683 // increment.
2684 if (!isa<SCEVConstant>(IncExpr)) {
2685 const SCEV *HeadExpr = SE.getSCEV(getWideOperand(Incs[0].IVOperand));
2686 if (isa<SCEVConstant>(SE.getMinusSCEV(OperExpr, HeadExpr)))
2687 return false;
2688 }
2689
2690 SmallPtrSet<const SCEV*, 8> Processed;
2691 return !isHighCostExpansion(IncExpr, Processed, SE);
2692}
2693
2694/// Return true if the number of registers needed for the chain is estimated to
2695/// be less than the number required for the individual IV users. First prohibit
2696/// any IV users that keep the IV live across increments (the Users set should
2697/// be empty). Next count the number and type of increments in the chain.
2698///
2699/// Chaining IVs can lead to considerable code bloat if ISEL doesn't
2700/// effectively use postinc addressing modes. Only consider it profitable it the
2701/// increments can be computed in fewer registers when chained.
2702///
2703/// TODO: Consider IVInc free if it's already used in another chains.
2704static bool
2705isProfitableChain(IVChain &Chain, SmallPtrSetImpl<Instruction*> &Users,
2706 ScalarEvolution &SE, const TargetTransformInfo &TTI) {
2707 if (StressIVChain)
2708 return true;
2709
2710 if (!Chain.hasIncs())
2711 return false;
2712
2713 if (!Users.empty()) {
2714 DEBUG(dbgs() << "Chain: " << *Chain.Incs[0].UserInst << " users:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Chain: " << *Chain.
Incs[0].UserInst << " users:\n"; for (Instruction *Inst
: Users) { dbgs() << " " << *Inst << "\n"
; }; } } while (false)
2715 for (Instruction *Inst : Users) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Chain: " << *Chain.
Incs[0].UserInst << " users:\n"; for (Instruction *Inst
: Users) { dbgs() << " " << *Inst << "\n"
; }; } } while (false)
2716 dbgs() << " " << *Inst << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Chain: " << *Chain.
Incs[0].UserInst << " users:\n"; for (Instruction *Inst
: Users) { dbgs() << " " << *Inst << "\n"
; }; } } while (false)
2717 })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Chain: " << *Chain.
Incs[0].UserInst << " users:\n"; for (Instruction *Inst
: Users) { dbgs() << " " << *Inst << "\n"
; }; } } while (false)
;
2718 return false;
2719 }
2720 assert(!Chain.Incs.empty() && "empty IV chains are not allowed")((!Chain.Incs.empty() && "empty IV chains are not allowed"
) ? static_cast<void> (0) : __assert_fail ("!Chain.Incs.empty() && \"empty IV chains are not allowed\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 2720, __PRETTY_FUNCTION__))
;
2721
2722 // The chain itself may require a register, so intialize cost to 1.
2723 int cost = 1;
2724
2725 // A complete chain likely eliminates the need for keeping the original IV in
2726 // a register. LSR does not currently know how to form a complete chain unless
2727 // the header phi already exists.
2728 if (isa<PHINode>(Chain.tailUserInst())
2729 && SE.getSCEV(Chain.tailUserInst()) == Chain.Incs[0].IncExpr) {
2730 --cost;
2731 }
2732 const SCEV *LastIncExpr = nullptr;
2733 unsigned NumConstIncrements = 0;
2734 unsigned NumVarIncrements = 0;
2735 unsigned NumReusedIncrements = 0;
2736 for (const IVInc &Inc : Chain) {
2737 if (Inc.IncExpr->isZero())
2738 continue;
2739
2740 // Incrementing by zero or some constant is neutral. We assume constants can
2741 // be folded into an addressing mode or an add's immediate operand.
2742 if (isa<SCEVConstant>(Inc.IncExpr)) {
2743 ++NumConstIncrements;
2744 continue;
2745 }
2746
2747 if (Inc.IncExpr == LastIncExpr)
2748 ++NumReusedIncrements;
2749 else
2750 ++NumVarIncrements;
2751
2752 LastIncExpr = Inc.IncExpr;
2753 }
2754 // An IV chain with a single increment is handled by LSR's postinc
2755 // uses. However, a chain with multiple increments requires keeping the IV's
2756 // value live longer than it needs to be if chained.
2757 if (NumConstIncrements > 1)
2758 --cost;
2759
2760 // Materializing increment expressions in the preheader that didn't exist in
2761 // the original code may cost a register. For example, sign-extended array
2762 // indices can produce ridiculous increments like this:
2763 // IV + ((sext i32 (2 * %s) to i64) + (-1 * (sext i32 %s to i64)))
2764 cost += NumVarIncrements;
2765
2766 // Reusing variable increments likely saves a register to hold the multiple of
2767 // the stride.
2768 cost -= NumReusedIncrements;
2769
2770 DEBUG(dbgs() << "Chain: " << *Chain.Incs[0].UserInst << " Cost: " << costdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Chain: " << *Chain.
Incs[0].UserInst << " Cost: " << cost << "\n"
; } } while (false)
2771 << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Chain: " << *Chain.
Incs[0].UserInst << " Cost: " << cost << "\n"
; } } while (false)
;
2772
2773 return cost < 0;
2774}
2775
2776/// Add this IV user to an existing chain or make it the head of a new chain.
2777void LSRInstance::ChainInstruction(Instruction *UserInst, Instruction *IVOper,
2778 SmallVectorImpl<ChainUsers> &ChainUsersVec) {
2779 // When IVs are used as types of varying widths, they are generally converted
2780 // to a wider type with some uses remaining narrow under a (free) trunc.
2781 Value *const NextIV = getWideOperand(IVOper);
2782 const SCEV *const OperExpr = SE.getSCEV(NextIV);
2783 const SCEV *const OperExprBase = getExprBase(OperExpr);
2784
2785 // Visit all existing chains. Check if its IVOper can be computed as a
2786 // profitable loop invariant increment from the last link in the Chain.
2787 unsigned ChainIdx = 0, NChains = IVChainVec.size();
2788 const SCEV *LastIncExpr = nullptr;
2789 for (; ChainIdx < NChains; ++ChainIdx) {
2790 IVChain &Chain = IVChainVec[ChainIdx];
2791
2792 // Prune the solution space aggressively by checking that both IV operands
2793 // are expressions that operate on the same unscaled SCEVUnknown. This
2794 // "base" will be canceled by the subsequent getMinusSCEV call. Checking
2795 // first avoids creating extra SCEV expressions.
2796 if (!StressIVChain && Chain.ExprBase != OperExprBase)
2797 continue;
2798
2799 Value *PrevIV = getWideOperand(Chain.Incs.back().IVOperand);
2800 if (!isCompatibleIVType(PrevIV, NextIV))
2801 continue;
2802
2803 // A phi node terminates a chain.
2804 if (isa<PHINode>(UserInst) && isa<PHINode>(Chain.tailUserInst()))
2805 continue;
2806
2807 // The increment must be loop-invariant so it can be kept in a register.
2808 const SCEV *PrevExpr = SE.getSCEV(PrevIV);
2809 const SCEV *IncExpr = SE.getMinusSCEV(OperExpr, PrevExpr);
2810 if (!SE.isLoopInvariant(IncExpr, L))
2811 continue;
2812
2813 if (Chain.isProfitableIncrement(OperExpr, IncExpr, SE)) {
2814 LastIncExpr = IncExpr;
2815 break;
2816 }
2817 }
2818 // If we haven't found a chain, create a new one, unless we hit the max. Don't
2819 // bother for phi nodes, because they must be last in the chain.
2820 if (ChainIdx == NChains) {
2821 if (isa<PHINode>(UserInst))
2822 return;
2823 if (NChains >= MaxChains && !StressIVChain) {
2824 DEBUG(dbgs() << "IV Chain Limit\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "IV Chain Limit\n"; } } while
(false)
;
2825 return;
2826 }
2827 LastIncExpr = OperExpr;
2828 // IVUsers may have skipped over sign/zero extensions. We don't currently
2829 // attempt to form chains involving extensions unless they can be hoisted
2830 // into this loop's AddRec.
2831 if (!isa<SCEVAddRecExpr>(LastIncExpr))
2832 return;
2833 ++NChains;
2834 IVChainVec.push_back(IVChain(IVInc(UserInst, IVOper, LastIncExpr),
2835 OperExprBase));
2836 ChainUsersVec.resize(NChains);
2837 DEBUG(dbgs() << "IV Chain#" << ChainIdx << " Head: (" << *UserInstdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "IV Chain#" << ChainIdx
<< " Head: (" << *UserInst << ") IV=" <<
*LastIncExpr << "\n"; } } while (false)
2838 << ") IV=" << *LastIncExpr << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "IV Chain#" << ChainIdx
<< " Head: (" << *UserInst << ") IV=" <<
*LastIncExpr << "\n"; } } while (false)
;
2839 } else {
2840 DEBUG(dbgs() << "IV Chain#" << ChainIdx << " Inc: (" << *UserInstdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "IV Chain#" << ChainIdx
<< " Inc: (" << *UserInst << ") IV+" <<
*LastIncExpr << "\n"; } } while (false)
2841 << ") IV+" << *LastIncExpr << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "IV Chain#" << ChainIdx
<< " Inc: (" << *UserInst << ") IV+" <<
*LastIncExpr << "\n"; } } while (false)
;
2842 // Add this IV user to the end of the chain.
2843 IVChainVec[ChainIdx].add(IVInc(UserInst, IVOper, LastIncExpr));
2844 }
2845 IVChain &Chain = IVChainVec[ChainIdx];
2846
2847 SmallPtrSet<Instruction*,4> &NearUsers = ChainUsersVec[ChainIdx].NearUsers;
2848 // This chain's NearUsers become FarUsers.
2849 if (!LastIncExpr->isZero()) {
2850 ChainUsersVec[ChainIdx].FarUsers.insert(NearUsers.begin(),
2851 NearUsers.end());
2852 NearUsers.clear();
2853 }
2854
2855 // All other uses of IVOperand become near uses of the chain.
2856 // We currently ignore intermediate values within SCEV expressions, assuming
2857 // they will eventually be used be the current chain, or can be computed
2858 // from one of the chain increments. To be more precise we could
2859 // transitively follow its user and only add leaf IV users to the set.
2860 for (User *U : IVOper->users()) {
2861 Instruction *OtherUse = dyn_cast<Instruction>(U);
2862 if (!OtherUse)
2863 continue;
2864 // Uses in the chain will no longer be uses if the chain is formed.
2865 // Include the head of the chain in this iteration (not Chain.begin()).
2866 IVChain::const_iterator IncIter = Chain.Incs.begin();
2867 IVChain::const_iterator IncEnd = Chain.Incs.end();
2868 for( ; IncIter != IncEnd; ++IncIter) {
2869 if (IncIter->UserInst == OtherUse)
2870 break;
2871 }
2872 if (IncIter != IncEnd)
2873 continue;
2874
2875 if (SE.isSCEVable(OtherUse->getType())
2876 && !isa<SCEVUnknown>(SE.getSCEV(OtherUse))
2877 && IU.isIVUserOrOperand(OtherUse)) {
2878 continue;
2879 }
2880 NearUsers.insert(OtherUse);
2881 }
2882
2883 // Since this user is part of the chain, it's no longer considered a use
2884 // of the chain.
2885 ChainUsersVec[ChainIdx].FarUsers.erase(UserInst);
2886}
2887
2888/// Populate the vector of Chains.
2889///
2890/// This decreases ILP at the architecture level. Targets with ample registers,
2891/// multiple memory ports, and no register renaming probably don't want
2892/// this. However, such targets should probably disable LSR altogether.
2893///
2894/// The job of LSR is to make a reasonable choice of induction variables across
2895/// the loop. Subsequent passes can easily "unchain" computation exposing more
2896/// ILP *within the loop* if the target wants it.
2897///
2898/// Finding the best IV chain is potentially a scheduling problem. Since LSR
2899/// will not reorder memory operations, it will recognize this as a chain, but
2900/// will generate redundant IV increments. Ideally this would be corrected later
2901/// by a smart scheduler:
2902/// = A[i]
2903/// = A[i+x]
2904/// A[i] =
2905/// A[i+x] =
2906///
2907/// TODO: Walk the entire domtree within this loop, not just the path to the
2908/// loop latch. This will discover chains on side paths, but requires
2909/// maintaining multiple copies of the Chains state.
2910void LSRInstance::CollectChains() {
2911 DEBUG(dbgs() << "Collecting IV Chains.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Collecting IV Chains.\n";
} } while (false)
;
2912 SmallVector<ChainUsers, 8> ChainUsersVec;
2913
2914 SmallVector<BasicBlock *,8> LatchPath;
2915 BasicBlock *LoopHeader = L->getHeader();
2916 for (DomTreeNode *Rung = DT.getNode(L->getLoopLatch());
2917 Rung->getBlock() != LoopHeader; Rung = Rung->getIDom()) {
2918 LatchPath.push_back(Rung->getBlock());
2919 }
2920 LatchPath.push_back(LoopHeader);
2921
2922 // Walk the instruction stream from the loop header to the loop latch.
2923 for (BasicBlock *BB : reverse(LatchPath)) {
2924 for (Instruction &I : *BB) {
2925 // Skip instructions that weren't seen by IVUsers analysis.
2926 if (isa<PHINode>(I) || !IU.isIVUserOrOperand(&I))
2927 continue;
2928
2929 // Ignore users that are part of a SCEV expression. This way we only
2930 // consider leaf IV Users. This effectively rediscovers a portion of
2931 // IVUsers analysis but in program order this time.
2932 if (SE.isSCEVable(I.getType()) && !isa<SCEVUnknown>(SE.getSCEV(&I)))
2933 continue;
2934
2935 // Remove this instruction from any NearUsers set it may be in.
2936 for (unsigned ChainIdx = 0, NChains = IVChainVec.size();
2937 ChainIdx < NChains; ++ChainIdx) {
2938 ChainUsersVec[ChainIdx].NearUsers.erase(&I);
2939 }
2940 // Search for operands that can be chained.
2941 SmallPtrSet<Instruction*, 4> UniqueOperands;
2942 User::op_iterator IVOpEnd = I.op_end();
2943 User::op_iterator IVOpIter = findIVOperand(I.op_begin(), IVOpEnd, L, SE);
2944 while (IVOpIter != IVOpEnd) {
2945 Instruction *IVOpInst = cast<Instruction>(*IVOpIter);
2946 if (UniqueOperands.insert(IVOpInst).second)
2947 ChainInstruction(&I, IVOpInst, ChainUsersVec);
2948 IVOpIter = findIVOperand(std::next(IVOpIter), IVOpEnd, L, SE);
2949 }
2950 } // Continue walking down the instructions.
2951 } // Continue walking down the domtree.
2952 // Visit phi backedges to determine if the chain can generate the IV postinc.
2953 for (BasicBlock::iterator I = L->getHeader()->begin();
2954 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
2955 if (!SE.isSCEVable(PN->getType()))
2956 continue;
2957
2958 Instruction *IncV =
2959 dyn_cast<Instruction>(PN->getIncomingValueForBlock(L->getLoopLatch()));
2960 if (IncV)
2961 ChainInstruction(PN, IncV, ChainUsersVec);
2962 }
2963 // Remove any unprofitable chains.
2964 unsigned ChainIdx = 0;
2965 for (unsigned UsersIdx = 0, NChains = IVChainVec.size();
2966 UsersIdx < NChains; ++UsersIdx) {
2967 if (!isProfitableChain(IVChainVec[UsersIdx],
2968 ChainUsersVec[UsersIdx].FarUsers, SE, TTI))
2969 continue;
2970 // Preserve the chain at UsesIdx.
2971 if (ChainIdx != UsersIdx)
2972 IVChainVec[ChainIdx] = IVChainVec[UsersIdx];
2973 FinalizeChain(IVChainVec[ChainIdx]);
2974 ++ChainIdx;
2975 }
2976 IVChainVec.resize(ChainIdx);
2977}
2978
2979void LSRInstance::FinalizeChain(IVChain &Chain) {
2980 assert(!Chain.Incs.empty() && "empty IV chains are not allowed")((!Chain.Incs.empty() && "empty IV chains are not allowed"
) ? static_cast<void> (0) : __assert_fail ("!Chain.Incs.empty() && \"empty IV chains are not allowed\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 2980, __PRETTY_FUNCTION__))
;
2981 DEBUG(dbgs() << "Final Chain: " << *Chain.Incs[0].UserInst << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Final Chain: " << *
Chain.Incs[0].UserInst << "\n"; } } while (false)
;
2982
2983 for (const IVInc &Inc : Chain) {
2984 DEBUG(dbgs() << " Inc: " << *Inc.UserInst << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Inc: " << *
Inc.UserInst << "\n"; } } while (false)
;
2985 auto UseI = find(Inc.UserInst->operands(), Inc.IVOperand);
2986 assert(UseI != Inc.UserInst->op_end() && "cannot find IV operand")((UseI != Inc.UserInst->op_end() && "cannot find IV operand"
) ? static_cast<void> (0) : __assert_fail ("UseI != Inc.UserInst->op_end() && \"cannot find IV operand\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 2986, __PRETTY_FUNCTION__))
;
2987 IVIncSet.insert(UseI);
2988 }
2989}
2990
2991/// Return true if the IVInc can be folded into an addressing mode.
2992static bool canFoldIVIncExpr(const SCEV *IncExpr, Instruction *UserInst,
2993 Value *Operand, const TargetTransformInfo &TTI) {
2994 const SCEVConstant *IncConst = dyn_cast<SCEVConstant>(IncExpr);
2995 if (!IncConst || !isAddressUse(UserInst, Operand))
2996 return false;
2997
2998 if (IncConst->getAPInt().getMinSignedBits() > 64)
2999 return false;
3000
3001 MemAccessTy AccessTy = getAccessType(UserInst);
3002 int64_t IncOffset = IncConst->getValue()->getSExtValue();
3003 if (!isAlwaysFoldable(TTI, LSRUse::Address, AccessTy, /*BaseGV=*/nullptr,
3004 IncOffset, /*HaseBaseReg=*/false))
3005 return false;
3006
3007 return true;
3008}
3009
3010/// Generate an add or subtract for each IVInc in a chain to materialize the IV
3011/// user's operand from the previous IV user's operand.
3012void LSRInstance::GenerateIVChain(const IVChain &Chain, SCEVExpander &Rewriter,
3013 SmallVectorImpl<WeakTrackingVH> &DeadInsts) {
3014 // Find the new IVOperand for the head of the chain. It may have been replaced
3015 // by LSR.
3016 const IVInc &Head = Chain.Incs[0];
3017 User::op_iterator IVOpEnd = Head.UserInst->op_end();
3018 // findIVOperand returns IVOpEnd if it can no longer find a valid IV user.
3019 User::op_iterator IVOpIter = findIVOperand(Head.UserInst->op_begin(),
3020 IVOpEnd, L, SE);
3021 Value *IVSrc = nullptr;
1
'IVSrc' initialized to a null pointer value
3022 while (IVOpIter != IVOpEnd) {
2
Loop condition is false. Execution continues on line 3039
3023 IVSrc = getWideOperand(*IVOpIter);
3024
3025 // If this operand computes the expression that the chain needs, we may use
3026 // it. (Check this after setting IVSrc which is used below.)
3027 //
3028 // Note that if Head.IncExpr is wider than IVSrc, then this phi is too
3029 // narrow for the chain, so we can no longer use it. We do allow using a
3030 // wider phi, assuming the LSR checked for free truncation. In that case we
3031 // should already have a truncate on this operand such that
3032 // getSCEV(IVSrc) == IncExpr.
3033 if (SE.getSCEV(*IVOpIter) == Head.IncExpr
3034 || SE.getSCEV(IVSrc) == Head.IncExpr) {
3035 break;
3036 }
3037 IVOpIter = findIVOperand(std::next(IVOpIter), IVOpEnd, L, SE);
3038 }
3039 if (IVOpIter == IVOpEnd) {
3
Taking false branch
3040 // Gracefully give up on this chain.
3041 DEBUG(dbgs() << "Concealed chain head: " << *Head.UserInst << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Concealed chain head: " <<
*Head.UserInst << "\n"; } } while (false)
;
3042 return;
3043 }
3044
3045 DEBUG(dbgs() << "Generate chain at: " << *IVSrc << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Generate chain at: " <<
*IVSrc << "\n"; } } while (false)
;
4
Within the expansion of the macro 'DEBUG':
a
Assuming 'DebugFlag' is not equal to 0
b
Assuming the condition is true
c
Forming reference to null pointer
3046 Type *IVTy = IVSrc->getType();
3047 Type *IntTy = SE.getEffectiveSCEVType(IVTy);
3048 const SCEV *LeftOverExpr = nullptr;
3049 for (const IVInc &Inc : Chain) {
3050 Instruction *InsertPt = Inc.UserInst;
3051 if (isa<PHINode>(InsertPt))
3052 InsertPt = L->getLoopLatch()->getTerminator();
3053
3054 // IVOper will replace the current IV User's operand. IVSrc is the IV
3055 // value currently held in a register.
3056 Value *IVOper = IVSrc;
3057 if (!Inc.IncExpr->isZero()) {
3058 // IncExpr was the result of subtraction of two narrow values, so must
3059 // be signed.
3060 const SCEV *IncExpr = SE.getNoopOrSignExtend(Inc.IncExpr, IntTy);
3061 LeftOverExpr = LeftOverExpr ?
3062 SE.getAddExpr(LeftOverExpr, IncExpr) : IncExpr;
3063 }
3064 if (LeftOverExpr && !LeftOverExpr->isZero()) {
3065 // Expand the IV increment.
3066 Rewriter.clearPostInc();
3067 Value *IncV = Rewriter.expandCodeFor(LeftOverExpr, IntTy, InsertPt);
3068 const SCEV *IVOperExpr = SE.getAddExpr(SE.getUnknown(IVSrc),
3069 SE.getUnknown(IncV));
3070 IVOper = Rewriter.expandCodeFor(IVOperExpr, IVTy, InsertPt);
3071
3072 // If an IV increment can't be folded, use it as the next IV value.
3073 if (!canFoldIVIncExpr(LeftOverExpr, Inc.UserInst, Inc.IVOperand, TTI)) {
3074 assert(IVTy == IVOper->getType() && "inconsistent IV increment type")((IVTy == IVOper->getType() && "inconsistent IV increment type"
) ? static_cast<void> (0) : __assert_fail ("IVTy == IVOper->getType() && \"inconsistent IV increment type\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3074, __PRETTY_FUNCTION__))
;
3075 IVSrc = IVOper;
3076 LeftOverExpr = nullptr;
3077 }
3078 }
3079 Type *OperTy = Inc.IVOperand->getType();
3080 if (IVTy != OperTy) {
3081 assert(SE.getTypeSizeInBits(IVTy) >= SE.getTypeSizeInBits(OperTy) &&((SE.getTypeSizeInBits(IVTy) >= SE.getTypeSizeInBits(OperTy
) && "cannot extend a chained IV") ? static_cast<void
> (0) : __assert_fail ("SE.getTypeSizeInBits(IVTy) >= SE.getTypeSizeInBits(OperTy) && \"cannot extend a chained IV\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3082, __PRETTY_FUNCTION__))
3082 "cannot extend a chained IV")((SE.getTypeSizeInBits(IVTy) >= SE.getTypeSizeInBits(OperTy
) && "cannot extend a chained IV") ? static_cast<void
> (0) : __assert_fail ("SE.getTypeSizeInBits(IVTy) >= SE.getTypeSizeInBits(OperTy) && \"cannot extend a chained IV\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3082, __PRETTY_FUNCTION__))
;
3083 IRBuilder<> Builder(InsertPt);
3084 IVOper = Builder.CreateTruncOrBitCast(IVOper, OperTy, "lsr.chain");
3085 }
3086 Inc.UserInst->replaceUsesOfWith(Inc.IVOperand, IVOper);
3087 DeadInsts.emplace_back(Inc.IVOperand);
3088 }
3089 // If LSR created a new, wider phi, we may also replace its postinc. We only
3090 // do this if we also found a wide value for the head of the chain.
3091 if (isa<PHINode>(Chain.tailUserInst())) {
3092 for (BasicBlock::iterator I = L->getHeader()->begin();
3093 PHINode *Phi = dyn_cast<PHINode>(I); ++I) {
3094 if (!isCompatibleIVType(Phi, IVSrc))
3095 continue;
3096 Instruction *PostIncV = dyn_cast<Instruction>(
3097 Phi->getIncomingValueForBlock(L->getLoopLatch()));
3098 if (!PostIncV || (SE.getSCEV(PostIncV) != SE.getSCEV(IVSrc)))
3099 continue;
3100 Value *IVOper = IVSrc;
3101 Type *PostIncTy = PostIncV->getType();
3102 if (IVTy != PostIncTy) {
3103 assert(PostIncTy->isPointerTy() && "mixing int/ptr IV types")((PostIncTy->isPointerTy() && "mixing int/ptr IV types"
) ? static_cast<void> (0) : __assert_fail ("PostIncTy->isPointerTy() && \"mixing int/ptr IV types\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3103, __PRETTY_FUNCTION__))
;
3104 IRBuilder<> Builder(L->getLoopLatch()->getTerminator());
3105 Builder.SetCurrentDebugLocation(PostIncV->getDebugLoc());
3106 IVOper = Builder.CreatePointerCast(IVSrc, PostIncTy, "lsr.chain");
3107 }
3108 Phi->replaceUsesOfWith(PostIncV, IVOper);
3109 DeadInsts.emplace_back(PostIncV);
3110 }
3111 }
3112}
3113
3114void LSRInstance::CollectFixupsAndInitialFormulae() {
3115 for (const IVStrideUse &U : IU) {
3116 Instruction *UserInst = U.getUser();
3117 // Skip IV users that are part of profitable IV Chains.
3118 User::op_iterator UseI =
3119 find(UserInst->operands(), U.getOperandValToReplace());
3120 assert(UseI != UserInst->op_end() && "cannot find IV operand")((UseI != UserInst->op_end() && "cannot find IV operand"
) ? static_cast<void> (0) : __assert_fail ("UseI != UserInst->op_end() && \"cannot find IV operand\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3120, __PRETTY_FUNCTION__))
;
3121 if (IVIncSet.count(UseI)) {
3122 DEBUG(dbgs() << "Use is in profitable chain: " << **UseI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Use is in profitable chain: "
<< **UseI << '\n'; } } while (false)
;
3123 continue;
3124 }
3125
3126 LSRUse::KindType Kind = LSRUse::Basic;
3127 MemAccessTy AccessTy;
3128 if (isAddressUse(UserInst, U.getOperandValToReplace())) {
3129 Kind = LSRUse::Address;
3130 AccessTy = getAccessType(UserInst);
3131 }
3132
3133 const SCEV *S = IU.getExpr(U);
3134 PostIncLoopSet TmpPostIncLoops = U.getPostIncLoops();
3135
3136 // Equality (== and !=) ICmps are special. We can rewrite (i == N) as
3137 // (N - i == 0), and this allows (N - i) to be the expression that we work
3138 // with rather than just N or i, so we can consider the register
3139 // requirements for both N and i at the same time. Limiting this code to
3140 // equality icmps is not a problem because all interesting loops use
3141 // equality icmps, thanks to IndVarSimplify.
3142 if (ICmpInst *CI = dyn_cast<ICmpInst>(UserInst))
3143 if (CI->isEquality()) {
3144 // Swap the operands if needed to put the OperandValToReplace on the
3145 // left, for consistency.
3146 Value *NV = CI->getOperand(1);
3147 if (NV == U.getOperandValToReplace()) {
3148 CI->setOperand(1, CI->getOperand(0));
3149 CI->setOperand(0, NV);
3150 NV = CI->getOperand(1);
3151 Changed = true;
3152 }
3153
3154 // x == y --> x - y == 0
3155 const SCEV *N = SE.getSCEV(NV);
3156 if (SE.isLoopInvariant(N, L) && isSafeToExpand(N, SE)) {
3157 // S is normalized, so normalize N before folding it into S
3158 // to keep the result normalized.
3159 N = normalizeForPostIncUse(N, TmpPostIncLoops, SE);
3160 Kind = LSRUse::ICmpZero;
3161 S = SE.getMinusSCEV(N, S);
3162 }
3163
3164 // -1 and the negations of all interesting strides (except the negation
3165 // of -1) are now also interesting.
3166 for (size_t i = 0, e = Factors.size(); i != e; ++i)
3167 if (Factors[i] != -1)
3168 Factors.insert(-(uint64_t)Factors[i]);
3169 Factors.insert(-1);
3170 }
3171
3172 // Get or create an LSRUse.
3173 std::pair<size_t, int64_t> P = getUse(S, Kind, AccessTy);
3174 size_t LUIdx = P.first;
3175 int64_t Offset = P.second;
3176 LSRUse &LU = Uses[LUIdx];
3177
3178 // Record the fixup.
3179 LSRFixup &LF = LU.getNewFixup();
3180 LF.UserInst = UserInst;
3181 LF.OperandValToReplace = U.getOperandValToReplace();
3182 LF.PostIncLoops = TmpPostIncLoops;
3183 LF.Offset = Offset;
3184 LU.AllFixupsOutsideLoop &= LF.isUseFullyOutsideLoop(L);
3185
3186 if (!LU.WidestFixupType ||
3187 SE.getTypeSizeInBits(LU.WidestFixupType) <
3188 SE.getTypeSizeInBits(LF.OperandValToReplace->getType()))
3189 LU.WidestFixupType = LF.OperandValToReplace->getType();
3190
3191 // If this is the first use of this LSRUse, give it a formula.
3192 if (LU.Formulae.empty()) {
3193 InsertInitialFormula(S, LU, LUIdx);
3194 CountRegisters(LU.Formulae.back(), LUIdx);
3195 }
3196 }
3197
3198 DEBUG(print_fixups(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { print_fixups(dbgs()); } } while (false)
;
3199}
3200
3201/// Insert a formula for the given expression into the given use, separating out
3202/// loop-variant portions from loop-invariant and loop-computable portions.
3203void
3204LSRInstance::InsertInitialFormula(const SCEV *S, LSRUse &LU, size_t LUIdx) {
3205 // Mark uses whose expressions cannot be expanded.
3206 if (!isSafeToExpand(S, SE))
3207 LU.RigidFormula = true;
3208
3209 Formula F;
3210 F.initialMatch(S, L, SE);
3211 bool Inserted = InsertFormula(LU, LUIdx, F);
3212 assert(Inserted && "Initial formula already exists!")((Inserted && "Initial formula already exists!") ? static_cast
<void> (0) : __assert_fail ("Inserted && \"Initial formula already exists!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3212, __PRETTY_FUNCTION__))
; (void)Inserted;
3213}
3214
3215/// Insert a simple single-register formula for the given expression into the
3216/// given use.
3217void
3218LSRInstance::InsertSupplementalFormula(const SCEV *S,
3219 LSRUse &LU, size_t LUIdx) {
3220 Formula F;
3221 F.BaseRegs.push_back(S);
3222 F.HasBaseReg = true;
3223 bool Inserted = InsertFormula(LU, LUIdx, F);
3224 assert(Inserted && "Supplemental formula already exists!")((Inserted && "Supplemental formula already exists!")
? static_cast<void> (0) : __assert_fail ("Inserted && \"Supplemental formula already exists!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3224, __PRETTY_FUNCTION__))
; (void)Inserted;
3225}
3226
3227/// Note which registers are used by the given formula, updating RegUses.
3228void LSRInstance::CountRegisters(const Formula &F, size_t LUIdx) {
3229 if (F.ScaledReg)
3230 RegUses.countRegister(F.ScaledReg, LUIdx);
3231 for (const SCEV *BaseReg : F.BaseRegs)
3232 RegUses.countRegister(BaseReg, LUIdx);
3233}
3234
3235/// If the given formula has not yet been inserted, add it to the list, and
3236/// return true. Return false otherwise.
3237bool LSRInstance::InsertFormula(LSRUse &LU, unsigned LUIdx, const Formula &F) {
3238 // Do not insert formula that we will not be able to expand.
3239 assert(isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy, F) &&((isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy
, F) && "Formula is illegal") ? static_cast<void>
(0) : __assert_fail ("isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy, F) && \"Formula is illegal\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3240, __PRETTY_FUNCTION__))
3240 "Formula is illegal")((isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy
, F) && "Formula is illegal") ? static_cast<void>
(0) : __assert_fail ("isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy, F) && \"Formula is illegal\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3240, __PRETTY_FUNCTION__))
;
3241
3242 if (!LU.InsertFormula(F, *L))
3243 return false;
3244
3245 CountRegisters(F, LUIdx);
3246 return true;
3247}
3248
3249/// Check for other uses of loop-invariant values which we're tracking. These
3250/// other uses will pin these values in registers, making them less profitable
3251/// for elimination.
3252/// TODO: This currently misses non-constant addrec step registers.
3253/// TODO: Should this give more weight to users inside the loop?
3254void
3255LSRInstance::CollectLoopInvariantFixupsAndFormulae() {
3256 SmallVector<const SCEV *, 8> Worklist(RegUses.begin(), RegUses.end());
3257 SmallPtrSet<const SCEV *, 32> Visited;
3258
3259 while (!Worklist.empty()) {
3260 const SCEV *S = Worklist.pop_back_val();
3261
3262 // Don't process the same SCEV twice
3263 if (!Visited.insert(S).second)
3264 continue;
3265
3266 if (const SCEVNAryExpr *N = dyn_cast<SCEVNAryExpr>(S))
3267 Worklist.append(N->op_begin(), N->op_end());
3268 else if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(S))
3269 Worklist.push_back(C->getOperand());
3270 else if (const SCEVUDivExpr *D = dyn_cast<SCEVUDivExpr>(S)) {
3271 Worklist.push_back(D->getLHS());
3272 Worklist.push_back(D->getRHS());
3273 } else if (const SCEVUnknown *US = dyn_cast<SCEVUnknown>(S)) {
3274 const Value *V = US->getValue();
3275 if (const Instruction *Inst = dyn_cast<Instruction>(V)) {
3276 // Look for instructions defined outside the loop.
3277 if (L->contains(Inst)) continue;
3278 } else if (isa<UndefValue>(V))
3279 // Undef doesn't have a live range, so it doesn't matter.
3280 continue;
3281 for (const Use &U : V->uses()) {
3282 const Instruction *UserInst = dyn_cast<Instruction>(U.getUser());
3283 // Ignore non-instructions.
3284 if (!UserInst)
3285 continue;
3286 // Ignore instructions in other functions (as can happen with
3287 // Constants).
3288 if (UserInst->getParent()->getParent() != L->getHeader()->getParent())
3289 continue;
3290 // Ignore instructions not dominated by the loop.
3291 const BasicBlock *UseBB = !isa<PHINode>(UserInst) ?
3292 UserInst->getParent() :
3293 cast<PHINode>(UserInst)->getIncomingBlock(
3294 PHINode::getIncomingValueNumForOperand(U.getOperandNo()));
3295 if (!DT.dominates(L->getHeader(), UseBB))
3296 continue;
3297 // Don't bother if the instruction is in a BB which ends in an EHPad.
3298 if (UseBB->getTerminator()->isEHPad())
3299 continue;
3300 // Don't bother rewriting PHIs in catchswitch blocks.
3301 if (isa<CatchSwitchInst>(UserInst->getParent()->getTerminator()))
3302 continue;
3303 // Ignore uses which are part of other SCEV expressions, to avoid
3304 // analyzing them multiple times.
3305 if (SE.isSCEVable(UserInst->getType())) {
3306 const SCEV *UserS = SE.getSCEV(const_cast<Instruction *>(UserInst));
3307 // If the user is a no-op, look through to its uses.
3308 if (!isa<SCEVUnknown>(UserS))
3309 continue;
3310 if (UserS == US) {
3311 Worklist.push_back(
3312 SE.getUnknown(const_cast<Instruction *>(UserInst)));
3313 continue;
3314 }
3315 }
3316 // Ignore icmp instructions which are already being analyzed.
3317 if (const ICmpInst *ICI = dyn_cast<ICmpInst>(UserInst)) {
3318 unsigned OtherIdx = !U.getOperandNo();
3319 Value *OtherOp = const_cast<Value *>(ICI->getOperand(OtherIdx));
3320 if (SE.hasComputableLoopEvolution(SE.getSCEV(OtherOp), L))
3321 continue;
3322 }
3323
3324 std::pair<size_t, int64_t> P = getUse(
3325 S, LSRUse::Basic, MemAccessTy());
3326 size_t LUIdx = P.first;
3327 int64_t Offset = P.second;
3328 LSRUse &LU = Uses[LUIdx];
3329 LSRFixup &LF = LU.getNewFixup();
3330 LF.UserInst = const_cast<Instruction *>(UserInst);
3331 LF.OperandValToReplace = U;
3332 LF.Offset = Offset;
3333 LU.AllFixupsOutsideLoop &= LF.isUseFullyOutsideLoop(L);
3334 if (!LU.WidestFixupType ||
3335 SE.getTypeSizeInBits(LU.WidestFixupType) <
3336 SE.getTypeSizeInBits(LF.OperandValToReplace->getType()))
3337 LU.WidestFixupType = LF.OperandValToReplace->getType();
3338 InsertSupplementalFormula(US, LU, LUIdx);
3339 CountRegisters(LU.Formulae.back(), Uses.size() - 1);
3340 break;
3341 }
3342 }
3343 }
3344}
3345
3346/// Split S into subexpressions which can be pulled out into separate
3347/// registers. If C is non-null, multiply each subexpression by C.
3348///
3349/// Return remainder expression after factoring the subexpressions captured by
3350/// Ops. If Ops is complete, return NULL.
3351static const SCEV *CollectSubexprs(const SCEV *S, const SCEVConstant *C,
3352 SmallVectorImpl<const SCEV *> &Ops,
3353 const Loop *L,
3354 ScalarEvolution &SE,
3355 unsigned Depth = 0) {
3356 // Arbitrarily cap recursion to protect compile time.
3357 if (Depth >= 3)
3358 return S;
3359
3360 if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
3361 // Break out add operands.
3362 for (const SCEV *S : Add->operands()) {
3363 const SCEV *Remainder = CollectSubexprs(S, C, Ops, L, SE, Depth+1);
3364 if (Remainder)
3365 Ops.push_back(C ? SE.getMulExpr(C, Remainder) : Remainder);
3366 }
3367 return nullptr;
3368 } else if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
3369 // Split a non-zero base out of an addrec.
3370 if (AR->getStart()->isZero() || !AR->isAffine())
3371 return S;
3372
3373 const SCEV *Remainder = CollectSubexprs(AR->getStart(),
3374 C, Ops, L, SE, Depth+1);
3375 // Split the non-zero AddRec unless it is part of a nested recurrence that
3376 // does not pertain to this loop.
3377 if (Remainder && (AR->getLoop() == L || !isa<SCEVAddRecExpr>(Remainder))) {
3378 Ops.push_back(C ? SE.getMulExpr(C, Remainder) : Remainder);
3379 Remainder = nullptr;
3380 }
3381 if (Remainder != AR->getStart()) {
3382 if (!Remainder)
3383 Remainder = SE.getConstant(AR->getType(), 0);
3384 return SE.getAddRecExpr(Remainder,
3385 AR->getStepRecurrence(SE),
3386 AR->getLoop(),
3387 //FIXME: AR->getNoWrapFlags(SCEV::FlagNW)
3388 SCEV::FlagAnyWrap);
3389 }
3390 } else if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(S)) {
3391 // Break (C * (a + b + c)) into C*a + C*b + C*c.
3392 if (Mul->getNumOperands() != 2)
3393 return S;
3394 if (const SCEVConstant *Op0 =
3395 dyn_cast<SCEVConstant>(Mul->getOperand(0))) {
3396 C = C ? cast<SCEVConstant>(SE.getMulExpr(C, Op0)) : Op0;
3397 const SCEV *Remainder =
3398 CollectSubexprs(Mul->getOperand(1), C, Ops, L, SE, Depth+1);
3399 if (Remainder)
3400 Ops.push_back(SE.getMulExpr(C, Remainder));
3401 return nullptr;
3402 }
3403 }
3404 return S;
3405}
3406
3407/// \brief Helper function for LSRInstance::GenerateReassociations.
3408void LSRInstance::GenerateReassociationsImpl(LSRUse &LU, unsigned LUIdx,
3409 const Formula &Base,
3410 unsigned Depth, size_t Idx,
3411 bool IsScaledReg) {
3412 const SCEV *BaseReg = IsScaledReg ? Base.ScaledReg : Base.BaseRegs[Idx];
3413 SmallVector<const SCEV *, 8> AddOps;
3414 const SCEV *Remainder = CollectSubexprs(BaseReg, nullptr, AddOps, L, SE);
3415 if (Remainder)
3416 AddOps.push_back(Remainder);
3417
3418 if (AddOps.size() == 1)
3419 return;
3420
3421 for (SmallVectorImpl<const SCEV *>::const_iterator J = AddOps.begin(),
3422 JE = AddOps.end();
3423 J != JE; ++J) {
3424
3425 // Loop-variant "unknown" values are uninteresting; we won't be able to
3426 // do anything meaningful with them.
3427 if (isa<SCEVUnknown>(*J) && !SE.isLoopInvariant(*J, L))
3428 continue;
3429
3430 // Don't pull a constant into a register if the constant could be folded
3431 // into an immediate field.
3432 if (isAlwaysFoldable(TTI, SE, LU.MinOffset, LU.MaxOffset, LU.Kind,
3433 LU.AccessTy, *J, Base.getNumRegs() > 1))
3434 continue;
3435
3436 // Collect all operands except *J.
3437 SmallVector<const SCEV *, 8> InnerAddOps(
3438 ((const SmallVector<const SCEV *, 8> &)AddOps).begin(), J);
3439 InnerAddOps.append(std::next(J),
3440 ((const SmallVector<const SCEV *, 8> &)AddOps).end());
3441
3442 // Don't leave just a constant behind in a register if the constant could
3443 // be folded into an immediate field.
3444 if (InnerAddOps.size() == 1 &&
3445 isAlwaysFoldable(TTI, SE, LU.MinOffset, LU.MaxOffset, LU.Kind,
3446 LU.AccessTy, InnerAddOps[0], Base.getNumRegs() > 1))
3447 continue;
3448
3449 const SCEV *InnerSum = SE.getAddExpr(InnerAddOps);
3450 if (InnerSum->isZero())
3451 continue;
3452 Formula F = Base;
3453
3454 // Add the remaining pieces of the add back into the new formula.
3455 const SCEVConstant *InnerSumSC = dyn_cast<SCEVConstant>(InnerSum);
3456 if (InnerSumSC && SE.getTypeSizeInBits(InnerSumSC->getType()) <= 64 &&
3457 TTI.isLegalAddImmediate((uint64_t)F.UnfoldedOffset +
3458 InnerSumSC->getValue()->getZExtValue())) {
3459 F.UnfoldedOffset =
3460 (uint64_t)F.UnfoldedOffset + InnerSumSC->getValue()->getZExtValue();
3461 if (IsScaledReg)
3462 F.ScaledReg = nullptr;
3463 else
3464 F.BaseRegs.erase(F.BaseRegs.begin() + Idx);
3465 } else if (IsScaledReg)
3466 F.ScaledReg = InnerSum;
3467 else
3468 F.BaseRegs[Idx] = InnerSum;
3469
3470 // Add J as its own register, or an unfolded immediate.
3471 const SCEVConstant *SC = dyn_cast<SCEVConstant>(*J);
3472 if (SC && SE.getTypeSizeInBits(SC->getType()) <= 64 &&
3473 TTI.isLegalAddImmediate((uint64_t)F.UnfoldedOffset +
3474 SC->getValue()->getZExtValue()))
3475 F.UnfoldedOffset =
3476 (uint64_t)F.UnfoldedOffset + SC->getValue()->getZExtValue();
3477 else
3478 F.BaseRegs.push_back(*J);
3479 // We may have changed the number of register in base regs, adjust the
3480 // formula accordingly.
3481 F.canonicalize(*L);
3482
3483 if (InsertFormula(LU, LUIdx, F))
3484 // If that formula hadn't been seen before, recurse to find more like
3485 // it.
3486 GenerateReassociations(LU, LUIdx, LU.Formulae.back(), Depth + 1);
3487 }
3488}
3489
3490/// Split out subexpressions from adds and the bases of addrecs.
3491void LSRInstance::GenerateReassociations(LSRUse &LU, unsigned LUIdx,
3492 Formula Base, unsigned Depth) {
3493 assert(Base.isCanonical(*L) && "Input must be in the canonical form")((Base.isCanonical(*L) && "Input must be in the canonical form"
) ? static_cast<void> (0) : __assert_fail ("Base.isCanonical(*L) && \"Input must be in the canonical form\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3493, __PRETTY_FUNCTION__))
;
3494 // Arbitrarily cap recursion to protect compile time.
3495 if (Depth >= 3)
3496 return;
3497
3498 for (size_t i = 0, e = Base.BaseRegs.size(); i != e; ++i)
3499 GenerateReassociationsImpl(LU, LUIdx, Base, Depth, i);
3500
3501 if (Base.Scale == 1)
3502 GenerateReassociationsImpl(LU, LUIdx, Base, Depth,
3503 /* Idx */ -1, /* IsScaledReg */ true);
3504}
3505
3506/// Generate a formula consisting of all of the loop-dominating registers added
3507/// into a single register.
3508void LSRInstance::GenerateCombinations(LSRUse &LU, unsigned LUIdx,
3509 Formula Base) {
3510 // This method is only interesting on a plurality of registers.
3511 if (Base.BaseRegs.size() + (Base.Scale == 1) <= 1)
3512 return;
3513
3514 // Flatten the representation, i.e., reg1 + 1*reg2 => reg1 + reg2, before
3515 // processing the formula.
3516 Base.unscale();
3517 Formula F = Base;
3518 F.BaseRegs.clear();
3519 SmallVector<const SCEV *, 4> Ops;
3520 for (const SCEV *BaseReg : Base.BaseRegs) {
3521 if (SE.properlyDominates(BaseReg, L->getHeader()) &&
3522 !SE.hasComputableLoopEvolution(BaseReg, L))
3523 Ops.push_back(BaseReg);
3524 else
3525 F.BaseRegs.push_back(BaseReg);
3526 }
3527 if (Ops.size() > 1) {
3528 const SCEV *Sum = SE.getAddExpr(Ops);
3529 // TODO: If Sum is zero, it probably means ScalarEvolution missed an
3530 // opportunity to fold something. For now, just ignore such cases
3531 // rather than proceed with zero in a register.
3532 if (!Sum->isZero()) {
3533 F.BaseRegs.push_back(Sum);
3534 F.canonicalize(*L);
3535 (void)InsertFormula(LU, LUIdx, F);
3536 }
3537 }
3538}
3539
3540/// \brief Helper function for LSRInstance::GenerateSymbolicOffsets.
3541void LSRInstance::GenerateSymbolicOffsetsImpl(LSRUse &LU, unsigned LUIdx,
3542 const Formula &Base, size_t Idx,
3543 bool IsScaledReg) {
3544 const SCEV *G = IsScaledReg ? Base.ScaledReg : Base.BaseRegs[Idx];
3545 GlobalValue *GV = ExtractSymbol(G, SE);
3546 if (G->isZero() || !GV)
3547 return;
3548 Formula F = Base;
3549 F.BaseGV = GV;
3550 if (!isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy, F))
3551 return;
3552 if (IsScaledReg)
3553 F.ScaledReg = G;
3554 else
3555 F.BaseRegs[Idx] = G;
3556 (void)InsertFormula(LU, LUIdx, F);
3557}
3558
3559/// Generate reuse formulae using symbolic offsets.
3560void LSRInstance::GenerateSymbolicOffsets(LSRUse &LU, unsigned LUIdx,
3561 Formula Base) {
3562 // We can't add a symbolic offset if the address already contains one.
3563 if (Base.BaseGV) return;
3564
3565 for (size_t i = 0, e = Base.BaseRegs.size(); i != e; ++i)
3566 GenerateSymbolicOffsetsImpl(LU, LUIdx, Base, i);
3567 if (Base.Scale == 1)
3568 GenerateSymbolicOffsetsImpl(LU, LUIdx, Base, /* Idx */ -1,
3569 /* IsScaledReg */ true);
3570}
3571
3572/// \brief Helper function for LSRInstance::GenerateConstantOffsets.
3573void LSRInstance::GenerateConstantOffsetsImpl(
3574 LSRUse &LU, unsigned LUIdx, const Formula &Base,
3575 const SmallVectorImpl<int64_t> &Worklist, size_t Idx, bool IsScaledReg) {
3576 const SCEV *G = IsScaledReg ? Base.ScaledReg : Base.BaseRegs[Idx];
3577 for (int64_t Offset : Worklist) {
3578 Formula F = Base;
3579 F.BaseOffset = (uint64_t)Base.BaseOffset - Offset;
3580 if (isLegalUse(TTI, LU.MinOffset - Offset, LU.MaxOffset - Offset, LU.Kind,
3581 LU.AccessTy, F)) {
3582 // Add the offset to the base register.
3583 const SCEV *NewG = SE.getAddExpr(SE.getConstant(G->getType(), Offset), G);
3584 // If it cancelled out, drop the base register, otherwise update it.
3585 if (NewG->isZero()) {
3586 if (IsScaledReg) {
3587 F.Scale = 0;
3588 F.ScaledReg = nullptr;
3589 } else
3590 F.deleteBaseReg(F.BaseRegs[Idx]);
3591 F.canonicalize(*L);
3592 } else if (IsScaledReg)
3593 F.ScaledReg = NewG;
3594 else
3595 F.BaseRegs[Idx] = NewG;
3596
3597 (void)InsertFormula(LU, LUIdx, F);
3598 }
3599 }
3600
3601 int64_t Imm = ExtractImmediate(G, SE);
3602 if (G->isZero() || Imm == 0)
3603 return;
3604 Formula F = Base;
3605 F.BaseOffset = (uint64_t)F.BaseOffset + Imm;
3606 if (!isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy, F))
3607 return;
3608 if (IsScaledReg)
3609 F.ScaledReg = G;
3610 else
3611 F.BaseRegs[Idx] = G;
3612 (void)InsertFormula(LU, LUIdx, F);
3613}
3614
3615/// GenerateConstantOffsets - Generate reuse formulae using symbolic offsets.
3616void LSRInstance::GenerateConstantOffsets(LSRUse &LU, unsigned LUIdx,
3617 Formula Base) {
3618 // TODO: For now, just add the min and max offset, because it usually isn't
3619 // worthwhile looking at everything inbetween.
3620 SmallVector<int64_t, 2> Worklist;
3621 Worklist.push_back(LU.MinOffset);
3622 if (LU.MaxOffset != LU.MinOffset)
3623 Worklist.push_back(LU.MaxOffset);
3624
3625 for (size_t i = 0, e = Base.BaseRegs.size(); i != e; ++i)
3626 GenerateConstantOffsetsImpl(LU, LUIdx, Base, Worklist, i);
3627 if (Base.Scale == 1)
3628 GenerateConstantOffsetsImpl(LU, LUIdx, Base, Worklist, /* Idx */ -1,
3629 /* IsScaledReg */ true);
3630}
3631
3632/// For ICmpZero, check to see if we can scale up the comparison. For example, x
3633/// == y -> x*c == y*c.
3634void LSRInstance::GenerateICmpZeroScales(LSRUse &LU, unsigned LUIdx,
3635 Formula Base) {
3636 if (LU.Kind != LSRUse::ICmpZero) return;
3637
3638 // Determine the integer type for the base formula.
3639 Type *IntTy = Base.getType();
3640 if (!IntTy) return;
3641 if (SE.getTypeSizeInBits(IntTy) > 64) return;
3642
3643 // Don't do this if there is more than one offset.
3644 if (LU.MinOffset != LU.MaxOffset) return;
3645
3646 assert(!Base.BaseGV && "ICmpZero use is not legal!")((!Base.BaseGV && "ICmpZero use is not legal!") ? static_cast
<void> (0) : __assert_fail ("!Base.BaseGV && \"ICmpZero use is not legal!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3646, __PRETTY_FUNCTION__))
;
3647
3648 // Check each interesting stride.
3649 for (int64_t Factor : Factors) {
3650 // Check that the multiplication doesn't overflow.
3651 if (Base.BaseOffset == INT64_MIN(-9223372036854775807L -1) && Factor == -1)
3652 continue;
3653 int64_t NewBaseOffset = (uint64_t)Base.BaseOffset * Factor;
3654 if (NewBaseOffset / Factor != Base.BaseOffset)
3655 continue;
3656 // If the offset will be truncated at this use, check that it is in bounds.
3657 if (!IntTy->isPointerTy() &&
3658 !ConstantInt::isValueValidForType(IntTy, NewBaseOffset))
3659 continue;
3660
3661 // Check that multiplying with the use offset doesn't overflow.
3662 int64_t Offset = LU.MinOffset;
3663 if (Offset == INT64_MIN(-9223372036854775807L -1) && Factor == -1)
3664 continue;
3665 Offset = (uint64_t)Offset * Factor;
3666 if (Offset / Factor != LU.MinOffset)
3667 continue;
3668 // If the offset will be truncated at this use, check that it is in bounds.
3669 if (!IntTy->isPointerTy() &&
3670 !ConstantInt::isValueValidForType(IntTy, Offset))
3671 continue;
3672
3673 Formula F = Base;
3674 F.BaseOffset = NewBaseOffset;
3675
3676 // Check that this scale is legal.
3677 if (!isLegalUse(TTI, Offset, Offset, LU.Kind, LU.AccessTy, F))
3678 continue;
3679
3680 // Compensate for the use having MinOffset built into it.
3681 F.BaseOffset = (uint64_t)F.BaseOffset + Offset - LU.MinOffset;
3682
3683 const SCEV *FactorS = SE.getConstant(IntTy, Factor);
3684
3685 // Check that multiplying with each base register doesn't overflow.
3686 for (size_t i = 0, e = F.BaseRegs.size(); i != e; ++i) {
3687 F.BaseRegs[i] = SE.getMulExpr(F.BaseRegs[i], FactorS);
3688 if (getExactSDiv(F.BaseRegs[i], FactorS, SE) != Base.BaseRegs[i])
3689 goto next;
3690 }
3691
3692 // Check that multiplying with the scaled register doesn't overflow.
3693 if (F.ScaledReg) {
3694 F.ScaledReg = SE.getMulExpr(F.ScaledReg, FactorS);
3695 if (getExactSDiv(F.ScaledReg, FactorS, SE) != Base.ScaledReg)
3696 continue;
3697 }
3698
3699 // Check that multiplying with the unfolded offset doesn't overflow.
3700 if (F.UnfoldedOffset != 0) {
3701 if (F.UnfoldedOffset == INT64_MIN(-9223372036854775807L -1) && Factor == -1)
3702 continue;
3703 F.UnfoldedOffset = (uint64_t)F.UnfoldedOffset * Factor;
3704 if (F.UnfoldedOffset / Factor != Base.UnfoldedOffset)
3705 continue;
3706 // If the offset will be truncated, check that it is in bounds.
3707 if (!IntTy->isPointerTy() &&
3708 !ConstantInt::isValueValidForType(IntTy, F.UnfoldedOffset))
3709 continue;
3710 }
3711
3712 // If we make it here and it's legal, add it.
3713 (void)InsertFormula(LU, LUIdx, F);
3714 next:;
3715 }
3716}
3717
3718/// Generate stride factor reuse formulae by making use of scaled-offset address
3719/// modes, for example.
3720void LSRInstance::GenerateScales(LSRUse &LU, unsigned LUIdx, Formula Base) {
3721 // Determine the integer type for the base formula.
3722 Type *IntTy = Base.getType();
3723 if (!IntTy) return;
3724
3725 // If this Formula already has a scaled register, we can't add another one.
3726 // Try to unscale the formula to generate a better scale.
3727 if (Base.Scale != 0 && !Base.unscale())
3728 return;
3729
3730 assert(Base.Scale == 0 && "unscale did not did its job!")((Base.Scale == 0 && "unscale did not did its job!") ?
static_cast<void> (0) : __assert_fail ("Base.Scale == 0 && \"unscale did not did its job!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 3730, __PRETTY_FUNCTION__))
;
3731
3732 // Check each interesting stride.
3733 for (int64_t Factor : Factors) {
3734 Base.Scale = Factor;
3735 Base.HasBaseReg = Base.BaseRegs.size() > 1;
3736 // Check whether this scale is going to be legal.
3737 if (!isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy,
3738 Base)) {
3739 // As a special-case, handle special out-of-loop Basic users specially.
3740 // TODO: Reconsider this special case.
3741 if (LU.Kind == LSRUse::Basic &&
3742 isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LSRUse::Special,
3743 LU.AccessTy, Base) &&
3744 LU.AllFixupsOutsideLoop)
3745 LU.Kind = LSRUse::Special;
3746 else
3747 continue;
3748 }
3749 // For an ICmpZero, negating a solitary base register won't lead to
3750 // new solutions.
3751 if (LU.Kind == LSRUse::ICmpZero &&
3752 !Base.HasBaseReg && Base.BaseOffset == 0 && !Base.BaseGV)
3753 continue;
3754 // For each addrec base reg, if its loop is current loop, apply the scale.
3755 for (size_t i = 0, e = Base.BaseRegs.size(); i != e; ++i) {
3756 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Base.BaseRegs[i]);
3757 if (AR && (AR->getLoop() == L || LU.AllFixupsOutsideLoop)) {
3758 const SCEV *FactorS = SE.getConstant(IntTy, Factor);
3759 if (FactorS->isZero())
3760 continue;
3761 // Divide out the factor, ignoring high bits, since we'll be
3762 // scaling the value back up in the end.
3763 if (const SCEV *Quotient = getExactSDiv(AR, FactorS, SE, true)) {
3764 // TODO: This could be optimized to avoid all the copying.
3765 Formula F = Base;
3766 F.ScaledReg = Quotient;
3767 F.deleteBaseReg(F.BaseRegs[i]);
3768 // The canonical representation of 1*reg is reg, which is already in
3769 // Base. In that case, do not try to insert the formula, it will be
3770 // rejected anyway.
3771 if (F.Scale == 1 && (F.BaseRegs.empty() ||
3772 (AR->getLoop() != L && LU.AllFixupsOutsideLoop)))
3773 continue;
3774 // If AllFixupsOutsideLoop is true and F.Scale is 1, we may generate
3775 // non canonical Formula with ScaledReg's loop not being L.
3776 if (F.Scale == 1 && LU.AllFixupsOutsideLoop)
3777 F.canonicalize(*L);
3778 (void)InsertFormula(LU, LUIdx, F);
3779 }
3780 }
3781 }
3782 }
3783}
3784
3785/// Generate reuse formulae from different IV types.
3786void LSRInstance::GenerateTruncates(LSRUse &LU, unsigned LUIdx, Formula Base) {
3787 // Don't bother truncating symbolic values.
3788 if (Base.BaseGV) return;
3789
3790 // Determine the integer type for the base formula.
3791 Type *DstTy = Base.getType();
3792 if (!DstTy) return;
3793 DstTy = SE.getEffectiveSCEVType(DstTy);
3794
3795 for (Type *SrcTy : Types) {
3796 if (SrcTy != DstTy && TTI.isTruncateFree(SrcTy, DstTy)) {
3797 Formula F = Base;
3798
3799 if (F.ScaledReg) F.ScaledReg = SE.getAnyExtendExpr(F.ScaledReg, SrcTy);
3800 for (const SCEV *&BaseReg : F.BaseRegs)
3801 BaseReg = SE.getAnyExtendExpr(BaseReg, SrcTy);
3802
3803 // TODO: This assumes we've done basic processing on all uses and
3804 // have an idea what the register usage is.
3805 if (!F.hasRegsUsedByUsesOtherThan(LUIdx, RegUses))
3806 continue;
3807
3808 F.canonicalize(*L);
3809 (void)InsertFormula(LU, LUIdx, F);
3810 }
3811 }
3812}
3813
3814namespace {
3815
3816/// Helper class for GenerateCrossUseConstantOffsets. It's used to defer
3817/// modifications so that the search phase doesn't have to worry about the data
3818/// structures moving underneath it.
3819struct WorkItem {
3820 size_t LUIdx;
3821 int64_t Imm;
3822 const SCEV *OrigReg;
3823
3824 WorkItem(size_t LI, int64_t I, const SCEV *R)
3825 : LUIdx(LI), Imm(I), OrigReg(R) {}
3826
3827 void print(raw_ostream &OS) const;
3828 void dump() const;
3829};
3830
3831} // end anonymous namespace
3832
3833void WorkItem::print(raw_ostream &OS) const {
3834 OS << "in formulae referencing " << *OrigReg << " in use " << LUIdx
3835 << " , add offset " << Imm;
3836}
3837
3838#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
3839LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void WorkItem::dump() const {
3840 print(errs()); errs() << '\n';
3841}
3842#endif
3843
3844/// Look for registers which are a constant distance apart and try to form reuse
3845/// opportunities between them.
3846void LSRInstance::GenerateCrossUseConstantOffsets() {
3847 // Group the registers by their value without any added constant offset.
3848 typedef std::map<int64_t, const SCEV *> ImmMapTy;
3849 DenseMap<const SCEV *, ImmMapTy> Map;
3850 DenseMap<const SCEV *, SmallBitVector> UsedByIndicesMap;
3851 SmallVector<const SCEV *, 8> Sequence;
3852 for (const SCEV *Use : RegUses) {
3853 const SCEV *Reg = Use; // Make a copy for ExtractImmediate to modify.
3854 int64_t Imm = ExtractImmediate(Reg, SE);
3855 auto Pair = Map.insert(std::make_pair(Reg, ImmMapTy()));
3856 if (Pair.second)
3857 Sequence.push_back(Reg);
3858 Pair.first->second.insert(std::make_pair(Imm, Use));
3859 UsedByIndicesMap[Reg] |= RegUses.getUsedByIndices(Use);
3860 }
3861
3862 // Now examine each set of registers with the same base value. Build up
3863 // a list of work to do and do the work in a separate step so that we're
3864 // not adding formulae and register counts while we're searching.
3865 SmallVector<WorkItem, 32> WorkItems;
3866 SmallSet<std::pair<size_t, int64_t>, 32> UniqueItems;
3867 for (const SCEV *Reg : Sequence) {
3868 const ImmMapTy &Imms = Map.find(Reg)->second;
3869
3870 // It's not worthwhile looking for reuse if there's only one offset.
3871 if (Imms.size() == 1)
3872 continue;
3873
3874 DEBUG(dbgs() << "Generating cross-use offsets for " << *Reg << ':';do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Generating cross-use offsets for "
<< *Reg << ':'; for (const auto &Entry : Imms
) dbgs() << ' ' << Entry.first; dbgs() << '\n'
; } } while (false)
3875 for (const auto &Entry : Imms)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Generating cross-use offsets for "
<< *Reg << ':'; for (const auto &Entry : Imms
) dbgs() << ' ' << Entry.first; dbgs() << '\n'
; } } while (false)
3876 dbgs() << ' ' << Entry.first;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Generating cross-use offsets for "
<< *Reg << ':'; for (const auto &Entry : Imms
) dbgs() << ' ' << Entry.first; dbgs() << '\n'
; } } while (false)
3877 dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Generating cross-use offsets for "
<< *Reg << ':'; for (const auto &Entry : Imms
) dbgs() << ' ' << Entry.first; dbgs() << '\n'
; } } while (false)
;
3878
3879 // Examine each offset.
3880 for (ImmMapTy::const_iterator J = Imms.begin(), JE = Imms.end();
3881 J != JE; ++J) {
3882 const SCEV *OrigReg = J->second;
3883
3884 int64_t JImm = J->first;
3885 const SmallBitVector &UsedByIndices = RegUses.getUsedByIndices(OrigReg);
3886
3887 if (!isa<SCEVConstant>(OrigReg) &&
3888 UsedByIndicesMap[Reg].count() == 1) {
3889 DEBUG(dbgs() << "Skipping cross-use reuse for " << *OrigReg << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Skipping cross-use reuse for "
<< *OrigReg << '\n'; } } while (false)
;
3890 continue;
3891 }
3892
3893 // Conservatively examine offsets between this orig reg a few selected
3894 // other orig regs.
3895 ImmMapTy::const_iterator OtherImms[] = {
3896 Imms.begin(), std::prev(Imms.end()),
3897 Imms.lower_bound((Imms.begin()->first + std::prev(Imms.end())->first) /
3898 2)
3899 };
3900 for (size_t i = 0, e = array_lengthof(OtherImms); i != e; ++i) {
3901 ImmMapTy::const_iterator M = OtherImms[i];
3902 if (M == J || M == JE) continue;
3903
3904 // Compute the difference between the two.
3905 int64_t Imm = (uint64_t)JImm - M->first;
3906 for (unsigned LUIdx : UsedByIndices.set_bits())
3907 // Make a memo of this use, offset, and register tuple.
3908 if (UniqueItems.insert(std::make_pair(LUIdx, Imm)).second)
3909 WorkItems.push_back(WorkItem(LUIdx, Imm, OrigReg));
3910 }
3911 }
3912 }
3913
3914 Map.clear();
3915 Sequence.clear();
3916 UsedByIndicesMap.clear();
3917 UniqueItems.clear();
3918
3919 // Now iterate through the worklist and add new formulae.
3920 for (const WorkItem &WI : WorkItems) {
3921 size_t LUIdx = WI.LUIdx;
3922 LSRUse &LU = Uses[LUIdx];
3923 int64_t Imm = WI.Imm;
3924 const SCEV *OrigReg = WI.OrigReg;
3925
3926 Type *IntTy = SE.getEffectiveSCEVType(OrigReg->getType());
3927 const SCEV *NegImmS = SE.getSCEV(ConstantInt::get(IntTy, -(uint64_t)Imm));
3928 unsigned BitWidth = SE.getTypeSizeInBits(IntTy);
3929
3930 // TODO: Use a more targeted data structure.
3931 for (size_t L = 0, LE = LU.Formulae.size(); L != LE; ++L) {
3932 Formula F = LU.Formulae[L];
3933 // FIXME: The code for the scaled and unscaled registers looks
3934 // very similar but slightly different. Investigate if they
3935 // could be merged. That way, we would not have to unscale the
3936 // Formula.
3937 F.unscale();
3938 // Use the immediate in the scaled register.
3939 if (F.ScaledReg == OrigReg) {
3940 int64_t Offset = (uint64_t)F.BaseOffset + Imm * (uint64_t)F.Scale;
3941 // Don't create 50 + reg(-50).
3942 if (F.referencesReg(SE.getSCEV(
3943 ConstantInt::get(IntTy, -(uint64_t)Offset))))
3944 continue;
3945 Formula NewF = F;
3946 NewF.BaseOffset = Offset;
3947 if (!isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy,
3948 NewF))
3949 continue;
3950 NewF.ScaledReg = SE.getAddExpr(NegImmS, NewF.ScaledReg);
3951
3952 // If the new scale is a constant in a register, and adding the constant
3953 // value to the immediate would produce a value closer to zero than the
3954 // immediate itself, then the formula isn't worthwhile.
3955 if (const SCEVConstant *C = dyn_cast<SCEVConstant>(NewF.ScaledReg))
3956 if (C->getValue()->isNegative() != (NewF.BaseOffset < 0) &&
3957 (C->getAPInt().abs() * APInt(BitWidth, F.Scale))
3958 .ule(std::abs(NewF.BaseOffset)))
3959 continue;
3960
3961 // OK, looks good.
3962 NewF.canonicalize(*this->L);
3963 (void)InsertFormula(LU, LUIdx, NewF);
3964 } else {
3965 // Use the immediate in a base register.
3966 for (size_t N = 0, NE = F.BaseRegs.size(); N != NE; ++N) {
3967 const SCEV *BaseReg = F.BaseRegs[N];
3968 if (BaseReg != OrigReg)
3969 continue;
3970 Formula NewF = F;
3971 NewF.BaseOffset = (uint64_t)NewF.BaseOffset + Imm;
3972 if (!isLegalUse(TTI, LU.MinOffset, LU.MaxOffset,
3973 LU.Kind, LU.AccessTy, NewF)) {
3974 if (!TTI.isLegalAddImmediate((uint64_t)NewF.UnfoldedOffset + Imm))
3975 continue;
3976 NewF = F;
3977 NewF.UnfoldedOffset = (uint64_t)NewF.UnfoldedOffset + Imm;
3978 }
3979 NewF.BaseRegs[N] = SE.getAddExpr(NegImmS, BaseReg);
3980
3981 // If the new formula has a constant in a register, and adding the
3982 // constant value to the immediate would produce a value closer to
3983 // zero than the immediate itself, then the formula isn't worthwhile.
3984 for (const SCEV *NewReg : NewF.BaseRegs)
3985 if (const SCEVConstant *C = dyn_cast<SCEVConstant>(NewReg))
3986 if ((C->getAPInt() + NewF.BaseOffset)
3987 .abs()
3988 .slt(std::abs(NewF.BaseOffset)) &&
3989 (C->getAPInt() + NewF.BaseOffset).countTrailingZeros() >=
3990 countTrailingZeros<uint64_t>(NewF.BaseOffset))
3991 goto skip_formula;
3992
3993 // Ok, looks good.
3994 NewF.canonicalize(*this->L);
3995 (void)InsertFormula(LU, LUIdx, NewF);
3996 break;
3997 skip_formula:;
3998 }
3999 }
4000 }
4001 }
4002}
4003
4004/// Generate formulae for each use.
4005void
4006LSRInstance::GenerateAllReuseFormulae() {
4007 // This is split into multiple loops so that hasRegsUsedByUsesOtherThan
4008 // queries are more precise.
4009 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4010 LSRUse &LU = Uses[LUIdx];
4011 for (size_t i = 0, f = LU.Formulae.size(); i != f; ++i)
4012 GenerateReassociations(LU, LUIdx, LU.Formulae[i]);
4013 for (size_t i = 0, f = LU.Formulae.size(); i != f; ++i)
4014 GenerateCombinations(LU, LUIdx, LU.Formulae[i]);
4015 }
4016 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4017 LSRUse &LU = Uses[LUIdx];
4018 for (size_t i = 0, f = LU.Formulae.size(); i != f; ++i)
4019 GenerateSymbolicOffsets(LU, LUIdx, LU.Formulae[i]);
4020 for (size_t i = 0, f = LU.Formulae.size(); i != f; ++i)
4021 GenerateConstantOffsets(LU, LUIdx, LU.Formulae[i]);
4022 for (size_t i = 0, f = LU.Formulae.size(); i != f; ++i)
4023 GenerateICmpZeroScales(LU, LUIdx, LU.Formulae[i]);
4024 for (size_t i = 0, f = LU.Formulae.size(); i != f; ++i)
4025 GenerateScales(LU, LUIdx, LU.Formulae[i]);
4026 }
4027 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4028 LSRUse &LU = Uses[LUIdx];
4029 for (size_t i = 0, f = LU.Formulae.size(); i != f; ++i)
4030 GenerateTruncates(LU, LUIdx, LU.Formulae[i]);
4031 }
4032
4033 GenerateCrossUseConstantOffsets();
4034
4035 DEBUG(dbgs() << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "After generating reuse formulae:\n"
; print_uses(dbgs()); } } while (false)
4036 "After generating reuse formulae:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "After generating reuse formulae:\n"
; print_uses(dbgs()); } } while (false)
4037 print_uses(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "After generating reuse formulae:\n"
; print_uses(dbgs()); } } while (false)
;
4038}
4039
4040/// If there are multiple formulae with the same set of registers used
4041/// by other uses, pick the best one and delete the others.
4042void LSRInstance::FilterOutUndesirableDedicatedRegisters() {
4043 DenseSet<const SCEV *> VisitedRegs;
4044 SmallPtrSet<const SCEV *, 16> Regs;
4045 SmallPtrSet<const SCEV *, 16> LoserRegs;
4046#ifndef NDEBUG
4047 bool ChangedFormulae = false;
4048#endif
4049
4050 // Collect the best formula for each unique set of shared registers. This
4051 // is reset for each use.
4052 typedef DenseMap<SmallVector<const SCEV *, 4>, size_t, UniquifierDenseMapInfo>
4053 BestFormulaeTy;
4054 BestFormulaeTy BestFormulae;
4055
4056 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4057 LSRUse &LU = Uses[LUIdx];
4058 DEBUG(dbgs() << "Filtering for use "; LU.print(dbgs()); dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Filtering for use "; LU.print
(dbgs()); dbgs() << '\n'; } } while (false)
;
4059
4060 bool Any = false;
4061 for (size_t FIdx = 0, NumForms = LU.Formulae.size();
4062 FIdx != NumForms; ++FIdx) {
4063 Formula &F = LU.Formulae[FIdx];
4064
4065 // Some formulas are instant losers. For example, they may depend on
4066 // nonexistent AddRecs from other loops. These need to be filtered
4067 // immediately, otherwise heuristics could choose them over others leading
4068 // to an unsatisfactory solution. Passing LoserRegs into RateFormula here
4069 // avoids the need to recompute this information across formulae using the
4070 // same bad AddRec. Passing LoserRegs is also essential unless we remove
4071 // the corresponding bad register from the Regs set.
4072 Cost CostF;
4073 Regs.clear();
4074 CostF.RateFormula(TTI, F, Regs, VisitedRegs, L, SE, DT, LU, &LoserRegs);
4075 if (CostF.isLoser()) {
4076 // During initial formula generation, undesirable formulae are generated
4077 // by uses within other loops that have some non-trivial address mode or
4078 // use the postinc form of the IV. LSR needs to provide these formulae
4079 // as the basis of rediscovering the desired formula that uses an AddRec
4080 // corresponding to the existing phi. Once all formulae have been
4081 // generated, these initial losers may be pruned.
4082 DEBUG(dbgs() << " Filtering loser "; F.print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Filtering loser "; F.print
(dbgs()); dbgs() << "\n"; } } while (false)
4083 dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Filtering loser "; F.print
(dbgs()); dbgs() << "\n"; } } while (false)
;
4084 }
4085 else {
4086 SmallVector<const SCEV *, 4> Key;
4087 for (const SCEV *Reg : F.BaseRegs) {
4088 if (RegUses.isRegUsedByUsesOtherThan(Reg, LUIdx))
4089 Key.push_back(Reg);
4090 }
4091 if (F.ScaledReg &&
4092 RegUses.isRegUsedByUsesOtherThan(F.ScaledReg, LUIdx))
4093 Key.push_back(F.ScaledReg);
4094 // Unstable sort by host order ok, because this is only used for
4095 // uniquifying.
4096 std::sort(Key.begin(), Key.end());
4097
4098 std::pair<BestFormulaeTy::const_iterator, bool> P =
4099 BestFormulae.insert(std::make_pair(Key, FIdx));
4100 if (P.second)
4101 continue;
4102
4103 Formula &Best = LU.Formulae[P.first->second];
4104
4105 Cost CostBest;
4106 Regs.clear();
4107 CostBest.RateFormula(TTI, Best, Regs, VisitedRegs, L, SE, DT, LU);
4108 if (CostF < CostBest)
4109 std::swap(F, Best);
4110 DEBUG(dbgs() << " Filtering out formula "; F.print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Filtering out formula "
; F.print(dbgs()); dbgs() << "\n" " in favor of formula "
; Best.print(dbgs()); dbgs() << '\n'; } } while (false)
4111 dbgs() << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Filtering out formula "
; F.print(dbgs()); dbgs() << "\n" " in favor of formula "
; Best.print(dbgs()); dbgs() << '\n'; } } while (false)
4112 " in favor of formula "; Best.print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Filtering out formula "
; F.print(dbgs()); dbgs() << "\n" " in favor of formula "
; Best.print(dbgs()); dbgs() << '\n'; } } while (false)
4113 dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Filtering out formula "
; F.print(dbgs()); dbgs() << "\n" " in favor of formula "
; Best.print(dbgs()); dbgs() << '\n'; } } while (false)
;
4114 }
4115#ifndef NDEBUG
4116 ChangedFormulae = true;
4117#endif
4118 LU.DeleteFormula(F);
4119 --FIdx;
4120 --NumForms;
4121 Any = true;
4122 }
4123
4124 // Now that we've filtered out some formulae, recompute the Regs set.
4125 if (Any)
4126 LU.RecomputeRegs(LUIdx, RegUses);
4127
4128 // Reset this to prepare for the next use.
4129 BestFormulae.clear();
4130 }
4131
4132 DEBUG(if (ChangedFormulae) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { if (ChangedFormulae) { dbgs() << "\n"
"After filtering out undesirable candidates:\n"; print_uses(
dbgs()); }; } } while (false)
4133 dbgs() << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { if (ChangedFormulae) { dbgs() << "\n"
"After filtering out undesirable candidates:\n"; print_uses(
dbgs()); }; } } while (false)
4134 "After filtering out undesirable candidates:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { if (ChangedFormulae) { dbgs() << "\n"
"After filtering out undesirable candidates:\n"; print_uses(
dbgs()); }; } } while (false)
4135 print_uses(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { if (ChangedFormulae) { dbgs() << "\n"
"After filtering out undesirable candidates:\n"; print_uses(
dbgs()); }; } } while (false)
4136 })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { if (ChangedFormulae) { dbgs() << "\n"
"After filtering out undesirable candidates:\n"; print_uses(
dbgs()); }; } } while (false)
;
4137}
4138
4139// This is a rough guess that seems to work fairly well.
4140static const size_t ComplexityLimit = UINT16_MAX(65535);
4141
4142/// Estimate the worst-case number of solutions the solver might have to
4143/// consider. It almost never considers this many solutions because it prune the
4144/// search space, but the pruning isn't always sufficient.
4145size_t LSRInstance::EstimateSearchSpaceComplexity() const {
4146 size_t Power = 1;
4147 for (const LSRUse &LU : Uses) {
4148 size_t FSize = LU.Formulae.size();
4149 if (FSize >= ComplexityLimit) {
4150 Power = ComplexityLimit;
4151 break;
4152 }
4153 Power *= FSize;
4154 if (Power >= ComplexityLimit)
4155 break;
4156 }
4157 return Power;
4158}
4159
4160/// When one formula uses a superset of the registers of another formula, it
4161/// won't help reduce register pressure (though it may not necessarily hurt
4162/// register pressure); remove it to simplify the system.
4163void LSRInstance::NarrowSearchSpaceByDetectingSupersets() {
4164 if (EstimateSearchSpaceComplexity() >= ComplexityLimit) {
4165 DEBUG(dbgs() << "The search space is too complex.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "The search space is too complex.\n"
; } } while (false)
;
4166
4167 DEBUG(dbgs() << "Narrowing the search space by eliminating formulae "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by eliminating formulae "
"which use a superset of registers used by other " "formulae.\n"
; } } while (false)
4168 "which use a superset of registers used by other "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by eliminating formulae "
"which use a superset of registers used by other " "formulae.\n"
; } } while (false)
4169 "formulae.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by eliminating formulae "
"which use a superset of registers used by other " "formulae.\n"
; } } while (false)
;
4170
4171 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4172 LSRUse &LU = Uses[LUIdx];
4173 bool Any = false;
4174 for (size_t i = 0, e = LU.Formulae.size(); i != e; ++i) {
4175 Formula &F = LU.Formulae[i];
4176 // Look for a formula with a constant or GV in a register. If the use
4177 // also has a formula with that same value in an immediate field,
4178 // delete the one that uses a register.
4179 for (SmallVectorImpl<const SCEV *>::const_iterator
4180 I = F.BaseRegs.begin(), E = F.BaseRegs.end(); I != E; ++I) {
4181 if (const SCEVConstant *C = dyn_cast<SCEVConstant>(*I)) {
4182 Formula NewF = F;
4183 NewF.BaseOffset += C->getValue()->getSExtValue();
4184 NewF.BaseRegs.erase(NewF.BaseRegs.begin() +
4185 (I - F.BaseRegs.begin()));
4186 if (LU.HasFormulaWithSameRegs(NewF)) {
4187 DEBUG(dbgs() << " Deleting "; F.print(dbgs()); dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; F.print(dbgs
()); dbgs() << '\n'; } } while (false)
;
4188 LU.DeleteFormula(F);
4189 --i;
4190 --e;
4191 Any = true;
4192 break;
4193 }
4194 } else if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(*I)) {
4195 if (GlobalValue *GV = dyn_cast<GlobalValue>(U->getValue()))
4196 if (!F.BaseGV) {
4197 Formula NewF = F;
4198 NewF.BaseGV = GV;
4199 NewF.BaseRegs.erase(NewF.BaseRegs.begin() +
4200 (I - F.BaseRegs.begin()));
4201 if (LU.HasFormulaWithSameRegs(NewF)) {
4202 DEBUG(dbgs() << " Deleting "; F.print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; F.print(dbgs
()); dbgs() << '\n'; } } while (false)
4203 dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; F.print(dbgs
()); dbgs() << '\n'; } } while (false)
;
4204 LU.DeleteFormula(F);
4205 --i;
4206 --e;
4207 Any = true;
4208 break;
4209 }
4210 }
4211 }
4212 }
4213 }
4214 if (Any)
4215 LU.RecomputeRegs(LUIdx, RegUses);
4216 }
4217
4218 DEBUG(dbgs() << "After pre-selection:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
4219 print_uses(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
;
4220 }
4221}
4222
4223/// When there are many registers for expressions like A, A+1, A+2, etc.,
4224/// allocate a single register for them.
4225void LSRInstance::NarrowSearchSpaceByCollapsingUnrolledCode() {
4226 if (EstimateSearchSpaceComplexity() < ComplexityLimit)
4227 return;
4228
4229 DEBUG(dbgs() << "The search space is too complex.\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "The search space is too complex.\n"
"Narrowing the search space by assuming that uses separated "
"by a constant offset will use the same registers.\n"; } } while
(false)
4230 "Narrowing the search space by assuming that uses separated "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "The search space is too complex.\n"
"Narrowing the search space by assuming that uses separated "
"by a constant offset will use the same registers.\n"; } } while
(false)
4231 "by a constant offset will use the same registers.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "The search space is too complex.\n"
"Narrowing the search space by assuming that uses separated "
"by a constant offset will use the same registers.\n"; } } while
(false)
;
4232
4233 // This is especially useful for unrolled loops.
4234
4235 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4236 LSRUse &LU = Uses[LUIdx];
4237 for (const Formula &F : LU.Formulae) {
4238 if (F.BaseOffset == 0 || (F.Scale != 0 && F.Scale != 1))
4239 continue;
4240
4241 LSRUse *LUThatHas = FindUseWithSimilarFormula(F, LU);
4242 if (!LUThatHas)
4243 continue;
4244
4245 if (!reconcileNewOffset(*LUThatHas, F.BaseOffset, /*HasBaseReg=*/ false,
4246 LU.Kind, LU.AccessTy))
4247 continue;
4248
4249 DEBUG(dbgs() << " Deleting use "; LU.print(dbgs()); dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting use "; LU.print
(dbgs()); dbgs() << '\n'; } } while (false)
;
4250
4251 LUThatHas->AllFixupsOutsideLoop &= LU.AllFixupsOutsideLoop;
4252
4253 // Transfer the fixups of LU to LUThatHas.
4254 for (LSRFixup &Fixup : LU.Fixups) {
4255 Fixup.Offset += F.BaseOffset;
4256 LUThatHas->pushFixup(Fixup);
4257 DEBUG(dbgs() << "New fixup has offset " << Fixup.Offset << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "New fixup has offset " <<
Fixup.Offset << '\n'; } } while (false)
;
4258 }
4259
4260 // Delete formulae from the new use which are no longer legal.
4261 bool Any = false;
4262 for (size_t i = 0, e = LUThatHas->Formulae.size(); i != e; ++i) {
4263 Formula &F = LUThatHas->Formulae[i];
4264 if (!isLegalUse(TTI, LUThatHas->MinOffset, LUThatHas->MaxOffset,
4265 LUThatHas->Kind, LUThatHas->AccessTy, F)) {
4266 DEBUG(dbgs() << " Deleting "; F.print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; F.print(dbgs
()); dbgs() << '\n'; } } while (false)
4267 dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; F.print(dbgs
()); dbgs() << '\n'; } } while (false)
;
4268 LUThatHas->DeleteFormula(F);
4269 --i;
4270 --e;
4271 Any = true;
4272 }
4273 }
4274
4275 if (Any)
4276 LUThatHas->RecomputeRegs(LUThatHas - &Uses.front(), RegUses);
4277
4278 // Delete the old use.
4279 DeleteUse(LU, LUIdx);
4280 --LUIdx;
4281 --NumUses;
4282 break;
4283 }
4284 }
4285
4286 DEBUG(dbgs() << "After pre-selection:\n"; print_uses(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
;
4287}
4288
4289/// Call FilterOutUndesirableDedicatedRegisters again, if necessary, now that
4290/// we've done more filtering, as it may be able to find more formulae to
4291/// eliminate.
4292void LSRInstance::NarrowSearchSpaceByRefilteringUndesirableDedicatedRegisters(){
4293 if (EstimateSearchSpaceComplexity() >= ComplexityLimit) {
4294 DEBUG(dbgs() << "The search space is too complex.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "The search space is too complex.\n"
; } } while (false)
;
4295
4296 DEBUG(dbgs() << "Narrowing the search space by re-filtering out "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by re-filtering out "
"undesirable dedicated registers.\n"; } } while (false)
4297 "undesirable dedicated registers.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by re-filtering out "
"undesirable dedicated registers.\n"; } } while (false)
;
4298
4299 FilterOutUndesirableDedicatedRegisters();
4300
4301 DEBUG(dbgs() << "After pre-selection:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
4302 print_uses(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
;
4303 }
4304}
4305
4306/// The function delete formulas with high registers number expectation.
4307/// Assuming we don't know the value of each formula (already delete
4308/// all inefficient), generate probability of not selecting for each
4309/// register.
4310/// For example,
4311/// Use1:
4312/// reg(a) + reg({0,+,1})
4313/// reg(a) + reg({-1,+,1}) + 1
4314/// reg({a,+,1})
4315/// Use2:
4316/// reg(b) + reg({0,+,1})
4317/// reg(b) + reg({-1,+,1}) + 1
4318/// reg({b,+,1})
4319/// Use3:
4320/// reg(c) + reg(b) + reg({0,+,1})
4321/// reg(c) + reg({b,+,1})
4322///
4323/// Probability of not selecting
4324/// Use1 Use2 Use3
4325/// reg(a) (1/3) * 1 * 1
4326/// reg(b) 1 * (1/3) * (1/2)
4327/// reg({0,+,1}) (2/3) * (2/3) * (1/2)
4328/// reg({-1,+,1}) (2/3) * (2/3) * 1
4329/// reg({a,+,1}) (2/3) * 1 * 1
4330/// reg({b,+,1}) 1 * (2/3) * (2/3)
4331/// reg(c) 1 * 1 * 0
4332///
4333/// Now count registers number mathematical expectation for each formula:
4334/// Note that for each use we exclude probability if not selecting for the use.
4335/// For example for Use1 probability for reg(a) would be just 1 * 1 (excluding
4336/// probabilty 1/3 of not selecting for Use1).
4337/// Use1:
4338/// reg(a) + reg({0,+,1}) 1 + 1/3 -- to be deleted
4339/// reg(a) + reg({-1,+,1}) + 1 1 + 4/9 -- to be deleted
4340/// reg({a,+,1}) 1
4341/// Use2:
4342/// reg(b) + reg({0,+,1}) 1/2 + 1/3 -- to be deleted
4343/// reg(b) + reg({-1,+,1}) + 1 1/2 + 2/3 -- to be deleted
4344/// reg({b,+,1}) 2/3
4345/// Use3:
4346/// reg(c) + reg(b) + reg({0,+,1}) 1 + 1/3 + 4/9 -- to be deleted
4347/// reg(c) + reg({b,+,1}) 1 + 2/3
4348
4349void LSRInstance::NarrowSearchSpaceByDeletingCostlyFormulas() {
4350 if (EstimateSearchSpaceComplexity() < ComplexityLimit)
4351 return;
4352 // Ok, we have too many of formulae on our hands to conveniently handle.
4353 // Use a rough heuristic to thin out the list.
4354
4355 // Set of Regs wich will be 100% used in final solution.
4356 // Used in each formula of a solution (in example above this is reg(c)).
4357 // We can skip them in calculations.
4358 SmallPtrSet<const SCEV *, 4> UniqRegs;
4359 DEBUG(dbgs() << "The search space is too complex.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "The search space is too complex.\n"
; } } while (false)
;
4360
4361 // Map each register to probability of not selecting
4362 DenseMap <const SCEV *, float> RegNumMap;
4363 for (const SCEV *Reg : RegUses) {
4364 if (UniqRegs.count(Reg))
4365 continue;
4366 float PNotSel = 1;
4367 for (const LSRUse &LU : Uses) {
4368 if (!LU.Regs.count(Reg))
4369 continue;
4370 float P = LU.getNotSelectedProbability(Reg);
4371 if (P != 0.0)
4372 PNotSel *= P;
4373 else
4374 UniqRegs.insert(Reg);
4375 }
4376 RegNumMap.insert(std::make_pair(Reg, PNotSel));
4377 }
4378
4379 DEBUG(dbgs() << "Narrowing the search space by deleting costly formulas\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by deleting costly formulas\n"
; } } while (false)
;
4380
4381 // Delete formulas where registers number expectation is high.
4382 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4383 LSRUse &LU = Uses[LUIdx];
4384 // If nothing to delete - continue.
4385 if (LU.Formulae.size() < 2)
4386 continue;
4387 // This is temporary solution to test performance. Float should be
4388 // replaced with round independent type (based on integers) to avoid
4389 // different results for different target builds.
4390 float FMinRegNum = LU.Formulae[0].getNumRegs();
4391 float FMinARegNum = LU.Formulae[0].getNumRegs();
4392 size_t MinIdx = 0;
4393 for (size_t i = 0, e = LU.Formulae.size(); i != e; ++i) {
4394 Formula &F = LU.Formulae[i];
4395 float FRegNum = 0;
4396 float FARegNum = 0;
4397 for (const SCEV *BaseReg : F.BaseRegs) {
4398 if (UniqRegs.count(BaseReg))
4399 continue;
4400 FRegNum += RegNumMap[BaseReg] / LU.getNotSelectedProbability(BaseReg);
4401 if (isa<SCEVAddRecExpr>(BaseReg))
4402 FARegNum +=
4403 RegNumMap[BaseReg] / LU.getNotSelectedProbability(BaseReg);
4404 }
4405 if (const SCEV *ScaledReg = F.ScaledReg) {
4406 if (!UniqRegs.count(ScaledReg)) {
4407 FRegNum +=
4408 RegNumMap[ScaledReg] / LU.getNotSelectedProbability(ScaledReg);
4409 if (isa<SCEVAddRecExpr>(ScaledReg))
4410 FARegNum +=
4411 RegNumMap[ScaledReg] / LU.getNotSelectedProbability(ScaledReg);
4412 }
4413 }
4414 if (FMinRegNum > FRegNum ||
4415 (FMinRegNum == FRegNum && FMinARegNum > FARegNum)) {
4416 FMinRegNum = FRegNum;
4417 FMinARegNum = FARegNum;
4418 MinIdx = i;
4419 }
4420 }
4421 DEBUG(dbgs() << " The formula "; LU.Formulae[MinIdx].print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " The formula "; LU.Formulae
[MinIdx].print(dbgs()); dbgs() << " with min reg num " <<
FMinRegNum << '\n'; } } while (false)
4422 dbgs() << " with min reg num " << FMinRegNum << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " The formula "; LU.Formulae
[MinIdx].print(dbgs()); dbgs() << " with min reg num " <<
FMinRegNum << '\n'; } } while (false)
;
4423 if (MinIdx != 0)
4424 std::swap(LU.Formulae[MinIdx], LU.Formulae[0]);
4425 while (LU.Formulae.size() != 1) {
4426 DEBUG(dbgs() << " Deleting "; LU.Formulae.back().print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; LU.Formulae
.back().print(dbgs()); dbgs() << '\n'; } } while (false
)
4427 dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; LU.Formulae
.back().print(dbgs()); dbgs() << '\n'; } } while (false
)
;
4428 LU.Formulae.pop_back();
4429 }
4430 LU.RecomputeRegs(LUIdx, RegUses);
4431 assert(LU.Formulae.size() == 1 && "Should be exactly 1 min regs formula")((LU.Formulae.size() == 1 && "Should be exactly 1 min regs formula"
) ? static_cast<void> (0) : __assert_fail ("LU.Formulae.size() == 1 && \"Should be exactly 1 min regs formula\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4431, __PRETTY_FUNCTION__))
;
4432 Formula &F = LU.Formulae[0];
4433 DEBUG(dbgs() << " Leaving only "; F.print(dbgs()); dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Leaving only "; F.print
(dbgs()); dbgs() << '\n'; } } while (false)
;
4434 // When we choose the formula, the regs become unique.
4435 UniqRegs.insert(F.BaseRegs.begin(), F.BaseRegs.end());
4436 if (F.ScaledReg)
4437 UniqRegs.insert(F.ScaledReg);
4438 }
4439 DEBUG(dbgs() << "After pre-selection:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
4440 print_uses(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
;
4441}
4442
4443
4444/// Pick a register which seems likely to be profitable, and then in any use
4445/// which has any reference to that register, delete all formulae which do not
4446/// reference that register.
4447void LSRInstance::NarrowSearchSpaceByPickingWinnerRegs() {
4448 // With all other options exhausted, loop until the system is simple
4449 // enough to handle.
4450 SmallPtrSet<const SCEV *, 4> Taken;
4451 while (EstimateSearchSpaceComplexity() >= ComplexityLimit) {
4452 // Ok, we have too many of formulae on our hands to conveniently handle.
4453 // Use a rough heuristic to thin out the list.
4454 DEBUG(dbgs() << "The search space is too complex.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "The search space is too complex.\n"
; } } while (false)
;
4455
4456 // Pick the register which is used by the most LSRUses, which is likely
4457 // to be a good reuse register candidate.
4458 const SCEV *Best = nullptr;
4459 unsigned BestNum = 0;
4460 for (const SCEV *Reg : RegUses) {
4461 if (Taken.count(Reg))
4462 continue;
4463 if (!Best) {
4464 Best = Reg;
4465 BestNum = RegUses.getUsedByIndices(Reg).count();
4466 } else {
4467 unsigned Count = RegUses.getUsedByIndices(Reg).count();
4468 if (Count > BestNum) {
4469 Best = Reg;
4470 BestNum = Count;
4471 }
4472 }
4473 }
4474
4475 DEBUG(dbgs() << "Narrowing the search space by assuming " << *Bestdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by assuming "
<< *Best << " will yield profitable reuse.\n"; }
} while (false)
4476 << " will yield profitable reuse.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "Narrowing the search space by assuming "
<< *Best << " will yield profitable reuse.\n"; }
} while (false)
;
4477 Taken.insert(Best);
4478
4479 // In any use with formulae which references this register, delete formulae
4480 // which don't reference it.
4481 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx) {
4482 LSRUse &LU = Uses[LUIdx];
4483 if (!LU.Regs.count(Best)) continue;
4484
4485 bool Any = false;
4486 for (size_t i = 0, e = LU.Formulae.size(); i != e; ++i) {
4487 Formula &F = LU.Formulae[i];
4488 if (!F.referencesReg(Best)) {
4489 DEBUG(dbgs() << " Deleting "; F.print(dbgs()); dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << " Deleting "; F.print(dbgs
()); dbgs() << '\n'; } } while (false)
;
4490 LU.DeleteFormula(F);
4491 --e;
4492 --i;
4493 Any = true;
4494 assert(e != 0 && "Use has no formulae left! Is Regs inconsistent?")((e != 0 && "Use has no formulae left! Is Regs inconsistent?"
) ? static_cast<void> (0) : __assert_fail ("e != 0 && \"Use has no formulae left! Is Regs inconsistent?\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4494, __PRETTY_FUNCTION__))
;
4495 continue;
4496 }
4497 }
4498
4499 if (Any)
4500 LU.RecomputeRegs(LUIdx, RegUses);
4501 }
4502
4503 DEBUG(dbgs() << "After pre-selection:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
4504 print_uses(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "After pre-selection:\n"; print_uses
(dbgs()); } } while (false)
;
4505 }
4506}
4507
4508/// If there are an extraordinary number of formulae to choose from, use some
4509/// rough heuristics to prune down the number of formulae. This keeps the main
4510/// solver from taking an extraordinary amount of time in some worst-case
4511/// scenarios.
4512void LSRInstance::NarrowSearchSpaceUsingHeuristics() {
4513 NarrowSearchSpaceByDetectingSupersets();
4514 NarrowSearchSpaceByCollapsingUnrolledCode();
4515 NarrowSearchSpaceByRefilteringUndesirableDedicatedRegisters();
4516 if (LSRExpNarrow)
4517 NarrowSearchSpaceByDeletingCostlyFormulas();
4518 else
4519 NarrowSearchSpaceByPickingWinnerRegs();
4520}
4521
4522/// This is the recursive solver.
4523void LSRInstance::SolveRecurse(SmallVectorImpl<const Formula *> &Solution,
4524 Cost &SolutionCost,
4525 SmallVectorImpl<const Formula *> &Workspace,
4526 const Cost &CurCost,
4527 const SmallPtrSet<const SCEV *, 16> &CurRegs,
4528 DenseSet<const SCEV *> &VisitedRegs) const {
4529 // Some ideas:
4530 // - prune more:
4531 // - use more aggressive filtering
4532 // - sort the formula so that the most profitable solutions are found first
4533 // - sort the uses too
4534 // - search faster:
4535 // - don't compute a cost, and then compare. compare while computing a cost
4536 // and bail early.
4537 // - track register sets with SmallBitVector
4538
4539 const LSRUse &LU = Uses[Workspace.size()];
4540
4541 // If this use references any register that's already a part of the
4542 // in-progress solution, consider it a requirement that a formula must
4543 // reference that register in order to be considered. This prunes out
4544 // unprofitable searching.
4545 SmallSetVector<const SCEV *, 4> ReqRegs;
4546 for (const SCEV *S : CurRegs)
4547 if (LU.Regs.count(S))
4548 ReqRegs.insert(S);
4549
4550 SmallPtrSet<const SCEV *, 16> NewRegs;
4551 Cost NewCost;
4552 for (const Formula &F : LU.Formulae) {
4553 // Ignore formulae which may not be ideal in terms of register reuse of
4554 // ReqRegs. The formula should use all required registers before
4555 // introducing new ones.
4556 int NumReqRegsToFind = std::min(F.getNumRegs(), ReqRegs.size());
4557 for (const SCEV *Reg : ReqRegs) {
4558 if ((F.ScaledReg && F.ScaledReg == Reg) ||
4559 is_contained(F.BaseRegs, Reg)) {
4560 --NumReqRegsToFind;
4561 if (NumReqRegsToFind == 0)
4562 break;
4563 }
4564 }
4565 if (NumReqRegsToFind != 0) {
4566 // If none of the formulae satisfied the required registers, then we could
4567 // clear ReqRegs and try again. Currently, we simply give up in this case.
4568 continue;
4569 }
4570
4571 // Evaluate the cost of the current formula. If it's already worse than
4572 // the current best, prune the search at that point.
4573 NewCost = CurCost;
4574 NewRegs = CurRegs;
4575 NewCost.RateFormula(TTI, F, NewRegs, VisitedRegs, L, SE, DT, LU);
4576 if (NewCost < SolutionCost) {
4577 Workspace.push_back(&F);
4578 if (Workspace.size() != Uses.size()) {
4579 SolveRecurse(Solution, SolutionCost, Workspace, NewCost,
4580 NewRegs, VisitedRegs);
4581 if (F.getNumRegs() == 1 && Workspace.size() == 1)
4582 VisitedRegs.insert(F.ScaledReg ? F.ScaledReg : F.BaseRegs[0]);
4583 } else {
4584 DEBUG(dbgs() << "New best at "; NewCost.print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "New best at "; NewCost.print
(dbgs()); dbgs() << ".\n Regs:"; for (const SCEV *S : NewRegs
) dbgs() << ' ' << *S; dbgs() << '\n'; } } while
(false)
4585 dbgs() << ".\n Regs:";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "New best at "; NewCost.print
(dbgs()); dbgs() << ".\n Regs:"; for (const SCEV *S : NewRegs
) dbgs() << ' ' << *S; dbgs() << '\n'; } } while
(false)
4586 for (const SCEV *S : NewRegs)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "New best at "; NewCost.print
(dbgs()); dbgs() << ".\n Regs:"; for (const SCEV *S : NewRegs
) dbgs() << ' ' << *S; dbgs() << '\n'; } } while
(false)
4587 dbgs() << ' ' << *S;do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "New best at "; NewCost.print
(dbgs()); dbgs() << ".\n Regs:"; for (const SCEV *S : NewRegs
) dbgs() << ' ' << *S; dbgs() << '\n'; } } while
(false)
4588 dbgs() << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "New best at "; NewCost.print
(dbgs()); dbgs() << ".\n Regs:"; for (const SCEV *S : NewRegs
) dbgs() << ' ' << *S; dbgs() << '\n'; } } while
(false)
;
4589
4590 SolutionCost = NewCost;
4591 Solution = Workspace;
4592 }
4593 Workspace.pop_back();
4594 }
4595 }
4596}
4597
4598/// Choose one formula from each use. Return the results in the given Solution
4599/// vector.
4600void LSRInstance::Solve(SmallVectorImpl<const Formula *> &Solution) const {
4601 SmallVector<const Formula *, 8> Workspace;
4602 Cost SolutionCost;
4603 SolutionCost.Lose();
4604 Cost CurCost;
4605 SmallPtrSet<const SCEV *, 16> CurRegs;
4606 DenseSet<const SCEV *> VisitedRegs;
4607 Workspace.reserve(Uses.size());
4608
4609 // SolveRecurse does all the work.
4610 SolveRecurse(Solution, SolutionCost, Workspace, CurCost,
4611 CurRegs, VisitedRegs);
4612 if (Solution.empty()) {
4613 DEBUG(dbgs() << "\nNo Satisfactory Solution\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\nNo Satisfactory Solution\n"
; } } while (false)
;
4614 return;
4615 }
4616
4617 // Ok, we've now made all our decisions.
4618 DEBUG(dbgs() << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4619 "The chosen solution requires "; SolutionCost.print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4620 dbgs() << ":\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4621 for (size_t i = 0, e = Uses.size(); i != e; ++i) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4622 dbgs() << " ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4623 Uses[i].print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4624 dbgs() << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4625 " ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4626 Solution[i]->print(dbgs());do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4627 dbgs() << '\n';do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
4628 })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\n" "The chosen solution requires "
; SolutionCost.print(dbgs()); dbgs() << ":\n"; for (size_t
i = 0, e = Uses.size(); i != e; ++i) { dbgs() << " ";
Uses[i].print(dbgs()); dbgs() << "\n" " "; Solution
[i]->print(dbgs()); dbgs() << '\n'; }; } } while (false
)
;
4629
4630 assert(Solution.size() == Uses.size() && "Malformed solution!")((Solution.size() == Uses.size() && "Malformed solution!"
) ? static_cast<void> (0) : __assert_fail ("Solution.size() == Uses.size() && \"Malformed solution!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4630, __PRETTY_FUNCTION__))
;
4631}
4632
4633/// Helper for AdjustInsertPositionForExpand. Climb up the dominator tree far as
4634/// we can go while still being dominated by the input positions. This helps
4635/// canonicalize the insert position, which encourages sharing.
4636BasicBlock::iterator
4637LSRInstance::HoistInsertPosition(BasicBlock::iterator IP,
4638 const SmallVectorImpl<Instruction *> &Inputs)
4639 const {
4640 Instruction *Tentative = &*IP;
4641 while (true) {
4642 bool AllDominate = true;
4643 Instruction *BetterPos = nullptr;
4644 // Don't bother attempting to insert before a catchswitch, their basic block
4645 // cannot have other non-PHI instructions.
4646 if (isa<CatchSwitchInst>(Tentative))
4647 return IP;
4648
4649 for (Instruction *Inst : Inputs) {
4650 if (Inst == Tentative || !DT.dominates(Inst, Tentative)) {
4651 AllDominate = false;
4652 break;
4653 }
4654 // Attempt to find an insert position in the middle of the block,
4655 // instead of at the end, so that it can be used for other expansions.
4656 if (Tentative->getParent() == Inst->getParent() &&
4657 (!BetterPos || !DT.dominates(Inst, BetterPos)))
4658 BetterPos = &*std::next(BasicBlock::iterator(Inst));
4659 }
4660 if (!AllDominate)
4661 break;
4662 if (BetterPos)
4663 IP = BetterPos->getIterator();
4664 else
4665 IP = Tentative->getIterator();
4666
4667 const Loop *IPLoop = LI.getLoopFor(IP->getParent());
4668 unsigned IPLoopDepth = IPLoop ? IPLoop->getLoopDepth() : 0;
4669
4670 BasicBlock *IDom;
4671 for (DomTreeNode *Rung = DT.getNode(IP->getParent()); ; ) {
4672 if (!Rung) return IP;
4673 Rung = Rung->getIDom();
4674 if (!Rung) return IP;
4675 IDom = Rung->getBlock();
4676
4677 // Don't climb into a loop though.
4678 const Loop *IDomLoop = LI.getLoopFor(IDom);
4679 unsigned IDomDepth = IDomLoop ? IDomLoop->getLoopDepth() : 0;
4680 if (IDomDepth <= IPLoopDepth &&
4681 (IDomDepth != IPLoopDepth || IDomLoop == IPLoop))
4682 break;
4683 }
4684
4685 Tentative = IDom->getTerminator();
4686 }
4687
4688 return IP;
4689}
4690
4691/// Determine an input position which will be dominated by the operands and
4692/// which will dominate the result.
4693BasicBlock::iterator
4694LSRInstance::AdjustInsertPositionForExpand(BasicBlock::iterator LowestIP,
4695 const LSRFixup &LF,
4696 const LSRUse &LU,
4697 SCEVExpander &Rewriter) const {
4698 // Collect some instructions which must be dominated by the
4699 // expanding replacement. These must be dominated by any operands that
4700 // will be required in the expansion.
4701 SmallVector<Instruction *, 4> Inputs;
4702 if (Instruction *I = dyn_cast<Instruction>(LF.OperandValToReplace))
4703 Inputs.push_back(I);
4704 if (LU.Kind == LSRUse::ICmpZero)
4705 if (Instruction *I =
4706 dyn_cast<Instruction>(cast<ICmpInst>(LF.UserInst)->getOperand(1)))
4707 Inputs.push_back(I);
4708 if (LF.PostIncLoops.count(L)) {
4709 if (LF.isUseFullyOutsideLoop(L))
4710 Inputs.push_back(L->getLoopLatch()->getTerminator());
4711 else
4712 Inputs.push_back(IVIncInsertPos);
4713 }
4714 // The expansion must also be dominated by the increment positions of any
4715 // loops it for which it is using post-inc mode.
4716 for (const Loop *PIL : LF.PostIncLoops) {
4717 if (PIL == L) continue;
4718
4719 // Be dominated by the loop exit.
4720 SmallVector<BasicBlock *, 4> ExitingBlocks;
4721 PIL->getExitingBlocks(ExitingBlocks);
4722 if (!ExitingBlocks.empty()) {
4723 BasicBlock *BB = ExitingBlocks[0];
4724 for (unsigned i = 1, e = ExitingBlocks.size(); i != e; ++i)
4725 BB = DT.findNearestCommonDominator(BB, ExitingBlocks[i]);
4726 Inputs.push_back(BB->getTerminator());
4727 }
4728 }
4729
4730 assert(!isa<PHINode>(LowestIP) && !LowestIP->isEHPad()((!isa<PHINode>(LowestIP) && !LowestIP->isEHPad
() && !isa<DbgInfoIntrinsic>(LowestIP) &&
"Insertion point must be a normal instruction") ? static_cast
<void> (0) : __assert_fail ("!isa<PHINode>(LowestIP) && !LowestIP->isEHPad() && !isa<DbgInfoIntrinsic>(LowestIP) && \"Insertion point must be a normal instruction\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4732, __PRETTY_FUNCTION__))
4731 && !isa<DbgInfoIntrinsic>(LowestIP) &&((!isa<PHINode>(LowestIP) && !LowestIP->isEHPad
() && !isa<DbgInfoIntrinsic>(LowestIP) &&
"Insertion point must be a normal instruction") ? static_cast
<void> (0) : __assert_fail ("!isa<PHINode>(LowestIP) && !LowestIP->isEHPad() && !isa<DbgInfoIntrinsic>(LowestIP) && \"Insertion point must be a normal instruction\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4732, __PRETTY_FUNCTION__))
4732 "Insertion point must be a normal instruction")((!isa<PHINode>(LowestIP) && !LowestIP->isEHPad
() && !isa<DbgInfoIntrinsic>(LowestIP) &&
"Insertion point must be a normal instruction") ? static_cast
<void> (0) : __assert_fail ("!isa<PHINode>(LowestIP) && !LowestIP->isEHPad() && !isa<DbgInfoIntrinsic>(LowestIP) && \"Insertion point must be a normal instruction\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4732, __PRETTY_FUNCTION__))
;
4733
4734 // Then, climb up the immediate dominator tree as far as we can go while
4735 // still being dominated by the input positions.
4736 BasicBlock::iterator IP = HoistInsertPosition(LowestIP, Inputs);
4737
4738 // Don't insert instructions before PHI nodes.
4739 while (isa<PHINode>(IP)) ++IP;
4740
4741 // Ignore landingpad instructions.
4742 while (IP->isEHPad()) ++IP;
4743
4744 // Ignore debug intrinsics.
4745 while (isa<DbgInfoIntrinsic>(IP)) ++IP;
4746
4747 // Set IP below instructions recently inserted by SCEVExpander. This keeps the
4748 // IP consistent across expansions and allows the previously inserted
4749 // instructions to be reused by subsequent expansion.
4750 while (Rewriter.isInsertedInstruction(&*IP) && IP != LowestIP)
4751 ++IP;
4752
4753 return IP;
4754}
4755
4756/// Emit instructions for the leading candidate expression for this LSRUse (this
4757/// is called "expanding").
4758Value *LSRInstance::Expand(const LSRUse &LU, const LSRFixup &LF,
4759 const Formula &F, BasicBlock::iterator IP,
4760 SCEVExpander &Rewriter,
4761 SmallVectorImpl<WeakTrackingVH> &DeadInsts) const {
4762 if (LU.RigidFormula)
4763 return LF.OperandValToReplace;
4764
4765 // Determine an input position which will be dominated by the operands and
4766 // which will dominate the result.
4767 IP = AdjustInsertPositionForExpand(IP, LF, LU, Rewriter);
4768 Rewriter.setInsertPoint(&*IP);
4769
4770 // Inform the Rewriter if we have a post-increment use, so that it can
4771 // perform an advantageous expansion.
4772 Rewriter.setPostInc(LF.PostIncLoops);
4773
4774 // This is the type that the user actually needs.
4775 Type *OpTy = LF.OperandValToReplace->getType();
4776 // This will be the type that we'll initially expand to.
4777 Type *Ty = F.getType();
4778 if (!Ty)
4779 // No type known; just expand directly to the ultimate type.
4780 Ty = OpTy;
4781 else if (SE.getEffectiveSCEVType(Ty) == SE.getEffectiveSCEVType(OpTy))
4782 // Expand directly to the ultimate type if it's the right size.
4783 Ty = OpTy;
4784 // This is the type to do integer arithmetic in.
4785 Type *IntTy = SE.getEffectiveSCEVType(Ty);
4786
4787 // Build up a list of operands to add together to form the full base.
4788 SmallVector<const SCEV *, 8> Ops;
4789
4790 // Expand the BaseRegs portion.
4791 for (const SCEV *Reg : F.BaseRegs) {
4792 assert(!Reg->isZero() && "Zero allocated in a base register!")((!Reg->isZero() && "Zero allocated in a base register!"
) ? static_cast<void> (0) : __assert_fail ("!Reg->isZero() && \"Zero allocated in a base register!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4792, __PRETTY_FUNCTION__))
;
4793
4794 // If we're expanding for a post-inc user, make the post-inc adjustment.
4795 Reg = denormalizeForPostIncUse(Reg, LF.PostIncLoops, SE);
4796 Ops.push_back(SE.getUnknown(Rewriter.expandCodeFor(Reg, nullptr)));
4797 }
4798
4799 // Expand the ScaledReg portion.
4800 Value *ICmpScaledV = nullptr;
4801 if (F.Scale != 0) {
4802 const SCEV *ScaledS = F.ScaledReg;
4803
4804 // If we're expanding for a post-inc user, make the post-inc adjustment.
4805 PostIncLoopSet &Loops = const_cast<PostIncLoopSet &>(LF.PostIncLoops);
4806 ScaledS = denormalizeForPostIncUse(ScaledS, Loops, SE);
4807
4808 if (LU.Kind == LSRUse::ICmpZero) {
4809 // Expand ScaleReg as if it was part of the base regs.
4810 if (F.Scale == 1)
4811 Ops.push_back(
4812 SE.getUnknown(Rewriter.expandCodeFor(ScaledS, nullptr)));
4813 else {
4814 // An interesting way of "folding" with an icmp is to use a negated
4815 // scale, which we'll implement by inserting it into the other operand
4816 // of the icmp.
4817 assert(F.Scale == -1 &&((F.Scale == -1 && "The only scale supported by ICmpZero uses is -1!"
) ? static_cast<void> (0) : __assert_fail ("F.Scale == -1 && \"The only scale supported by ICmpZero uses is -1!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4818, __PRETTY_FUNCTION__))
4818 "The only scale supported by ICmpZero uses is -1!")((F.Scale == -1 && "The only scale supported by ICmpZero uses is -1!"
) ? static_cast<void> (0) : __assert_fail ("F.Scale == -1 && \"The only scale supported by ICmpZero uses is -1!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4818, __PRETTY_FUNCTION__))
;
4819 ICmpScaledV = Rewriter.expandCodeFor(ScaledS, nullptr);
4820 }
4821 } else {
4822 // Otherwise just expand the scaled register and an explicit scale,
4823 // which is expected to be matched as part of the address.
4824
4825 // Flush the operand list to suppress SCEVExpander hoisting address modes.
4826 // Unless the addressing mode will not be folded.
4827 if (!Ops.empty() && LU.Kind == LSRUse::Address &&
4828 isAMCompletelyFolded(TTI, LU, F)) {
4829 Value *FullV = Rewriter.expandCodeFor(SE.getAddExpr(Ops), Ty);
4830 Ops.clear();
4831 Ops.push_back(SE.getUnknown(FullV));
4832 }
4833 ScaledS = SE.getUnknown(Rewriter.expandCodeFor(ScaledS, nullptr));
4834 if (F.Scale != 1)
4835 ScaledS =
4836 SE.getMulExpr(ScaledS, SE.getConstant(ScaledS->getType(), F.Scale));
4837 Ops.push_back(ScaledS);
4838 }
4839 }
4840
4841 // Expand the GV portion.
4842 if (F.BaseGV) {
4843 // Flush the operand list to suppress SCEVExpander hoisting.
4844 if (!Ops.empty()) {
4845 Value *FullV = Rewriter.expandCodeFor(SE.getAddExpr(Ops), Ty);
4846 Ops.clear();
4847 Ops.push_back(SE.getUnknown(FullV));
4848 }
4849 Ops.push_back(SE.getUnknown(F.BaseGV));
4850 }
4851
4852 // Flush the operand list to suppress SCEVExpander hoisting of both folded and
4853 // unfolded offsets. LSR assumes they both live next to their uses.
4854 if (!Ops.empty()) {
4855 Value *FullV = Rewriter.expandCodeFor(SE.getAddExpr(Ops), Ty);
4856 Ops.clear();
4857 Ops.push_back(SE.getUnknown(FullV));
4858 }
4859
4860 // Expand the immediate portion.
4861 int64_t Offset = (uint64_t)F.BaseOffset + LF.Offset;
4862 if (Offset != 0) {
4863 if (LU.Kind == LSRUse::ICmpZero) {
4864 // The other interesting way of "folding" with an ICmpZero is to use a
4865 // negated immediate.
4866 if (!ICmpScaledV)
4867 ICmpScaledV = ConstantInt::get(IntTy, -(uint64_t)Offset);
4868 else {
4869 Ops.push_back(SE.getUnknown(ICmpScaledV));
4870 ICmpScaledV = ConstantInt::get(IntTy, Offset);
4871 }
4872 } else {
4873 // Just add the immediate values. These again are expected to be matched
4874 // as part of the address.
4875 Ops.push_back(SE.getUnknown(ConstantInt::getSigned(IntTy, Offset)));
4876 }
4877 }
4878
4879 // Expand the unfolded offset portion.
4880 int64_t UnfoldedOffset = F.UnfoldedOffset;
4881 if (UnfoldedOffset != 0) {
4882 // Just add the immediate values.
4883 Ops.push_back(SE.getUnknown(ConstantInt::getSigned(IntTy,
4884 UnfoldedOffset)));
4885 }
4886
4887 // Emit instructions summing all the operands.
4888 const SCEV *FullS = Ops.empty() ?
4889 SE.getConstant(IntTy, 0) :
4890 SE.getAddExpr(Ops);
4891 Value *FullV = Rewriter.expandCodeFor(FullS, Ty);
4892
4893 // We're done expanding now, so reset the rewriter.
4894 Rewriter.clearPostInc();
4895
4896 // An ICmpZero Formula represents an ICmp which we're handling as a
4897 // comparison against zero. Now that we've expanded an expression for that
4898 // form, update the ICmp's other operand.
4899 if (LU.Kind == LSRUse::ICmpZero) {
4900 ICmpInst *CI = cast<ICmpInst>(LF.UserInst);
4901 DeadInsts.emplace_back(CI->getOperand(1));
4902 assert(!F.BaseGV && "ICmp does not support folding a global value and "((!F.BaseGV && "ICmp does not support folding a global value and "
"a scale at the same time!") ? static_cast<void> (0) :
__assert_fail ("!F.BaseGV && \"ICmp does not support folding a global value and \" \"a scale at the same time!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4903, __PRETTY_FUNCTION__))
4903 "a scale at the same time!")((!F.BaseGV && "ICmp does not support folding a global value and "
"a scale at the same time!") ? static_cast<void> (0) :
__assert_fail ("!F.BaseGV && \"ICmp does not support folding a global value and \" \"a scale at the same time!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4903, __PRETTY_FUNCTION__))
;
4904 if (F.Scale == -1) {
4905 if (ICmpScaledV->getType() != OpTy) {
4906 Instruction *Cast =
4907 CastInst::Create(CastInst::getCastOpcode(ICmpScaledV, false,
4908 OpTy, false),
4909 ICmpScaledV, OpTy, "tmp", CI);
4910 ICmpScaledV = Cast;
4911 }
4912 CI->setOperand(1, ICmpScaledV);
4913 } else {
4914 // A scale of 1 means that the scale has been expanded as part of the
4915 // base regs.
4916 assert((F.Scale == 0 || F.Scale == 1) &&(((F.Scale == 0 || F.Scale == 1) && "ICmp does not support folding a global value and "
"a scale at the same time!") ? static_cast<void> (0) :
__assert_fail ("(F.Scale == 0 || F.Scale == 1) && \"ICmp does not support folding a global value and \" \"a scale at the same time!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4918, __PRETTY_FUNCTION__))
4917 "ICmp does not support folding a global value and "(((F.Scale == 0 || F.Scale == 1) && "ICmp does not support folding a global value and "
"a scale at the same time!") ? static_cast<void> (0) :
__assert_fail ("(F.Scale == 0 || F.Scale == 1) && \"ICmp does not support folding a global value and \" \"a scale at the same time!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4918, __PRETTY_FUNCTION__))
4918 "a scale at the same time!")(((F.Scale == 0 || F.Scale == 1) && "ICmp does not support folding a global value and "
"a scale at the same time!") ? static_cast<void> (0) :
__assert_fail ("(F.Scale == 0 || F.Scale == 1) && \"ICmp does not support folding a global value and \" \"a scale at the same time!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 4918, __PRETTY_FUNCTION__))
;
4919 Constant *C = ConstantInt::getSigned(SE.getEffectiveSCEVType(OpTy),
4920 -(uint64_t)Offset);
4921 if (C->getType() != OpTy)
4922 C = ConstantExpr::getCast(CastInst::getCastOpcode(C, false,
4923 OpTy, false),
4924 C, OpTy);
4925
4926 CI->setOperand(1, C);
4927 }
4928 }
4929
4930 return FullV;
4931}
4932
4933/// Helper for Rewrite. PHI nodes are special because the use of their operands
4934/// effectively happens in their predecessor blocks, so the expression may need
4935/// to be expanded in multiple places.
4936void LSRInstance::RewriteForPHI(
4937 PHINode *PN, const LSRUse &LU, const LSRFixup &LF, const Formula &F,
4938 SCEVExpander &Rewriter, SmallVectorImpl<WeakTrackingVH> &DeadInsts) const {
4939 DenseMap<BasicBlock *, Value *> Inserted;
4940 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
4941 if (PN->getIncomingValue(i) == LF.OperandValToReplace) {
4942 BasicBlock *BB = PN->getIncomingBlock(i);
4943
4944 // If this is a critical edge, split the edge so that we do not insert
4945 // the code on all predecessor/successor paths. We do this unless this
4946 // is the canonical backedge for this loop, which complicates post-inc
4947 // users.
4948 if (e != 1 && BB->getTerminator()->getNumSuccessors() > 1 &&
4949 !isa<IndirectBrInst>(BB->getTerminator()) &&
4950 !isa<CatchSwitchInst>(BB->getTerminator())) {
4951 BasicBlock *Parent = PN->getParent();
4952 Loop *PNLoop = LI.getLoopFor(Parent);
4953 if (!PNLoop || Parent != PNLoop->getHeader()) {
4954 // Split the critical edge.
4955 BasicBlock *NewBB = nullptr;
4956 if (!Parent->isLandingPad()) {
4957 NewBB = SplitCriticalEdge(BB, Parent,
4958 CriticalEdgeSplittingOptions(&DT, &LI)
4959 .setMergeIdenticalEdges()
4960 .setDontDeleteUselessPHIs());
4961 } else {
4962 SmallVector<BasicBlock*, 2> NewBBs;
4963 SplitLandingPadPredecessors(Parent, BB, "", "", NewBBs, &DT, &LI);
4964 NewBB = NewBBs[0];
4965 }
4966 // If NewBB==NULL, then SplitCriticalEdge refused to split because all
4967 // phi predecessors are identical. The simple thing to do is skip
4968 // splitting in this case rather than complicate the API.
4969 if (NewBB) {
4970 // If PN is outside of the loop and BB is in the loop, we want to
4971 // move the block to be immediately before the PHI block, not
4972 // immediately after BB.
4973 if (L->contains(BB) && !L->contains(PN))
4974 NewBB->moveBefore(PN->getParent());
4975
4976 // Splitting the edge can reduce the number of PHI entries we have.
4977 e = PN->getNumIncomingValues();
4978 BB = NewBB;
4979 i = PN->getBasicBlockIndex(BB);
4980 }
4981 }
4982 }
4983
4984 std::pair<DenseMap<BasicBlock *, Value *>::iterator, bool> Pair =
4985 Inserted.insert(std::make_pair(BB, static_cast<Value *>(nullptr)));
4986 if (!Pair.second)
4987 PN->setIncomingValue(i, Pair.first->second);
4988 else {
4989 Value *FullV = Expand(LU, LF, F, BB->getTerminator()->getIterator(),
4990 Rewriter, DeadInsts);
4991
4992 // If this is reuse-by-noop-cast, insert the noop cast.
4993 Type *OpTy = LF.OperandValToReplace->getType();
4994 if (FullV->getType() != OpTy)
4995 FullV =
4996 CastInst::Create(CastInst::getCastOpcode(FullV, false,
4997 OpTy, false),
4998 FullV, LF.OperandValToReplace->getType(),
4999 "tmp", BB->getTerminator());
5000
5001 PN->setIncomingValue(i, FullV);
5002 Pair.first->second = FullV;
5003 }
5004 }
5005}
5006
5007/// Emit instructions for the leading candidate expression for this LSRUse (this
5008/// is called "expanding"), and update the UserInst to reference the newly
5009/// expanded value.
5010void LSRInstance::Rewrite(const LSRUse &LU, const LSRFixup &LF,
5011 const Formula &F, SCEVExpander &Rewriter,
5012 SmallVectorImpl<WeakTrackingVH> &DeadInsts) const {
5013 // First, find an insertion point that dominates UserInst. For PHI nodes,
5014 // find the nearest block which dominates all the relevant uses.
5015 if (PHINode *PN = dyn_cast<PHINode>(LF.UserInst)) {
5016 RewriteForPHI(PN, LU, LF, F, Rewriter, DeadInsts);
5017 } else {
5018 Value *FullV =
5019 Expand(LU, LF, F, LF.UserInst->getIterator(), Rewriter, DeadInsts);
5020
5021 // If this is reuse-by-noop-cast, insert the noop cast.
5022 Type *OpTy = LF.OperandValToReplace->getType();
5023 if (FullV->getType() != OpTy) {
5024 Instruction *Cast =
5025 CastInst::Create(CastInst::getCastOpcode(FullV, false, OpTy, false),
5026 FullV, OpTy, "tmp", LF.UserInst);
5027 FullV = Cast;
5028 }
5029
5030 // Update the user. ICmpZero is handled specially here (for now) because
5031 // Expand may have updated one of the operands of the icmp already, and
5032 // its new value may happen to be equal to LF.OperandValToReplace, in
5033 // which case doing replaceUsesOfWith leads to replacing both operands
5034 // with the same value. TODO: Reorganize this.
5035 if (LU.Kind == LSRUse::ICmpZero)
5036 LF.UserInst->setOperand(0, FullV);
5037 else
5038 LF.UserInst->replaceUsesOfWith(LF.OperandValToReplace, FullV);
5039 }
5040
5041 DeadInsts.emplace_back(LF.OperandValToReplace);
5042}
5043
5044/// Rewrite all the fixup locations with new values, following the chosen
5045/// solution.
5046void LSRInstance::ImplementSolution(
5047 const SmallVectorImpl<const Formula *> &Solution) {
5048 // Keep track of instructions we may have made dead, so that
5049 // we can remove them after we are done working.
5050 SmallVector<WeakTrackingVH, 16> DeadInsts;
5051
5052 SCEVExpander Rewriter(SE, L->getHeader()->getModule()->getDataLayout(),
5053 "lsr");
5054#ifndef NDEBUG
5055 Rewriter.setDebugType(DEBUG_TYPE"loop-reduce");
5056#endif
5057 Rewriter.disableCanonicalMode();
5058 Rewriter.enableLSRMode();
5059 Rewriter.setIVIncInsertPos(L, IVIncInsertPos);
5060
5061 // Mark phi nodes that terminate chains so the expander tries to reuse them.
5062 for (const IVChain &Chain : IVChainVec) {
5063 if (PHINode *PN = dyn_cast<PHINode>(Chain.tailUserInst()))
5064 Rewriter.setChainedPhi(PN);
5065 }
5066
5067 // Expand the new value definitions and update the users.
5068 for (size_t LUIdx = 0, NumUses = Uses.size(); LUIdx != NumUses; ++LUIdx)
5069 for (const LSRFixup &Fixup : Uses[LUIdx].Fixups) {
5070 Rewrite(Uses[LUIdx], Fixup, *Solution[LUIdx], Rewriter, DeadInsts);
5071 Changed = true;
5072 }
5073
5074 for (const IVChain &Chain : IVChainVec) {
5075 GenerateIVChain(Chain, Rewriter, DeadInsts);
5076 Changed = true;
5077 }
5078 // Clean up after ourselves. This must be done before deleting any
5079 // instructions.
5080 Rewriter.clear();
5081
5082 Changed |= DeleteTriviallyDeadInstructions(DeadInsts);
5083}
5084
5085LSRInstance::LSRInstance(Loop *L, IVUsers &IU, ScalarEvolution &SE,
5086 DominatorTree &DT, LoopInfo &LI,
5087 const TargetTransformInfo &TTI)
5088 : IU(IU), SE(SE), DT(DT), LI(LI), TTI(TTI), L(L), Changed(false),
5089 IVIncInsertPos(nullptr) {
5090 // If LoopSimplify form is not available, stay out of trouble.
5091 if (!L->isLoopSimplifyForm())
5092 return;
5093
5094 // If there's no interesting work to be done, bail early.
5095 if (IU.empty()) return;
5096
5097 // If there's too much analysis to be done, bail early. We won't be able to
5098 // model the problem anyway.
5099 unsigned NumUsers = 0;
5100 for (const IVStrideUse &U : IU) {
5101 if (++NumUsers > MaxIVUsers) {
5102 (void)U;
5103 DEBUG(dbgs() << "LSR skipping loop, too many IV Users in " << U << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "LSR skipping loop, too many IV Users in "
<< U << "\n"; } } while (false)
;
5104 return;
5105 }
5106 // Bail out if we have a PHI on an EHPad that gets a value from a
5107 // CatchSwitchInst. Because the CatchSwitchInst cannot be split, there is
5108 // no good place to stick any instructions.
5109 if (auto *PN = dyn_cast<PHINode>(U.getUser())) {
5110 auto *FirstNonPHI = PN->getParent()->getFirstNonPHI();
5111 if (isa<FuncletPadInst>(FirstNonPHI) ||
5112 isa<CatchSwitchInst>(FirstNonPHI))
5113 for (BasicBlock *PredBB : PN->blocks())
5114 if (isa<CatchSwitchInst>(PredBB->getFirstNonPHI()))
5115 return;
5116 }
5117 }
5118
5119#ifndef NDEBUG
5120 // All dominating loops must have preheaders, or SCEVExpander may not be able
5121 // to materialize an AddRecExpr whose Start is an outer AddRecExpr.
5122 //
5123 // IVUsers analysis should only create users that are dominated by simple loop
5124 // headers. Since this loop should dominate all of its users, its user list
5125 // should be empty if this loop itself is not within a simple loop nest.
5126 for (DomTreeNode *Rung = DT.getNode(L->getLoopPreheader());
5127 Rung; Rung = Rung->getIDom()) {
5128 BasicBlock *BB = Rung->getBlock();
5129 const Loop *DomLoop = LI.getLoopFor(BB);
5130 if (DomLoop && DomLoop->getHeader() == BB) {
5131 assert(DomLoop->getLoopPreheader() && "LSR needs a simplified loop nest")((DomLoop->getLoopPreheader() && "LSR needs a simplified loop nest"
) ? static_cast<void> (0) : __assert_fail ("DomLoop->getLoopPreheader() && \"LSR needs a simplified loop nest\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 5131, __PRETTY_FUNCTION__))
;
5132 }
5133 }
5134#endif // DEBUG
5135
5136 DEBUG(dbgs() << "\nLSR on loop ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\nLSR on loop "; L->getHeader
()->printAsOperand(dbgs(), false); dbgs() << ":\n"; }
} while (false)
5137 L->getHeader()->printAsOperand(dbgs(), /*PrintType=*/false);do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\nLSR on loop "; L->getHeader
()->printAsOperand(dbgs(), false); dbgs() << ":\n"; }
} while (false)
5138 dbgs() << ":\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "\nLSR on loop "; L->getHeader
()->printAsOperand(dbgs(), false); dbgs() << ":\n"; }
} while (false)
;
5139
5140 // First, perform some low-level loop optimizations.
5141 OptimizeShadowIV();
5142 OptimizeLoopTermCond();
5143
5144 // If loop preparation eliminates all interesting IV users, bail.
5145 if (IU.empty()) return;
5146
5147 // Skip nested loops until we can model them better with formulae.
5148 if (!L->empty()) {
5149 DEBUG(dbgs() << "LSR skipping outer loop " << *L << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "LSR skipping outer loop "
<< *L << "\n"; } } while (false)
;
5150 return;
5151 }
5152
5153 // Start collecting data and preparing for the solver.
5154 CollectChains();
5155 CollectInterestingTypesAndFactors();
5156 CollectFixupsAndInitialFormulae();
5157 CollectLoopInvariantFixupsAndFormulae();
5158
5159 assert(!Uses.empty() && "IVUsers reported at least one use")((!Uses.empty() && "IVUsers reported at least one use"
) ? static_cast<void> (0) : __assert_fail ("!Uses.empty() && \"IVUsers reported at least one use\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 5159, __PRETTY_FUNCTION__))
;
5160 DEBUG(dbgs() << "LSR found " << Uses.size() << " uses:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "LSR found " << Uses
.size() << " uses:\n"; print_uses(dbgs()); } } while (false
)
5161 print_uses(dbgs()))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("loop-reduce")) { dbgs() << "LSR found " << Uses
.size() << " uses:\n"; print_uses(dbgs()); } } while (false
)
;
5162
5163 // Now use the reuse data to generate a bunch of interesting ways
5164 // to formulate the values needed for the uses.
5165 GenerateAllReuseFormulae();
5166
5167 FilterOutUndesirableDedicatedRegisters();
5168 NarrowSearchSpaceUsingHeuristics();
5169
5170 SmallVector<const Formula *, 8> Solution;
5171 Solve(Solution);
5172
5173 // Release memory that is no longer needed.
5174 Factors.clear();
5175 Types.clear();
5176 RegUses.clear();
5177
5178 if (Solution.empty())
5179 return;
5180
5181#ifndef NDEBUG
5182 // Formulae should be legal.
5183 for (const LSRUse &LU : Uses) {
5184 for (const Formula &F : LU.Formulae)
5185 assert(isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy,((isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy
, F) && "Illegal formula generated!") ? static_cast<
void> (0) : __assert_fail ("isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy, F) && \"Illegal formula generated!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 5186, __PRETTY_FUNCTION__))
5186 F) && "Illegal formula generated!")((isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy
, F) && "Illegal formula generated!") ? static_cast<
void> (0) : __assert_fail ("isLegalUse(TTI, LU.MinOffset, LU.MaxOffset, LU.Kind, LU.AccessTy, F) && \"Illegal formula generated!\""
, "/tmp/buildd/llvm-toolchain-snapshot-5.0~svn303373/lib/Transforms/Scalar/LoopStrengthReduce.cpp"
, 5186, __PRETTY_FUNCTION__))
;
5187 };
5188#endif
5189
5190 // Now that we've decided what we want, make it so.
5191 ImplementSolution(Solution);
5192}
5193
5194void LSRInstance::print_factors_and_types(raw_ostream &OS) const {
5195 if (Factors.empty() && Types.empty()) return;
5196
5197 OS << "LSR has identified the following interesting factors and types: ";
5198 bool First = true;
5199
5200 for (int64_t Factor : Factors) {
5201 if (!First) OS << ", ";
5202 First = false;
5203 OS << '*' << Factor;
5204 }
5205
5206 for (Type *Ty : Types) {
5207 if (!First) OS << ", ";
5208 First = false;
5209 OS << '(' << *Ty << ')';
5210 }
5211 OS << '\n';
5212}
5213
5214void LSRInstance::print_fixups(raw_ostream &OS) const {
5215 OS << "LSR is examining the following fixup sites:\n";
5216 for (const LSRUse &LU : Uses)
5217 for (const LSRFixup &LF : LU.Fixups) {
5218 dbgs() << " ";
5219 LF.print(OS);
5220 OS << '\n';
5221 }
5222}
5223
5224void LSRInstance::print_uses(raw_ostream &OS) const {
5225 OS << "LSR is examining the following uses:\n";
5226 for (const LSRUse &LU : Uses) {
5227 dbgs() << " ";
5228 LU.print(OS);
5229 OS << '\n';
5230 for (const Formula &F : LU.Formulae) {
5231 OS << " ";
5232 F.print(OS);
5233 OS << '\n';
5234 }
5235 }
5236}
5237
5238void LSRInstance::print(raw_ostream &OS) const {
5239 print_factors_and_types(OS);
5240 print_fixups(OS);
5241 print_uses(OS);
5242}
5243
5244#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
5245LLVM_DUMP_METHOD__attribute__((noinline)) __attribute__((__used__)) void LSRInstance::dump() const {
5246 print(errs()); errs() << '\n';
5247}
5248#endif
5249
5250namespace {
5251
5252class LoopStrengthReduce : public LoopPass {
5253public:
5254 static char ID; // Pass ID, replacement for typeid
5255
5256 LoopStrengthReduce();
5257
5258private:
5259 bool runOnLoop(Loop *L, LPPassManager &LPM) override;
5260 void getAnalysisUsage(AnalysisUsage &AU) const override;
5261};
5262
5263} // end anonymous namespace
5264
5265LoopStrengthReduce::LoopStrengthReduce() : LoopPass(ID) {
5266 initializeLoopStrengthReducePass(*PassRegistry::getPassRegistry());
5267}
5268
5269void LoopStrengthReduce::getAnalysisUsage(AnalysisUsage &AU) const {
5270 // We split critical edges, so we change the CFG. However, we do update
5271 // many analyses if they are around.
5272 AU.addPreservedID(LoopSimplifyID);
5273
5274 AU.addRequired<LoopInfoWrapperPass>();
5275 AU.addPreserved<LoopInfoWrapperPass>();
5276 AU.addRequiredID(LoopSimplifyID);
5277 AU.addRequired<DominatorTreeWrapperPass>();
5278 AU.addPreserved<DominatorTreeWrapperPass>();
5279 AU.addRequired<ScalarEvolutionWrapperPass>();
5280 AU.addPreserved<ScalarEvolutionWrapperPass>();
5281 // Requiring LoopSimplify a second time here prevents IVUsers from running
5282 // twice, since LoopSimplify was invalidated by running ScalarEvolution.
5283 AU.addRequiredID(LoopSimplifyID);
5284 AU.addRequired<IVUsersWrapperPass>();
5285 AU.addPreserved<IVUsersWrapperPass>();
5286 AU.addRequired<TargetTransformInfoWrapperPass>();
5287}
5288
5289static bool ReduceLoopStrength(Loop *L, IVUsers &IU, ScalarEvolution &SE,
5290 DominatorTree &DT, LoopInfo &LI,
5291 const TargetTransformInfo &TTI) {
5292 bool Changed = false;
5293
5294 // Run the main LSR transformation.
5295 Changed |= LSRInstance(L, IU, SE, DT, LI, TTI).getChanged();
5296
5297 // Remove any extra phis created by processing inner loops.
5298 Changed |= DeleteDeadPHIs(L->getHeader());
5299 if (EnablePhiElim && L->isLoopSimplifyForm()) {
5300 SmallVector<WeakTrackingVH, 16> DeadInsts;
5301 const DataLayout &DL = L->getHeader()->getModule()->getDataLayout();
5302 SCEVExpander Rewriter(SE, DL, "lsr");
5303#ifndef NDEBUG
5304 Rewriter.setDebugType(DEBUG_TYPE"loop-reduce");
5305#endif
5306 unsigned numFolded = Rewriter.replaceCongruentIVs(L, &DT, DeadInsts, &TTI);
5307 if (numFolded) {
5308 Changed = true;
5309 DeleteTriviallyDeadInstructions(DeadInsts);
5310 DeleteDeadPHIs(L->getHeader());
5311 }
5312 }
5313 return Changed;
5314}
5315
5316bool LoopStrengthReduce::runOnLoop(Loop *L, LPPassManager & /*LPM*/) {
5317 if (skipLoop(L))
5318 return false;
5319
5320 auto &IU = getAnalysis<IVUsersWrapperPass>().getIU();
5321 auto &SE = getAnalysis<ScalarEvolutionWrapperPass>().getSE();
5322 auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
5323 auto &LI = getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
5324 const auto &TTI = getAnalysis<TargetTransformInfoWrapperPass>().getTTI(
5325 *L->getHeader()->getParent());
5326 return ReduceLoopStrength(L, IU, SE, DT, LI, TTI);
5327}
5328
5329PreservedAnalyses LoopStrengthReducePass::run(Loop &L, LoopAnalysisManager &AM,
5330 LoopStandardAnalysisResults &AR,
5331 LPMUpdater &) {
5332 if (!ReduceLoopStrength(&L, AM.getResult<IVUsersAnalysis>(L, AR), AR.SE,
5333 AR.DT, AR.LI, AR.TTI))
5334 return PreservedAnalyses::all();
5335
5336 return getLoopPassPreservedAnalyses();
5337}
5338
5339char LoopStrengthReduce::ID = 0;
5340INITIALIZE_PASS_BEGIN(LoopStrengthReduce, "loop-reduce",static void *initializeLoopStrengthReducePassOnce(PassRegistry
&Registry) {
5341 "Loop Strength Reduction", false, false)static void *initializeLoopStrengthReducePassOnce(PassRegistry
&Registry) {
5342INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)initializeTargetTransformInfoWrapperPassPass(Registry);
5343INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)initializeDominatorTreeWrapperPassPass(Registry);
5344INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass)initializeScalarEvolutionWrapperPassPass(Registry);
5345INITIALIZE_PASS_DEPENDENCY(IVUsersWrapperPass)initializeIVUsersWrapperPassPass(Registry);
5346INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)initializeLoopInfoWrapperPassPass(Registry);
5347INITIALIZE_PASS_DEPENDENCY(LoopSimplify)initializeLoopSimplifyPass(Registry);
5348INITIALIZE_PASS_END(LoopStrengthReduce, "loop-reduce",PassInfo *PI = new PassInfo( "Loop Strength Reduction", "loop-reduce"
, &LoopStrengthReduce::ID, PassInfo::NormalCtor_t(callDefaultCtor
<LoopStrengthReduce>), false, false); Registry.registerPass
(*PI, true); return PI; } static llvm::once_flag InitializeLoopStrengthReducePassFlag
; void llvm::initializeLoopStrengthReducePass(PassRegistry &
Registry) { llvm::call_once(InitializeLoopStrengthReducePassFlag
, initializeLoopStrengthReducePassOnce, std::ref(Registry)); }
5349 "Loop Strength Reduction", false, false)PassInfo *PI = new PassInfo( "Loop Strength Reduction", "loop-reduce"
, &LoopStrengthReduce::ID, PassInfo::NormalCtor_t(callDefaultCtor
<LoopStrengthReduce>), false, false); Registry.registerPass
(*PI, true); return PI; } static llvm::once_flag InitializeLoopStrengthReducePassFlag
; void llvm::initializeLoopStrengthReducePass(PassRegistry &
Registry) { llvm::call_once(InitializeLoopStrengthReducePassFlag
, initializeLoopStrengthReducePassOnce, std::ref(Registry)); }
5350
5351Pass *llvm::createLoopStrengthReducePass() { return new LoopStrengthReduce(); }