LoopIdiomRecognize.cpp revision 95ae676bc89b4cb9166576b74f1220ab5b0ff0ad 
1//===-- LoopIdiomRecognize.cpp - Loop idiom recognition -------------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This pass implements an idiom recognizer that transforms simple loops into a
11// non-loop form.  In cases that this kicks in, it can be a significant
12// performance win.
13//
14//===----------------------------------------------------------------------===//
15
16#define DEBUG_TYPE "loop-idiom"
17#include "llvm/Transforms/Scalar.h"
18#include "llvm/Analysis/AliasAnalysis.h"
19#include "llvm/Analysis/LoopPass.h"
20#include "llvm/Analysis/ScalarEvolutionExpressions.h"
21#include "llvm/Analysis/ScalarEvolutionExpander.h"
22#include "llvm/Analysis/ValueTracking.h"
23#include "llvm/Target/TargetData.h"
24#include "llvm/Transforms/Utils/Local.h"
25#include "llvm/Support/Debug.h"
26#include "llvm/Support/IRBuilder.h"
27#include "llvm/Support/raw_ostream.h"
28using namespace llvm;
29
30// TODO: Recognize "N" size array multiplies: replace with call to blas or
31// something.
32
33namespace {
34  class LoopIdiomRecognize : public LoopPass {
35    Loop *CurLoop;
36    const TargetData *TD;
37    ScalarEvolution *SE;
38  public:
39    static char ID;
40    explicit LoopIdiomRecognize() : LoopPass(ID) {
41      initializeLoopIdiomRecognizePass(*PassRegistry::getPassRegistry());
42    }
43
44    bool runOnLoop(Loop *L, LPPassManager &LPM);
45
46    bool processLoopStore(StoreInst *SI, const SCEV *BECount);
47
48    bool processLoopStoreOfSplatValue(StoreInst *SI, unsigned StoreSize,
49                                      Value *SplatValue,
50                                      const SCEVAddRecExpr *Ev,
51                                      const SCEV *BECount);
52
53    /// This transformation requires natural loop information & requires that
54    /// loop preheaders be inserted into the CFG.
55    ///
56    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
57      AU.addRequired<LoopInfo>();
58      AU.addPreserved<LoopInfo>();
59      AU.addRequiredID(LoopSimplifyID);
60      AU.addPreservedID(LoopSimplifyID);
61      AU.addRequiredID(LCSSAID);
62      AU.addPreservedID(LCSSAID);
63      AU.addRequired<AliasAnalysis>();
64      AU.addPreserved<AliasAnalysis>();
65      AU.addRequired<ScalarEvolution>();
66      AU.addPreserved<ScalarEvolution>();
67      AU.addPreserved<DominatorTree>();
68    }
69  };
70}
71
72char LoopIdiomRecognize::ID = 0;
73INITIALIZE_PASS_BEGIN(LoopIdiomRecognize, "loop-idiom", "Recognize loop idioms",
74                      false, false)
75INITIALIZE_PASS_DEPENDENCY(LoopInfo)
76INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
77INITIALIZE_PASS_DEPENDENCY(LCSSA)
78INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
79INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
80INITIALIZE_PASS_END(LoopIdiomRecognize, "loop-idiom", "Recognize loop idioms",
81                    false, false)
82
83Pass *llvm::createLoopIdiomPass() { return new LoopIdiomRecognize(); }
84
85/// DeleteDeadInstruction - Delete this instruction.  Before we do, go through
86/// and zero out all the operands of this instruction.  If any of them become
87/// dead, delete them and the computation tree that feeds them.
88///
89static void DeleteDeadInstruction(Instruction *I, ScalarEvolution &SE) {
90  SmallVector<Instruction*, 32> NowDeadInsts;
91
92  NowDeadInsts.push_back(I);
93
94  // Before we touch this instruction, remove it from SE!
95  do {
96    Instruction *DeadInst = NowDeadInsts.pop_back_val();
97
98    // This instruction is dead, zap it, in stages.  Start by removing it from
99    // SCEV.
100    SE.forgetValue(DeadInst);
101
102    for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) {
103      Value *Op = DeadInst->getOperand(op);
104      DeadInst->setOperand(op, 0);
105
106      // If this operand just became dead, add it to the NowDeadInsts list.
107      if (!Op->use_empty()) continue;
108
109      if (Instruction *OpI = dyn_cast<Instruction>(Op))
110        if (isInstructionTriviallyDead(OpI))
111          NowDeadInsts.push_back(OpI);
112    }
113
114    DeadInst->eraseFromParent();
115
116  } while (!NowDeadInsts.empty());
117}
118
119bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) {
120  CurLoop = L;
121
122  // We only look at trivial single basic block loops.
123  // TODO: eventually support more complex loops, scanning the header.
124  if (L->getBlocks().size() != 1)
125    return false;
126
127  // The trip count of the loop must be analyzable.
128  SE = &getAnalysis<ScalarEvolution>();
129  if (!SE->hasLoopInvariantBackedgeTakenCount(L))
130    return false;
131  const SCEV *BECount = SE->getBackedgeTakenCount(L);
132  if (isa<SCEVCouldNotCompute>(BECount)) return false;
133
134  // We require target data for now.
135  TD = getAnalysisIfAvailable<TargetData>();
136  if (TD == 0) return false;
137
138  BasicBlock *BB = L->getHeader();
139  DEBUG(dbgs() << "loop-idiom Scanning: F[" << BB->getParent()->getName()
140               << "] Loop %" << BB->getName() << "\n");
141
142  bool MadeChange = false;
143  for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ) {
144    // Look for store instructions, which may be memsets.
145    StoreInst *SI = dyn_cast<StoreInst>(I++);
146    if (SI == 0 || SI->isVolatile()) continue;
147
148    WeakVH InstPtr(SI);
149    if (!processLoopStore(SI, BECount)) continue;
150
151    MadeChange = true;
152
153    // If processing the store invalidated our iterator, start over from the
154    // head of the loop.
155    if (InstPtr == 0)
156      I = BB->begin();
157  }
158
159  return MadeChange;
160}
161
162/// scanBlock - Look over a block to see if we can promote anything out of it.
163bool LoopIdiomRecognize::processLoopStore(StoreInst *SI, const SCEV *BECount) {
164  Value *StoredVal = SI->getValueOperand();
165  Value *StorePtr = SI->getPointerOperand();
166
167  // Reject stores that are so large that they overflow an unsigned.
168  uint64_t SizeInBits = TD->getTypeSizeInBits(StoredVal->getType());
169  if ((SizeInBits & 7) || (SizeInBits >> 32) != 0)
170    return false;
171
172  // See if the pointer expression is an AddRec like {base,+,1} on the current
173  // loop, which indicates a strided store.  If we have something else, it's a
174  // random store we can't handle.
175  const SCEVAddRecExpr *Ev = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(StorePtr));
176  if (Ev == 0 || Ev->getLoop() != CurLoop || !Ev->isAffine())
177    return false;
178
179  // Check to see if the stride matches the size of the store.  If so, then we
180  // know that every byte is touched in the loop.
181  unsigned StoreSize = (unsigned)SizeInBits >> 3;
182  const SCEVConstant *Stride = dyn_cast<SCEVConstant>(Ev->getOperand(1));
183  if (Stride == 0 || StoreSize != Stride->getValue()->getValue())
184    return false;
185
186  // If the stored value is a byte-wise value (like i32 -1), then it may be
187  // turned into a memset of i8 -1, assuming that all the consequtive bytes
188  // are stored.  A store of i32 0x01020304 can never be turned into a memset.
189  if (Value *SplatValue = isBytewiseValue(StoredVal))
190    return processLoopStoreOfSplatValue(SI, StoreSize, SplatValue, Ev, BECount);
191
192  // Handle the memcpy case here.
193  errs() << "Found strided store: " << *Ev << "\n";
194
195
196  return false;
197}
198
199/// processLoopStoreOfSplatValue - We see a strided store of a memsetable value.
200/// If we can transform this into a memset in the loop preheader, do so.
201bool LoopIdiomRecognize::
202processLoopStoreOfSplatValue(StoreInst *SI, unsigned StoreSize,
203                             Value *SplatValue,
204                             const SCEVAddRecExpr *Ev, const SCEV *BECount) {
205  // Okay, we have a strided store "p[i]" of a splattable value.  We can turn
206  // this into a memset in the loop preheader now if we want.  However, this
207  // would be unsafe to do if there is anything else in the loop that may read
208  // or write to the aliased location.  Check for an alias.
209
210  // FIXME: Need to get a base pointer that is valid.
211  //  if (LoopCanModRefLocation(SI->getPointerOperand())
212
213
214  // FIXME: TODO safety check.
215
216  // Okay, everything looks good, insert the memset.
217  BasicBlock *Preheader = CurLoop->getLoopPreheader();
218
219  IRBuilder<> Builder(Preheader->getTerminator());
220
221  // The trip count of the loop and the base pointer of the addrec SCEV is
222  // guaranteed to be loop invariant, which means that it should dominate the
223  // header.  Just insert code for it in the preheader.
224  SCEVExpander Expander(*SE);
225
226  unsigned AddrSpace = SI->getPointerAddressSpace();
227  Value *BasePtr =
228    Expander.expandCodeFor(Ev->getStart(), Builder.getInt8PtrTy(AddrSpace),
229                           Preheader->getTerminator());
230
231  // The # stored bytes is (BECount+1)*Size.  Expand the trip count out to
232  // pointer size if it isn't already.
233  const Type *IntPtr = TD->getIntPtrType(SI->getContext());
234  unsigned BESize = SE->getTypeSizeInBits(BECount->getType());
235  if (BESize < TD->getPointerSizeInBits())
236    BECount = SE->getZeroExtendExpr(BECount, IntPtr);
237  else if (BESize > TD->getPointerSizeInBits())
238    BECount = SE->getTruncateExpr(BECount, IntPtr);
239
240  const SCEV *NumBytesS = SE->getAddExpr(BECount, SE->getConstant(IntPtr, 1),
241                                         true, true /*nooverflow*/);
242  if (StoreSize != 1)
243    NumBytesS = SE->getMulExpr(NumBytesS, SE->getConstant(IntPtr, StoreSize),
244                               true, true /*nooverflow*/);
245
246  Value *NumBytes =
247    Expander.expandCodeFor(NumBytesS, IntPtr, Preheader->getTerminator());
248
249  Value *NewCall =
250    Builder.CreateMemSet(BasePtr, SplatValue, NumBytes, SI->getAlignment());
251
252  DEBUG(dbgs() << "  Formed memset: " << *NewCall << "\n"
253               << "    from store to: " << *Ev << " at: " << *SI << "\n");
254  (void)NewCall;
255
256  // Okay, the memset has been formed.  Zap the original store and anything that
257  // feeds into it.
258  DeleteDeadInstruction(SI, *SE);
259  return true;
260}
261
262