aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Transforms/Scalar/Sink.cpp
blob: 903b675fdd56e3ae56e166677801ae5b3e1d4073 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
//===-- Sink.cpp - Code Sinking -------------------------------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This pass moves instructions into successor blocks, when possible, so that
// they aren't executed on paths where their results aren't needed.
//
//===----------------------------------------------------------------------===//

#include "llvm/Transforms/Scalar.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/CFG.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;

#define DEBUG_TYPE "sink"

STATISTIC(NumSunk, "Number of instructions sunk");
STATISTIC(NumSinkIter, "Number of sinking iterations");

namespace {
  class Sinking : public FunctionPass {
    DominatorTree *DT;
    LoopInfo *LI;
    AliasAnalysis *AA;
    const DataLayout *DL;

  public:
    static char ID; // Pass identification
    Sinking() : FunctionPass(ID) {
      initializeSinkingPass(*PassRegistry::getPassRegistry());
    }

    bool runOnFunction(Function &F) override;

    void getAnalysisUsage(AnalysisUsage &AU) const override {
      AU.setPreservesCFG();
      FunctionPass::getAnalysisUsage(AU);
      AU.addRequired<AliasAnalysis>();
      AU.addRequired<DominatorTreeWrapperPass>();
      AU.addRequired<LoopInfo>();
      AU.addPreserved<DominatorTreeWrapperPass>();
      AU.addPreserved<LoopInfo>();
    }
  private:
    bool ProcessBlock(BasicBlock &BB);
    bool SinkInstruction(Instruction *I, SmallPtrSetImpl<Instruction*> &Stores);
    bool AllUsesDominatedByBlock(Instruction *Inst, BasicBlock *BB) const;
    bool IsAcceptableTarget(Instruction *Inst, BasicBlock *SuccToSinkTo) const;
  };
} // end anonymous namespace

char Sinking::ID = 0;
INITIALIZE_PASS_BEGIN(Sinking, "sink", "Code sinking", false, false)
INITIALIZE_PASS_DEPENDENCY(LoopInfo)
INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(Sinking, "sink", "Code sinking", false, false)

FunctionPass *llvm::createSinkingPass() { return new Sinking(); }

/// AllUsesDominatedByBlock - Return true if all uses of the specified value
/// occur in blocks dominated by the specified block.
bool Sinking::AllUsesDominatedByBlock(Instruction *Inst,
                                      BasicBlock *BB) const {
  // Ignoring debug uses is necessary so debug info doesn't affect the code.
  // This may leave a referencing dbg_value in the original block, before
  // the definition of the vreg.  Dwarf generator handles this although the
  // user might not get the right info at runtime.
  for (Use &U : Inst->uses()) {
    // Determine the block of the use.
    Instruction *UseInst = cast<Instruction>(U.getUser());
    BasicBlock *UseBlock = UseInst->getParent();
    if (PHINode *PN = dyn_cast<PHINode>(UseInst)) {
      // PHI nodes use the operand in the predecessor block, not the block with
      // the PHI.
      unsigned Num = PHINode::getIncomingValueNumForOperand(U.getOperandNo());
      UseBlock = PN->getIncomingBlock(Num);
    }
    // Check that it dominates.
    if (!DT->dominates(BB, UseBlock))
      return false;
  }
  return true;
}

bool Sinking::runOnFunction(Function &F) {
  DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
  LI = &getAnalysis<LoopInfo>();
  AA = &getAnalysis<AliasAnalysis>();
  DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
  DL = DLP ? &DLP->getDataLayout() : nullptr;

  bool MadeChange, EverMadeChange = false;

  do {
    MadeChange = false;
    DEBUG(dbgs() << "Sinking iteration " << NumSinkIter << "\n");
    // Process all basic blocks.
    for (Function::iterator I = F.begin(), E = F.end();
         I != E; ++I)
      MadeChange |= ProcessBlock(*I);
    EverMadeChange |= MadeChange;
    NumSinkIter++;
  } while (MadeChange);

  return EverMadeChange;
}

bool Sinking::ProcessBlock(BasicBlock &BB) {
  // Can't sink anything out of a block that has less than two successors.
  if (BB.getTerminator()->getNumSuccessors() <= 1 || BB.empty()) return false;

  // Don't bother sinking code out of unreachable blocks. In addition to being
  // unprofitable, it can also lead to infinite looping, because in an
  // unreachable loop there may be nowhere to stop.
  if (!DT->isReachableFromEntry(&BB)) return false;

  bool MadeChange = false;

  // Walk the basic block bottom-up.  Remember if we saw a store.
  BasicBlock::iterator I = BB.end();
  --I;
  bool ProcessedBegin = false;
  SmallPtrSet<Instruction *, 8> Stores;
  do {
    Instruction *Inst = I;  // The instruction to sink.

    // Predecrement I (if it's not begin) so that it isn't invalidated by
    // sinking.
    ProcessedBegin = I == BB.begin();
    if (!ProcessedBegin)
      --I;

    if (isa<DbgInfoIntrinsic>(Inst))
      continue;

    if (SinkInstruction(Inst, Stores))
      ++NumSunk, MadeChange = true;

    // If we just processed the first instruction in the block, we're done.
  } while (!ProcessedBegin);

  return MadeChange;
}

static bool isSafeToMove(Instruction *Inst, AliasAnalysis *AA,
                         SmallPtrSetImpl<Instruction *> &Stores) {

  if (Inst->mayWriteToMemory()) {
    Stores.insert(Inst);
    return false;
  }

  if (LoadInst *L = dyn_cast<LoadInst>(Inst)) {
    AliasAnalysis::Location Loc = AA->getLocation(L);
    for (Instruction *S : Stores)
      if (AA->getModRefInfo(S, Loc) & AliasAnalysis::Mod)
        return false;
  }

  if (isa<TerminatorInst>(Inst) || isa<PHINode>(Inst))
    return false;

  return true;
}

/// IsAcceptableTarget - Return true if it is possible to sink the instruction
/// in the specified basic block.
bool Sinking::IsAcceptableTarget(Instruction *Inst,
                                 BasicBlock *SuccToSinkTo) const {
  assert(Inst && "Instruction to be sunk is null");
  assert(SuccToSinkTo && "Candidate sink target is null");

  // It is not possible to sink an instruction into its own block.  This can
  // happen with loops.
  if (Inst->getParent() == SuccToSinkTo)
    return false;

  // If the block has multiple predecessors, this would introduce computation
  // on different code paths.  We could split the critical edge, but for now we
  // just punt.
  // FIXME: Split critical edges if not backedges.
  if (SuccToSinkTo->getUniquePredecessor() != Inst->getParent()) {
    // We cannot sink a load across a critical edge - there may be stores in
    // other code paths.
    if (!isSafeToSpeculativelyExecute(Inst, DL))
      return false;

    // We don't want to sink across a critical edge if we don't dominate the
    // successor. We could be introducing calculations to new code paths.
    if (!DT->dominates(Inst->getParent(), SuccToSinkTo))
      return false;

    // Don't sink instructions into a loop.
    Loop *succ = LI->getLoopFor(SuccToSinkTo);
    Loop *cur = LI->getLoopFor(Inst->getParent());
    if (succ != nullptr && succ != cur)
      return false;
  }

  // Finally, check that all the uses of the instruction are actually
  // dominated by the candidate
  return AllUsesDominatedByBlock(Inst, SuccToSinkTo);
}

/// SinkInstruction - Determine whether it is safe to sink the specified machine
/// instruction out of its current block into a successor.
bool Sinking::SinkInstruction(Instruction *Inst,
                              SmallPtrSetImpl<Instruction *> &Stores) {

  // Don't sink static alloca instructions.  CodeGen assumes allocas outside the
  // entry block are dynamically sized stack objects.
  if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
    if (AI->isStaticAlloca())
      return false;

  // Check if it's safe to move the instruction.
  if (!isSafeToMove(Inst, AA, Stores))
    return false;

  // FIXME: This should include support for sinking instructions within the
  // block they are currently in to shorten the live ranges.  We often get
  // instructions sunk into the top of a large block, but it would be better to
  // also sink them down before their first use in the block.  This xform has to
  // be careful not to *increase* register pressure though, e.g. sinking
  // "x = y + z" down if it kills y and z would increase the live ranges of y
  // and z and only shrink the live range of x.

  // SuccToSinkTo - This is the successor to sink this instruction to, once we
  // decide.
  BasicBlock *SuccToSinkTo = nullptr;

  // Instructions can only be sunk if all their uses are in blocks
  // dominated by one of the successors.
  // Look at all the postdominators and see if we can sink it in one.
  DomTreeNode *DTN = DT->getNode(Inst->getParent());
  for (DomTreeNode::iterator I = DTN->begin(), E = DTN->end();
      I != E && SuccToSinkTo == nullptr; ++I) {
    BasicBlock *Candidate = (*I)->getBlock();
    if ((*I)->getIDom()->getBlock() == Inst->getParent() &&
        IsAcceptableTarget(Inst, Candidate))
      SuccToSinkTo = Candidate;
  }

  // If no suitable postdominator was found, look at all the successors and
  // decide which one we should sink to, if any.
  for (succ_iterator I = succ_begin(Inst->getParent()),
      E = succ_end(Inst->getParent()); I != E && !SuccToSinkTo; ++I) {
    if (IsAcceptableTarget(Inst, *I))
      SuccToSinkTo = *I;
  }

  // If we couldn't find a block to sink to, ignore this instruction.
  if (!SuccToSinkTo)
    return false;

  DEBUG(dbgs() << "Sink" << *Inst << " (";
        Inst->getParent()->printAsOperand(dbgs(), false);
        dbgs() << " -> ";
        SuccToSinkTo->printAsOperand(dbgs(), false);
        dbgs() << ")\n");

  // Move the instruction.
  Inst->moveBefore(SuccToSinkTo->getFirstInsertionPt());
  return true;
}