PeepholeOptimizer.cpp revision 40a5eb18b031fa1a5e9697e21e251e613d441cc5 
1//===-- PeepholeOptimizer.cpp - Peephole Optimizations --------------------===//
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// Perform peephole optimizations on the machine code:
11//
12// - Optimize Extensions
13//
14//     Optimization of sign / zero extension instructions. It may be extended to
15//     handle other instructions with similar properties.
16//
17//     On some targets, some instructions, e.g. X86 sign / zero extension, may
18//     leave the source value in the lower part of the result. This optimization
19//     will replace some uses of the pre-extension value with uses of the
20//     sub-register of the results.
21//
22// - Optimize Comparisons
23//
24//     Optimization of comparison instructions. For instance, in this code:
25//
26//       sub r1, 1
27//       cmp r1, 0
28//       bz  L1
29//
30//     If the "sub" instruction all ready sets (or could be modified to set) the
31//     same flag that the "cmp" instruction sets and that "bz" uses, then we can
32//     eliminate the "cmp" instruction.
33//
34//===----------------------------------------------------------------------===//
35
36#define DEBUG_TYPE "peephole-opt"
37#include "llvm/CodeGen/Passes.h"
38#include "llvm/CodeGen/MachineDominators.h"
39#include "llvm/CodeGen/MachineInstrBuilder.h"
40#include "llvm/CodeGen/MachineRegisterInfo.h"
41#include "llvm/Target/TargetInstrInfo.h"
42#include "llvm/Target/TargetRegisterInfo.h"
43#include "llvm/Support/CommandLine.h"
44#include "llvm/ADT/SmallPtrSet.h"
45#include "llvm/ADT/Statistic.h"
46using namespace llvm;
47
48// Optimize Extensions
49static cl::opt<bool>
50Aggressive("aggressive-ext-opt", cl::Hidden,
51           cl::desc("Aggressive extension optimization"));
52
53static cl::opt<bool>
54DisablePeephole("disable-peephole", cl::Hidden, cl::init(false),
55                cl::desc("Disable the peephole optimizer"));
56
57STATISTIC(NumReuse,      "Number of extension results reused");
58STATISTIC(NumEliminated, "Number of compares eliminated");
59
60namespace {
61  class PeepholeOptimizer : public MachineFunctionPass {
62    const TargetMachine   *TM;
63    const TargetInstrInfo *TII;
64    MachineRegisterInfo   *MRI;
65    MachineDominatorTree  *DT;  // Machine dominator tree
66
67  public:
68    static char ID; // Pass identification
69    PeepholeOptimizer() : MachineFunctionPass(ID) {
70      initializePeepholeOptimizerPass(*PassRegistry::getPassRegistry());
71    }
72
73    virtual bool runOnMachineFunction(MachineFunction &MF);
74
75    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
76      AU.setPreservesCFG();
77      MachineFunctionPass::getAnalysisUsage(AU);
78      if (Aggressive) {
79        AU.addRequired<MachineDominatorTree>();
80        AU.addPreserved<MachineDominatorTree>();
81      }
82    }
83
84  private:
85    bool OptimizeCmpInstr(MachineInstr *MI, MachineBasicBlock *MBB,
86                          MachineBasicBlock::iterator &MII);
87    bool OptimizeExtInstr(MachineInstr *MI, MachineBasicBlock *MBB,
88                          SmallPtrSet<MachineInstr*, 8> &LocalMIs);
89  };
90}
91
92char PeepholeOptimizer::ID = 0;
93INITIALIZE_PASS_BEGIN(PeepholeOptimizer, "peephole-opts",
94                "Peephole Optimizations", false, false)
95INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
96INITIALIZE_PASS_END(PeepholeOptimizer, "peephole-opts",
97                "Peephole Optimizations", false, false)
98
99FunctionPass *llvm::createPeepholeOptimizerPass() {
100  return new PeepholeOptimizer();
101}
102
103/// OptimizeExtInstr - If instruction is a copy-like instruction, i.e. it reads
104/// a single register and writes a single register and it does not modify the
105/// source, and if the source value is preserved as a sub-register of the
106/// result, then replace all reachable uses of the source with the subreg of the
107/// result.
108///
109/// Do not generate an EXTRACT that is used only in a debug use, as this changes
110/// the code. Since this code does not currently share EXTRACTs, just ignore all
111/// debug uses.
112bool PeepholeOptimizer::
113OptimizeExtInstr(MachineInstr *MI, MachineBasicBlock *MBB,
114                 SmallPtrSet<MachineInstr*, 8> &LocalMIs) {
115  LocalMIs.insert(MI);
116
117  unsigned SrcReg, DstReg, SubIdx;
118  if (!TII->isCoalescableExtInstr(*MI, SrcReg, DstReg, SubIdx))
119    return false;
120
121  if (TargetRegisterInfo::isPhysicalRegister(DstReg) ||
122      TargetRegisterInfo::isPhysicalRegister(SrcReg))
123    return false;
124
125  MachineRegisterInfo::use_nodbg_iterator UI = MRI->use_nodbg_begin(SrcReg);
126  if (++UI == MRI->use_nodbg_end())
127    // No other uses.
128    return false;
129
130  // The source has other uses. See if we can replace the other uses with use of
131  // the result of the extension.
132  SmallPtrSet<MachineBasicBlock*, 4> ReachedBBs;
133  UI = MRI->use_nodbg_begin(DstReg);
134  for (MachineRegisterInfo::use_nodbg_iterator UE = MRI->use_nodbg_end();
135       UI != UE; ++UI)
136    ReachedBBs.insert(UI->getParent());
137
138  // Uses that are in the same BB of uses of the result of the instruction.
139  SmallVector<MachineOperand*, 8> Uses;
140
141  // Uses that the result of the instruction can reach.
142  SmallVector<MachineOperand*, 8> ExtendedUses;
143
144  bool ExtendLife = true;
145  UI = MRI->use_nodbg_begin(SrcReg);
146  for (MachineRegisterInfo::use_nodbg_iterator UE = MRI->use_nodbg_end();
147       UI != UE; ++UI) {
148    MachineOperand &UseMO = UI.getOperand();
149    MachineInstr *UseMI = &*UI;
150    if (UseMI == MI)
151      continue;
152
153    if (UseMI->isPHI()) {
154      ExtendLife = false;
155      continue;
156    }
157
158    // It's an error to translate this:
159    //
160    //    %reg1025 = <sext> %reg1024
161    //     ...
162    //    %reg1026 = SUBREG_TO_REG 0, %reg1024, 4
163    //
164    // into this:
165    //
166    //    %reg1025 = <sext> %reg1024
167    //     ...
168    //    %reg1027 = COPY %reg1025:4
169    //    %reg1026 = SUBREG_TO_REG 0, %reg1027, 4
170    //
171    // The problem here is that SUBREG_TO_REG is there to assert that an
172    // implicit zext occurs. It doesn't insert a zext instruction. If we allow
173    // the COPY here, it will give us the value after the <sext>, not the
174    // original value of %reg1024 before <sext>.
175    if (UseMI->getOpcode() == TargetOpcode::SUBREG_TO_REG)
176      continue;
177
178    MachineBasicBlock *UseMBB = UseMI->getParent();
179    if (UseMBB == MBB) {
180      // Local uses that come after the extension.
181      if (!LocalMIs.count(UseMI))
182        Uses.push_back(&UseMO);
183    } else if (ReachedBBs.count(UseMBB)) {
184      // Non-local uses where the result of the extension is used. Always
185      // replace these unless it's a PHI.
186      Uses.push_back(&UseMO);
187    } else if (Aggressive && DT->dominates(MBB, UseMBB)) {
188      // We may want to extend the live range of the extension result in order
189      // to replace these uses.
190      ExtendedUses.push_back(&UseMO);
191    } else {
192      // Both will be live out of the def MBB anyway. Don't extend live range of
193      // the extension result.
194      ExtendLife = false;
195      break;
196    }
197  }
198
199  if (ExtendLife && !ExtendedUses.empty())
200    // Extend the liveness of the extension result.
201    std::copy(ExtendedUses.begin(), ExtendedUses.end(),
202              std::back_inserter(Uses));
203
204  // Now replace all uses.
205  bool Changed = false;
206  if (!Uses.empty()) {
207    SmallPtrSet<MachineBasicBlock*, 4> PHIBBs;
208
209    // Look for PHI uses of the extended result, we don't want to extend the
210    // liveness of a PHI input. It breaks all kinds of assumptions down
211    // stream. A PHI use is expected to be the kill of its source values.
212    UI = MRI->use_nodbg_begin(DstReg);
213    for (MachineRegisterInfo::use_nodbg_iterator
214           UE = MRI->use_nodbg_end(); UI != UE; ++UI)
215      if (UI->isPHI())
216        PHIBBs.insert(UI->getParent());
217
218    const TargetRegisterClass *RC = MRI->getRegClass(SrcReg);
219    for (unsigned i = 0, e = Uses.size(); i != e; ++i) {
220      MachineOperand *UseMO = Uses[i];
221      MachineInstr *UseMI = UseMO->getParent();
222      MachineBasicBlock *UseMBB = UseMI->getParent();
223      if (PHIBBs.count(UseMBB))
224        continue;
225
226      unsigned NewVR = MRI->createVirtualRegister(RC);
227      BuildMI(*UseMBB, UseMI, UseMI->getDebugLoc(),
228              TII->get(TargetOpcode::COPY), NewVR)
229        .addReg(DstReg, 0, SubIdx);
230
231      UseMO->setReg(NewVR);
232      ++NumReuse;
233      Changed = true;
234    }
235  }
236
237  return Changed;
238}
239
240/// OptimizeCmpInstr - If the instruction is a compare and the previous
241/// instruction it's comparing against all ready sets (or could be modified to
242/// set) the same flag as the compare, then we can remove the comparison and use
243/// the flag from the previous instruction.
244bool PeepholeOptimizer::OptimizeCmpInstr(MachineInstr *MI,
245                                         MachineBasicBlock *MBB,
246                                         MachineBasicBlock::iterator &NextIter){
247  // If this instruction is a comparison against zero and isn't comparing a
248  // physical register, we can try to optimize it.
249  unsigned SrcReg;
250  int CmpMask, CmpValue;
251  if (!TII->AnalyzeCompare(MI, SrcReg, CmpMask, CmpValue) ||
252      TargetRegisterInfo::isPhysicalRegister(SrcReg))
253    return false;
254
255  // Attempt to optimize the comparison instruction.
256  if (TII->OptimizeCompareInstr(MI, SrcReg, CmpMask, CmpValue, MRI, NextIter)) {
257    ++NumEliminated;
258    return true;
259  }
260
261  return false;
262}
263
264bool PeepholeOptimizer::runOnMachineFunction(MachineFunction &MF) {
265  TM  = &MF.getTarget();
266  TII = TM->getInstrInfo();
267  MRI = &MF.getRegInfo();
268  DT  = Aggressive ? &getAnalysis<MachineDominatorTree>() : 0;
269
270  bool Changed = false;
271
272  SmallPtrSet<MachineInstr*, 8> LocalMIs;
273  for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
274    MachineBasicBlock *MBB = &*I;
275    LocalMIs.clear();
276
277    for (MachineBasicBlock::iterator
278           MII = I->begin(), MIE = I->end(); MII != MIE; ) {
279      MachineInstr *MI = &*MII;
280
281      if (MI->getDesc().isCompare() &&
282          !MI->getDesc().hasUnmodeledSideEffects()) {
283        if (!DisablePeephole && OptimizeCmpInstr(MI, MBB, MII))
284          Changed = true;
285        else
286          ++MII;
287      } else {
288        Changed |= OptimizeExtInstr(MI, MBB, LocalMIs);
289        ++MII;
290      }
291    }
292  }
293
294  return Changed;
295}
296