DelaySlotFiller.cpp revision e837dead3c8dc3445ef6a0e2322179c57e264a13
15821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//===-- DelaySlotFiller.cpp - SPARC delay slot filler ---------------------===//
25821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
35821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//                     The LLVM Compiler Infrastructure
45821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
55821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// This file is distributed under the University of Illinois Open Source
65821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// License. See LICENSE.TXT for details.
75821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
85821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//===----------------------------------------------------------------------===//
95f1c94371a64b3196d4be9466099bb892df9b88eTorne (Richard Coles)//
105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// This is a simple local pass that attempts to fill delay slots with useful
115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// instructions. If no instructions can be moved into the delay slot, then a
125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// NOP is placed.
135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//===----------------------------------------------------------------------===//
145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#define DEBUG_TYPE "delay-slot-filler"
165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "Sparc.h"
17c2e0dbddbe15c98d52c4786dac06cb8952a8ae6dTorne (Richard Coles)#include "llvm/CodeGen/MachineFunctionPass.h"
18c2e0dbddbe15c98d52c4786dac06cb8952a8ae6dTorne (Richard Coles)#include "llvm/CodeGen/MachineInstrBuilder.h"
19c2e0dbddbe15c98d52c4786dac06cb8952a8ae6dTorne (Richard Coles)#include "llvm/Support/CommandLine.h"
20cedac228d2dd51db4b79ea1e72c7f249408ee061Torne (Richard Coles)#include "llvm/Target/TargetMachine.h"
215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "llvm/Target/TargetInstrInfo.h"
225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "llvm/Target/TargetRegisterInfo.h"
235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "llvm/ADT/SmallSet.h"
24cedac228d2dd51db4b79ea1e72c7f249408ee061Torne (Richard Coles)#include "llvm/ADT/Statistic.h"
255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)using namespace llvm;
275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
28cedac228d2dd51db4b79ea1e72c7f249408ee061Torne (Richard Coles)STATISTIC(FilledSlots, "Number of delay slots filled");
29cedac228d2dd51db4b79ea1e72c7f249408ee061Torne (Richard Coles)
305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)static cl::opt<bool> DisableDelaySlotFiller(
315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  "disable-sparc-delay-filler",
325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  cl::init(false),
33cedac228d2dd51db4b79ea1e72c7f249408ee061Torne (Richard Coles)  cl::desc("Disable the Sparc delay slot filler."),
34cedac228d2dd51db4b79ea1e72c7f249408ee061Torne (Richard Coles)  cl::Hidden);
355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)namespace {
375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  struct Filler : public MachineFunctionPass {
385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    /// Target machine description which we query for reg. names, data
395f1c94371a64b3196d4be9466099bb892df9b88eTorne (Richard Coles)    /// layout, etc.
40    ///
41    TargetMachine &TM;
42    const TargetInstrInfo *TII;
43
44    static char ID;
45    Filler(TargetMachine &tm)
46      : MachineFunctionPass(ID), TM(tm), TII(tm.getInstrInfo()) { }
47
48    virtual const char *getPassName() const {
49      return "SPARC Delay Slot Filler";
50    }
51
52    bool runOnMachineBasicBlock(MachineBasicBlock &MBB);
53    bool runOnMachineFunction(MachineFunction &F) {
54      bool Changed = false;
55      for (MachineFunction::iterator FI = F.begin(), FE = F.end();
56           FI != FE; ++FI)
57        Changed |= runOnMachineBasicBlock(*FI);
58      return Changed;
59    }
60
61    bool isDelayFiller(MachineBasicBlock &MBB,
62                       MachineBasicBlock::iterator candidate);
63
64    void insertCallUses(MachineBasicBlock::iterator MI,
65                        SmallSet<unsigned, 32>& RegUses);
66
67    void insertDefsUses(MachineBasicBlock::iterator MI,
68                        SmallSet<unsigned, 32>& RegDefs,
69                        SmallSet<unsigned, 32>& RegUses);
70
71    bool IsRegInSet(SmallSet<unsigned, 32>& RegSet,
72                    unsigned Reg);
73
74    bool delayHasHazard(MachineBasicBlock::iterator candidate,
75                        bool &sawLoad, bool &sawStore,
76                        SmallSet<unsigned, 32> &RegDefs,
77                        SmallSet<unsigned, 32> &RegUses);
78
79    MachineBasicBlock::iterator
80    findDelayInstr(MachineBasicBlock &MBB, MachineBasicBlock::iterator slot);
81
82    bool needsUnimp(MachineBasicBlock::iterator I, unsigned &StructSize);
83
84  };
85  char Filler::ID = 0;
86} // end of anonymous namespace
87
88/// createSparcDelaySlotFillerPass - Returns a pass that fills in delay
89/// slots in Sparc MachineFunctions
90///
91FunctionPass *llvm::createSparcDelaySlotFillerPass(TargetMachine &tm) {
92  return new Filler(tm);
93}
94
95
96/// runOnMachineBasicBlock - Fill in delay slots for the given basic block.
97/// We assume there is only one delay slot per delayed instruction.
98///
99bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
100  bool Changed = false;
101
102  for (MachineBasicBlock::iterator I = MBB.begin(); I != MBB.end(); ++I)
103    if (I->getDesc().hasDelaySlot()) {
104      MachineBasicBlock::iterator D = MBB.end();
105      MachineBasicBlock::iterator J = I;
106
107      if (!DisableDelaySlotFiller)
108        D = findDelayInstr(MBB, I);
109
110      ++FilledSlots;
111      Changed = true;
112
113      if (D == MBB.end())
114        BuildMI(MBB, ++J, I->getDebugLoc(), TII->get(SP::NOP));
115      else
116        MBB.splice(++J, &MBB, D);
117      unsigned structSize = 0;
118      if (needsUnimp(I, structSize)) {
119        MachineBasicBlock::iterator J = I;
120        ++J; //skip the delay filler.
121        BuildMI(MBB, ++J, I->getDebugLoc(),
122                TII->get(SP::UNIMP)).addImm(structSize);
123      }
124    }
125  return Changed;
126}
127
128MachineBasicBlock::iterator
129Filler::findDelayInstr(MachineBasicBlock &MBB,
130                       MachineBasicBlock::iterator slot)
131{
132  SmallSet<unsigned, 32> RegDefs;
133  SmallSet<unsigned, 32> RegUses;
134  bool sawLoad = false;
135  bool sawStore = false;
136
137  MachineBasicBlock::iterator I = slot;
138
139  if (slot->getOpcode() == SP::RET)
140    return MBB.end();
141
142  if (slot->getOpcode() == SP::RETL) {
143    --I;
144    if (I->getOpcode() != SP::RESTORErr)
145      return MBB.end();
146    //change retl to ret
147    slot->setDesc(TII->get(SP::RET));
148    return I;
149  }
150
151  //Call's delay filler can def some of call's uses.
152  if (slot->getDesc().isCall())
153    insertCallUses(slot, RegUses);
154  else
155    insertDefsUses(slot, RegDefs, RegUses);
156
157  bool done = false;
158
159  while (!done) {
160    done = (I == MBB.begin());
161
162    if (!done)
163      --I;
164
165    // skip debug value
166    if (I->isDebugValue())
167      continue;
168
169
170    if (I->hasUnmodeledSideEffects()
171        || I->isInlineAsm()
172        || I->isLabel()
173        || I->getDesc().hasDelaySlot()
174        || isDelayFiller(MBB, I))
175      break;
176
177    if (delayHasHazard(I, sawLoad, sawStore, RegDefs, RegUses)) {
178      insertDefsUses(I, RegDefs, RegUses);
179      continue;
180    }
181
182    return I;
183  }
184  return MBB.end();
185}
186
187bool Filler::delayHasHazard(MachineBasicBlock::iterator candidate,
188                            bool &sawLoad,
189                            bool &sawStore,
190                            SmallSet<unsigned, 32> &RegDefs,
191                            SmallSet<unsigned, 32> &RegUses)
192{
193
194  if (candidate->isImplicitDef() || candidate->isKill())
195    return true;
196
197  if (candidate->getDesc().mayLoad()) {
198    sawLoad = true;
199    if (sawStore)
200      return true;
201  }
202
203  if (candidate->getDesc().mayStore()) {
204    if (sawStore)
205      return true;
206    sawStore = true;
207    if (sawLoad)
208      return true;
209  }
210
211  for (unsigned i = 0, e = candidate->getNumOperands(); i!= e; ++i) {
212    const MachineOperand &MO = candidate->getOperand(i);
213    if (!MO.isReg())
214      continue; // skip
215
216    unsigned Reg = MO.getReg();
217
218    if (MO.isDef()) {
219      //check whether Reg is defined or used before delay slot.
220      if (IsRegInSet(RegDefs, Reg) || IsRegInSet(RegUses, Reg))
221        return true;
222    }
223    if (MO.isUse()) {
224      //check whether Reg is defined before delay slot.
225      if (IsRegInSet(RegDefs, Reg))
226        return true;
227    }
228  }
229  return false;
230}
231
232
233void Filler::insertCallUses(MachineBasicBlock::iterator MI,
234                            SmallSet<unsigned, 32>& RegUses)
235{
236
237  switch(MI->getOpcode()) {
238  default: llvm_unreachable("Unknown opcode.");
239  case SP::CALL: break;
240  case SP::JMPLrr:
241  case SP::JMPLri:
242    assert(MI->getNumOperands() >= 2);
243    const MachineOperand &Reg = MI->getOperand(0);
244    assert(Reg.isReg() && "JMPL first operand is not a register.");
245    assert(Reg.isUse() && "JMPL first operand is not a use.");
246    RegUses.insert(Reg.getReg());
247
248    const MachineOperand &RegOrImm = MI->getOperand(1);
249    if (RegOrImm.isImm())
250        break;
251    assert(RegOrImm.isReg() && "JMPLrr second operand is not a register.");
252    assert(RegOrImm.isUse() && "JMPLrr second operand is not a use.");
253    RegUses.insert(RegOrImm.getReg());
254    break;
255  }
256}
257
258//Insert Defs and Uses of MI into the sets RegDefs and RegUses.
259void Filler::insertDefsUses(MachineBasicBlock::iterator MI,
260                            SmallSet<unsigned, 32>& RegDefs,
261                            SmallSet<unsigned, 32>& RegUses)
262{
263  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
264    const MachineOperand &MO = MI->getOperand(i);
265    if (!MO.isReg())
266      continue;
267
268    unsigned Reg = MO.getReg();
269    if (Reg == 0)
270      continue;
271    if (MO.isDef())
272      RegDefs.insert(Reg);
273    if (MO.isUse())
274      RegUses.insert(Reg);
275
276  }
277}
278
279//returns true if the Reg or its alias is in the RegSet.
280bool Filler::IsRegInSet(SmallSet<unsigned, 32>& RegSet, unsigned Reg)
281{
282  if (RegSet.count(Reg))
283    return true;
284  // check Aliased Registers
285  for (const unsigned *Alias = TM.getRegisterInfo()->getAliasSet(Reg);
286       *Alias; ++ Alias)
287    if (RegSet.count(*Alias))
288      return true;
289
290  return false;
291}
292
293// return true if the candidate is a delay filler.
294bool Filler::isDelayFiller(MachineBasicBlock &MBB,
295                           MachineBasicBlock::iterator candidate)
296{
297  if (candidate == MBB.begin())
298    return false;
299  if (candidate->getOpcode() == SP::UNIMP)
300    return true;
301  const MCInstrDesc &prevdesc = (--candidate)->getDesc();
302  return prevdesc.hasDelaySlot();
303}
304
305bool Filler::needsUnimp(MachineBasicBlock::iterator I, unsigned &StructSize)
306{
307  if (!I->getDesc().isCall())
308    return false;
309
310  unsigned structSizeOpNum = 0;
311  switch (I->getOpcode()) {
312  default: llvm_unreachable("Unknown call opcode.");
313  case SP::CALL: structSizeOpNum = 1; break;
314  case SP::JMPLrr:
315  case SP::JMPLri: structSizeOpNum = 2; break;
316  }
317
318  const MachineOperand &MO = I->getOperand(structSizeOpNum);
319  if (!MO.isImm())
320    return false;
321  StructSize = MO.getImm();
322  return true;
323}
324