llvm.org GIT mirror llvm / release_29 lib / Target / CellSPU / SPUInstrInfo.cpp
release_29

Tree @release_29 (Download .tar.gz)

SPUInstrInfo.cpp @release_29raw · history · blame

  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
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
//===- SPUInstrInfo.cpp - Cell SPU Instruction Information ----------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains the Cell SPU implementation of the TargetInstrInfo class.
//
//===----------------------------------------------------------------------===//

#include "SPURegisterNames.h"
#include "SPUInstrInfo.h"
#include "SPUInstrBuilder.h"
#include "SPUTargetMachine.h"
#include "SPUGenInstrInfo.inc"
#include "SPUHazardRecognizers.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/MC/MCContext.h"

using namespace llvm;

namespace {
  //! Predicate for an unconditional branch instruction
  inline bool isUncondBranch(const MachineInstr *I) {
    unsigned opc = I->getOpcode();

    return (opc == SPU::BR
            || opc == SPU::BRA
            || opc == SPU::BI);
  }

  //! Predicate for a conditional branch instruction
  inline bool isCondBranch(const MachineInstr *I) {
    unsigned opc = I->getOpcode();

    return (opc == SPU::BRNZr32
            || opc == SPU::BRNZv4i32
            || opc == SPU::BRZr32
            || opc == SPU::BRZv4i32
            || opc == SPU::BRHNZr16
            || opc == SPU::BRHNZv8i16
            || opc == SPU::BRHZr16
            || opc == SPU::BRHZv8i16);
  }
}

SPUInstrInfo::SPUInstrInfo(SPUTargetMachine &tm)
  : TargetInstrInfoImpl(SPUInsts, sizeof(SPUInsts)/sizeof(SPUInsts[0])),
    TM(tm),
    RI(*TM.getSubtargetImpl(), *this)
{ /* NOP */ }

/// CreateTargetHazardRecognizer - Return the hazard recognizer to use for
/// this target when scheduling the DAG.
ScheduleHazardRecognizer *SPUInstrInfo::CreateTargetHazardRecognizer(
  const TargetMachine *TM,
  const ScheduleDAG *DAG) const {
  const TargetInstrInfo *TII = TM->getInstrInfo();
  assert(TII && "No InstrInfo?");
  return new SPUHazardRecognizer(*TII);
}

unsigned
SPUInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
                                  int &FrameIndex) const {
  switch (MI->getOpcode()) {
  default: break;
  case SPU::LQDv16i8:
  case SPU::LQDv8i16:
  case SPU::LQDv4i32:
  case SPU::LQDv4f32:
  case SPU::LQDv2f64:
  case SPU::LQDr128:
  case SPU::LQDr64:
  case SPU::LQDr32:
  case SPU::LQDr16: {
    const MachineOperand MOp1 = MI->getOperand(1);
    const MachineOperand MOp2 = MI->getOperand(2);
    if (MOp1.isImm() && MOp2.isFI()) {
      FrameIndex = MOp2.getIndex();
      return MI->getOperand(0).getReg();
    }
    break;
  }
  }
  return 0;
}

unsigned
SPUInstrInfo::isStoreToStackSlot(const MachineInstr *MI,
                                 int &FrameIndex) const {
  switch (MI->getOpcode()) {
  default: break;
  case SPU::STQDv16i8:
  case SPU::STQDv8i16:
  case SPU::STQDv4i32:
  case SPU::STQDv4f32:
  case SPU::STQDv2f64:
  case SPU::STQDr128:
  case SPU::STQDr64:
  case SPU::STQDr32:
  case SPU::STQDr16:
  case SPU::STQDr8: {
    const MachineOperand MOp1 = MI->getOperand(1);
    const MachineOperand MOp2 = MI->getOperand(2);
    if (MOp1.isImm() && MOp2.isFI()) {
      FrameIndex = MOp2.getIndex();
      return MI->getOperand(0).getReg();
    }
    break;
  }
  }
  return 0;
}

void SPUInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
                               MachineBasicBlock::iterator I, DebugLoc DL,
                               unsigned DestReg, unsigned SrcReg,
                               bool KillSrc) const
{
  // We support cross register class moves for our aliases, such as R3 in any
  // reg class to any other reg class containing R3.  This is required because
  // we instruction select bitconvert i64 -> f64 as a noop for example, so our
  // types have no specific meaning.

  BuildMI(MBB, I, DL, get(SPU::LRr128), DestReg)
    .addReg(SrcReg, getKillRegState(KillSrc));
}

void
SPUInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator MI,
                                  unsigned SrcReg, bool isKill, int FrameIdx,
                                  const TargetRegisterClass *RC,
                                  const TargetRegisterInfo *TRI) const
{
  unsigned opc;
  bool isValidFrameIdx = (FrameIdx < SPUFrameLowering::maxFrameOffset());
  if (RC == SPU::GPRCRegisterClass) {
    opc = (isValidFrameIdx ? SPU::STQDr128 : SPU::STQXr128);
  } else if (RC == SPU::R64CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::STQDr64 : SPU::STQXr64);
  } else if (RC == SPU::R64FPRegisterClass) {
    opc = (isValidFrameIdx ? SPU::STQDr64 : SPU::STQXr64);
  } else if (RC == SPU::R32CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::STQDr32 : SPU::STQXr32);
  } else if (RC == SPU::R32FPRegisterClass) {
    opc = (isValidFrameIdx ? SPU::STQDr32 : SPU::STQXr32);
  } else if (RC == SPU::R16CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::STQDr16 : SPU::STQXr16);
  } else if (RC == SPU::R8CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::STQDr8 : SPU::STQXr8);
  } else if (RC == SPU::VECREGRegisterClass) {
    opc = (isValidFrameIdx) ? SPU::STQDv16i8 : SPU::STQXv16i8;
  } else {
    llvm_unreachable("Unknown regclass!");
  }

  DebugLoc DL;
  if (MI != MBB.end()) DL = MI->getDebugLoc();
  addFrameReference(BuildMI(MBB, MI, DL, get(opc))
                    .addReg(SrcReg, getKillRegState(isKill)), FrameIdx);
}

void
SPUInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
                                   MachineBasicBlock::iterator MI,
                                   unsigned DestReg, int FrameIdx,
                                   const TargetRegisterClass *RC,
                                   const TargetRegisterInfo *TRI) const
{
  unsigned opc;
  bool isValidFrameIdx = (FrameIdx < SPUFrameLowering::maxFrameOffset());
  if (RC == SPU::GPRCRegisterClass) {
    opc = (isValidFrameIdx ? SPU::LQDr128 : SPU::LQXr128);
  } else if (RC == SPU::R64CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::LQDr64 : SPU::LQXr64);
  } else if (RC == SPU::R64FPRegisterClass) {
    opc = (isValidFrameIdx ? SPU::LQDr64 : SPU::LQXr64);
  } else if (RC == SPU::R32CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::LQDr32 : SPU::LQXr32);
  } else if (RC == SPU::R32FPRegisterClass) {
    opc = (isValidFrameIdx ? SPU::LQDr32 : SPU::LQXr32);
  } else if (RC == SPU::R16CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::LQDr16 : SPU::LQXr16);
  } else if (RC == SPU::R8CRegisterClass) {
    opc = (isValidFrameIdx ? SPU::LQDr8 : SPU::LQXr8);
  } else if (RC == SPU::VECREGRegisterClass) {
    opc = (isValidFrameIdx) ? SPU::LQDv16i8 : SPU::LQXv16i8;
  } else {
    llvm_unreachable("Unknown regclass in loadRegFromStackSlot!");
  }

  DebugLoc DL;
  if (MI != MBB.end()) DL = MI->getDebugLoc();
  addFrameReference(BuildMI(MBB, MI, DL, get(opc), DestReg), FrameIdx);
}

//! Branch analysis
/*!
  \note This code was kiped from PPC. There may be more branch analysis for
  CellSPU than what's currently done here.
 */
bool
SPUInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
                            MachineBasicBlock *&FBB,
                            SmallVectorImpl<MachineOperand> &Cond,
                            bool AllowModify) const {
  // If the block has no terminators, it just falls into the block after it.
  MachineBasicBlock::iterator I = MBB.end();
  if (I == MBB.begin())
    return false;
  --I;
  while (I->isDebugValue()) {
    if (I == MBB.begin())
      return false;
    --I;
  }
  if (!isUnpredicatedTerminator(I))
    return false;

  // Get the last instruction in the block.
  MachineInstr *LastInst = I;

  // If there is only one terminator instruction, process it.
  if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
    if (isUncondBranch(LastInst)) {
      // Check for jump tables
      if (!LastInst->getOperand(0).isMBB())
        return true;
      TBB = LastInst->getOperand(0).getMBB();
      return false;
    } else if (isCondBranch(LastInst)) {
      // Block ends with fall-through condbranch.
      TBB = LastInst->getOperand(1).getMBB();
      DEBUG(errs() << "Pushing LastInst:               ");
      DEBUG(LastInst->dump());
      Cond.push_back(MachineOperand::CreateImm(LastInst->getOpcode()));
      Cond.push_back(LastInst->getOperand(0));
      return false;
    }
    // Otherwise, don't know what this is.
    return true;
  }

  // Get the instruction before it if it's a terminator.
  MachineInstr *SecondLastInst = I;

  // If there are three terminators, we don't know what sort of block this is.
  if (SecondLastInst && I != MBB.begin() &&
      isUnpredicatedTerminator(--I))
    return true;

  // If the block ends with a conditional and unconditional branch, handle it.
  if (isCondBranch(SecondLastInst) && isUncondBranch(LastInst)) {
    TBB =  SecondLastInst->getOperand(1).getMBB();
    DEBUG(errs() << "Pushing SecondLastInst:         ");
    DEBUG(SecondLastInst->dump());
    Cond.push_back(MachineOperand::CreateImm(SecondLastInst->getOpcode()));
    Cond.push_back(SecondLastInst->getOperand(0));
    FBB = LastInst->getOperand(0).getMBB();
    return false;
  }

  // If the block ends with two unconditional branches, handle it.  The second
  // one is not executed, so remove it.
  if (isUncondBranch(SecondLastInst) && isUncondBranch(LastInst)) {
    TBB = SecondLastInst->getOperand(0).getMBB();
    I = LastInst;
    if (AllowModify)
      I->eraseFromParent();
    return false;
  }

  // Otherwise, can't handle this.
  return true;
}

// search MBB for branch hint labels and branch hit ops
static void removeHBR( MachineBasicBlock &MBB) {
  for (MachineBasicBlock::iterator I = MBB.begin(); I != MBB.end(); ++I){
    if (I->getOpcode() == SPU::HBRA ||
        I->getOpcode() == SPU::HBR_LABEL){
      I=MBB.erase(I);
    }
  }
}

unsigned
SPUInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
  MachineBasicBlock::iterator I = MBB.end();
  removeHBR(MBB);
  if (I == MBB.begin())
    return 0;
  --I;
  while (I->isDebugValue()) {
    if (I == MBB.begin())
      return 0;
    --I;
  }
  if (!isCondBranch(I) && !isUncondBranch(I))
    return 0;

  // Remove the first branch.
  DEBUG(errs() << "Removing branch:                ");
  DEBUG(I->dump());
  I->eraseFromParent();
  I = MBB.end();
  if (I == MBB.begin())
    return 1;

  --I;
  if (!(isCondBranch(I) || isUncondBranch(I)))
    return 1;

  // Remove the second branch.
  DEBUG(errs() << "Removing second branch:         ");
  DEBUG(I->dump());
  I->eraseFromParent();
  return 2;
}

/** Find the optimal position for a hint branch instruction in a basic block.
 * This should take into account:
 *   -the branch hint delays
 *   -congestion of the memory bus
 *   -dual-issue scheduling (i.e. avoid insertion of nops)
 * Current implementation is rather simplistic.
 */
static MachineBasicBlock::iterator findHBRPosition(MachineBasicBlock &MBB)
{
   MachineBasicBlock::iterator J = MBB.end();
	for( int i=0; i<8; i++) {
		if( J == MBB.begin() ) return J;
		J--;
	}
	return J;
}

unsigned
SPUInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
                           MachineBasicBlock *FBB,
                           const SmallVectorImpl<MachineOperand> &Cond,
                           DebugLoc DL) const {
  // Shouldn't be a fall through.
  assert(TBB && "InsertBranch must not be told to insert a fallthrough");
  assert((Cond.size() == 2 || Cond.size() == 0) &&
         "SPU branch conditions have two components!");

  MachineInstrBuilder MIB;
  //TODO: make a more accurate algorithm.
  bool haveHBR = MBB.size()>8;
  
  removeHBR(MBB);
  MCSymbol *branchLabel = MBB.getParent()->getContext().CreateTempSymbol();
  // Add a label just before the branch
  if (haveHBR)
    MIB = BuildMI(&MBB, DL, get(SPU::HBR_LABEL)).addSym(branchLabel);

  // One-way branch.
  if (FBB == 0) {
    if (Cond.empty()) {
      // Unconditional branch
      MIB = BuildMI(&MBB, DL, get(SPU::BR));
      MIB.addMBB(TBB);

      DEBUG(errs() << "Inserted one-way uncond branch: ");
      DEBUG((*MIB).dump());

      // basic blocks have just one branch so it is safe to add the hint a its
      if (haveHBR) {
        MIB = BuildMI( MBB, findHBRPosition(MBB), DL, get(SPU::HBRA));
        MIB.addSym(branchLabel);
        MIB.addMBB(TBB);
      }	
    } else {
      // Conditional branch
      MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
      MIB.addReg(Cond[1].getReg()).addMBB(TBB);

      if (haveHBR) {
        MIB = BuildMI(MBB, findHBRPosition(MBB), DL, get(SPU::HBRA));
        MIB.addSym(branchLabel);
        MIB.addMBB(TBB);
      }	

      DEBUG(errs() << "Inserted one-way cond branch:   ");
      DEBUG((*MIB).dump());
    }
    return 1;
  } else {
    MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
    MachineInstrBuilder MIB2 = BuildMI(&MBB, DL, get(SPU::BR));

    // Two-way Conditional Branch.
    MIB.addReg(Cond[1].getReg()).addMBB(TBB);
    MIB2.addMBB(FBB);

    if (haveHBR) {
      MIB = BuildMI( MBB, findHBRPosition(MBB), DL, get(SPU::HBRA));
      MIB.addSym(branchLabel);
      MIB.addMBB(FBB);
    }	

    DEBUG(errs() << "Inserted conditional branch:    ");
    DEBUG((*MIB).dump());
    DEBUG(errs() << "part 2: ");
    DEBUG((*MIB2).dump());
   return 2;
  }
}

//! Reverses a branch's condition, returning false on success.
bool
SPUInstrInfo::ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond)
  const {
  // Pretty brainless way of inverting the condition, but it works, considering
  // there are only two conditions...
  static struct {
    unsigned Opc;               //! The incoming opcode
    unsigned RevCondOpc;        //! The reversed condition opcode
  } revconds[] = {
    { SPU::BRNZr32, SPU::BRZr32 },
    { SPU::BRNZv4i32, SPU::BRZv4i32 },
    { SPU::BRZr32, SPU::BRNZr32 },
    { SPU::BRZv4i32, SPU::BRNZv4i32 },
    { SPU::BRHNZr16, SPU::BRHZr16 },
    { SPU::BRHNZv8i16, SPU::BRHZv8i16 },
    { SPU::BRHZr16, SPU::BRHNZr16 },
    { SPU::BRHZv8i16, SPU::BRHNZv8i16 }
  };

  unsigned Opc = unsigned(Cond[0].getImm());
  // Pretty dull mapping between the two conditions that SPU can generate:
  for (int i = sizeof(revconds)/sizeof(revconds[0]) - 1; i >= 0; --i) {
    if (revconds[i].Opc == Opc) {
      Cond[0].setImm(revconds[i].RevCondOpc);
      return false;
    }
  }

  return true;
}