llvm.org GIT mirror llvm / 73675f1 lib / Target / AMDGPU / AMDGPURegisterBankInfo.cpp
73675f1

Tree @73675f1 (Download .tar.gz)

AMDGPURegisterBankInfo.cpp @73675f1raw · 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
//===- AMDGPURegisterBankInfo.cpp -------------------------------*- C++ -*-==//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
/// \file
/// This file implements the targeting of the RegisterBankInfo class for
/// AMDGPU.
/// \todo This should be generated by TableGen.
//===----------------------------------------------------------------------===//

#include "AMDGPURegisterBankInfo.h"
#include "AMDGPUInstrInfo.h"
#include "SIRegisterInfo.h"
#include "llvm/CodeGen/GlobalISel/RegisterBank.h"
#include "llvm/CodeGen/GlobalISel/RegisterBankInfo.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/IR/Constants.h"

#define GET_TARGET_REGBANK_IMPL
#include "AMDGPUGenRegisterBank.inc"

// This file will be TableGen'ed at some point.
#include "AMDGPUGenRegisterBankInfo.def"

using namespace llvm;

AMDGPURegisterBankInfo::AMDGPURegisterBankInfo(const TargetRegisterInfo &TRI)
    : AMDGPUGenRegisterBankInfo(),
      TRI(static_cast<const SIRegisterInfo*>(&TRI)) {

  // HACK: Until this is fully tablegen'd
  static bool AlreadyInit = false;
  if (AlreadyInit)
    return;

  AlreadyInit = true;

  const RegisterBank &RBSGPR = getRegBank(AMDGPU::SGPRRegBankID);
  (void)RBSGPR;
  assert(&RBSGPR == &AMDGPU::SGPRRegBank);

  const RegisterBank &RBVGPR = getRegBank(AMDGPU::VGPRRegBankID);
  (void)RBVGPR;
  assert(&RBVGPR == &AMDGPU::VGPRRegBank);

}

unsigned AMDGPURegisterBankInfo::copyCost(const RegisterBank &Dst,
                                          const RegisterBank &Src,
                                          unsigned Size) const {
  if (Dst.getID() == AMDGPU::SGPRRegBankID &&
      Src.getID() == AMDGPU::VGPRRegBankID)
    return std::numeric_limits<unsigned>::max();

  // SGPRRegBank with size 1 is actually vcc or another 64-bit sgpr written by
  // the valu.
  if (Size == 1 && Dst.getID() == AMDGPU::SCCRegBankID &&
      Src.getID() == AMDGPU::SGPRRegBankID)
    return std::numeric_limits<unsigned>::max();

  return RegisterBankInfo::copyCost(Dst, Src, Size);
}

const RegisterBank &AMDGPURegisterBankInfo::getRegBankFromRegClass(
    const TargetRegisterClass &RC) const {

  if (TRI->isSGPRClass(&RC))
    return getRegBank(AMDGPU::SGPRRegBankID);

  return getRegBank(AMDGPU::VGPRRegBankID);
}

RegisterBankInfo::InstructionMappings
AMDGPURegisterBankInfo::getInstrAlternativeMappings(
    const MachineInstr &MI) const {

  const MachineFunction &MF = *MI.getParent()->getParent();
  const MachineRegisterInfo &MRI = MF.getRegInfo();


  InstructionMappings AltMappings;
  switch (MI.getOpcode()) {
  case TargetOpcode::G_LOAD: {
    unsigned Size = getSizeInBits(MI.getOperand(0).getReg(), MRI, *TRI);
    // FIXME: Should we be hard coding the size for these mappings?
    const InstructionMapping &SSMapping = getInstructionMapping(
        1, 1, getOperandsMapping(
                  {AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size),
                   AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 64)}),
        2); // Num Operands
    AltMappings.push_back(&SSMapping);

    const InstructionMapping &VVMapping = getInstructionMapping(
        2, 1, getOperandsMapping(
                  {AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size),
                   AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 64)}),
        2); // Num Operands
    AltMappings.push_back(&VVMapping);

    // FIXME: Should this be the pointer-size (64-bits) or the size of the
    // register that will hold the bufffer resourc (128-bits).
    const InstructionMapping &VSMapping = getInstructionMapping(
        3, 1, getOperandsMapping(
                  {AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size),
                   AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 64)}),
        2); // Num Operands
    AltMappings.push_back(&VSMapping);

    return AltMappings;

  }
  case TargetOpcode::G_ICMP: {
    unsigned Size = getSizeInBits(MI.getOperand(2).getReg(), MRI, *TRI);
    const InstructionMapping &SSMapping = getInstructionMapping(1, 1,
      getOperandsMapping({AMDGPU::getValueMapping(AMDGPU::SCCRegBankID, 1),
                          nullptr, // Predicate operand.
                          AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size),
                          AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size)}),
      4); // Num Operands
    AltMappings.push_back(&SSMapping);

    const InstructionMapping &SVMapping = getInstructionMapping(2, 1,
      getOperandsMapping({AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 1),
                          nullptr, // Predicate operand.
                          AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size),
                          AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size)}),
      4); // Num Operands
    AltMappings.push_back(&SVMapping);

    const InstructionMapping &VSMapping = getInstructionMapping(3, 1,
      getOperandsMapping({AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 1),
                          nullptr, // Predicate operand.
                          AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size),
                          AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size)}),
      4); // Num Operands
    AltMappings.push_back(&VSMapping);

    const InstructionMapping &VVMapping = getInstructionMapping(4, 1,
      getOperandsMapping({AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 1),
                          nullptr, // Predicate operand.
                          AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size),
                          AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size)}),
      4); // Num Operands
    AltMappings.push_back(&VVMapping);

    return AltMappings;
  }
  default:
    break;
  }
  return RegisterBankInfo::getInstrAlternativeMappings(MI);
}

void AMDGPURegisterBankInfo::applyMappingImpl(
    const OperandsMapper &OpdMapper) const {
  return applyDefaultMapping(OpdMapper);
}

static bool isInstrUniform(const MachineInstr &MI) {
  if (!MI.hasOneMemOperand())
    return false;

  const MachineMemOperand *MMO = *MI.memoperands_begin();
  return AMDGPUInstrInfo::isUniformMMO(MMO);
}

bool AMDGPURegisterBankInfo::isSALUMapping(const MachineInstr &MI) const {
  const MachineFunction &MF = *MI.getParent()->getParent();
  const MachineRegisterInfo &MRI = MF.getRegInfo();
  for (unsigned i = 0, e = MI.getNumOperands();i != e; ++i) {
    unsigned Reg = MI.getOperand(i).getReg();
    const RegisterBank *Bank = getRegBank(Reg, MRI, *TRI);
    if (Bank && Bank->getID() != AMDGPU::SGPRRegBankID)
      return false;
  }
  return true;
}

const RegisterBankInfo::InstructionMapping &
AMDGPURegisterBankInfo::getDefaultMappingSOP(const MachineInstr &MI) const {
  const MachineFunction &MF = *MI.getParent()->getParent();
  const MachineRegisterInfo &MRI = MF.getRegInfo();
  SmallVector<const ValueMapping*, 8> OpdsMapping(MI.getNumOperands());

  for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
    unsigned Size = getSizeInBits(MI.getOperand(i).getReg(), MRI, *TRI);
    OpdsMapping[i] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
  }
  return getInstructionMapping(1, 1, getOperandsMapping(OpdsMapping),
                               MI.getNumOperands());
}

const RegisterBankInfo::InstructionMapping &
AMDGPURegisterBankInfo::getDefaultMappingVOP(const MachineInstr &MI) const {
  const MachineFunction &MF = *MI.getParent()->getParent();
  const MachineRegisterInfo &MRI = MF.getRegInfo();
  SmallVector<const ValueMapping*, 8> OpdsMapping(MI.getNumOperands());
  unsigned OpdIdx = 0;

  unsigned Size0 = getSizeInBits(MI.getOperand(0).getReg(), MRI, *TRI);
  OpdsMapping[OpdIdx++] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size0);

  if (MI.getOperand(OpdIdx).isIntrinsicID())
    OpdsMapping[OpdIdx++] = nullptr;

  unsigned Reg1 = MI.getOperand(OpdIdx).getReg();
  unsigned Size1 = getSizeInBits(Reg1, MRI, *TRI);
  unsigned Bank1 = getRegBankID(Reg1, MRI, *TRI);
  OpdsMapping[OpdIdx++] = AMDGPU::getValueMapping(Bank1, Size1);

  for (unsigned e = MI.getNumOperands(); OpdIdx != e; ++OpdIdx) {
    unsigned Size = getSizeInBits(MI.getOperand(OpdIdx).getReg(), MRI, *TRI);
    OpdsMapping[OpdIdx] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size);
  }

  return getInstructionMapping(1, 1, getOperandsMapping(OpdsMapping),
                               MI.getNumOperands());
}

const RegisterBankInfo::InstructionMapping &
AMDGPURegisterBankInfo::getInstrMappingForLoad(const MachineInstr &MI) const {

  const MachineFunction &MF = *MI.getParent()->getParent();
  const MachineRegisterInfo &MRI = MF.getRegInfo();
  SmallVector<const ValueMapping*, 8> OpdsMapping(MI.getNumOperands());
  unsigned Size = getSizeInBits(MI.getOperand(0).getReg(), MRI, *TRI);
  unsigned PtrSize = getSizeInBits(MI.getOperand(1).getReg(), MRI, *TRI);

  const ValueMapping *ValMapping;
  const ValueMapping *PtrMapping;

  if (isInstrUniform(MI)) {
    // We have a uniform instruction so we want to use an SMRD load
    ValMapping = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
    PtrMapping = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, PtrSize);
  } else {
    ValMapping = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size);
    // FIXME: What would happen if we used SGPRRegBankID here?
    PtrMapping = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, PtrSize);
  }

  OpdsMapping[0] = ValMapping;
  OpdsMapping[1] = PtrMapping;
  const RegisterBankInfo::InstructionMapping &Mapping = getInstructionMapping(
      1, 1, getOperandsMapping(OpdsMapping), MI.getNumOperands());
  return Mapping;

  // FIXME: Do we want to add a mapping for FLAT load, or should we just
  // handle that during instruction selection?
}

unsigned
AMDGPURegisterBankInfo::getRegBankID(unsigned Reg,
                                     const MachineRegisterInfo &MRI,
                                     const TargetRegisterInfo &TRI,
                                     unsigned Default) const {

  const RegisterBank *Bank = getRegBank(Reg, MRI, TRI);
  return Bank ? Bank->getID() : Default;
}

///
/// This function must return a legal mapping, because
/// AMDGPURegisterBankInfo::getInstrAlternativeMappings() is not called
/// in RegBankSelect::Mode::Fast.  Any mapping that would cause a
/// VGPR to SGPR generated is illegal.
///
const RegisterBankInfo::InstructionMapping &
AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
  const RegisterBankInfo::InstructionMapping &Mapping = getInstrMappingImpl(MI);

  if (Mapping.isValid())
    return Mapping;

  const MachineFunction &MF = *MI.getParent()->getParent();
  const MachineRegisterInfo &MRI = MF.getRegInfo();
  SmallVector<const ValueMapping*, 8> OpdsMapping(MI.getNumOperands());

  switch (MI.getOpcode()) {
  default:
    return getInvalidInstructionMapping();
  case AMDGPU::G_AND:
  case AMDGPU::G_OR:
  case AMDGPU::G_XOR:
  case AMDGPU::G_SHL:
    if (isSALUMapping(MI))
      return getDefaultMappingSOP(MI);
    // Fall-through

  case AMDGPU::G_FADD:
  case AMDGPU::G_FPTOSI:
  case AMDGPU::G_FPTOUI:
  case AMDGPU::G_FMUL:
    return getDefaultMappingVOP(MI);
  case AMDGPU::G_IMPLICIT_DEF: {
    unsigned Size = MRI.getType(MI.getOperand(0).getReg()).getSizeInBits();
    OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
    break;
  }
  case AMDGPU::G_FCONSTANT:
  case AMDGPU::G_CONSTANT: {
    unsigned Size = MRI.getType(MI.getOperand(0).getReg()).getSizeInBits();
    OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
    break;
  }
  case AMDGPU::G_BITCAST: {
    unsigned Size = MRI.getType(MI.getOperand(0).getReg()).getSizeInBits();
    unsigned BankID = getRegBankID(MI.getOperand(1).getReg(), MRI, *TRI);
    OpdsMapping[0] = OpdsMapping[1] = AMDGPU::getValueMapping(BankID, Size);
    break;
  }
  case AMDGPU::G_TRUNC: {
    unsigned Dst = MI.getOperand(0).getReg();
    unsigned Src = MI.getOperand(1).getReg();
    unsigned Bank = getRegBankID(Src, MRI, *TRI);
    unsigned DstSize = getSizeInBits(Dst, MRI, *TRI);
    unsigned SrcSize = getSizeInBits(Src, MRI, *TRI);
    OpdsMapping[0] = AMDGPU::getValueMapping(Bank, DstSize);
    OpdsMapping[1] = AMDGPU::getValueMapping(Bank, SrcSize);
    break;
  }
  case AMDGPU::G_ZEXT: {
    unsigned Dst = MI.getOperand(0).getReg();
    unsigned Src = MI.getOperand(1).getReg();
    unsigned DstSize = getSizeInBits(Dst, MRI, *TRI);
    unsigned SrcSize = getSizeInBits(Src, MRI, *TRI);
    unsigned SrcBank = getRegBankID(Src, MRI, *TRI,
                                    SrcSize == 1 ? AMDGPU::SGPRRegBankID :
                                    AMDGPU::VGPRRegBankID);
    unsigned DstBank = SrcBank;
    if (SrcSize == 1) {
      if (SrcBank == AMDGPU::SGPRRegBankID)
        DstBank = AMDGPU::VGPRRegBankID;
      else
        DstBank = AMDGPU::SGPRRegBankID;
    }

    OpdsMapping[0] = AMDGPU::getValueMapping(DstBank, DstSize);
    OpdsMapping[1] = AMDGPU::getValueMapping(SrcBank, SrcSize);
    break;
  }
  case AMDGPU::G_FCMP: {
    unsigned Size = MRI.getType(MI.getOperand(2).getReg()).getSizeInBits();
    unsigned Op2Bank = getRegBankID(MI.getOperand(2).getReg(), MRI, *TRI);
    OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 1);
    OpdsMapping[1] = nullptr; // Predicate Operand.
    OpdsMapping[2] = AMDGPU::getValueMapping(Op2Bank, Size);
    OpdsMapping[3] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size);
    break;
  }
  case AMDGPU::G_GEP: {
    for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
      if (!MI.getOperand(i).isReg())
        continue;

      unsigned Size = MRI.getType(MI.getOperand(i).getReg()).getSizeInBits();
      OpdsMapping[i] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
    }
    break;
  }
  case AMDGPU::G_STORE: {
    assert(MI.getOperand(0).isReg());
    unsigned Size = MRI.getType(MI.getOperand(0).getReg()).getSizeInBits();
    // FIXME: We need to specify a different reg bank once scalar stores
    // are supported.
    const ValueMapping *ValMapping =
        AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size);
    // FIXME: Depending on the type of store, the pointer could be in
    // the SGPR Reg bank.
    // FIXME: Pointer size should be based on the address space.
    const ValueMapping *PtrMapping =
        AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 64);

    OpdsMapping[0] = ValMapping;
    OpdsMapping[1] = PtrMapping;
    break;
  }

  case AMDGPU::G_ICMP: {
    unsigned Size = MRI.getType(MI.getOperand(2).getReg()).getSizeInBits();
    unsigned Op2Bank = getRegBankID(MI.getOperand(2).getReg(), MRI, *TRI);
    unsigned Op3Bank = getRegBankID(MI.getOperand(3).getReg(), MRI, *TRI);
    unsigned Op0Bank = Op2Bank == AMDGPU::SGPRRegBankID &&
                       Op3Bank == AMDGPU::SGPRRegBankID ?
                       AMDGPU::SCCRegBankID : AMDGPU::VGPRRegBankID;
    OpdsMapping[0] = AMDGPU::getValueMapping(Op0Bank, 1);
    OpdsMapping[1] = nullptr; // Predicate Operand.
    OpdsMapping[2] = AMDGPU::getValueMapping(Op2Bank, Size);
    OpdsMapping[3] = AMDGPU::getValueMapping(Op3Bank, Size);
    break;
  }
  case AMDGPU::G_INTRINSIC: {
    switch(MI.getOperand(1).getIntrinsicID()) {
    default:
      return getInvalidInstructionMapping();
    case Intrinsic::maxnum:
    case Intrinsic::minnum:
    case Intrinsic::amdgcn_cvt_pkrtz:
      return getDefaultMappingVOP(MI);
    }
    break;
  }
  case AMDGPU::G_INTRINSIC_W_SIDE_EFFECTS: {
    switch (MI.getOperand(0).getIntrinsicID()) {
    default:
      return getInvalidInstructionMapping();
    case Intrinsic::amdgcn_exp_compr:
      OpdsMapping[0] = nullptr; // IntrinsicID
      // FIXME: These are immediate values which can't be read from registers.
      OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      OpdsMapping[2] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      // FIXME: Could we support packed types here?
      OpdsMapping[3] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 32);
      OpdsMapping[4] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 32);
      // FIXME: These are immediate values which can't be read from registers.
      OpdsMapping[5] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      OpdsMapping[6] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      break;
    case Intrinsic::amdgcn_exp:
      OpdsMapping[0] = nullptr; // IntrinsicID
      // FIXME: These are immediate values which can't be read from registers.
      OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      OpdsMapping[2] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      // FIXME: Could we support packed types here?
      OpdsMapping[3] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 32);
      OpdsMapping[4] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 32);
      OpdsMapping[5] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 32);
      OpdsMapping[6] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 32);
      // FIXME: These are immediate values which can't be read from registers.
      OpdsMapping[7] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      OpdsMapping[8] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
      break;
    }
    break;
  }
  case AMDGPU::G_LOAD:
    return getInstrMappingForLoad(MI);
  }

  return getInstructionMapping(1, 1, getOperandsMapping(OpdsMapping),
                               MI.getNumOperands());
}