llvm.org GIT mirror llvm / release_35@215010 lib / Target / AArch64 / AArch64TargetTransformInfo.cpp
release_35@215010

Tree @release_35@215010 (Download .tar.gz)

AArch64TargetTransformInfo.cpp @release_35@215010raw · 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
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
//===-- AArch64TargetTransformInfo.cpp - AArch64 specific TTI pass --------===//
//
//                     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 a TargetTransformInfo analysis pass specific to the
/// AArch64 target machine. It uses the target's detailed information to provide
/// more precise answers to certain TTI queries, while letting the target
/// independent and default TTI implementations handle the rest.
///
//===----------------------------------------------------------------------===//

#include "AArch64.h"
#include "AArch64TargetMachine.h"
#include "MCTargetDesc/AArch64AddressingModes.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Support/Debug.h"
#include "llvm/Target/CostTable.h"
#include "llvm/Target/TargetLowering.h"
#include <algorithm>
using namespace llvm;

#define DEBUG_TYPE "aarch64tti"

// Declare the pass initialization routine locally as target-specific passes
// don't have a target-wide initialization entry point, and so we rely on the
// pass constructor initialization.
namespace llvm {
void initializeAArch64TTIPass(PassRegistry &);
}

namespace {

class AArch64TTI final : public ImmutablePass, public TargetTransformInfo {
  const AArch64TargetMachine *TM;
  const AArch64Subtarget *ST;
  const AArch64TargetLowering *TLI;

  /// Estimate the overhead of scalarizing an instruction. Insert and Extract
  /// are set if the result needs to be inserted and/or extracted from vectors.
  unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract) const;

public:
  AArch64TTI() : ImmutablePass(ID), TM(nullptr), ST(nullptr), TLI(nullptr) {
    llvm_unreachable("This pass cannot be directly constructed");
  }

  AArch64TTI(const AArch64TargetMachine *TM)
      : ImmutablePass(ID), TM(TM), ST(TM->getSubtargetImpl()),
        TLI(TM->getTargetLowering()) {
    initializeAArch64TTIPass(*PassRegistry::getPassRegistry());
  }

  void initializePass() override { pushTTIStack(this); }

  void getAnalysisUsage(AnalysisUsage &AU) const override {
    TargetTransformInfo::getAnalysisUsage(AU);
  }

  /// Pass identification.
  static char ID;

  /// Provide necessary pointer adjustments for the two base classes.
  void *getAdjustedAnalysisPointer(const void *ID) override {
    if (ID == &TargetTransformInfo::ID)
      return (TargetTransformInfo *)this;
    return this;
  }

  /// \name Scalar TTI Implementations
  /// @{
  unsigned getIntImmCost(int64_t Val) const;
  unsigned getIntImmCost(const APInt &Imm, Type *Ty) const override;
  unsigned getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
                         Type *Ty) const override;
  unsigned getIntImmCost(Intrinsic::ID IID, unsigned Idx, const APInt &Imm,
                         Type *Ty) const override;
  PopcntSupportKind getPopcntSupport(unsigned TyWidth) const override;

  /// @}

  /// \name Vector TTI Implementations
  /// @{

  unsigned getNumberOfRegisters(bool Vector) const override {
    if (Vector) {
      if (ST->hasNEON())
        return 32;
      return 0;
    }
    return 31;
  }

  unsigned getRegisterBitWidth(bool Vector) const override {
    if (Vector) {
      if (ST->hasNEON())
        return 128;
      return 0;
    }
    return 64;
  }

  unsigned getMaximumUnrollFactor() const override { return 2; }

  unsigned getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const
      override;

  unsigned getVectorInstrCost(unsigned Opcode, Type *Val, unsigned Index) const
      override;

  unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty,
                                  OperandValueKind Opd1Info = OK_AnyValue,
                                  OperandValueKind Opd2Info = OK_AnyValue) const
      override;

  unsigned getAddressComputationCost(Type *Ty, bool IsComplex) const override;

  unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy) const
      override;

  unsigned getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
                           unsigned AddressSpace) const override;
  /// @}
};

} // end anonymous namespace

INITIALIZE_AG_PASS(AArch64TTI, TargetTransformInfo, "aarch64tti",
                   "AArch64 Target Transform Info", true, true, false)
char AArch64TTI::ID = 0;

ImmutablePass *
llvm::createAArch64TargetTransformInfoPass(const AArch64TargetMachine *TM) {
  return new AArch64TTI(TM);
}

/// \brief Calculate the cost of materializing a 64-bit value. This helper
/// method might only calculate a fraction of a larger immediate. Therefore it
/// is valid to return a cost of ZERO.
unsigned AArch64TTI::getIntImmCost(int64_t Val) const {
  // Check if the immediate can be encoded within an instruction.
  if (Val == 0 || AArch64_AM::isLogicalImmediate(Val, 64))
    return 0;

  if (Val < 0)
    Val = ~Val;

  // Calculate how many moves we will need to materialize this constant.
  unsigned LZ = countLeadingZeros((uint64_t)Val);
  return (64 - LZ + 15) / 16;
}

/// \brief Calculate the cost of materializing the given constant.
unsigned AArch64TTI::getIntImmCost(const APInt &Imm, Type *Ty) const {
  assert(Ty->isIntegerTy());

  unsigned BitSize = Ty->getPrimitiveSizeInBits();
  if (BitSize == 0)
    return ~0U;

  // Sign-extend all constants to a multiple of 64-bit.
  APInt ImmVal = Imm;
  if (BitSize & 0x3f)
    ImmVal = Imm.sext((BitSize + 63) & ~0x3fU);

  // Split the constant into 64-bit chunks and calculate the cost for each
  // chunk.
  unsigned Cost = 0;
  for (unsigned ShiftVal = 0; ShiftVal < BitSize; ShiftVal += 64) {
    APInt Tmp = ImmVal.ashr(ShiftVal).sextOrTrunc(64);
    int64_t Val = Tmp.getSExtValue();
    Cost += getIntImmCost(Val);
  }
  // We need at least one instruction to materialze the constant.
  return std::max(1U, Cost);
}

unsigned AArch64TTI::getIntImmCost(unsigned Opcode, unsigned Idx,
                                 const APInt &Imm, Type *Ty) const {
  assert(Ty->isIntegerTy());

  unsigned BitSize = Ty->getPrimitiveSizeInBits();
  // There is no cost model for constants with a bit size of 0. Return TCC_Free
  // here, so that constant hoisting will ignore this constant.
  if (BitSize == 0)
    return TCC_Free;

  unsigned ImmIdx = ~0U;
  switch (Opcode) {
  default:
    return TCC_Free;
  case Instruction::GetElementPtr:
    // Always hoist the base address of a GetElementPtr.
    if (Idx == 0)
      return 2 * TCC_Basic;
    return TCC_Free;
  case Instruction::Store:
    ImmIdx = 0;
    break;
  case Instruction::Add:
  case Instruction::Sub:
  case Instruction::Mul:
  case Instruction::UDiv:
  case Instruction::SDiv:
  case Instruction::URem:
  case Instruction::SRem:
  case Instruction::And:
  case Instruction::Or:
  case Instruction::Xor:
  case Instruction::ICmp:
    ImmIdx = 1;
    break;
  // Always return TCC_Free for the shift value of a shift instruction.
  case Instruction::Shl:
  case Instruction::LShr:
  case Instruction::AShr:
    if (Idx == 1)
      return TCC_Free;
    break;
  case Instruction::Trunc:
  case Instruction::ZExt:
  case Instruction::SExt:
  case Instruction::IntToPtr:
  case Instruction::PtrToInt:
  case Instruction::BitCast:
  case Instruction::PHI:
  case Instruction::Call:
  case Instruction::Select:
  case Instruction::Ret:
  case Instruction::Load:
    break;
  }

  if (Idx == ImmIdx) {
    unsigned NumConstants = (BitSize + 63) / 64;
    unsigned Cost = AArch64TTI::getIntImmCost(Imm, Ty);
    return (Cost <= NumConstants * TCC_Basic)
      ? static_cast<unsigned>(TCC_Free) : Cost;
  }
  return AArch64TTI::getIntImmCost(Imm, Ty);
}

unsigned AArch64TTI::getIntImmCost(Intrinsic::ID IID, unsigned Idx,
                                 const APInt &Imm, Type *Ty) const {
  assert(Ty->isIntegerTy());

  unsigned BitSize = Ty->getPrimitiveSizeInBits();
  // There is no cost model for constants with a bit size of 0. Return TCC_Free
  // here, so that constant hoisting will ignore this constant.
  if (BitSize == 0)
    return TCC_Free;

  switch (IID) {
  default:
    return TCC_Free;
  case Intrinsic::sadd_with_overflow:
  case Intrinsic::uadd_with_overflow:
  case Intrinsic::ssub_with_overflow:
  case Intrinsic::usub_with_overflow:
  case Intrinsic::smul_with_overflow:
  case Intrinsic::umul_with_overflow:
    if (Idx == 1) {
      unsigned NumConstants = (BitSize + 63) / 64;
      unsigned Cost = AArch64TTI::getIntImmCost(Imm, Ty);
      return (Cost <= NumConstants * TCC_Basic)
        ? static_cast<unsigned>(TCC_Free) : Cost;
    }
    break;
  case Intrinsic::experimental_stackmap:
    if ((Idx < 2) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
      return TCC_Free;
    break;
  case Intrinsic::experimental_patchpoint_void:
  case Intrinsic::experimental_patchpoint_i64:
    if ((Idx < 4) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
      return TCC_Free;
    break;
  }
  return AArch64TTI::getIntImmCost(Imm, Ty);
}

AArch64TTI::PopcntSupportKind
AArch64TTI::getPopcntSupport(unsigned TyWidth) const {
  assert(isPowerOf2_32(TyWidth) && "Ty width must be power of 2");
  if (TyWidth == 32 || TyWidth == 64)
    return PSK_FastHardware;
  // TODO: AArch64TargetLowering::LowerCTPOP() supports 128bit popcount.
  return PSK_Software;
}

unsigned AArch64TTI::getCastInstrCost(unsigned Opcode, Type *Dst,
                                    Type *Src) const {
  int ISD = TLI->InstructionOpcodeToISD(Opcode);
  assert(ISD && "Invalid opcode");

  EVT SrcTy = TLI->getValueType(Src);
  EVT DstTy = TLI->getValueType(Dst);

  if (!SrcTy.isSimple() || !DstTy.isSimple())
    return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);

  static const TypeConversionCostTblEntry<MVT> ConversionTbl[] = {
    // LowerVectorINT_TO_FP:
    { ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i32, 1 },
    { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i32, 1 },
    { ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i64, 1 },
    { ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i32, 1 },
    { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i32, 1 },
    { ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i64, 1 },

    // Complex: to v2f32
    { ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i8,  3 },
    { ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i16, 3 },
    { ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i64, 2 },
    { ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i8,  3 },
    { ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i16, 3 },
    { ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i64, 2 },

    // Complex: to v4f32
    { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i8,  4 },
    { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i16, 2 },
    { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i8,  3 },
    { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i16, 2 },

    // Complex: to v2f64
    { ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i8,  4 },
    { ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i16, 4 },
    { ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i32, 2 },
    { ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i8,  4 },
    { ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i16, 4 },
    { ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i32, 2 },


    // LowerVectorFP_TO_INT
    { ISD::FP_TO_SINT, MVT::v2i32, MVT::v2f32, 1 },
    { ISD::FP_TO_SINT, MVT::v4i32, MVT::v4f32, 1 },
    { ISD::FP_TO_SINT, MVT::v2i64, MVT::v2f64, 1 },
    { ISD::FP_TO_UINT, MVT::v2i32, MVT::v2f32, 1 },
    { ISD::FP_TO_UINT, MVT::v4i32, MVT::v4f32, 1 },
    { ISD::FP_TO_UINT, MVT::v2i64, MVT::v2f64, 1 },

    // Complex, from v2f32: legal type is v2i32 (no cost) or v2i64 (1 ext).
    { ISD::FP_TO_SINT, MVT::v2i64, MVT::v2f32, 2 },
    { ISD::FP_TO_SINT, MVT::v2i16, MVT::v2f32, 1 },
    { ISD::FP_TO_SINT, MVT::v2i8,  MVT::v2f32, 1 },
    { ISD::FP_TO_UINT, MVT::v2i64, MVT::v2f32, 2 },
    { ISD::FP_TO_UINT, MVT::v2i16, MVT::v2f32, 1 },
    { ISD::FP_TO_UINT, MVT::v2i8,  MVT::v2f32, 1 },

    // Complex, from v4f32: legal type is v4i16, 1 narrowing => ~2
    { ISD::FP_TO_SINT, MVT::v4i16, MVT::v4f32, 2 },
    { ISD::FP_TO_SINT, MVT::v4i8,  MVT::v4f32, 2 },
    { ISD::FP_TO_UINT, MVT::v4i16, MVT::v4f32, 2 },
    { ISD::FP_TO_UINT, MVT::v4i8,  MVT::v4f32, 2 },

    // Complex, from v2f64: legal type is v2i32, 1 narrowing => ~2.
    { ISD::FP_TO_SINT, MVT::v2i32, MVT::v2f64, 2 },
    { ISD::FP_TO_SINT, MVT::v2i16, MVT::v2f64, 2 },
    { ISD::FP_TO_SINT, MVT::v2i8,  MVT::v2f64, 2 },
    { ISD::FP_TO_UINT, MVT::v2i32, MVT::v2f64, 2 },
    { ISD::FP_TO_UINT, MVT::v2i16, MVT::v2f64, 2 },
    { ISD::FP_TO_UINT, MVT::v2i8,  MVT::v2f64, 2 },
  };

  int Idx = ConvertCostTableLookup<MVT>(
      ConversionTbl, array_lengthof(ConversionTbl), ISD, DstTy.getSimpleVT(),
      SrcTy.getSimpleVT());
  if (Idx != -1)
    return ConversionTbl[Idx].Cost;

  return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
}

unsigned AArch64TTI::getVectorInstrCost(unsigned Opcode, Type *Val,
                                      unsigned Index) const {
  assert(Val->isVectorTy() && "This must be a vector type");

  if (Index != -1U) {
    // Legalize the type.
    std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Val);

    // This type is legalized to a scalar type.
    if (!LT.second.isVector())
      return 0;

    // The type may be split. Normalize the index to the new type.
    unsigned Width = LT.second.getVectorNumElements();
    Index = Index % Width;

    // The element at index zero is already inside the vector.
    if (Index == 0)
      return 0;
  }

  // All other insert/extracts cost this much.
  return 2;
}

unsigned AArch64TTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty,
                                          OperandValueKind Opd1Info,
                                          OperandValueKind Opd2Info) const {
  // Legalize the type.
  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Ty);

  int ISD = TLI->InstructionOpcodeToISD(Opcode);

  switch (ISD) {
  default:
    return TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty, Opd1Info,
                                                       Opd2Info);
  case ISD::ADD:
  case ISD::MUL:
  case ISD::XOR:
  case ISD::OR:
  case ISD::AND:
    // These nodes are marked as 'custom' for combining purposes only.
    // We know that they are legal. See LowerAdd in ISelLowering.
    return 1 * LT.first;
  }
}

unsigned AArch64TTI::getAddressComputationCost(Type *Ty, bool IsComplex) const {
  // Address computations in vectorized code with non-consecutive addresses will
  // likely result in more instructions compared to scalar code where the
  // computation can more often be merged into the index mode. The resulting
  // extra micro-ops can significantly decrease throughput.
  unsigned NumVectorInstToHideOverhead = 10;

  if (Ty->isVectorTy() && IsComplex)
    return NumVectorInstToHideOverhead;

  // In many cases the address computation is not merged into the instruction
  // addressing mode.
  return 1;
}

unsigned AArch64TTI::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
                                      Type *CondTy) const {

  int ISD = TLI->InstructionOpcodeToISD(Opcode);
  // We don't lower vector selects well that are wider than the register width.
  if (ValTy->isVectorTy() && ISD == ISD::SELECT) {
    // We would need this many instructions to hide the scalarization happening.
    unsigned AmortizationCost = 20;
    static const TypeConversionCostTblEntry<MVT::SimpleValueType>
    VectorSelectTbl[] = {
      { ISD::SELECT, MVT::v16i1, MVT::v16i16, 16 * AmortizationCost },
      { ISD::SELECT, MVT::v8i1, MVT::v8i32, 8 * AmortizationCost },
      { ISD::SELECT, MVT::v16i1, MVT::v16i32, 16 * AmortizationCost },
      { ISD::SELECT, MVT::v4i1, MVT::v4i64, 4 * AmortizationCost },
      { ISD::SELECT, MVT::v8i1, MVT::v8i64, 8 * AmortizationCost },
      { ISD::SELECT, MVT::v16i1, MVT::v16i64, 16 * AmortizationCost }
    };

    EVT SelCondTy = TLI->getValueType(CondTy);
    EVT SelValTy = TLI->getValueType(ValTy);
    if (SelCondTy.isSimple() && SelValTy.isSimple()) {
      int Idx =
          ConvertCostTableLookup(VectorSelectTbl, ISD, SelCondTy.getSimpleVT(),
                                 SelValTy.getSimpleVT());
      if (Idx != -1)
        return VectorSelectTbl[Idx].Cost;
    }
  }
  return TargetTransformInfo::getCmpSelInstrCost(Opcode, ValTy, CondTy);
}

unsigned AArch64TTI::getMemoryOpCost(unsigned Opcode, Type *Src,
                                   unsigned Alignment,
                                   unsigned AddressSpace) const {
  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);

  if (Opcode == Instruction::Store && Src->isVectorTy() && Alignment != 16 &&
      Src->getVectorElementType()->isIntegerTy(64)) {
    // Unaligned stores are extremely inefficient. We don't split
    // unaligned v2i64 stores because the negative impact that has shown in
    // practice on inlined memcpy code.
    // We make v2i64 stores expensive so that we will only vectorize if there
    // are 6 other instructions getting vectorized.
    unsigned AmortizationCost = 6;

    return LT.first * 2 * AmortizationCost;
  }

  if (Src->isVectorTy() && Src->getVectorElementType()->isIntegerTy(8) &&
      Src->getVectorNumElements() < 8) {
    // We scalarize the loads/stores because there is not v.4b register and we
    // have to promote the elements to v.4h.
    unsigned NumVecElts = Src->getVectorNumElements();
    unsigned NumVectorizableInstsToAmortize = NumVecElts * 2;
    // We generate 2 instructions per vector element.
    return NumVectorizableInstsToAmortize * NumVecElts * 2;
  }

  return LT.first;
}