Training courses

Kernel and Embedded Linux

Bootlin training courses

Embedded Linux, kernel,
Yocto Project, Buildroot, real-time,
graphics, boot time, debugging...

Bootlin logo

Elixir Cross Referencer

  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
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
//===-- SIMCCodeEmitter.cpp - SI Code Emitter -----------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
/// \file
/// The SI code emitter produces machine code that can be executed
/// directly on the GPU device.
//
//===----------------------------------------------------------------------===//

#include "AMDGPU.h"
#include "MCTargetDesc/AMDGPUFixupKinds.h"
#include "MCTargetDesc/AMDGPUMCCodeEmitter.h"
#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
#include "SIDefines.h"
#include "Utils/AMDGPUBaseInfo.h"
#include "llvm/MC/MCCodeEmitter.h"
#include "llvm/MC/MCContext.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCFixup.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCInstrDesc.h"
#include "llvm/MC/MCInstrInfo.h"
#include "llvm/MC/MCRegisterInfo.h"
#include "llvm/MC/MCSubtargetInfo.h"
#include "llvm/MC/MCSymbol.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <cassert>
#include <cstdint>
#include <cstdlib>

using namespace llvm;

namespace {

class SIMCCodeEmitter : public  AMDGPUMCCodeEmitter {
  const MCRegisterInfo &MRI;

  /// Encode an fp or int literal
  uint32_t getLitEncoding(const MCOperand &MO, const MCOperandInfo &OpInfo,
                          const MCSubtargetInfo &STI) const;

public:
  SIMCCodeEmitter(const MCInstrInfo &mcii, const MCRegisterInfo &mri,
                  MCContext &ctx)
      : AMDGPUMCCodeEmitter(mcii), MRI(mri) {}
  SIMCCodeEmitter(const SIMCCodeEmitter &) = delete;
  SIMCCodeEmitter &operator=(const SIMCCodeEmitter &) = delete;

  /// Encode the instruction and write it to the OS.
  void encodeInstruction(const MCInst &MI, raw_ostream &OS,
                         SmallVectorImpl<MCFixup> &Fixups,
                         const MCSubtargetInfo &STI) const override;

  /// \returns the encoding for an MCOperand.
  uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
                             SmallVectorImpl<MCFixup> &Fixups,
                             const MCSubtargetInfo &STI) const override;

  /// Use a fixup to encode the simm16 field for SOPP branch
  ///        instructions.
  unsigned getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
                             SmallVectorImpl<MCFixup> &Fixups,
                             const MCSubtargetInfo &STI) const override;

  unsigned getSMEMOffsetEncoding(const MCInst &MI, unsigned OpNo,
                                 SmallVectorImpl<MCFixup> &Fixups,
                                 const MCSubtargetInfo &STI) const override;

  unsigned getSDWASrcEncoding(const MCInst &MI, unsigned OpNo,
                              SmallVectorImpl<MCFixup> &Fixups,
                              const MCSubtargetInfo &STI) const override;

  unsigned getSDWAVopcDstEncoding(const MCInst &MI, unsigned OpNo,
                                  SmallVectorImpl<MCFixup> &Fixups,
                                  const MCSubtargetInfo &STI) const override;

  unsigned getAVOperandEncoding(const MCInst &MI, unsigned OpNo,
                                SmallVectorImpl<MCFixup> &Fixups,
                                const MCSubtargetInfo &STI) const override;
};

} // end anonymous namespace

MCCodeEmitter *llvm::createSIMCCodeEmitter(const MCInstrInfo &MCII,
                                           const MCRegisterInfo &MRI,
                                           MCContext &Ctx) {
  return new SIMCCodeEmitter(MCII, MRI, Ctx);
}

// Returns the encoding value to use if the given integer is an integer inline
// immediate value, or 0 if it is not.
template <typename IntTy>
static uint32_t getIntInlineImmEncoding(IntTy Imm) {
  if (Imm >= 0 && Imm <= 64)
    return 128 + Imm;

  if (Imm >= -16 && Imm <= -1)
    return 192 + std::abs(Imm);

  return 0;
}

static uint32_t getLit16IntEncoding(uint16_t Val, const MCSubtargetInfo &STI) {
  uint16_t IntImm = getIntInlineImmEncoding(static_cast<int16_t>(Val));
  return IntImm == 0 ? 255 : IntImm;
}

static uint32_t getLit16Encoding(uint16_t Val, const MCSubtargetInfo &STI) {
  uint16_t IntImm = getIntInlineImmEncoding(static_cast<int16_t>(Val));
  if (IntImm != 0)
    return IntImm;

  if (Val == 0x3800) // 0.5
    return 240;

  if (Val == 0xB800) // -0.5
    return 241;

  if (Val == 0x3C00) // 1.0
    return 242;

  if (Val == 0xBC00) // -1.0
    return 243;

  if (Val == 0x4000) // 2.0
    return 244;

  if (Val == 0xC000) // -2.0
    return 245;

  if (Val == 0x4400) // 4.0
    return 246;

  if (Val == 0xC400) // -4.0
    return 247;

  if (Val == 0x3118 && // 1.0 / (2.0 * pi)
      STI.getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm])
    return 248;

  return 255;
}

static uint32_t getLit32Encoding(uint32_t Val, const MCSubtargetInfo &STI) {
  uint32_t IntImm = getIntInlineImmEncoding(static_cast<int32_t>(Val));
  if (IntImm != 0)
    return IntImm;

  if (Val == FloatToBits(0.5f))
    return 240;

  if (Val == FloatToBits(-0.5f))
    return 241;

  if (Val == FloatToBits(1.0f))
    return 242;

  if (Val == FloatToBits(-1.0f))
    return 243;

  if (Val == FloatToBits(2.0f))
    return 244;

  if (Val == FloatToBits(-2.0f))
    return 245;

  if (Val == FloatToBits(4.0f))
    return 246;

  if (Val == FloatToBits(-4.0f))
    return 247;

  if (Val == 0x3e22f983 && // 1.0 / (2.0 * pi)
      STI.getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm])
    return 248;

  return 255;
}

static uint32_t getLit64Encoding(uint64_t Val, const MCSubtargetInfo &STI) {
  uint32_t IntImm = getIntInlineImmEncoding(static_cast<int64_t>(Val));
  if (IntImm != 0)
    return IntImm;

  if (Val == DoubleToBits(0.5))
    return 240;

  if (Val == DoubleToBits(-0.5))
    return 241;

  if (Val == DoubleToBits(1.0))
    return 242;

  if (Val == DoubleToBits(-1.0))
    return 243;

  if (Val == DoubleToBits(2.0))
    return 244;

  if (Val == DoubleToBits(-2.0))
    return 245;

  if (Val == DoubleToBits(4.0))
    return 246;

  if (Val == DoubleToBits(-4.0))
    return 247;

  if (Val == 0x3fc45f306dc9c882 && // 1.0 / (2.0 * pi)
      STI.getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm])
    return 248;

  return 255;
}

uint32_t SIMCCodeEmitter::getLitEncoding(const MCOperand &MO,
                                         const MCOperandInfo &OpInfo,
                                         const MCSubtargetInfo &STI) const {
  int64_t Imm;
  if (MO.isExpr()) {
    const auto *C = dyn_cast<MCConstantExpr>(MO.getExpr());
    if (!C)
      return 255;

    Imm = C->getValue();
  } else {

    assert(!MO.isFPImm());

    if (!MO.isImm())
      return ~0;

    Imm = MO.getImm();
  }

  switch (OpInfo.OperandType) {
  case AMDGPU::OPERAND_REG_IMM_INT32:
  case AMDGPU::OPERAND_REG_IMM_FP32:
  case AMDGPU::OPERAND_REG_INLINE_C_INT32:
  case AMDGPU::OPERAND_REG_INLINE_C_FP32:
  case AMDGPU::OPERAND_REG_INLINE_AC_INT32:
  case AMDGPU::OPERAND_REG_INLINE_AC_FP32:
    return getLit32Encoding(static_cast<uint32_t>(Imm), STI);

  case AMDGPU::OPERAND_REG_IMM_INT64:
  case AMDGPU::OPERAND_REG_IMM_FP64:
  case AMDGPU::OPERAND_REG_INLINE_C_INT64:
  case AMDGPU::OPERAND_REG_INLINE_C_FP64:
    return getLit64Encoding(static_cast<uint64_t>(Imm), STI);

  case AMDGPU::OPERAND_REG_IMM_INT16:
  case AMDGPU::OPERAND_REG_INLINE_C_INT16:
  case AMDGPU::OPERAND_REG_INLINE_AC_INT16:
    return getLit16IntEncoding(static_cast<uint16_t>(Imm), STI);
  case AMDGPU::OPERAND_REG_IMM_FP16:
  case AMDGPU::OPERAND_REG_INLINE_C_FP16:
  case AMDGPU::OPERAND_REG_INLINE_AC_FP16:
    // FIXME Is this correct? What do inline immediates do on SI for f16 src
    // which does not have f16 support?
    return getLit16Encoding(static_cast<uint16_t>(Imm), STI);
  case AMDGPU::OPERAND_REG_IMM_V2INT16:
  case AMDGPU::OPERAND_REG_IMM_V2FP16: {
    if (!isUInt<16>(Imm) && STI.getFeatureBits()[AMDGPU::FeatureVOP3Literal])
      return getLit32Encoding(static_cast<uint32_t>(Imm), STI);
    if (OpInfo.OperandType == AMDGPU::OPERAND_REG_IMM_V2FP16)
      return getLit16Encoding(static_cast<uint16_t>(Imm), STI);
    LLVM_FALLTHROUGH;
  }
  case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
  case AMDGPU::OPERAND_REG_INLINE_AC_V2INT16:
    return getLit16IntEncoding(static_cast<uint16_t>(Imm), STI);
  case AMDGPU::OPERAND_REG_INLINE_C_V2FP16:
  case AMDGPU::OPERAND_REG_INLINE_AC_V2FP16: {
    uint16_t Lo16 = static_cast<uint16_t>(Imm);
    uint32_t Encoding = getLit16Encoding(Lo16, STI);
    return Encoding;
  }
  default:
    llvm_unreachable("invalid operand size");
  }
}

void SIMCCodeEmitter::encodeInstruction(const MCInst &MI, raw_ostream &OS,
                                       SmallVectorImpl<MCFixup> &Fixups,
                                       const MCSubtargetInfo &STI) const {
  verifyInstructionPredicates(MI,
                              computeAvailableFeatures(STI.getFeatureBits()));

  uint64_t Encoding = getBinaryCodeForInstr(MI, Fixups, STI);
  const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
  unsigned bytes = Desc.getSize();

  for (unsigned i = 0; i < bytes; i++) {
    OS.write((uint8_t) ((Encoding >> (8 * i)) & 0xff));
  }

  // NSA encoding.
  if (AMDGPU::isGFX10(STI) && Desc.TSFlags & SIInstrFlags::MIMG) {
    int vaddr0 = AMDGPU::getNamedOperandIdx(MI.getOpcode(),
                                            AMDGPU::OpName::vaddr0);
    int srsrc = AMDGPU::getNamedOperandIdx(MI.getOpcode(),
                                           AMDGPU::OpName::srsrc);
    assert(vaddr0 >= 0 && srsrc > vaddr0);
    unsigned NumExtraAddrs = srsrc - vaddr0 - 1;
    unsigned NumPadding = (-NumExtraAddrs) & 3;

    for (unsigned i = 0; i < NumExtraAddrs; ++i)
      OS.write((uint8_t)getMachineOpValue(MI, MI.getOperand(vaddr0 + 1 + i),
                                          Fixups, STI));
    for (unsigned i = 0; i < NumPadding; ++i)
      OS.write(0);
  }

  if ((bytes > 8 && STI.getFeatureBits()[AMDGPU::FeatureVOP3Literal]) ||
      (bytes > 4 && !STI.getFeatureBits()[AMDGPU::FeatureVOP3Literal]))
    return;

  // Check for additional literals in SRC0/1/2 (Op 1/2/3)
  for (unsigned i = 0, e = Desc.getNumOperands(); i < e; ++i) {

    // Check if this operand should be encoded as [SV]Src
    if (!AMDGPU::isSISrcOperand(Desc, i))
      continue;

    // Is this operand a literal immediate?
    const MCOperand &Op = MI.getOperand(i);
    if (getLitEncoding(Op, Desc.OpInfo[i], STI) != 255)
      continue;

    // Yes! Encode it
    int64_t Imm = 0;

    if (Op.isImm())
      Imm = Op.getImm();
    else if (Op.isExpr()) {
      if (const auto *C = dyn_cast<MCConstantExpr>(Op.getExpr()))
        Imm = C->getValue();

    } else if (!Op.isExpr()) // Exprs will be replaced with a fixup value.
      llvm_unreachable("Must be immediate or expr");

    for (unsigned j = 0; j < 4; j++) {
      OS.write((uint8_t) ((Imm >> (8 * j)) & 0xff));
    }

    // Only one literal value allowed
    break;
  }
}

unsigned SIMCCodeEmitter::getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
                                            SmallVectorImpl<MCFixup> &Fixups,
                                            const MCSubtargetInfo &STI) const {
  const MCOperand &MO = MI.getOperand(OpNo);

  if (MO.isExpr()) {
    const MCExpr *Expr = MO.getExpr();
    MCFixupKind Kind = (MCFixupKind)AMDGPU::fixup_si_sopp_br;
    Fixups.push_back(MCFixup::create(0, Expr, Kind, MI.getLoc()));
    return 0;
  }

  return getMachineOpValue(MI, MO, Fixups, STI);
}

unsigned SIMCCodeEmitter::getSMEMOffsetEncoding(const MCInst &MI, unsigned OpNo,
                                                SmallVectorImpl<MCFixup> &Fixups,
                                                const MCSubtargetInfo &STI) const {
  auto Offset = MI.getOperand(OpNo).getImm();
  // VI only supports 20-bit unsigned offsets.
  assert(!AMDGPU::isVI(STI) || isUInt<20>(Offset));
  return Offset;
}

unsigned
SIMCCodeEmitter::getSDWASrcEncoding(const MCInst &MI, unsigned OpNo,
                                    SmallVectorImpl<MCFixup> &Fixups,
                                    const MCSubtargetInfo &STI) const {
  using namespace AMDGPU::SDWA;

  uint64_t RegEnc = 0;

  const MCOperand &MO = MI.getOperand(OpNo);

  if (MO.isReg()) {
    unsigned Reg = MO.getReg();
    RegEnc |= MRI.getEncodingValue(Reg);
    RegEnc &= SDWA9EncValues::SRC_VGPR_MASK;
    if (AMDGPU::isSGPR(AMDGPU::mc2PseudoReg(Reg), &MRI)) {
      RegEnc |= SDWA9EncValues::SRC_SGPR_MASK;
    }
    return RegEnc;
  } else {
    const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
    uint32_t Enc = getLitEncoding(MO, Desc.OpInfo[OpNo], STI);
    if (Enc != ~0U && Enc != 255) {
      return Enc | SDWA9EncValues::SRC_SGPR_MASK;
    }
  }

  llvm_unreachable("Unsupported operand kind");
  return 0;
}

unsigned
SIMCCodeEmitter::getSDWAVopcDstEncoding(const MCInst &MI, unsigned OpNo,
                                        SmallVectorImpl<MCFixup> &Fixups,
                                        const MCSubtargetInfo &STI) const {
  using namespace AMDGPU::SDWA;

  uint64_t RegEnc = 0;

  const MCOperand &MO = MI.getOperand(OpNo);

  unsigned Reg = MO.getReg();
  if (Reg != AMDGPU::VCC && Reg != AMDGPU::VCC_LO) {
    RegEnc |= MRI.getEncodingValue(Reg);
    RegEnc &= SDWA9EncValues::VOPC_DST_SGPR_MASK;
    RegEnc |= SDWA9EncValues::VOPC_DST_VCC_MASK;
  }
  return RegEnc;
}

unsigned
SIMCCodeEmitter::getAVOperandEncoding(const MCInst &MI, unsigned OpNo,
                                      SmallVectorImpl<MCFixup> &Fixups,
                                      const MCSubtargetInfo &STI) const {
  unsigned Reg = MI.getOperand(OpNo).getReg();
  uint64_t Enc = MRI.getEncodingValue(Reg);

  // VGPR and AGPR have the same encoding, but SrcA and SrcB operands of mfma
  // instructions use acc[0:1] modifier bits to distinguish. These bits are
  // encoded as a virtual 9th bit of the register for these operands.
  if (MRI.getRegClass(AMDGPU::AGPR_32RegClassID).contains(Reg) ||
      MRI.getRegClass(AMDGPU::AReg_64RegClassID).contains(Reg) ||
      MRI.getRegClass(AMDGPU::AReg_96RegClassID).contains(Reg) ||
      MRI.getRegClass(AMDGPU::AReg_128RegClassID).contains(Reg) ||
      MRI.getRegClass(AMDGPU::AReg_160RegClassID).contains(Reg) ||
      MRI.getRegClass(AMDGPU::AReg_192RegClassID).contains(Reg) ||
      MRI.getRegClass(AMDGPU::AReg_256RegClassID).contains(Reg) ||
      MRI.getRegClass(AMDGPU::AGPR_LO16RegClassID).contains(Reg))
    Enc |= 512;

  return Enc;
}

static bool needsPCRel(const MCExpr *Expr) {
  switch (Expr->getKind()) {
  case MCExpr::SymbolRef: {
    auto *SE = cast<MCSymbolRefExpr>(Expr);
    MCSymbolRefExpr::VariantKind Kind = SE->getKind();
    return Kind != MCSymbolRefExpr::VK_AMDGPU_ABS32_LO &&
           Kind != MCSymbolRefExpr::VK_AMDGPU_ABS32_HI;
  }
  case MCExpr::Binary: {
    auto *BE = cast<MCBinaryExpr>(Expr);
    if (BE->getOpcode() == MCBinaryExpr::Sub)
      return false;
    return needsPCRel(BE->getLHS()) || needsPCRel(BE->getRHS());
  }
  case MCExpr::Unary:
    return needsPCRel(cast<MCUnaryExpr>(Expr)->getSubExpr());
  case MCExpr::Target:
  case MCExpr::Constant:
    return false;
  }
  llvm_unreachable("invalid kind");
}

uint64_t SIMCCodeEmitter::getMachineOpValue(const MCInst &MI,
                                            const MCOperand &MO,
                                       SmallVectorImpl<MCFixup> &Fixups,
                                       const MCSubtargetInfo &STI) const {
  if (MO.isReg())
    return MRI.getEncodingValue(MO.getReg());

  if (MO.isExpr() && MO.getExpr()->getKind() != MCExpr::Constant) {
    // FIXME: If this is expression is PCRel or not should not depend on what
    // the expression looks like. Given that this is just a general expression,
    // it should probably be FK_Data_4 and whatever is producing
    //
    //    s_add_u32 s2, s2, (extern_const_addrspace+16
    //
    // And expecting a PCRel should instead produce
    //
    // .Ltmp1:
    //   s_add_u32 s2, s2, (extern_const_addrspace+16)-.Ltmp1
    MCFixupKind Kind;
    if (needsPCRel(MO.getExpr()))
      Kind = FK_PCRel_4;
    else
      Kind = FK_Data_4;

    const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
    uint32_t Offset = Desc.getSize();
    assert(Offset == 4 || Offset == 8);

    Fixups.push_back(
      MCFixup::create(Offset, MO.getExpr(), Kind, MI.getLoc()));
  }

  // Figure out the operand number, needed for isSrcOperand check
  unsigned OpNo = 0;
  for (unsigned e = MI.getNumOperands(); OpNo < e; ++OpNo) {
    if (&MO == &MI.getOperand(OpNo))
      break;
  }

  const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
  if (AMDGPU::isSISrcOperand(Desc, OpNo)) {
    uint32_t Enc = getLitEncoding(MO, Desc.OpInfo[OpNo], STI);
    if (Enc != ~0U &&
        (Enc != 255 || Desc.getSize() == 4 || Desc.getSize() == 8))
      return Enc;

  } else if (MO.isImm())
    return MO.getImm();

  llvm_unreachable("Encoding of this operand type is not supported yet.");
  return 0;
}

#define ENABLE_INSTR_PREDICATE_VERIFIER
#include "AMDGPUGenMCCodeEmitter.inc"