aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp
blob: 4deb2a9485e4d41314faeb262a52f2d09d036072 (plain)
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
//===-- GCNPreRAOptimizations.cpp -----------------------------------------===//
//
// 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
/// This pass combines split register tuple initialization into a single pseudo:
///
///   undef %0.sub1:sreg_64 = S_MOV_B32 1
///   %0.sub0:sreg_64 = S_MOV_B32 2
/// =>
///   %0:sreg_64 = S_MOV_B64_IMM_PSEUDO 0x200000001
///
/// This is to allow rematerialization of a value instead of spilling. It is
/// supposed to be done after register coalescer to allow it to do its job and
/// before actual register allocation to allow rematerialization.
///
/// Right now the pass only handles 64 bit SGPRs with immediate initializers,
/// although the same shall be possible with other register classes and
/// instructions if necessary.
///
/// This pass also adds register allocation hints to COPY.
/// The hints will be post-processed by SIRegisterInfo::getRegAllocationHints.
/// When using True16, we often see COPY moving a 16-bit value between a VGPR_32
/// and a VGPR_16. If we use the VGPR_16 that corresponds to the lo16 bits of
/// the VGPR_32, the COPY can be completely eliminated.
///
//===----------------------------------------------------------------------===//

#include "GCNPreRAOptimizations.h"
#include "AMDGPU.h"
#include "GCNSubtarget.h"
#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
#include "SIRegisterInfo.h"
#include "llvm/CodeGen/LiveIntervals.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/InitializePasses.h"

using namespace llvm;

#define DEBUG_TYPE "amdgpu-pre-ra-optimizations"

namespace {

class GCNPreRAOptimizationsImpl {
private:
  const SIInstrInfo *TII;
  const SIRegisterInfo *TRI;
  MachineRegisterInfo *MRI;
  LiveIntervals *LIS;

  bool processReg(Register Reg);

public:
  GCNPreRAOptimizationsImpl(LiveIntervals *LS) : LIS(LS) {}
  bool run(MachineFunction &MF);
};

class GCNPreRAOptimizationsLegacy : public MachineFunctionPass {
public:
  static char ID;

  GCNPreRAOptimizationsLegacy() : MachineFunctionPass(ID) {
    initializeGCNPreRAOptimizationsLegacyPass(*PassRegistry::getPassRegistry());
  }

  bool runOnMachineFunction(MachineFunction &MF) override;

  StringRef getPassName() const override {
    return "AMDGPU Pre-RA optimizations";
  }

  void getAnalysisUsage(AnalysisUsage &AU) const override {
    AU.addRequired<LiveIntervalsWrapperPass>();
    AU.setPreservesAll();
    MachineFunctionPass::getAnalysisUsage(AU);
  }
};
} // End anonymous namespace.

INITIALIZE_PASS_BEGIN(GCNPreRAOptimizationsLegacy, DEBUG_TYPE,
                      "AMDGPU Pre-RA optimizations", false, false)
INITIALIZE_PASS_DEPENDENCY(LiveIntervalsWrapperPass)
INITIALIZE_PASS_END(GCNPreRAOptimizationsLegacy, DEBUG_TYPE,
                    "Pre-RA optimizations", false, false)

char GCNPreRAOptimizationsLegacy::ID = 0;

char &llvm::GCNPreRAOptimizationsID = GCNPreRAOptimizationsLegacy::ID;

FunctionPass *llvm::createGCNPreRAOptimizationsLegacyPass() {
  return new GCNPreRAOptimizationsLegacy();
}

bool GCNPreRAOptimizationsImpl::processReg(Register Reg) {
  MachineInstr *Def0 = nullptr;
  MachineInstr *Def1 = nullptr;
  uint64_t Init = 0;
  bool Changed = false;
  SmallSet<Register, 32> ModifiedRegs;
  bool IsAGPRDst = TRI->isAGPRClass(MRI->getRegClass(Reg));

  for (MachineInstr &I : MRI->def_instructions(Reg)) {
    switch (I.getOpcode()) {
    default:
      return false;
    case AMDGPU::V_ACCVGPR_WRITE_B32_e64:
      break;
    case AMDGPU::COPY: {
      // Some subtargets cannot do an AGPR to AGPR copy directly, and need an
      // intermdiate temporary VGPR register. Try to find the defining
      // accvgpr_write to avoid temporary registers.

      if (!IsAGPRDst)
        return false;

      Register SrcReg = I.getOperand(1).getReg();

      if (!SrcReg.isVirtual())
        break;

      // Check if source of copy is from another AGPR.
      bool IsAGPRSrc = TRI->isAGPRClass(MRI->getRegClass(SrcReg));
      if (!IsAGPRSrc)
        break;

      // def_instructions() does not look at subregs so it may give us a
      // different instruction that defines the same vreg but different subreg
      // so we have to manually check subreg.
      Register SrcSubReg = I.getOperand(1).getSubReg();
      for (auto &Def : MRI->def_instructions(SrcReg)) {
        if (SrcSubReg != Def.getOperand(0).getSubReg())
          continue;

        if (Def.getOpcode() == AMDGPU::V_ACCVGPR_WRITE_B32_e64) {
          MachineOperand DefSrcMO = Def.getOperand(1);

          // Immediates are not an issue and can be propagated in
          // postrapseudos pass. Only handle cases where defining
          // accvgpr_write source is a vreg.
          if (DefSrcMO.isReg() && DefSrcMO.getReg().isVirtual()) {
            // Propagate source reg of accvgpr write to this copy instruction
            I.getOperand(1).setReg(DefSrcMO.getReg());
            I.getOperand(1).setSubReg(DefSrcMO.getSubReg());

            // Reg uses were changed, collect unique set of registers to update
            // live intervals at the end.
            ModifiedRegs.insert(DefSrcMO.getReg());
            ModifiedRegs.insert(SrcReg);

            Changed = true;
          }

          // Found the defining accvgpr_write, stop looking any further.
          break;
        }
      }
      break;
    }
    case AMDGPU::S_MOV_B32:
      if (I.getOperand(0).getReg() != Reg || !I.getOperand(1).isImm() ||
          I.getNumOperands() != 2)
        return false;

      switch (I.getOperand(0).getSubReg()) {
      default:
        return false;
      case AMDGPU::sub0:
        if (Def0)
          return false;
        Def0 = &I;
        Init |= Lo_32(I.getOperand(1).getImm());
        break;
      case AMDGPU::sub1:
        if (Def1)
          return false;
        Def1 = &I;
        Init |= static_cast<uint64_t>(I.getOperand(1).getImm()) << 32;
        break;
      }
      break;
    }
  }

  // For AGPR reg, check if live intervals need to be updated.
  if (IsAGPRDst) {
    if (Changed) {
      for (Register RegToUpdate : ModifiedRegs) {
        LIS->removeInterval(RegToUpdate);
        LIS->createAndComputeVirtRegInterval(RegToUpdate);
      }
    }

    return Changed;
  }

  // For SGPR reg, check if we can combine instructions.
  if (!Def0 || !Def1 || Def0->getParent() != Def1->getParent())
    return Changed;

  LLVM_DEBUG(dbgs() << "Combining:\n  " << *Def0 << "  " << *Def1
                    << "    =>\n");

  if (SlotIndex::isEarlierInstr(LIS->getInstructionIndex(*Def1),
                                LIS->getInstructionIndex(*Def0)))
    std::swap(Def0, Def1);

  LIS->RemoveMachineInstrFromMaps(*Def0);
  LIS->RemoveMachineInstrFromMaps(*Def1);
  auto NewI = BuildMI(*Def0->getParent(), *Def0, Def0->getDebugLoc(),
                      TII->get(AMDGPU::S_MOV_B64_IMM_PSEUDO), Reg)
                  .addImm(Init);

  Def0->eraseFromParent();
  Def1->eraseFromParent();
  LIS->InsertMachineInstrInMaps(*NewI);
  LIS->removeInterval(Reg);
  LIS->createAndComputeVirtRegInterval(Reg);

  LLVM_DEBUG(dbgs() << "  " << *NewI);

  return true;
}

bool GCNPreRAOptimizationsLegacy::runOnMachineFunction(MachineFunction &MF) {
  if (skipFunction(MF.getFunction()))
    return false;
  LiveIntervals *LIS = &getAnalysis<LiveIntervalsWrapperPass>().getLIS();
  return GCNPreRAOptimizationsImpl(LIS).run(MF);
}

PreservedAnalyses
GCNPreRAOptimizationsPass::run(MachineFunction &MF,
                               MachineFunctionAnalysisManager &MFAM) {
  LiveIntervals *LIS = &MFAM.getResult<LiveIntervalsAnalysis>(MF);
  GCNPreRAOptimizationsImpl(LIS).run(MF);
  return PreservedAnalyses::all();
}

bool GCNPreRAOptimizationsImpl::run(MachineFunction &MF) {
  const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
  TII = ST.getInstrInfo();
  MRI = &MF.getRegInfo();
  TRI = ST.getRegisterInfo();

  bool Changed = false;

  for (unsigned I = 0, E = MRI->getNumVirtRegs(); I != E; ++I) {
    Register Reg = Register::index2VirtReg(I);
    if (!LIS->hasInterval(Reg))
      continue;
    const TargetRegisterClass *RC = MRI->getRegClass(Reg);
    if ((RC->MC->getSizeInBits() != 64 || !TRI->isSGPRClass(RC)) &&
        (ST.hasGFX90AInsts() || !TRI->isAGPRClass(RC)))
      continue;

    Changed |= processReg(Reg);
  }

  if (!ST.useRealTrue16Insts())
    return Changed;

  // Add RA hints to improve True16 COPY elimination.
  for (const MachineBasicBlock &MBB : MF) {
    for (const MachineInstr &MI : MBB) {
      if (MI.getOpcode() != AMDGPU::COPY)
        continue;
      Register Dst = MI.getOperand(0).getReg();
      Register Src = MI.getOperand(1).getReg();
      if (Dst.isVirtual() &&
          MRI->getRegClass(Dst) == &AMDGPU::VGPR_16RegClass &&
          Src.isPhysical() &&
          TRI->getRegClassForReg(*MRI, Src) == &AMDGPU::VGPR_32RegClass)
        MRI->setRegAllocationHint(Dst, 0, TRI->getSubReg(Src, AMDGPU::lo16));
      if (Src.isVirtual() &&
          MRI->getRegClass(Src) == &AMDGPU::VGPR_16RegClass &&
          Dst.isPhysical() &&
          TRI->getRegClassForReg(*MRI, Dst) == &AMDGPU::VGPR_32RegClass)
        MRI->setRegAllocationHint(Src, 0, TRI->getSubReg(Dst, AMDGPU::lo16));
      if (!Dst.isVirtual() || !Src.isVirtual())
        continue;
      if (MRI->getRegClass(Dst) == &AMDGPU::VGPR_32RegClass &&
          MRI->getRegClass(Src) == &AMDGPU::VGPR_16RegClass) {
        MRI->setRegAllocationHint(Dst, AMDGPURI::Size32, Src);
        MRI->setRegAllocationHint(Src, AMDGPURI::Size16, Dst);
      }
      if (MRI->getRegClass(Dst) == &AMDGPU::VGPR_16RegClass &&
          MRI->getRegClass(Src) == &AMDGPU::VGPR_32RegClass)
        MRI->setRegAllocationHint(Dst, AMDGPURI::Size16, Src);
    }
  }

  return Changed;
}