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
| //===-- SIFixupVectorISel.cpp - Fixup post ISel vector issues -------------===//
//
// 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
/// SIFixupVectorISel pass cleans up post ISEL Vector issues.
/// Currently this will convert GLOBAL_{LOAD|STORE}_*
/// and GLOBAL_Atomic_* instructions into their _SADDR variants,
/// feeding the sreg into the saddr field of the new instruction.
/// We currently handle a REG_SEQUENCE feeding the vaddr
/// and decompose it into a base and index.
///
/// Transform:
/// %17:vgpr_32, %19:sreg_64_xexec = V_ADD_I32_e64 %21:sgpr_32, %22:vgpr_32
/// %18:vgpr_32, %20:sreg_64_xexec = V_ADDC_U32_e64 %25:vgpr_32,
/// %24:vgpr_32, %19:sreg_64_xexec
/// %16:vreg_64 = REG_SEQUENCE %17:vgpr_32, %sub0, %18:vgpr_32, %sub1
/// %11:vreg_64 = COPY %16:vreg_64
/// %10:vgpr_32 = GLOBAL_LOAD_DWORD killed %11:vreg_64, 16, 0, 0
/// Into:
/// %4:sreg_64_xexec = S_LOAD_DWORDX2_IMM %1:sgpr_64, 36, 0
/// %14:vreg_64 = REG_SEQUENCE %6:vgpr_32, %sub0, %15:vgpr_32, %sub1
/// %10:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR %14:vreg_64, %4:sreg_64_xexec,16...
///
//===----------------------------------------------------------------------===//
//
#include "AMDGPU.h"
#include "AMDGPUSubtarget.h"
#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/Support/Debug.h"
#include "llvm/Target/TargetMachine.h"
#define DEBUG_TYPE "si-fixup-vector-isel"
using namespace llvm;
static cl::opt<bool> EnableGlobalSGPRAddr(
"amdgpu-enable-global-sgpr-addr",
cl::desc("Enable use of SGPR regs for GLOBAL LOAD/STORE instructions"),
cl::init(false));
STATISTIC(NumSGPRGlobalOccurs, "Number of global ld/st opportunities");
STATISTIC(NumSGPRGlobalSaddrs, "Number of global sgpr instructions converted");
namespace {
class SIFixupVectorISel : public MachineFunctionPass {
public:
static char ID;
public:
SIFixupVectorISel() : MachineFunctionPass(ID) {
initializeSIFixupVectorISelPass(*PassRegistry::getPassRegistry());
}
bool runOnMachineFunction(MachineFunction &MF) override;
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
MachineFunctionPass::getAnalysisUsage(AU);
}
};
} // End anonymous namespace.
INITIALIZE_PASS(SIFixupVectorISel, DEBUG_TYPE,
"SI Fixup Vector ISel", false, false)
char SIFixupVectorISel::ID = 0;
char &llvm::SIFixupVectorISelID = SIFixupVectorISel::ID;
FunctionPass *llvm::createSIFixupVectorISelPass() {
return new SIFixupVectorISel();
}
static bool findSRegBaseAndIndex(MachineOperand *Op,
unsigned &BaseReg,
unsigned &IndexReg,
MachineRegisterInfo &MRI,
const SIRegisterInfo *TRI) {
SmallVector<MachineOperand *, 8> Worklist;
Worklist.push_back(Op);
while (!Worklist.empty()) {
MachineOperand *WOp = Worklist.pop_back_val();
if (!WOp->isReg() || !Register::isVirtualRegister(WOp->getReg()))
continue;
MachineInstr *DefInst = MRI.getUniqueVRegDef(WOp->getReg());
switch (DefInst->getOpcode()) {
default:
continue;
case AMDGPU::COPY:
Worklist.push_back(&DefInst->getOperand(1));
break;
case AMDGPU::REG_SEQUENCE:
if (DefInst->getNumOperands() != 5)
continue;
Worklist.push_back(&DefInst->getOperand(1));
Worklist.push_back(&DefInst->getOperand(3));
break;
case AMDGPU::V_ADD_I32_e64:
// The V_ADD_* and its analogous V_ADDCV_* are generated by
// a previous pass which lowered from an ADD_64_PSEUDO,
// which generates subregs to break up the 64 bit args.
if (DefInst->getOperand(2).getSubReg() != AMDGPU::NoSubRegister)
continue;
BaseReg = DefInst->getOperand(2).getReg();
if (DefInst->getOperand(3).getSubReg() != AMDGPU::NoSubRegister)
continue;
IndexReg = DefInst->getOperand(3).getReg();
// Chase the IndexReg.
MachineInstr *MI = MRI.getUniqueVRegDef(IndexReg);
if (!MI || !MI->isCopy())
continue;
// Make sure the reg class is 64 bit for Index.
// If the Index register is a subreg, we want it to reference
// a 64 bit register which we will use as the Index reg.
const TargetRegisterClass *IdxRC, *BaseRC;
IdxRC = MRI.getRegClass(MI->getOperand(1).getReg());
if (AMDGPU::getRegBitWidth(IdxRC->getID()) != 64)
continue;
IndexReg = MI->getOperand(1).getReg();
// Chase the BaseReg.
MI = MRI.getUniqueVRegDef(BaseReg);
if (!MI || !MI->isCopy())
continue;
// Make sure the register class is 64 bit for Base.
BaseReg = MI->getOperand(1).getReg();
BaseRC = MRI.getRegClass(BaseReg);
if (AMDGPU::getRegBitWidth(BaseRC->getID()) != 64)
continue;
// Make sure Base is SReg and Index is VReg.
if (!TRI->isSGPRReg(MRI, BaseReg))
return false;
if (!TRI->hasVGPRs(MRI.getRegClass(IndexReg)))
return false;
// clear any killed flags on Index and Base regs, used later.
MRI.clearKillFlags(IndexReg);
MRI.clearKillFlags(BaseReg);
return true;
}
}
return false;
}
// Identify Global LOAD|STORE/ATOMIC and try to convert to _SADDR.
static bool fixupGlobalSaddr(MachineBasicBlock &MBB,
MachineFunction &MF,
MachineRegisterInfo &MRI,
const GCNSubtarget &ST,
const SIInstrInfo *TII,
const SIRegisterInfo *TRI) {
if (!EnableGlobalSGPRAddr)
return false;
bool FuncModified = false;
MachineBasicBlock::iterator I, Next;
for (I = MBB.begin(); I != MBB.end(); I = Next) {
Next = std::next(I);
MachineInstr &MI = *I;
int NewOpcd = AMDGPU::getGlobalSaddrOp(MI.getOpcode());
if (NewOpcd < 0)
continue;
// Update our statistics on opportunities seen.
++NumSGPRGlobalOccurs;
LLVM_DEBUG(dbgs() << "Global Mem opp " << MI << '\n');
// Need a Base and Index or we cant transform to _SADDR.
unsigned BaseReg = 0;
unsigned IndexReg = 0;
MachineOperand *Op = TII->getNamedOperand(MI, AMDGPU::OpName::vaddr);
if (!findSRegBaseAndIndex(Op, BaseReg, IndexReg, MRI, TRI))
continue;
++NumSGPRGlobalSaddrs;
FuncModified = true;
// Create the new _SADDR Memory instruction.
bool HasVdst = TII->getNamedOperand(MI, AMDGPU::OpName::vdst) != nullptr;
MachineOperand *VData = TII->getNamedOperand(MI, AMDGPU::OpName::vdata);
MachineInstr *NewGlob = nullptr;
NewGlob = BuildMI(MBB, I, MI.getDebugLoc(), TII->get(NewOpcd));
if (HasVdst)
NewGlob->addOperand(MF, MI.getOperand(0));
NewGlob->addOperand(MF, MachineOperand::CreateReg(IndexReg, false));
if (VData)
NewGlob->addOperand(MF, *VData);
NewGlob->addOperand(MF, MachineOperand::CreateReg(BaseReg, false));
NewGlob->addOperand(*TII->getNamedOperand(MI, AMDGPU::OpName::offset));
MachineOperand *Glc = TII->getNamedOperand(MI, AMDGPU::OpName::glc);
// Atomics dont have a GLC, so omit the field if not there.
if (Glc)
NewGlob->addOperand(MF, *Glc);
MachineOperand *DLC = TII->getNamedOperand(MI, AMDGPU::OpName::dlc);
if (DLC)
NewGlob->addOperand(MF, *DLC);
NewGlob->addOperand(*TII->getNamedOperand(MI, AMDGPU::OpName::slc));
// _D16 have an vdst_in operand, copy it in.
MachineOperand *VDstInOp = TII->getNamedOperand(MI,
AMDGPU::OpName::vdst_in);
if (VDstInOp)
NewGlob->addOperand(MF, *VDstInOp);
NewGlob->copyImplicitOps(MF, MI);
NewGlob->cloneMemRefs(MF, MI);
// Remove the old Global Memop instruction.
MI.eraseFromParent();
LLVM_DEBUG(dbgs() << "New Global Mem " << *NewGlob << '\n');
}
return FuncModified;
}
bool SIFixupVectorISel::runOnMachineFunction(MachineFunction &MF) {
if (skipFunction(MF.getFunction()))
return false;
MachineRegisterInfo &MRI = MF.getRegInfo();
const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
const SIRegisterInfo *TRI = ST.getRegisterInfo();
bool FuncModified = false;
for (MachineBasicBlock &MBB : MF) {
// Cleanup missed Saddr opportunites from ISel.
FuncModified |= fixupGlobalSaddr(MBB, MF, MRI, ST, TII, TRI);
}
return FuncModified;
}
|