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
|
//===-- R600ISelLowering.cpp - R600 DAG Lowering Implementation -----------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// Most of the DAG lowering is handled in AMDILISelLowering.cpp. This file
// is mostly EmitInstrWithCustomInserter().
//
//===----------------------------------------------------------------------===//
#include "R600ISelLowering.h"
#include "R600InstrInfo.h"
#include "R600MachineFunctionInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
using namespace llvm;
R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
AMDGPUTargetLowering(TM),
TII(static_cast<const R600InstrInfo*>(TM.getInstrInfo()))
{
setOperationAction(ISD::MUL, MVT::i64, Expand);
// setSchedulingPreference(Sched::VLIW);
addRegisterClass(MVT::v4f32, &AMDIL::R600_Reg128RegClass);
addRegisterClass(MVT::f32, &AMDIL::R600_Reg32RegClass);
addRegisterClass(MVT::v4i32, &AMDIL::R600_Reg128RegClass);
addRegisterClass(MVT::i32, &AMDIL::R600_Reg32RegClass);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4f32, Legal);
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4f32, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4i32, Legal);
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4i32, Legal);
}
MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
MachineInstr * MI, MachineBasicBlock * BB) const
{
MachineFunction * MF = BB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MachineBasicBlock::iterator I = *MI;
switch (MI->getOpcode()) {
default: return AMDGPUTargetLowering::EmitInstrWithCustomInserter(MI, BB);
case AMDIL::TGID_X:
addLiveIn(MI, MF, MRI, TII, AMDIL::T1_X);
break;
case AMDIL::TGID_Y:
addLiveIn(MI, MF, MRI, TII, AMDIL::T1_Y);
break;
case AMDIL::TGID_Z:
addLiveIn(MI, MF, MRI, TII, AMDIL::T1_Z);
break;
case AMDIL::TIDIG_X:
addLiveIn(MI, MF, MRI, TII, AMDIL::T0_X);
break;
case AMDIL::TIDIG_Y:
addLiveIn(MI, MF, MRI, TII, AMDIL::T0_Y);
break;
case AMDIL::TIDIG_Z:
addLiveIn(MI, MF, MRI, TII, AMDIL::T0_Z);
break;
case AMDIL::NGROUPS_X:
lowerImplicitParameter(MI, *BB, MRI, 0);
break;
case AMDIL::NGROUPS_Y:
lowerImplicitParameter(MI, *BB, MRI, 1);
break;
case AMDIL::NGROUPS_Z:
lowerImplicitParameter(MI, *BB, MRI, 2);
break;
case AMDIL::GLOBAL_SIZE_X:
lowerImplicitParameter(MI, *BB, MRI, 3);
break;
case AMDIL::GLOBAL_SIZE_Y:
lowerImplicitParameter(MI, *BB, MRI, 4);
break;
case AMDIL::GLOBAL_SIZE_Z:
lowerImplicitParameter(MI, *BB, MRI, 5);
break;
case AMDIL::LOCAL_SIZE_X:
lowerImplicitParameter(MI, *BB, MRI, 6);
break;
case AMDIL::LOCAL_SIZE_Y:
lowerImplicitParameter(MI, *BB, MRI, 7);
break;
case AMDIL::LOCAL_SIZE_Z:
lowerImplicitParameter(MI, *BB, MRI, 8);
break;
case AMDIL::R600_LOAD_CONST:
{
int64_t RegIndex = MI->getOperand(1).getImm();
unsigned ConstantReg = AMDIL::R600_CReg32RegClass.getRegister(RegIndex);
BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDIL::COPY))
.addOperand(MI->getOperand(0))
.addReg(ConstantReg);
break;
}
case AMDIL::LOAD_INPUT:
{
int64_t RegIndex = MI->getOperand(1).getImm();
addLiveIn(MI, MF, MRI, TII,
AMDIL::R600_TReg32RegClass.getRegister(RegIndex));
break;
}
case AMDIL::STORE_OUTPUT:
{
int64_t OutputIndex = MI->getOperand(1).getImm();
unsigned OutputReg = AMDIL::R600_TReg32RegClass.getRegister(OutputIndex);
BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDIL::COPY), OutputReg)
.addOperand(MI->getOperand(0));
if (!MRI.isLiveOut(OutputReg)) {
MRI.addLiveOut(OutputReg);
}
break;
}
case AMDIL::RESERVE_REG:
{
R600MachineFunctionInfo * MFI = MF->getInfo<R600MachineFunctionInfo>();
int64_t ReservedIndex = MI->getOperand(0).getImm();
unsigned ReservedReg =
AMDIL::R600_TReg32RegClass.getRegister(ReservedIndex);
MFI->ReservedRegs.push_back(ReservedReg);
break;
}
}
MI->eraseFromParent();
return BB;
}
void R600TargetLowering::lowerImplicitParameter(MachineInstr *MI, MachineBasicBlock &BB,
MachineRegisterInfo & MRI, unsigned dword_offset) const
{
MachineBasicBlock::iterator I = *MI;
unsigned offsetReg = MRI.createVirtualRegister(&AMDIL::R600_TReg32_XRegClass);
MRI.setRegClass(MI->getOperand(0).getReg(), &AMDIL::R600_TReg32_XRegClass);
BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDIL::MOV), offsetReg)
.addReg(AMDIL::ALU_LITERAL_X)
.addImm(dword_offset * 4);
BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDIL::VTX_READ_eg))
.addOperand(MI->getOperand(0))
.addReg(offsetReg)
.addImm(0);
}
|
