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
|
//===-- R600LowerShaderInstructions.cpp - TODO: Add brief description -------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// TODO: Add full description
//
//===----------------------------------------------------------------------===//
#include "AMDGPU.h"
#include "AMDGPULowerShaderInstructions.h"
#include "AMDGPUUtil.h"
#include "AMDIL.h"
#include "AMDILInstrInfo.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
using namespace llvm;
namespace {
class R600LowerShaderInstructionsPass : public MachineFunctionPass,
public AMDGPULowerShaderInstructionsPass {
private:
static char ID;
TargetMachine &TM;
void lowerEXPORT_REG_FAKE(MachineInstr &MI, MachineBasicBlock &MBB,
MachineBasicBlock::iterator I);
void lowerLOAD_INPUT(MachineInstr & MI);
bool lowerSTORE_OUTPUT(MachineInstr & MI, MachineBasicBlock &MBB,
MachineBasicBlock::iterator I);
public:
R600LowerShaderInstructionsPass(TargetMachine &tm) :
MachineFunctionPass(ID), TM(tm) { }
bool runOnMachineFunction(MachineFunction &MF);
const char *getPassName() const { return "R600 Lower Shader Instructions"; }
};
} /* End anonymous namespace */
char R600LowerShaderInstructionsPass::ID = 0;
FunctionPass *llvm::createR600LowerShaderInstructionsPass(TargetMachine &tm) {
return new R600LowerShaderInstructionsPass(tm);
}
#define INSTR_CASE_FLOAT_V(inst) \
case AMDIL:: inst##_v4f32: \
#define INSTR_CASE_FLOAT_S(inst) \
case AMDIL:: inst##_f32:
#define INSTR_CASE_FLOAT(inst) \
INSTR_CASE_FLOAT_V(inst) \
INSTR_CASE_FLOAT_S(inst)
bool R600LowerShaderInstructionsPass::runOnMachineFunction(MachineFunction &MF)
{
MRI = &MF.getRegInfo();
for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
BB != BB_E; ++BB) {
MachineBasicBlock &MBB = *BB;
for (MachineBasicBlock::iterator I = MBB.begin(); I != MBB.end();) {
MachineInstr &MI = *I;
bool deleteInstr = false;
switch (MI.getOpcode()) {
default: break;
case AMDIL::RESERVE_REG:
case AMDIL::EXPORT_REG:
deleteInstr = true;
break;
case AMDIL::LOAD_INPUT:
lowerLOAD_INPUT(MI);
deleteInstr = true;
break;
case AMDIL::STORE_OUTPUT:
deleteInstr = lowerSTORE_OUTPUT(MI, MBB, I);
break;
}
++I;
if (deleteInstr) {
MI.eraseFromParent();
}
}
}
return false;
}
/* The goal of this function is to replace the virutal destination register of
* a LOAD_INPUT instruction with the correct physical register that will.
*
* XXX: I don't think this is the right way things assign physical registers,
* but I'm not sure of another way to do this.
*/
void R600LowerShaderInstructionsPass::lowerLOAD_INPUT(MachineInstr &MI)
{
MachineOperand &dst = MI.getOperand(0);
MachineOperand &arg = MI.getOperand(1);
int64_t inputIndex = arg.getImm();
const TargetRegisterClass * inputClass = TM.getRegisterInfo()->getRegClass(AMDIL::R600_TReg32RegClassID);
unsigned newRegister = inputClass->getRegister(inputIndex);
unsigned dstReg = dst.getReg();
AMDGPU::utilAddLiveIn(MI.getParent()->getParent(), *MRI, TM.getInstrInfo(), newRegister, dstReg);
}
bool R600LowerShaderInstructionsPass::lowerSTORE_OUTPUT(MachineInstr &MI,
MachineBasicBlock &MBB, MachineBasicBlock::iterator I)
{
MachineOperand &valueOp = MI.getOperand(1);
MachineOperand &indexOp = MI.getOperand(2);
unsigned valueReg = valueOp.getReg();
int64_t outputIndex = indexOp.getImm();
const TargetRegisterClass * outputClass = TM.getRegisterInfo()->getRegClass(AMDIL::R600_TReg32RegClassID);
unsigned newRegister = outputClass->getRegister(outputIndex);
BuildMI(MBB, I, MBB.findDebugLoc(I), TM.getInstrInfo()->get(AMDIL::COPY),
newRegister)
.addReg(valueReg);
if (!MRI->isLiveOut(newRegister))
MRI->addLiveOut(newRegister);
return true;
}
|
