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
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2017-2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
#include "Compiler/HandleLoadStoreInstructions.hpp"
#include "Compiler/CISACodeGen/helper.h"
#include "Compiler/CodeGenPublic.h"
#include "Compiler/IGCPassSupport.h"
#include "common/LLVMWarningsPush.hpp"
#include "llvmWrapper/IR/DerivedTypes.h"
#include <llvmWrapper/IR/IRBuilder.h>
#include <llvm/IR/Module.h>
#include <llvm/IR/Function.h>
#include <llvm/IR/Instructions.h>
#include "common/LLVMWarningsPop.hpp"
#include "Probe/Assertion.h"
using namespace llvm;
using namespace IGC;
// Register pass to igc-opt
#define PASS_FLAG "igc-dp-to-fp-load-store"
#define PASS_DESCRIPTION "Convert load/store on doubles into store/loads on i32 or float types"
#define PASS_CFG_ONLY false
#define PASS_ANALYSIS false
IGC_INITIALIZE_PASS_BEGIN(HandleLoadStoreInstructions, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
IGC_INITIALIZE_PASS_END(HandleLoadStoreInstructions, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
char HandleLoadStoreInstructions::ID = 0;
HandleLoadStoreInstructions::HandleLoadStoreInstructions() : FunctionPass(ID)
{
initializeHandleLoadStoreInstructionsPass(*PassRegistry::getPassRegistry());
}
bool HandleLoadStoreInstructions::runOnFunction(llvm::Function& F)
{
m_changed = false;
visit(F);
return m_changed;
}
void HandleLoadStoreInstructions::visitLoadInst(llvm::LoadInst& I)
{
IGCLLVM::IRBuilder<> builder(&I);
llvm::Value* ptrv = llvm::cast<llvm::LoadInst>(I).getPointerOperand();
if (I.getType()->isDoubleTy() ||
(I.getType()->isVectorTy() &&
cast<VectorType>(I.getType())->getElementType()->isDoubleTy()))
{
// scalar/vector double instruction
// Found an instruction of type
// %11 = load <2 x double> addrspace(8585216)* %10, align 16
uint32_t numVectorElements = 1;
llvm::Type* doubleDstType = builder.getDoubleTy();
if (I.getType()->isVectorTy())
{
numVectorElements = (uint32_t)cast<IGCLLVM::FixedVectorType>(I.getType())->getNumElements();
doubleDstType = IGCLLVM::FixedVectorType::get(builder.getDoubleTy(), numVectorElements);
}
uint as = ptrv->getType()->getPointerAddressSpace();
BufferType bufType = GetBufferType(as);
llvm::Value* newInst = nullptr;
// WA: driver does not support gather4(untyped surface read) from constant buffer
if (bufType == CONSTANT_BUFFER)
{
llvm::Type* floatyptr = llvm::PointerType::get(builder.getFloatTy(), as);
llvm::Value* byteOffset = nullptr;
if (!isa<ConstantPointerNull>(ptrv))
{
if (ConstantExpr * ptrExpr = dyn_cast<ConstantExpr>(ptrv))
{
IGC_ASSERT(ptrExpr->getOpcode() == Instruction::IntToPtr);
byteOffset = ptrExpr->getOperand(0);
}
else if (IntToPtrInst * i2p = dyn_cast<IntToPtrInst>(ptrv))
{
byteOffset = i2p->getOperand(0);
}
else
{
byteOffset = builder.CreatePtrToInt(ptrv, builder.getInt32Ty());
}
}
Value* vec = UndefValue::get(IGCLLVM::FixedVectorType::get(builder.getFloatTy(), numVectorElements * 2));
for (unsigned int i = 0; i < numVectorElements * 2; ++i)
{
Value* offset = builder.getInt32(i * 4);
if (byteOffset)
{
offset = builder.CreateAdd(byteOffset, offset);
}
// new IntToPtr and new load
// cannot use irbuilder to create IntToPtr. It may create ConstantExpr instead of instruction
Value* i2p = llvm::IntToPtrInst::Create(Instruction::IntToPtr, offset, floatyptr, "splitDouble", &I);
Value* data = builder.CreateLoad(i2p);
vec = builder.CreateInsertElement(vec, data, builder.getInt32(i));
}
newInst = builder.CreateBitCast(vec, doubleDstType);
}
// End of WA
else
{
// double to <floatx2> ; <doublex2> to <floatx4>
llvm::Type* dataType = IGCLLVM::FixedVectorType::get(builder.getFloatTy(), numVectorElements * 2);
llvm::PointerType* ptrType = llvm::PointerType::get(dataType, ptrv->getType()->getPointerAddressSpace());
llvm::Value* newPtrv = builder.CreateBitCast(ptrv, ptrType);
Value* newLoad = builder.CreateLoad(newPtrv);
newInst = builder.CreateBitCast(newLoad, doubleDstType);
}
I.replaceAllUsesWith(newInst);
I.eraseFromParent();
m_changed = true;
}
}
void HandleLoadStoreInstructions::visitStoreInst(llvm::StoreInst& I)
{
IGCLLVM::IRBuilder<> builder(&I);
llvm::Value* ptrv = llvm::cast<llvm::StoreInst>(I).getPointerOperand();
if (I.getValueOperand()->getType()->isDoubleTy() ||
(I.getValueOperand()->getType()->isVectorTy() &&
cast<VectorType>(I.getValueOperand()->getType())->getElementType()->isDoubleTy()))
{
// scalar/vector double instruction
uint32_t numVectorElements = 1;
if (I.getValueOperand()->getType()->isVectorTy())
{
numVectorElements = (uint32_t)cast<IGCLLVM::FixedVectorType>(I.getValueOperand()->getType())->getNumElements();
}
// %9 = bitcast double addrspace(8519681)* %8 to <2 x float> addrspace(8519681)*
llvm::Type* floatDatType = IGCLLVM::FixedVectorType::get(builder.getFloatTy(), numVectorElements * 2);
llvm::PointerType* floatPtrType = llvm::PointerType::get(floatDatType, ptrv->getType()->getPointerAddressSpace());
llvm::Value* newPtrv = builder.CreateBitCast(ptrv, floatPtrType);
I.setOperand(I.getPointerOperandIndex(), newPtrv);
// %10 = bitcast double %4 to <2 x float>
llvm::Value* srcFloatInst = builder.CreateBitCast(I.getValueOperand(), floatDatType);
I.setOperand(0, srcFloatInst);
m_changed = true;
}
}
|
