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
| //===- NVPTXLowerAggrCopies.cpp - ------------------------------*- C++ -*--===//
//
// 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
// Lower aggregate copies, memset, memcpy, memmov intrinsics into loops when
// the size is large or is not a compile-time constant.
//
//===----------------------------------------------------------------------===//
#include "NVPTXLowerAggrCopies.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/CodeGen/StackProtector.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "llvm/Support/Debug.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
#define DEBUG_TYPE "nvptx"
using namespace llvm;
namespace {
// actual analysis class, which is a functionpass
struct NVPTXLowerAggrCopies : public FunctionPass {
static char ID;
NVPTXLowerAggrCopies() : FunctionPass(ID) {}
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addPreserved<StackProtector>();
AU.addRequired<TargetTransformInfoWrapperPass>();
}
bool runOnFunction(Function &F) override;
static const unsigned MaxAggrCopySize = 128;
StringRef getPassName() const override {
return "Lower aggregate copies/intrinsics into loops";
}
};
char NVPTXLowerAggrCopies::ID = 0;
bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
SmallVector<LoadInst *, 4> AggrLoads;
SmallVector<MemIntrinsic *, 4> MemCalls;
const DataLayout &DL = F.getParent()->getDataLayout();
LLVMContext &Context = F.getParent()->getContext();
const TargetTransformInfo &TTI =
getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
// Collect all aggregate loads and mem* calls.
for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE;
++II) {
if (LoadInst *LI = dyn_cast<LoadInst>(II)) {
if (!LI->hasOneUse())
continue;
if (DL.getTypeStoreSize(LI->getType()) < MaxAggrCopySize)
continue;
if (StoreInst *SI = dyn_cast<StoreInst>(LI->user_back())) {
if (SI->getOperand(0) != LI)
continue;
AggrLoads.push_back(LI);
}
} else if (MemIntrinsic *IntrCall = dyn_cast<MemIntrinsic>(II)) {
// Convert intrinsic calls with variable size or with constant size
// larger than the MaxAggrCopySize threshold.
if (ConstantInt *LenCI = dyn_cast<ConstantInt>(IntrCall->getLength())) {
if (LenCI->getZExtValue() >= MaxAggrCopySize) {
MemCalls.push_back(IntrCall);
}
} else {
MemCalls.push_back(IntrCall);
}
}
}
}
if (AggrLoads.size() == 0 && MemCalls.size() == 0) {
return false;
}
//
// Do the transformation of an aggr load/copy/set to a loop
//
for (LoadInst *LI : AggrLoads) {
auto *SI = cast<StoreInst>(*LI->user_begin());
Value *SrcAddr = LI->getOperand(0);
Value *DstAddr = SI->getOperand(1);
unsigned NumLoads = DL.getTypeStoreSize(LI->getType());
ConstantInt *CopyLen =
ConstantInt::get(Type::getInt32Ty(Context), NumLoads);
createMemCpyLoopKnownSize(/* ConvertedInst */ SI,
/* SrcAddr */ SrcAddr, /* DstAddr */ DstAddr,
/* CopyLen */ CopyLen,
/* SrcAlign */ LI->getAlignment(),
/* DestAlign */ SI->getAlignment(),
/* SrcIsVolatile */ LI->isVolatile(),
/* DstIsVolatile */ SI->isVolatile(), TTI);
SI->eraseFromParent();
LI->eraseFromParent();
}
// Transform mem* intrinsic calls.
for (MemIntrinsic *MemCall : MemCalls) {
if (MemCpyInst *Memcpy = dyn_cast<MemCpyInst>(MemCall)) {
expandMemCpyAsLoop(Memcpy, TTI);
} else if (MemMoveInst *Memmove = dyn_cast<MemMoveInst>(MemCall)) {
expandMemMoveAsLoop(Memmove);
} else if (MemSetInst *Memset = dyn_cast<MemSetInst>(MemCall)) {
expandMemSetAsLoop(Memset);
}
MemCall->eraseFromParent();
}
return true;
}
} // namespace
namespace llvm {
void initializeNVPTXLowerAggrCopiesPass(PassRegistry &);
}
INITIALIZE_PASS(NVPTXLowerAggrCopies, "nvptx-lower-aggr-copies",
"Lower aggregate copies, and llvm.mem* intrinsics into loops",
false, false)
FunctionPass *llvm::createLowerAggrCopies() {
return new NVPTXLowerAggrCopies();
}
|