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
|
//===- AMDGPUAliasAnalysis ------------------------------------------------===//
//
// 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
/// This is the AMGPU address space based alias analysis pass.
//===----------------------------------------------------------------------===//
#include "AMDGPUAliasAnalysis.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/Instructions.h"
using namespace llvm;
#define DEBUG_TYPE "amdgpu-aa"
AnalysisKey AMDGPUAA::Key;
// Register this pass...
char AMDGPUAAWrapperPass::ID = 0;
char AMDGPUExternalAAWrapper::ID = 0;
INITIALIZE_PASS(AMDGPUAAWrapperPass, "amdgpu-aa",
"AMDGPU Address space based Alias Analysis", false, true)
INITIALIZE_PASS(AMDGPUExternalAAWrapper, "amdgpu-aa-wrapper",
"AMDGPU Address space based Alias Analysis Wrapper", false, true)
ImmutablePass *llvm::createAMDGPUAAWrapperPass() {
return new AMDGPUAAWrapperPass();
}
ImmutablePass *llvm::createAMDGPUExternalAAWrapperPass() {
return new AMDGPUExternalAAWrapper();
}
void AMDGPUAAWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesAll();
}
static AliasResult getAliasResult(unsigned AS1, unsigned AS2) {
static_assert(AMDGPUAS::MAX_AMDGPU_ADDRESS <= 7, "Addr space out of range");
if (AS1 > AMDGPUAS::MAX_AMDGPU_ADDRESS || AS2 > AMDGPUAS::MAX_AMDGPU_ADDRESS)
return AliasResult::MayAlias;
#define ASMay AliasResult::MayAlias
#define ASNo AliasResult::NoAlias
// This array is indexed by address space value enum elements 0 ... to 7
static const AliasResult ASAliasRules[8][8] = {
/* Flat Global Region Group Constant Private Const32 Buf Fat Ptr */
/* Flat */ {ASMay, ASMay, ASNo, ASMay, ASMay, ASMay, ASMay, ASMay},
/* Global */ {ASMay, ASMay, ASNo, ASNo, ASMay, ASNo, ASMay, ASMay},
/* Region */ {ASNo, ASNo, ASMay, ASNo, ASNo, ASNo, ASNo, ASNo},
/* Group */ {ASMay, ASNo, ASNo, ASMay, ASNo, ASNo, ASNo, ASNo},
/* Constant */ {ASMay, ASMay, ASNo, ASNo, ASNo, ASNo, ASMay, ASMay},
/* Private */ {ASMay, ASNo, ASNo, ASNo, ASNo, ASMay, ASNo, ASNo},
/* Constant 32-bit */ {ASMay, ASMay, ASNo, ASNo, ASMay, ASNo, ASNo, ASMay},
/* Buffer Fat Ptr */ {ASMay, ASMay, ASNo, ASNo, ASMay, ASNo, ASMay, ASMay}
};
#undef ASMay
#undef ASNo
return ASAliasRules[AS1][AS2];
}
AliasResult AMDGPUAAResult::alias(const MemoryLocation &LocA,
const MemoryLocation &LocB,
AAQueryInfo &AAQI) {
unsigned asA = LocA.Ptr->getType()->getPointerAddressSpace();
unsigned asB = LocB.Ptr->getType()->getPointerAddressSpace();
AliasResult Result = getAliasResult(asA, asB);
if (Result == AliasResult::NoAlias)
return Result;
// In general, FLAT (generic) pointers could be aliased to LOCAL or PRIVATE
// pointers. However, as LOCAL or PRIVATE pointers point to local objects, in
// certain cases, it's still viable to check whether a FLAT pointer won't
// alias to a LOCAL or PRIVATE pointer.
MemoryLocation A = LocA;
MemoryLocation B = LocB;
// Canonicalize the location order to simplify the following alias check.
if (asA != AMDGPUAS::FLAT_ADDRESS) {
std::swap(asA, asB);
std::swap(A, B);
}
if (asA == AMDGPUAS::FLAT_ADDRESS &&
(asB == AMDGPUAS::LOCAL_ADDRESS || asB == AMDGPUAS::PRIVATE_ADDRESS)) {
const auto *ObjA =
getUnderlyingObject(A.Ptr->stripPointerCastsForAliasAnalysis());
if (const LoadInst *LI = dyn_cast<LoadInst>(ObjA)) {
// If a generic pointer is loaded from the constant address space, it
// could only be a GLOBAL or CONSTANT one as that address space is soley
// prepared on the host side, where only GLOBAL or CONSTANT variables are
// visible. Note that this even holds for regular functions.
if (LI->getPointerAddressSpace() == AMDGPUAS::CONSTANT_ADDRESS)
return AliasResult::NoAlias;
} else if (const Argument *Arg = dyn_cast<Argument>(ObjA)) {
const Function *F = Arg->getParent();
switch (F->getCallingConv()) {
case CallingConv::AMDGPU_KERNEL:
// In the kernel function, kernel arguments won't alias to (local)
// variables in shared or private address space.
return AliasResult::NoAlias;
default:
// TODO: In the regular function, if that local variable in the
// location B is not captured, that argument pointer won't alias to it
// as well.
break;
}
}
}
// Forward the query to the next alias analysis.
return AAResultBase::alias(LocA, LocB, AAQI);
}
bool AMDGPUAAResult::pointsToConstantMemory(const MemoryLocation &Loc,
AAQueryInfo &AAQI, bool OrLocal) {
unsigned AS = Loc.Ptr->getType()->getPointerAddressSpace();
if (AS == AMDGPUAS::CONSTANT_ADDRESS ||
AS == AMDGPUAS::CONSTANT_ADDRESS_32BIT)
return true;
const Value *Base = getUnderlyingObject(Loc.Ptr);
AS = Base->getType()->getPointerAddressSpace();
if (AS == AMDGPUAS::CONSTANT_ADDRESS ||
AS == AMDGPUAS::CONSTANT_ADDRESS_32BIT)
return true;
if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(Base)) {
if (GV->isConstant())
return true;
} else if (const Argument *Arg = dyn_cast<Argument>(Base)) {
const Function *F = Arg->getParent();
// Only assume constant memory for arguments on kernels.
switch (F->getCallingConv()) {
default:
return AAResultBase::pointsToConstantMemory(Loc, AAQI, OrLocal);
case CallingConv::AMDGPU_LS:
case CallingConv::AMDGPU_HS:
case CallingConv::AMDGPU_ES:
case CallingConv::AMDGPU_GS:
case CallingConv::AMDGPU_VS:
case CallingConv::AMDGPU_PS:
case CallingConv::AMDGPU_CS:
case CallingConv::AMDGPU_KERNEL:
case CallingConv::SPIR_KERNEL:
break;
}
unsigned ArgNo = Arg->getArgNo();
/* On an argument, ReadOnly attribute indicates that the function does
not write through this pointer argument, even though it may write
to the memory that the pointer points to.
On an argument, ReadNone attribute indicates that the function does
not dereference that pointer argument, even though it may read or write
the memory that the pointer points to if accessed through other pointers.
*/
if (F->hasParamAttribute(ArgNo, Attribute::NoAlias) &&
(F->hasParamAttribute(ArgNo, Attribute::ReadNone) ||
F->hasParamAttribute(ArgNo, Attribute::ReadOnly))) {
return true;
}
}
return AAResultBase::pointsToConstantMemory(Loc, AAQI, OrLocal);
}
|
