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
|
//===--------- llvm/unittests/Target/AMDGPU/AMDGPUUnitTests.cpp -----------===//
//
// 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
//
//===----------------------------------------------------------------------===//
#include "AMDGPUUnitTests.h"
#include "AMDGPUTargetMachine.h"
#include "GCNSubtarget.h"
#include "llvm/MC/TargetRegistry.h"
#include "llvm/Support/TargetSelect.h"
#include "llvm/TargetParser/TargetParser.h"
#include "gtest/gtest.h"
#include "AMDGPUGenSubtargetInfo.inc"
using namespace llvm;
std::once_flag flag;
void InitializeAMDGPUTarget() {
std::call_once(flag, []() {
LLVMInitializeAMDGPUTargetInfo();
LLVMInitializeAMDGPUTarget();
LLVMInitializeAMDGPUTargetMC();
});
}
std::unique_ptr<const GCNTargetMachine>
llvm::createAMDGPUTargetMachine(std::string TStr, StringRef CPU, StringRef FS) {
InitializeAMDGPUTarget();
std::string Error;
const Target *T = TargetRegistry::lookupTarget(TStr, Error);
if (!T)
return nullptr;
TargetOptions Options;
return std::unique_ptr<GCNTargetMachine>(
static_cast<GCNTargetMachine *>(T->createTargetMachine(
TStr, CPU, FS, Options, std::nullopt, std::nullopt)));
}
static cl::opt<bool> PrintCpuRegLimits(
"print-cpu-reg-limits", cl::NotHidden, cl::init(false),
cl::desc("force printing per AMDGPU CPU register limits"));
static bool checkMinMax(std::stringstream &OS, unsigned Occ, unsigned MinOcc,
unsigned MaxOcc,
std::function<unsigned(unsigned)> GetOcc,
std::function<unsigned(unsigned)> GetMinGPRs,
std::function<unsigned(unsigned)> GetMaxGPRs) {
bool MinValid = true, MaxValid = true, RangeValid = true;
unsigned MinGPRs = GetMinGPRs(Occ);
unsigned MaxGPRs = GetMaxGPRs(Occ);
unsigned RealOcc;
if (MinGPRs >= MaxGPRs)
RangeValid = false;
else {
RealOcc = GetOcc(MinGPRs);
for (unsigned NumRegs = MinGPRs + 1; NumRegs <= MaxGPRs; ++NumRegs) {
if (RealOcc != GetOcc(NumRegs)) {
RangeValid = false;
break;
}
}
}
if (RangeValid && RealOcc > MinOcc && RealOcc <= MaxOcc) {
if (MinGPRs > 0 && GetOcc(MinGPRs - 1) <= RealOcc)
MinValid = false;
if (GetOcc(MaxGPRs + 1) >= RealOcc)
MaxValid = false;
}
std::stringstream MinStr;
MinStr << (MinValid ? ' ' : '<') << ' ' << std::setw(3) << MinGPRs << " (O"
<< GetOcc(MinGPRs) << ") " << (RangeValid ? ' ' : 'R');
OS << std::left << std::setw(15) << MinStr.str() << std::setw(3) << MaxGPRs
<< " (O" << GetOcc(MaxGPRs) << ')' << (MaxValid ? "" : " >");
return MinValid && MaxValid && RangeValid;
}
static const std::pair<StringRef, StringRef>
EmptyFS = {"", ""},
W32FS = {"+wavefrontsize32", "w32"},
W64FS = {"+wavefrontsize64", "w64"};
using TestFuncTy =
function_ref<bool(std::stringstream &, unsigned, const GCNSubtarget &)>;
static bool testAndRecord(std::stringstream &Table, const GCNSubtarget &ST,
TestFuncTy test) {
bool Success = true;
unsigned MaxOcc = ST.getMaxWavesPerEU();
for (unsigned Occ = MaxOcc; Occ > 0; --Occ) {
Table << std::right << std::setw(3) << Occ << " ";
Success = test(Table, Occ, ST) && Success;
Table << '\n';
}
return Success;
}
static void testGPRLimits(const char *RegName, bool TestW32W64,
TestFuncTy test) {
SmallVector<StringRef> CPUs;
AMDGPU::fillValidArchListAMDGCN(CPUs);
std::map<std::string, SmallVector<std::string>> TablePerCPUs;
for (auto CPUName : CPUs) {
auto CanonCPUName =
AMDGPU::getArchNameAMDGCN(AMDGPU::parseArchAMDGCN(CPUName));
auto *FS = &EmptyFS;
while (true) {
auto TM = createAMDGPUTargetMachine("amdgcn-amd-", CPUName, FS->first);
if (!TM)
break;
GCNSubtarget ST(TM->getTargetTriple(), std::string(TM->getTargetCPU()),
std::string(TM->getTargetFeatureString()), *TM);
if (TestW32W64 &&
ST.getFeatureBits().test(AMDGPU::FeatureWavefrontSize32))
FS = &W32FS;
std::stringstream Table;
bool Success = testAndRecord(Table, ST, test);
if (!Success || PrintCpuRegLimits)
TablePerCPUs[Table.str()].push_back((CanonCPUName + FS->second).str());
if (FS != &W32FS)
break;
FS = &W64FS;
}
}
std::stringstream OS;
for (auto &P : TablePerCPUs) {
for (auto &CPUName : P.second)
OS << ' ' << CPUName;
OS << ":\nOcc Min" << RegName << " Max" << RegName << '\n'
<< P.first << '\n';
}
auto ErrStr = OS.str();
EXPECT_TRUE(ErrStr.empty()) << ErrStr;
}
TEST(AMDGPU, TestVGPRLimitsPerOccupancy) {
auto test = [](std::stringstream &OS, unsigned Occ, const GCNSubtarget &ST) {
unsigned MaxVGPRNum = ST.getAddressableNumVGPRs();
return checkMinMax(
OS, Occ, ST.getOccupancyWithNumVGPRs(MaxVGPRNum), ST.getMaxWavesPerEU(),
[&](unsigned NumGPRs) { return ST.getOccupancyWithNumVGPRs(NumGPRs); },
[&](unsigned Occ) { return ST.getMinNumVGPRs(Occ); },
[&](unsigned Occ) { return ST.getMaxNumVGPRs(Occ); });
};
testGPRLimits("VGPR", true, test);
}
|
