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
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
|
#include "clang/Basic/Cuda.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/Twine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/VersionTuple.h"
namespace clang {
struct CudaVersionMapEntry {
const char *Name;
CudaVersion Version;
llvm::VersionTuple TVersion;
};
#define CUDA_ENTRY(major, minor) \
{ \
#major "." #minor, CudaVersion::CUDA_##major##minor, \
llvm::VersionTuple(major, minor) \
}
static const CudaVersionMapEntry CudaNameVersionMap[] = {
CUDA_ENTRY(7, 0),
CUDA_ENTRY(7, 5),
CUDA_ENTRY(8, 0),
CUDA_ENTRY(9, 0),
CUDA_ENTRY(9, 1),
CUDA_ENTRY(9, 2),
CUDA_ENTRY(10, 0),
CUDA_ENTRY(10, 1),
CUDA_ENTRY(10, 2),
CUDA_ENTRY(11, 0),
CUDA_ENTRY(11, 1),
CUDA_ENTRY(11, 2),
CUDA_ENTRY(11, 3),
CUDA_ENTRY(11, 4),
CUDA_ENTRY(11, 5),
CUDA_ENTRY(11, 6),
CUDA_ENTRY(11, 7),
CUDA_ENTRY(11, 8),
CUDA_ENTRY(12, 0),
CUDA_ENTRY(12, 1),
CUDA_ENTRY(12, 2),
CUDA_ENTRY(12, 3),
CUDA_ENTRY(12, 4),
CUDA_ENTRY(12, 5),
{"", CudaVersion::NEW, llvm::VersionTuple(std::numeric_limits<int>::max())},
{"unknown", CudaVersion::UNKNOWN, {}} // End of list tombstone.
};
#undef CUDA_ENTRY
const char *CudaVersionToString(CudaVersion V) {
for (auto *I = CudaNameVersionMap; I->Version != CudaVersion::UNKNOWN; ++I)
if (I->Version == V)
return I->Name;
return CudaVersionToString(CudaVersion::UNKNOWN);
}
CudaVersion CudaStringToVersion(const llvm::Twine &S) {
std::string VS = S.str();
for (auto *I = CudaNameVersionMap; I->Version != CudaVersion::UNKNOWN; ++I)
if (I->Name == VS)
return I->Version;
return CudaVersion::UNKNOWN;
}
CudaVersion ToCudaVersion(llvm::VersionTuple Version) {
for (auto *I = CudaNameVersionMap; I->Version != CudaVersion::UNKNOWN; ++I)
if (I->TVersion == Version)
return I->Version;
return CudaVersion::UNKNOWN;
}
namespace {
struct OffloadArchToStringMap {
OffloadArch arch;
const char *arch_name;
const char *virtual_arch_name;
};
} // namespace
#define SM2(sm, ca) {OffloadArch::SM_##sm, "sm_" #sm, ca}
#define SM(sm) SM2(sm, "compute_" #sm)
#define GFX(gpu) {OffloadArch::GFX##gpu, "gfx" #gpu, "compute_amdgcn"}
static const OffloadArchToStringMap arch_names[] = {
// clang-format off
{OffloadArch::UNUSED, "", ""},
SM2(20, "compute_20"), SM2(21, "compute_20"), // Fermi
SM(30), {OffloadArch::SM_32_, "sm_32", "compute_32"}, SM(35), SM(37), // Kepler
SM(50), SM(52), SM(53), // Maxwell
SM(60), SM(61), SM(62), // Pascal
SM(70), SM(72), // Volta
SM(75), // Turing
SM(80), SM(86), // Ampere
SM(87), // Jetson/Drive AGX Orin
SM(89), // Ada Lovelace
SM(90), // Hopper
SM(90a), // Hopper
GFX(600), // gfx600
GFX(601), // gfx601
GFX(602), // gfx602
GFX(700), // gfx700
GFX(701), // gfx701
GFX(702), // gfx702
GFX(703), // gfx703
GFX(704), // gfx704
GFX(705), // gfx705
GFX(801), // gfx801
GFX(802), // gfx802
GFX(803), // gfx803
GFX(805), // gfx805
GFX(810), // gfx810
{OffloadArch::GFX9_GENERIC, "gfx9-generic", "compute_amdgcn"},
GFX(900), // gfx900
GFX(902), // gfx902
GFX(904), // gfx903
GFX(906), // gfx906
GFX(908), // gfx908
GFX(909), // gfx909
GFX(90a), // gfx90a
GFX(90c), // gfx90c
GFX(940), // gfx940
GFX(941), // gfx941
GFX(942), // gfx942
{OffloadArch::GFX10_1_GENERIC, "gfx10-1-generic", "compute_amdgcn"},
GFX(1010), // gfx1010
GFX(1011), // gfx1011
GFX(1012), // gfx1012
GFX(1013), // gfx1013
{OffloadArch::GFX10_3_GENERIC, "gfx10-3-generic", "compute_amdgcn"},
GFX(1030), // gfx1030
GFX(1031), // gfx1031
GFX(1032), // gfx1032
GFX(1033), // gfx1033
GFX(1034), // gfx1034
GFX(1035), // gfx1035
GFX(1036), // gfx1036
{OffloadArch::GFX11_GENERIC, "gfx11-generic", "compute_amdgcn"},
GFX(1100), // gfx1100
GFX(1101), // gfx1101
GFX(1102), // gfx1102
GFX(1103), // gfx1103
GFX(1150), // gfx1150
GFX(1151), // gfx1151
GFX(1152), // gfx1152
{OffloadArch::GFX12_GENERIC, "gfx12-generic", "compute_amdgcn"},
GFX(1200), // gfx1200
GFX(1201), // gfx1201
{OffloadArch::AMDGCNSPIRV, "amdgcnspirv", "compute_amdgcn"},
{OffloadArch::Generic, "generic", ""},
// clang-format on
};
#undef SM
#undef SM2
#undef GFX
const char *OffloadArchToString(OffloadArch A) {
auto result = std::find_if(
std::begin(arch_names), std::end(arch_names),
[A](const OffloadArchToStringMap &map) { return A == map.arch; });
if (result == std::end(arch_names))
return "unknown";
return result->arch_name;
}
const char *OffloadArchToVirtualArchString(OffloadArch A) {
auto result = std::find_if(
std::begin(arch_names), std::end(arch_names),
[A](const OffloadArchToStringMap &map) { return A == map.arch; });
if (result == std::end(arch_names))
return "unknown";
return result->virtual_arch_name;
}
OffloadArch StringToOffloadArch(llvm::StringRef S) {
auto result = std::find_if(
std::begin(arch_names), std::end(arch_names),
[S](const OffloadArchToStringMap &map) { return S == map.arch_name; });
if (result == std::end(arch_names))
return OffloadArch::UNKNOWN;
return result->arch;
}
CudaVersion MinVersionForOffloadArch(OffloadArch A) {
if (A == OffloadArch::UNKNOWN)
return CudaVersion::UNKNOWN;
// AMD GPUs do not depend on CUDA versions.
if (IsAMDOffloadArch(A))
return CudaVersion::CUDA_70;
switch (A) {
case OffloadArch::SM_20:
case OffloadArch::SM_21:
case OffloadArch::SM_30:
case OffloadArch::SM_32_:
case OffloadArch::SM_35:
case OffloadArch::SM_37:
case OffloadArch::SM_50:
case OffloadArch::SM_52:
case OffloadArch::SM_53:
return CudaVersion::CUDA_70;
case OffloadArch::SM_60:
case OffloadArch::SM_61:
case OffloadArch::SM_62:
return CudaVersion::CUDA_80;
case OffloadArch::SM_70:
return CudaVersion::CUDA_90;
case OffloadArch::SM_72:
return CudaVersion::CUDA_91;
case OffloadArch::SM_75:
return CudaVersion::CUDA_100;
case OffloadArch::SM_80:
return CudaVersion::CUDA_110;
case OffloadArch::SM_86:
return CudaVersion::CUDA_111;
case OffloadArch::SM_87:
return CudaVersion::CUDA_114;
case OffloadArch::SM_89:
case OffloadArch::SM_90:
return CudaVersion::CUDA_118;
case OffloadArch::SM_90a:
return CudaVersion::CUDA_120;
default:
llvm_unreachable("invalid enum");
}
}
CudaVersion MaxVersionForOffloadArch(OffloadArch A) {
// AMD GPUs do not depend on CUDA versions.
if (IsAMDOffloadArch(A))
return CudaVersion::NEW;
switch (A) {
case OffloadArch::UNKNOWN:
return CudaVersion::UNKNOWN;
case OffloadArch::SM_20:
case OffloadArch::SM_21:
return CudaVersion::CUDA_80;
case OffloadArch::SM_30:
case OffloadArch::SM_32_:
return CudaVersion::CUDA_102;
case OffloadArch::SM_35:
case OffloadArch::SM_37:
return CudaVersion::CUDA_118;
default:
return CudaVersion::NEW;
}
}
bool CudaFeatureEnabled(llvm::VersionTuple Version, CudaFeature Feature) {
return CudaFeatureEnabled(ToCudaVersion(Version), Feature);
}
bool CudaFeatureEnabled(CudaVersion Version, CudaFeature Feature) {
switch (Feature) {
case CudaFeature::CUDA_USES_NEW_LAUNCH:
return Version >= CudaVersion::CUDA_92;
case CudaFeature::CUDA_USES_FATBIN_REGISTER_END:
return Version >= CudaVersion::CUDA_101;
}
llvm_unreachable("Unknown CUDA feature.");
}
} // namespace clang
|
