Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
14 | pmbaty | 1 | //===--- Cuda.h - Utilities for compiling CUDA code ------------*- C++ -*-===// |
2 | // |
||
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
||
4 | // See https://llvm.org/LICENSE.txt for license information. |
||
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
||
6 | // |
||
7 | //===----------------------------------------------------------------------===// |
||
8 | |||
9 | #ifndef LLVM_CLANG_BASIC_CUDA_H |
||
10 | #define LLVM_CLANG_BASIC_CUDA_H |
||
11 | |||
12 | namespace llvm { |
||
13 | class StringRef; |
||
14 | class Twine; |
||
15 | class VersionTuple; |
||
16 | } // namespace llvm |
||
17 | |||
18 | namespace clang { |
||
19 | |||
20 | enum class CudaVersion { |
||
21 | UNKNOWN, |
||
22 | CUDA_70, |
||
23 | CUDA_75, |
||
24 | CUDA_80, |
||
25 | CUDA_90, |
||
26 | CUDA_91, |
||
27 | CUDA_92, |
||
28 | CUDA_100, |
||
29 | CUDA_101, |
||
30 | CUDA_102, |
||
31 | CUDA_110, |
||
32 | CUDA_111, |
||
33 | CUDA_112, |
||
34 | CUDA_113, |
||
35 | CUDA_114, |
||
36 | CUDA_115, |
||
37 | CUDA_116, |
||
38 | CUDA_117, |
||
39 | CUDA_118, |
||
40 | FULLY_SUPPORTED = CUDA_115, |
||
41 | PARTIALLY_SUPPORTED = |
||
42 | CUDA_118, // Partially supported. Proceed with a warning. |
||
43 | NEW = 10000, // Too new. Issue a warning, but allow using it. |
||
44 | }; |
||
45 | const char *CudaVersionToString(CudaVersion V); |
||
46 | // Input is "Major.Minor" |
||
47 | CudaVersion CudaStringToVersion(const llvm::Twine &S); |
||
48 | |||
49 | enum class CudaArch { |
||
50 | UNUSED, |
||
51 | UNKNOWN, |
||
52 | SM_20, |
||
53 | SM_21, |
||
54 | SM_30, |
||
55 | SM_32, |
||
56 | SM_35, |
||
57 | SM_37, |
||
58 | SM_50, |
||
59 | SM_52, |
||
60 | SM_53, |
||
61 | SM_60, |
||
62 | SM_61, |
||
63 | SM_62, |
||
64 | SM_70, |
||
65 | SM_72, |
||
66 | SM_75, |
||
67 | SM_80, |
||
68 | SM_86, |
||
69 | SM_87, |
||
70 | SM_89, |
||
71 | SM_90, |
||
72 | GFX600, |
||
73 | GFX601, |
||
74 | GFX602, |
||
75 | GFX700, |
||
76 | GFX701, |
||
77 | GFX702, |
||
78 | GFX703, |
||
79 | GFX704, |
||
80 | GFX705, |
||
81 | GFX801, |
||
82 | GFX802, |
||
83 | GFX803, |
||
84 | GFX805, |
||
85 | GFX810, |
||
86 | GFX900, |
||
87 | GFX902, |
||
88 | GFX904, |
||
89 | GFX906, |
||
90 | GFX908, |
||
91 | GFX909, |
||
92 | GFX90a, |
||
93 | GFX90c, |
||
94 | GFX940, |
||
95 | GFX1010, |
||
96 | GFX1011, |
||
97 | GFX1012, |
||
98 | GFX1013, |
||
99 | GFX1030, |
||
100 | GFX1031, |
||
101 | GFX1032, |
||
102 | GFX1033, |
||
103 | GFX1034, |
||
104 | GFX1035, |
||
105 | GFX1036, |
||
106 | GFX1100, |
||
107 | GFX1101, |
||
108 | GFX1102, |
||
109 | GFX1103, |
||
110 | Generic, // A processor model named 'generic' if the target backend defines a |
||
111 | // public one. |
||
112 | LAST, |
||
113 | |||
114 | CudaDefault = CudaArch::SM_35, |
||
115 | HIPDefault = CudaArch::GFX803, |
||
116 | }; |
||
117 | |||
118 | static inline bool IsNVIDIAGpuArch(CudaArch A) { |
||
119 | return A >= CudaArch::SM_20 && A < CudaArch::GFX600; |
||
120 | } |
||
121 | |||
122 | static inline bool IsAMDGpuArch(CudaArch A) { |
||
123 | // Generic processor model is for testing only. |
||
124 | return A >= CudaArch::GFX600 && A < CudaArch::Generic; |
||
125 | } |
||
126 | |||
127 | const char *CudaArchToString(CudaArch A); |
||
128 | const char *CudaArchToVirtualArchString(CudaArch A); |
||
129 | |||
130 | // The input should have the form "sm_20". |
||
131 | CudaArch StringToCudaArch(llvm::StringRef S); |
||
132 | |||
133 | /// Get the earliest CudaVersion that supports the given CudaArch. |
||
134 | CudaVersion MinVersionForCudaArch(CudaArch A); |
||
135 | |||
136 | /// Get the latest CudaVersion that supports the given CudaArch. |
||
137 | CudaVersion MaxVersionForCudaArch(CudaArch A); |
||
138 | |||
139 | // Various SDK-dependent features that affect CUDA compilation |
||
140 | enum class CudaFeature { |
||
141 | // CUDA-9.2+ uses a new API for launching kernels. |
||
142 | CUDA_USES_NEW_LAUNCH, |
||
143 | // CUDA-10.1+ needs explicit end of GPU binary registration. |
||
144 | CUDA_USES_FATBIN_REGISTER_END, |
||
145 | }; |
||
146 | |||
147 | CudaVersion ToCudaVersion(llvm::VersionTuple); |
||
148 | bool CudaFeatureEnabled(llvm::VersionTuple, CudaFeature); |
||
149 | bool CudaFeatureEnabled(CudaVersion, CudaFeature); |
||
150 | |||
151 | } // namespace clang |
||
152 | |||
153 | #endif |