[flang][cuda] Adapt ExternalNameConversion to work in gpu module (#117039)
[llvm-project.git] / clang / test / CodeGen / attr-cpuspecific.c
blob628892d5809b4ee19912ae7dc215aeadb48982cc
1 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm -o - %s | FileCheck %s --check-prefixes=CHECK,LINUX
2 // RUN: %clang_cc1 -triple x86_64-apple-macos -emit-llvm -o - %s | FileCheck %s --check-prefixes=CHECK,LINUX
3 // RUN: %clang_cc1 -triple x86_64-windows-pc -fms-compatibility -emit-llvm -o - %s | FileCheck %s --check-prefixes=CHECK,WINDOWS
5 #ifdef _WIN64
6 #define ATTR(X) __declspec(X)
7 #else
8 #define ATTR(X) __attribute__((X))
9 #endif // _WIN64
11 // Each version should have an IFunc and an alias.
12 // LINUX: @SingleVersion = weak_odr alias void (), ptr @SingleVersion.ifunc
13 // LINUX: @TwoVersions = weak_odr alias void (), ptr @TwoVersions.ifunc
14 // LINUX: @OrderDispatchUsageSpecific = weak_odr alias void (), ptr @OrderDispatchUsageSpecific.ifunc
15 // LINUX: @TwoVersionsSameAttr = weak_odr alias void (), ptr @TwoVersionsSameAttr.ifunc
16 // LINUX: @ThreeVersionsSameAttr = weak_odr alias void (), ptr @ThreeVersionsSameAttr.ifunc
17 // LINUX: @OrderSpecificUsageDispatch = weak_odr alias void (), ptr @OrderSpecificUsageDispatch.ifunc
18 // LINUX: @NoSpecifics = weak_odr alias void (), ptr @NoSpecifics.ifunc
19 // LINUX: @HasGeneric = weak_odr alias void (), ptr @HasGeneric.ifunc
20 // LINUX: @HasParams = weak_odr alias void (i32, double), ptr @HasParams.ifunc
21 // LINUX: @HasParamsAndReturn = weak_odr alias i32 (i32, double), ptr @HasParamsAndReturn.ifunc
22 // LINUX: @GenericAndPentium = weak_odr alias i32 (i32, double), ptr @GenericAndPentium.ifunc
23 // LINUX: @DispatchFirst = weak_odr alias i32 (), ptr @DispatchFirst.ifunc
25 // LINUX: @SingleVersion.ifunc = weak_odr ifunc void (), ptr @SingleVersion.resolver
26 // LINUX: @TwoVersions.ifunc = weak_odr ifunc void (), ptr @TwoVersions.resolver
27 // LINUX: @OrderDispatchUsageSpecific.ifunc = weak_odr ifunc void (), ptr @OrderDispatchUsageSpecific.resolver
28 // LINUX: @TwoVersionsSameAttr.ifunc = weak_odr ifunc void (), ptr @TwoVersionsSameAttr.resolver
29 // LINUX: @ThreeVersionsSameAttr.ifunc = weak_odr ifunc void (), ptr @ThreeVersionsSameAttr.resolver
30 // LINUX: @OrderSpecificUsageDispatch.ifunc = weak_odr ifunc void (), ptr @OrderSpecificUsageDispatch.resolver
31 // LINUX: @NoSpecifics.ifunc = weak_odr ifunc void (), ptr @NoSpecifics.resolver
32 // LINUX: @HasGeneric.ifunc = weak_odr ifunc void (), ptr @HasGeneric.resolver
33 // LINUX: @HasParams.ifunc = weak_odr ifunc void (i32, double), ptr @HasParams.resolver
34 // LINUX: @HasParamsAndReturn.ifunc = weak_odr ifunc i32 (i32, double), ptr @HasParamsAndReturn.resolver
35 // LINUX: @GenericAndPentium.ifunc = weak_odr ifunc i32 (i32, double), ptr @GenericAndPentium.resolver
36 // LINUX: @DispatchFirst.ifunc = weak_odr ifunc i32 (), ptr @DispatchFirst.resolver
38 ATTR(cpu_specific(ivybridge))
39 void SingleVersion(void){}
40 // LINUX: define{{.*}} void @SingleVersion.S() #[[S:[0-9]+]]
41 // WINDOWS: define dso_local void @SingleVersion.S() #[[S:[0-9]+]]
43 ATTR(cpu_dispatch(ivybridge))
44 void SingleVersion(void);
45 // LINUX: define weak_odr ptr @SingleVersion.resolver()
46 // LINUX: call void @__cpu_indicator_init
47 // LINUX: %[[FEAT_INIT:.+]] = load i32, ptr getelementptr inbounds ({ i32, i32, i32, [1 x i32] }, ptr @__cpu_model, i32 0, i32 3, i32 0), align 4
48 // LINUX: %[[FEAT_JOIN:.+]] = and i32 %[[FEAT_INIT]], 525311
49 // LINUX: %[[FEAT_CHECK:.+]] = icmp eq i32 %[[FEAT_JOIN]], 525311
50 // LINUX: ret ptr @SingleVersion.S
51 // LINUX: call void @llvm.trap
52 // LINUX: unreachable
54 // WINDOWS: define weak_odr dso_local void @SingleVersion() comdat
55 // WINDOWS: call void @__cpu_indicator_init()
56 // WINDOWS: %[[FEAT_INIT:.+]] = load i32, ptr getelementptr inbounds ({ i32, i32, i32, [1 x i32] }, ptr @__cpu_model, i32 0, i32 3, i32 0), align 4
57 // WINDOWS: %[[FEAT_JOIN:.+]] = and i32 %[[FEAT_INIT]], 525311
58 // WINDOWS: %[[FEAT_CHECK:.+]] = icmp eq i32 %[[FEAT_JOIN]], 525311
59 // WINDOWS: call void @SingleVersion.S()
60 // WINDOWS-NEXT: ret void
61 // WINDOWS: call void @llvm.trap
62 // WINDOWS: unreachable
64 ATTR(cpu_specific(ivybridge))
65 void NotCalled(void){}
66 // LINUX: define{{.*}} void @NotCalled.S() #[[S]]
67 // WINDOWS: define dso_local void @NotCalled.S() #[[S:[0-9]+]]
69 // Done before any of the implementations. Also has an undecorated forward
70 // declaration.
71 void TwoVersions(void);
73 ATTR(cpu_dispatch(ivybridge, knl))
74 void TwoVersions(void);
75 // LINUX: define weak_odr ptr @TwoVersions.resolver()
76 // LINUX: call void @__cpu_indicator_init
77 // LINUX: %[[FEAT_INIT:.+]] = load i32, ptr getelementptr inbounds ({ i32, i32, i32, [1 x i32] }, ptr @__cpu_model, i32 0, i32 3, i32 0), align 4
78 // LINUX: %[[FEAT_JOIN:.+]] = and i32 %[[FEAT_INIT]], 9422847
79 // LINUX: %[[FEAT_CHECK:.+]] = icmp eq i32 %[[FEAT_JOIN]], 9422847
80 // LINUX: ret ptr @TwoVersions.Z
81 // LINUX: ret ptr @TwoVersions.S
82 // LINUX: call void @llvm.trap
83 // LINUX: unreachable
85 // WINDOWS: define weak_odr dso_local void @TwoVersions() comdat
86 // WINDOWS: call void @__cpu_indicator_init()
87 // WINDOWS: %[[FEAT_INIT:.+]] = load i32, ptr getelementptr inbounds ({ i32, i32, i32, [1 x i32] }, ptr @__cpu_model, i32 0, i32 3, i32 0), align 4
88 // WINDOWS: %[[FEAT_JOIN:.+]] = and i32 %[[FEAT_INIT]], 9422847
89 // WINDOWS: %[[FEAT_CHECK:.+]] = icmp eq i32 %[[FEAT_JOIN]], 9422847
90 // WINDOWS: call void @TwoVersions.Z()
91 // WINDOWS-NEXT: ret void
92 // WINDOWS: call void @TwoVersions.S()
93 // WINDOWS-NEXT: ret void
94 // WINDOWS: call void @llvm.trap
95 // WINDOWS: unreachable
97 ATTR(cpu_specific(ivybridge))
98 void TwoVersions(void){}
99 // CHECK: define {{.*}}void @TwoVersions.S() #[[S]]
101 ATTR(cpu_specific(knl))
102 void TwoVersions(void){}
103 // CHECK: define {{.*}}void @TwoVersions.Z() #[[K:[0-9]+]]
105 ATTR(cpu_specific(ivybridge, knl))
106 void TwoVersionsSameAttr(void){}
107 // CHECK: define {{.*}}void @TwoVersionsSameAttr.S() #[[S]]
108 // CHECK: define {{.*}}void @TwoVersionsSameAttr.Z() #[[K]]
110 ATTR(cpu_specific(atom, ivybridge, knl))
111 void ThreeVersionsSameAttr(void){}
112 // CHECK: define {{.*}}void @ThreeVersionsSameAttr.O() #[[O:[0-9]+]]
113 // CHECK: define {{.*}}void @ThreeVersionsSameAttr.S() #[[S]]
114 // CHECK: define {{.*}}void @ThreeVersionsSameAttr.Z() #[[K]]
116 ATTR(cpu_specific(knl))
117 void CpuSpecificNoDispatch(void) {}
118 // CHECK: define {{.*}}void @CpuSpecificNoDispatch.Z() #[[K:[0-9]+]]
120 ATTR(cpu_dispatch(knl))
121 void OrderDispatchUsageSpecific(void);
122 // LINUX: define weak_odr ptr @OrderDispatchUsageSpecific.resolver()
123 // LINUX: call void @__cpu_indicator_init
124 // LINUX: ret ptr @OrderDispatchUsageSpecific.Z
125 // LINUX: call void @llvm.trap
126 // LINUX: unreachable
128 // WINDOWS: define weak_odr dso_local void @OrderDispatchUsageSpecific() comdat
129 // WINDOWS: call void @__cpu_indicator_init()
130 // WINDOWS: call void @OrderDispatchUsageSpecific.Z()
131 // WINDOWS-NEXT: ret void
132 // WINDOWS: call void @llvm.trap
133 // WINDOWS: unreachable
135 // CHECK: define {{.*}}void @OrderDispatchUsageSpecific.Z()
137 ATTR(cpu_specific(knl))
138 void OrderSpecificUsageDispatch(void) {}
139 // CHECK: define {{.*}}void @OrderSpecificUsageDispatch.Z() #[[K:[0-9]+]]
141 void usages(void) {
142 SingleVersion();
143 // LINUX: @SingleVersion.ifunc()
144 // WINDOWS: @SingleVersion()
145 TwoVersions();
146 // LINUX: @TwoVersions.ifunc()
147 // WINDOWS: @TwoVersions()
148 TwoVersionsSameAttr();
149 // LINUX: @TwoVersionsSameAttr.ifunc()
150 // WINDOWS: @TwoVersionsSameAttr()
151 ThreeVersionsSameAttr();
152 // LINUX: @ThreeVersionsSameAttr.ifunc()
153 // WINDOWS: @ThreeVersionsSameAttr()
154 CpuSpecificNoDispatch();
155 // LINUX: @CpuSpecificNoDispatch.ifunc()
156 // WINDOWS: @CpuSpecificNoDispatch()
157 OrderDispatchUsageSpecific();
158 // LINUX: @OrderDispatchUsageSpecific.ifunc()
159 // WINDOWS: @OrderDispatchUsageSpecific()
160 OrderSpecificUsageDispatch();
161 // LINUX: @OrderSpecificUsageDispatch.ifunc()
162 // WINDOWS: @OrderSpecificUsageDispatch()
165 // LINUX: declare void @CpuSpecificNoDispatch.ifunc()
167 // has an extra config to emit!
168 ATTR(cpu_dispatch(ivybridge, knl, atom))
169 void TwoVersionsSameAttr(void);
170 // LINUX: define weak_odr ptr @TwoVersionsSameAttr.resolver()
171 // LINUX: ret ptr @TwoVersionsSameAttr.Z
172 // LINUX: ret ptr @TwoVersionsSameAttr.S
173 // LINUX: ret ptr @TwoVersionsSameAttr.O
174 // LINUX: call void @llvm.trap
175 // LINUX: unreachable
177 // WINDOWS: define weak_odr dso_local void @TwoVersionsSameAttr() comdat
178 // WINDOWS: call void @TwoVersionsSameAttr.Z
179 // WINDOWS-NEXT: ret void
180 // WINDOWS: call void @TwoVersionsSameAttr.S
181 // WINDOWS-NEXT: ret void
182 // WINDOWS: call void @TwoVersionsSameAttr.O
183 // WINDOWS-NEXT: ret void
184 // WINDOWS: call void @llvm.trap
185 // WINDOWS: unreachable
187 ATTR(cpu_dispatch(atom, ivybridge, knl))
188 void ThreeVersionsSameAttr(void){}
189 // LINUX: define weak_odr ptr @ThreeVersionsSameAttr.resolver()
190 // LINUX: call void @__cpu_indicator_init
191 // LINUX: ret ptr @ThreeVersionsSameAttr.Z
192 // LINUX: ret ptr @ThreeVersionsSameAttr.S
193 // LINUX: ret ptr @ThreeVersionsSameAttr.O
194 // LINUX: call void @llvm.trap
195 // LINUX: unreachable
197 // WINDOWS: define weak_odr dso_local void @ThreeVersionsSameAttr() comdat
198 // WINDOWS: call void @__cpu_indicator_init
199 // WINDOWS: call void @ThreeVersionsSameAttr.Z
200 // WINDOWS-NEXT: ret void
201 // WINDOWS: call void @ThreeVersionsSameAttr.S
202 // WINDOWS-NEXT: ret void
203 // WINDOWS: call void @ThreeVersionsSameAttr.O
204 // WINDOWS-NEXT: ret void
205 // WINDOWS: call void @llvm.trap
206 // WINDOWS: unreachable
208 ATTR(cpu_dispatch(knl))
209 void OrderSpecificUsageDispatch(void);
210 // LINUX: define weak_odr ptr @OrderSpecificUsageDispatch.resolver()
211 // LINUX: ret ptr @OrderSpecificUsageDispatch.Z
213 // WINDOWS: define weak_odr dso_local void @OrderSpecificUsageDispatch() comdat
214 // WINDOWS: call void @__cpu_indicator_init
215 // WINDOWS: call void @OrderSpecificUsageDispatch.Z
216 // WINDOWS-NEXT: ret void
218 // No Cpu Specific options.
219 ATTR(cpu_dispatch(atom, ivybridge, knl))
220 void NoSpecifics(void);
221 // LINUX: define weak_odr ptr @NoSpecifics.resolver()
222 // LINUX: call void @__cpu_indicator_init
223 // LINUX: ret ptr @NoSpecifics.Z
224 // LINUX: ret ptr @NoSpecifics.S
225 // LINUX: ret ptr @NoSpecifics.O
226 // LINUX: call void @llvm.trap
227 // LINUX: unreachable
229 // WINDOWS: define weak_odr dso_local void @NoSpecifics() comdat
230 // WINDOWS: call void @__cpu_indicator_init
231 // WINDOWS: call void @NoSpecifics.Z
232 // WINDOWS-NEXT: ret void
233 // WINDOWS: call void @NoSpecifics.S
234 // WINDOWS-NEXT: ret void
235 // WINDOWS: call void @NoSpecifics.O
236 // WINDOWS-NEXT: ret void
237 // WINDOWS: call void @llvm.trap
238 // WINDOWS: unreachable
240 ATTR(cpu_dispatch(atom, generic, ivybridge, knl))
241 void HasGeneric(void);
242 // LINUX: define weak_odr ptr @HasGeneric.resolver()
243 // LINUX: call void @__cpu_indicator_init
244 // LINUX: ret ptr @HasGeneric.Z
245 // LINUX: ret ptr @HasGeneric.S
246 // LINUX: ret ptr @HasGeneric.O
247 // LINUX: ret ptr @HasGeneric.A
248 // LINUX-NOT: call void @llvm.trap
250 // WINDOWS: define weak_odr dso_local void @HasGeneric() comdat
251 // WINDOWS: call void @__cpu_indicator_init
252 // WINDOWS: call void @HasGeneric.Z
253 // WINDOWS-NEXT: ret void
254 // WINDOWS: call void @HasGeneric.S
255 // WINDOWS-NEXT: ret void
256 // WINDOWS: call void @HasGeneric.O
257 // WINDOWS-NEXT: ret void
258 // WINDOWS: call void @HasGeneric.A
259 // WINDOWS-NEXT: ret void
260 // WINDOWS-NOT: call void @llvm.trap
262 ATTR(cpu_dispatch(atom, generic, ivybridge, knl))
263 void HasParams(int i, double d);
264 // LINUX: define weak_odr ptr @HasParams.resolver()
265 // LINUX: call void @__cpu_indicator_init
266 // LINUX: ret ptr @HasParams.Z
267 // LINUX: ret ptr @HasParams.S
268 // LINUX: ret ptr @HasParams.O
269 // LINUX: ret ptr @HasParams.A
270 // LINUX-NOT: call void @llvm.trap
272 // WINDOWS: define weak_odr dso_local void @HasParams(i32 %0, double %1) comdat
273 // WINDOWS: call void @__cpu_indicator_init
274 // WINDOWS: call void @HasParams.Z(i32 %0, double %1)
275 // WINDOWS-NEXT: ret void
276 // WINDOWS: call void @HasParams.S(i32 %0, double %1)
277 // WINDOWS-NEXT: ret void
278 // WINDOWS: call void @HasParams.O(i32 %0, double %1)
279 // WINDOWS-NEXT: ret void
280 // WINDOWS: call void @HasParams.A(i32 %0, double %1)
281 // WINDOWS-NEXT: ret void
282 // WINDOWS-NOT: call void @llvm.trap
284 ATTR(cpu_dispatch(atom, generic, ivybridge, knl))
285 int HasParamsAndReturn(int i, double d);
286 // LINUX: define weak_odr ptr @HasParamsAndReturn.resolver()
287 // LINUX: call void @__cpu_indicator_init
288 // LINUX: ret ptr @HasParamsAndReturn.Z
289 // LINUX: ret ptr @HasParamsAndReturn.S
290 // LINUX: ret ptr @HasParamsAndReturn.O
291 // LINUX: ret ptr @HasParamsAndReturn.A
292 // LINUX-NOT: call void @llvm.trap
294 // WINDOWS: define weak_odr dso_local i32 @HasParamsAndReturn(i32 %0, double %1) comdat
295 // WINDOWS: call void @__cpu_indicator_init
296 // WINDOWS: %[[RET:.+]] = musttail call i32 @HasParamsAndReturn.Z(i32 %0, double %1)
297 // WINDOWS-NEXT: ret i32 %[[RET]]
298 // WINDOWS: %[[RET:.+]] = musttail call i32 @HasParamsAndReturn.S(i32 %0, double %1)
299 // WINDOWS-NEXT: ret i32 %[[RET]]
300 // WINDOWS: %[[RET:.+]] = musttail call i32 @HasParamsAndReturn.O(i32 %0, double %1)
301 // WINDOWS-NEXT: ret i32 %[[RET]]
302 // WINDOWS: %[[RET:.+]] = musttail call i32 @HasParamsAndReturn.A(i32 %0, double %1)
303 // WINDOWS-NEXT: ret i32 %[[RET]]
304 // WINDOWS-NOT: call void @llvm.trap
306 ATTR(cpu_dispatch(atom, generic, pentium))
307 int GenericAndPentium(int i, double d);
308 // LINUX: define weak_odr ptr @GenericAndPentium.resolver()
309 // LINUX: call void @__cpu_indicator_init
310 // LINUX: ret ptr @GenericAndPentium.O
311 // LINUX: ret ptr @GenericAndPentium.B
312 // LINUX-NOT: ret ptr @GenericAndPentium.A
313 // LINUX-NOT: call void @llvm.trap
315 // WINDOWS: define weak_odr dso_local i32 @GenericAndPentium(i32 %0, double %1) comdat
316 // WINDOWS: call void @__cpu_indicator_init
317 // WINDOWS: %[[RET:.+]] = musttail call i32 @GenericAndPentium.O(i32 %0, double %1)
318 // WINDOWS-NEXT: ret i32 %[[RET]]
319 // WINDOWS: %[[RET:.+]] = musttail call i32 @GenericAndPentium.B(i32 %0, double %1)
320 // WINDOWS-NEXT: ret i32 %[[RET]]
321 // WINDOWS-NOT: call i32 @GenericAndPentium.A
322 // WINDOWS-NOT: call void @llvm.trap
324 ATTR(cpu_dispatch(atom, pentium))
325 int DispatchFirst(void);
326 // LINUX: define weak_odr ptr @DispatchFirst.resolver
327 // LINUX: ret ptr @DispatchFirst.O
328 // LINUX: ret ptr @DispatchFirst.B
330 // WINDOWS: define weak_odr dso_local i32 @DispatchFirst() comdat
331 // WINDOWS: %[[RET:.+]] = musttail call i32 @DispatchFirst.O()
332 // WINDOWS-NEXT: ret i32 %[[RET]]
333 // WINDOWS: %[[RET:.+]] = musttail call i32 @DispatchFirst.B()
334 // WINDOWS-NEXT: ret i32 %[[RET]]
336 ATTR(cpu_specific(atom))
337 int DispatchFirst(void) {return 0;}
338 // LINUX: define{{.*}} i32 @DispatchFirst.O
339 // LINUX: ret i32 0
341 // WINDOWS: define dso_local i32 @DispatchFirst.O()
342 // WINDOWS: ret i32 0
344 ATTR(cpu_specific(pentium))
345 int DispatchFirst(void) {return 1;}
346 // LINUX: define{{.*}} i32 @DispatchFirst.B
347 // LINUX: ret i32 1
349 // WINDOWS: define dso_local i32 @DispatchFirst.B
350 // WINDOWS: ret i32 1
352 ATTR(cpu_specific(knl))
353 void OrderDispatchUsageSpecific(void) {}
355 // CHECK: attributes #[[S]] = {{.*}}"target-features"="+avx,+cmov,+crc32,+cx16,+cx8,+f16c,+fsgsbase,+fxsr,+mmx,+pclmul,+popcnt,+rdrnd,+sahf,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt"
356 // CHECK-SAME: "tune-cpu"="ivybridge"
357 // CHECK: attributes #[[K]] = {{.*}}"target-features"="+adx,+aes,+avx,+avx2,+avx512cd,+avx512f,+bmi,+bmi2,+cmov,+crc32,+cx16,+cx8,+evex512,+f16c,+fma,+fsgsbase,+fxsr,+invpcid,+lzcnt,+mmx,+movbe,+pclmul,+popcnt,+prfchw,+rdrnd,+rdseed,+sahf,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt"
358 // CHECK-SAME: "tune-cpu"="knl"
359 // CHECK: attributes #[[O]] = {{.*}}"target-features"="+cmov,+cx16,+cx8,+fxsr,+mmx,+movbe,+sahf,+sse,+sse2,+sse3,+ssse3,+x87"
360 // CHECK-SAME: "tune-cpu"="atom"