1 | // REQUIRES: x86-registered-target |
2 | // REQUIRES: nvptx-registered-target |
3 | |
4 | // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fsyntax-only -verify %s |
5 | // RUN: %clang_cc1 -triple nvptx64-nvidia-cuda -fsyntax-only -fcuda-is-device -verify %s |
6 | |
7 | #include "Inputs/cuda.h" |
8 | |
9 | // Opaque return types used to check that we pick the right overloads. |
10 | struct HostReturnTy {}; |
11 | struct HostReturnTy2 {}; |
12 | struct DeviceReturnTy {}; |
13 | struct DeviceReturnTy2 {}; |
14 | struct HostDeviceReturnTy {}; |
15 | struct TemplateReturnTy {}; |
16 | |
17 | typedef HostReturnTy (*HostFnPtr)(); |
18 | typedef DeviceReturnTy (*DeviceFnPtr)(); |
19 | typedef HostDeviceReturnTy (*HostDeviceFnPtr)(); |
20 | typedef void (*GlobalFnPtr)(); // __global__ functions must return void. |
21 | |
22 | // CurrentReturnTy is {HostReturnTy,DeviceReturnTy} during {host,device} |
23 | // compilation. |
24 | #ifdef __CUDA_ARCH__ |
25 | typedef DeviceReturnTy CurrentReturnTy; |
26 | #else |
27 | typedef HostReturnTy CurrentReturnTy; |
28 | #endif |
29 | |
30 | // CurrentFnPtr is a function pointer to a {host,device} function during |
31 | // {host,device} compilation. |
32 | typedef CurrentReturnTy (*CurrentFnPtr)(); |
33 | |
34 | // Host and unattributed functions can't be overloaded. |
35 | __host__ void hh() {} // expected-note {{previous definition is here}} |
36 | void hh() {} // expected-error {{redefinition of 'hh'}} |
37 | |
38 | // H/D overloading is OK. |
39 | __host__ HostReturnTy dh() { return HostReturnTy(); } |
40 | __device__ DeviceReturnTy dh() { return DeviceReturnTy(); } |
41 | |
42 | // H/HD and D/HD are not allowed. |
43 | __host__ __device__ int hdh() { return 0; } // expected-note {{previous declaration is here}} |
44 | __host__ int hdh() { return 0; } |
45 | // expected-error@-1 {{__host__ function 'hdh' cannot overload __host__ __device__ function 'hdh'}} |
46 | |
47 | __host__ int hhd() { return 0; } // expected-note {{previous declaration is here}} |
48 | __host__ __device__ int hhd() { return 0; } |
49 | // expected-error@-1 {{__host__ __device__ function 'hhd' cannot overload __host__ function 'hhd'}} |
50 | |
51 | __host__ __device__ int hdd() { return 0; } // expected-note {{previous declaration is here}} |
52 | __device__ int hdd() { return 0; } |
53 | // expected-error@-1 {{__device__ function 'hdd' cannot overload __host__ __device__ function 'hdd'}} |
54 | |
55 | __device__ int dhd() { return 0; } // expected-note {{previous declaration is here}} |
56 | __host__ __device__ int dhd() { return 0; } |
57 | // expected-error@-1 {{__host__ __device__ function 'dhd' cannot overload __device__ function 'dhd'}} |
58 | |
59 | // Same tests for extern "C" functions. |
60 | extern "C" __host__ int chh() { return 0; } // expected-note {{previous definition is here}} |
61 | extern "C" int chh() { return 0; } // expected-error {{redefinition of 'chh'}} |
62 | |
63 | // H/D overloading is OK. |
64 | extern "C" __device__ DeviceReturnTy cdh() { return DeviceReturnTy(); } |
65 | extern "C" __host__ HostReturnTy cdh() { return HostReturnTy(); } |
66 | |
67 | // H/HD and D/HD overloading is not allowed. |
68 | extern "C" __host__ __device__ int chhd1() { return 0; } // expected-note {{previous declaration is here}} |
69 | extern "C" __host__ int chhd1() { return 0; } |
70 | // expected-error@-1 {{__host__ function 'chhd1' cannot overload __host__ __device__ function 'chhd1'}} |
71 | |
72 | extern "C" __host__ int chhd2() { return 0; } // expected-note {{previous declaration is here}} |
73 | extern "C" __host__ __device__ int chhd2() { return 0; } |
74 | // expected-error@-1 {{__host__ __device__ function 'chhd2' cannot overload __host__ function 'chhd2'}} |
75 | |
76 | // Helper functions to verify calling restrictions. |
77 | __device__ DeviceReturnTy d() { return DeviceReturnTy(); } |
78 | // expected-note@-1 1+ {{'d' declared here}} |
79 | // expected-note@-2 1+ {{candidate function not viable: call to __device__ function from __host__ function}} |
80 | // expected-note@-3 0+ {{candidate function not viable: call to __device__ function from __host__ __device__ function}} |
81 | |
82 | __host__ HostReturnTy h() { return HostReturnTy(); } |
83 | // expected-note@-1 1+ {{'h' declared here}} |
84 | // expected-note@-2 1+ {{candidate function not viable: call to __host__ function from __device__ function}} |
85 | // expected-note@-3 0+ {{candidate function not viable: call to __host__ function from __host__ __device__ function}} |
86 | // expected-note@-4 1+ {{candidate function not viable: call to __host__ function from __global__ function}} |
87 | |
88 | __global__ void g() {} |
89 | // expected-note@-1 1+ {{'g' declared here}} |
90 | // expected-note@-2 1+ {{candidate function not viable: call to __global__ function from __device__ function}} |
91 | // expected-note@-3 0+ {{candidate function not viable: call to __global__ function from __host__ __device__ function}} |
92 | // expected-note@-4 1+ {{candidate function not viable: call to __global__ function from __global__ function}} |
93 | |
94 | extern "C" __device__ DeviceReturnTy cd() { return DeviceReturnTy(); } |
95 | // expected-note@-1 1+ {{'cd' declared here}} |
96 | // expected-note@-2 1+ {{candidate function not viable: call to __device__ function from __host__ function}} |
97 | // expected-note@-3 0+ {{candidate function not viable: call to __device__ function from __host__ __device__ function}} |
98 | |
99 | extern "C" __host__ HostReturnTy ch() { return HostReturnTy(); } |
100 | // expected-note@-1 1+ {{'ch' declared here}} |
101 | // expected-note@-2 1+ {{candidate function not viable: call to __host__ function from __device__ function}} |
102 | // expected-note@-3 0+ {{candidate function not viable: call to __host__ function from __host__ __device__ function}} |
103 | // expected-note@-4 1+ {{candidate function not viable: call to __host__ function from __global__ function}} |
104 | |
105 | __host__ void hostf() { |
106 | DeviceFnPtr fp_d = d; // expected-error {{reference to __device__ function 'd' in __host__ function}} |
107 | DeviceReturnTy ret_d = d(); // expected-error {{no matching function for call to 'd'}} |
108 | DeviceFnPtr fp_cd = cd; // expected-error {{reference to __device__ function 'cd' in __host__ function}} |
109 | DeviceReturnTy ret_cd = cd(); // expected-error {{no matching function for call to 'cd'}} |
110 | |
111 | HostFnPtr fp_h = h; |
112 | HostReturnTy ret_h = h(); |
113 | HostFnPtr fp_ch = ch; |
114 | HostReturnTy ret_ch = ch(); |
115 | |
116 | HostFnPtr fp_dh = dh; |
117 | HostReturnTy ret_dh = dh(); |
118 | HostFnPtr fp_cdh = cdh; |
119 | HostReturnTy ret_cdh = cdh(); |
120 | |
121 | GlobalFnPtr fp_g = g; |
122 | g(); // expected-error {{call to global function 'g' not configured}} |
123 | g<<<0, 0>>>(); |
124 | } |
125 | |
126 | __device__ void devicef() { |
127 | DeviceFnPtr fp_d = d; |
128 | DeviceReturnTy ret_d = d(); |
129 | DeviceFnPtr fp_cd = cd; |
130 | DeviceReturnTy ret_cd = cd(); |
131 | |
132 | HostFnPtr fp_h = h; // expected-error {{reference to __host__ function 'h' in __device__ function}} |
133 | HostReturnTy ret_h = h(); // expected-error {{no matching function for call to 'h'}} |
134 | HostFnPtr fp_ch = ch; // expected-error {{reference to __host__ function 'ch' in __device__ function}} |
135 | HostReturnTy ret_ch = ch(); // expected-error {{no matching function for call to 'ch'}} |
136 | |
137 | DeviceFnPtr fp_dh = dh; |
138 | DeviceReturnTy ret_dh = dh(); |
139 | DeviceFnPtr fp_cdh = cdh; |
140 | DeviceReturnTy ret_cdh = cdh(); |
141 | |
142 | GlobalFnPtr fp_g = g; // expected-error {{reference to __global__ function 'g' in __device__ function}} |
143 | g(); // expected-error {{no matching function for call to 'g'}} |
144 | g<<<0,0>>>(); // expected-error {{reference to __global__ function 'g' in __device__ function}} |
145 | } |
146 | |
147 | __global__ void globalf() { |
148 | DeviceFnPtr fp_d = d; |
149 | DeviceReturnTy ret_d = d(); |
150 | DeviceFnPtr fp_cd = cd; |
151 | DeviceReturnTy ret_cd = cd(); |
152 | |
153 | HostFnPtr fp_h = h; // expected-error {{reference to __host__ function 'h' in __global__ function}} |
154 | HostReturnTy ret_h = h(); // expected-error {{no matching function for call to 'h'}} |
155 | HostFnPtr fp_ch = ch; // expected-error {{reference to __host__ function 'ch' in __global__ function}} |
156 | HostReturnTy ret_ch = ch(); // expected-error {{no matching function for call to 'ch'}} |
157 | |
158 | DeviceFnPtr fp_dh = dh; |
159 | DeviceReturnTy ret_dh = dh(); |
160 | DeviceFnPtr fp_cdh = cdh; |
161 | DeviceReturnTy ret_cdh = cdh(); |
162 | |
163 | GlobalFnPtr fp_g = g; // expected-error {{reference to __global__ function 'g' in __global__ function}} |
164 | g(); // expected-error {{no matching function for call to 'g'}} |
165 | g<<<0,0>>>(); // expected-error {{reference to __global__ function 'g' in __global__ function}} |
166 | } |
167 | |
168 | __host__ __device__ void hostdevicef() { |
169 | DeviceFnPtr fp_d = d; |
170 | DeviceReturnTy ret_d = d(); |
171 | DeviceFnPtr fp_cd = cd; |
172 | DeviceReturnTy ret_cd = cd(); |
173 | #if !defined(__CUDA_ARCH__) |
174 | // expected-error@-5 {{reference to __device__ function 'd' in __host__ __device__ function}} |
175 | // expected-error@-5 {{reference to __device__ function 'd' in __host__ __device__ function}} |
176 | // expected-error@-5 {{reference to __device__ function 'cd' in __host__ __device__ function}} |
177 | // expected-error@-5 {{reference to __device__ function 'cd' in __host__ __device__ function}} |
178 | #endif |
179 | |
180 | HostFnPtr fp_h = h; |
181 | HostReturnTy ret_h = h(); |
182 | HostFnPtr fp_ch = ch; |
183 | HostReturnTy ret_ch = ch(); |
184 | #if defined(__CUDA_ARCH__) |
185 | // expected-error@-5 {{reference to __host__ function 'h' in __host__ __device__ function}} |
186 | // expected-error@-5 {{reference to __host__ function 'h' in __host__ __device__ function}} |
187 | // expected-error@-5 {{reference to __host__ function 'ch' in __host__ __device__ function}} |
188 | // expected-error@-5 {{reference to __host__ function 'ch' in __host__ __device__ function}} |
189 | #endif |
190 | |
191 | CurrentFnPtr fp_dh = dh; |
192 | CurrentReturnTy ret_dh = dh(); |
193 | CurrentFnPtr fp_cdh = cdh; |
194 | CurrentReturnTy ret_cdh = cdh(); |
195 | |
196 | GlobalFnPtr fp_g = g; |
197 | #if defined(__CUDA_ARCH__) |
198 | // expected-error@-2 {{reference to __global__ function 'g' in __host__ __device__ function}} |
199 | #endif |
200 | |
201 | g(); |
202 | #if defined (__CUDA_ARCH__) |
203 | // expected-error@-2 {{reference to __global__ function 'g' in __host__ __device__ function}} |
204 | #else |
205 | // expected-error@-4 {{call to global function 'g' not configured}} |
206 | #endif |
207 | |
208 | g<<<0,0>>>(); |
209 | #if defined(__CUDA_ARCH__) |
210 | // expected-error@-2 {{reference to __global__ function 'g' in __host__ __device__ function}} |
211 | #endif |
212 | } |
213 | |
214 | // Test for address of overloaded function resolution in the global context. |
215 | HostFnPtr fp_h = h; |
216 | HostFnPtr fp_ch = ch; |
217 | CurrentFnPtr fp_dh = dh; |
218 | CurrentFnPtr fp_cdh = cdh; |
219 | GlobalFnPtr fp_g = g; |
220 | |
221 | |
222 | // Test overloading of destructors |
223 | // Can't mix H and unattributed destructors |
224 | struct d_h { |
225 | ~d_h() {} // expected-note {{previous definition is here}} |
226 | __host__ ~d_h() {} // expected-error {{destructor cannot be redeclared}} |
227 | }; |
228 | |
229 | // HD is OK |
230 | struct d_hd { |
231 | __host__ __device__ ~d_hd() {} |
232 | }; |
233 | |
234 | // Test overloading of member functions |
235 | struct m_h { |
236 | void operator delete(void *ptr); // expected-note {{previous declaration is here}} |
237 | __host__ void operator delete(void *ptr); // expected-error {{class member cannot be redeclared}} |
238 | }; |
239 | |
240 | // D/H overloading is OK |
241 | struct m_dh { |
242 | __device__ void operator delete(void *ptr); |
243 | __host__ void operator delete(void *ptr); |
244 | }; |
245 | |
246 | // HD by itself is OK |
247 | struct m_hd { |
248 | __device__ __host__ void operator delete(void *ptr); |
249 | }; |
250 | |
251 | struct m_hhd { |
252 | __host__ void operator delete(void *ptr) {} // expected-note {{previous declaration is here}} |
253 | __host__ __device__ void operator delete(void *ptr) {} |
254 | // expected-error@-1 {{__host__ __device__ function 'operator delete' cannot overload __host__ function 'operator delete'}} |
255 | }; |
256 | |
257 | struct m_hdh { |
258 | __host__ __device__ void operator delete(void *ptr) {} // expected-note {{previous declaration is here}} |
259 | __host__ void operator delete(void *ptr) {} |
260 | // expected-error@-1 {{__host__ function 'operator delete' cannot overload __host__ __device__ function 'operator delete'}} |
261 | }; |
262 | |
263 | struct m_dhd { |
264 | __device__ void operator delete(void *ptr) {} // expected-note {{previous declaration is here}} |
265 | __host__ __device__ void operator delete(void *ptr) {} |
266 | // expected-error@-1 {{__host__ __device__ function 'operator delete' cannot overload __device__ function 'operator delete'}} |
267 | }; |
268 | |
269 | struct m_hdd { |
270 | __host__ __device__ void operator delete(void *ptr) {} // expected-note {{previous declaration is here}} |
271 | __device__ void operator delete(void *ptr) {} |
272 | // expected-error@-1 {{__device__ function 'operator delete' cannot overload __host__ __device__ function 'operator delete'}} |
273 | }; |
274 | |
275 | // __global__ functions can't be overloaded based on attribute |
276 | // difference. |
277 | struct G { |
278 | friend void friend_of_g(G &arg); // expected-note {{previous declaration is here}} |
279 | private: |
280 | int x; // expected-note {{declared private here}} |
281 | }; |
282 | __global__ void friend_of_g(G &arg) { int x = arg.x; } |
283 | // expected-error@-1 {{__global__ function 'friend_of_g' cannot overload __host__ function 'friend_of_g'}} |
284 | // expected-error@-2 {{'x' is a private member of 'G'}} |
285 | void friend_of_g(G &arg) { int x = arg.x; } |
286 | |
287 | // HD functions are sometimes allowed to call H or D functions -- this |
288 | // is an artifact of the source-to-source splitting performed by nvcc |
289 | // that we need to mimic. During device mode compilation in nvcc, host |
290 | // functions aren't present at all, so don't participate in |
291 | // overloading. But in clang, H and D functions are present in both |
292 | // compilation modes. Clang normally uses the target attribute as a |
293 | // tiebreaker between overloads with otherwise identical priority, but |
294 | // in order to match nvcc's behavior, we sometimes need to wholly |
295 | // discard overloads that would not be present during compilation |
296 | // under nvcc. |
297 | |
298 | template <typename T> TemplateReturnTy template_vs_function(T arg) { |
299 | return TemplateReturnTy(); |
300 | } |
301 | __device__ DeviceReturnTy template_vs_function(float arg) { |
302 | return DeviceReturnTy(); |
303 | } |
304 | |
305 | // Here we expect to call the templated function during host compilation, even |
306 | // if -fcuda-disable-target-call-checks is passed, and even though C++ overload |
307 | // rules prefer the non-templated function. |
308 | __host__ __device__ void test_host_device_calls_template(void) { |
309 | #ifdef __CUDA_ARCH__ |
310 | typedef DeviceReturnTy ExpectedReturnTy; |
311 | #else |
312 | typedef TemplateReturnTy ExpectedReturnTy; |
313 | #endif |
314 | |
315 | ExpectedReturnTy ret1 = template_vs_function(1.0f); |
316 | ExpectedReturnTy ret2 = template_vs_function(2.0); |
317 | } |
318 | |
319 | // Calls from __host__ and __device__ functions should always call the |
320 | // overloaded function that matches their mode. |
321 | __host__ void test_host_calls_template_fn() { |
322 | TemplateReturnTy ret1 = template_vs_function(1.0f); |
323 | TemplateReturnTy ret2 = template_vs_function(2.0); |
324 | } |
325 | |
326 | __device__ void test_device_calls_template_fn() { |
327 | DeviceReturnTy ret1 = template_vs_function(1.0f); |
328 | DeviceReturnTy ret2 = template_vs_function(2.0); |
329 | } |
330 | |
331 | // If we have a mix of HD and H-only or D-only candidates in the overload set, |
332 | // normal C++ overload resolution rules apply first. |
333 | template <typename T> TemplateReturnTy template_vs_hd_function(T arg) |
334 | #ifdef __CUDA_ARCH__ |
335 | //expected-note@-2 {{declared here}} |
336 | #endif |
337 | { |
338 | return TemplateReturnTy(); |
339 | } |
340 | __host__ __device__ HostDeviceReturnTy template_vs_hd_function(float arg) { |
341 | return HostDeviceReturnTy(); |
342 | } |
343 | |
344 | __host__ __device__ void test_host_device_calls_hd_template() { |
345 | HostDeviceReturnTy ret1 = template_vs_hd_function(1.0f); |
346 | TemplateReturnTy ret2 = template_vs_hd_function(1); |
347 | #ifdef __CUDA_ARCH__ |
348 | // expected-error@-2 {{reference to __host__ function 'template_vs_hd_function<int>' in __host__ __device__ function}} |
349 | #endif |
350 | } |
351 | |
352 | __host__ void test_host_calls_hd_template() { |
353 | HostDeviceReturnTy ret1 = template_vs_hd_function(1.0f); |
354 | TemplateReturnTy ret2 = template_vs_hd_function(1); |
355 | } |
356 | |
357 | __device__ void test_device_calls_hd_template() { |
358 | HostDeviceReturnTy ret1 = template_vs_hd_function(1.0f); |
359 | // Host-only function template is not callable with strict call checks, |
360 | // so for device side HD function will be the only choice. |
361 | HostDeviceReturnTy ret2 = template_vs_hd_function(1); |
362 | } |
363 | |
364 | // Check that overloads still work the same way on both host and |
365 | // device side when the overload set contains only functions from one |
366 | // side of compilation. |
367 | __device__ DeviceReturnTy device_only_function(int arg) { return DeviceReturnTy(); } |
368 | __device__ DeviceReturnTy2 device_only_function(float arg) { return DeviceReturnTy2(); } |
369 | #ifndef __CUDA_ARCH__ |
370 | // expected-note@-3 {{'device_only_function' declared here}} |
371 | // expected-note@-3 {{'device_only_function' declared here}} |
372 | #endif |
373 | __host__ HostReturnTy host_only_function(int arg) { return HostReturnTy(); } |
374 | __host__ HostReturnTy2 host_only_function(float arg) { return HostReturnTy2(); } |
375 | #ifdef __CUDA_ARCH__ |
376 | // expected-note@-3 {{'host_only_function' declared here}} |
377 | // expected-note@-3 {{'host_only_function' declared here}} |
378 | #endif |
379 | |
380 | __host__ __device__ void test_host_device_single_side_overloading() { |
381 | DeviceReturnTy ret1 = device_only_function(1); |
382 | DeviceReturnTy2 ret2 = device_only_function(1.0f); |
383 | #ifndef __CUDA_ARCH__ |
384 | // expected-error@-3 {{reference to __device__ function 'device_only_function' in __host__ __device__ function}} |
385 | // expected-error@-3 {{reference to __device__ function 'device_only_function' in __host__ __device__ function}} |
386 | #endif |
387 | HostReturnTy ret3 = host_only_function(1); |
388 | HostReturnTy2 ret4 = host_only_function(1.0f); |
389 | #ifdef __CUDA_ARCH__ |
390 | // expected-error@-3 {{reference to __host__ function 'host_only_function' in __host__ __device__ function}} |
391 | // expected-error@-3 {{reference to __host__ function 'host_only_function' in __host__ __device__ function}} |
392 | #endif |
393 | } |
394 | |
395 | // Verify that we allow overloading function templates. |
396 | template <typename T> __host__ T template_overload(const T &a) { return a; }; |
397 | template <typename T> __device__ T template_overload(const T &a) { return a; }; |
398 | |
399 | __host__ void test_host_template_overload() { |
400 | template_overload(1); // OK. Attribute-based overloading picks __host__ variant. |
401 | } |
402 | __device__ void test_device_template_overload() { |
403 | template_overload(1); // OK. Attribute-based overloading picks __device__ variant. |
404 | } |
405 | |