[CUDA][HIP} Add a test for constexpr default ctor

[tra]

test/SemaCUDA/constexpr-ctor.cu

14–27 ↗

(On Diff #224225)

Do we really need three identical templates? If they are needed to let compiler emit multiple diagnostics, perhaps we could just add another template parameter so we can get different instantiations.

[yaxunl]

test/SemaCUDA/constexpr-ctor.cu

14–27 ↗

(On Diff #224225)

the error is emitted on the default ctor of the template. If we do not use different templates, all errors are emitted on the same line, we cannot make sure some instantiations do not cause error and some instantiations should cause error.

Adding template parameter will not help, because the error will still be emitted to the same line.

[tra]

test/SemaCUDA/constexpr-ctor.cu

14–27 ↗

(On Diff #224225)

Having multiple errors attributed to the same source is OK. You can specify expected count. E.g.
dev-error 2 {{something}}. Single template appears to be sufficient (https://godbolt.org/z/G3WvYD).

You'll have different note diags emitted for individual errors, so checking them would both provide more info about that exactly the problem is and will verify that they are reported in the correct locations. Right now you are describing what/where triggers an error that as a comment. Letting compiler verify that instead would be better, IMO.

[yaxunl]

test/SemaCUDA/constexpr-ctor.cu

14–27 ↗

(On Diff #224225)

Yes I can use note. will do.

[yaxunl]

revised by Artem's comments

[tra]

LGTM.
Sometimes I wish it would be possible to specify some of the -verify diagnostics in temporal order, as opposed to tying them to locations. I.e. in this case it would be way more useful to see the call stack leading to the error at B::B. Oh, well.

clang/test/SemaCUDA/constexpr-ctor.cu

+// RUN: %clang_cc1 -std=c++11 -triple nvptx64-nvidia-cuda -fsyntax-only \
+// RUN:            -fcuda-is-device -verify=dev %s
+// RUN: %clang_cc1 -std=c++11 -triple x86_64-unknown-linux-gnu -fsyntax-only \
+// RUN:            -verify=host %s
+
+// host-no-diagnostics
+
+#include "Inputs/cuda.h"
+
+struct A {
+  A(); // dev-note 2{{'A' declared here}}
+};
+
+template <class T, int x> struct B {
+  T a;
+  constexpr B() = default; // dev-error 2{{reference to __host__ function 'A' in __host__ __device__ function}}
+};
+
+__host__ void f() { B<A, 1> x; }
+__device__ void f() { B<A, 1> x; } // dev-note{{called by 'f'}}
+
+struct foo {
+  __host__ foo() { B<A, 2> x; }
+  __device__ foo() { B<A, 2> x; } // dev-note{{called by 'foo'}}
+};
+
+__host__ void g() { foo x; }
+__device__ void g() { foo x; } // dev-note{{called by 'g'}}
+
+struct bar {
+  __host__ bar() { B<A, 3> x; }
+  __device__ bar() { B<A, 3> x; } // no error since no instantiation of bar
+};