================
@@ -0,0 +1,7 @@
+// Verify that a .c file compiled with -fsycl is an error.
+// RUN: not %clang -### -fsycl %s 2>&1 | FileCheck -check-prefix ERR %s
+// ERR: error: invalid argument '{{.*}}sycl-c-warn.c' not allowed with '-fsycl'
+
+// Verify that explicitly forcing -x c with -fsycl is also an error.
----------------
srividya-sundaram wrote:

I moved the C-input rejection into the frontend based on the feedback. But that 
caused a handful of existing tests to fail.

I looked through the failures and none of them are actually testing C-language 
behavior. Most are using .c files for historical reasons. I fixed most of them 
by adding `-x c++`.

The two tricky cases are `SemaSPIRV/BuiltIns/ids_and_ranges.c` and 
`SemaSPIRV/BuiltIns/subgroup-errors.c`. 

These are SPIR-V builtin tests that seem to use `-fsycl-is-device` mainly as a 
convenient way to get a SPIR-V compilation environment. They already have 
equivalent `-x cl` RUN lines that exercise the same builtins.

Adding `-x c++` to the `-fsycl-is-device` RUN lines changes the diagnostics 
because the existing // expected-error annotations were written against C-mode 
diagnostics, which differ from C++ mode.

Can we just remove the `-fsycl-is-device` RUN lines from those two tests given 
the `-x cl` lines already cover the same SPIRV builtin validation or
Or keep `-fsycl-is-device` but add `-x c++` and update the expected diagnostics 
to the correct C++ wording?

https://github.com/llvm/llvm-project/pull/200318
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to