[SYCL] Fix __divdc3 declaration for SYCL device

[jinge90]

No description provided.

[bader]

Can you write a test to catch this kind of error?

[github-actions]

@intel/llvm-gatekeepers please consider merging

[jinge90]

Can you write a test to catch this kind of error?

Hi, @bader
I added explicit declarations for __mulsc3, __muldc3, __divsc3, __divdc3 in current test code.
Before the fix, building std::complex e2e test, will lead to:
./complex_utils.hpp:14:20: error: conflicting types for '__divdc3'
14 | double complex __divdc3(double a, double b, double c, double d);
| ^
../llvm/build/install/bin/../include/sycl/stl_wrappers/cmath:219:54: note: previous declaration is here
219 | extern __DPCPP_SYCL_EXTERNAL_LIBC double complex __divdc3(float a,
| ^
1 error generated.

After the fix, no warning or compilation error happens.
Is thik OK to you?
Thanks very much.

[bader]

Can you write a test to catch this kind of error?

Hi, @bader I added explicit declarations for __mulsc3, __muldc3, __divsc3, __divdc3 in current test code. Before the fix, building std::complex e2e test, will lead to: ./complex_utils.hpp:14:20: error: conflicting types for '__divdc3' 14 | double complex __divdc3(double a, double b, double c, double d); | ^ ../llvm/build/install/bin/../include/sycl/stl_wrappers/cmath:219:54: note: previous declaration is here 219 | extern __DPCPP_SYCL_EXTERNAL_LIBC double complex __divdc3(float a, | ^ 1 error generated.

After the fix, no warning or compilation error happens. Is thik OK to you? Thanks very much.

Adding GLIBC declarations to the test doesn't sound right to me.
Can't we just compile the test with -Werror=narrowing? If I understand correctly, w/o your fix std::complex<double> division uses 32-bit FP data types instead of 64-bit. I hope we can catch this issue withe compiler diagnostics.

[jinge90]

Can you write a test to catch this kind of error?

Hi, @bader I added explicit declarations for __mulsc3, __muldc3, __divsc3, __divdc3 in current test code. Before the fix, building std::complex e2e test, will lead to: ./complex_utils.hpp:14:20: error: conflicting types for '__divdc3' 14 | double complex __divdc3(double a, double b, double c, double d); | ^ ../llvm/build/install/bin/../include/sycl/stl_wrappers/cmath:219:54: note: previous declaration is here 219 | extern __DPCPP_SYCL_EXTERNAL_LIBC double complex __divdc3(float a, | ^ 1 error generated.
After the fix, no warning or compilation error happens. Is thik OK to you? Thanks very much.

Adding GLIBC declarations to the test doesn't sound right to me. Can't we just compile the test with -Werror=narrowing? If I understand correctly, w/o your fix std::complex<double> division uses 32-bit FP data types instead of 64-bit. I hope we can catch this issue withe compiler diagnostics.

Hi, @bader
I find -Werror=narrowing can't work but -Werror=implicit-float-conversion does work, add a new test in sycl/test/regression/ ,could you help review it?
Thanks very much.