[HIP] Add signbit(long double) decl

[ashi1]

Summary

Similarly to isinfinite, isinf, and isnan, a _MSC_VER version of signbit(long double) is required for MSVC headers.

[arsenm]

Testcase?

[ashi1]

In D93062#2446926, @arsenm wrote:

Testcase?

Not too sure how to properly write a testcase for this. I've reduced the failing case, but it requires the user to copy/install the hip headers from the HIP project.

test.cpp:

#include <hip/hip_runtime.h>

Running command:

C:\hip/bin/clang -std=c++14 -fms-extensions -fms-compatibility -isystem C:\hip/include -DHIP_ROCclr=1 -x hip test.cpp -o test.exe

In file included from test.cpp:1:
In file included from C:\hip/include\hip/hip_runtime.h:60:
In file included from C:\hip/include\hip/hcc_detail/hip_runtime.h:602:
In file included from C:\hip\lib\clang\12.0.0\include/cuda_wrappers/complex:75:
C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.28.29333\include\complex:249:16: error:

call to 'signbit' is ambiguous
  return (_STD signbit)(_Left);

...

[tra]

For CUDA we have tests in the test-suite (https://github.com/llvm/llvm-test-suite/tree/main/External/CUDA) and a handful of buildbots running them (e.g. http://lab.llvm.org:8011/#/builders/55).
AMD should probably set up some public build/test bots for HIP, too.
In this case, manual testing and rollback if something breaks is about all we can do.

[ashi1]

I was able to shrink down the testcase, but it still requires 2019 MSVC to be installed and Windows Kits 10 at a minimum.

test.cpp:

#include <__clang_cuda_math_forward_declares.h>
#include <cuda_wrappers/complex>
int main() {}

clang cc1 command:

C:\hip\bin\clang.exe -cc1 -triple amdgcn-amd-amdhsa -aux-triple x86_64-pc-windows-msvc -emit-obj -fcuda-is-device -isystem C:\hip\lib\clang\12.0.0\include -internal-isystem "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.28.29333\include" -internal-isystem "C:\Program Files (x86)\Windows Kits\10\Include\10.0.19041.0\ucrt" -std=c++14 -fms-extensions -fms-compatibility -fms-compatibility-version=19.28.29336 -x hip test.cpp

In file included from test.cpp:3:
In file included from C:\hip\lib\clang\12.0.0\include\cuda_wrappers/complex:75:
C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.28.29333\include\complex:249:16: error: call to 'signbit' is ambiguous

return (_STD signbit)(_Left);

[ashi1]

In D93062#2496542, @tra wrote:

For CUDA we have tests in the test-suite (https://github.com/llvm/llvm-test-suite/tree/main/External/CUDA) and a handful of buildbots running them (e.g. http://lab.llvm.org:8011/#/builders/55).
AMD should probably set up some public build/test bots for HIP, too.
In this case, manual testing and rollback if something breaks is about all we can do.

I will work with our team and look into adding HIP to the test-suite and AMD buildbots to the public llvm testing.

For now, is it OK to merge this patch?

[yaxunl]

Let's land this fix now. It will take some time to set up the buildbot. I think we already have this issue covered in HIP directed tests, right?

[tra]

Go ahead.

clang/lib/Headers/__clang_cuda_math_forward_declares.h

 __DEVICE__ double rint(double);
 __DEVICE__ float rint(float);
 __DEVICE__ double round(double);
 __DEVICE__ float round(float);
 __DEVICE__ double scalbln(double, long);
 __DEVICE__ float scalbln(float, long);
 __DEVICE__ double scalbn(double, int);
 __DEVICE__ float scalbn(float, int);
+#ifdef _MSC_VER
+__DEVICE__ bool signbit(long double);
+#endif
 __DEVICE__ bool signbit(double);
 __DEVICE__ bool signbit(float);
 __DEVICE__ double sin(double);
 __DEVICE__ float sin(float);
 __DEVICE__ double sinh(double);
 __DEVICE__ float sinh(float);
 __DEVICE__ double sqrt(double);
 __DEVICE__ float sqrt(float);