[SYCL][XPTI] Fix test

[vladimirlaz]

No description provided.

[bader]

Thanks!

[bader]

@vladimirlaz, I still see a lot of failed tests in pre-commit.
They are not related to this particular change, but we need to fix them too.

[vladimirlaz]

@vladimirlaz, I still see a lot of failed tests in pre-commit. They are not related to this particular change, but we need to fix them too.

I am currently looking into the failures.

[bader]

SYCL :: Matrix/joint_matrix_tensorcore_double.cpp - this one impacts a lot of other PRs, so we need to do something about it ASAP.

Timed Out Tests (1): SYCL :: ESIMD/api/simd_view_copy_move_assign.cpp - I think I've seen another PR with ESIMD test killed by timeout. @intel/dpcpp-esimd-reviewers, please, take a look.

SYCL :: Basic/free_function_queries/free_function_queries.cpp - compilation error - this is something new, which you should be able to investigate yourself.

include\sycl\CL/sycl/id.hpp(367,44): error: no template named 'declptr' in namespace 'sycl::ext::oneapi::experimental::detail'; did you mean '::sycl::detail::declptr'?
return sycl::detail::Builder::getElement(detail::declptr<id>());

[bader]

It looks like SYCL :: Basic/free_function_queries/free_function_queries.cpp and SYCL :: Matrix/joint_matrix_tensorcore_double.cpp are caused by intel/llvm#5169.
@capatober, FYI.

[vladimirlaz]

SYCL :: Matrix/joint_matrix_tensorcore_double.cpp - this one impacts a lot of other PRs, so we need to do something about it ASAP.

Timed Out Tests (1): SYCL :: ESIMD/api/simd_view_copy_move_assign.cpp - I think I've seen another PR with ESIMD test killed by timeout. @intel/dpcpp-esimd-reviewers, please, take a look.

SYCL :: Basic/free_function_queries/free_function_queries.cpp - compilation error - this is something new, which you should be able to investigate yourself.

include\sycl\CL/sycl/id.hpp(367,44): error: no template named 'declptr' in namespace 'sycl::ext::oneapi::experimental::detail'; did you mean '::sycl::detail::declptr'? return sycl::detail::Builder::getElement(detail::declptr());

The same tests have passed on Linux which is weird. It looks like the nightly compiler build for windows is not available/delayed.
@tfzhu, @DoyleLi could you comment?

[bader]

@vladimirlaz the fix doesn't work - https://github.com/intel/llvm/runs/4615454713?check_suite_focus=true#step:5:1810.
I suggest we revert this change to fix pre-commit again.

[bader]

SYCL :: Matrix/joint_matrix_tensorcore_double.cpp - this one impacts a lot of other PRs, so we need to do something about it ASAP.

The same tests have passed on Linux which is weird. It looks like the nightly compiler build for windows is not available/delayed. @tfzhu, @DoyleLi could you comment?

SYCL :: Matrix/joint_matrix_tensorcore_double.cpp fails on Linux/CUDA and it looks like we need to update the test. intel/llvm#5169 moved some functions to a different namespace.

[bader]

SYCL :: Matrix/joint_matrix_tensorcore_double.cpp - this one impacts a lot of other PRs, so we need to do something about it ASAP.

The same tests have passed on Linux which is weird. It looks like the nightly compiler build for windows is not available/delayed. @tfzhu, @DoyleLi could you comment?

SYCL :: Matrix/joint_matrix_tensorcore_double.cpp fails on Linux/CUDA and it looks like we need to update the test. intel/llvm#5169 moved some functions to a different namespace.

Or maybe intel/llvm#5169 exposed a bug in matrix API implementation.
@yubingex007-a11y, please, take a look.
Meanwhile, I'll disable the test to unblock llvm-test-suite validation on CUDA.

[andreyfe1]

Hi all,
Changes in intel/llvm#5169 affect only group sort extension. It shouldn't affect any other parts of repo since things, which were moved into different namespace, are used by group sort extension only.

I've spent some time to look at the failure for Matrix/joint_matrix_tensorcore_double.cpp and find following.

1. Look at the code. It's called from the sycl::ext::oneapi::experimental::matrix and calls function from the detail namespace. I'd expect the full path for detail is sycl::ext::oneapi::experimental::matrix::detail, but no, it's from sycl::ext::oneapi::detail that looks strange for me.
   Possible ways to fix:
   a) move local detail to sycl::ext::oneapi::experimental::matrix::detail
   b) use fully qualified namespace that is required by some C++ guidelines (example 1, example 2)

2. Look at the code. We have using namespace in headers! It should definitely be fixed. As a C++ developer I wouldn't expect to see using namespace in the C++ code especially in library and/or headers that are used by multiple users.

So, we have any mix for namespaces in the https://github.com/intel/llvm/blob/sycl/sycl/include/sycl/ext/oneapi/matrix/matrix-tensorcore.hpp file. In my opinion, it's accidentally worked without errors previously.

[JackAKirk]

Hi all, Changes in intel/llvm#5169 affect only group sort extension. It shouldn't affect any other parts of repo since things, which were moved into different namespace, are used by group sort extension only.

I've spent some time to look at the failure for Matrix/joint_matrix_tensorcore_double.cpp and find following.

1. Look at [the code](https://github.com/intel/llvm/blob/sycl/sycl/include/sycl/ext/oneapi/matrix/matrix-tensorcore.hpp#L238). It's called from the `sycl::ext::oneapi::experimental::matrix` and calls function from the `detail` namespace. I'd expect the full path for `detail` is `sycl::ext::oneapi::experimental::matrix::detail`, but no, it's from `sycl::ext::oneapi::detail` that looks strange for me.
   Possible ways to fix:
   a) move local `detail` to `sycl::ext::oneapi::experimental::matrix::detail`
   b) use fully qualified namespace that is required by some C++ guidelines ([example 1](https://google.github.io/styleguide/cppguide.html#Namespaces), [example 2](https://docs.microsoft.com/en-us/cpp/cpp/namespaces-cpp?view=msvc-170#using_directives))

2. Look at [the code](https://github.com/intel/llvm/blob/sycl/sycl/include/sycl/ext/oneapi/matrix/matrix-tensorcore.hpp#L54). We have `using namespace` in headers! It should definitely be fixed. As a C++ developer I wouldn't expect to see `using namespace` in the C++ code especially in library and/or headers that are used by multiple users.

So, we have any mix for namespaces in the https://github.com/intel/llvm/blob/sycl/sycl/include/sycl/ext/oneapi/matrix/matrix-tensorcore.hpp file. In my opinion, it's accidentally worked without errors previously.

Apologies for introducing this bug, and thanks for the comment. It is addressed here: intel/llvm#5217

[andreyfe1]

@JackAKirk, never mind:)
It happens since we have multiple contributors for the repo and multiple files that can be related to each other (even unexpectedly).