💊 CI failures summary and remediations

As of commit 72466d2 (more details on the Dr. CI page):

✅ None of the CI failures appear to be your fault 💚

• 1/1 tentatively recognized as flaky ❄️
  • Click here to rerun these jobs

❄️ 1 failure tentatively classified as flaky

but reruns have not yet been triggered to confirm:

pytorch_windows_vs2019_py36_cpu_test2 (1/1)

Step: "Test" (full log | diagnosis details | 🔁 rerun) ❄️

 if 0 NEQ 0 (exit /b 0  )   
 call conda install -y -q -c conda-forge cmake   
 if 0 NEQ 0 (exit /b 0  )  
)  
The system cannot find the drive specified. 
Collecting package metadata (current_repodata.json): ...working... done 
Solving environment: ...working... failed with initial frozen solve. Retrying with flexible solve. 
Collecting package metadata (repodata.json): ...working... done 
Solving environment: ...working... done 
 
CondaHTTPError: HTTP 000 CONNECTION FAILED for url <https://repo.anaconda.com/pkgs/main/win-64/llvmlite-0.29.0-py36ha925a31_0.conda> 
Elapsed: - 
 
An HTTP error occurred when trying to retrieve this URL. 
HTTP errors are often intermittent, and a simple retry will get you on your way. 
 
 
 
## Package Plan ## 
 
  environment location: C:\Jenkins\Miniconda3 

Reviewer: as of 8/1/2020, cuda 11 + windows is not yet available in the CI

ROCm failure in 72d716e seems irrelevant, because all CI passed in 874403e.

cc @ptrblck

@xwang233 I'm trying to add Windows CUDA 11 CI with #42420 and now it is blocked by the same error as described in #42406.

@xwang233 I'll try to cherry-pick your fix there.

@peterjc123 Thanks for checking the build!

New error after cherry-picking your fix:

C:/Users/circleci/project/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu(875): error: expression must have a constant value
C:/Users/circleci/project/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu(875): note: the value of variable "num_matrices"
(831): here cannot be used as a constant

C:/Users/circleci/project/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu(879): error: identifier "search_end_matrix_indices" is undefined

C:/Users/circleci/project/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu(903): error: identifier "getTensorCudaDataType" is undefined

C:/Users/circleci/project/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu(903): error: identifier "getTensorCudaDataType" is undefined

4 errors detected in the compilation of "C:/Users/circleci/project/aten/src/ATen/native/sparse/cuda/SparseCUDATensorMath.cu".

See https://app.circleci.com/pipelines/github/pytorch/pytorch/195766/workflows/bc3ba331-dd46-4ec5-9356-02cdfa015561/jobs/6468496.
Seems to be caused by using VLA which is not supported in MSVC and the wrong preprocessor condition in

I have reverted the code in SparseCUDATensorMath.cu.

@xwang233 So the new api of cusparse is still not supported on Windows, right? I'll have a bit more testing with 0c00fa3.

The problem in #42406 will be fixed, since it crashes cuda11 + windows build.

The change in bmm-sparse-dense is optional at this moment, since it doesn't crash any build. bmm-sparse-dense just raises TORCH_CHECK error in windows now regardless of cuda version. The cusparse API itself should be fine on cuda11 + windows. It would be nice if you can find a fix. If not, we can leave that in the next PR.

You can check the table in the first post of this thread.

The problem in #42406 will be fixed, since it crashes cuda11 + windows build.

The change in bmm-sparse-dense is optional at this moment, since it doesn't crash any build. bmm-sparse-dense just raises TORCH_CHECK error in windows now regardless of cuda version. The cusparse API itself should be fine on cuda11 + windows. It would be nice if you can find a fix. If not, we can leave that in the next PR.

You can check the table in the first post of this thread.

For the latest build in the pr, it seems that it is supported. But yes, it's okay to bring them in with later commits.

@ezyang merged this pull request in c8cb5e5.