Skip to content

nvfuser complex fails opinfo tests #1730

@jjsjann123

Description

@jjsjann123

🐛 Describe the bug

running with PYTORCH_TEST_WITH_SLOW=1 python test_jit_cuda_fuser.py -v TestCudaFuserOpInfoCUDA

root@cd04ced199a7:/opt/pytorch/pytorch# grep "FAIL:" /jiej/playground/opinfo_log_0524_all 
FAIL: test_nvfuser_correctness___rpow___cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness__masked_std_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness__masked_std_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness__masked_var_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness__masked_var_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness_pow_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness_std_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness_std_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness_var_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_correctness_var_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values___radd___cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values___radd___cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values___rpow___cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values___rpow___cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values__masked_mean_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values__masked_mean_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values__masked_normalize_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values__masked_normalize_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_add_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_add_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_nn_functional_normalize_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_nn_functional_normalize_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_pow_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_pow_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_reciprocal_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_reciprocal_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_rsub_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_rsub_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_sub_cuda_complex128 (__main__.TestCudaFuserOpInfoCUDA)
FAIL: test_nvfuser_extremal_values_sub_cuda_complex64 (__main__.TestCudaFuserOpInfoCUDA)

I'm trying to push code base to upstream. So I'll be disabling these tests (will also cherry-pick that into our devel branch).

Versions

This fails on our devel which is on dd23252

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions