I was looking through some logs and noticed the following slow tests. I'm not sure if a timestamp skip means that the test before or the test after was slow; but I've seen big jumps on "skipped" tests under the former interpretation, so this issue is written under the assumption that it's the after.
test_baddbmm_broadcast_lhs_coef, test_baddbmm_coef, test_baddbmm_scalar_broadcast_lhs, test_baddbmm_scalar_broadcast_lhs_coef, test_bmm
Oct 29 00:52:20 test_backward_no_grad (__main__.TestAutograd) ... ok
Oct 29 00:56:29 test_baddbmm (__main__.TestAutograd) ... ok
Oct 29 00:59:32 test_baddbmm_broadcast_lhs (__main__.TestAutograd) ... ok
Oct 29 01:02:29 test_baddbmm_broadcast_lhs_coef (__main__.TestAutograd) ... ok
Oct 29 01:06:42 test_baddbmm_coef (__main__.TestAutograd) ... ok
Oct 29 01:09:36 test_baddbmm_scalar_broadcast_lhs (__main__.TestAutograd) ... ok
Oct 29 01:12:48 test_baddbmm_scalar_broadcast_lhs_coef (__main__.TestAutograd) ... ok
Oct 29 01:17:28 test_bmm (__main__.TestAutograd) ... ok
test_transpose:
Oct 29 01:29:58 test_to_sparse (__main__.TestUncoalescedSparse) ... ok
Oct 29 01:31:29 test_transpose (__main__.TestUncoalescedSparse) ... ok
test_matmul_1d_4d (not replicated in python3.5)
Oct 29 01:20:43 test_matmul (__main__.TestAutograd) ... ok
Oct 29 01:20:43 test_matmul_1d_2d (__main__.TestAutograd) ... ok
Oct 29 01:24:50 test_matmul_1d_4d (__main__.TestAutograd) ... ok
Oct 29 01:24:50 test_matmul_2d_1d (__main__.TestAutograd) ... ok
test_fft_ifft_rfft_irfft (less pronounced in python3.5)
Oct 29 01:17:42 test_expm1 (__main__.TestAutograd) ... ok
Oct 29 01:17:42 test_expm1_scalar (__main__.TestAutograd) ... ok
Oct 29 01:20:28 test_fft_ifft_rfft_irfft (__main__.TestAutograd) ... ok
Oct 29 01:20:28 test_fill (__main__.TestAutograd) ... ok
test_softmax_spatial_dtype, test_softmax_spatial_special: (only special replicated in python3.5)
Oct 29 00:50:54 test_softmax_spatial_cuda (__main__.TestNN) ... skipped 'Excluded from CUDA tests'
Oct 29 00:51:04 test_softmax_spatial_dtype (__main__.TestNN) ... ok
Oct 29 00:51:04 test_softmax_spatial_dtype_cuda (__main__.TestNN) ... skipped 'Excluded from CUDA tests'
Oct 29 00:52:03 test_softmax_spatial_special (__main__.TestNN) ... ok
Oct 29 00:52:03 test_softmax_spatial_special_cuda (__main__.TestNN) ... skipped 'Excluded from CUDA tests'
test_conv_double_backward, test_conv_double_backward_groups (generally the convolution tests are slow-ish)
Oct 29 00:41:37 test_container_copy (__main__.TestNN) ... ok
Oct 29 00:41:37 test_contig_wrong_stride_cudnn (__main__.TestNN) ... skipped 'needs cudnn'
Oct 29 00:48:49 test_conv_double_backward (__main__.TestNN) ... ok
Oct 29 00:48:49 test_conv_double_backward_cuda (__main__.TestNN) ... skipped 'CUDA unavailable'
Oct 29 00:49:02 test_conv_double_backward_groups (__main__.TestNN) ... ok
test_AdaptiveMaxPool3d_single, test_AdaptiveMaxPool3d_single_nonatomic
Oct 29 00:26:44 test_AdaptiveMaxPool3d_indices_cuda (__main__.TestNN) ... skipped 'CUDA unavailable'
Oct 29 00:29:05 test_AdaptiveMaxPool3d_single (__main__.TestNN) ... ok
Oct 29 00:29:05 test_AdaptiveMaxPool3d_single_cuda (__main__.TestNN) ... skipped 'Excluded from CUDA tests'
Oct 29 00:31:31 test_AdaptiveMaxPool3d_single_nonatomic (__main__.TestNN) ... ok
test_AdaptiveAvgPool3d_tuple_none
Oct 29 00:23:44 test_AdaptiveAvgPool3d_tuple (__main__.TestNN) ... ok
Oct 29 00:23:44 test_AdaptiveAvgPool3d_tuple_cuda (__main__.TestNN) ... skipped 'Excluded from CUDA tests'
Oct 29 00:25:53 test_AdaptiveAvgPool3d_tuple_none (__main__.TestNN) ... ok
Oct 29 00:25:53 test_AdaptiveAvgPool3d_tuple_none_cuda (__main__.TestNN) ... skipped 'Excluded from CUDA tests'
Source log: https://circleci.com/gh/pytorch/pytorch/137930?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link
Python 3.5 source log: https://circleci.com/gh/pytorch/pytorch/137928?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link
cc @zdevito
I was looking through some logs and noticed the following slow tests. I'm not sure if a timestamp skip means that the test before or the test after was slow; but I've seen big jumps on "skipped" tests under the former interpretation, so this issue is written under the assumption that it's the after.
test_baddbmm_broadcast_lhs_coef, test_baddbmm_coef, test_baddbmm_scalar_broadcast_lhs, test_baddbmm_scalar_broadcast_lhs_coef, test_bmm
test_transpose:
test_matmul_1d_4d (not replicated in python3.5)
test_fft_ifft_rfft_irfft (less pronounced in python3.5)
test_softmax_spatial_dtype, test_softmax_spatial_special: (only special replicated in python3.5)
test_conv_double_backward, test_conv_double_backward_groups (generally the convolution tests are slow-ish)
test_AdaptiveMaxPool3d_single, test_AdaptiveMaxPool3d_single_nonatomic
test_AdaptiveAvgPool3d_tuple_none
Source log: https://circleci.com/gh/pytorch/pytorch/137930?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link
Python 3.5 source log: https://circleci.com/gh/pytorch/pytorch/137928?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link
cc @zdevito