Fix view size error when backpropagating through lfilter

[yoyololicon]

This commit solves the view size error when the gradient tensor propagates to the filter is not in a contiguous format.

The bug can be reproduced by the following script (provided by @forgi86) using the latest version of torchaudio.

import torch
import torchaudio.functional

order = 5
in_channels = 10
seq_len = 1000
batch_size = 32

b_coeff = torch.randn((in_channels, order)) * 0.05
b_coeff = b_coeff.requires_grad_(True)
a_coeff = torch.randn((in_channels, order - 1)) * 0.05
a_coeff = a_coeff.requires_grad_(True)

ones = torch.ones_like(a_coeff[..., :1])
a_coeffs_with_ones = torch.cat((ones, a_coeff), dim=-1)

u = torch.randn((batch_size, in_channels, seq_len))

y = torchaudio.functional.lfilter(u, a_coeffs_with_ones, b_coeff, clamp=False, batching=True)
y = y.transpose(-2, -1).reshape(batch_size, in_channels // 2, 2, seq_len)

loss = torch.sum(y**2)
loss.backward()

RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/audio/3794

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 7bd9b72 with merge base 1980f8a ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.