Pad CV-CUDA Backend

[justincdavis]

Summary

Add the CV-CUDA backend kernel for the pad transform

Testing

python3 -m pytest test/test_transforms_v2.py::TestPad

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9289

• 📄 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 b68b3a7 with merge base 1e53952 ():
💚 Looks good so far! There are no failures yet. 💚

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

[zy1git]

I left one minor comment.

[zy1git]

We can use rewrite this line to if not _pad_mode_to_cvcuda_border:

[NicolasHug]

@zy1git , I'm surprised to see device="cpu" here. Can you check if there's a particular reason that it was set in the first place? Is it still relevant?
We should probably remove it, at least when make_input is make_image_cvcuda.

[zy1git]

I checked out the make_image function in the common_utils.py file and find that the default is "cpu" for device:

def make_image(
    size=DEFAULT_SIZE,
    *,
    color_space="RGB",
    batch_dims=(),
    dtype=None,
    device="cpu",
    memory_format=torch.contiguous_format,
):

And the function make_image_cvcuda called the function make_image:

def make_image_cvcuda(*args, batch_dims=(1,), **kwargs):
    return to_cvcuda_tensor(make_image(*args, batch_dims=batch_dims, **kwargs))

Thus, I think it is safe to remove the device="cpu". I am not sure why it was set in the first place. Maybe the previous author want to be explicit for clarity/readability?

[zy1git]

address one comment

[zy1git]

I checked out the make_image function in the common_utils.py file and find that the default is "cpu" for device:

def make_image(
    size=DEFAULT_SIZE,
    *,
    color_space="RGB",
    batch_dims=(),
    dtype=None,
    device="cpu",
    memory_format=torch.contiguous_format,
):

And the function make_image_cvcuda called the function make_image:

def make_image_cvcuda(*args, batch_dims=(1,), **kwargs):
    return to_cvcuda_tensor(make_image(*args, batch_dims=batch_dims, **kwargs))

Thus, I think it is safe to remove the device="cpu". I am not sure why it was set in the first place. Maybe the previous author want to be explicit for clarity/readability?