Expose IResizeLayer in dynamo.conversion.impl

[narendasan]

[github-actions]

This issue has not seen activity for 90 days, Remove stale label or comment or this will be closed in 10 days

[bowang007]

torch.ops.aten.resize.default is doing something like resizing the data's memory instead of what we expect for resize a image.
For resizing an input from [4, 4] to [2, 2]:

input:  [tensor([[-0.0766,  0.3599, -0.7820,  0.0715],
        [ 0.6648, -0.2868,  1.6206, -1.5967],
        [-0.0517, -0.3060,  0.2485, -0.2226],
        [ 0.9132,  0.2043,  0.5740,  0.4163]])]
trt:  [tensor([[-0.0766,  0.0715],
        [ 0.9132,  0.4163]], device='cuda:0')]
torch:  [tensor([[-0.0766,  0.3599],
        [-0.7820,  0.0715]])]

Similarly, if we resize an input from [4, 4] to [8, 8], we have:

[ 0.0000e+00,  0.0000e+00,  0.0000e+00,  0.0000e+00,  0.0000e+00,
          0.0000e+00,  0.0000e+00,  0.0000e+00],

for filling output.

TensorRT's IResizeLayer is more likely to be torch.nn.functional.interpolate, which emphasize more on interpolating.

[peri044]

PR linked : #2488