Mac Mx compatibility

[Marx2]

There's ComfyUI node SD Ultimate Upscaler which fails on Mac based computers. It fails with an error:

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.
Fix is described here:
ssitu/ComfyUI_UltimateSDUpscale#143 which references this

It's just adding one line in one function:
x = x.contiguous()

spandrel/libs/spandrel/spandrel/architectures/ESRGAN/__arch/RRDB.py

Line 136 in afd06dd

def forward(self, x):

In short it modify code in this library, and then everything works as expected.

I'd like to know if this fix can be applied here, or this library is used in a wrong way by Upscaler, and fix should land in SD Ultimate Upscaler?

[joeyballentine]

The fix probably can be applied here. Does .contiguous just need to be applied to the passed in tensor before doing anything else, or is it only under the shuffle case? And is this the only model affected by it?

[Marx2]

I've met this problem multiple times in the past, trying other people's workflows in ComfyUI. Unfortunately I don't know if it's model specific. What I know it's platform specific. Most code is written for NVidia GPUs, which don't signal such an error.