[primTorch] elementwise CPU strides are incorrect #77553
Description
a = torch.testing.make_tensor((2, 2), dtype=torch.long, device='cpu', noncontiguous=True).transpose(0, -1)
a.stride()
: (2, 4)
torch.cos(a).stride()
: (2, 1)
a = torch.testing.make_tensor((2, 2), dtype=torch.long, device='cuda', noncontiguous=True).transpose(0, -1)
a.stride()
: (2, 4)
torch.cos(a).stride()
: (1, 2)
Sometimes the CPU does preserve strides as expected, but for other inputs it doesn't.