🐛 Describe the bug

Possible regression on #94190?
Calling torch.rand_like() on a tensor on mps device after it has gone through permute() doesn't work.

import torch

# Works
print("cpu")
t = torch.ones((3,2,2)).permute(2,0,1)
print(torch.rand_like(t))

# Not random
print("mps")
t = torch.ones((3,2,2)).to('mps').permute(2,0,1)
print(torch.rand_like(t))

# Somehow alters state?
t + 100
print(torch.rand_like(t))

Outputs:

cpu
tensor([[[0.4733, 0.7733],
         [0.6867, 0.6849],
         [0.9873, 0.9970]],

        [[0.8636, 0.9555],
         [0.6707, 0.1060],
         [0.9323, 0.1317]]])
mps
tensor([[[0., 0.],
         [0., 0.],
         [0., 0.]],

        [[0., 0.],
         [0., 0.],
         [0., 0.]]], device='mps:0')
tensor([[[101., 101.],
         [101., 101.],
         [101., 101.]],

        [[101., 101.],
         [101., 101.],
         [101., 101.]]], device='mps:0')

Tried on 2.2.2 and nightly

Versions

Collecting environment information...
PyTorch version: 2.2.2
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 14.4.1 (arm64)
GCC version: Could not collect
Clang version: 15.0.0 (clang-1500.3.9.4)
CMake version: version 3.29.0
Libc version: N/A

Python version: 3.11.8 (main, Apr 4 2024, 20:29:28) [Clang 15.0.0 (clang-1500.3.9.4)] (64-bit runtime)
Python platform: macOS-14.4.1-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Apple M3 Max

Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] torch==2.2.2
[conda] Could not collect

cc @kulinseth @albanD @malfet @DenisVieriu97 @jhavukainen