pytorch - 💡(How to fix) Fix [TMA] Failure when running slice+view operations with TMA enabled on H100 [1 pull requests]

import torch, torch._inductor.config as cfg
cfg.triton.use_tensor_descriptor = True
cfg.assume_aligned_inputs = True

offset = 1

def f(x):
    return x[2:].view(dtype=torch.float32) + 1

x = torch.randn((128 + offset) * 2, dtype=torch.bfloat16, device="cuda")
torch.compile(f)(x); torch.cuda.synchronize()

---

Traceback (most recent call last):
  File "/opt/pytorch/pytorch/tma_test.py", line 19, in <module>
    torch.compile(f)(x); torch.cuda.synchronize()
                         ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/pytorch/torch/cuda/__init__.py", line 1215, in synchronize
    return torch._C._cuda_synchronize()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.AcceleratorError: CUDA error: misaligned address
Search for `cudaErrorMisalignedAddress' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1