NVFuser `FusionComputeAtMultiBCast_CUDA` and `FusionDetectSelfMappedDomains_CUDA` does not raise exception on sm_80+

[malfet]

🐛 Describe the bug

See here for example (only change that test is compiled with sm_86 and run on A10G )
https://github.com/pytorch/pytorch/actions/runs/3222501479/jobs/5272211415

[ RUN      ] NVFuserTest.FusionDetectSelfMappedDomains_CUDA
/home/nshulga/git/pytorch/pytorch/torch/csrc/jit/codegen/cuda/test/test_gpu.cpp:3828: Failure
Expected: tv1->computeAt(tv4, 1) throws an exception.
  Actual: it doesn't.
[  FAILED  ] NVFuserTest.FusionDetectSelfMappedDomains_CUDA (0 ms)

Versions

CI

cc @ngimel @seemethere @pytorch/pytorch-dev-infra

[davidberard98]

Same issue observed when running this internally (fbcode), but on V100.

[malfet]

@davidberard98 I was told that upgrading nvfuser should help.
@jjsjann123 any luck with pushing the update?

[jjsjann123]

Was chatting with David on the upstream PR.

Have PR squashed here: #87779 . Working through CI. Will update once it's green.
Also got a node with Maxwell card. 🤞 Hopefully that will repro the CI failure I saw in my draft.

[ZainRizvi]

Seems like this bug was fixed by your PR @jjsjann123?

[jjsjann123]

since nvfuser is no longer enabled in torchscript. we can safely close this issue.