Arm backend: Add FuseViewCopyTransform and FuseConstantsPass in arm_pass_manager

[AdrianLundell]

These passes both removes redundant ops from the graph:

• FuseViewCopyTransform pass is added from backends/transforms to merge sequential view ops.
• FuseConstantOpsPass is created to compute ops with constant inputs AOT.
  • This is not done in cases where the result is a larger tensor, to avoid increasing the constant memory size.
  • For BI, ops are quantized with the q/dq-ops as to not change the behaviour of the graph.
  • Pass order is important: the pass must be placed after all passes which may add constant ops, but before the
    InsertTableOpsPass, since it doesn't handle TOSA _table-ops.

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/8997

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 9d82c07 with merge base d3cca89 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[kirklandsign]

Sorry @zingo would you mind if we revert and land this again from internal? This is breaking our internal tests.

cc @zonglinpeng do you think a forward fix is possible?

[zonglinpeng]

Sorry @zingo would you mind if we revert and land this again from internal? This is breaking our internal tests.

cc @zonglinpeng do you think a forward fix is possible?

I'm not familiar with arm passes. But based on the error message ModuleNotFoundError: No module named 'executorch.backends.transforms.fuse_view_copy, fuse_view_copy needs to added or change reference as the fix.

[zingo]

Of course, if there is a problem let's revert so it works again, and we can figure out the problem much calmer. Unfortunately I'm not at my desk right now so I cannot do it right now. Feel free to go ahead if you can. 🙏

[zingo]

Created #9070 if still needed
I cant self review it so please feel free to review and merge if this is still an issue.

[kirklandsign]

Hi @zingo

We have the following internal error log:

        debug_mode = "ALL" if logger.level <= logging.DEBUG else None
>       outputs_np, status = tosa_reference_model.run(
            graph,
            inputs_np,
            verbosity=_tosa_refmodel_loglevel(logger.level),
            tosa_profile=tosa_profile,
            initialize_variable_tensor_from_numpy=1,  # True
            debug_mode=debug_mode,
        )
E       AttributeError: 'NoneType' object has no attribute 'run'

Do you have idea?

[kirklandsign]

Hi @zingo

Seems that all 3 models in this test https://github.com/pytorch/executorch/blob/main/backends/arm/test/passes/test_fuse_constant_ops_pass.py#L110-L115 failed

[AdrianLundell]

Hi @kirklandsign , can you give more context here:

1. The first error indicates that the tosa_reference_model is none, but it's impossible to say why without knowing how it was created.
2. Where is the unittest failing, as they passed when running on the PR?
   Thanks!

[kirklandsign]

Hi @AdrianLundell

Sorry I don't have the full context about how the CI is run on our internal CI. Maybe @digantdesai will give more context.

I'm not sure why tosa_reference_model is none here.