Convert aten.repeat to ttnn.repeat

[jdh8]

Ticket

Subproblem of tenstorrent/tt-metal#12853

Problem description

This happens when repeating non-aligned innermost dimension. Luckily, this does not happen in our input variations.

E       RuntimeError: TT_FATAL @ /home/runner/work/tt-metal/tt-metal/ttnn/cpp/ttnn/operations/data_movement/repeat/repeat.cpp:37: (input_tensor.get_legacy_shape()[dim] * input_tensor.element_size()) % input_tensor.buffer()->alignment() == 0
E       info:
E       Current repeat implementation requires aligned last dim when repeating on last dim
E       backtrace:
E        --- /home/jdh8/venv/lib/python3.8/site-packages/ttnn/_ttnn.cpython-38-x86_64-linux-gnu.so(+0x4f082a) [0x7f60cb3dd82a]
E        --- /home/jdh8/venv/lib/python3.8/site-packages/ttnn/_ttnn.cpython-38-x86_64-linux-gnu.so(+0x4f15b0) [0x7f60cb3de5b0]
E        --- /home/jdh8/venv/lib/python3.8/site-packages/ttnn/_ttnn.cpython-38-x86_64-linux-gnu.so(+0x4f219f) [0x7f60cb3df19f]
E        --- tt::tt_metal::Device::push_work(std::__1::shared_ptr<std::__1::function<void ()>>, bool)
E        --- ttnn::operations::data_movement::RepeatOperation::invoke(unsigned char, tt::tt_metal::Tensor const&, ttnn::types::Shape const&, std::__1::optional<tt::tt_metal::MemoryConfig> const&)
E        --- /home/jdh8/venv/lib/python3.8/site-packages/ttnn/_ttnn.cpython-38-x86_64-linux-gnu.so(+0x17df968) [0x7f60cc6cc968]
E        --- /home/jdh8/venv/lib/python3.8/site-packages/ttnn/_ttnn.cpython-38-x86_64-linux-gnu.so(+0x17e0753) [0x7f60cc6cd753]
E        --- /home/jdh8/venv/lib/python3.8/site-packages/ttnn/_ttnn.cpython-38-x86_64-linux-gnu.so(+0x17e0dbf) [0x7f60cc6cddbf]
E        --- tt::tt_metal::Device::push_work(std::__1::shared_ptr<std::__1::function<void ()>>, bool)

What's changed

• Restore conversion aten.repeat → ttnn.repeat
• Reenable test for aten.repeat → ttnn.repeat

[ayerofieiev-tt]

@jdh8 can you please review this list?
https://github.com/tenstorrent/pytorch2.0_ttnn/blob/main/docs/cumulative_input_variations.md#atenrepeatdefault

We have this for every op.
I think it is great to have it in lowering tests.
If some test fails - mark it xfailed and add a check in the lowering pass for it to fallback.

Wdyt?

[jdh8]

I just refined the conversion so that our use cases should pass.

[jdh8]

These cases are not in the list of input variations. I leave them for future improvement.

[jdh8]

I'm merging this because it may help #146