Convert aten.bitwise_not to ttnn.bitwise_not

[jdh8]

Ticket

Depends on tenstorrent/tt-metal#12902

Problem description

ttnn.bitwise_not takes an extraneous value: int. I'm fixing this in the kernel.

E       TypeError: __call__(): incompatible function arguments. The following argument types are supported:
E           1. (self: ttnn._ttnn.operations.unary.bitwise_not_t, input_tensor: ttnn._ttnn.tensor.Tensor, value: int, *, memory_config: Optional[ttnn._ttnn.tensor.MemoryConfig] = None, output_tensor: Optional[ttnn._ttnn.tensor.Tensor] = None, queue_id: int = 0) -> ttnn._ttnn.tensor.Tensor
E       
E       Invoked with: <ttnn._ttnn.operations.unary.bitwise_not_t object at 0x7f58fc4d5530>, ttnn.Tensor([[-81.00000, 66.00000,  ...,  0.00000,  0.00000],
E                    [58.00000, -75.00000,  ...,  0.00000,  0.00000],
E                    ...,
E                    [ 0.00000,  0.00000,  ...,  0.00000,  0.00000],
E                    [ 0.00000,  0.00000,  ...,  0.00000,  0.00000]], shape=Shape([4[32], 4[32]]), dtype=DataType::BFLOAT16, layout=Layout::TILE)

../tt-metal/ttnn/ttnn/decorators.py:326: TypeError

What's changed

• Lower aten.bitwise_not to ttnn.bitwise_not
• Fix parameters of ttnn.bitwise_not in the kernel (either of the following)
  • #12901: Remove extraneous value: int parameter from ttnn.bitwise_not tt-metal#12902
  • #13537: Remove extra param in ttnn.bitwise_not tt-metal#13579
• Passing test for the conversion

[jdh8]

Here is a problem. I make this tensor in torch.int32, but the test complains that it gets a BFLOAT16 tensor.

E       RuntimeError: TT_FATAL @ /home/jdh8/tt-metal/ttnn/cpp/ttnn/operations/eltwise/unary/device/unary_device_operation.cpp:42: input_datatype == DataType::INT32
E       info:
E       Unsupported input data type '0' for UnaryOpType '63' (Bitwise operation).
E       backtrace:
E        --- /home/jdh8/tt-metal/ttnn/ttnn/_ttnn.so(+0x4933a9) [0x7f0db03ee3a9]
E        --- ttnn::operations::unary::UnaryDeviceOperation::validate_on_program_cache_miss(ttnn::operations::unary::operation_attributes_t const&, ttnn::operations::unary::tensor_args_t const&)

Previous error message also confirms this.

tenstorrent/tt-metal#12902 does not fix this error.

[ayerofieiev-tt]

@kevinwuTT ptal

[jdh8]

This error still persists after merging tenstorrent/tt-metal#13579.