[bugfix] [Relay] fix broadcast in PyTorch frontend

[jikechao]

broadcast_shapes is a newly introduced operator after PyTorch 1.8.0. If you install PyTorch < 1.8.0, the usage torch.broadcast_shapes will lead to a crash and throw: AttributeError: module 'torch' has no attribute 'broadcast_shapes'.

This PR replaces broadcast_shapes with an old operator broadcast_tensors, which was introduced since PyTorch 1.0.0.

Reference: PyTorch Documentation

Reproducible script

import torch
from tvm import relay
import tvm

m = torch.nn.L1Loss()
input_shape_ = [10, 3]
input_data=[torch.randn(input_shape_, dtype=torch.float64), torch.randn(input_shape_, dtype=torch.float64)]
trace = torch.jit.trace(m, input_data)

#input_shapes = [('input0', torch.Size(input_shape))]
input_shapes = [('input0', torch.Size(input_shape_)), ('input1', torch.Size(input_shape_))]
mod, params = relay.frontend.from_pytorch(trace, input_shapes)

with tvm.transform.PassContext(opt_level=3):
    exe = relay.create_executor('vm', mod=mod, params=params, device=tvm.device('llvm', 0), target='llvm').evaluate()

BTW, this bug can be triggered by existing test cases. Thus, we needn't add new test cases.

cc @echuraev @vvchernov @yongwww

[tvm-bot]

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

• cc @shingjan _{See #10317 for details}

_{Generated by tvm-bot}

[echuraev]

Could you please add a unit test?

[jikechao]

@echuraev Thanks for your review. Test cases have been added. These test cases can cover the bugging line. However only when the PyTorch version < 1.8.0 in your environments, these test cases can trigger this crash.

[jikechao]

@shingjan @echuraev Thanks for your scrutiny. These redundant test cases have been removed.

[echuraev]

LGTM. Thanks