Migrate to upstream torch::lazy OpKind, Node, Value, and Output

[wonjoo-wj]

[WIP] Migrate to upstream torch::lazy OpKind, Node, Value, and Output.

This PR has gotten a bit big due to torch::lazy::OpKind migration that was required for the IR related classes. The most important changes are in the ir.h file, specifically the new Node, Value and Output classes.

[wonjoo-wj]

Builds succeeds on cloudtop but fails on CircleCI:

FAILED: /tmp/pytorch/xla/build/temp.linux-x86_64-3.7/torch_xla/csrc/ir.o 
clang++-8 -MMD -MF /tmp/pytorch/xla/build/temp.linux-x86_64-3.7/torch_xla/csrc/ir.o.d -Wsign-compare -DNDEBUG -g 
...
/tmp/pytorch/xla/torch_xla/csrc/ir.cpp:104:7: error: no matching constructor for initialization of 'torch::lazy::Node'
    : torch::lazy::Node(op, num_outputs, hash_seed),

Seems like there is a change in PyTorch. Trying to pull the latest in PyTorch and rebuild on cloudtop.

[wonjoo-wj]

Builds successfully but currently CPU and GPU tests failing with:

+ python3 /tmp/pytorch/xla/test/../../test/test_view_ops.py -v TestViewOpsXLA
Traceback (most recent call last):
  File "/tmp/pytorch/xla/test/../../test/test_view_ops.py", line 16, in <module>
    from torch.testing._internal.common_device_type import \
  File "/opt/conda/lib/python3.7/site-packages/torch/testing/_internal/common_device_type.py", line 560, in <module>
    mod = runpy.run_path(path, init_globals=globals())  # type: ignore[func-returns-value]
  File "/opt/conda/lib/python3.7/runpy.py", line 263, in run_path
    pkg_name=pkg_name, script_name=fname)
  File "/opt/conda/lib/python3.7/runpy.py", line 96, in _run_module_code
    mod_name, mod_spec, pkg_name, script_name)
  File "/opt/conda/lib/python3.7/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/tmp/pytorch/xla/test/pytorch_test_base.py", line 8, in <module>
    import torch_xla
  File "/opt/conda/lib/python3.7/site-packages/torch_xla-1.11-py3.7-linux-x86_64.egg/torch_xla/__init__.py", line 102, in <module>
    import _XLAC
ImportError: /opt/conda/lib/python3.7/site-packages/torch_xla-1.11-py3.7-linux-x86_64.egg/_XLAC.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZNK9torch_xla2ir6Output6HasherclERKS1_

[wonjoo-wj]

Can reproduce the issue locally while trying to import torch_xla:

>>> import torch_xla
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/google/home/wonjoo/anaconda3/lib/python3.8/site-packages/torch_xla-1.11-py3.8-linux-x86_64.egg/torch_xla/__init__.py", line 104, in <module>
    import _XLAC
ImportError: /usr/local/google/home/wonjoo/anaconda3/lib/python3.8/site-packages/torch_xla-1.11-py3.8-linux-x86_64.egg/_XLAC.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZNK9torch_xla2ir6Output6HasherclERKS1_

Seems caused by this PR, as running with master branch works fine locally. Looking more into it.

[JackCaoG]

Builds successfully but currently CPU and GPU tests failing with:

+ python3 /tmp/pytorch/xla/test/../../test/test_view_ops.py -v TestViewOpsXLA
Traceback (most recent call last):
  File "/tmp/pytorch/xla/test/../../test/test_view_ops.py", line 16, in <module>
    from torch.testing._internal.common_device_type import \
  File "/opt/conda/lib/python3.7/site-packages/torch/testing/_internal/common_device_type.py", line 560, in <module>
    mod = runpy.run_path(path, init_globals=globals())  # type: ignore[func-returns-value]
  File "/opt/conda/lib/python3.7/runpy.py", line 263, in run_path
    pkg_name=pkg_name, script_name=fname)
  File "/opt/conda/lib/python3.7/runpy.py", line 96, in _run_module_code
    mod_name, mod_spec, pkg_name, script_name)
  File "/opt/conda/lib/python3.7/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/tmp/pytorch/xla/test/pytorch_test_base.py", line 8, in <module>
    import torch_xla
  File "/opt/conda/lib/python3.7/site-packages/torch_xla-1.11-py3.7-linux-x86_64.egg/torch_xla/__init__.py", line 102, in <module>
    import _XLAC
ImportError: /opt/conda/lib/python3.7/site-packages/torch_xla-1.11-py3.7-linux-x86_64.egg/_XLAC.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZNK9torch_xla2ir6Output6HasherclERKS1_

This usually means you have this function defined in the header but it couldn't find the implementation. Search for torch_xla2ir6Output6Hasher

[JackCaoG]

I think we don't need this pr anymore