handle new param from torch.compile (Inductor pattern matcher), enable_log (#100814)
This PR puts a placeholder param handler for a new param being passed in from Inductor, enable log.
Fixes this error below, where I've been unable to run torch.compile on NanoGPT due to this error:
~~~
File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/torch/_inductor/fx_passes/fuse_attention.py", line 219, in _sfdp_init
register_replacement(
File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/torch/_inductor/pattern_matcher.py", line 658, in register_replacement
search_gm = trace_fn(search_fn, example_inputs)
File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/torch/_inductor/pattern_matcher.py", line 828, in training_graph
aot_function(
torch._dynamo.exc.BackendCompilerFailed: backend='compile_fn' raised:
TypeError: patched_aot_function() got an unexpected keyword argument 'enable_log'
~~~
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100814
Approved by: https://github.com/fegin