Ph3 \ddlmZmZmZmZddlZddlmcm Z gdZ ddejjdeejdeeeeffdZ ddejjdeejdeeeefdefd Z dd eed eedeeeffd Zdeefd ZdZy))AnyDictListOptionalN)compilelist_mode_options list_optionscudagraph_mark_step_begingmexample_inputsoptionsc$ddlm}||||S)au Compile a given FX graph with TorchInductor. This allows compiling FX graphs captured without using TorchDynamo. Args: gm: The FX graph to compile. example_inputs: List of tensor inputs. options: Optional dict of config options. See `torch._inductor.config`. Returns: Callable with same behavior as gm but faster.  compile_fxconfig_patchesr)r r r rs cC:\Users\daisl\Desktop\realtime-object-detection\venv\Lib\site-packages\torch/_inductor/__init__.pyrr s"' b. AAreturncddlm}d}d}t|jjt j jjr|jj}t j jj|j_|j|jj)tj|jj}|jj)tj|jj}|||dni|||d}||||S)aG Ahead-of-time compile a given FX graph with TorchInductor into a shared library. Args: gm: The FX graph to compile. example_inputs: List of tensor inputs. options: Optional dict of config options. See `torch._inductor.config`. Returns: Path to the generated shared library r)compile_fx_aot)zaot_inductor.serialized_in_specz aot_inductor.serialized_out_specr)rr isinstancegraph_codegentorchfx_PyTreeCodeGenCodeGen recompile pytree_infoin_specpytreetreespec_dumpsout_spec)r r r rserialized_in_specserialized_out_speccodegens r aot_compiler*!s +"((##UXX^^%B%BC((##!HHNN224     & & 2!'!6!6w7J7J7R7R!S     ' ' 3"("7"78K8K8T8T"U  ?0B0C   /A0C     rmodedynamicc0iddiddidddd}|r||S|S)aReturns a dictionary describing the optimizations that each of the available modes passed to `torch.compile()` performs. Args: mode (str, optional): The mode to return the optimizations for. If None, returns optimizations for all modes dynamic (bool, optional): Whether dynamic shape is enabled. Example:: >>> torch._inductor.list_mode_options() triton.cudagraphsT max_autotune)r/r.)defaultzreduce-overheadzmax-autotune-no-cudagraphsz max-autotune)r+r, mode_optionss rrrUsG   D' !!% 0L""&< 7<7rc`ddlm}|j}t|j S)zReturns a dictionary describing the optimizations and debug configurations that are available to `torch.compile()`. The options are documented in `torch._inductor.config`. Example:: >>> torch._inductor.list_options() r)config)torch._inductorr4shallow_copy_dictlistkeys)r4current_configs rr r xs)'%+%=%=%?N ##% &&rcddlm}|y)zJIndicates that a new iteration of inference or training is about to begin.rmark_step_beginN)cudagraph_treesr<r;s rr r s 0r)N)NN)typingrrrrtorch.fxrtorch.utils._pytreeutils_pytreer$__all__r GraphModuleTensorstrrr*boolrr r r1rrrHs,,$$ W )-B B&Bd38n %B6)-1 1&1d38n %1 1j;? 8 3- 8)1$ 8 #s(^ 8F'd3i'$r