8hSSKrSSKrSSKJrJr SSKrSSKJs Jr SSK J r SSK J r SSK JrJrJrJrJrJr SSKJr SSKJr SSKJr \R4"\5r"S S \ R:5r"S S \ 5rg) N)AnyCallable)ir)KernelTemplate)Bufferget_free_symbolsget_symbolic_inputsgm_original_output_stridesir_node_to_tensorLayout) benchmarker)do_bench_using_profiling)Vc ^\rSrSrSrS\S\\S\S\S\ S\ 4S S 4 U4S jjr S \4S jr S \\ S\ RS \4SjrS \4SjrS \R&4SjrS \\\ 44SjrS \4SjrSrU=r$)SubgraphChoiceCallerz Represents a Subgraph Autotuning choice, and the subgraph can be any arbitrary GraphModule. Compiles the Subgraph down to a module for benchmarking. name input_nodeslayout description make_fx_graph.returnNc L>[TU]XX45 /Ul[R UR Hn[ [UR5SS95S:Xde[ [UR5SS95S:XdeURR5 URR[U55 M SSS5 U"UR6Ul[UR5 [!UR 5Ulg!,(df  NR=f)NT) unbacked_onlyr)super__init__example_inputsr fake_moderlenrget_size get_stridedata freeze_layoutappendr gmr r sym_inputs)selfrrrrrinp __class__s Z/var/www/fran/franai/venv/lib/python3.13/site-packages/torch/_inductor/codegen/subgraph.pyrSubgraphChoiceCaller.__init__s F@  [[''+CLLN$OPTUUUU+CNN,.>?[s B D D#c"SURS3$)NzSubgraphCaller()rr's r*__str__SubgraphChoiceCaller.__str__8s  1--argsoutc x^^ ^ SSKJs Jn SSKJn U"UR UR [RR[RR[RR[RR[RR[RRSUR3S9 nUR H@nXeR"UR'UR$R'UR5 MB UR Vs/sH?n[)[RR*R,R/U55PMA snm [1T 5S:Xa[3TUR 5Hup[5U[6R85(dM&[5U [6R85(deUR:U R::XdeUR=5U R=5:XaMe [1T 5S:Xa[3TUR 5Hup[5U[6R85(dM&[5U [6R85(deUR:U R::XdeUR=5U R=5:XaMe [R>"U5 URASSSS9 URB"UR 6 URE5n U RFm T "/T QTQ5 SSS5 SSS5 [RH(a[KUU U 4Sj5$[LRN"UU U 4S j5$s snf!,(df  N]=f!,(df  Nf=f) Nr) GraphLowering benchmark_) r%r shape_env cpp_wrapperaot_modeextern_node_serializer is_inference is_backwardrFATEN) max_autotunemax_autotune_gemmmax_autotune_gemm_backendsc>T"/TQTQ5$Nr3bm_funcr&sr*0SubgraphChoiceCaller.benchmark..zsGT"/TQTQ5$rCrDrEsr*rGrH{s9M:9M9M1Nr2)(torch._inductor.config _inductorconfigtorch._inductor.graphr6r%rrgraph _shape_envr9r:r;r<r=rr& graph_inputsgraph_input_namesr$intsizevarsr8 size_hintrzip isinstancetorchTensorshapestrideset_graph_handlerpatchruncompile_to_modulecall/profile_bandwidth_with_do_bench_using_profilingrr benchmark_gpu) r'r4r3inductor_configr6bm_graph_loweringsym_inpsym_varar example_inpmodrFr&s ` @@r* benchmarkSubgraphChoiceCaller.benchmark;s 987)ww..gg((++WW%%#$77#A#A--++dii[)  G;B * *7<< 8  / / 6 6w|| D'  ?? *   **44W= >* z?a $'tT-@-@#Ab%,,//%k5<<@@@@88{'8'888899;+*<*<*>>>> $B z?a $'tT-@-@#Ab%,,//%k5<<@@@@88{'8'888899;+*<*<*>>>> $B !2 3 &&""'+1' "%%t':':;'99;((,*,t,-4  A A+,QR R(()NOOS :4 3s,0AN/N+ANN+ N( $N++ N9c SRURRSS5S/URVs/sHn[ UR 55PM snQURVs/sHn[ UR 55PM snQ[ URR5P5$s snfs snf)N-_r) joinrrsplitrstrr r!r%rN)r'r(s r*hash_keySubgraphChoiceCaller.hash_key}sxx   a(+ 151A1AB1A##clln%1AB 483C3CD3CC#cnn&'3CD DGGMM"    CDs #B7 -#B< c [RR[R"URUR UR URURS95$)N)rrr%r subgraph_name) r TensorBoxcreateSubgraphBufferrrr%rrr/s r* output_node SubgraphChoiceCaller.output_nodesN||""   {{ ,,77#22"ii    r2c SURS.$)zRInformation returned here is logged to the autotune log file when that is enabled.subgraph)backend kernel_namer.r/s r* info_dictSubgraphChoiceCaller.info_dicts"99  r2c SUR3$)N subgraph_r.r/s r*autoheuristic_id%SubgraphChoiceCaller.autoheuristic_ids499+&&r2)rr%r&)__name__ __module__ __qualname____firstlineno____doc__rqlistrr rrrr0rWrXfloatrirrrrvrydictrr__static_attributes__ __classcell__)r)s@r*rrs @@&\@ @  @  S) @ @2..@PtCy@Pu||@P@PD #   R\\   4S> '#''r2rcv\rSrSrSr\R "5rS\S\ S\ 44Sjr S\ \ S\S \ S \4S jrS rg )SubgraphTemplatez A template for subgraph evaluation to be used in autotuning. This class allows creating customized subgraphs that can be appended as choices during the autotuning process, enabling the selection of optimal implementations for complex operations. rr.cVUS[[R53UlX lg)zd Initialize a subgraph template. Args: name: The name of this template graph: The FX graph rmN)nextr index_counterrr)r'rrs r*rSubgraphTemplate.__init__s)fAd#3#A#ABCD *r2rrkwargsrc D[URUUSURS9$)a Generate a SubgraphChoiceCaller instance for autotuning. Args: input_nodes: List of input nodes to the subgraph layout: Memory layout information for the output example_inputs: Example tensor inputs used to trace and benchmark the subgraph **kwargs: Additional keyword arguments Returns: SubgraphChoiceCaller: A callable object that can be used for autotuning )rrrrr)rrr)r'rrrs r*generateSubgraphTemplate.generates*&$#,,   r2)rrN)rrrrr itertoolscountrrqrrrrrr rrrrDr2r*rrsdOO%M + + S) + &\      r2r) rloggingtypingrrrWrJrKrLtorch._inductorrtorch._inductor.codegen.commonrtorch._inductor.irrrr r r r $torch._inductor.runtime.benchmarkingr torch._inductor.utilsrtorch._inductor.virtualizedr getLoggerrlog ChoiceCallerrrrDr2r*rsc ''9=:)!A'2??A'H3 ~3 r2