-`i:ddlZddlZddlmZddlmZeeZdejd<dejd<dej j _ dZ d e efd Zdd ed efdZddZdZeedrKddlmZddlmZddlmZmZeegej j jd<e e_ee_dSdS)N) init_loggeris_torch_equal1PYTORCH_NVML_BASED_CUDA_CHECKTORCHINDUCTOR_COMPILE_THREADSc ddlmcm ddlm}m}m}m}m}ddl m dttf fd }t jj|r1 jjj | jjjj}n j}|jrt|jd|rm|jdjj|vrT|j|jr4t|jd|r|jdjj|vT|g}g} t/t1|jD]} |j| } t| |r$| |d|j| <Ct| |r||qt| |r'| || |t1|dksJdS)Nr)EnterSubgraphLineExitSubgraphLineMemoryPlanningLineMemoryPlanningStateSubgraphPythonWrapperCodegenVreturncddl}g}|d}|d}|D]}t|jr2|jjdt|It|jr2|jjdt||| |S)Nr_none_shape) itertoolscount isinstanceNoneAsConstantBufferappendgraphnamenextShapeAsConstantBufferget_name) graph_outputsrnames shape_counter none_counternoderirs e/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/vllm/env_override.pyget_output_namesz3memory_plan_reuse_patched..get_output_names-s!**  q)) ! . .D$ 788 .  GG4 3E3EGGHHHHD"":;; .  IID4G4GIIJJJJ T]]__---- )torch._inductor.ir _inductorr%torch._inductor.codegen.wrapperr r r rrtorch._inductor.virtualizedrliststrrr wrapper_codepartition_signatures output_nodesr'linesr$rpoprangelenplanr)selfr r r rrr' out_namesplanning_statespast_planning_statesilinerr%s @@r&memory_plan_reuse_patchedr>"s#########.----- 49        17')EFF/ G 5 A$$ G 5 B  G,,..   tz"~'9 : : JrN  $I 5 5   tz"~'9 : : JrN  $I 5 5 +*,,-O 3tz?? # #??z!} d. / / ? IIob&9::DJqMM / 0 0 ?  " "#6#6#8#8 9 9 9 9 . / / ? ' '(;(;(=(= > > > 3 3 5 5666   1 $ $ $ $ $ $r(skip_cudagraphsc ddlm}ddlm}mmddlmddlm }g}|j } dtdtffd tt!|t!|D]\}} |} |D].} | | j/| |} |jd |D} |fd | j| jzD| z }|fd |D}||D]} | jfd | z D}||fd |D}fd|D}fd|D}| ||fd| D} fd| D}fd|D}||}|||||| |}||||| z }|dddS)z Gets signature for each graph partition, including input nodes, output nodes, and whether deallocating an input within graph partition. r) dependencies)GraphPartitionSignatureMutationOutput NoneLayoutr) OrderedSetbuf_namercj|d}|dSt|jjr?t|jr(j|dx}r |SdSdS)z Checks if buf_name is NoneLayout. Buffers with NoneLayout is not allocated so graph partition should not take it as inputs or outputs. NFT) name_to_bufgetrr$layoutmutation_real_name)rFbuf real_namerCrDis_none_layoutr8s r&rNz=get_graph_partition_signature_patched..is_none_layoutws ""8T22 ;5 choz 2 2 #(N33 1!4884HHH  1&~i0004ur(cg|] }|j S) read_writes).0r$s r& z9get_graph_partition_signature_patched..s 4 4 4$T  4 4 4r(c>g|]}|j|jSrP)r)rRxrNs r&rSz9get_graph_partition_signature_patched..s=)>!&11Fr(c3NK|]}j||V dSNrKrIrRrr8s r& z8get_graph_partition_signature_patched..G+ + 8.s.   |## ###r(c*i|]}|v||SrPrPr]s r& z9get_graph_partition_signature_patched..s5   |## ,t$###r(c"i|] }|v||v SrPrPrRrbuffer_names_to_freer^s r&r`z9get_graph_partition_signature_patched..s6   |## $..###r(c$g|] }|v|v | SrPrPrbs r&rSz9get_graph_partition_signature_patched..s9   |##4H(H(H (H(H(Hr(c3NK|]}j||V dSrWrXrYs r&rZz8get_graph_partition_signature_patched..r[r(c6g|]}| |SrPrP)rRrrNr^s r&rSz9get_graph_partition_signature_patched..s>   !>$''     r(c0g|]}|jjv|SrP)r constants)rRrrs r&rSz9get_graph_partition_signature_patched..s/   dag>O6O6OD6O6O6Or(Nr))torch._inductorrAr*rBrCrDr-rtorch.utils._ordered_setrErr'get_name_to_nodesr/boolzipreversedupdateoutputs_by_namekeys intersection ReadWrites merge_listreadswrites last_usage!get_graph_partition_symbol_inputsrunion)r8 partitionsr?rArBrE signaturesunmet_output_names partitionskip_cudagraph output_namesr$returned_output_namesrQpartition_input_namesextra_input_names input_nodesinput_deallocationextra_output_namesr2constant_names symbol_inputspartition_signaturerCrDrrcrNr^s` @@@@@@r&%get_graph_partition_signature_patchedrfs=-,,,,,VVVVVVVVVV------333333J#AG$<$<$>$>??))++L(&)h77&&c c ! >)3  = =D    4 9 9 ; ; < < < < , 9 9:L M M#-88 4 4) 4 4 4   J(.1CC    !+ + + + + @U+ + + ! ! 1;  9 9D ' ' 8 8 8 8     - <    $$%6777    -        -        -    $$%7888 * + + + + @U+ + + ! !      -       2   >>y+VV 55          -...288 !6 6   ddd r(F should_logrc*ddlmcm}ddlm}m}ddlm}m}m }|j } t| tjjj r| jx} r| } t| tjjr | d| jn| } | tjjjvs| tjjjvr#t| tjjsJdStjjjjs |jdSdt0d|dzd dfd } |r|n| }t||r t3fd |jDS|j J|s|d | dSt|j |jr|d| dSt|j |jr|d| dSt=|j ddr|d| dS||j r|d| dSdS)zBReturn True if we should partition the inductor graph on this noderN)BaseSchedulerNodeFusedSchedulerNode)&_unstable_customized_partition_wrapperis_cudagraph_unsafe_opmaybe_log_cudagraph_partition.Tmsgr$rcdSrWrP)rr$s r&noop_logz*should_partition_patched..noop_log9sr(c3BK|]}|VdSrW)should_partition)rRsnoder8s r&rZz+should_partition_patched..?s1IIE4((//IIIIIIr(z non gpu ops)r$zDeviceCopy opszConditional opsunbacked_bindingszunbacked binding opszCUDAGraph-unsafe custom opsF)r*r+r%torch._inductor.schedulerrrtorch._inductor.utilsrrrr$rtorchFallbackKernel op_overloadr_ops OpOverload _overloadnameconfigcustom_should_partition_opstriton cudagraphswrapperr/anysnodesis_gpu DeviceCopy Conditionalgetattr)r8r$rr%rrrrrir_nodeopop_overload_packet_nameop_overload_namerlog_partition_reasons` r&should_partition_patchedrs$########iG'5?-<==!!#%''))"ej344 )& ; ;)9 ; ; ;(  $%A B B5?#9#UUUb%*"788 8 884 O " ) 4 2 : Btc!2T!9d=GT88H$*++JIIIIT[IIIIII 9 ;;==]6666t$)R]++-D9999t$)R^,,.T::::tty-t443$????tdi((:FFFFt 5r(cddlmcm}ddlm}t |_t|_| dd5||j |_ ddddS#1swxYwYdS)z (Re)initializes the scheduler member. When initializing the scheduler, no CUBIN files should be generated (to avoid biasing any benchmarks and pessimizing fusion decisions). rN) Schedulerztriton.store_cubinF) torch._inductor.configr+rrrrrrget_graph_partition_signaturepatch operations scheduler)r8rrs r&_update_scheduler_patchedr[s,++++++++333333!9I.SI+ *E 2 244"4?33444444444444444444sA!!A%(A%cddlm}|ds |dr/ddl}ttjdrddlm}||_dSdSdS)z;Workaround for TorchInductor autotune get_raw_stream() bug.rr2.9.0z2.9.1N_cuda_getCurrentRawStream)r) vllm.utils.torch_utilsrbuiltinshasattrr_Ctorch._Crget_raw_stream)rr_get_raw_streams r&_patch_get_raw_stream_if_neededrts555555~g6.."9"96 588 9 9 6 M M M M M M&5H # # # 6 6 66r(r)PythonWrapperCodegen) GraphLowering)_Config _ConfigEntry)defaultr)F)rN)osr vllm.loggerrrr__name__loggerenvironr+rcompile_threadsr>r.rlrrrrr,rtorch._inductor.graphrtorch.utils._config_modulerr_configmemory_plan_reuse_update_schedulerrPr(r&rs ######111111 X  /2 *+/2 *+)*&8%8%8%HJ'+DzJJJJDPPTPdPPPPf44442 6 6 6  !!!>' @DDDDDD333333@@@@@@@@EQLEEEO"#@A.G*&?M### @ @r(