g^(h6ddlZddlZddlZddlZddlZddlmZddlmZddl m Z m Z ddl Z ddl Z ddlmZddlmZddlmcmZddl mZddlmZddlmZdd lmZmZmZdd lm Z dd l!m"Z"m#Z#m$Z$ejJe&Z'd Z(ed Z)edejTde fdZ+d dZ,dZ-edejTde fdZ.Gddej^Z0edejTde fdZ1edZ2dZ3e jhjjZ5e5jle5jne5jpe5jre5jte5jve5jxe5jze5j|e5j~e5je5je5je5jje5jje5je5je5je5je5je5je5jhZMeeMZMedZNde e ejffdZPdZQdaRdZSdZTd!dZUy)"N)contextmanager)partial)CallableUnion)SymInt)get_decompositions) bind_symbols) aot_function aot_modulemake_boxed_compiler)strip_overloads)default_partition draw_graph#min_cut_rematerialization_partitionc|jjdtjjj D]+}tjjj |_-|j|S)N call_functionoptarget) graph find_nodestorchopsaten_to_copytor recompile)fx_gnodes X/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/torch/_functorch/compilers.py _canonicalizer"$s` %% 599>>#:#:&(iinn'' ( NN Kc#Ktjjd} dtjj|y#tjj|wxYww)NF)r_C_jit_set_autocast_mode)old_jit_autocast_flags r!_disable_jit_autocastr(-sI!HH;;EB?  ''(=>''(=>s A+A A+!A((A+rreturncDt5t||jjdtj j jD]l}t|jdk(st|jdk(s5d|jvsDtj j j|_ n|jjD]X}i}|jjD]0\}}t|tj r |j"}|||<2||_ Z|jj%|j'tj(j+|}tj,j/|jtj(j1|j3}tj(j5|}t7d|Ds||ddd|S#1swYSxYw)a Compiles the :attr:`fx_g` with Torchscript compiler. .. warning:: This API is experimental and likely to change. Args: fx_g(fx.GraphModule): The input Fx graph module to be compiled. Returns: Torch scripted model. rrr dtypec3dK|](}t|tjj*ywN) isinstancer _subclasses FakeTensor).0ts r! zts_compile..`s#M1:a!2!2!=!=>Ms.0N)r(rrrrrrrlenargskwargsrrnodesitemsr.devicetypelintrjitscriptr%_jit_pass_remove_mutationfreezeevaloptimize_for_inferenceany)rinpsr new_kwargskvfs r! ts_compilerH6s  JJ))uyy~~'>'>*  0D499~"s4;;'71'<DKKAW#iinn//  0 JJ$$ %DJ ))+ "1a.A ! 1  "%DK  %   II  T " **1773 II  QVVX & II , ,Q /MMM tH9: H;: HsA&H2H HE1HHcLt|jt||||S)N) clear_meta)printcoder)r_namerJs r!_draw_graph_compilerOes $))tTj1 Kr#c6ttt|S)NrN)r rrOrQs r!draw_graph_compilerRks w':F GGr#c|S)z Returns the :attr:`fx_g` Fx graph module as it is. This is a no-op compiler and can be used to check accuracy. .. warning:: This API is experimental and likely to change. rrMs r!noprVos  Kr#c(eZdZfdZfdZxZS)DebugInterpretercTt|jg||_t||yr-)r modulesymbol_mappingsuperrun)selfr5 __class__s r!r]zDebugInterpreter.run}s%*4;;>>  Tr#c fd fd fd fd}t|}d|jvrtj|jd\}}tj|\}}t |t |k(sJt |dt |t tt |||D]/\ }} t| tjs"||| fd1|S)Nct|ts|Stj|jj j j}|jsJ|t|Sr-) r.rsympyexpandr exprxreplacer[ is_numberint)nirr^s r! subst_symintz/DebugInterpreter.run_node..subst_symintsSb&)  RWW\\2243F3FGHA;; ! !;q6Mr#c,tfd|DS)Nc3.K|] }|ywr-rT)r1rhrjs r!r3zHDebugInterpreter.run_node..subst_symint_tuple..s8bb)8s)tuple)nisrjs r!subst_symint_tuplez5DebugInterpreter.run_node..subst_symint_tuples8C88 8r#c|jdkDr`t|jD]H}|j||j|k7s-|j |dkDsHyy)Nrr FT)numelrangendimstridesize)abidxrjs r!check_significant_stridesz.check_significant_stridessgAGGI&* =%C$QXXc]3qxx}D(59$ % r#c "t|sJ|j|jk(s(J|d|jd|j|j|jk(sGJ|d|jd|jd|j||}|sGJ|d|jd|jd|jy)Nz:  != z aka )callabler+rurt)nvrvdesc same_stridesryros r!checkz(DebugInterpreter.run_node..checksD> !>88rxx' NDF82bhhZtBHH:)N N'"2779-: [&BGGI;e,>rwwy,I+J$rwwykZ [:4Rz+DebugInterpreter.run_node..ss'$:M:M9N&Or#) r\run_nodemetapytree tree_flattenr4ziprrr.rTensor)r^nrrin_vals_n_specr_vals_r_specr}r~ryrrjror_s` @@@@r!rzDebugInterpreter.run_nodes  9  a G Q  AFF?$11!&&-@OFG$11!4OFGv;#f+- P#f+d3v;-/P P- s6{!3VVD Q 2r!"ell3b"OP Qr#)__name__ __module__ __qualname__r]r __classcell__)r_s@r!rXrX|s//r#rXc,t|jS)z Returns a (slow) interpreter over the FX graph module that also checks various debugging properties (e.g., that tracing strides matched real strides.) )rXr]rUs r! debug_noprs D ! % %%r#ct|tjj|}tjj |j }|Sr-)rrr<r=r?r@)rrMrGs r!simple_ts_compilers=D A "A Hr#c"t|tSr-)r r)rGs r!nnc_jitrs , --r#c0t|j|Sr-)rKrLrUs r! print_compilers $)) Kr#fnc ttttd}|j|t |t j jr t|fi|St|fi|S)a Wrapper function over :func:`aot_function` and :func:`aot_module` to perform memory efficient fusion. It uses the :func:`min_cut_rematerialization_partition` partitioner to perform efficient recomputation. It uses NVFuser to compile the generated forward and backward graphs. .. warning:: This API is experimental and likely to change. Args: fn (Union[Callable, nn.Module]): A Python function or a ``nn.Module`` that takes one ore more arguments. Must return one or more Tensors. **kwargs: Any other overrides you want to make to the settings Returns: Returns a ``Callable`` or ``nn.Module`` that retains the eager behavior of the original :attr:`fn`, but whose forward and backward graphs have gone through recomputation optimizations, and the graphs have been compiled with nvfuser.  fw_compiler bw_compiler partition_fndecompositions) rHrdefault_decompositionsupdater.rnnModuler r )rr6configs r!memory_efficient_fusionrsW6"!;0 F  MM&"ehhoo&"'''B)&))r#c|jdtd|Dcgc]}|j|jfc}dddlm}|j |t||Scc}w)NfooaQ ############################################################## # To minimize FX graph, copy and paste the below and run it # ############################################################## import torch import torch.fx as fx from functorch.compile import minifier, check_nvfuser_subprocess, check_nvfuser_correctness_subprocess inps = a? inps = [torch.ones(shape, dtype=dtype, device='cuda') for (shape, dtype) in inps] from foo import FxModule mod = FxModule().cuda() with torch.jit.fuser("fuser2"): # check_nvfuser_subprocess can be replaced with check_nvfuser_correctness_subprocess minifier(fx.symbolic_trace(mod), inps, check_nvfuser_subprocess) r)FxModule) to_folderrKshaper+rrcudarH)rrCrrs r! debug_compilersoNN5  &**!''177 *+, (HJOOt dD !! +sA/ c Jg}t|d5}tj|}g}|D]}t|dk(r|}|t j }n|\}}}} } | t jt jt jt jt jt jtthvrt jdd|| | }nt j || | }|j| ddd|S#1swY|SxYw)zZ Return a random input for the given inputs meta generated from _save_fx_default. rbr r)r+r9N)openpickleloadr4randomrandrrgint32int64booluint8floatrandintappend) input_data_pathinputsrG inputs_metarr:inputr_strider+r9s r! get_inputsr3sF ot $!kk!n  !D4yA~V[[]+6:3eWeVIIKKKKJJIIKK  "MM!QU6RE!JJuE&IE MM% ' !!. M/!. Ms C>DD"ct ddlm} fd  fd fd} fd} fd}||||||tS) aO The forward, backward, and joint computation graph will be stored in {folder_name}/{current_name}/{current_name}_forward_{graph_index}, {folder_name}/{current_name}/{current_name}_backward_{graph_index}, and {folder_name}/{current_name}/{current_name}_joint_{graph_index} respectively. The input shape of the graphs will be stored in the .input files. These files can be loaded with pickle, and is a list of format (type, shape, stride, dtype, device). In the case of type = int or float, it is just (type,). For joint graph input, it is a nested list [[],[]] where the two inner lists have the same format. If dump_example_input is True, example_inputs will be stored in .pt file. Since each function might produce multiple graphs, the graph_index is used to distinguish difference graphs r)aot_module_simplifiedcg}t|dkDr1t|dtr||dz }||dz }|S|D]}t|tk(st|t k(r|j t|fC|j t||j|j|j|jf|S)Nrr ) r4r.rmr:rgrrrrtr+r9)r5 input_metaargget_input_metas r!rz(_save_fx_default..get_input_metads t9q=ZQ7 .a1 1J .a1 1J  CCyC49#5!!49,/!!#Y 3::<CJJO  r#ct|jjdk(r,tjtj d|t ytj|}|jjtjjj|j|}tjdd|j!ddd|dt t#j$|t'ddd|dt dd|dt ddr7tj(|ddd|dt dd|dt d yy) Nrz!No nodes in graph {%s}_{%s}_{%s}./T)exist_okrMz.inputwbz.pt)r4rr7logloggingWARNING graph_indexcopydeepcopy set_codegenrfxCodeGenrosmakedirsrrdumprsave) gm_to_saver5 type_namegmr current_namedump_example_input folder_namers r!graph_saver_helperz,_save_fx_default..graph_saver_helperss z%% &! + GG3    ]]: & UXX^^3356 #D)  {m1\N3dC m1\N!L>9+Q{m T    -qa ~Qyk;-WXYeXffghqgrrsttAAGH    JJ-qa ~Qyk;-WXYeXffghqgrrsttAADE  r#c||d|S)NforwardrT)rfw_argsrs r!graph_saver_forwardz-_save_fx_default..graph_saver_forwards2w 2 r#c.||dtdz a|S)Nbackwardr )r)rbw_argsrs r!graph_saver_backwardz._save_fx_default..graph_saver_backwards2w 3q  r#c0||dt||S)Njoint)r)r joint_argsrs r!graph_saver_jointz+_save_fx_default..graph_saver_joints2z73 Z00r#r)functorch.compilerr) rrrrexample_inputsrrrrrrs ``` @@r!_save_fx_defaultrRsC 8 !F 1 ! '(&-  r#c*datt|||S)as Dump the forward, backward, and joint computation graph. Example Usage: save_fx_func = graph_dumper_aot(current_name, folder_name, dump_example_input = False) optimize_ctx = torchdynamo.optimize( save_fx_func ) with torch.enable_grad(): with optimize_ctx: result = forward_and_backward_pass(model, example_inputs) r)rrr)rrrs r!graph_dumper_aotrsK #\;@R SSr#)T)F)Vrrrrr contextlibr functoolsrtypingrrrbrtorch.fxrtorch.nnrtorch.utils._pytreeutils_pytreerr torch._decompr%torch.fx.experimental.symbolic_shapesr aot_autogradr r r compile_utilsr partitionersrrr getLoggerrrr"r( GraphModulerHrOrRrV InterpreterrXrrrrrdetach gelu_backwardleaky_relu_backwardsigmoid_backwardthreshold_backwardhardtanh_backwardhardsigmoid_backwardhardswish_backward tanh_backward silu_backward elu_backwardcudnn_batch_normcudnn_batch_norm_backward masked_fillScalarrelu leaky_reluhardtanh hardswish hardsigmoid conj_physical is_same_sizerrrrrrrrrrTr#r!rs]  %" $$,>GG*g! ??+ R^^+ h+ + \ H bnn H  4r~~4n&BNN&(&&  . yy~~KK""HHOOMMNN-2,,BC $* h !"$*N": >YzTr#