9iY ddlZddlZddlmZddlZddlmcmcmcm Z ddl mcmcm Z ddl mZddlmZmZddlmZddlmZddlmZddlmZdd lmZmZejj ZGd d ejZd ed edede e!e"efde#eeff dZ$d ed ede e!e"efde#ej%e&zej%e'zfdzfdZ(d ed edefdZ)d ed ede'fdZ*d ede+e'fdZ,d ed ede!fdZ-dede!defdZ.deddfdZ/dZ0e0dej%dej%dej%fdZ1e0dej%dej%dej%fdZ2e0dej%dej%dej%fd Z3d ede4fd!Z5d ed ed"e'defd#Z6dS)$N)Callable)FakeQuantizeBase ObserverBase)_is_activation_post_process)getattr_from_fqn) GraphModule)Node)NSNodeTargetType NSResultsTypeceZdZejZejZejZejZejZ dS)NodeInputOrOutputTypeN) __name__ __module__ __qualname__enumautoFP32INT8FP16UNKNOWN FP32_OR_INT8G/var/www/icac/venv/lib/python3.11/site-packages/torch/ao/ns/fx/utils.pyrrsQ 49;;D 49;;D 49;;DdikkG 49;;LLLrrnodegm logger_clsnode_type_to_io_type_mapreturnc |d}|d}|d}|d}|d}|d} |d} |d} |jd kr|j|vrtjtjfS|j|vrtjtjfS|j|vrtjtjfS|j|vr^t ||d } t| tstd t| t| |||\} }||fStj tj fS|jd kr|jd krtd |jdt|jts$tdt|jt||jtfd| D}t|t t"fs|r^t ||d } t| tstd t| t| |||\} }||fStfd|D}tfd| D}|rtjtjfS|rtjtjfStj tj fS|jdkr|jdkrht ||d }t|tstd t|t||||\} }|tjfS|jdkrt ||d }t|tstd t|t||||\} }t ||d}|t$jurt|d|tjfS|j| vr^t ||d } t| tstd t| t| |||\} }||fStj tj fStj tj fS)Nfuns_io_type_fp32funs_io_type_fp16funs_io_type_int8funs_io_type_fp32_or_int8mods_io_type_fp32mods_io_type_int8mods_io_type_fp32_or_int8meths_io_type_fp32_or_int8 call_functionrExpected Node, got call_modulezExpected call_module, got ''zExpected str, but got c38K|]}t|VdSN isinstance.0 target_typemods r z7get_node_first_input_and_output_type..RsA1 1  sK ( (1 1 1 1 1 1 rc38K|]}t|VdSr/r0r2s rr6z7get_node_first_input_and_output_type..fA) )  sK ( () ) ) ) ) ) rc38K|]}t|VdSr/r0r2s rr6z7get_node_first_input_and_output_type..jr8r call_method dequantizetor z handling needs to be added)optargetrrrrget_normalized_nth_inputr1r AssertionErrortype$get_node_first_input_and_output_typerstrranyrrtorchfloat16)rrrrFUNS_IO_TYPE_FP32FUNS_IO_TYPE_FP16FUNS_IO_TYPE_INT8FUNS_IO_TYPE_FP32_OR_INT8MODS_IO_TYPE_FP32MODS_IO_TYPE_INT8MODS_IO_TYPE_FP32_OR_INT8METHS_IO_TYPE_FP32_OR_INT8 first_arg_prev_node_input_typeprev_node_output_type"is_known_fp32_or_int8_input_moduleis_known_fp32_input_moduleis_known_int8_input_module prev_nodecur_node_dtype_targetr5s @rrBrB&sf11DE01DE01DE 89T U01DE01DE 89T U!9:V!W w/!! ;+ + +).0E0JK K ;+ + +).0E0JK K [- - -).0E0JK K [5 5 50r1==Ii.. N$%L4 ??%L%LMMM52z+C %%*+@A A)13H3PQ Q M ! ! 7m # # !Itw!I!I!IJJ J$+s++ O !M$t{:K:K!M!MNN Nr4;//-01 1 1 1 81 1 1 . . * sZ7GH I I B1 B 1r1==Ii.. N$%L4 ??%L%LMMM52z+C %%*+@A A%() ) ) ) 0) ) ) & & "&)) ) ) ) 0) ) ) & & " & R).0E0JK K ' R).0E0JK K)13H3PQ Q M ! ! ;, & &1r1==Ii.. N$%L4 ??%L%LMMM52z+C %%*+@+EF F [D 1r1==Ii.. N$%L4 ??%L%LMMM52z+C %% %=T2q$I$I !$EM99$,III*+@+EF F [6 6 60r1==Ii.. N$%L4 ??%L%LMMM52z+C %%*+@A A%-/D/LMM%-/D/LMMrct||d}t|tsdS|d}d}|jdkrf|jt jur|||ddS|jtjtj tj tj fvr|||ddSdS|jd krt|jts$td t|jt||jtt jt jt jt(jt jt jt jt jt jt jt jt jt jt jt j t j!t j"t j#t(j$t(j%t(j&t(jt(j't(j(frj)j*fStWfd |D}|rtY|||SdS) z{ Returns the qparams (scale, zero_point) of the first input to `node`, if they can be inferred from the graph. rNr(cjt|||}t|||}t|tstdt |t|jt s$tdt |jt|tstdt |t|jt s$tdt |jt||j}t||j}||fS)Nr+Expected str, got )r?r1r r@rAr>rCr)rr scale_arg_idx zp_arg_idx scale_nodezp_node scale_objzp_objs r _get_scale_zp_from_function_argsz@get_node_input_qparams.._get_scale_zp_from_function_argss"-dB FF *4Z@@*d++ K !ItJ7G7G!I!IJJ J*+S11 Q !Od:;L6M6M!O!OPP P'4(( H !FtG}}!F!FGG G'.#.. N !Ld7>6J6J!L!LMM M$R):;; !"gn556""rr*r r,rYc38K|]}t|VdSr/r0)r3r4 module_objs rr6z)get_node_input_qparams..sA1 1  z; / /1 1 1 1 1 1 r)-r?r1r r=r>rEquantize_per_tensortoqaddadd_relumulmul_relurCr@rArnnqLinearConv1dConv2dnniq ConvReLU2dConv3d BatchNorm2d BatchNorm3dConvTranspose1dConvTranspose2dELU GroupNormInstanceNorm1dInstanceNorm2dInstanceNorm3d LayerNorm Hardswish LeakyReLUReLU6BNReLU2dBNReLU3d ConvReLU1d ConvReLU3d LinearReLUscale zero_pointrDget_node_input_qparams)rrrrUrMr`rRrds @rrrs)r155I i & &t 89T U # # #|&&  u8 8 833Ir1aHH H  #'3<#,!O O O33Ir1aHH Ht  & &)*C00 P !Nd9;K6L6L!N!NOO O%b)*:;;      ## """      1    =:$j&;< <-01 1 1 1 81 1 1 . . * . S))R9QRR R 4rcx|jdkrt||j}t|rt |jdkr$t dt |jt|jdts*t dt|jd|jd}t|jts$t dt|jt||j}t|rt |jdkr$t dt |jt|jdts*t dt|jd|jd}|S)a If node is not an observer, returns it. If node is an observer, navigates up the graph and returns the first parent which is not an observer. For example, graph: (node_non_obs), node = node_non_obs : returns node_non_obs graph: (node_non_obs -> obs0), node = obs0 : returns node_non_obs graph: (node_non_obs -> obs0 -> fq0), node = fq0 : returns node_non_obs r,r z)Expected node.args to have length 1, got rr+rY) r=rr>rlenargsr@r1r rArCrrnode_objs rreturn_first_non_observer_noders w-#B 44 &x 0 0 $49~~""$PDIPPdilD11 Q$%O4 ! ;M;M%O%OPPP9Q>Q&&(TC NNTT"$)A,55U()StDIaL?Q?Q)S)STTTy| Krc~|jdkr1t||j}t|tjrdSdS)aO Assumes that all non-param args occur first. Returns the number of non-param args expected for a node. For example, for F.linear(x, weight, bias) Returns 1, because x is a non-param arg and weight and bias are params. For lstm_mod(x, hid) Returns 2, because both x and hid are non-param args. r,rar )r=rr>r1nnLSTMrs rget_number_of_non_param_argsr+sB" w-#B 44 h ( ( 1 1rc^tjdkrgSjdkrjtjtjjjtjfvs4jtj tjjj tj fvrfdtdD}|SdgS)a- Returns the indices of args of the node which we should attach loggers to, if input logging is enabled. For example, * for (x + y), returns [0, 1] * for (1 + y), returns [1] * for (x + 1), returns [0] * for (linear(x, w, b)) returns [0] * by default, returns [0] rr*cVg|]%}tj|tu#|&Sr)rArr )r3irs r z4get_arg_indices_of_inputs_to_log..Xs1DDDdil););t)C)C!)C)C)Crra) rrr=r>rErgops quantizedoperatorrirange)rresults` r get_arg_indices_of_inputs_to_logrEs 49~~  w/!!  59#6#:HLIII ;59ei&9&=x|L L LDDDDU1XXDDD 3Jrc4d}|jdvrtj|j}nr|jdkrgt |jt s$t dt|jt||j}tj|}|S)z Returns a string representation of the type of the function or module pointed to by this node, or '' for other node types. )r*r:r,rY) r=rEtypenamer>r1rCr@rAr)rrr4 target_mods rget_target_type_strr]s K w222nT[11 M ! !$+s++ K !Id4;6G6G!I!IJJ J%b$+66 nZ00 rresults model_namec,i}|D]|\}}d}|D]S}|D]<\}}||kr1t|dkrtd|dd}<=T||||<w|||<}|S)a  Rekeys the layer name of a results dictionary to use node names from `model_name`. For example, transforms {'base_op_1_0': {'node_output': {'model_a': [{'ref_node_name': 'linear1', ...}]}}} into {'linear1': {'node_output': {'model_a': [{'ref_node_name': 'linear1', ...}]}}} Note: we cannot use these node names directly because they are not guaranteed to be consistent across models. This is why we extract the results first and rekey afterwards. Nrz(Expected list_of_results to be not empty ref_node_name)itemsvaluesrr@) rr new_resultsold_layer_nameresult_type_to_resultsnew_layer_namemodel_name_to_resultscur_model_namelist_of_resultss r'rekey_logger_info_on_node_name_of_modelrms,K29--// A A..%;%B%B%D%D   !3H3N3N3P3P  /!Z//?++q00,-WXXX%4Q%7%HNN    %*@K ' '*@K ' ' rc d}|D]X}|D]A}|D]*\}}t|dkr|dd|}n+|r|D]|}|D]c}||}|D]D\}}||kr tt|D]}||d}|||d<Ed{dSdS)ay If `fqn` entries are filled in for one of the models in `results`, copies them over to any models which do not have them filled out. A common use case benefitting from this is comparing a model prepared by quantization to a quantized model. In this case, the model prepared by quantization would have `fqn` entries, and the quantized model would not. Nrfqn)rrrr) rmodel_name_with_fqnsrrr model_resultsref_model_resultsrrs rmaybe_add_missing_fqnsrsq ").."2"2%;%B%B%D%D   !-B-H-H-J-J  ) M}%%))$Q'.:/9,   6&-nn&6&6 6 6 ")?)F)F)H)H 6 6%$9:N$O!1F1L1L1N1N66-J !%999 "3}#5#56666/25925 a(//6 6 6 6 6 6 6rcfdS)Ncx|^}}}t|trt|ts*t|trQt|tr.innersSB r5 ! ! %jU&;&; % r4  %%/D%9%9 %GBKK ; ;S/w//uuh9&99::::N EL ) ) %jU\.J.J % %]]__ %]]__ 8u{ " "bh%+&=&=4%W%%q(%f%%%rr)rrs`@r8maybe_dequantize_first_two_tensor_args_and_handle_tuplesrs)&&&&&&2 Lrxyctj|}tj||z }dtj||z zS)z Computes the SQNR between `x` and `y`. Args: x: Tensor or tuple of tensors y: Tensor or tuple of tensors Return: float or tuple of floats )rEnormlog10)rrPsPns r compute_sqnrrs= AB AE  B  BG$$ $$rctj||z dz|dzz S)z Computes the normalized L2 error between `x` and `y`. Args: x: Tensor or tuple of tensors y: Tensor or tuple of tensors Return: float or tuple of floats ra)rEsqrtsumrrs rcompute_normalized_l2_errorrs9 :A!|((**adZZ\\9 : ::rc|dd}|dd}tjj||S)z Computes the cosine similarity between `x` and `y`. Args: x: Tensor or tuple of tensors y: Tensor or tuple of tensors Return: float or tuple of floats r )reshaperEr functionalcosine_similarityrs rcompute_cosine_similarityrsE !RA !RA 8  0 0A 6 66rc|jdkrM|jtjtjt jt jtjtjfvrdSdS)Nr*FT)r=r>rErgrircatstack)rs rop_type_supports_shadowingr sM w/!! ; I I L L I K    5 4ridxc  ||d}||\}}t|t|z|kr2td|dt|t|z|t|kr||St||St|jt|jz|kr>1 DK..0011*==  : : : ty>>C ,, , 3 3 XXXS^^c$+FVFV5VXX  TY  9S> ! ! !s49~~-J **,,--j9 9 9 9 :s.B F&F7B FAFBI)#AI)(I))7rrcollections.abcrrEtorch.ao.nn.intrinsic.quantizedaor intrinsicrrotorch.ao.nn.quantizedrktorch.nntorch.ao.quantizationrrtorch.ao.quantization.observerrtorch.ao.quantization.utilsrtorch.fxrtorch.fx.graphr ns_typesr r rrfEnumrdictrCsetrrBrrintrrrrrrrrrrrrboolrr?rrrrsC $$$$$$ ...............############@@@@@@@@FFFFFF888888 55555555 i     DI   DN DNDNDN#3,<(=#=> DN  "7 78 DNDNDNDNNU UU#3,<(=#=>U 5<% !3 34t; UUUUp# ## ####L        44DI0 d       % %%%%%%P 6M 6d 6 6 6 6F:: %EL %U\ %el % % %:9 % : ;5< ;EL ;U\ ; ; ;:9 ;:77%,75<777:97& T d    *:4*:[*:s*:t*:*:*:*:*:*:r