o I h(@sUddlZddlmZmZmZmZddlmmm Z ddl mZddl m Z mZmZgdZddZddZd d Zd d Zd dZiejejfeejejejfeejejfeejejejfeejejfeejejejfeejejfee jejejfee jejejfee jej ejfeej ejfee j!ejejfee j"ejejfee j#ej$ejfeej%ejfeej&ejfeZ'e(e)eej*effe+d<dddZ,ddZ-ddZ.ddZ/dede(eeej*efffddZ0dS)N)AnyCallableOptionalUnion)get_combined_dict MatchAllNodePattern) fuse_conv_bnfuse_conv_bn_relufuse_linear_bnfuse_convtranspose_bnget_fuser_methodget_fuser_method_newcCs|j|jks Jdtjtjtjtjtjtji}|rL|j |j ks$Jd|j s+Jd|j s2Jd| t|d}|durC|||Std||ftj||S)aReturn the fused the conv and bn modules. Given the conv and bn modules, fuses them and returns the fused module Args: is_qat: a flag for whether we are using quantization aware training fusion or post training quantization fusion conv: Module instance of type conv2d/conv3d bn: Spatial BN instance that needs to be fused with the conv Examples:: >>> m1 = nn.Conv2d(10, 20, 3) >>> b1 = nn.BatchNorm2d(20) >>> # xdoctest: +SKIP >>> m2 = fuse_conv_bn(m1, b1) :Conv and BN both must be in the same mode (train or eval).z?Output channel of Conv2d must match num_features of BatchNorm2dz7Only support fusing BatchNorm2d with affine set to TruezGOnly support fusing BatchNorm2d with tracking_running_stats set to TrueNCannot fuse train modules: )trainingnnConv1dnniConvBn1dConv2dConvBn2dConv3dConvBn3d num_features out_channelsaffinetrack_running_statsgettypeNotImplementedErrorutilsfuse_conv_bn_eval)is_qatconvbnfused_module_class_mapfused_module_classr(_/var/www/vscode/kcb/lib/python3.10/site-packages/torch/ao/quantization/fuser_method_mappings.pyr s* r cCs|j|jkr|jksJdJdd}|r[tjtjtjtjtjtji}|j |j ks1Jd|j s8Jd|j s?Jd| t|d}|durQ||||Std|||ftjtjtjtjtjtji}| t|d}|durtjj||}|||Std|||f)aJReturn the fused conv and bv modules. Given the conv and bn modules, fuses them and returns the fused module Args: is_qat: a flag for whether we are using quantization aware training fusion or post training quantization fusion conv: Module instance of type conv2d/conv3d bn: Spatial BN instance that needs to be fused with the conv Examples:: >>> m1 = nn.Conv2d(10, 20, 3) >>> b1 = nn.BatchNorm2d(20) >>> r1 = nn.ReLU(inplace=False) >>> # xdoctest: +SKIP >>> m2 = fuse_conv_bn_relu(m1, b1, r1) rNz;Output channel of Conv must match num_features of BatchNormz5Only support fusing BatchNorm with affine set to TruezEOnly support fusing BatchNorm with tracking_running_stats set to TruerzCannot fuse eval modules: )rrrr ConvBnReLU1dr ConvBnReLU2dr ConvBnReLU3drrrrrrr ConvReLU1d ConvReLU2d ConvReLU3dr!fusionr")r#r$r%relu fused_modulemap_to_fused_module_trainmap_to_fused_module_eval fused_convr(r(r)r @s@  r cCsd|j|jks Jd|r*|j|jksJd|jsJd|js$Jdt||Stjj ||S)aReturn the fused linear and bn modules. Given the linear and bn modules, fuses them and returns the fused module Args: is_qat: a flag for whether we are using quantization aware training fusion or post training quantization fusion linear: Module instance of type Linear bn: BatchNorm1d instance that needs to be fused with the linear layer Examples:: >>> m1 = nn.Linear(20, 10) >>> b1 = nn.BatchNorm1d(10) >>> # xdoctest: +SKIP >>> m2 = fuse_linear_bn(m1, b1) z>> m1 = nn.ConvTranspose2d(10, 20, 3) >>> b1 = nn.BatchNorm2d(20) >>> # xdoctest: +SKIP >>> m2 = fuse_convtranspose_bn(m1, b1) zCConvTranspose and BN both must be in the same mode (train or eval).z8Fusing ConvTranspose+BatchNorm not yet supported in QAT.T) transpose)r Exceptionrr!r0r")r#convtr%r(r(r)r sr cfdd}|S)a!Return a sequential wrapped that for is_qat and two modules. Given a sequential class for two modules, return a function that takes is_qat, and then two modules as argument, that ignores the is_qat flag and always returns the sequential that combines the two input modules cs ||SNr()r#m1m2 sequentialr(r) fuser_methods z*_sequential_wrapper2..fuser_methodr()rBrCr(rAr)_sequential_wrapper2s rD _DEFAULT_OP_LIST_TO_FUSER_METHODcCs>|duri}tt|}||d}|dusJd|d|S)zGet fuser method for the given list of module types. Get fuser method for the given list of module types, return None if fuser method does not exist Ndid not find fuser method for:  )rrEr)op_listadditional_fuser_method_mapping all_mappingsrCr(r(r)r s r cr=)Ncs |||Sr>r()r#xyfr(r)reverseds z_reverse2..reversedr(rNrOr(rMr) _reverse2s rQcr=)Ncs|\}}||||Sr>r()r#rKwrLzrMr(r)rOsz_reverse3..reversedr(rPr(rMr) _reverse3s rTcCs:t|ttfrdd|D}ttj|}|S|tg}|S)aQReturn a list of valid patterns generated from the op_pattern. Returns a list of valid patterns generated from the op_pattern, since MatchAllNode can match all types of nodes, e.g. pattern (torch.nn.Conv2d, torch.add) should also be able to match keys like (MatchAllNode, torch.add) and (torch.nn.Conv2d, MatchAllNode) Example Input: (torch.add, (torch.nn.ReLU, torch.nn.Conv2d)) Example Output: [(torch.add, (torch.nn.ReLU, torch.nn.Conv2d)), (torch.add, (torch.nn.ReLU, MatchAllNode)), (torch.add, (MatchAllNode, torch.nn.Conv2d)), (torch.add, (MatchAllNode, MatchAllNode)), (MatchAllNode, (torch.nn.ReLU, torch.nn.Conv2d)), (MatchAllNode, (torch.nn.ReLU, MatchAllNode)), (MatchAllNode, (MatchAllNode, torch.nn.Conv2d)), (MatchAllNode, (MatchAllNode, MatchAllNode)), ] cSsg|]}t|qSr()_get_valid_patterns).0 sub_patternr(r(r) sz'_get_valid_patterns..) isinstancetuplelist itertoolsproductr) op_pattern sub_combsresultr(r(r)rUs rUr^fuser_method_mappingcCsJt|}d}|D]}||d}|durnq|dus#Jd|d|S)zGet fuser method. This will be made default after we deprecate the get_fuser_method Would like to implement this first and have a separate PR for deprecation NrFrG)rUr)r^ra op_patternsrCr(r(r)rs  rr>)1r\typingrrrrtorch.ao.nn.intrinsicaor intrinsicrtorch.nntorch.ao.quantization.utilsrrr__all__r r r r rDr BatchNorm1dReLUr BatchNorm2dr BatchNorm3dr-r.r/Linear LinearReLUBNReLU2dBNReLU3dConvTranspose1dConvTranspose2dConvTranspose3drEdictrZ Sequential__annotations__r rQrTrUrr(r(r(r)sj   ,7"