o h~[@sddlZddlZddlZddlZddlZddlZddlZddlZddlZddl m Z ddl m Z ddl mZmZmZmZddlZddlmZddlZddlmZddlmZddlmZddlmZmZd d lm Z d d lm!Z!m"Z"e#e$Z%ed Z&e d Z'e'(Z)e)rddl*Z+gZ,dZ-e)rgdZ,e+j.j/j012ddZ3d4dde,DZ-gdZ5GdddZ6ddZ7dZ8GdddZ9e:dddZ;dd d!d"ZGd'd(d(e?Z@d)d*ZAdGd+d,ZB dHddd-d.d/ZCd0d1ZDd2d3ZEd4d5ZF dHddd-d6d7ZGd8eeeHd9eeHd:eeHfd;d<ZId=e&d:eee&ge&ffd>d?ZJeJejKZLeJeMd@ZNeJdZOeJdZPeJdZQGdAdBdBZRGdCdDdDZSGdEdFdFZTdS)IN)Counter) import_module)CallableOptionalSequenceTypeVar) rand_strided)is_float_dtype)StorageWeakRef)ContentStoreReaderContentStoreWriter)config) clone_inputs get_debug_dirTztorch._inductor.config)z1//caffe2/torch/fb/sparsenn:sparsenn_operators_gpuz-//caffe2/torch/fb/sparsenn:sparsenn_operatorsz///deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpuz+//deeplearning/fbgemm/fbgemm_gpu:sparse_opszfbcode:// cCg|]}d|dqS)ztorch.ops.load_library("z").0xrrO/var/www/html/ai/venv/lib/python3.10/site-packages/torch/_dynamo/debug_utils.py 0r)buck2runz@mode/dev-nosanc@s&eZdZddZddZd ddZdS) BuckTargetWritercCstjtj|\|_|_|jdd|_|jddd|j|_|j|jdd|_|jdd|_|j}||dddd}d|d |j|_ dS) Nz.pyr/.zfbcode.zfbcode/r:) ospathsplitabspathsubdirpy_filereplacetargetfind cmd_line_path)selffilenametmprrr__init__7szBuckTargetWriter.__init__c CsDdddtD}td|jd|jdtd|d|jd S) NrcSr)z "z",rrrrrrFrz*BuckTargetWriter.build..za load("@fbcode_macros//build_defs:python_binary.bzl", "python_binary") python_binary( name="z", srcs = ["z"], compile = False, deps = [ "//caffe2:torch", "//caffe2/functorch:functorch", "//triton:triton", "z", ], cpp_deps = [ z ], main_module = "z", ) )join extra_depstextwrapdedentr+r) cur_targetr%)r.extra_cpp_depsrrrbuildEs zBuckTargetWriter.buildTcCsntj|jd}t|d}||Wdn1swYt|jg}|r5t dd||S)NTARGETSwzFFound an example that reproduces the error. Run this cmd to repro - %s ) r$r%r2r(openwriter8BUCK_CMD_PREFIXr-logwarning)r. print_msg target_filefd cmd_splitrrrr=]s  zBuckTargetWriter.writeN)T)__name__ __module__ __qualname__r1r8r=rrrrr6srcCsLtjtd}|durtdt}tj|s$tj |dd|S)Nminifierz /minifier_T)exist_ok) r$r%r2rtempfile gettempdirgetpassgetuserexistsmakedirs)r%rrr minifier_dirks  rPc@seZdZejjejjejjejjejj ejj ejj ejj ejj ejjejjejjejjejjejjejjejjejjejjejjejjgZeddZeddZdS)NNModuleToStringcCsLt}|D]\}}t|tjvr||qt|dkr$td|dS)Nrz-We have not tested reprs of some modules - %sT) setnamed_childrentyperR safe_reprsaddlenr?r@)gm cant_convert_modulerrrcan_convert_to_strings   z&NNModuleToString.can_convert_to_stringcCsddlm}d}td}|D]+\}}|}t|d}|dur-|jr-|d}||dd|d|d 7}q|j D]X\}} | durKqB| t krcdd l m} | jt ks^Jt| } n t| rvd t| jd | jd } n dt| jd | jd } | jr| d} ||dd|d| d7}qB|j D].\} } | durqd}| jrd}dt| jd | j|d} ||dd| d| d 7}q|||jdd 7}|S)Nr) _addindent z from torch.nn import * class Repro(torch.nn.Module): def __init__(self): super().__init__() z.cuda()zself.z = r) PRINT_OPTSz torch.randn(z, dtype=)ztorch.randint(1, size=zself.register_buffer('z', z) rz, device="cuda"ztorch.nn.Parameter(torch.randn(z))rQ)torch.nn.modules.moduler^r4r5rT__repr__next parametersis_cuda_buffersitemsnumelMAX_CONSTANT_NUMEL_INLINEtorch._tensor_strra thresholdreprtorchis_floating_pointlistshapedtype _parameterscode)rYr^tab model_str module_namer\ module_str example_param buffer_namebufferra tensor_str param_nameparam maybe_devicerrrconvertsF            zNNModuleToString.convertN)rErFrGronnLinearConv1dConv2dConv3d BatchNorm1d BatchNorm2d BatchNorm3d LayerNormDropoutSoftmaxReLUGELUIdentity MaxPool2d EmbeddingTanhConvTranspose1dGLULSTMFlattenAdaptiveAvgPool2drV staticmethodr]rrrrrrRws6  rRcCstjsdSd}z%tjddgtjd}|jd}d dd |D}||d7}Wn t y;|d 7}Ynwt d d t tj D}|d 7}|D]\}}|d|d|d7}qR|d7}|S)Nz:# torch.cuda.is_available()==False, no GPU info collected z# CUDA Info: nvccz --version)stdoutrrcSs g|] }|dvrd|dqS))r#  r)rsrrrrs z-_cuda_system_info_comment..z# nvcc not found css|] }tj|VqdSN)rocudaget_device_name)rirrr s  z,_cuda_system_info_comment..z# GPU Hardware Info: rz : r)ror is_available subprocessrPIPErdecoder&r2FileNotFoundErrorrrange device_countri)rwcuda_version_outcuda_version_linescomment gpu_namesnamecountrrr_cuda_system_info_comments&    rF) stable_outputcCsHddl}ddl}|r dSd|jjd|jjd|jjdS)Nrz*# config omitted due to stable_output=TruezYimport torch._dynamo.config import torch._inductor.config import torch._functorch.config r)torch._functorch.configtorch._inductor.config_dynamorcodegen_config _inductor _functorch)rrorrrgenerate_config_strings   rcCstjtdS)Nzminifier_launcher.py)r$r%r2rPrrrrget_minifier_repro_pathsrc Cst}td|trt|zt|d}||WdWdS1s)wYWdStyF}z t|t d|d}~ww)NzWriting minified repro to: %sr:z(Could not write to {minified_repro_path}) rr?r@use_buckrr=r<OSError exceptionNotImplementedError)contentsminified_repro_pathrCerrrhelper_for_dump_minify s    &  rc@s eZdZdS) AccuracyErrorN)rErFrGrrrrrsrcCsBt|}tt|D]}t||tjr||||jq |S)z This clone inputs is different from utils clone_input. In case of minifier, all the tensors are leaf tensors while creating a new graph. So, we set the requires_grad field w/o checking the leafness of the tensor. )rrrX isinstanceroTensorrequires_grad_ requires_grad)example_inputs cloned_inputsidxrrrclone_inputs_retaining_gradnesss rc Csddlm}ddlm}m}m}t|}|st|}t |dr%| dt |dd}t |d d} t |d sP|dus>| durP||}|durI||_ | durP| |_ ||} |rX| S|| rd|| } | ||| d|S) z Runs a forward and possibly backward iteration for a given mod and args. When disable_clone is True, we will use args as-is without cloning. This is higher fidelity but we may destroy the args in the process. r)make_boxed_funcr )collect_resultsreduce_to_scalar_lossrequires_bwd_pass zero_gradTnamed_parametersN named_buffers _boxed_call)torch._functorch.aot_autogradrtestingrrrcopydeepcopyrhasattrrgetattrrrbackward) rYargsonly_fwd disable_clonerrrrorig_named_parametersorig_named_buffersoutlossrrrrun_fwd_maybe_bwd)s.       r require_fp64 ignore_non_fpc Cs(ddlm}ddlm}m}ddlm} t||r#|||_|||_ t||r2|||_|||_ t |||} d} t j rfzt t|t|\} } t | | |} Wntye|r^tdtdYnwzt |||}Wnty}z tdWYd}~dSd}~ww| | || t jd|d }|S) aa Check two models have same accuracy. require_fp64: if True, raise an error if we unable to calculate the fp64 reference ignore_non_fp: if True, do not compare outputs which are not floating point. This is mostly useful for the minifier (which wants to avoid quantizing floating point error into integer/boolean error) r )OptimizedModule)"named_buffers_for_optimized_module%named_parameters_for_optimized_module)sameNzCould not generate fp64 outputszWhile minifying the program in accuracy minification mode, ran into a runtime exception which is likely an unrelated issue. Skipping this graph.T)tol equal_nanr) eval_framerrrrutilsrrrrrrsame_two_models_use_fp64 cast_to_fp64rrr Exception RuntimeErrorr?r@rrepro_tolerance)rYopt_gmrrrrrrrrreffp64_ref fp64_model fp64_examplesresrpassingrrrsame_two_modelsPsP           rcCs|jjD]1}|jdkr5|jtjjjjkr5t |j dksJt |j dr5|j dtj kr5|j dtj f|_ q|j ||S)N call_functionr`r r)graphnodesopr+roopsprimsconvert_element_typedefaultrXrr float64lint recompile)modelnoderrr!cast_convert_element_type_to_fp64s   rcsBddlm}|}tjkrt|}|fdd|}||fS)Nr)tree_mapcs"t|tjr|r|S|Sr)rrorrptorrsrrs  zcast_to..)torch.utils._pytreerr rorr)rsrinputsrrr rcast_tos    rcCsttj||Sr)rror)rrrrrrsrc Cs^z|t|t|}t||||||d WSty.}z tdWYd}~dSd}~ww)NrzWhile minifying the program in accuracy minification mode, ran into a runtime exception which is likely an unrelated issue. Skipping this graphF)rrrrrr?r)rYr compiler_fnrrr compiled_gmrrrrbackend_accuracy_failss&  rstriderrreturncCs|dur|St|Sr)rmake_contiguous_strides_for)rrrrrr_stride_or_defaultsrdcs fddS)Ncs|dur|SSrrr rrrr sz_mk_defaulter..rrrrr _mk_defaulters rcpuc@s4eZdZddZdddddZddZd d ZdS) NopInputReadercCs d|_dS)Nrtotal)r.rrrr1s zNopInputReader.__init__Ndevice dtype_hintcCs|jd7_dS)Nr r)r. storage_hashnbytesrr rrrstorageszNopInputReader.storagecOdSrrr.rkwargsrrrtensorzNopInputReader.tensorcOr$rrr%rrrsymintr(zNopInputReader.symintrErFrGr1r#r'r)rrrrrs  rc@sLeZdZd ddddZdddddZ d dddddd d Zd d ZdS) InputReaderN)pbarcCs8|dur td|durt|nd|_g|_||_dS)Nz0no save_dir specified, will generate random data)r?r@r storerr,)r.save_dirr,rrrr1s   zInputReader.__init__rcCs|jdur |jdt|}t|}|jdur=|dur=z|j|}Wn ty-Ynw||jkr;t d||j|St d|||j f}t d|d}t |||| S)Nr zdevice mismatch: %s != %sz1could not load %s, generating random data insteadrr)r,update_device_or_default_dtype_or_defaultr- read_storagerrr?r@itemsizerruntyped_storage)r.r!r"rr r#rrrrrrr#s"       zInputReader.storage)storage_offsetrsris_leafc Kst||d}t|}t|}t|}t|}tjg||j|d} t| ||||Wdn1s7wY|sut | j tj d} Wdn1sTwYt| ||||Wdn1spwYtj j| |ksJtj| ||j| | S)Nr/)rsrr) memory_format)r_storage_offset_or_defaultr2_is_leaf_or_default_requires_grad_or_defaultror'rno_gradset_ enable_gradclonepreserve_format _subclasses meta_utils safe_is_leaf_utilsset_tensor_metadatarappend) r.r#rrrr6rsrr7metadatatrrrr'%s,      zInputReader.tensorcCs|j||Sr)rrF)r.valrrrr)Fs zInputReader.symintrr*rrrrr+s  !r+c@sLeZdZddddZddZdddd efd d Zdd d ZdddZdS) InputWriterF stable_hashcCs:g|_t|_||_|durt||dnd|_i|_dS)NrK)_lines itertoolsrstorage_counterr.r r- seen_storages)r.r.rLrrrr1Xs   zInputWriter.__init__cCs*dg}|dd|jD|d|S)Nzdef load_args(reader):css|]}d|VqdS)r_Nr)rlrrrrhsz$InputWriter.lines..zload_args._version = 0)extendrMrF)r.rrrrlinesds  zInputWriter.linesNr  device_hintrc Cst|}|j|}|dur|Sdt|j}d}tdt|kr'd|}d}|j}|jdkr9|dus7J|}td|krDd|}| } d} |j dur[|jjdkr[|j |} |j |d| d| ||d||j|<|S) Nbufrz , dtype_hint=metaz , device=z = reader.storage(, rb)r rPgetrerOr2rrUr1r"r- write_storagerMrF) r.r5r rVwsvmaybe_dtype_hintrrr"r!rrrr#ss0        zInputWriter.storagec Cs4|j||j|jd}g}td|jd|kr$|tt |t d|jkr4|d|jt d| krF|d| t j|}|rZ|dd|Dtd|jkrj|d|jt jj|}td|kr|d||jd d |tt |jg|d |dS) NrUr/zdtype=zstorage_offset=css"|] \}}|d|VqdS)=Nr)rkr]rrrrs z%InputWriter.tensor..zrequires_grad=zis_leaf=zreader.tensor(rY) # )r#r5rsrrrrrrFstrtupler2r9r6rorDget_tensor_metadatarRrir;rrArBrCr:rMr2)r.rrHr#rtensor_metadatar7rrrr's2  zInputWriter.tensorcCs0t|tjr |jj}|jd|d|dS)Nzreader.symint(ra)rroSymIntrhintrMrF)r.rrIrrrr)s zInputWriter.symint)rN) rErFrGr1rTrbr#r'r)rrrrrJWs   rJ)FF)F)Ur functoolsrLrNloggingr$rrJr4 collectionsr importlibrtypingrrrrrotorch._prims_common _prims_commonrtorch._subclasses.meta_utilstorch._dynamo.testingrr torch.multiprocessing.reductionsr torch.utils._content_storer r rrrr getLoggerrEr?rinductor_config is_fbcoderlibfb.py.build_infolibfbr3 extra_importspy build_info BuildInfoget_build_ruler*r6r2r>rrPrkrR lru_cacherrrrrrrrrrrrrintrrfloat32r2rr1r9r;r:rr+rJrrrrs        5 d  + E )   T