}Yh0|ddlZddlZddlmZmZmZddlZddlmZ dej fdZ de ddfdZ dej fdZ dd ed ed eed edeedeedefdZGddZGddZ ddeeefdeeedeeeedfffdZ ddededede fdZdS) N)AnyOptionalUnion)_get_device_indexreturncrtjdkrtjdStjdS)Nwin32z nvcuda.dllz libcuda.so.1sysplatformctypesCDLLc/var/www/tools.fuzzalab.pt/emblema-extractor/venv/lib/python3.11/site-packages/torch/cuda/_utils.py_get_cuda_libraryr s/ |w{<((({>***rresultc |dkrdStj}t}||tj||j|jnd}td|)NrUnknown CUDA error CUDA error: )r c_char_prcuGetErrorStringbyrefvaluedecode RuntimeError)rerr_strlibcuda error_messages r _check_cudar s {{oG!!G VV\'%:%:;;;")-"; AU 5m55 6 66rcrtjdkrtjdStjdS)Nr znvrtc64_120_0.dllz libnvrtc.sor rrr_get_nvrtc_libraryr" s2 |w{.///{=)))r kernel_source kernel_namecompute_capability header_codecuda_include_dirs nvcc_optionsc ddl}tddtddffd }|dsd|}|r |dz|z}n|}|d } |A|j|j} | j | j }g} | d ||r/|D],} | d | -|r-|D]*} | | d +dd l m }d |D}| d|Dt| }t!j|z| }t!j}|t!j|| |dddd|||}|krt!j}|t!j|t!j|j}||t7d|jt!j}||t!j|t!j|j}|||t!j||jS)a Compiles a CUDA kernel using NVRTC and returns the PTX code. Args: kernel_source (str): The CUDA kernel source code as a string kernel_name (str): The name of the kernel function to compile compute_capability (str, None): The compute capability to target (e.g., "86"). If None, will detect from current device. header_code (str, optional): Additional header code to prepend to the kernel source cuda_include_dirs (list, None): List of directories containing CUDA headers nvcc_options (list, None): Additional options to pass to NVRTC Returns: str: The compiled PTX code rNrrc|krotj}|tj||j|jnd}t d|dS)Nrr)r rnvrtcGetErrorStringrrrr)rrr NVRTC_SUCCESSlibnvrtcs r check_nvrtcz#_nvrtc_compile..check_nvrtcJs ] " "o''G  ( (g1F1F G G G=, $$&&&)  =m==>> > # "rz extern "C"z extern "C"  utf-8z--gpu-architecture=sm_z-I)COMMON_NVCC_FLAGScg|] }|dk| S)z--expt-relaxed-constexprr.0flags r z"_nvrtc_compile..ys*d6P.P.P.P.P.Prc8g|]}|dS)r1)encoder4s rr7z"_nvrtc_compile..|s$LLLTDKK((LLLrz.cuzKernel compilation failed: ) torch.cudar"intstrip startswithr9cudaget_device_propertiescurrent_devicemajorminorappendtorch.utils.cpp_extensionr2extendlenr rc_void_pnvrtcCreateProgramrnvrtcCompileProgramc_size_tnvrtcGetProgramLogSizecreate_string_bufferrnvrtcGetProgramLogrrnvrtcGetPTXSize nvrtcGetPTXnvrtcDestroyProgram)r$r%r&r'r(r)torchr/ full_source source_bytespropsoptions directoryoptionr2nvrtc_compatible_flags num_options options_arrayprogreslog_sizelogptx_sizeptxr-r.s @@r_nvrtc_compilera)s0"##HM ?C ?D ? ? ? ? ? ? ?    + +L 9 965m55 $!D(=8 # %%g..L! 001J1J1L1LMM % :U[::G NN@,>@@GGIIJJJ6* 6 6I NN+ ++2244 5 5 5 53" 3 3F NN6==11 2 2 2 2<;;;;;* NNLL5KLLLMMMg,,K_{2W=M ?  DK## L       & & ( (          & &t[- H HC m?$$''fl8.D.DEEE)(.99##D#...N#):J:J:L:LNNOOO  HK((v|H/E/EFFGGG  %hn 5 5CK$$T3//000   d!3!3444 9rc8eZdZdejddfdZdeddfdZdS) _CudaModulemodulerNc"||_i|_dSN)_module_kernels)selfrds r__init__z_CudaModule.__init__s 02 rname _CudaKernelc ||jvr |j|Sddlm}|}tj} t |tj||j| dt||j}||j|<|S#t$r}td|d|d}~wwxYw)Nr)rr1zNo kernel named 'z' in this module) rhtorch.cuda._utilsrr rGr cuModuleGetFunctionrrgr9rlrAttributeError)rirkrrfunckernelerrs r __getattr__z_CudaModule.__getattr__s 4= =& & 877777##%%   V ++L&& dkk'6J6J    !t|44F"(DM$ M V V V !KT!K!K!KLLRU U VsA.B** C 4CC )__name__ __module__ __qualname__r rGrjstrrtrrrrcrcsb3v343333VV VVVVVVrrcc eZdZdZdejdejddfdZ dd eeeefd eeeefd e e d ed e e ddf dZ dS)rlzT Represents a compiled CUDA kernel that can be called with PyTorch tensors. rqrdrNc"||_||_dSrf)rqrd)rirqrds rrjz_CudaKernel.__init__s  rr|r|rgridblockargs shared_memstreamcddl}|jj}|sg}g}g} |D]l} t | |jr| js*| jr| stdtj | } | | | tj| t | tr_utilsr isinstanceTensoris_cudais_cpu is_pinned ValueErrorr rGdata_ptrrCrr;c_intfloatc_float TypeErrortyperF enumeratecastr:current_streamr cuLaunchKernelrq_as_parameter_)rir}r~rrrrQrprocessed_argsc_argsargptrrr c_args_arrayis r__call__z_CudaKernel.__call__sg&  *#5577 D13 K KC#u|,, K{CJ3==??$Yocllnn55%%c*** fl3//0000C%% K S)) fl5112222C'' K .-- fl7334444 Id3ii I IJJJ#f++588 '' @ @FAs$k#v??LOO >    Z..00F  " " QQQaaa%       r)r{r{NrN) rurvrw__doc__r rGrjtupler;rlistrrrrrrlrlsV_fo$ &/&/# $ P P CcM"P S#s]#P tn P  P  P  P P P P P P rrlr` kernel_namesc ddl}t}t|tr|d}t j}|j}|5t| t j ||dddn #1swxYwY|st|Si}|D]q}t j}t| t j |||dt||||<r|S)a, Loads a CUDA module from PTX code and returns a module object that can access kernels. Args: ptx (bytes or str): The PTX code to load kernel_names (list, optional): List of kernel names to extract from the module. If None, will return a module object with __getattr__. Returns: object: If kernel_names is None, returns a module object with __getattr__ to access kernels. If kernel_names is provided, returns a dict mapping kernel names to _CudaKernel objects. rNr1)r:rrrxr9r rGr>rr cuModuleLoadDatarrcrorl) r`rrQrrdrkernelsrkrqs r_cuda_load_modulers  !!G#s"jj!!_  F Z & & ( (F IIG,,V\&-A-A3GGHHHIIIIIIIIIIIIIII #6"""G22    ' ' T""FDKK,@,@     $D&11 Ns+6B--B14B1Fdeviceoptional allow_cpuct|tr|St|trtj|}t|tjr;|r|jdvrt d|n|jdkrt d|tjs&t|tj jr|j St|||S)aGet the device index from :attr:`device`, which can be a torch.device object, a Python integer, or ``None``. If :attr:`device` is a torch.device object, returns the device index if it is a CUDA device. Note that for a CUDA device without a specified index, i.e., ``torch.device('cuda')``, this will return the current default CUDA device if :attr:`optional` is ``True``. If :attr:`allow_cpu` is ``True``, CPU devices will be accepted and ``-1`` will be returned in this case. If :attr:`device` is a Python integer, it is returned as is. If :attr:`device` is ``None``, this will return the current default CUDA device if :attr:`optional` is ``True``. )r>cpuz(Expected a cuda or cpu device, but got: r>z!Expected a cuda device, but got: ) rr;rxrQrrrjit is_scriptingr>idx_torch_get_device_index)rrrs rrrNs &# &#&f%%&%,''K  K{/11 !TF!T!TUUU2 [F " "IIIJJ J 9 ! ! # # fej/ 0 0 :  "68Y ? ??r)Nr#NNrf)FF)r r typingrrrrQ torch._utilsrrrrr;r r"rxrbytesrarcrldictrboolrrrrsE '''''''''' FEEEEE+6;++++ 7 7 7 7 7 7*FK****)-(,#' yyyy! y y  ~ y 4. y yyyyxVVVVVVVV:Y Y Y Y Y Y Y Y zAE-- sEz -*249*=- ;S-/0 01----b