+
    NåûjYb  ã                   ó¶  € ^ RI t ^ RIt^ RIt^ RIHt ^ RIt^ RIHt  ^ RIH	t
HtHt Rt]'       d   ]P"                  P$                  e   RtRtRt^ RIHt R R	 ltR
 R ltR R ltR R ltR R ltR R ltR R ltR R ltR R ltR R ltR R lt R(R R llt! ! R  R!4      t" ! R" R#4      t#R)R$ R% llt$R*R& R' lltR#   ] d    Rt
RtRtRt L­i ; i)+é    N©ÚAny)ÚVersion)Ú__version__ÚdriverÚruntimeTF)Ú_get_device_indexc                ó4   € V ^8„  d   QhR\         R\        RR/# )é   ÚversionÚmessageÚreturnN)ÚintÚstr)Úformats   "ÚMC:\Ai_Coding\apps\studio\ComfyUI\.venv\Lib\site-packages\torch\cuda\_utils.pyÚ__annotate__r   &   s!   € ÷ $ñ $¬3ð $¼ð $Àñ $ó    c                 ó¾   € \         f   \        V4      h\        \         4      P                  pVR,          V R,          V R,          ^
,          38  d   \        V4      hR # )N:Nr   Niè  )Ú_cuda_bindings_versionÚRuntimeErrorr   Úrelease)r   r   r   s   && r   Ú_ensure_cuda_bindings_versionr   &   sR   € ÜÒ%Ü˜7Ó#Ð#ÜÔ,Ó-×5Ñ5€GØˆr…{�g •o w°¥~¸Õ';Ð<Ô<Ü˜7Ó#Ð#ñ =r   c                ó8   € V ^8„  d   QhR\         P                  /# ©r   r   ©ÚctypesÚCDLL)r   s   "r   r   r   .   s   € ÷ ñ ¤&§+¡+ñ r   c                  ó  €  ^ RI p \        P                  ! \        V P	                  R4      ^ ,          4      4      pVP                  Vn        VP                  Vn        VP                   Vn        VP$                  Vn        VP(                  Vn        V#   \
        \        3 dj    \        P                  R8X  d<   \        P                  ! R\        P                  P                  ^ ,           R24      p L¸\        P                  ! R4      p LÐi ; i)r   NÚamdhip64Úwin32Ú	amdhip64_ú.dllzlibamdhip64.so)Úrocm_sdkr   r   r   Úfind_librariesÚImportErrorÚ
IndexErrorÚsysÚplatformÚtorchr   ÚhipÚhipGetErrorStringÚcuGetErrorStringÚhipModuleLoadDataÚcuModuleLoadDataÚhipModuleGetFunctionÚcuModuleGetFunctionÚhipModuleLaunchKernelÚcuLaunchKernelÚhipFuncSetAttributeÚcuFuncSetAttribute)r$   Úlibs     r   Ú_get_hip_runtime_libraryr7   .   sÒ   € ð	0ãä�kŠkœ#˜h×5Ñ5°jÓAÀ!ÕDÓEÓFˆð ×0Ñ0€CÔØ×0Ñ0€CÔØ!×6Ñ6€CÔØ×2Ñ2€CÔØ ×4Ñ4€CÔØ€Jøô œÐ$ô 0Ü�<‰<˜7Ô"Ü—+’+ 	¬%¯-©-×*;Ñ*;¸AÕ*>Ð)?¸tÐDÓEŠCä—+’+Ð.Ó/ŠCð	0ús   ‚9B ÂADÃ3DÄDc                ó8   € V ^8„  d   QhR\         P                  /# r   r   )r   s   "r   r   r   E   s   € ÷ +ñ +œ6Ÿ;™;ñ +r   c                  ó„   € \         P                  R 8X  d   \        P                  ! R4      # \        P                  ! R4      # )r!   z
nvcuda.dllzlibcuda.so.1)r(   r)   r   r   © r   r   Ú_get_cuda_libraryr;   E   s,   € Ü
‡|�|�wÔÜ�{Š{˜<Ó(Ð(ä�{Š{˜>Ó*Ð*r   c                ó8   € V ^8„  d   QhR\         P                  /# r   r   )r   s   "r   r   r   M   s   € ÷ #ñ #¤&§+¡+ñ #r   c                  ój   € \         P                  P                  '       d   \        4       # \	        4       # ©N)r*   r   r+   r7   r;   r:   r   r   Ú_get_gpu_runtime_libraryr?   M   s$   € Ü‡}�}××ÐÜ'Ó)Ð)ä Ó"Ð"r   c                ó(   € V ^8„  d   QhR\         RR/# ©r   Úresultr   N©r   )r   s   "r   r   r   U   s   € ÷ 	7ñ 	7œð 	7 ñ 	7r   c                 ó  € V ^ 8X  d   R# \         P                  ! 4       p\        4       pVP                  V \         P                  ! V4      4       VP
                  e   VP
                  P                  4       MRp\        RV 24      h)r   NúUnknown CUDA errorúCUDA error: )r   Úc_char_pr?   r-   ÚbyrefÚvalueÚdecoder   )rB   Úerr_strÚlibcudaÚerror_messages   &   r   Ú_check_cudarN   U   sn   € Ø�„{ÙÜ�oŠoÓ€GÜ&Ó(€GØ×Ñ˜V¤V§\¢\°'Ó%:Ô;à")§-¡-Ò";ˆ�‰×ÑÔÐAUð ô ˜ m _Ð5Ó
6Ð6r   c                ó0   € V ^8„  d   QhR\         R\         /# ©r   rB   r   r   )r   s   "r   r   r   a   s   € ÷ (ñ (¤ð (¬ñ (r   c                ó^  € \         '       g   \        R4      hV vrVP                  '       dÖ   \        V\        P
                  4      '       d>   \        P                  ! V4      w  r4\        P
                  P                  pW58X  d   TMRpM@\        V\        P                  4      '       d   \        P                  ! V4      ^,          pMRp\        V\        4      '       d   VP                  4       p\        RV RV R24      h\        V4      ^ 8X  d   R# \        V4      ^8X  d
   V^ ,          # V# )a  Check a cuda.bindings (cuda-python) call result for errors.

All cuda.bindings driver/runtime calls return ``(error, *outputs)``. This
helper unpacks the tuple, raises on non-success, and returns the
outputs (``None`` for zero outputs, scalar for one, tuple otherwise).
úcuda.bindings is not availablez	<unknown>zunknown error typerF   ú (Ú)N)Ú_HAS_CUDA_BINDINGSr   rI   Ú
isinstanceÚ_cuda_bindings_driverÚCUresultr-   ÚCUDA_SUCCESSÚ_cuda_bindings_runtimeÚcudaError_tÚcudaGetErrorStringÚbytesrJ   Úlen)rB   ÚerrÚoutÚerr2ÚnameÚsuccessrK   s   &      r   Ú_check_cuda_bindingsrd   a   sô   € ÷ ÓÜÐ;Ó<Ð<ð €I€CØ
‡y‡y€yä�cÔ0×9Ñ9×:Ò:ä.×?Ò?ÀÓD‰JˆDä+×4Ñ4×AÑAˆGð #œo‘d°;‰Gä˜Ô3×?Ñ?×@Ò@ä,×?Ò?ÀÓDÀQÕG‰Gà*ˆGÜ�gœu×%Ò%Ø—n‘nÓ&ˆGÜ˜\¨#¨¨b°°	¸Ð;Ó<Ð<ä
ˆ3ƒx�1„}ÙÜ
ˆ3ƒx�1„}Ø�1�vˆØ€Jr   c                ó0   € V ^8„  d   QhR\         R\         /# rP   r   )r   s   "r   r   r   Œ   s   € ÷ ñ ¬ð ´ñ r   c                ón  € \         '       g   \        R4      hV vrV\        P                  P                  8w  dQ   \        P
                  ! V4      w  r4\        V\        4      '       d   VP                  4       p\        RV RV R24      h\        V4      ^ 8X  d   R# \        V4      ^8X  d
   V^ ,          # V# )z×Check a cuda.bindings (cuda-python) driver call result for errors.

Like ``_check_cuda_bindings`` but for the CUDA driver API, whose calls
return ``(CUresult, *outputs)`` and report errors via ``cuGetErrorString``.
rR   zCUDA driver error: rS   rT   N)
rU   r   rW   rX   rY   r-   rV   r]   rJ   r^   )rB   r_   r`   Ú_rK   s   &    r   Ú_check_cuda_bindings_driverrh   Œ   s«   € ÷ ÓÜÐ;Ó<Ð<Ø€I€CàÜ ×)Ñ)×6Ñ6ô	7ô "×2Ò2Øóñ 	ˆô
 �gœu×%Ò%Ø—n‘nÓ&ˆGÜÐ0°°°R¸°yÀÐBÓCÐCÜ
ˆ3ƒx�1„}ÙÜ
ˆ3ƒx�1„}Ø�1�vˆØ€Jr   c                ó8   € V ^8„  d   QhR\         P                  /# r   r   )r   s   "r   r   r   ¨   s   € ÷ ñ œVŸ[™[ñ r   c                  ó4  €  ^ RI p \        P                  ! \        V P	                  R4      ^ ,          4      4      pVP                  Vn        VP                  Vn        VP"                  Vn        VP&                  Vn        VP*                  Vn        VP.                  Vn        VP2                  Vn        VP6                  Vn        VP:                  Vn        VP>                  Vn         V#   \
        \        3 d     \        P                  R8X  dq   RP                  R\        P                  P                  ^ ,          R\        P                  P                  ^,          .4      p\        P                  ! RT R24      p ELB\        P                  ! R4      p EL[i ; i)r   NÚhiprtcr!   Ú Ú0r#   zlibhiprtc.so)!r$   r   r   r   r%   r&   r'   r(   r)   Újoinr*   r   r+   ÚhiprtcGetErrorStringÚnvrtcGetErrorStringÚhiprtcCreateProgramÚnvrtcCreateProgramÚhiprtcDestroyProgramÚnvrtcDestroyProgramÚhiprtcCompileProgramÚnvrtcCompileProgramÚhiprtcGetCodeSizeÚnvrtcGetCUBINSizeÚhiprtcGetCodeÚnvrtcGetCUBINÚhiprtcGetProgramLogSizeÚnvrtcGetProgramLogSizeÚhiprtcGetProgramLogÚnvrtcGetProgramLogÚhiprtcAddNameExpressionÚnvrtcAddNameExpressionÚhiprtcGetLoweredNameÚnvrtcGetLoweredName)r$   r6   Úversion_strs      r   Ú_get_hiprtc_libraryr„   ¨   sA  € ð.ãä�kŠkœ#˜h×5Ñ5°hÓ?ÀÕBÓCÓDˆð "×6Ñ6€CÔØ ×4Ñ4€CÔØ!×6Ñ6€CÔØ!×6Ñ6€CÔØ×1Ñ1€CÔØ×)Ñ)€CÔØ!$×!<Ñ!<€CÔØ ×4Ñ4€CÔØ!$×!<Ñ!<€CÔØ!×6Ñ6€CÔØ€Jøô) œÐ$ô .Ü�<‰<˜7Ô"ØŸ'™'Ø”e—m‘m×'Ñ'¨Õ*¨C´·±×1BÑ1BÀ1Õ1EÐFóˆKô —+’+  { m°4Ð8Ó9‹Cä—+’+˜nÓ-‹Cð.ús   ‚9C' Ã'BFÅ=FÆFc                ó8   € V ^8„  d   QhR\         P                  /# r   r   )r   s   "r   r   r   Å   s   € ÷ 6ñ 6œFŸK™Kñ 6r   c                  ó6  € \        \        P                  P                  P	                  R 4      ^ ,          4      p \
        P                  R8X  d	   RV  R2.pMRV  2R.pV F  p \        P                  ! V4      u # 	  \        R4      h  \         d     K7  i ; i)Ú.r!   Únvrtc64_z0_0.dllzlibnvrtc.so.zlibnvrtc.soz Could not find any NVRTC library)
r   r*   r   ÚcudaÚsplitr(   r)   r   r   ÚOSError)Úmajor_versionÚ
nvrtc_libsÚlib_names      r   Ú_get_nvrtc_libraryr�   Å   s�   € ÜœŸ™×*Ñ*×0Ñ0°Ó5°aÕ8Ó9€MÜ
‡|�|�wÔà�}�o WÐ-ð
‰
ð
 ˜=˜/Ð*Øð
ˆ
ó ˆð	Ü—;’;˜xÓ(Ò(ñ ô
 Ð4Ó
5Ð5øô ô 	Úð	ús   Á$B	Â	BÂBc                ó8   € V ^8„  d   QhR\         P                  /# r   r   )r   s   "r   r   r   Ø   s   € ÷ $ñ $œfŸk™kñ $r   c                  ój   € \         P                  P                  '       d   \        4       # \	        4       # r>   )r*   r   r+   r„   r�   r:   r   r   Ú_get_gpu_rtc_libraryr’   Ø   s&   € ô ‡}�}××ÐÜ"Ó$Ð$ä!Ó#Ð#r   c                ó:   € V ^8„  d   QhR\         \        ,          /# r   )Úlistr   )r   s   "r   r   r   á   s   € ÷ ñ ¤t¬C¥yñ r   c                 óº   € ^ RI Hp Hp R0pV Uu. uF  q3V9  g   K  VNK  	  pp\        P                  P
                  '       d   VP                  V 4       V# u upi )z�
Get HIPCC/NVCC flags that are compatible with NVRTC compilation.

Returns:
    List of HIPCC/NVCC flags that can be safely used with NVRTC.
)ÚCOMMON_HIPCC_FLAGSÚCOMMON_NVCC_FLAGSz--expt-relaxed-constexpr)Útorch.utils.cpp_extensionr–   r—   r*   r   r+   Úextend)r–   r—   Únvrtc_unsupported_flagsÚflagÚcompatible_flagss        r   Ú_get_gpu_rtc_compatible_flagsr�   á   sh   € ÷ Pð 	#ðÐñ +óÙ*�Ð:QÑ.Q�ˆÑ*ð ð ô ‡}�}××ÐØ×ÑÐ 2Ô3àÐùòs
   �A�Ac                ó¸   € V ^8„  d   QhR\         R\         R\         R,          R\        R,          R\        R,          R\        R\        \        \         3,          /# )	r   Úkernel_sourceÚkernel_nameÚcompute_capabilityNÚcuda_include_dirsÚnvcc_optionsÚauto_pchr   )r   r”   ÚboolÚtupler]   )r   s   "r   r   r   ù   sl   € ÷ S$ñ S$ÜðS$äðS$ô ˜d�
ðS$ô ˜d•{ð	S$ô
 ˜•+ðS$ô ðS$ô Œ5”#ˆ:ÕñS$r   c           
     ó¨	  aa€ ^ RI p\        4       o^ oR VV3R llpV P                  R4      pVfx   VP                  P	                  VP                  P                  4       4      p	VP                  P                  '       d   V	P                   pMV	P                   V	P                   2p. p
VP                  P                  '       d$   V
P                  RV 2P                  4       4       M"V
P                  RV 2P                  4       4       ^ RIHp V! R4      pV F%  pV
P                  R	V 2P                  4       4       K'  	  V'       d,   V F%  pV
P                  R	V 2P                  4       4       K'  	  V'       d^   \        VP                  P                  4      R
8  d#   \        RVP                  P                   24      hVf   . pVP                  R4       V'       d*   V F#  pV
P                  VP                  R4      4       K%  	  \!        4       pT
P#                  V Uu. uF  pVP                  R4      NK  	  up4       \%        V
4      p\&        P(                  V,          ! V
!  p\*        P,                  ! RRV4      R,           p\&        P.                  ! 4       pV! SP1                  \&        P2                  ! V4      VVP                  4       ^ RR4      4       VP                  R4      pV! SP5                  VV4      4       SP7                  VVV4      pVS8w  d”   \&        P8                  ! 4       pSP;                  V\&        P2                  ! V4      4       \&        P<                  ! VP>                  4      pSPA                  VV4       \C        RVP>                  PE                  4        24      h\&        P8                  ! 4       pV! SPG                  V\&        P2                  ! V4      4      4       \&        P<                  ! VP>                  4      pV! SPI                  VV4      4       \&        P(                  ! 4       pV! SPK                  VV\&        P2                  ! V4      4      4       VP>                  e   VP>                  PE                  4       pMRpSPM                  \&        P2                  ! V4      4       VPN                  V3# u upi )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.
    cuda_include_dirs (list, None): List of directories containing CUDA headers
    nvcc_options (list, None): Additional options to pass to NVRTC
    auto_pch (bool): Enable automatic precompiled headers (CUDA 12.8+)

Returns:
    Tuple[bytes, str]: The compiled PTX code and mangled kernel name
Nc                ó(   € V ^8„  d   QhR\         RR/# rA   rC   )r   s   "r   r   Ú$_nvrtc_compile.<locals>.__annotate__  s   € ÷ 	?ñ 	?œCð 	? Dñ 	?r   c                 óü   <€ V S8w  dt   \         P                  ! 4       pSP                  V \         P                  ! V4      4       VP                  e   VP                  P                  4       MRp\        RV 24      hR # )NrE   rF   )r   rG   rp   rH   rI   rJ   r   )rB   rK   rM   ÚNVRTC_SUCCESSÚlibnvrtcs   &  €€r   Úcheck_nvrtcÚ#_nvrtc_compile.<locals>.check_nvrtc  so   ø€ Ø�]Ô"Ü—o’oÓ'ˆGØ×(Ñ(¨´·²¸gÓ1FÔGð —=‘=Ò,ð —‘×$Ñ$Ô&à)ð ô
  ¨m¨_Ð=Ó>Ð>ñ #r   úutf-8z--offload-arch=z--gpu-architecture=sm_)Úinclude_pathsr‰   z-Iz12.8zPCH requires CUDA 12.8+, got z--pchz[^\w.-]rg   z.cuzKernel compilation failed:
rl   )(Ú
torch.cudar’   Úencoder‰   Úget_device_propertiesÚcurrent_devicer   r+   ÚgcnArchNameÚmajorÚminorÚappendr˜   r°   r   ÚAssertionErrorr�   r™   r^   r   rG   ÚreÚsubÚc_void_prr   rH   r€   rv   Úc_size_tr|   Úcreate_string_bufferrI   r~   r   rJ   rx   rz   r‚   rt   Úraw) rŸ   r    r¡   r¢   r£   r¤   r*   r­   Úsource_bytesÚpropsÚoptionsr°   Úcuda_include_pathsÚ	cuda_pathÚ	directoryÚoptionÚnvrtc_compatible_flagsr›   Únum_optionsÚoptions_arrayÚsource_nameÚprogÚc_kernel_nameÚresÚlog_sizeÚlogÚbinary_sizeÚbinaryÚc_mangled_nameÚmangled_namer«   r¬   s    &&&&&&                        @@r   Ú_nvrtc_compilerÔ   ù   s   ù€ ó0 ô $Ó%€Hð €M÷	?ñ 	?ð !×'Ñ'¨Ó0€Lð Ò!Ø—
‘
×0Ñ0°·±×1JÑ1JÓ1LÓMˆØ�=‰=××ÐØ$)×$5Ñ$5Ð#6Ñà$)§K¡K =°·±°Ð!>Ðð €GØ‡}�}××ÐØ�‰˜Ð);Ð(<Ð=×DÑDÓFÕGà�‰Ð/Ð0BÐ/CÐD×KÑKÓMÔNõ 8á& vÓ.ÐÛ'ˆ	Ø�‰˜˜I˜;Ð'×.Ñ.Ó0Ö1ñ (÷ Û*ˆIØ�N‰N˜R 	˜{Ð+×2Ñ2Ó4Ö5ñ +÷ Üˆu�}‰}×!Ñ!Ó" VÔ+Ü Ð#@ÀÇÁ×ASÑASÐ@TÐ!UÓVÐVØÒØˆLØ×Ñ˜GÔ$÷ Û"ˆFØ�N‰N˜6Ÿ=™=¨Ó1Ö2ñ #ô ;Ó<ÐØ‡N�NÑ5KÓLÑ5K¨T�D—K‘K Ö(Ñ5KÑLÔMô �g“,€KÜ—_‘_ {Ö2°WÑ=€Mô —&’&˜ S¨+Ó6¸Õ>€Kô �?Š?Ó€DÙØ×#Ñ#Ü�LŠL˜ÓØØ×ÑÓ ØØØó	
ô	ð  ×&Ñ& wÓ/€MÙ�×/Ñ/°°mÓDÔEð ×
&Ñ
& t¨[¸-Ó
H€Cð ˆmÔä—?’?Ó$ˆØ×'Ñ'¨¬f¯lªl¸8Ó.DÔEÜ×)Ò)¨(¯.©.Ó9ˆØ×#Ñ# D¨#Ô.ÜÐ9¸#¿)¹)×:JÑ:JÓ:LÐ9MÐNÓOÐOô —/’/Ó#€KÙ�×*Ñ*¨4´·²¸kÓ1JÓKÔLÜ×(Ò(¨×):Ñ):Ó;€FÙ�×&Ñ& t¨VÓ4Ô5ô —_’_Ó&€NÙØ×$Ñ$ T¨=¼&¿,º,À~Ó:VÓWôð ×ÑÒ'Ø%×+Ñ+×2Ñ2Ó4‰àˆà× Ñ ¤§¢¨dÓ!3Ô4ð �:‰:�|Ð#Ð#ùòw Ms   È'Sc                   óD   a € ] tR tRt o V 3R lR ltV 3R lR ltRtV tR# )Ú_CudaModulei�  c                ó8   <€ V ^8„  d   QhRS[ P                  RR/# )r   Úmoduler   N©r   r¼   )r   Ú__classdict__s   "€r   r   Ú_CudaModule.__annotate__�  s   ø€ ÷ 3ñ 3™vŸ™ð 3°4ñ 3r   c                ó    € Wn         / V n        R # r>   )Ú_moduleÚ_kernels)ÚselfrØ   s   &&r   Ú__init__Ú_CudaModule.__init__�  s   € ØŒØ02ˆŽr   c                ó$   <€ V ^8„  d   QhRS[ RR/# )r   rb   r   Ú_CudaKernel)r   )r   rÚ   s   "€r   r   rÛ   ”  s   ø€ ÷ Vñ V¡ð V¨ñ Vr   c           	     ó°  € WP                   9   d   V P                   V,          # ^ RIHp V! 4       p\        P                  ! 4       p \        VP                  \        P                  ! V4      V P                  VP                  R4      4      4       \        W@P                  4      pWPP                   V&   V#   \         d   p\        RT R24      ThRp?ii ; i)r   )r?   r¯   zNo kernel named 'z' in this moduleN)rÞ   Útorch.cuda._utilsr?   r   r¼   rN   r1   rH   rÝ   r²   rã   r   ÚAttributeError)rß   rb   r?   rL   ÚfuncÚkernelr_   s   &&     r   Ú__getattr__Ú_CudaModule.__getattr__”  s¹   € Ø—=‘=Ô Ø—=‘= Õ&Ð&õ 	?á*Ó,ˆä�ŠÓ ˆð	VÜØ×+Ñ+Ü—L’L Ó&¨¯©°d·k±kÀ'Ó6Jóôô
 ! §|¡|Ó4ˆFØ"(�M‰M˜$ÑØˆMøäô 	VÜ Ð#4°T°FÐ:JÐ!KÓLÐRUÐUûð	Vús   ÁA-B5 Â5CÃ CÃC)rÞ   rÝ   N)Ú__name__Ú
__module__Ú__qualname__Ú__firstlineno__rà   ré   Ú__static_attributes__Ú__classdictcell__©rÚ   s   @r   rÖ   rÖ   �  s   ø‡ € ÷3ð 3÷Vö Vr   rÖ   c                   ó^   a € ] tR tRt o RtV 3R lR ltRV 3R lR lltV 3R lR	 ltR
tV t	R# )rã   i­  zL
Represents a compiled CUDA kernel that can be called with PyTorch tensors.
c                óR   <€ V ^8„  d   QhRS[ P                  RS[ P                  RR/# )r   rç   rØ   r   NrÙ   )r   rÚ   s   "€r   r   Ú_CudaKernel.__annotate__²  s*   ø€ ÷ 'ñ '™VŸ_™_ð '±f·o±oð 'È$ñ 'r   c                ó,   € Wn         W n        ^ V n        R# )r   N)rç   rØ   Ú_max_shared_mem_bytes)rß   rç   rØ   s   &&&r   rà   Ú_CudaKernel.__init__²  s   € ØŒ	ØŒØ%&ˆÖ"r   Nc                óŒ   <€ V ^8„  d   QhRS[ S[S[S[3,          RS[ S[S[S[3,          RS[R,          RS[RS[R,          RR/# )r   ÚgridÚblockÚargsNÚ
shared_memÚstreamr   )r¦   r   r”   r   )r   rÚ   s   "€r   r   rô   ·  sm   ø€ ÷ _
ñ _
á‘C™™c�MÕ"ð_
ñ ‘S™#™s�]Õ#ð_
ñ �T�kð	_
ñ
 ð_
ñ �d•
ð_
ð 
ñ_
r   c                óú  € ^ RI pVP                  P                  P                  4       pV'       g   . p. p. p	V EF|  p
\	        W¦P
                  4      '       d¢   V
P                  '       g4   V
P                  '       d   V
P                  4       '       g   \        R4      h\        P                  ! V
P                  4       4      pVP                  V4       V	P                  \        P                  ! V4      4       KÀ  \	        V
\        4      '       d?   \        P                   ! V
4      pV	P                  \        P                  ! V4      4       EK  \	        V
\"        4      '       d?   \        P$                  ! V
4      pV	P                  \        P                  ! V4      4       EKh  \'        R\)        V
4       24      h	  \        P                  \+        V	4      ,          ! 4       p\-        V	4       F,  w  rú\        P.                  ! V
\        P                  4      Wï&   K.  	  Vf   ^ RIpVP                  P3                  4       pVR
8¼  dW   V P4                  ^ 8X  g   W@P4                  8”  d6   V P4                  ^ 8X  d   RMRV P4                   R2p\7        RV RV R	24      h\9        VP;                  V P<                  V^ ,          V^,          V^,          V^ ,          V^,          V^,          VVP>                  VR4      4       R# )a¶  
Call the compiled CUDA kernel

Args:
    grid (tuple): Grid dimensions (grid_x, grid_y, grid_z)
    block (tuple): Block dimensions (block_x, block_y, block_z)
    args (list): List of arguments to pass to the kernel.
                 PyTorch tensor arguments will be automatically converted to pointers.
    shared_mem (int): Shared memory size in bytes
    stream (torch.cuda.Stream): CUDA stream to use. If None, uses current stream.
Nz?All tensor arguments must be CUDA tensors or pinned CPU tensorszUnsupported argument type: znot configuredzonly z bytes configuredzKernel requires z' bytes of shared memory (>= 48KB), but ze. Call kernel.set_shared_memory_config(shared_mem) after compilation and before launching the kernel.é À  ) r*   r‰   Ú_utilsr?   rV   ÚTensorÚis_cudaÚis_cpuÚ	is_pinnedÚ
ValueErrorr   r¼   Údata_ptrr¸   rH   r   Úc_intÚfloatÚc_doubleÚ	TypeErrorÚtyper^   Ú	enumerateÚcastr±   Úcurrent_streamrö   r   rN   r3   rç   Ú_as_parameter_)rß   rù   rú   rû   rü   rý   r*   rL   Úprocessed_argsÚc_argsÚargÚptrr  r	  Úc_args_arrayÚiÚconfigured_msgs   &&&&&&           r   Ú__call__Ú_CudaKernel.__call__·  sX  € ó& 	à—*‘*×#Ñ#×<Ñ<Ó>ˆçØˆDð 13ˆØˆäˆCÜ˜#Ÿ|™|×,Ò,Ø—{—{�{¨C¯J¯J¨J¸3¿=¹=¿?º?Ü$ØYóð ô —o’o c§l¡l£nÓ5�Ø×%Ñ% cÔ*Ø—‘œfŸlšl¨3Ó/Ö0Ü˜C¤×%Ò%äŸš SÓ)�à—‘œfŸlšl¨5Ó1×2Ü˜C¤×'Ò'ä!Ÿ?š?¨3Ó/�à—‘œfŸlšl¨8Ó4×5äÐ"=¼dÀ3»i¸[Ð IÓJÐJñ+ ô0 Ÿ™¬#¨f«+Ö5Ó8ˆÜ Ö'‰FˆAÜ$Ÿkšk¨#¬v¯©Ó?ˆL‹Oñ (ð Š>ãà—Z‘Z×.Ñ.Ó0ˆFð ˜Ô"Ø×&Ñ&¨!Ô+¨z×<VÑ<VÔ/Vð ×-Ñ-°Ô2ñ !à˜T×7Ñ7Ð8Ð8IÐJð ô
 Ø" : ,ð /Ø%Ð&ð '3ð3óð ô 	Ø×"Ñ"Ø—	‘	Ø�Q•Ø�Q•Ø�Q•Ø�a•Ø�a•Ø�a•ØØ×%Ñ%ØØóö	
r   c                ó$   <€ V ^8„  d   QhRS[ RR/# )r   Úshared_mem_bytesr   NrC   )r   rÚ   s   "€r   r   rô     s   ø€ ÷ ,6ñ ,6¹ð ,6Àñ ,6r   c                óÎ  € VR8  d	   Wn         R# \        4       p\        P                  P	                  4       p\        P
                  P                  '       d;   VP                  P                  R^4      ^ ,          pVR8X  d   RpMVR8X  d   RpMRpM\        VRR4      pW8”  d   \        RV R	V R
24      h^p\        VP                  V P                  VV4      4       Wn         R# )é0   NÚ:Úgfx950Úgfx1250i   Úshared_memory_per_block_optinrÿ   zRequested shared memory (z bytes) exceeds device limit (z= bytes). Consider reducing block size or shared memory usage.i € i   )rö   r?   r*   r‰   r³   r   r+   rµ   rŠ   Úgetattrr   rN   r5   rç   )rß   r  rL   Údevice_propsÚgcn_archÚmax_shared_memÚ+cudaFuncAttributeMaxDynamicSharedMemorySizes   &&     r   Úset_shared_memory_configÚ$_CudaKernel.set_shared_memory_config  s÷   € Ø˜iÔ'à)9Ô&Ùä*Ó,ˆô —z‘z×7Ñ7Ó9ˆä�=‰=××Ðð $×/Ñ/×5Ñ5°c¸1Ó=¸aÕ@ˆHØ˜8Ô#Ø!+‘Ø˜YÔ&Ø!+‘à!&‘ä$ØÐ=¸uóˆNð Ô,ÜØ+Ð,<Ð+=ð >!Ø!/Ð 0ð 1GðGóð ð 78Ð3ÜØ×&Ñ&Ø—	‘	Ø;Ø óô	
ð &6Ö"r   )rö   rç   rØ   )©é   r)  r)  r(  Nr   N)
rë   rì   rí   rî   Ú__doc__rà   r  r&  rï   rð   rñ   s   @r   rã   rã   ­  s+   ø‡ € ñ÷'ð '÷
_
ò _
÷B,6ö ,6r   rã   c          	      ó¦   € V ^8„  d   QhR\         \        ,          R\        \         ,          R,          R\        \        \         R3,          ,          /# )r   ÚptxÚkernel_namesNr   rã   )r   r]   r”   rÖ   Údict)r   s   "r   r   r   G  s@   € ÷ -ñ -Ü	Œu�ð-Ü$(¬¥I°Õ$4ð-ä”4œ˜]Ð*Õ+Õ+ñ-r   c           
     ót  € ^ RI p\        4       p\        V \        4      '       d   V P	                  R4      p \
        P                  ! 4       pVP                  P                  4       pV;_uu_ 4        \        VP                  \
        P                  ! V4      V 4      4       RRR4       V'       g   \        V4      # / pV Fc  p\
        P                  ! 4       p\        VP                  \
        P                  ! V4      WGP	                  R4      4      4       \        W„4      Wg&   Ke  	  V#   + '       g   i     L�; i)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.
Nr¯   )r±   r?   rV   r   r²   r   r¼   r‰   r  rN   r/   rH   rÖ   r1   rã   )	r,  r-  r*   rL   rØ   rý   Úkernelsrb   rç   s	   &&       r   Ú_cuda_load_moduler1  G  sí   € ó  ô 'Ó(€Gô �#”s×ÒØ�j‰j˜Ó!ˆô �_Š_Ó€Fà�Z‰Z×&Ñ&Ó(€Fß	ŠÜ�G×,Ñ,¬V¯\ª\¸&Ó-AÀ3ÓGÔH÷ 
÷ Ü˜6Ó"Ð"ð €GÛˆÜ�ŠÓ ˆÜØ×'Ñ'Ü—’˜TÓ" F¯K©K¸Ó,@óô	
ô
 $ DÓ1ˆ‹ñ ð €N÷! 
�ús   Á/0D'Ä'D7	c                óH   € V ^8„  d   QhR\         R\        R\        R\        /# )r   ÚdeviceÚoptionalÚ	allow_cpur   )r   r¥   r   )r   s   "r   r   r   w  s2   € ÷ @ñ @Üð@Üð@Ü48ð@äñ@r   c                ó$  € \        V \        4      '       d   V # \        V \        4      '       d   \        P                  ! V 4      p \        V \        P                  4      '       dH   V'       d!   V P
                  R9  d   \        RV  24      hMV P
                  R8w  d   \        RV  24      h\        P                  P                  4       '       g7   \        V \        P                  P                  4      '       d   V P                  # \        WV4      # )a�  Get 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‰   z(Expected a cuda or cpu device, but got: z!Expected a cuda device, but got: )r‰   Úcpu)rV   r   r   r*   r3  r  r  ÚjitÚis_scriptingr‰   ÚidxÚ_torch_get_device_index)r3  r4  r5  s   &&&r   r	   r	   w  sÉ   € ô  �&œ#×ÒØˆÜ�&œ#×ÒÜ—’˜fÓ%ˆÜ�&œ%Ÿ,™,×'Ò'ßØ�{‰{ /Ô1Ü Ð#KÈFÈ8Ð!TÓUÐUð 2à�[‰[˜FÔ"ÜÐ@ÀÀÐIÓJÐJÜ�9‰9×!Ñ!×#Ò#Ü�fœeŸj™j×/Ñ/×0Ò0Ø—:‘:ÐÜ" 6°YÓ?Ð?r   )NNNFr>   )FF)%r   rº   r(   Útypingr   r*   Útorch._vendor.packaging.versionr   Úcuda.bindingsr   r   r   rW   r   rZ   rU   r&   r   r+   Útorch._utilsr	   r;  r   r7   r;   r?   rN   rd   rh   r„   r�   r’   r�   rÔ   rÖ   rã   r1  r:   r   r   Ú<module>r@     sð   ðÛ Û 	Û 
Ý ã Ý 3ð÷ñ ð Ð÷ ˜%Ÿ-™-×+Ñ+Ò7ð !ÐØ!ÐØÐõ Fõ$õõ.+õ#õ	7õ(õVõ8õ:6õ&$õ÷0S$÷lVñ V÷<W6ñ W6÷t-÷`@ñ @øðK ô Ø!ÐØ ÐØ!ÐØÒð	ús   žC ÃCÃC