
    shK5                     j   % S r SSKrSSKrSSKrSSKJrJrJrJrJ	r	  SSK
r
SSKrSSKrSSKr\R                  R                  5       r\=(       a    \R                  R#                  5       S:  r\(       a  \R&                  " S5      OSr\(       a
  \" S 5      rO	\" S 5      r\" S 5      r\" S	 5      r\" S
 5      r\" S 5      r\" S 5      r\" S 5      r\" S 5      r\" S 5      r\" S 5      r\" S 5      r\" S 5      r \" S 5      r!S r"S r#\" S 5      r$\" S 5      r%S r&S r'S r(\" S 5      r)\*\+S'   \" S 5      r,\*\+S'   \" S 5      r-\*\+S '   \" S! 5      r.\*\+S"'   \=(       a    \(       + r/\*\+S#'   \" S$ 5      r0\*\+S%'   S& r1\" S' 5      r2\*\+S('   \" S) 5      r3\*\+S*'   \(       a    SSK4r5\5R                  Rm                  5       r7OS+r7S+q9S, r:\Rv                  S- 5       r<\Rv                  S;S. j5       r=\Rv                  S/ 5       r>S;S0 jr?S1 r@S2 rAS3 rBS4 rCS5 rDS6 rE\D" 5       rF\E" 5       rGS7\R                  R                  S4S8 jrJS7\R                  \R                  R                  S4S9 jrLS: rM\(       d"  \R                  R                  5       (       a   egg! \8 a    S+r7S+r Nf = f)<z>This file is allowed to initialize CUDA context when imported.    N)LazyVal
TEST_NUMBATEST_WITH_ROCM	TEST_CUDA
IS_WINDOWS   zcuda:0c                      [         $ N)r        w/Users/tiagomarins/Projetos/claudeai/copy_bank/venv/lib/python3.13/site-packages/torch/testing/_internal/common_cuda.py<lambda>r      s    r   c                      [         =(       a@    [        R                  R                  R	                  [        R
                  " S[        S95      $ )N      ?device)r   torchbackendscudnnis_acceptabletensorCUDA_DEVICEr   r   r   r   r      s1    !wu~~/C/C/Q/QRWR^R^_ajuRv/w!wr   c                  l    [         (       a(  [        R                  R                  R	                  5       $ S$ )Nr   )
TEST_CUDNNr   r   r   versionr   r   r   r   r      s$    zzU^^%9%9%A%A%C%XWX%Xr   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ )N)      r   cudais_availableget_device_capabilityr   r   r   r   r      ,    ejj557hEJJ<\<\<^bh<hhr   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ )N)   r   r   r   r   r   r   r      r#   r   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ )N)   r   r   r   r   r   r   r      r#   r   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ )N)r'   r   r   r   r   r   r   r      r#   r   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ )N)   r   r   r   r   r   r   r       r#   r   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ Nr*   	   r   r   r   r   r   r   !   r#   r   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ )N)r.   r   r   r   r   r   r   r   "   r#   r   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:  $ )N)
   r   r   r   r   r   r   r   #   s,    uzz668jUZZ=]=]=_cj=jjr   c                      [         R                  R                  5       =(       aO    [         R                  R                  5       S   S:H  =(       a$    [         R                  R                  5       S   S:  $ )Nr   r1      r   r   r   r   r   r   %   sX    %**113 @

8X8X8Z[\8]ac8c @jj668;a?@r   c                      [         R                  R                  5       =(       a.    [         R                  R                  5       S;   =(       d    [        $ )N))r'   r   )r*   r'   )r   r    r!   r"   IS_THORr   r   r   r   r   '   s2    EJJ335}5::;[;[;]aq;q;|u|}r   c                      [         R                  R                  5       =(       a!    [         R                  R                  5       S:H  $ r,   r   r   r   r   r   r   (   s,    %**113d

8X8X8Z^d8ddr   c                     ^  [         (       a>  [        R                  R                  S5      R                  m [        U 4S jS 5       5      $ g)Nr    c              3   ,   >#    U  H	  oT;   v   M     g 7fr
   r   ).0archgcn_arch_names     r   	<genexpr>CDNA2OrLater.<locals>.<genexpr>-   s     J5IT=(5Is   >   gfx90agfx942F)r   r   r    get_device_propertiesgcnArchNameany)r;   s   @r   CDNA2OrLaterrC   *   s7    ~

88@LLJ5IJJJr   c                     [         R                  R                  5       (       d  g[         R                  R                  S5      R                  n[
        R                  R                  SU5      nX :H  $ )NFr    /PYTORCH_DEBUG_FLASH_ATTENTION_GCN_ARCH_OVERRIDE)r   r    r!   r@   rA   osenvironget)matching_archr;   r:   s      r   evaluate_gfx_arch_exactrJ   0   sP    ::""$$JJ44V<HHM::>>K][D  r   c                      [        S5      $ )Ngfx90a:sramecc+:xnack-rJ   r   r   r   r   r   7       67OPr   c                      [        S5      $ )Ngfx942:sramecc+:xnack-rM   r   r   r   r   r   8   rN   r   c                      [         (       a  [        S5      =(       d    [        S5      $ [        (       a  [        (       + =(       a    [        $ g)NrL   rP   F)r   rJ   r   r   SM80OrLaterr   r   r   *evaluate_platform_supports_flash_attentionrS   :   s4    ~&'?@uD[\tDuuy~-+-r   c                  l    [         (       a  [        S5      =(       d    [        S5      $ [        (       a  gg)NrL   rP   TF)r   rJ   r   r   r   r   .evaluate_platform_supports_efficient_attentionrU   A   s*    ~&'?@uD[\tDuuyr   c                  R    [         (       + =(       a    [        =(       a	    [        S:  $ )Ni_ )r   rR   TEST_CUDNN_VERSIONr   r   r   *evaluate_platform_supports_cudnn_attentionrX   H   s    QKQ5G55PQr   c                      [        5       $ r
   )rS   r   r   r   r   r   K       :d:fr   !PLATFORM_SUPPORTS_FLASH_ATTENTIONc                      [        5       $ r
   )rU   r   r   r   r   r   L   s    <j<lr   #PLATFORM_SUPPORTS_MEM_EFF_ATTENTIONc                      [        5       $ r
   )rX   r   r   r   r   r   M   rZ   r   !PLATFORM_SUPPORTS_CUDNN_ATTENTIONc                  B    [         =(       d    [        =(       d    [        $ r
   )r[   r_   r]   r   r   r   r   r   O   s    :[ ;V2S;V2U;Vr   !PLATFORM_SUPPORTS_FUSED_ATTENTIONPLATFORM_SUPPORTS_FUSED_SDPAc                  (    [         =(       a    [        $ r
   )r   rR   r   r   r   r   r   U   s    y/H[/Hr   PLATFORM_SUPPORTS_BF16c                  2   [         R                  R                  5       (       a  [         R                  R                  (       a  [        S [         R                  R                  R                  S5      S S  5       5      n S/nU S:  a  UR                  S/5        U S:  a  UR                  S5        U H2  nU[         R                  R                  S	5      R                  ;   d  M2    g
   g[        =(       d!    [         R                  R                  5       S:H  $ g)Nc              3   8   #    U  H  n[        U5      v   M     g 7fr
   int)r9   vs     r   r<   1evaluate_platform_supports_fp8.<locals>.<genexpr>Z   s      R1QAQ1Q   .r   gfx94)r%   r   gfx120)r%   r   gfx95r   Tr-   F)r   r    r!   r   hiptuplesplitextendappendr@   rA   SM90OrLaterr"   )ROCM_VERSIONarchsr:   s      r   evaluate_platform_supports_fp8rx   W   s    zz  ==  R1B1B1H1H1Mbq1Q RRLIEv%hZ(v%W%5::;;A>JJJ 
  N%**"B"B"D"NNr   c                      [        5       $ r
   )rx   r   r   r   r   r   g   s    .L.Nr   PLATFORM_SUPPORTS_FP8c                  (    [         =(       a    [        $ r
   )r   SM100OrLaterr   r   r   r   r   i   s    )2L2Lr   PLATFORM_SUPPORTS_MX_GEMMFc                      [         (       d   S5       e[        (       dI  [        [        R                  R                  5       5       H  n [        R                  " SSU  3S9  M     Sqg g )Nz?CUDA must be available when calling initialize_cuda_context_rngr3   zcuda:r   T)r   __cuda_ctx_rng_initializedranger   r    device_countrandn)is    r   initialize_cuda_context_rngr   {   sS    9WWW9%%uzz..01AKKE!+. 2%)"	 &r   c               #     #    [         R                  R                  R                  R                  n  S[         R                  R                  R                  l        [         R                  R
                  R                  S S S SS9   S v   S S S 5        U [         R                  R                  R                  l        g ! , (       d  f       N8= f! U [         R                  R                  R                  l        f = f7f)NFenabled	benchmarkdeterministic
allow_tf32r   r   r    matmulr   r   flagsold_allow_tf32_matmuls    r   tf32_offr      s     !NN//66AAF05""-^^!!''TXej'k l 1F""- lk 1F""-5   /C;AC B<
C *C;<
C
C +C88C;c              #   6  #    [         R                  R                  R                  R                  nU R
                  n S[         R                  R                  R                  l        Xl        [         R                  R                  R                  S S S SS9   S v   S S S 5        U[         R                  R                  R                  l        X0l        g ! , (       d  f       N>= f! U[         R                  R                  R                  l        X0l        f = f7f)NTr   )r   r   r    r   r   	precisionr   r   )selftf32_precisionr   old_precisions       r   tf32_onr      s     !NN//66AANNM'04""-'^^!!''TXei'j k 1F""-&	 kj 1F""-&s5   ;DAC% CC% $0D
C"C% %1DDc               #     #    [         R                  R                  R                  R                  n  S[         R                  R                  R                  l        [         R                  R
                  R                  SSSSS9   Sv   SSS5        U [         R                  R                  R                  l        g! , (       d  f       N8= f! U [         R                  R                  R                  l        f = f7f)z}
Context manager to temporarily enable TF32 for CUDA operations.
Restores the previous TF32 state after exiting the context.
TNr   r   r   s    r   tf32_enabledr      s      "NN//66AAF04""-^^!!''D ( 
 

 1F""-
 

 1F""-r   c                 ,   ^ ^^ S mU 4S jmUU4S jnU$ )Nc                 Z    [        5          U" 5         S S S 5        g ! , (       d  f       g = fr
   r   )r   function_calls     r   with_tf32_disabled+tf32_on_and_off.<locals>.with_tf32_disabled   s    ZO ZZs   
*c                 `   > [        U T5         U" 5         S S S 5        g ! , (       d  f       g = fr
   )r   )r   r   r   s     r   with_tf32_enabled*tf32_on_and_off.<locals>.with_tf32_enabled   s    T>*O +**s   
-c                    >^ ^ [         R                  " T 5      R                  n[        UR	                  5       5      m[
        R                  " T 5      UU UU4S j5       nU$ )Nc                    >^ [        TU 5       H
  u  p#UTU'   M     [        R                  R                  5       nST;   a/  U=(       a&    [        R                  " TS   5      R
                  S:H  nST;   a/  U=(       a&    TS   [        R                  [        R                  1;   nU(       a#  T" TS   UU4S j5        T" TS   UU4S j5        g T" S0 TD6  g )Nr   r    dtyper   c                     > T " S0 TD6$ Nr   r   fkwargss   r   r   Ctf32_on_and_off.<locals>.wrapper.<locals>.wrapped.<locals>.<lambda>   s    1;v;r   c                     > T " S0 TD6$ r   r   r   s   r   r   r      s    !+f+r   r   )zipr   r    is_tf32_supportedr   typefloat32	complex64)	argsr   kri   cond	arg_namesr   r   r   s	    `   r   wrapped1tf32_on_and_off.<locals>.wrapper.<locals>.wrapped   s    It,q	 -:://1D6!OfX.>!?!D!D!N& UU]]EOO4T!T"6&>3FG!&.2EFFr   )inspect	signature
parametersrq   keys	functoolswraps)r   paramsr   r   r   r   s   `  @r   wrapper tf32_on_and_off.<locals>.wrapper   sK    ""1%00&++-(				 
	 r   r   )r   r   r   r   s   ` @@r   tf32_on_and_offr      s    ( Nr   c                 F   ^  [         R                  " T 5      U 4S j5       nU$ )Nc                  ^   > [        5          T" U 0 UD6sS S S 5        $ ! , (       d  f       g = fr
   r   )r   r   r   s     r   r   with_tf32_off.<locals>.wrapped   s    Zd%f% ZZs   
,)r   r   )r   r   s   ` r   with_tf32_offr      s%    __Q& & Nr   c                  b   S[         R                  R                  5       ;  a  g[         R                  R                  5       R                  S5      n [         R                  R                  5       U [	        S5      -   S  R                  S5      S   n[        S UR                  S5       5       5      $ )NMagmar   r   zMagma 
r   c              3   8   #    U  H  n[        U5      v   M     g 7fr
   rg   r9   xs     r   r<   %_get_magma_version.<locals>.<genexpr>   s     8!7AQ!7rk   rl   )r   
__config__showfindlenrr   rq   )positionversion_strs     r   _get_magma_versionr      s    e&&++--$$&++H5H""'')(S]*B*CDJJ4PQRSK8!2!23!7888r   c                      [         R                  R                  c  g[        [         R                  R                  5      n [	        S U R                  S5       5       5      $ )Nr   c              3   8   #    U  H  n[        U5      v   M     g 7fr
   rg   r   s     r   r<   *_get_torch_cuda_version.<locals>.<genexpr>       9!8AQ!8rk   rl   )r   r   r    strrq   rr   )cuda_versions    r   _get_torch_cuda_versionr      sE    }}!u}}))*L9!3!3C!8999r   c                      [         (       d  g[        [        R                  R                  5      n U R                  S5      S   n [        S U R                  S5       5       5      $ )Nr   -r   c              3   8   #    U  H  n[        U5      v   M     g 7fr
   rg   r   s     r   r<   *_get_torch_rocm_version.<locals>.<genexpr>
  r   rk   rl   )r   r   r   r   rp   rr   rq   )rocm_versions    r   _get_torch_rocm_versionr     sO    >u}}(()L%%c*1-L9!3!3C!8999r   c                      [         (       + $ r
   )r   r   r   r   !_check_cusparse_generic_availabler     s    r   c                  4   [         (       d  g[        R                  R                  (       d  g[	        [        R                  R                  5      n U R                  S5      S   n [        S U R                  S5       5       5      nUS L =(       d    US:  (       + $ )NFr   r   c              3   8   #    U  H  n[        U5      v   M     g 7fr
   rg   r   s     r   r<   5_check_hipsparse_generic_available.<locals>.<genexpr>  s     G/F!s1vv/Frk   rl   )r   r3   )r   r   r   rp   r   rr   rq   )r   rocm_version_tuples     r   "_check_hipsparse_generic_availabler     sx    >==u}}(()L%%c*1-LG|/A/A#/FGG"d*I.@6.IJJr   r    c                    [         R                  R                  [         R                  R                  SS5      [         R                  R                  SS5      5      R	                  U S9n[         R                  R                  [         R                  R                  SS5      [         R                  R                  SS5      5      R	                  U S9n[         R
                  " 5          [        UR                  5       UR                  5       5       H  u  pVUR                  U5        M     S S S 5        SS0nUb  UR                  U5        U" UR                  5       40 UD6nU" UR                  5       40 UD6n	X4X4$ ! , (       d  f       NX= f)Nr*   r   lrr   )
r   nn
SequentialLineartono_gradr   r   copy_update)
r   optimizer_ctoroptimizer_kwargsmod_controlmod_scalingcsr   opt_controlopt_scalings
             r   !_create_scaling_models_optimizersr     s,    ((%%ehhooa&;UXX__QPQ=RSVV^dVeK((%%ehhooa&;UXX__QPQ=RSVV^dVeK	..0+2H2H2JKDAGGAJ L 
 C[F#&' !7!7!9DVDK !7!7!9DVDK[== 
s   (AE==
Fc           
         [         R                  " SXS9[         R                  " SXS94[         R                  " SXS9[         R                  " SXS94[         R                  " SXS9[         R                  " SXS94[         R                  " SXS9[         R                  " SXS94/n[         R                  R                  5       R	                  U 5      nSn[        XUS9XEU4-   $ )N)r*   r*   )r   r   r   )r   r   r   )r   r   r   MSELossr   r   )r   r   r   r   dataloss_fn	skip_iters          r   _create_scaling_caser   1  s    [[u<ekk&X]>mn[[u<ekk&X]>mn[[u<ekk&X]>mn[[u<ekk&X]>mnpD
 hh ##F+GI,GW		"# #r   c                 H    [         (       d  U $ [        R                  " U 5      $ r
   )IS_SM89unittestexpectedFailure)funcs    r   xfailIfSM89r  @  s    w4BH$<$<T$BBr   )gh㈵>)N__doc__r   r   
torch.cuda$torch.testing._internal.common_utilsr   r   r   r   r   r   
contextlibrF   r  r    is_initialized"CUDA_ALREADY_INITIALIZED_ON_IMPORTr   TEST_MULTIGPUr   r   r   rW   SM53OrLaterSM60OrLaterSM70OrLaterSM75OrLaterrR   SM89OrLaterru   r|   r5   	IS_JETSONr  rC   rJ   GFX90A_ExactGFX942_ExactrS   rU   rX   r[   bool__annotations__r]   r_   ra   rb   rd   rx   rz   r}   
numba.cudanumbar!   TEST_NUMBA_CUDA	Exceptionr   r   contextmanagerr   r   r   r   r   r   r   r   r   r   TEST_CUSPARSE_GENERICTEST_HIPSPARSE_GENERICoptimSGDr   floatr   r  r   r   r   <module>r"     s   F    k k   	  &+ZZ%>%>%@ " <ejj5571<(1ell8$t*+JwxJXY hihihihihihihijk
 @ A}~	
d
e! PQPQR +22f*g !4 g,34l,m #T m*12f*g !4 g*1 3V +W !4 W &/%E~3E d E&'HI  I  &&NO t O")*L"M 4 M**113
 O # * F F 
' 
' F FXJ9::	K :; ;=  .4EKKOOfj >$ !'ekk%++//lp #C
 *zz((***** *m  
s   2J% %
J21J2