
    i             -       v   d dl Z d dlmZ d dlmZ d dlmZ d dlmZm	Z	m
Z
mZ d dlmZ d dlZd dlmZ d dlmZmZmZ d dlmZmZmZmZ d d	lmZ d d
lmZmZ d dlmZm Z m!Z!m"Z"m#Z#m$Z$m%Z%m&Z&m'Z'm(Z(m)Z) d dl*m+Z+m,Z,m-Z-m.Z.m/Z/ d dl0m1Z1m2Z2 d dl3m4Z5 d dl6m7Z8  e
d      Z9 ed      Z:ejv                  jx                  Z<ejz                  j}                  ddd      Z? e@d      \  ZAZBZCdeee:e9f   gee:e9f   f   fdZDde#fdZEd ZFd ZG eDe<j                  e<j                  g       e/       dddej                  ddfd              ZK eDe<j                  j                  e<j                  j                  g       e/       d               ZO eDe<j                  j                  e<j                  j                  g       e/       dd d!              ZP eDe<j                         e/       d"               ZQ eDe<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  g       e/d#d$      d%               ZT eDe<j                  j                  e<j                  j                  g       e/       d&               ZUd' ZVd$d(ed)eWeX   d*eYfd+ZZ eDe<j                  j                  e<j                  j                  g       e/       d,               Z\dZ]d)eWeX   fd-Z^ eDe<j                  j                  e<j                  j                  g       e/       d.               Z` eDe<j                  j                        dd/d0       Zc eDe<j                  j                        ej                  dddd1d2       Ze eDe<j                  j                  e<j                  j                  g       e/       ej                  dddd1d3              Zg eDe<j                  j                  e<j                  j                  g       e/       ej                  dddd1d4              Zj eDe<j                  j                  e<j                  j                  g       e/       ddddd1d5              Zl eDe<j                  j                  e<j                  j                  g       e/       d(ed)eWeX   d6eXd7eXfd8              Zn eDe<j                  j                        d$d9       Zpd: Zq eDe<j                  j                        d;        Zs eDe<j                        	 	 	 d%d<ed=ed>ed?e	e   d@e	eu   dAe	ej                     fdB       Zw eDe<j                        	 d&dCedDedEedAe	ej                     fdF       Zy eDe<j                        dGdGddHd<edCedDedEedAe	ej                     f
dI       Z{ eDe<j                        	 	 	 	 	 	 	 d'dJej(                  dKej(                  d?e	e   dLe	e   dAe	ej                     dMeYdNeXdOeXdPeXfdQ       Z} eDe<j                  j                        dRdSd(ed)eXdTedUej(                  dVeudWeYdefdX       Z eDe<j                   j                        dRdSd(ed)eXdTedUej(                  dVeudWeYdefdY       Z e/        eDe<j                  j                        dZ               Z eDe<j                  j                        dddd ddd[d\edVeud]e	e   d$e	e   d^e	e   d_eXd`eYdefda       Z eDe<j                  j                  e<j                  j                  g       e/       db               Z eDe<j                  j                        d$dc       Z eDe<j                  j                  e<j                  j                  g       e/       dd               Z eDe<j                  j                        d$de       Z eDe<j                  j                        df        Z eDe<j                  j                        dg        Z eDe<j"                  j                        dh        Z eDe<j"                  j&                        di        Z eDe<j*                  j                        dj        Z eDe<j.                  j                        ddddddkdl       Z eDe<j2                  j                        d(dm       Z eDe<j4                  j                        d%dn       Z eDe<j8                  j                        d(do       Z eDe<j:                  j                        dp        Z eDe<j>                  j&                        dq        Zd(edreufdsZd(edtedueufdvZ	 d)dwedreudxeYfdyZd*dtedreudzeufd{Zdted|ed}eYdreufd~Z	 d+deuded<edeufdZdeufdZ eDe<jP                  j                  e<jP                  jR                  g       e/dd      d,dtedeudeYfd              Z eDe<jV                  j                  e<jX                  j                  g       e/       d<edefd              Z eDe<j\                  g       e/dd      d<efd              ZdedefdZ eDe<jb                         e/       d(edtedeYdefd              Z eDe<jd                         e/       d$d(edtedeYdefd              Z eDe<jf                         e/       d$d(edeYdefd              Z eDe<jh                         e/       d$d(edeYdefd              Z eDe<jj                  j                        d-dtedeYdeYfd       Z eDe<jl                  j                  e<jl                  j                  g       e/       d<ededefd              Z eDe<jn                  j                        d$dtedeYfd       Z eDe<jr                  j                  e<jr                  j                  g       e/ddd      dddd(edeYdeYdeeeef   fd              Z eDe<jx                  j                  e<jx                  j                  g       e/       dddeded|edeYdef
d              Z eDe<j|                  j                  e<j|                  j                  g       e/ddd      dRddtedeYdeeeef   fd              Z eDe<j                  j                  e<j                  j                  g       e/ddd      dRdddtedeYdeYdeeeef   fd              Z eDe<j                  j                  e<j                  j                  g       e/       dRdddeded|ed}eYdeYdefd              Z eDe<j                         e/ddd      	 	 d.dededeYdeYdeeeef   f
d              ZdeudeeYeYf   fdZ eDe<j                  j                  e<j                  j                  g       e/dd      d/dtedeudeeef   fd              Z eDe<j                  j                  e<j                  j                  g       e/dddd      dtedeeeeef   fd              Z eDe<j                  j                        	 	 	 d0dtedeYdeYde	eu   fd       ZdededeeWeX   eWeX   f   fdZdededue	eu   deeef   fdZd<ededeYfdZ eDe<j                        dRdddddddted|ed}eYdeYde	e   de	e   de	e   de	e   deeeeef   fd       Z eDe<j                  j                  e<j                  j                  g      dRddddted|edeYd}eYdeYde	e   defdń       Z eDe<j                         e/dddRȫ      	 	 	 d1d(edtedeYdeYdeYdeeef   fdʄ              Z eDe<j                  j                        d˄        Z eDe<j                         e/       	 	 d2d<ededed}eYdeYdefd̄              Zd̈́ Zd΄ Z eDe<j                         e/       dτ               Z eDe<j                         e/       dЄ               Zdф Z eDe<j                         e/dҫ      dӄ               Z eDe<j                         e/dҫ      dԄ               ZdՄ Z eDe<j                         e/       dք               Z eDe<j                         e/       dׄ               Z eDe<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  g       e/dҫ      d؄               Zdل Z eDe<j                         e/       dڄ               Z eDe<j                         e/       dۄ               Z eDe<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  g       e/dҫ      d܄               Z eDe<j                         e/       d3d(ededefdބ              Z eDe<j                         e/       ded(edededef
d              Z eDe<j                  j                  e<j                  j                  g       e/dRȫ      dGdGdd              Z eDe<j                  j                  e<j                  j                  g       e/       dd/d              Z eDe<j                  j                        d4d       Z eDe<j                  j                        d4d       Z  eDe<j                  j                  e<j                  j                  g       e/       d&d              Z eDe<j                  j                        	 	 d-d       Z eDe<j
                         e/dRȫ      d               Zd Zd5dZ	 d&dej(                  d=ej(                  deeWeX   eXf   deeWeX   eXf   deeWeX   eXf   deYdeXde	eeWeX   eXf      fdZ	d Z
 eDe<j                  j                        dej(                  d=ej(                  d?e	ej(                     de	ej(                     de	ej(                     deYdedefd       Z eDe<j                  j                        dej(                  d=ej(                  d?ej(                  deWeX   deWeX   deWeX   deYdeWeX   deXfd       Zej                  j                   rejz                  j}                  ddd      Z eDejv                  j$                  j&                  j                        d        Z eDejv                  j$                  j*                  j                        d        Zej                  j.                  rNejz                  j}                  ddd      Z eDejv                  j2                  j4                        d         Zejz                  j}                  ddd      Z eDejv                  j:                  j<                  j                         eDejv                  j:                  j>                  j                        d               Z  eDejv                  j:                  j<                  jB                        d        Z" eDejv                  j:                  jF                  j                         eDejv                  j:                  jF                  jH                        d               Z% eDejv                  j:                  jF                  jB                         eDejv                  j:                  jF                  jL                        d               Z' eDejv                  j:                  jP                  j                         eDejv                  j:                  jR                  j                        d               Z*ejz                  j}                  ddd      Z+ eDejv                  jX                  jZ                        	 	 	 	 d6d
       Z. eDejv                  jX                  j^                        d        Z0d Z1 eDe<jd                  j                        	 	 	 	 	 d7d       Z3d Z4 eDe<jj                  j                        d        Z6 eDe<jn                         e/       	 	 	 	 	 d7d              Z8 eDe<jr                         e/dҫ      d               Z: eDe<jv                  j                        d        Z< eDe<jz                  j                        d        Z> eDe<j~                  j                        d        Z@ eDe<j                         e/dҫ      d               ZBdedzeufdZC eDe<j                         e/dd$      d               ZE eDe<j                         e/dҫ      d               ZG eDe<j                         e/dd$      d               ZI eDe<j                         e/dҫ      d               ZK eDe<j                  j(                        d&d       ZM eDe<j                  j                  e<j                  j                  g       e/       d               ZO eDe<j                  j                  e<j                  j                  g       e/       dddeXfd               ZP eDejv                  jx                  j                  j                  ejv                  jx                  j                  j                  g       e/       d!               ZQ eDe<j                  j(                  e<j                  j(                  g      d"        ZT eDe<j                  j                  g      d#        ZV eDe<j                  j                  e<j                  j                  g       e/dRȫ      dGdGdd$              ZX eDe<j                  j(                  g      d%        ZZ eDe<j                  j                  e<j                  j                  g      ddd&d'       Z] eDe<j                  j                  g      ddd&d(       Z_ eDe<j                  g       e/       d)               Za eDe<j                  g      d*        Zc eDe<j                  g      d+        Ze eDe<j                  g      d,        Zg eDe<j                  g      d-        Zi eDe<j                  g      d.        Zjd/eXd0eXdeXfd1Zkd2 Zl eDe<j                  g      d?e	e   fd3       Zn eDe<j                  g      d4        Zp eDe<j                  g      d5        Zr eDe<j                  j                        d6        Zt eDe<j                         e/       d7               Zv eDe<j                  j                        	 	 	 	 	 	 d8d8       Zx eDe<j                  j                        d9        Zzd)d:Z{ eDe<j                  j                  e<j                  j                  g       e/       d9dd;d<              Z} eDe<j                  j                  e<j                  j                  g      d=        Z eDe<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  g       e/d#d$      d:d>              Z eDe<j
                  j                        d?        Z eDe<j                  j                        d@        Z eDe<j                  j                        dA        Z eDe<j                  j                  e<j                  j                  e<j                  j(                  e<j                  j(                  e<j                  j                  e<j                  j                  e<j                   j                  g      dB        Z eDe<j$                  j                  e<j&                  j                  e<j$                  j(                  e<j&                  j(                  g      d	dC       Z eDe<j*                  j                  e<j,                  j                  g      d	dD       Z eDe<j0                  j                  e<j0                  j2                  g      dE        ZdF Z eDe<j8                  j(                  e<j8                  j                  g      dG        Z eDe<j<                  j(                  e<j<                  j                  g      dH        Z eDe<j@                  j                        dI        Z eDe<jD                  j(                  e<jD                  j                  g      dJ        Z eDe<jH                  j(                  e<jH                  j                  g      dK        Z eDe<jL                  j                        dL        Z eDe<jP                  j(                         e/       d	defdM              Z eDe<jT                  g       e/       	 d;dN              Z eDe<jX                  g      	 d;dO       Z eDe<j\                  g      	 d;dP       Z eDe<j`                  j                  e<jb                  j                  g      d$dQ       Z eDe<jf                  j                        dR        Z eDe<jj                  j                        dS        Z eDe<jn                        dT        Z eDe<jr                         e/       dU               Z eDe<jv                        dV        Z eDe<jz                  j                        d$dW       Zd(dXZ eDe<j                  j                        dY        Z eDe<j                  j                        dZ        Zd[ ZÐd\ ZĐd] ZŐd^ Z	 d$d<ed_eXd`eXdaeXdbeXdceXddeXdeeXdfeXdgeXdheXdieXdjeXdkeXdleXdmeXdneXdoeXdpeXdqeXdeudreYf,dsZǐdt Zd<eded_eXd`eXdaeXdbeXdceXddeXdeeXdfeXdgeXdheXdleXdmeXdneXdoeXdpeXdqeXdeuf&duZɐdv Z eDe<j                  j                        dw        Z eDe<j                  j                        	 	 	 	 d6dx       Z eDe<j                  j                        dy        Z eDe<j                         e/dd$      	 	 	 	 d6dz              Z eDe<j                         e/dҫ      d{               Zd<ed|efd}Z G d~ de      Zd<ed|edeXfdZ eDe<j                  j                        d        Z eDe<j                         e/       d               Z eDe<j                         e/dҐd      d               Z eDe<j                  j                  g      d        Z eDe<j                  j                        	 	 	 	 	 d<d       Z eDe<j                  j                  e<j                  j                  g       e/       ddddddd              Z eDe<j                  j                  e<j                  j                  g       e/       ddddddd              Z eDe<j                  j                        d        Z eDe<j                  j                        d=d       Zd)d)eXdeXdeYfdZd Zd Z eDe<j                  j                        d$d       Zd$dZd&dZd Zd&dZd>dZ eDe<j                  j                        d        Z eDe<j                        d        Z eDe<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  g       e/       d&d              Z eDe<j                  j                  e<j                  j                  e<j                  j                  e<j                  j                  g      d&d       Z eDe<j                  g      	 	 	 	 d?dededededeYdeYde	e   fd       ZdedeeXdf   fdZ eDe<j                  g      	 	 	 	 d?dededede	e   deYdedeYdeYde	e   fd       Z  eDe<j                  g      	 	 	 	 	 d@dededede	e   dedeYdeYde	e   fd       Z eDe<j                  g      	 d&dededededededededeXdeXdedeYdedede	e   fd       Z eDe<j
                  g      	 	 	 	 dAdededededeYde	e   de	e   fd       Z eDe<j                  g      	 	 d(dedededededededeYde	e   de	e   fd       Z eDe<j                  g      	 	 	 	 	 dBdededede	e   dedeYde	e   de	e   deeef   fd       Z
 eDe<j                  g      	 	 	 dCdededede	e   deYdeYde	e   fd       Z eDe<j                  g      	 	 dDdedededede	e   dedededededeWeY   deYde	e   fd       Z eDe<j                  g      	 d&dedededededededededededeXdeXdedeYde	e   f d       Z eDe<j"                  g      	 	 	 	 	 d<dededede	e   de	e   deXdeXdedeYdeYde	e   de	eX   de	eX   de	e   de	e   fd       Z eDe<j&                  g      	 	 	 d%dededededededededeXdeXdedeYdedede	e   de	eX   de	eX   f"d       Z eDe<j*                  g      	 	 	 	 	 dEdededed?e	e   de	e   de	e   de	eX   de	eX   dedeXdeYde	e   de	e   de	e   de	eX   fd˄       Z eDe<j.                  g      	 	 	 d>dedededed?e	e   de	e   de	e   dej0                  dej0                  dededededeXdeYde	e   de	eX   deYf$dτ       Z eDe<j4                  j                  g      	 	 	 	 dFd(ej(                  dEej(                  dej(                  dej(                  d?e	ej(                     de	ej(                     dAe	ej                     deYfdԄ       Z eDe<j8                  j:                  e<j8                  j<                  g       e/       d)dՄ              Z eDe<j@                  j:                        d)dք       Z! eDe<jD                  j                  e<jD                  j                  g       e/       d$dd/dׄ              Z#d؄ Z$dل Z% eDe<jL                  j                  e<jN                  j                  g      d&dڄ       Z& eDe<jP                  j                  e<jR                  j                  g      d(dۄ       Z( eDe<jT                  j                  e<jV                  j                  g      	 	 d(dedeeeXej0                  f      deeeXej0                  f      de	e   de	e   f
d       Z* eDe<jX                  j                  e<jZ                  j                  g      d%d       Z, eDe<j\                  j                  e<j\                  j^                  e<j\                  j                  e<j\                  j`                  g      dGd       Z1d Z2 eDe<jf                  j                        	 	 d(d       Z4 eDe<jj                  j                        d        Z5 eDe<jl                  j                        d        Z6d Z7d Z8 eDe<jr                  j                  e<jt                  j                  g      d9d       Z; eDe<jx                  j                        dHd       Z< eDe<jz                  j                        dId       Z> eDe<j~                         e/       	 dJd              Z@ eDe<j                  j                  e<j                  j                  g       e/d#d$      d:d              ZBej                  ZDd ZE eDe<j                  j                        d        ZF eDe<j                  j                        d        ZG eDe<j                  j                        d        ZI eDe<j                  j                        d        ZJ eDe<j                  j(                  e<j                  j                  g       e/       dddd              ZM eDe<j                  g       e/       dKd              ZO eDe<j                  j                  e<j                  j                  g      	 	 d(d       ZR eDe<j                  j                  g      	 	 d(d       ZT eDe<j                  j                        d        ZU eDe<j                  j                  e<j                  j                  g       e/       d%d              ZV eDejv                  jx                  j                        d        ZW eDejv                  jx                  j                        d        ZX eDe<j                         e/       dddddd              ZZd Z[ eDe<j                        d        Z] eDe<j                        	 dLd        Z_ eDe<j                        	 dLd       Za eDe<j                        	 dLd       Zc eDe<j                         e/       dddd              Ze eDe<j                         e/       deXd(edefd              Zg eDe<j                        d(efd       Zi eDe<j                         e/dRȫ      d(edefd              Zj eDe<j                         e/       d(edefd	              Zkd
 Zl	 	 	 	 	 dMdedede	ej(                     de	ej(                     de	e   d?e	e   de	ej(                     dAe	ej                     deYfdZm eDe<j                         e/       	 	 	 d%dedede	e   d?e	e   dAe	ej                     defd              Zo eDe<j                  g      	 	 	 	 	 dMdej(                  dej(                  dej(                  dej(                  de	ej(                     d?e	ej(                     de	ej(                     dAe	ej                     deYfd       Zq eDe<j                         e/       ded)eXdeYdefd              Zs eDe<j                         e/       dd              Zu eDe<j                         e/       	 	 	 dNd=ed$edeXdeYdeYdefd              Zv eDe<j                  j                        	 dOd#ed^eWe   deWeX   defd       Zxd Zyd Zz eye<j                          eye<j                          eye<j                          eye<j                          eye<j                          eye<j                           eye<j                          eye<j                          eye<j                          eze<j                          eze<j
                          eze<j                          eze<j                          eze<j                          eze<j                          eze<j                          eze<j                          eze<j                          eze<j                          eze<j                          eze<j                         d Z eDe<j"                         e/       d               Z eDe<j$                         e/       dGd d!              Z eDe<j&                         e/       dGd d"              Z ee<j"                        Z ee<j$                        Z ee<j&                        Zd dl0Zd dlZd dlZd# Z e        y(P      N)Sequence)Enum)wraps)CallableOptionalTypeVarUnion)	ParamSpec)SymBoolSymFloatTensor)_add_op_to_registry_convert_out_paramsglobal_decomposition_table
meta_table)
OpOverload)_prim_elementwise_meta$ELEMENTWISE_PRIM_TYPE_PROMOTION_KIND)BoolLikecorresponding_complex_dtypecorresponding_real_dtypeelementwise_dtypesELEMENTWISE_TYPE_PROMOTION_KIND	FloatLikeIntLikemake_contiguous_strides_forNumbersuggest_memory_format
TensorLike)_maybe_convert_to_dtype_maybe_resize_out_resize_output_check_safe_copy_outout_wrapper)_broadcast_shapes_maybe_broadcast)_config)_pytree_T_PatenIMPLMeta   returnc                       fd}|S )Nc                 V     t                 fd}t        j                  |        S )Nc                 (    t        t        |        y N)r   r   )opfns    S/var/www/html/engine/venv/lib/python3.12/site-packages/torch/_meta_registrations.pyregisterz0register_meta.<locals>.wrapper.<locals>.register:   s    
B3    )r   pytree	tree_map_)r5   r7   r4   s   ` r6   wrapperzregister_meta.<locals>.wrapper7   s)     $	4 	2&	r8    )r4   r;   s   ` r6   register_metar=   6   s     Nr8   type_promotionc                     t        j                  |d| i\  }}|D cg c]  }t        ||       }}t        | }t	        |dt
        j                  iS c c}w )Ntype_promotion_kindr>   )utilsr   r    r&   r   r   DEFAULT)r>   args_result_dtypexs        r6   elementwise_metarG   C   ss    
 ..	*OA| ?CC#A|4CDC T"D "	BJJ  Ds   Ac                     t         j                  t         j                  t         j                  t         j                  t         j
                  t         j                  i}|j                  | |       S r3   )torch	complex32halfcfloatfloatcdoubledoubleget)dtypefrom_complexs     r6   toRealValueTyperS   W   sE    ekku||L
 E5))r8   c                 l     t        t         g|       t        j                   k(   fd       y )Nc                      d d  S )Nzoutput with shape z# doesn't match the broadcast shape r<   )broadcasted_shape
self_shapes   r6   <lambda>z)check_inplace_broadcast.<locals>.<lambda>d   s    $ZL0STeSfg r8   )tupler%   rI   _check)rW   
args_shaperV   s   ` @r6   check_inplace_broadcastr\   `   s0    /
HZHI	LLZ'gr8   Fc	                 <   	 t         t        j                        r(t        j                   j	                         dk(  d        t        t        j                        r(t        j                  j	                         dk(  d        t        d  fD              rZt        j                  t        j                               		nFt        j                  t        j                        	fd       nxs t        j                         t        t        j                        sJ t        j                  t        t               fd       t        t              sJ t        j                  dk\  d        t        j                  f|d||	      S )
Nr   c                       yNz:linspace only supports 0-dimensional start and end tensorsr<   r<   r8   r6   rX   z(meta_linspace_logspace.<locals>.<lambda>x       r8   c                       yr_   r<   r<   r8   r6   rX   z(meta_linspace_logspace.<locals>.<lambda>}   r`   r8   c              3   <   K   | ]  }t        |t                y wr3   )
isinstancecomplex).0args     r6   	<genexpr>z)meta_linspace_logspace.<locals>.<genexpr>   s     
C:c7#
Cs   c                      d  d S )Nzlinspace(): inferred dtype z& can't be safely cast to passed dtype r<   )default_complex_dtyperQ   s   r6   rX   z(meta_linspace_logspace.<locals>.<lambda>   s    56K5LLrsxryz r8   c                      dt              j                   dt               j                   dt              j                   dS )Nz4received an invalid combination of arguments - got (, ))type__name__)endstartstepss   r6   rX   z(meta_linspace_logspace.<locals>.<lambda>   sD     u+r$s),,-RU0D0D/EQH r8   c                       y)Nz$number of steps must be non-negativer<   r<   r8   r6   rX   z(meta_linspace_logspace.<locals>.<lambda>   r`   r8   metarQ   layoutdevice
pin_memoryrequires_grad)rc   rI   r   rZ   dimanyrA   r   get_default_dtypeis_complex_dtyperQ   _check_typer   empty)
rp   ro   rq   baserQ   rv   ru   rw   rx   ri   s
   ``` `    @r6   meta_linspace_logspacer   h   sL    %&IIK1P	
 #u||$GGINP	

 
CsE/B
CC % A A##%!
 =)ELL&&u-z
 2002eU[[))) 
5'"	H
 eW%%%	LL!KL;;	# r8   c                    t        j                  j                  t         j                  k(  fd       t        j                  | j                         dk(  xr j                         dk7   d        | j                  j                        S )Nc                  "    d j                    S )Nz2take(): Expected a long tensor for index, but got rQ   indexs   r6   rX   zmeta_take.<locals>.<lambda>   s    DU[[MR r8   r   c                       y)Nz*take(): tried to take from an empty tensorr<   r<   r8   r6   rX   zmeta_take.<locals>.<lambda>   r`   r8   )rI   rZ   rQ   long_check_indexnumel	new_emptyshape)selfr   s    `r6   	meta_taker      sm     
LLuzz!R
 
ZZ\Q55;;=A#56< >>%++&&r8   ry   c                T     j                   }j                   }t        j                  ||k(  d        t        j                   j                        dk(  xr j                        dk(   fd       t	         j
                  j
                        } j                  |      S )Nc                       y)Nz=linalg.cross: inputs must have the same number of dimensions.r<   r<   r8   r6   rX   zlinalg_cross.<locals>.<lambda>   r`   r8   r.   c                  V    d  dj                          dj                          S )Nzlinalg.cross: inputs dimension z must have length 3. Got  and size)ry   otherr   s   r6   rX   zlinalg_cross.<locals>.<lambda>   s6    -cU 399S>"%

3'8: r8   )ndimrI   rZ   r   r%   r   r   )r   r   ry   x_dy_d	out_shapes   ```   r6   linalg_crossr      s     ))C
**C	LLs
O 
LL		#!4

31 4	
 "$**ekk:I>>)$$r8   c                 |    t        | d       t        | d       t        j                  | t        j                        S )Nzlinalg.matrix_expmemory_format)squareCheckInputscheckFloatingOrComplexrI   
empty_likecontiguous_formatr   s    r6   linalg_matrix_expr      s3     d/04!45D0G0GHHr8   valuesindicesc                 Z   t        j                  | j                  | j                  | j                        }t        j                  | j                  | j                  t         j
                        }| j                         dk7  r%| j                  dk7  rt        || j                         ||fS )Nrv   rQ   r   )	rI   r~   r   rv   rQ   int64r   r   maybe_wrap_dim)r   ry   r   r   s       r6   	cummaxminr      sp    
 [[DKKtzzJFkk$**T[[LGzz|qTYY!^sDII&7?r8   c                 x    t        || j                         t        j                  | t        j                        S Nr   )r   r   rI   r   r   )r   ry   s     r6   logcumsumexpr      s+     3		"D0G0GHHr8   c                D   |j                   }t        |      }||z
  }t        t        |            }t        |      D 	cg c]  }	d }
}	|D ]  }d|
|<   	 g g }}|D ]*  }|
|   s|j	                  |       |j	                  |       , ||z   }t        |      }|j                         |d | }|j                  fdd       |||d  z   }|j                  |      }dgt        |j                  |d        z   }|j                  |      }|j                  d      }||d<   t        |      }t        t        |            D ]  }|||      ||dz   <    | j                  |t        j                         t        |      D 	cg c]  }	d }}	d}|dz
  }|dk\  r0|| j                  d      z  |||   <   ||||      z  }|dz  }|dk\  r0t        ||      D ]  }| j                  d||z
  z         |||   <   ! | j                  ||| j                                | S c c}	w c c}	w )	NFTc                     |    S r3   r<   )rF   self_stridess    r6   rX   z_exec_fft.<locals>.<lambda>   s    <? r8   keyreverser   r      r   )r   lenlistrangeappendstridesortpermuter   reshaper   resize_rI   r   as_strided_storage_offset)outr   	out_sizesry   forwardr   signal_ndim
batch_dimsdim_permuterD   is_transformed_dimdleftright	batch_endtmpinputbatched_sizes
batch_sizebatched_out_sizesiout_stridesbatch_numelr   s                          @r6   	_exec_fftr      sZ   99Dc(K#J uT{#K).t5A%55 % $1% b%D !!$KKNLLO	
 ,KD	I;;=L
jy
!CHH*DH9IJ//KLL%E D4JK 899MMM-(EAJ!M!]+3s8_ 5#,SV#4!a% 5KK!1H1HKI $Dk*1*K*KQA
q&&1CJJqM&AKN#yQ00	Q q& :t$ G&)jja*n1E&FKN#GOOI{C,>,>,@AJW 6@ +s   	H 	Hr   ry   exclude_lastc                     t        |      }| j                         |d t        |      t        |      z
   j	                  fd       |S )Nc                     |    S r3   r<   )r   r   s    r6   rX   z_sort_dims.<locals>.<lambda>"  s    l1o r8   )r   )r   r   r   intr   )r   ry   r   sorted_dimsr   s       @r6   
_sort_dimsr     sL    s)K;;=L6#k"S%667<<% =  r8   c                 
   t        j                  | j                  j                         |s| j	                         S t        | |      }| j                  | j                               }t        || | j                         ||      S )Nr   )	rI   rZ   rQ   
is_complexcloner   r   r   r   )r   ry   normalizationr   r   r   s         r6   meta_fft_c2cr   )  sb     
LL&&'zz|T3'K
..
%CS$		['JJr8   c                 f    t        |       t        kD  st        |       dk\  r| d   dk(  r	| d   dk(  ryy)N   r   r   FT)r   cufft_max_ndimr   s    r6   use_optimized_cufft_pathr   8  s3    
3x. SX]s1v{s1vQR{r8   c                 z   t        j                  | j                  j                         t	        | j                               }t	        |      }|d   }||   dz  dz   }t	        |      }|||<   |r|||<   t        |       dk(  st        |       dk(  rz| j                  |t        j                  | j                              }	| }
t        |       dk(  rt        |      rt        |	|
||d       nt        |      dk(  r|n|}t        |	|
||gd       t        |      dkD  r0| j                  |t        j                  | j                              }
|d d }|rx|
|	}
}	|
j                         |j                  fd	d
       t        t         t        |            }|t        |      |z
  d  }t        |	|
||d       |d t        |      |z
   }|rx|s:|	j                  |      ||   k7  r#|
j#                  |t         j$                         |
}	|	S | j                  |t        j                  | j                              S )Nr   r   r   cudaxpur   Tr   c                     |    S r3   r<   )r   stridess    r6   rX   zmeta_fft_r2c.<locals>.<lambda>f  s    '!* r8   r   r   )rI   rZ   rQ   is_floating_pointr   r   device_hintr   rA   r   r   r   r   r   r   minr   r   r   )r   ry   r   onesidedinput_sizesr   last_dimlast_dim_halfsizeonesided_sizesoutputworking_tensortarget_sizesr   max_dims	last_dimsr   s                  @r6   meta_fft_r2cr   ?  s>    
LL--.tyy{#K[!I2wH#H-2Q6+&N0N8/	(4F"k$&75&@ U>>tzzJ   
 t&+CC+HfnidK ),CA9>LfnlXJPTU3x!|!%U%F%Ftzz%R "0 "
 cr(K)7(//1  ,d !  ~s;/?@'K(88(C(EF	NNIt **GC,<x,GH  {{8$	((;;&&y@W@W&X' ~~U>>tzzJ  
 	
r8   )	generatorc                B    t        |t        j                  | g            S r3   )r!   rI   Size)nr   r   s      r6   meta_randpermr   |  s    S%**aS/22r8   rQ   ru   rv   rw   c                6    t        j                  | ||||      S Nr   rI   r~   )r   rQ   ru   rv   rw   s        r6   meta_randperm_defaultr    s      ;;	vf r8   c                x     dt        j                   kD   fd       t        j                  |||||      S )Nr   c                      d d  S Nz:random_ expects 'from' to be less than 'to', but got from=z >= to=r<   highlows   r6   rX   zmeta_randint.<locals>.<lambda>      LSEQXY]X^_ r8   r   rI   rZ   r~   )r  r   rQ   ru   rv   rw   r  s   `     @r6   meta_randintr
    s>     C	LLs
_ ;;E&J r8   c                t     t        j                   kD   fd       t        j                  |||||      S )Nc                      d d  S r  r<   r  s   r6   rX   z"meta_randint_low.<locals>.<lambda>  r  r8   r   r	  )r  r  r   rQ   ru   rv   rw   s   ``     r6   meta_randint_lowr    s9     
LLs
_ ;;E&J r8   c                6    t        j                  | ||||      S r   r   )r   rQ   ru   rv   rw   s        r6   meta_rand_defaultr    s      ;;E&J r8   r   lastdimc                    t        j                  | j                  j                         t	        |       dk(  rt        | j                               }|||d   <   | j                  |t        | j                              }t        |      r.t        || j                  t         j                        ||d      S t        |      dkD  rt        | |d d d|      }n | j                  t         j                        }t        ||||d   gd      S | }t        |      dkD  r|d d }t        | ||d      }|dd  }t        |j                               }|||d   <   | j                  |t        | j                              }	t        |	|||d      S )	Nr   r   r   r   Fr   r   r   )rI   rZ   rQ   r   r   r   r   r   rS   r   r   r   r   r   r   )
r   ry   r   r  r   r   tempr   c2c_dimsr   s
             r6   meta_fft_c2rr    sk    
LL&&'4F"%	$	#b'	1LM#C(

)@)@
A  3x!|#D#cr(Aw?zz0G0GzHVT9s2wiOO s8a<3BxH xNEbc(C&	$	#b'nnYodjj.InJeYUCCr8   c                 J   ddl m}  ||       s#t        j                  |       dk(  rt	        d      t        |t              ra|j                  | |      }| j                         |j                         k7  r.t        j                  j                  || j                                | S )Nr   )free_unbacked_symbolsr   zQmore than one element of the written-to tensor refers to a single memory location)%torch.fx.experimental.symbolic_shapesr  rI   _debug_has_internal_overlapRuntimeErrorrc   r   tor   r+   expand_copydefault)r   srcnon_blockingr  intermediates        r6   
meta_copy_r     s     L "$'E,M,Md,SWX,X_
 	
 #vvvdL199;,++--$$\499;?Kr8   c                     t        | j                               }t        | j                               }|| j                         k\  rdn
||   ||   z  }|j	                  |d       |j	                  ||       ||fS Nr   )r   r   r   ry   insert)tensorry   result_sizesresult_strides
new_strides        r6   inferUnsqueezeGeometryr(    sq    &L&--/*NVZZ\)|C/@>RUCV/VJQ#z*''r8   c                     t        || j                         dz         }t        | |      \  }}| j                  ||       | S r"  )r   ry   r(  r   )r   ry   g_sizes	g_stridess       r6   meta_unsqueeze_r,    s>    
dhhj1n
-C/c:GYWi(Kr8   r   weight_metabias_activation_opt	out_dtypec                 8   t        | j                        }|*|j                  d      |j                  d      k(  sJ d       |j                  d      | j                  d      dz  k(  sJ |j                  d      |d<   t        | j                        dk(  sJ d       d| j                  d      f}|7| j                  t
        j                  k(  r|t
        j                  k(  sJ d       | j                  ||| j                  n|      j                  ||      }|S )	Nr   zoutput size mismatchr   r   r   z*we can only handle the squashed input case9out_dtype is only supported for i8i8->i32 linear operatorr   )
r   r   r   r   rQ   rI   int8int32r   
as_strided)	r   r-  r.  r/  r0  r1  output_sizestransposed_stridesr   s	            r6   meta_sparse_structured_linearr9    s    $L{{1~1-E/EE-;;q>UZZ^a////{{1~L u{{q N"NN UZZ]+{{ejj(Y%++-E 	
G	
E __&.ekkI   j12 
 Mr8   mat1	mat1_metamat2c                    t        | j                        dk(  sJ t        |j                        dk(  sJ t        |j                        dk(  sJ | j                  d      |j                  d      dz  k(  sJ | j                  d      |j                  d      g}|7|j                  t        j
                  k(  r|t        j                  k(  sJ d       |j                  |||j                  n|      }|S )Nr   r   r   r3  r   r   r   r   rQ   rI   r4  r5  r   )r:  r;  r<  r1  r7  r   s         r6   meta_sparse_structured_mmr?  6  s     tzz?ay1$$$tzz?a99Q<499Q<!++++IIaL$))A,/LzzUZZ'I,D 	
G	
D ^^%-djj9  F
 Mr8   r   )alphabetar1  c                |   t        | j                        dk(  sJ d       t        |j                        dk(  sJ t        |j                        dk(  sJ t        |j                        dk(  sJ | j                  d      |j                  d      k(  sJ d       |j                  d      |j                  d      dz  k(  sJ |j                  d      |j                  d      g}|7|j                  t        j
                  k(  r|t        j                  k(  sJ d       |j                  |||j                  n|      }|S )Nr   zEonly input broadcasted to columns of mat1 * mat2 product is supportedr   r   r3  r   r>  )	r   r:  r;  r<  r@  rA  r1  r7  r   s	            r6   meta_sparse_structured_addmmrC  O  s/    u{{q  O  tzz?ay1$$$tzz?a::a=DIIaL( O( 99Q<499Q<!++++IIaL$))A,/LzzUZZ'I,D 	
G	
D ^^%-djj9  F
 Mr8   compressed_Adense_Br@  transpose_resultalg_idsplit_ksplit_k_modec	                 L   |j                   t        j                  t        j                  t        j                  t        j
                  t        j                  hv sJ d       | j                   |j                   k(  sJ d       t        |j                        dk(  sJ d       | j                   t        j
                  t        j                  fv }	|	rdnd}
|	r|j                         rJ d       |j                  d      }|j                  d	      }| j                         d
z  |
|z  z  }|||j                  d      k(  sJ |I|	r@|t        j                  t        j                  t        j                  t        j                  hv sJ d       |r||fn||f}|j                  ||      S )Nz;_cslt_sparse_mm only supports fp16, bf16, int8, and fp8e4m3zinputs must have the same dtyper   z'_cslt_sparse_mm only supports 2d inputs
   	   z.dense input must be transposed for 8bit dtypesr   r      z\out_dtype is not supported for {compressed_A.dtype} x {dense_B.dtype} -> {out_dtype} matmul!r   )rQ   rI   float32float16bfloat16r4  float8_e4m3fnr   r   is_contiguousr   r   r5  r   )rD  rE  r/  r@  r1  rF  rG  rH  rI  is_8bit_input_typecompression_factorkr   moutput_shapes                  r6   meta__cslt_sparse_mmrX  r  s    ==

  E EE  .Q0QQ.w}}"M$MM"%++

E<O<O/PP1q((* 	
<	
* 	QAQA					"(:Q(>?ADIIaL   !iMMNNKK	4
 '
 	
 k	
 
 .Aq6Aq6L\;;r8   T)include_selfr   sourcereducerY  c                L    t        j                  | t         j                        S r   rI   r   r   r   ry   r   rZ  r[  rY  s         r6   meta_index_reducer_    s     D0G0GHHr8   c                    | S r3   r<   r^  s         r6   meta_index_reduce_ra    s	     Kr8   c                     t        | j                               }| j                         dkD  r|j                         ||<   | j	                  |      S Nr   )r   r   ry   r   r   )r   ry   r   result_sizes       r6   meta_index_selectre    s@     tyy{#KxxzA~ ;;=C>>+&&r8   )lengthsr   offsetsaxisunsafeinitialdatarf  rg  rh  ri  c                     |t        d       fd}| ||j                        S |+|j                  d d |j                  d   dz
  fz   }	 ||	      S t        d      )Nz?segment_reduce(): indices based reduction is not supported yet.c                     t        j                  | j                  dz   d  z   j                  dt         j                        S )Nr   rs   rQ   rv   r   )rI   r~   r   rQ   r   )lengths_shaperh  rk  s    r6   segment_reduce_lengths_tensorz:meta_segment_reduce.<locals>.segment_reduce_lengths_tensor  s>    {{DJJtaxz22**11	
 	
r8   r   r   z<segment_reduce(): Either lengths or offsets must be defined.)NotImplementedErrorr   r  )
rk  r[  rf  r   rg  rh  ri  rj  rp  ro  s
   `    `    r6   meta_segment_reducerr    s|     !M
 	

 ,W]];; cr*gmmB.?!.C-EE,];;
U
VVr8   c                 $    | j                  d      S Nr<   r   r   s    r6   meta_maxrv         >>"r8   c                     t        j                  | j                  |f      }t        | ||      }| j	                  |      | j	                  |t
        j                        fS Nr   rA   reduction_dimsr   _compute_reduction_shaper   rI   r   r   ry   keepdimrW  s       r6   meta_max_dimr    R    


tzzC6
2C+D#w?L|$|5::6 r8   c                 $    | j                  d      S rt  ru  r   s    r6   meta_minr    rw  r8   c                     t        j                  | j                  |f      }t        | ||      }| j	                  |      | j	                  |t
        j                        fS ry  rz  r}  s       r6   meta_min_dimr    r  r8   c                     | j                         rt        | j                        }nt        | t        j
                        \  }}t        j                  | |      S Nr@   r   )r   r   rQ   r   r   INT_TO_FLOATrI   r   )r   rE   rD   s      r6   
meta_angler    sI    /

;, ? L L
< D55r8   c                     t        j                  || j                         | j                         |j	                  t        j
                  |             S r3   )rI   _resize_output_r   rv   copy_angle)r   r   s     r6   meta_angle_outr    s6    	#tyy{DKK899U[[&''r8   c                      y r3   r<   )vals    r6   assert_asyncr        
r8   c                      y r3   r<   )r  
assert_msgs     r6   assert_async_metar  "  r  r8   c                      y r3   r<   )ss    r6   
print_metar  '  r  r8   rQ   ru   rv   rw   r   c                 0    t        j                  dd      S )Nr   rs   rv   r   r  s        r6   make_dep_tokenr  ,  s     ;;q((r8   c                 h    ddl m} t        | t        t        f      rt        d       || ||       y )Nr   )constrain_range'Constraining SymFloat or Symbool is nyir   max)r  r  rc   r   r   
ValueError)r   r   r  r  s       r6   sym_constrain_ranger  8  s/     F$7+,BCCDcs+r8   c                 6    t         j                  | ||       |S Nr  )r+   r  r   r   r  	dep_tokens       r6   functional_sym_constrain_ranger  B  s    Ts4r8   c                 (   ddl m} ||t        j                  |        y t	        | t
        t        f      rt        d      t        |       t        u r5|t        j                  | |k\         |t        j                  | |k         y  || ||       y )Nr   )_constrain_range_for_sizer  r  )r  r  rI   _check_is_sizerc   r   r   r  rm   r   rZ   )r   r   r  r  s       r6   sym_constrain_range_for_sizer  H  s     P
{s{T"$7+,BCCDzS?LL%?LL%d5r8   c                 6    t         j                  | ||       |S r  )r+   r  r  s       r6   'functional_sym_constrain_range_for_sizer  \  s    %%d%=r8   c                     |S r3   r<   )r  r  r  s      r6   functional_assert_async_metar  b  s    r8   f_namec                     | j                         dk\  s
J | d       | j                  d      | j                  d      k(  s.J | d| j                  d       d| j                  d       d       y )Nr   z3: The input tensor must have at least 2 dimensions.r   z5: A must be batches of square matrices, but they are  by 	 matrices)ry   r   )r   r  s     r6   r   r   h  s}    88:? (EF? 99R=DIIbM) (G		RTVZ[_[d[deg[hZiirs)r8   Anamec                     t        j                   j                  j                  k(   fd       t        j                   j                  j                  k(   fd       t        j                  j	                  d      j	                  d      k(  fd       t        j                  j	                  d       j	                  d      k(   fd       y )Nc                  >    dj                    d j                    dS )Nz:Expected b and A to be on the same device, but found b on z
 and A on 	 instead.r  r  r   s   r6   rX   z(linearSolveCheckInputs.<locals>.<lambda>w  s%    H{{m:ahhZy: r8   c                  >    dj                    d j                    dS )Nz=Expected b and A to have the same dtype, but found b of type z and A of type r  r   r  s   r6   rX   z(linearSolveCheckInputs.<locals>.<lambda>  s%    Kzzl/!'')= r8   r   r  c                  R    d j                  d       d j                  d       dS )Nz3A must be batches of square matrices, but they are r  r  r   r  r   r  s   r6   rX   z(linearSolveCheckInputs.<locals>.<lambda>  s0    FF2J<tAFF2J<yB r8   c                      d d j                  d       d j                  d       dj                  d       dj                  d       
S )NzIncompatible matrix sizes for z: each A matrix is r   r  z but each b matrix is r  r   )r  r  r   s   r6   rX   z(linearSolveCheckInputs.<locals>.<lambda>  sR    ,TF 3D$TYYr]O4		"H r8   )rI   rZ   rv   rQ   r   )r   r  r  s   ```r6   linearSolveCheckInputsr  t  s    	LLqxx	
 
LL

agg	
 
LL	r
affRj 	
 
LL	r
diim#	
r8   tallow_low_precision_dtypesc                 J   | j                   t        j                  | j                         xs | j	                         fd       |sYt        j                  t        j
                  t        j                  t        j                  t        j                  fv fd       y y )Nc                       d  S )Nz<: Expected a floating point or complex tensor as input. Got r<   rQ   r  s   r6   rX   z(checkFloatingOrComplex.<locals>.<lambda>  s    6(VW\V]^ r8   c                       d  S )Nz*: Low precision dtypes not supported. Got r<   r  s   r6   rX   z(checkFloatingOrComplex.<locals>.<lambda>  s    vhHP r8   )	rQ   rI   rZ   r   r   rM   rO   rL   rN   )r  r  r  rQ   s    ` @r6   r   r     sn    
 GGE	LL	/^ &ekk5<<u}}MMP	
 &r8   arg_namec                 ^    t        j                  | j                         dk\  fd       y )Nr   c                       d  dS )Nz: The input tensor z! must have at least 2 dimensions.r<   )r  r  s   r6   rX   zcheckIsMatrix.<locals>.<lambda>  s    6(-hZ7XY r8   )rI   rZ   ry   )r  r  r  s    ``r6   checkIsMatrixr    s    	LL	1Yr8   Br   c                      t                t               t        j                  r# j	                  d      j	                  d      k(  n" j	                  d      j	                  d      k(   fd       y )Nr  r   c                       drdnd d j                  d       d j                  d       dj                  d       dj                  d       d	S )
Nz2: Incompatible shapes of A and B for the equation zAX = BzXA = Bz (r  rF   r   r   rl   r   )r  r  r  r   s   r6   rX   z#checkInputsSolver.<locals>.<lambda>  s[    hHxX.AaffRj\qvvbzl!AFF2J<qJ r8   )r   r  rI   rZ   r   )r  r  r   r  s   ````r6   checkInputsSolverr    sY    a !V	LL$(r
affRj affRjAFF2J.F	
r8   resultfn_nameresult_namec                 r     t        j                  j                  j                  k(   fd       y )Nc            	      L      d d dj                    dj                    	S )Nz: Expected z5 and input tensors to be on the same device, but got z on z and input on r  )r  r   r  r  s   r6   rX   z!checkSameDevice.<locals>.<lambda>  s5    i{;-/dm4nU\\NL r8   )rI   rZ   rv   )r  r  r   r  s   ````r6   checkSameDevicer    s&     
LL%	
r8   UPLOc                       j                         }t        j                  t               dk(  xr |dk(  xs |dk(   fd       y )Nr   ULc                      d  S )Nz1Expected UPLO argument to be 'L' or 'U', but got r<   )r  s   r6   rX   zcheckUplo.<locals>.<lambda>  s    CD6J r8   )upperrI   rZ   r   )r  UPLO_uppercases   ` r6   	checkUplor    s<    ZZ\N	LLD	QKNc1J^s5JJr8   eigenvalueseigenvectorsr  	compute_vc                 T   t        | d       t        |       t        | j                        }|r/| j	                  |      }|j                  |t        |d             n| j	                  dg      }|j                          | j	                  |t        | j                              }||fS )Nzlinalg.eighF	row_majorr   r   )
r   r  r   r   r   r   r   poprS   rQ   )r  r  r  r   vecsvalss         r6   meta__linalg_eighr    s     a'dOME{{5! ;EU ST{{A3	IIK;;uOAGG$<;=D:r8   c                     t        | d       t        j                  | j                        r| j                  nt        j                  | j                        }| j                  | j                  d d |      S )Nzlinalg.eigvalsr   r   r   rA   r|   rQ   r   r   r   )r   complex_dtypes     r6   meta__linalg_eigvalsr    sc     e-. !!%++. 	..u{{; 
 ??5;;s+=?AAr8   c                 0   t        | d       t        j                  | j                        r| j                  nt        j                  | j                        }| j                  | j                  d d |      }| j                  | j                  |      }||fS )Nz
linalg.eigr   r   r  )r   r  r   vectorss       r6   meta_linalg_eigr    s     e\* !!%++. 	..u{{; 
 __U[["-]_CFooekko?G7?r8   r  c                 v    | j                   j                  t        j                        j	                  dd      S )Nr   r  r   )mTr   rI   r   	transpose)r  s    r6   cloneBatchedColumnMajorr    s*    66<<e&=&=<>HHRPPr8   r  c                     t        |       S r3   )r  )r   r  r  s      r6   _cholesky_solve_helperr    s     #4((r8   c                      t        j                   j                  dk\   fd       t        j                  j                  dk\  fd       t         d      \  }}t	        |||      S )Nr   c                  $    d j                    dS )Nz-b should have at least 2 dimensions, but has  dimensions insteadr   r   s   r6   rX   z cholesky_solve.<locals>.<lambda>  s    ?		{J]^ r8   c                  $    d j                    dS )Nz-u should have at least 2 dimensions, but has r  r  r  s   r6   rX   z cholesky_solve.<locals>.<lambda>  s    ?xGZ[ r8   cholesky_solve)rI   rZ   r   !_linalg_broadcast_batch_dims_namer  )r   r  r  self_broadcastedA_broadcasteds   ``   r6   r  r    sh     
LL		Q^ 
LL	![ 'Ha!'#m ""2M5IIr8   c                     | j                         dk(  r%t        j                  | t        j                        S t	        | d       t        |       S )Nr   r   cholesky)r   rI   r   legacy_contiguous_formatr   r  r   r  s     r6   r  r    s@     zz|qE4R4RSSdJ'"4((r8   c                 0    t        | d       t        |       S )Ncholesky_inverse)r   r  r   s     r6   r  r  &  s     d./"4((r8   check_errorsc                    t        | d       t        | d       | j                  }t        |      }t	        |d      }| j                  |      }|j                  ||       | j                  |d|dz
   t        j                        }||fS )Nzlinalg.choleskyFr   r   r   )	r   r   r   r   r   r   r   rI   r5  )r  r  r  A_shaper   	L_stridesr  infoss           r6   linalg_cholesky_exr  .  s    a*+1/0ggGw<D ,GU;I	GAMM'9% KKD1H-U[[KAEe8Or8   tauc                 @    t        j                   j                  dk\  d        t        j                   j                  d       j                  d      k\  d        t        j                   j                  d      j                  d      k\  d        t        j                   j                  j                  z
  dk(   fd        j                  dkD  r: j                  d d }j                  d d t        j                  |k(  fd	       t        j                  j
                   j
                  k(   fd
       t        d d       t        j                   j                  t         j                  d       j
                   j                        S )Nr   c                       y)NzHtorch.linalg.householder_product: input must have at least 2 dimensions.r<   r<   r8   r6   rX   z,linalg_householder_product.<locals>.<lambda>G  r`   r8   r  r   c                       y)Nzbtorch.linalg.householder_product: input.shape[-2] must be greater than or equal to input.shape[-1]r<   r<   r8   r6   rX   z,linalg_householder_product.<locals>.<lambda>K  r`   r8   c                       y)Nz`torch.linalg.householder_product: input.shape[-1] must be greater than or equal to tau.shape[-1]r<   r<   r8   r6   rX   z,linalg_householder_product.<locals>.<lambda>O  r`   r8   r   c                  <    dj                    d j                    S )Nzptorch.linalg.householder_product: Expected tau to have one dimension less than input, but got tau.ndim equal to  and input.ndim is equal to r  r   r	  s   r6   rX   z,linalg_householder_product.<locals>.<lambda>T  '    )),
2Nuzzl\ r8   c                      d  S )Nzltorch.linalg.householder_product: Expected batch dimensions of tau to be equal to input.shape[:-2], but got r<   actual_batch_tau_shapes   r6   rX   z,linalg_householder_product.<locals>.<lambda>^      66L5MO r8   c                  <    dj                    d j                    S )Nz,torch.linalg.householder_product: tau dtype z does not match input dtype r   r  s   r6   rX   z,linalg_householder_product.<locals>.<lambda>f  s#    :399+*5;;-9 r8   z torch.linalg.householder_productr	  Fr  r   r   rQ   rv   )
rI   rZ   r   r   r   rQ   r  empty_stridedr   rv   )r   r	  expected_batch_tau_shaper  s   `` @r6   linalg_householder_productr  @  sK   
 
LL

aZ 
LL

2%**R.(t 
LL

2#((2,&r
 
LL

SXX"	
 zzA~#(;;s#3 !$3B"&>>	
 
LL		U[[ 	
 6UEJ[[*5;;%Hkk||	 r8   c                 2   t        | d       t        | dd       | j                  | j                        }|j	                  | j                  t        | j                  d             | j                  | j                  d d t        j                        }||fS )Nzlinalg.inv_exF)r  r  r  r   r   r   r   r   r   r   rI   r5  )r  r  r  r  s       r6   linalg_inv_ex_metar  v  sq    a)1o%P	AGGAMM!''6qww%PQKKEKKK8Ee8Or8   LDpivotsinfo)	hermitianr  r!  c                   t        | d       t        | d       t        j                  | j                  t        | j                  d      | j                  | j                        }| j                  | j                  d d t        j                        }| j                  | j                  d d t        j                        }|||fS )Nztorch.linalg.ldl_factor_exFr  r  r   r   r  )
r   r   rI   r  r   r   rQ   rv   r   r   )r   r!  r  r  r  r   s         r6   linalg_ldl_factor_ex_metar#    s     d894!=>			ZZ*4::Gjj{{	
B ^^DJJsO599^=F>>$**Sb/>;Dvtr8   )r!  c                d    t         d       t         d       t         d       t        j                  j
                  dk\  fd        j                  d d }t        j                  |j                  k(  fd       t        j                  t        j                  j                        fd       t        j                   j                  j                  k(   fd       t               \  }}t        j                  |t        |d	      j                  j                  
      S )Nztorch.linalg.ldl_solver   c                  $    d j                    dS )NzMtorch.linalg.ldl_solve: Expected B to have at least 2 dimensions, but it has r  r  )r  s   r6   rX   z'linalg_ldl_solve_meta.<locals>.<lambda>      &&!46 r8   r   c                  $    d j                    dS )Nzjtorch.linalg.ldl_solve: Expected LD.shape[:-1] and pivots.shape to be the same, but got pivots with shape  insteadr   r  s   r6   rX   z'linalg_ldl_solve_meta.<locals>.<lambda>      ))/h@ r8   c                  "    d j                    S )Nz<torch.linalg.ldl_solve: Expected pivots to be integers. Got r   r*  s   r6   rX   z'linalg_ldl_solve_meta.<locals>.<lambda>  s    Nv||n] r8   c                  <    dj                    d j                    S )Nz!torch.linalg.ldl_solve: LD dtype z does not match b dtype r   )r  r  s   r6   rX   z'linalg_ldl_solve_meta.<locals>.<lambda>  s"    3BHH:=UVWV]V]U^_ r8   Fr  r  )r   r   r  rI   rZ   r   r   rA   is_integer_dtyperQ   _linalg_broadcast_batch_dimsr  r   rv   )r  r  r  r!  expected_pivots_shapeB_broadcast_sizerD   s   ```    r6   linalg_ldl_solve_metar2    s     b232781b":;	LL	!	
 HHSbM	LL-	
 
LLv||,] 
LL
AGG_ 7q"=a*+;uMggxx	 r8   Pr  )pivotr4  c                h    t        j                   j                  dk\   fd       t         j                        }|d   }|d   }t        ||      }||d<   |r j                  |      }n j                  dg      }||d<    j                  |      }||d<   ||d<    j                  |      }|||fS )Nr   c                  $    d j                    dS )Nz@linalg.lu: Expected tensor with 2 or more dimensions. Got size: r(  r)  r  s   r6   rX   z linalg_lu_meta.<locals>.<lambda>  s    RSTSZSZR[[cd r8   r  r   r   )rI   rZ   r   r   r   r   r   )	r  r4  sizesrV  r   rU  r3  r  r  s	   `        r6   linalg_lu_metar8    s     
LL	!d
 MEb	Ab	AAq	AE"IKKKKE"I	EAE"IE"I	EAa7Nr8   LU)r4  r  c                    t        j                   j                  dk\   fd       t         j                        }|d   }|d   }t        j
                  |t        |d       j                   j                        }|j                          t        ||      |d<    j                  |t         j                        }|j                           j                  |t         j                        }|||fS )	Nr   c                  $    d j                    dS )NzFtorch.lu_factor: Expected tensor with 2 or more dimensions. Got size: r(  r)  r  s   r6   rX   z*linalg_lu_factor_ex_meta.<locals>.<lambda>  s    XYZY`Y`Xaaij r8   r  r   Fr  r  r   )rI   rZ   r   r   r   r  r   rQ   rv   r  r   r   r   )	r  r4  r  r7  rV  r   r9  r  r   s	   `        r6   linalg_lu_factor_ex_metar<    s     
LL	!j
 MEb	Ab	A			*5EBggxx	
B 
IIKAq	E"I[[eii[0F 
IIK;;uEII;.Dvtr8   )r   adjointr=  c                    t         d       t        j                   j                  j                  k(   fd       t        j                  j                  t        j                  k(  d        t         d       t         |d       t        j                   j                  d      j                  d      k(  d        t        j                   j                  d d j                  k(  fd       t               \  }}t        j                  |t        ||       j                  j                  	      }|j                         d
k7  r"|s |j                         r|j                         }|S )Nztorch.linalg.lu_solvec                  >    dj                    d j                    dS )NzPlinalg.lu_solve: Expected LU and B to have the same dtype, but found LU of type  and B of type r(  r   )r  r9  s   r6   rX   z&linalg_lu_solve_meta.<locals>.<lambda>  s(    $$&HH:_QWWIXO r8   c                       y)NzElinalg.lu_solve: pivots should be a Tensor of scalar type torch.int32r<   r<   r8   r6   rX   z&linalg_lu_solve_meta.<locals>.<lambda>  r`   r8   zlinalg.lu_solver   c                       y)NzYlinalg.lu_solve: Number of pivots per batch should be same as the dimension of the matrixr<   r<   r8   r6   rX   z&linalg_lu_solve_meta.<locals>.<lambda>   r`   r8   c                  $    d j                    dS )Nzclinalg.lu_solve: Expected LU.shape[:-1] and pivots.shape to be the same, but got pivots with shape r(  r)  r*  s   r6   rX   z&linalg_lu_solve_meta.<locals>.<lambda>&  r+  r8   r  r  r   )r   rI   rZ   rQ   r   r   r  r   r   r/  r  r   rv   r   r   conj)r9  r  r  r   r=  r1  rD   r  s   ```     r6   linalg_lu_solve_metarE    s.    267	LL
AGG	
 
LL		!W b12b!T#45	LL
v{{2&k 
LL
"%	
 7q"=a  *+;4xPggxx	F ||~4[[]FMr8   unpack_dataunpack_pivotsc                     t        j                   j                  dk\   fd       |r2t        j                  |j                  t         j                  k(  d        t         j                        }|d   }|d   }t        ||      }||d<   |r j                  |      }n j                  dg      }|r2||d<    j                  |      }	||d<   ||d<    j                  |      }
n$ j                  dg      }	 j                  dg      }
||	|
fS )Nr   c                  $    d j                    dS )NzFtorch.lu_unpack: Expected tensor with 2 or more dimensions. Got size: r(  r)  )r9  s   r6   rX   z lu_unpack_meta.<locals>.<lambda>F  s    XY[YaYaXbbjk r8   c                       	 y)Nztorch.lu_unpack: LU_pivots is expected to be a contiguous tensor of torch.int32 dtype.
Note: this function is intended to be used with the output produced by torch.linalg.lu_factorr<   r<   r8   r6   rX   z lu_unpack_meta.<locals>.<lambda>K  s    p r8   r  r   r   )	rI   rZ   r   rQ   r5  r   r   r   r   )r9  r  rF  rG  r7  rV  r   rU  r3  r  r  s   `          r6   lu_unpack_metarK  <  s     
LL
1k LLEKK'	
 NEb	Ab	AAq	AE"ILLLL!b	LLb	b	LLLL!LL!a7Nr8   modec                       dk(  rd}d}||fS  dk(  rd}d}||fS  dk(  rd}d}||fS t        j                  d fd       fS )NreducedTcompleteFrc                      d  dS )Nzqr received unrecognized mode 'z=' but expected one of 'reduced' (default), 'r', or 'complete'r<   )rL  s   r6   rX   z _parse_qr_mode.<locals>.<lambda>s  s    1$ 8N O r8   rI   rZ   )rL  	compute_qrN  s   `  r6   _parse_qr_moderT  f  s    y	 g 
		 g 
	 g 		
 gr8   QRc                    t        | d       t        | d       t        |      \  }}| j                  d   }| j                  d   }t	        ||      }|rMt        | j                        }|r|n||d<   | j                  |      }|j                  |t        |d             n| j                  dg      }t        | j                        }	|s|s|n||	d<   | j                  |	      }
|
j                  |	t        |	d             ||
fS )Nz	linalg.qrr  r   Fr  r   )	r  r   rT  r   r   r   r   r   r   )r  rL  rS  reduced_moderV  r   rU  Q_shaperU  R_shaperV  s              r6   linalg_qr_metar[  {  s     ![!1k*,T2I|	A	AAq	Aqww-'aQKK 	g:7eTUKK 177mG#9!!GBK	GAMM'6w%PQa4Kr8   sign	logabsdetc                    t        | d       t        | dd       | j                  }| j                  |d d       }| j                  |d d t	        | j
                              }t        j                  |t        |d      | j
                  | j                        }| j                  |d d t        j                        }||||fS )Nzlinalg.slogdetFr  r   r  r   )r   r   r   r   rS   rQ   rI   r  r   rv   r5  )r  r   r\  r]  r9  r  s         r6   _linalg_slogdetr_    s     a)*1.6GGE;;uSbz"DE#2Joagg.FGI			*5%8ggxx	
B [[s5;;[7FB&&r8   full_matrices
compute_uvdriverc                 b   t        | d       t        | d       t        | j                  d d       }| j                  d   }| j                  d   }t	        ||      }|r|||r|n|gz   }| j                  |      }	|	j                  |t        |d             ||r|n||gz   }
| j                  |
      }t        |       dk(  }|j                  |
t        |
|             n$| j                  dg      }	| j                  dg      }| j                  ||gz   t        | j                              }|	||fS )	Nz
linalg.svdr  r   Fr  r   r   r   )r  r   r   r   r   r   r   r   r   rS   rQ   )r  r`  ra  rb  r   rV  r   rU  U_shaper  V_shapeVis_cudaSs                 r6   _linalg_svd_metari    s#    !\"1l+aggcrl#J	A	AAq	A11==KK 	g:7eTU]1==KK 
 a.F*	g:7gVW KKKK 	
J!$OAGG,DEAa7Nr8   arg1arg2c                    | j                   d d }|j                   d d }t        ||      }t        |      }|| j                  d      | j                  d      gz  }t        |      }||j                  d      |j                  d      gz  }||fS )Nr  r   )r   r%   r   r   )rj  rk  arg1_batch_sizesarg2_batch_sizesexpand_batch_portionarg1_expand_sizearg2_expand_sizes          r6   r/  r/    s    
 zz#2zz#2,-=?OP012		"66012		"66---r8   c                     |rt        | ||       t        | |      \  }}|| j                  k(  r| n| j                  |      }||j                  k(  r|n|j                  |      }||fS r3   )r  r/  r   expand)rj  rk  r  rp  rq  arg1_broadcastedarg2_broadcasteds          r6   r  r    sv     tT40)EdD)Q&& !DJJ.DKK@P4Q  !DJJ.DKK@P4Q  ---r8   r   c                     | j                   d d }|j                  dk(  xs- | j                  dz
  |j                  k(  xr |j                   |k(  }|S )Nr   r   )r   r   )r   r   expected_batched_rhs_shapevector_cases       r6   linalg_solve_is_vector_rhsry    sS    !&Sb!1**/ 

Q%**$R8R)R  r8   )r   r  r  r9  r  r   c                    t         d       t        j                   j                  j                  k(   fd       t	               }|rj                  d      n}	t         |	|d       t        |	       \  }
}t        j                  |xs | d        |r|
d d n|
}t        j                  |t        ||       j                  j                        } j                  }t        j                  |t        |d       j                   j                        } j                  |d d t        j                        } j                  |d d t        j                        }||||f}||||f}t        d	 |D              rbt        ||      D ]S  \  }}t!        ||j                         |j#                  |j                  |j%                                t'        ||d
       U |S )Nzlinalg.solvec                  >    d j                    dj                    dS )NzKlinalg.solve: Expected A and B to have the same dtype, but found A of type r@  r(  r   )r  r  s   r6   rX   z"_linalg_solve_ex.<locals>.<lambda>  s%    Ywwiqwwix9 r8   r   c                       	 y)Nzlinalg.solve: Vector broadcasting of the left hand side is not supported for left=False. In this case linalg.solve is equivalent to B / A.squeeze(-1)r<   r<   r8   r6   rX   z"_linalg_solve_ex.<locals>.<lambda>  s    K r8   r  Fr   r  c              3   $   K   | ]  }|d u 
 y wr3   r<   )re   rF   s     r6   rg   z#_linalg_solve_ex.<locals>.<genexpr>/  s     
&Q1D=
&s   )	copy_fromcopy_toexact_dtype)r   rI   rZ   rQ   ry  	unsqueezer  r/  r  r   rv   r   r   r5  allzipr!   r   r   r#   )r  r  r   r  r  r9  r  r   rx  B_B_broad_shaperD   result_shaperesult_r   LU_pivots_info_r   resrP  os   ``                    r6   _linalg_solve_exr    s    1n-	LL	177	
 -Q2K'RQBaT>23B:M1	LLK	
 *5="%-L!!*<TBggxx	G GGE


*5%8ggxx	C kk%*EKKk8GKKcr
%++K6E2vt
$CC%
(C

&#
&&SM 	FDAqa)MM!''188:.QuE	F Jr8   )r   unitriangularr   r  r   c                   || j                  dg      }t        |t              sJ t        | ||d       t	        || d       \  }}|j                  dd      j                         xr |j                         }|rt        ||j                        }|S t        ||j                        r=|j                  |j                  dd      j                         |j                  dd       |S )Nr   zlinalg.solve_triangularr  r   )r   rc   r   r  r  r  rR  is_conjr!   r   r"   r   
transpose_)	r  r  r  r   r  r   r  A_avoid_copy_As	            r6   linalg_solve_triangular_metar  9  s     {kk1#c:&&&aD";<.q!T:FB<<B'557HBJJLLRXX. J  RXX.KKR,223NN2r"Jr8   XM)r  r  c                     t        j                   j                  dk\   fd       t        j                  j                  dk\  fd       t         d       j                  t         j
                  k(  rt               \  }}t        j                  |t        |d       j                   j                        }t        j                  |t        |d      j                  j                        }||fS j                  t         j                  k(  sj                  t         j                  k(  r+t        j                         } j                  dg      }||fS t        j                  dd	        fS )
Nr   c                  $    d j                    dS )NzMtorch.triangular_solve: Expected b to have at least 2 dimensions, but it has r  r  r   s   r6   rX   z'triangular_solve_meta.<locals>.<lambda>^  s    ))$79 r8   c                  $    d j                    dS )NzMtorch.triangular_solve: Expected A to have at least 2 dimensions, but it has r  r  r  s   r6   rX   z'triangular_solve_meta.<locals>.<lambda>e  r&  r8   triangular_solveFr  r  r   c                       y)Nz+triangular_solve: Got an unexpected layout.r<   r<   r8   r6   rX   z'triangular_solve_meta.<locals>.<lambda>  r`   r8   )rI   rZ   r   r  ru   stridedr/  r  r   rQ   rv   
sparse_csr
sparse_bsrr   r   )	r   r  r  r  r  self_broadcast_sizeA_broadcast_sizesolutioncloned_coefficients	   ``       r6   triangular_solve_metar  S  sL    
LL		Q	
 
LL	!	
 4$67xx5== 0LTST0U--&&$./BeT**;;	
 #00!./?5Q''88	
 ''' 
U%%	%U5E5E)E##D)!^^QC0 ''' 	UQR'''r8   c                 l   t        | d       t        | d       | j                  | j                  d d       }| j                  | j                        }|j	                  | j                  t        | j                  d             | j                  | j                  d d t        j                        }|||fS )Nz
linalg.detr  Fr  r   r   r  )r  detr9  r  s       r6   _linalg_det_metar    s    a&1l+
++aggcrl
#C	
QWW	BNN17775QR[["U[[[9FF?r8   c                 0    t        j                   j                  dk\  d        t        j                  j                  dk\  d        |rdndt        j                  j                     j                  d   k\  fd       t        j                  j                      j                  d   k(  fd       t        j                  j                  d    j                  d   k  d        t        j                   j                  j                  z
  d	k(   fd
       t        j                   j                  j                  k(   fd        j                  dkD  re j                  d d }j                  d d t        j                  |k(  fd       j                  d d t        j                  |k(  fd       t        j                  j                   j                  k(   fd       t        j                  j                   j                  k(   fd       t        d d       t        d d       t        j                  j                  t        j                  d      j                  j                        S )Nr   c                       y)Nz3torch.ormqr: input must have at least 2 dimensions.r<   r<   r8   r6   rX   zormqr.<locals>.<lambda>  r`   r8   c                       y)Nz3torch.ormqr: other must have at least 2 dimensions.r<   r<   r8   r6   rX   zormqr.<locals>.<lambda>  r`   r8   r  r   c                      d  dS )Ntorch.ormqr: other.shape[z0] must be greater than or equal to tau.shape[-1]r<   left_size_conditions   r6   rX   zormqr.<locals>.<lambda>  s    +,?+@@pq r8   c                      d  dS )Nr  z"] must be equal to input.shape[-2]r<   r  s   r6   rX   zormqr.<locals>.<lambda>  s    +,?+@@bc r8   c                       y)NzHtorch.ormqr: tau.shape[-1] must be less than or equal to input.shape[-1]r<   r<   r8   r6   rX   zormqr.<locals>.<lambda>  r`   r8   r   c                  <    dj                    d j                    S )Nz[torch.ormqr: Expected tau to have one dimension less than input, but got tau.ndim equal to r  r  r  s   r6   rX   zormqr.<locals>.<lambda>  r  r8   c                  <    dj                    d j                    S )Nzhtorch.ormqr: Expected other to have the same number of dimensions as input, but got other.ndim equal to r  r  r   r   s   r6   rX   zormqr.<locals>.<lambda>  s+    ++0::,6RSXS]S]R^` r8   c                      d  S )NzWtorch.ormqr: Expected batch dimensions of tau to be equal to input.shape[:-2], but got r<   r  s   r6   rX   zormqr.<locals>.<lambda>  r  r8   c                      d  S )NzYtorch.ormqr: Expected batch dimensions of other to be equal to input.shape[:-2], but got r<   )actual_batch_other_shapes   r6   rX   zormqr.<locals>.<lambda>  s    66N5OQ r8   c                  <    d j                    dj                    S )NzPtorch.ormqr: Expected input and tau to have the same dtype, but input has dtype z and tau has dtype r   r  s   r6   rX   zormqr.<locals>.<lambda>  s'    ##(;;-/B399+O r8   c                  <    d j                    dj                    S )NzRtorch.ormqr: Expected input and other to have the same dtype, but input has dtype z and other has dtype r   r  s   r6   rX   zormqr.<locals>.<lambda>  s'    ##(;;-/DU[[MS r8   ztorch.ormqrr	  r   Fr  r  )	rI   rZ   r   r   rQ   r  r  r   rv   )	r   r	  r   r   r  expected_batch_shaper  r  r  s	   ```   @@@r6   ormqrr    s    
LL

aV 
LL

aV !%""	LL'(CIIbM9q 
LL'(EKKO;c
 
LL		"R(Z
 
LL

SXX"	
 
LL

ejj 	
 zzA~${{3B/!$3B"&::	
 $);;s#3 $(<<	
 
LL		U[[ 	
 
LLu{{"	
 M3u5M5%9[[*5;;%Hkk||	 r8   c                t    t        j                  t              dz  k(  fd        j                  }|dz   k(  }|}| }|r*t	        d|      D ]  }|xr  j                  |      dk7  } n)t	        d|      D ]  }|xr  j                  |      dk7  } t        j                  |xs | fd       y )Nr   c                  ,    dd z   dt               S )Nzpadding size is expected to be r   z, but got: r   )ry   paddings   r6   rX   z,_padding_check_valid_input.<locals>.<lambda>  s    1!c'+c'l^T r8   r   r   c                  :    d dz    d dz    dj                    S )N	Expected r   zD or r   zcD (batch mode) tensor with possibly 0 batch size and other non-zero dimensions for input, but got: r)  )ry   r   s   r6   rX   z,_padding_check_valid_input.<locals>.<lambda>  s2    aycAgY /AAFO r8   )rI   rZ   r   r   r   r   )r   r  ry   	input_dimis_batch_modevalid_batch_modevalid_non_batch_moder   s   ```     r6   _padding_check_valid_inputr    s    	LLGCT
 

I#'*M$,,q)$ 	GA/FEJJqMQ4F	G q)$ 	OA#7#NEJJqMQ<N 	O 
LL00	
r8   c                   	
 d}dd} j                   dk(  r j                  d      }dz  |dz  }t         |d       |\  	
 j                  |      } j                        	z   
z   |r&t        j                  	k  xr 
k   	
fd       t        j                  dk\  fd        j                   dk(  r j                  |f      S  j                  ||f      S )Nr   r   r.   r   c                  4    d d d  dj                    S NzcArgument #4: Padding size should be less than the corresponding input dimension, but got: padding (rk   ) at dimension 
 of input r)  dim_wr   pad_lpad_rs   r6   rX   z_pad1d_common.<locals>.<lambda>"  2    %%*G2eWOE7*UZU`U`Tac r8   c                      d  d S )Nz
input (W: z%) is too small. Calculated output W: r<   )input_woutput_ws   r6   rX   z_pad1d_common.<locals>.<lambda>*  s    *WI%J8*U r8   r   )r   r   r  rI   rZ   r   )r   r  is_reflection	dim_planenbatchnplaner  r  r  r  r  s   `     @@@@@r6   _pad1d_commonr    s    IEFzzQA
Q	ug15LE5ZZ	"FjjG&HGO/	
 
LLAU
 zzQ1229::r8   c                     t        | |d      S NTr  )r  r   r  s     r6   meta_reflection_pad1dr  3       t<<r8   c                      t        j                   j                  t         j                  k7   fd       t	         |d      S )Nc                  @    d j                   j                          dS )Nz)"replication_pad1d" not implemented for ''rQ   __str__r   s   r6   rX   z(meta_replication_pad1d.<locals>.<lambda>>       =ekk>Q>Q>S=TTUX r8   Fr  )rI   rZ   rQ   boolr  r  s   ` r6   meta_replication_pad1dr  9  5     
LLuzz!X u==r8   c                    d|s#t        j                  t        |      dk(  d        j                  dk(  rdz  |\  j	                        }|z   z   |r&t        j                  |k  xr |k  fd       t        j                   j	                        k(   fd       j                  j                        S )Nr   r   c                       y)Nz padding size is expected to be 2r<   r<   r8   r6   rX   z(_pad1d_backward_common.<locals>.<lambda>F  r`   r8   r.   c                  4    d d d  dj                    S r  r)  r  s   r6   rX   z(_pad1d_backward_common.<locals>.<lambda>S  r  r8   c                  2    d dj                          S Nz(grad_output width unexpected. Expected: , Got: r   r  grad_outputr  s   r6   rX   z(_pad1d_backward_common.<locals>.<lambda>[  "    :8*GKL\L\]bLcKde r8   rI   rZ   r   r   r   r   r   )	r  r   r  r  r  r  r  r  r  s	   ``   @@@@r6   _pad1d_backward_commonr  C  s    ES\Q&(RSzzQ
LE5jjG&HGO/	
 
LLK$$U++e
 ??5;;''r8   
grad_inputc                      t        | ||d      S r  r  r  r   r  s      r6   meta_reflection_pad1d_backwardr  a  s     "+ugTRRr8   c                      t        | ||d      S )NFr  r  r  s      r6   meta_replication_pad1d_backwardr  g  s     "+ugUSSr8   c                   	
 ddd}d}t         |d        j                  }|dk(  r  j                  d      }dz  dz  |dz  }|\   j                  |      } j                        	 j                        
	z   z   
z   z   |rLt        j                  
k  xr 
k   fd       t        j                  	k  xr 	k   fd       t        j                  dk\  xs dk\  	
fd        j                  d	k(  r j                  |f      S  j                  ||f      S )
Nr   r   r   r      c                  4    d d d  dj                    S r  r)  r  s   r6   rX   z_pad2d_common.<locals>.<lambda>  r  r8   c                  4    d d d  dj                    S NzcArgument #6: Padding size should be less than the corresponding input dimension, but got: padding (rk   r  r  r)  dim_hr   pad_bpad_ts   r6   rX   z_pad2d_common.<locals>.<lambda>  r  r8   c                       d  d d d S )Nz
input (H:  W: z%) is too small. Calculated output H: r<   )input_hr  output_hr  s   r6   rX   z_pad2d_common.<locals>.<lambda>  s*    	gY /$$,:T(= r8   r.   r  r   r   rI   rZ   r   )r   r  r  
dim_slicesr  r   r  r  r  r  r  r  r  r   r  r  r  s   `      @@@@@@@@@@r6   _pad2d_commonr  m  sU   EEJFug15::DqyA

a
!(E5%ZZ
#FjjGjjG&H&HGO/	
 	GO/	
 
LLA&Q	
 zzQ(;<<(CDDr8   c                     t        | |d      S r  )r  r  s     r6   meta_reflection_pad2dr
    r  r8   c                      t        j                   j                  t         j                  k7   fd       t	         |d      S )Nc                  @    d j                   j                          dS )Nz)"replication_pad2d" not implemented for 'r  r  r  s   r6   rX   z(meta_replication_pad2d.<locals>.<lambda>  r  r8   Fr  )rI   rZ   rQ   r  r  r  s   ` r6   meta_replication_pad2dr    r  r8   c                     ddd}|j                   }|j                         dk(  rdz  dz  |dz  }|\  }}}}|   }	|   }
|	|z   |z   |
|z   |z   t        j                   j	                        k(   fd       t        j                   j	                        k(   fd       |j                  |j                         S )Nr   r   r   r  c                  2    d dj                          S r  r   r  s   r6   rX   z%meta_pad2d_backward.<locals>.<lambda>  r  r8   c                  2    d dj                          S Nz)grad_output height unexpected. Expected: r  r   r  r  r  s   r6   rX   z%meta_pad2d_backward.<locals>.<lambda>  "    ;H:W[M]M]^cMdLef r8   )r   ry   rI   rZ   r   r   )r  r   r  r  rW   r  r  r  r   r  r  r  r  r  r  s   `          @@@@r6   meta_pad2d_backwardr    s     EEIJxxzQ

Q	!(E5%GG&H&H	LLK$$U++e 
LLK$$U++f >>$**%%r8   c          	      $   	
 d	ddd}t         |d        j                  dk(  }|r% j                  d      }	dz  	dz  dz  |dz  }|\   j                  |      } j                        
 j                         j                  	      
z   z   z   z   z   z   |rrt        j                  k  xr k  	 fd       t        j                  k  xr k   fd       t        j                  
k  xr 
k   fd	       t        j                  dk\  xs dk\  xs dk\  
fd
       |r j                  |f      S  j                  |f      S )Nr.   r   r   r   r      c                  4    d d d  dj                    S r  r)  r  s   r6   rX   z_pad3d_common.<locals>.<lambda>  r  r8   c                  4    d d d  dj                    S r  r)  r  s   r6   rX   z_pad3d_common.<locals>.<lambda>  r  r8   c                  4    d d d  dj                    S )NzcArgument #8: Padding size should be less than the corresponding input dimension, but got: padding (rk   r  r  r)  )dim_dr   pad_bkpad_fs   r6   rX   z_pad3d_common.<locals>.<lambda>  s2    %%*G2fX_UG:V[VaVaUbd r8   c                  ,    d  d d d d d S )Nz
input (D:  H: r  z%) is too small. Calculated output D: r<   )input_dr  r  output_dr  r  s   r6   rX   z_pad3d_common.<locals>.<lambda>  s7    	gYd7) <$$,:T(4zK r8   r  )r   r  r  r  
batch_moder  r  r  r  r  r  r  r  r   r  r  r   r  r  r  r  r  s   `      @@@@@@@@@@@@@@@r6   _pad3d_commonr"    s   EEEIug15qJA


Q	07-E5%vZZ	"FjjGjjGjjG'H&H&HGO/	
 	GO/	
 	GO0 0	
 
LLA7Q7(a-	
 	
 (HMNN(HEFFr8   c                     t        | |d      S r  )r"  r  s     r6   meta_reflection_pad3dr$    r  r8   c                      t        j                   j                  t         j                  k7   fd       t	         |d      S )Nc                  @    d j                   j                          dS )Nz)"replication_pad3d" not implemented for 'r  r  r  s   r6   rX   z(meta_replication_pad3d.<locals>.<lambda>"  r  r8   Fr  )rI   rZ   rQ   r  r"  r  s   ` r6   meta_replication_pad3dr'    r  r8   c                     t        j                  t        |      dk(  d        |j                  dkD  sJ  j                  |j                  k(  sJ ddd|j                  dk(  rdz  dz  dz  |\  }}}}}}|j	                        }	|j	                        }
|j	                        }|	|z   |z   |
|z   |z   ||z   |z   t        j                   j	                        k(   fd       t        j                   j	                        k(   fd       t        j                   j	                        k(   fd	       |j                  |j                        S )
N   c                       y)Nz padding size is expected to be 6r<   r<   r8   r6   rX   z%meta_pad3d_backward.<locals>.<lambda>1  r`   r8   r.   r   r   r  c                  2    d dj                          S r  r   r  s   r6   rX   z%meta_pad3d_backward.<locals>.<lambda>I  r  r8   c                  2    d dj                          S r  r   r  s   r6   rX   z%meta_pad3d_backward.<locals>.<lambda>M  r  r8   c                  2    d dj                          S )Nz(grad_output depth unexpected. Expected: r  r   )r  r  r   s   r6   rX   z%meta_pad3d_backward.<locals>.<lambda>Q  r  r8   r  )r  r   r  r  r  r  r   r  r  r  r  r  r  r  r  r   r  r  s   `           @@@@@@r6   meta_pad3d_backwardr.  '  s_    
LLW"$NO::>>uzz)))EEEzzQ


07-E5%vjjGjjGjjG'H&H&H	LLK$$U++e 
LLK$$U++f 
LLK$$U++e
 ??5;;''r8   pc                 J   t        j                  | j                         d        | j                  d      }|dk  r0| j	                  dg      j                  t         j                        S | j	                  ||dz
  z  dz  f      j                  t         j                        S )Nc                       y)Nz(_pdist_forward requires contiguous inputr<   r<   r8   r6   rX   z%meta__pdist_forward.<locals>.<lambda>[  r`   r8   r   r   r   r   )rI   rZ   rR  r   r   r  r  )r   r/  r   s      r6   meta__pdist_forwardr2  W  s     
LLP 			!AAv~~qc"%%E4R4R%SS~~qAE{a/125588 6 
 	
r8   gradpdistc                     t        j                  |j                         d        t        j                  |j                         d        t        j                  |t         j                        S )Nc                       y)Nz._pdist_backward requires self to be contiguousr<   r<   r8   r6   rX   z&meta__pdist_backward.<locals>.<lambda>j  r`   r8   c                       y)Nz/_pdist_backward requires pdist to be contiguousr<   r<   r8   r6   rX   z&meta__pdist_backward.<locals>.<lambda>m  r`   r8   r   )rI   rZ   rR  r   r  )r3  r   r/  r4  s       r6   meta__pdist_backwardr8  f  sW     
LLV 
LLX D0N0NOOr8   )rA  r@  c          
      0    ddl m}m} j                  d      }j                  d      }j                  d      }	 |t	        j
                   | j                  |||	f                  r j                  |||	f       t	        j                  j                         dk(  d        t	        j                  j                         dk(  d        t        j                  sGt	        j                   j                  j                  cxk(  xr j                  k(  nc  fd       j                  }
j                  |
d   |
d   t	        j                  d   k(  xr d   k(  fd	        j                   j                               S )
Nr   )guard_or_truesym_eqr   r   r.   c                       yNzbatch1 must be a 3D tensorr<   r<   r8   r6   rX   zmeta_baddbmm.<locals>.<lambda>|  r`   r8   c                       yNzbatch2 must be a 3D tensorr<   r<   r8   r6   rX   zmeta_baddbmm.<locals>.<lambda>}  r`   r8   c                  V    dj                    d j                    dj                    S )Nz+Input dtypes must be the same, got: input: z
, batch1: z
, batch2: r   )batch1batch2r   s   r6   rX   zmeta_baddbmm.<locals>.<lambda>  s0    A$**ZX^XdXdWeeopvp|p|o}~ r8   c            	      .    d d d d    d d    d	S Nz@Expected size for first two dimensions of batch2 tensor to be: [rk   z] but got: [r   r   ].r<   batch2_sizesbscontraction_sizes   r6   rX   zmeta_baddbmm.<locals>.<lambda>  s:    t2&'|LO3DB|TUFWWY[ r8   )r  r:  r;  r   rI   sym_notr   rs  rZ   ry   
exp_config&skip_dtype_check_in_meta_registrationsrQ   r   )r   rA  rB  rA  r@  r:  r;  dim1dim2dim3batch1_sizesrG  rH  rI  s   ```        @@@r6   meta_baddbmmrQ  r  s7    L;;q>D;;q>D;;q>DU]]6$**tT46H#IJK{{D$-.	LL"$HI	LL"$HI<<JJ&,,6&,,6~	
 <<L<<L	aB#A	LLQ2E,q/5E"E	
 >>$))+&&r8   c                L    t        j                  | t         j                        S r   r]  r   r   s     r6   meta_bernoullirT    s     D0G0GHHr8   c                     | S r3   r<   r   r/  r   s      r6   meta_bernoulli_rW        Kr8   c                 L    t        j                  | t         j                        S r   r]  rV  s      r6   meta_bernoulli_prZ    s     D0G0GHHr8   c                 ,    t        j                  |       S r3   rI   r   rS  s     r6   meta_poissonr]         D!!r8   c                     t        j                  |
| j                         k  d        t        j                  | t         j                        }t        j                  |       |fS )Nc                       y)NzJError in fused_moving_avg_obs_fake_quant_cpu: ch_axis must be < self.dim()r<   r<   r8   r6   rX   z6meta__fused_moving_avg_obs_fq_helper.<locals>.<lambda>  r`   r8   r   )rI   rZ   ry   r   r  )r   observer_onfake_quant_onrunning_minrunning_maxscale
zero_pointaveraging_const	quant_min	quant_maxch_axisper_row_fake_quantsymmetric_quantmasks                 r6   $meta__fused_moving_avg_obs_fq_helperrn    sO      
LL$((*\ D

3DT"D))r8   c                 H   t        j                  | j                         dk(  d        t        j                  |j                         dk(  d        | j                  \  |j                  \  t        j                  k(  fd       | j	                        S )Nr   c                       y)Nza must be 2Dr<   r<   r8   r6   rX   zmeta_mm.<locals>.<lambda>  r`   r8   c                       y)Nzb must be 2Dr<   r<   r8   r6   rX   zmeta_mm.<locals>.<lambda>  r`   r8   c            	      "    d d  d d d	S )Nz/a and b must have same reduction dim, but got [rk   z] X [rE  r<   )M1M2Nr3  s   r6   rX   zmeta_mm.<locals>.<lambda>  s(    A!Brd%PRtSUVWUXXZ[ r8   )rI   rZ   ry   r   r   )abrs  rt  ru  r3  s     @@@@r6   meta_mmrx    sz     
LLA56	LLA56GGEArGGEB	LL
b[ ;;q!r8   c                      |r(t         fdt         j                        D              S t        j                   j
                        S )Nc              3   H   K   | ]  }|vrj                   |   nd   yw)r   Nr)  )re   r   dimsr   s     r6   rg   z+_compute_reduction_shape.<locals>.<genexpr>  s$     UqatmTZZ]:Us   ")rY   r   r   rA   compute_reduction_output_shaper   )r   r{  r~  s   `` r6   r|  r|    s7    UE$))DTUUU//

DAAr8   c                    t        | t        j                  j                        r| j                  j
                  S t        | d      rEt        | j                  d      r/| j                  j
                  dk7  r| j                  j
                  S y)Nrv   rm   rs   r   )rc   rI   _subclasses
FakeTensorfake_devicerm   hasattrrv   )r$  s    r6   r   r     sg    &%++667!!&&&!FMM6*MM&(}}!!!r8   input_tensorr   r  dilationis_transposedgroupsoutput_paddingc                 $   dt         dt         dt         dt         dt         dt         fd}dt         dt         dt         dt         dt         dt         dt         fd	}	|j                  d
d  }
| j                  d
d  |r||j                  d   z  }n<|j                  d   }|j                  d   |z  | j                  d   k7  rt        d      | j                  d   |gt        |t              r|gt              z  }n t        |      dk(  r|d   gt              z  }t        |t              r|gt              z  }n t        |      dk(  r|d   gt              z  }t        |t              r|gt              z  }n t        |      dk(  r|d   gt              z  }d }|rCt        |t              r|gt              z  }n#t        |      dk(  r|d   gt              z  }n|}t        t                    D ]]  }|r/j                   |	|   ||   ||   |
|   ||   ||                4j                   ||   ||   ||   |
|   ||                _ ddlm	} t        j                   |d
d  D cg c]  }|dkD  	 c} fd       S c c}w )Nlnr/  r   rU  r  r/   c                 6    | d|z  z   ||dz
  z  z
  dz
  |z  dz   S )a  
        Formula to apply to calculate the length of some dimension of the output

        See: https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html

        Args:
            ln: length of the dimension
            p: padding in that dim
            d: dilation in that dim
            k: kernel size in that dim
            s: stride in that dim
        Returns:
            The output length
        r   r   r<   )r  r/  r   rU  r  s        r6   _formulaz+calc_conv_nd_return_shape.<locals>._formula  s.     QU
Q!a%[(1,2Q66r8   r4   c                 <    | dz
  |z  d|z  z
  ||dz
  z  z   |z   dz   S )a  
        Formula to apply to calculate the length of some dimension of the output
        if transposed convolution is used.
        See: https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html

        Args:
            ln: length of the dimension
            p: padding in that dim
            d: dilation in that dim
            k: kernel size in that dim
            s: stride in that dim
            op: output padding in that dim

        Returns:
            The output length
        r   r   r<   )r  r/  r   rU  r  r4   s         r6   _formula_transposedz6calc_conv_nd_return_shape.<locals>._formula_transposed	  s2    " Q!|a!e#a1q5k1B6::r8   r   r   r   zInvalid channel dimensions)sym_orc                  .    dt                ddd   dS )NzGiven input size per channel: z&. Calculated output size per channel: r   z. Output size is too small)r   )r{  	ret_shapes   r6   rX   z+calc_conv_nd_return_shape.<locals>.<lambda>L	  s*    0d =//8}o >#$ r8   )r   r   r  rc   r   r   r   r   r  r  rI   rZ   )r  r-  r   r  r  r  r  r  r  r  kernel_sizeout_channelsoutput_padding_listr   r  rF   r{  r  s                   @@r6   calc_conv_nd_return_shaper    s   7S 7S 7S 7S 7S 7S 7"; ; ; ; ; ; ;QT ;& ,,qr"Kab!DQ/||A<<?V#|'9'9!'<<;<<##A&5I&'"CI%	V	)s4y('7#)c$i'	W	1:,T*(G$:D	)	X!	QK=3t9,/3ng.#1"2SY"> A%#1!#4"5D	"A"03t9 #GAJQKN1I'*	 a'!*hqk;q>6RS9U" =	LL	!".1Q./	$  /s   2Jc                 b    t         j                  j                  |       t         j                  k(  S r3   rI   _prims_commonr   channels_lasttens    r6   is_channels_lastr  T	  s$    44S9U=P=PPPr8   running_meanrunning_vartrainingexponential_average_factorepsilonc                 r     j                   }||j                   n|j                   }	||j                   n|j                   }
 fd} j                  |      j                   |             }|r# j                  |	      } j                  |
      }n" j                  d      } j                  d      }|||fS )Nc                      t               rt        j                  S  j                  t        j                        rt        j                  S t        j                  S r   )r  rI   r  rR  r   )r  s   r6   pick_memory_formatz2meta_miopen_batch_norm.<locals>.pick_memory_formatk	  sF    L)&&&%%E4K4K%L***&&&r8   r   r   )r   r   r  )r  r-  r/  r  r  r  r  r  r   save_mean_shapesave_var_shaper  r   	save_meansave_vars   `              r6   meta_miopen_batch_normr  X	  s     ""I -9,Dl((&,,O*5*A[&&v||N' 
 
 
+
.
.=O=Q
.
RC **?;	)).9 **40	))$/	8##r8   c	           
            fd}	t         ||||||r|nd       }
d}d} j                  |      dk(  rd|
|<    j                  |
      }|j                   |	             }|S )Nc                  d   t               dk(  r&t               st              r+t        j                  S t               rt        j                  S  j	                  t        j
                        rt        j
                  S  j	                  t        j                        rt        j                  S y Nr   r   )r   r  rI   r  rR  r   preserve_format)r  r-  s   r6   r  z%meta_conv.<locals>.pick_memory_format	  s    |$.-1A&1I***-***%%E4K4K%L***''e6K6K'L((( Mr8   r   r   r   )r  r   r   r  )r  r-  r/  r   r  r  r  r  r  r  	shape_outinput_channels_dimoutput_channels_dimr   s   ``            r6   	meta_convr  ~	  s    
) *'T	I +,1)*	%&

 
 
+C
&&13&
4CJr8   mkldnnc
           
          t        | ||||d|g       }
| j                  |
      }t        j                  }| j	                         dk(  rt        j
                  }|j                  |      }|S )NFr  r   )r  r   rI   r  ry   channels_last_3dr  )r  r-  r/  r  r   r  r  attrscalars	algorithmr  r   out_memory_formats                r6   meta_mkldnn_convolution_defaultr  	  sp     .&&'8UFB
	 $$Y/!//" % 6 6ff#4f5
r8   c                 b    | j                  g | j                  d d |j                  d         S Nr   r   r   r   )r  r-  r/  r  r  r  s         r6   meta_linear_pointwise_defaultr  	  s5     %%&Q(:(:3B(?&Qa&QRRr8   mklc                 b    | j                  g | j                  d d |j                  d         S r  r  )r  packed_weightorig_weightr/  r   s        r6   meta_mkl_linearr  	  s:    ))@,$$Sb)@;+<+<Q+?@ r8   onednnc           
         t        | ||||	d|
d       }|| j                  }|t        j                  t        j                  t        j
                  t        j                  t        j                  fv sJ | j                  ||      }t        |      dv sJ d       t        j                  t        j                  t        j                  dt        |         }|j                  |      }|S )NFr   )r.   r  r  z-Expect output to be 3d/4d/5d for conv1d/2d/3dr   )r  rQ   rI   rN  rP  uint8r4  rQ  r   r   r   r  r  r  )rF   x_scalex_zpww_scalew_zpr/  r   r  r  r  output_scaleoutput_zero_pointoutput_dtyper  r  r  r  r   formats                       r6   meta_qconv_pointwiser  	  s    * .	
	 77LMMNNKKJJ 
 
 	
 
 kk)<k89~* 	
;	
* &&""%%
 i.	
 ff6f*
r8   c                     |dk(  sJ |S )Nsumr<   )rF   r  r  r  r  r  accumr/  r   r  r  r  r  r  r  accum_scaleaccum_zero_pointbinary_op_namer@  unary_op_nameunary_op_argsunary_op_algorithms                         r6   meta_qconv2d_pointwise_binaryr  
  s    2 &&&r8   c                    t        | j                        }|j                  d   |d<   |	t        j                  t        j                  t        j
                  t        j                  t        j                  fv sJ | j                  ||	      }|S )Nr   r   r   )	r   r   rI   rN  rP  r4  r  rQ  r   )rF   r  r  r  r  r  r/  r  r  r  post_op_namepost_op_argspost_op_algorithmrW  r   s                  r6   meta_qlinear_pointwiser  ,
  sx    " AGG}771:RMMNNJJKK 
 
 	
 
 kk,lk;
r8   c                 *   |dk(  r|S t        | j                        }|j                  d   |d<   |
t        j                  t        j                  t        j
                  t        j                  t        j                  fv sJ | j                  ||
      }|S )Nr  r   r   r   )	r   r   rI   rN  rP  r  r4  rQ  r   )rF   r  r  r  r  r  x_2r/  r  r  r  x2_scalex2_zpr  r@  r  r  r  rW  r   s                       r6   meta_qlinear_pointwise_binaryr  J
  s    , U"JAGG}771:RMMNNKKJJ 
 
 	
 
 kk,lk;
r8   c                 v    t        | j                        }|j                  d   |d<   | j                  |      }|S )Nr   r   )r   r   r   )rF   r  r/  rW  r   s        r6   meta_linear_dynamic_fp16r  o
  s6     AGG}771:Rkk,'
r8   	quantizedr  r   c                 .   t        | |||||      \  }}}| j                         dk(  r| j                  d      nd}	t        j                  }
| j                         dk(  r|||g}n|	|||g}t        j
                  || j                  | j                  |
      S Nr  r   r.   rn  )#max_pool2d_checks_and_compute_shapery   r   rI   r  r~   rQ   rv   r   r  r   r  r  	ceil_modenInputPlaneoutputHeightoutputWidthr  r   r   s               r6   meta_quantized_max_pool2dr  
  s     0;9
		
 $)99;!#3B++99;!{;DK{CD{{++<<'	
 	
r8   c                    t        j                  | j                         dk(  d| j                          d       t        j                  |j                         dk(  d|j                          d       t        j                  | j                  t         j                  t         j
                  t         j                  fv d| j                          t        j                  |j                  t         j                  k(  d|j                          t        j                  |j                  t         j                  k(  d|j                          t        j                  |j                  | j                  k(  d|j                          | j                  | j                  d	      |j                  d	      | j                  
      S )Nr   zx must be a 2D tensor, got Dzw must be a 2D tensor, got #expected x to be f32/f16/bf16, got expected w to be uint8, got z q_group_size must be int64, got z5q_scale_and_zeros must have the same dtype as x, got r   r   )rI   rZ   ry   rQ   rN  rO  rP  r  r   r   r   rF   r  q_group_sizeq_scale_and_zeross       r6   meta_int4mm_packed_weight_cpur  
  s@   QUUW\%@	#KLQUUW\%@	#KLGGu}}ennEE1!'';	
 	QWW+/KAGG9-UV%++-.|/A/A.BC	
 	##qww.CDUD[D[C\]	
 {{166!9affQiqww{??r8   c                      t        j                   j                         k(  xr  j                     k(   fd       y )Nc                  j    d  d d ddj                          d dj                      z   S )NzExpected a tensor of dimension z and tensor.size[z] == rk   zbut got : dimension z] = ry   r   )ry   dim_sizer   r$  s   r6   rX   z check_dim_size.<locals>.<lambda>
  sP    1#6GzQVW[V\\^_ .?zfll[cNdMe
fg r8   )rI   rZ   ry   r   )r$  ry   r  r   s   ````r6   check_dim_sizer  
  s6    	LL

>X 6$ >	gr8   c                     d } |d|      \  }}	t        j                  t        |      dv d        t        j                   j                  t         j                  t         j
                  t         j                  t         j                  fv fd       t        |      dk(  r||	}}
n%t        |      dk(  r|d   |d   }}
n |d|      \  }
} |d	|      \  }}t        j                  |d u xs |dk7  d
         j                         dk(  r j                  d      nd} j                  d      } j                  d      } j                  d      }t        ||||
d|      }t        ||	||d|      }t        j                         }t         ||	|
|||dd||||||        j                         dk(  r|||g}n||||g}t        j                  | j                   j                  |      S )Nc                      t        j                  t        |      dv  fd       |d   }t        |      dk(  r|n|d   }||fS )Nr   r   c                      d  dS )Nzavg_pool2d: 4 must either be a single int, or a tuple of two intsr<   r  s   r6   rX   z1meta_avg_pool2d.<locals>.unpack.<locals>.<lambda>
      l4&(\] r8   r   r   rI   rZ   r   r  r  HWs   `   r6   unpackzmeta_avg_pool2d.<locals>.unpack
  G    H]	
 FSQACF!tr8   r  r   r   r   c                       yNzOavg_pool2d: stride must either be omitted, a single int, or a tuple of two intsr<   r<   r8   r6   rX   z!meta_avg_pool2d.<locals>.<lambda>
  r`   r8   c                  @    d j                   j                          dS )Nz""avg_pool2d" not implemented for 'r  r  r  s   r6   rX   z!meta_avg_pool2d.<locals>.<lambda>
      6u{{7J7J7L6MQQ r8   r   r   r   r  c                       yNzdivisor must be not zeror<   r<   r8   r6   rX   z!meta_avg_pool2d.<locals>.<lambda>
  r`   r8   r  r  r  r   r.   rn  )rI   rZ   r   rQ   r  uint16uint32uint64ry   r   pooling_output_shaperA   r   pool2d_shape_checkr~   rv   )r   r  r   r  r  count_include_paddivisor_overrider	  kHkWdHdWpadHpadWr  r  inputHeight
inputWidthr  r  r   r   s   `                     r6   meta_avg_pool2dr"  
  s    M;/FB	LLFy a 
LLEKKu||U\\RRQ 6{aRB	V	F1IB&)B	7+JD$	LLD 9$4$9*
  %yy{a/UZZ^QF**R.K**R.KBJ'Rr1iPL&z2tRINK//6M



		$ yy{a\;7\;?;;kk||#	 r8   c                     t        | ||||||dd|	|
||||       | j                         }|	}t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       y )Nr   r.   r   )r  ry   r  )r   
gradOutputr  r  r  r  r  r  r  r  r   r!  r  r  
mem_formatr   nOutputPlanes                    r6   avg_pool2d_backward_shape_checkr'    s    " 



		$ 99;DL:tTAX|<:tTAX|<:tTAX{;r8   c                    t        j                  t        |      dk(  xs t        |      dk(  d        |d   }t        |      dk(  r|n|d   }	t        j                  t        |      dk(  xs t        |      dk(  xs t        |      dk(  d        t        |      dk(  r|n|d   }
t        |      dk(  r|	nt        |      dk(  r|
n|d   }t        j                  t        |      dk(  xs t        |      dk(  d        |d   }t        |      dk(  r|n|d   }t        j                  |d u xs |dk7  d        |j                  }|j	                         dk(  r|d	   nd}|d
   }|d   }|d   }t        ||||
d|      }t        ||	||d|      }t        j                  |      }t        || |||	|
|||||||||       t        j                  ||j                  |j                  |      S )Nr   r   c                       y)NzKavg_pool2d: kernel_size must either be a single int, or a tuple of two intsr<   r<   r8   r6   rX   z*meta_avg_pool2d_backward.<locals>.<lambda>F  r`   r8   r   c                       yr  r<   r<   r8   r6   rX   z*meta_avg_pool2d_backward.<locals>.<lambda>L  r`   r8   c                       y)NzGavg_pool2d: padding must either be a single int, or a tuple of two intsr<   r<   r8   r6   rX   z*meta_avg_pool2d_backward.<locals>.<lambda>R  r`   r8   c                       yr  r<   r<   r8   r6   rX   z*meta_avg_pool2d_backward.<locals>.<lambda>Y  r`   r8   r  r  r  r  r   rn  )rI   rZ   r   r   ry   r  rA   r   r'  r~   rQ   rv   )gradOutput_r   r  r   r  r  r  r  r  r  r  r  r  r  
input_sizer  r  r   r!  r  r  r%  s                         r6   meta_avg_pool2d_backwardr/  8  s    
LLKA6[!1Q!6] 
QB;1$+a.B	LLFq@CK1,@Fq0@a 6{aVAYB6{a3v;!+;RB	LLG.S\Q.Y 1:Dw<1$4'!*D	LLD 9$4$9*
 J$yy{a/Z^QFR.KR.KBJ'Rr1iPL&z2tRINK,,U3J#



$ ;;kk|| 	 r8   c                     t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }	t        j                  | xs t        |      dv d        t        j                   j                  t         j                  t         j
                  t         j                  t         j                  fv fd       |s|n|d   }
|s|nt        |      dk(  r|
n|d   }|s|	nt        |      dk(  r|
n|d   }t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }t        j                   j                  d	v d
        t        j                  | xs |dk7  d         j                  d      } j                  d      } j                  d      } j                  d      } j                  d      }t        ||||
d|      }t        ||||d|      }t        ||	||d|      }t         ||||	|
|||||ddd||||||dd        j                  dk(  r j                  ||||f      S  j                  |||||f      S )Nr   r.   c                       yNzFavg_pool3d: kernel_size must be a single int, or a tuple of three intsr<   r<   r8   r6   rX   z!meta_avg_pool3d.<locals>.<lambda>  r`   r8   r   r   r   c                       yNzJavg_pool3d: stride must be omitted, a single int, or a tuple of three intsr<   r<   r8   r6   rX   z!meta_avg_pool3d.<locals>.<lambda>  r`   r8   c                  @    d j                   j                          dS )Nz""avg_pool3d" not implemented for 'r  r  r  s   r6   rX   z!meta_avg_pool3d.<locals>.<lambda>  r  r8   c                       yNzBavg_pool3d: padding must be a single int, or a tuple of three intsr<   r<   r8   r6   rX   z!meta_avg_pool3d.<locals>.<lambda>  r`   r8   r  r  c                       yNz9non-empty 4D or 5D (batch mode) tensor expected for inputr<   r<   r8   r6   rX   z!meta_avg_pool3d.<locals>.<lambda>  r`   r8   c                       yr  r<   r<   r8   r6   rX   z!meta_avg_pool3d.<locals>.<lambda>  r`   r8   r  r  r  r   zavg_pool3d()T)check_input_sizer  )rI   rZ   r   rQ   r  r  r  r  r   r   r  pool3d_shape_checkr   )r   r  r   r  r  r  r  kTr  r  dTr  r  padTr  r  r  nslicesitimeiheightiwidthotimeoheightowidths   `                       r6   meta_avg_pool3drI    s    
LLKF"X 
QB;1$+a.B;1$+a.B	LL
+c&kV+\ 
LLEKKu||U\\RRQ vayBc&kQ&6F1IBc&kQ&6F1IB	LLGT 1:Dw<1$4'!*Dw<1$4'!*D	LL

fK
 
LL5 0A 5*
 ZZ]FjjnGJJrNEjjnGZZ^F D"aCE"7Bb!YGG!&"dB9EF





			-2 zzQ@AAHIIr8   c                    t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }	t        |      dk(  r|n|d   }
t        j                  | xs t        |      dv d        |s|n|d   }|s|	nt        |      dk(  r|n|d   }|s|
nt        |      dk(  r|n|d   }t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }t        j                  |j                  dv d	        t        j                  | xs |dk7  d
        |j	                  d      }|j	                  d      }|j	                  d      }|j	                  d      }t        ||||d|      }t        ||	||d|      }t        ||
||d|      }t        || |||	|
||||||||||||d       |j                  |j                        S )Nr1  c                       yr3  r<   r<   r8   r6   rX   z*meta_avg_pool3d_backward.<locals>.<lambda>  r`   r8   r   r   r   c                       yr5  r<   r<   r8   r6   rX   z*meta_avg_pool3d_backward.<locals>.<lambda>  r`   r8   c                       yr8  r<   r<   r8   r6   rX   z*meta_avg_pool3d_backward.<locals>.<lambda>  r`   r8   r9  c                       yr;  r<   r<   r8   r6   rX   z*meta_avg_pool3d_backward.<locals>.<lambda>  r`   r8   c                       yr  r<   r<   r8   r6   rX   z*meta_avg_pool3d_backward.<locals>.<lambda>  r`   r8   r  r  r  r   zavg_pool3d_backward())	rI   rZ   r   r   r   r  avg_pool3d_backward_shape_checkr   r   )r  r   r  r   r  r  r  r  r?  r  r  r@  r  r  rA  r  r  rB  rC  rD  rE  otime_for_shape_checkoheight_for_shape_checkowidth_for_shape_checks                           r6   meta_avg_pool3d_backwardrT    s    
LLKF"X 
QB;1$+a.B;1$+a.B	LL
+c&kV+\ vayBc&kQ&6F1IBc&kQ&6F1IB	LLGT 1:Dw<1$4'!*Dw<1$4'!*D	LL

fK
 
LL5 0A 5*
 jjnGJJrNEjjnGZZ^F0D"aS27Bb!YW1&"dB9U#





', ??5;;''r8   c                 ,    t        j                   j                  dk(  xs  j                  dk(   fd        j                  d d t	        |      z   }t        j                         }t        j                  | j                   j                  |      S )Nr.   r  c                  "    d j                    S )Nz"Expected 3D or 4D tensor, but got r)  r   s   r6   rX   z*meta_adaptive_avg_pool2d.<locals>.<lambda>/      4TZZLA r8   r  rn  )
rI   rZ   r   r   rY   rA   r   r~   rQ   rv   )r   output_sizerW  r   s   `   r6   meta_adaptive_avg_pool2drY  +  s|    	LL		Q($))q.A ::cr?U;%77L//5M ;;jj{{#	 r8   c                      t        j                   j                  dk(  xs  j                  dk(   fd        j                   j                  d d t        |      z         S )Nr  r  c                  "    d j                    S )Nz"Expected 4D or 5D tensor, but got r)  r   s   r6   rX   z*meta_adaptive_avg_pool3d.<locals>.<lambda>A  rW  r8   r  )rI   rZ   r   r   r   rY   )r   rX  s   ` r6   meta_adaptive_avg_pool3dr\  =  sO    	LL		Q($))q.A >>$**Sb/E+,>>??r8   c                      j                   }t        d|      D ].  t        j                   j	                        dkD   fd       0 t        j                  |dk(  xs |dk(  fd       t        j                  j
                   j
                  k(   fd       t        j                  }t              rt        j                  }j                  j                        j                  |      S )	Nr   r   c                  *    d j                    d dS )Nz{adaptive_avg_pool2d_backward(): Expected grad_output to have non-zero                       size for non-batch dimensions,  with dimension  being emptyr)  )grad_outr   s   r6   rX   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>L  s&     66>nn5EEUVWUXXdf r8   r.   r  c                  "    d j                    S )NzBadaptive_avg_pool2d_backward(): Expected 3D or 4D tensor, but got r)  r   s   r6   rX   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>Q  s    TUYU_U_T`a r8   c                  <    dj                    d j                    S Nexpected dtype z! for `grad_output` but got dtype r   )ra  r   s   r6   rX   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>U  s    /$**-Nx~~N^_ r8   r   )r   r   rI   rZ   r   rQ   r   r  r  r   r   r  )ra  r   r   r   r   s   ``  @r6   "meta__adaptive_avg_pool2d_backwardrf  F  s    ==D1d^ 
MM!q f	

 
LL	TQYa 
LL

hnn$_ ++M++>>$**%((}(EEr8   c                 d    t        | d       t        j                  |t        j                        S )Nadaptive_avg_pool3d_backwardr   )!_adaptive_pool_empty_output_checkrI   r   r  r  r   s     r6   "meta__adaptive_avg_pool3d_backwardrk  ]  s(     &k3QRD0N0NOOr8   r  c                       j                   }t        d|      D ]/  t        j                   j	                        dkD   fd       1 y )Nr   r   c                  .      dj                    d dS )Nzc(): Expected grad_output to have non-zero size for non-batch dimensions, but grad_output has sizes r_  r`  r)  )r  r  r   s   r6   rX   z3_adaptive_pool_empty_output_check.<locals>.<lambda>i  s/    * --8->->,??OPQsR^` r8   )r   r   rI   rZ   r   )r  r  r   r   s   `` @r6   ri  ri  d  sG    D1d^ 
Q!#	

r8   c                      j                   }t        j                  |dv  fd       t        d|      D ].  t        j                   j	                        dkD   fd       0 t        j                  t        |      dk(  d        d}d}d} j                   dk(  r j	                  d      }|dz  } j	                  |dz
        }|\  }} j                   d	k(  r;|||f} j                  |      }	 j                  |t        j                  
      }
|	|
fS ||||f}t        j                         } j                  |      j                  |      }	 j                  |t        j                  
      j                  |      }
|	|
fS )Nr.   r  c                  "    d j                    S )Nz:adaptive_max_pool2d(): Expected 3D or 4D tensor, but got: r)  r  s   r6   rX   z*meta_adaptive_max_pool2d.<locals>.<lambda>v      LU[[MZ r8   r   r   c                  *    dj                    d  dS )Nzjadaptive_max_pool2d(): Expected input to have non-zero size for non-batch dimensions, but input has sizes r_  r`  r)  r   r   s   r6   rX   z*meta_adaptive_max_pool2d.<locals>.<lambda>{  %    '',{{m3CA3lT r8   r   c                       y)NzCadaptive_max_pool2d(): internal error: output_size.size() must be 2r<   r<   r8   r6   rX   z*meta_adaptive_max_pool2d.<locals>.<lambda>  r`   r8   r  r.   r   r   )r   rI   rZ   r   r   r   r   r   rA   r   r  )r   rX  r   dimHsizeBsizeDosizeHosizeWr   r   r   r   r   s   `           @r6   meta_adaptive_max_pool2dr{  p  s|    ::D	LLZ 1d^ 
JJqMA	

 
LLKAU
 DEEzzQ

1	JJtax E NFFzzQFF+	ooi(//)5;;/?G|E662	33E:ooi(++-+H//)5;;/?BB' C 
 G|r8   c                 N     j                   }t        j                  |dv  fd       t         d       t        j                  j                   j                  k(   fd       t        j                        }j                  j                        j                  |      S )Nro  c                  "    d j                    S )NzKadaptive_max_pooling2d_backward(): Expected 3D or 4D grad_output, but got: r)  r  s   r6   rX   z3meta_adaptive_max_pool2d_backward.<locals>.<lambda>  s    ]^i^o^o]pq r8   adaptive_max_pool2d_backwardc                  <    dj                    d j                    S rd  r   )r  r   s   r6   rX   z3meta_adaptive_max_pool2d_backward.<locals>.<lambda>  s!    /%++.OP[PaPaObc r8   r   )
r   rI   rZ   ri  rQ   rA   r   r   r   r  )r  r   r   r   r   s   ``   r6   !meta_adaptive_max_pool2d_backwardr    s     D	LLq
 &k3QR	LL{(((c
 //6M??5;;'***GGr8   c                      j                   }t        j                  |dv  fd       t        d|      D ].  t        j                   j	                        dkD   fd       0 t        j                  t        |      dk(  d        d}d}d}|dk(  r j	                  d      }|dz  } j	                  |      }|\  }}}|d	k(  r||||f}	n|||||f}	 j                  |	      }
 j                  |	t        j                  
      }|
|fS )Nr9  c                  "    d j                    S )Nz:adaptive_max_pool3d(): Expected 4D or 5D tensor, but got: r)  r  s   r6   rX   z*meta_adaptive_max_pool3d.<locals>.<lambda>  rq  r8   r   r   c                  *    dj                    d  dS )Nzjadaptive_max_pool3d(): Expected input to have non-zero size for non-batch dimensions, but input has sizes r_  r`  r)  rs  s   r6   rX   z*meta_adaptive_max_pool3d.<locals>.<lambda>  rt  r8   r.   c                       y)NzCadaptive_max_pool3d(): internal error: output_size.size() must be 3r<   r<   r8   r6   rX   z*meta_adaptive_max_pool3d.<locals>.<lambda>  r`   r8   r  r  r   )r   rI   rZ   r   r   r   r   r   )r   rX  r   dimDrw  rx  osizeTry  rz  r   r   r   r   s   `           @r6   meta_adaptive_max_pool3dr    s    ::D	LLZ 1d^ 
JJqMA	

 
LLKAU
 DEEqy

1	JJtE(FFFqyFFF3	E666:	
//)
$Cooiu{{o;G<r8   c                 P    t        | d       |j                  |j                        S )Nadaptive_max_pool3d_backward)ri  r   r   )r  r   r   s      r6   !meta_adaptive_max_pool3d_backwardr    s"     &k3QR??5;;''r8   c                 >    |t        d      | j                  |      S )Nz:cannot repeat_interleave a meta tensor without output_size)r  r   )repeatsrX  s     r6   meta_repeat_interleave_Tensorr    s%    WXX[))r8   c                 &   | j                   j                  sJ |j                   j                  sJ t        | j                  t	        | j                               |j                  t	        |j                               t
        j                        }|S Nr>   )rQ   r   rG   r  r   r   rB   )realimagr  s      r6   meta_complexr    sp     ::''''::''''+DJJ78+DJJ786>>F
 Mr8   )
fill_valuer  c                d    | j                  || j                         ft        j                        S ry  )r   ry   rI   r   )r   r   r  s      r6   nonzero_staticr    s&     >>4,EJJ>??r8   c                 
   t        j                  t        j                  d        t        j                  | j                         | j                         fd| j                         ft         j                  | j                        S )Nc                       y)NaY  The register_meta function for torch.nonzero() raises unimplemented by default, as a correct data-independent implementation does not exist. This implementation returns a fake value, assuming all elements of the tensor are non-zero. To enable this registration, please set 'torch.fx.experimental._config.meta_nonzero_assume_all_nonzero' to True.r<   r<   r8   r6   rX   znonzero.<locals>.<lambda>  r`   r8   r   rQ   rv   )	rI   _check_not_implementedrK  meta_nonzero_assume_all_nonzeror  r   ry   r   rv   r   s    r6   nonzeror    sf     
  22	S 	txxz"	
DJJLjj{{	 r8   c           
          t        j                  t              d        g }t              D ]  \  ft        j                  j                  t         j
                  t         j                  t         j                  t         j                  fv d        j                  t         j                  t         j                  fv rȉj                         }t        |      t        j                  j                  z    j                  k   fd       t        j                        D ]`  t        j                  j                      j                  z      k(   fd       |j                  |j                  d             b ]|j                         p|j                          |t        j                  t               j                  k   fd       dd lm} t%         |j&                         t               j                  k  r*j                  d        t               j                  k  r*d}d}D ]  |dk(  rd}|dk(  rd	} n d
}|sg }g }t              D ]*  \  	|j                         |j                         , t              D ]*  \  	|j                         |j                         ,  j)                  |       |g g g t              D ]\  \  }	@rj                   j                  |	          )j                   j                  |	          Ht%        j                        ^ fd}
 j+                  z   z         }ddlm}  | j1                         dk(        r|S  |
       }t3        j4                  |      }t%        |      t%        t        t        |                  k7  r~t3        j6                  |j                  |      }t3        j8                  |      }t3        j6                  |t3        j:                  |            }|j=                  |j?                         |      }|S )Nc                       y)Nz#at least one index must be providedr<   r<   r8   r6   rX   z#meta_index_Tensor.<locals>.<lambda>  r`   r8   c                       y)Nz?tensors used as indices must be long, int, byte or bool tensorsr<   r<   r8   r6   rX   z#meta_index_Tensor.<locals>.<lambda>  r`   r8   c                  "    d j                    S )N)too many indices for tensor of dimension r  r   s   r6   rX   z#meta_index_Tensor.<locals>.<lambda>%  s    G		{S r8   c            	      N    dj                    d  dj                    dz    S )NzThe shape of the mask 
 at index z0 does not match the shape of the indexed tensor r)  )r   r   jrU  r   s   r6   rX   z#meta_index_Tensor.<locals>.<lambda>*  s<    "8ZPQs SJJN**U_`ade`e_f!h r8   r   c                  <    dj                    dt                dS )Nr  z (got rl   )r   r   )r   r   s   r6   rX   z#meta_index_Tensor.<locals>.<lambda>5  s!    ;DII;fSQX\NZ[\ r8   r   Fr   Tc                     z   z   }t        | j                               }dgt              z  |t              t        | j                        t              z
   | j	                  ||      S )zI
        This follows restride_src in TensorAdvancedIndexing.cpp
        r   )r   r   r   r   r6  )r   r   r   after_shapebefore_shapereplacement_shapes      r6   _restride_srcz(meta_index_Tensor.<locals>._restride_srcv  so     00;>t{{}%KL#PSQ
 K
L!C

Oc+6F$FG ug..r8   guard_or_false) rI   rZ   r  	enumeraterQ   r   r   r4  r  r   r   r   r   r   r   selecttorch._refs_refsr   r&   r   r   r  r  r   rA   3compute_elementwise_output_logical_to_physical_perm
apply_permr   invert_permr6  r   )r   r   r  r  refsstatehas_contiguous_subspacer{  transposed_indicesry   r  r   r  restrided_selfperm
perm_shaper'  r  r  r   r   r  rU  r  s   ``               @@@@@@@r6   meta_index_Tensorr    s   	LLg MN &(Fg& !5LL

EIIuzz5::NNY {{uzz5::66--/K""

Ndii/S uzz* 8A&&A$**QU*;;h
 MM'..A"678 e$MM% /!0 G	LLG		!\
 (4(('23G
g,
"t g,
" E# 'A: aZ} ' #'
 #!'* 	1HAu A"))%0	1 "'* 	1HAu}A"))%0	1 ||D!$ !LK#%( 2
U= ""4::c?3##DJJsO4 $U[[ 12	/ ..(99KG
HCDdjjla'(

 #4(NDD^TD DzT%D	*++%%cii6
66zB
%%j%2C2CD2IJ
nnSXXZ4Jr8   c                     d }d }d }|
d   r| j                  |j                               }|
d   r| j                  |j                               }|
d   r| j                  |      }|||fS )Nr   r   r   r   r   )grad_output_input_weight_bias_sizes_optr   r  r  
transposedr  r  output_maskbackend_grad_inputbackend_grad_weightbackend_grad_biass                 r6   meta_convolution_backwardr    sy      1~)33FKKMB1~*44W\\^D1~(22>B 35FGGr8   c                   j                  d      }j                  d      }| j                  ||f      } t        j                  j	                         dk(  d        t        j                  j	                         dk(  d        t        j                  j                  d      j                  d      k(  fd       t        j                  j                  d      j                  d      k(  fd       t        j                  | j                  d      |k(  xr | j                  d      |k(  d	        | j                  | j                               S )
Nr   r   r.   c                       yr=  r<   r<   r8   r6   rX   zmeta_addbmm.<locals>.<lambda>  r`   r8   c                       yr?  r<   r<   r8   r6   rX   zmeta_addbmm.<locals>.<lambda>  r`   r8   r   c                  P    d j                  d       dj                  d       S )Nz8batch1 and batch2 must have same number of batches, got r   r   r   rA  rB  s   r6   rX   zmeta_addbmm.<locals>.<lambda>  s.    J6;;WX>JZZ_`f`k`klm`n_op r8   c            
          d j                  d       d j                  d       dj                  d       dj                  d       d	S )Nz#Incompatible matrix sizes for bmm (r   rF   r   r   rl   r   r  s   r6   rX   zmeta_addbmm.<locals>.<lambda>  sQ    1&++a.1A6;;q>BR S;;q>"!FKKN#316 r8   c                       y)Nz.self tensor does not match matmul output shaper<   r<   r8   r6   rX   zmeta_addbmm.<locals>.<lambda>  r`   r8   )r   rs  rI   rZ   ry   r   )r   rA  rB  rA  r@  rM  rN  s    ``    r6   meta_addbmmr    s    ;;q>D;;q>D;;d|$D	LL"$HI	LL"$HI	LLA&++a.(p 
LLA&++a.(	
 
LL		!51!5@ >>$))+&&r8   c                 @    | j                  | j                               S r3   r  )r   r  kwargss      r6   meta_randint_liker    s    >>$))+&&r8   )
grad_scale	found_infc       	         n    | |||||fD ])  t        j                  t        t              fd       + y )Nc                       dt                S Nz'exponent must be a tensor list but got rm   ls   r6   rX   z#meta__fused_adam_.<locals>.<lambda>      =d1gYG r8   rI   rZ   rc   r   )r   gradsexp_avgsexp_avg_sqsmax_exp_avg_sqsstate_stepslrbeta1beta2weight_decayepsamsgradmaximizer  r  r  s                  @r6   meta__fused_adam_r    s:    & E8[/;O 
q$G	

r8   c       	             | |||||fD ])  t        j                  t        t              fd       + d } ||        ||       ||       ||       ||      fS )Nc                       dt                S r  r  r  s   r6   rX   z"meta__fused_adam.<locals>.<lambda>  r  r8   c                 R    | D cg c]  }t        j                  |       c}S c c}w r3   r\  )tensor_listr  s     r6   empty_like_listz)meta__fused_adam.<locals>.empty_like_list  s!    -89  #999s   $r  )r   r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  s                   @r6   meta__fused_adamr    ss    & E8[/;O 
q$G	

: 	!$( r8   c                 j    t        j                   j                         dk(  d        t        j                  j                         dk(  d        t        j                   j                  t         j                  u  fd       t        j                  j                  t         j                  u fd       t        j                   j                  d      j                  d      k(   fd        j                   j                  d      j                  d      ft         j                  	      S )
Nr   c                       y)Nza must be a 2D tensorr<   r<   r8   r6   rX   zmeta__int_mm.<locals>.<lambda>  r`   r8   c                       y)Nzb must be a 2D tensorr<   r<   r8   r6   rX   zmeta__int_mm.<locals>.<lambda>  r`   r8   c                  "    d j                    S )Nzexpected self to be int8, got r   )rv  s   r6   rX   zmeta__int_mm.<locals>.<lambda>      0	: r8   c                  "    d j                    S )Nzexpected mat2 to be int8, got r   )rw  s   r6   rX   zmeta__int_mm.<locals>.<lambda>  r  r8   r   r   c            
          d j                  d       d j                  d       dj                  d       dj                  d       d	S )Nz'Incompatible matrix sizes for _int_mm (r   rF   r   r   rl   r   rv  rw  s   r6   rX   zmeta__int_mm.<locals>.<lambda>"  sM    5affQi[!&&) M66!9+Qqvvayk, r8   r   )rI   rZ   ry   rQ   r4  r   r   r5  r  s   ``r6   meta__int_mmr    s     
LLA>?	LLA>?	LL	5::: 
LL	5::: 
LL	q	QVVAY	
 ;;q	166!9-U[[;AAr8   c                 f    t        j                   j                         dk(  d        t        j                   j                  t         j                  u  fd        j                  d      } j                  d      dz  } j                  |dz  ||dz  z  d|dz  ft         j                  	      S )
Nr   c                       yNzw must be a 2D tensorr<   r<   r8   r6   rX   z2meta__convert_weight_to_int4pack.<locals>.<lambda>,  r`   r8   c                  "    d j                    S Nr  r   r  s   r6   rX   z2meta__convert_weight_to_int4pack.<locals>.<lambda>/      .qwwi8 r8   r   r      rM      r   )rI   rZ   ry   rQ   r  r   r   r5  r  inner_k_tilesr   rU  s   `   r6    meta__convert_weight_to_int4packr  *  s    	LLA>?	LL	5;;8 	
q	A	q	AA;;F-"$%Q		
 kk   r8   c                 J    t        j                   j                         dk(  d        t        j                   j                  t         j                  u  fd        j                  d      } j                  d      } j                  ||dz  ft         j                        S )Nr   c                       yr  r<   r<   r8   r6   rX   z:meta__convert_weight_to_int4pack_for_cpu.<locals>.<lambda>@  r`   r8   c                  "    d j                    S Nzexpected w to be int32, got r   r  s   r6   rX   z:meta__convert_weight_to_int4pack_for_cpu.<locals>.<lambda>C  r   r8   r   r   r   )rI   rZ   ry   rQ   r5  r   r   r  r  s   `   r6   (meta__convert_weight_to_int4pack_for_cpur
  >  s    	LLA>?	LL	5;;8 	
q	A	q	A;;	
AFkk   r8   c                 .    t        j                   j                         dk(  d        t        j                  j                         dk(  d        t        j                   j                  t         j                  t         j
                  t         j                  fv  fd       t        j                  j                  t         j                  u fd        j                   j                  d      j                  d      dz   j                  	      S )
Nr   c                       yNzx must be a 2D tensorr<   r<   r8   r6   rX   z*meta__weight_int4pack_mm.<locals>.<lambda>O  r`   r8   r  c                       y)Nzw must be a 4D tensorr<   r<   r8   r6   rX   z*meta__weight_int4pack_mm.<locals>.<lambda>P  r`   r8   c                  "    d j                    S Nr  r   rF   s   r6   rX   z*meta__weight_int4pack_mm.<locals>.<lambda>S      5aggY? r8   c                  "    d j                    S r	  r   r  s   r6   rX   z*meta__weight_int4pack_mm.<locals>.<lambda>W  r   r8   r   r  r   
rI   rZ   ry   rQ   rN  rO  rP  r5  r   r   r  s   ``  r6   meta__weight_int4pack_mmr  M  s    	LLA>?	LLA>?	LL	EMM5==%..AA? 
LL	5;;8 ;;qvvay!&&)a-qww;??r8   c                 (    t        j                   j                         dk(  d        t        j                  j                         dk(  d        t        j                   j                  t         j                  t         j
                  t         j                  fv  fd       t        j                  j                  t         j                  u fd        j                   j                  d      j                  d       j                        S )Nr   c                       yr  r<   r<   r8   r6   rX   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>^  r`   r8   c                       yr  r<   r<   r8   r6   rX   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>_  r`   r8   c                  "    d j                    S r  r   r  s   r6   rX   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>b  r  r8   c                  "    d j                    S r  r   r  s   r6   rX   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>f  r   r8   r   r   )
rI   rZ   ry   rQ   rN  rO  rP  r  r   r   r  s   ``  r6    meta__weight_int4pack_mm_for_cpur  \      	LLA>?	LLA>?	LL	EMM5==%..AA? 
LL	5;;8 ;;qvvay!&&)177;;;r8   c                 (    t        j                   j                         dk(  d        t        j                  j                         dk(  d        t        j                   j                  t         j                  t         j
                  t         j                  fv  fd       t        j                  j                  t         j                  u fd        j                   j                  d      j                  d       j                        S )Nr   c                       yr  r<   r<   r8   r6   rX   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>m  r`   r8   c                       yr  r<   r<   r8   r6   rX   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>n  r`   r8   c                  "    d j                    S r  r   r  s   r6   rX   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>q  r  r8   c                  "    d j                    S r	  r   r  s   r6   rX   z;_weight_int4pack_mm_with_scales_and_zeros.<locals>.<lambda>u  r   r8   r   r   r  )rF   r  r  qScaleqZeross   ``   r6   )_weight_int4pack_mm_with_scales_and_zerosr$  k  r  r8   rv  rw  c                     | |z   dz
  |z  |z  S r"  r<   r  s     r6   kai_roundupr&  z  s    UQY1!!r8   c                   	
 | dk(  ry||k(  r(d}d}d}dddd fdfd} ||||||      S |dz  d	k(  rC||z  d	k(  r:d}d}d}dddd		fd
}	
fdd 
	fd	fd |||||||      S y y y )Nr  r  rM  r   c                 8    t        ||z  d      }t        | |      S )Nr  r&  )rU  krsrkr_sr_roundedup4s       r6   kai_k_roundedupz3get_kai_packed_weight_size.<locals>.kai_k_roundedup  s#     $/rBw#: "1&677r8   c                 X     | ||      }|dz  dk(  sJ d       ||dz  z   z   z   z  S )Nr   r   zk_internal must be evenr<   )	rU  nrr*  r+  
k_internalr-  kai_num_bytes_biaskai_num_bytes_multiplier_rhskai_num_bytes_sum_rhss	        r6   9kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0z]get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0  sY     -QB7
"Q1,G.GG,1_23+, )) r8   c                 >    t        | |      |z  }| ||||      z  S r3   r)  )r   rU  r/  r*  r+  num_rowsr4  s         r6   7kai_get_rhs_packed_size_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0z[get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_size_rhs_pack_nxk_qsi4cxp_qsu4cxs1s0  s6     'q"-3 O2r2r8   r  r   c                 |    ||z  dk(  sJ |	z  dk(  sJ |z  dk(  sJ t        | |      |z  }| |||||      z  S rc  r)  )
r   rU  r/  r*  r+  blr6  kai_bl_multiple_of;kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0kai_nr_multiple_ofs
          r6   9kai_get_rhs_packed_size_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0z]get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_size_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0  sp     RA~%~//A555//A555&q"-3 Q2r2rr8   c                     ||z  dk(  sJ |
z  dk(  sJ |z  dk(  sJ  	       } | |      } ||      }|||z  z   z   z  S rc  r<   )rU  r/  r*  r+  r9  num_bytes_multiplier_rhsnum_blocks_per_rownum_bytes_per_blockr:  #kai_get_bf16_datatype_size_in_bytesr<  kai_num_blocks_per_rowr1  kai_num_bytes_per_blockr3  s           r6   r;  z_get_kai_packed_weight_size.<locals>.kai_get_rhs_packed_stride_rhs_pack_nxk_qsi4c32p_qsu4c32s1s0  s     RA~%~//A555//A555 ,O+P(%;Ar%B"&=0'# (+==+,() r8   c                       y)Nr   r<   r<   r8   r6   rB  zGget_kai_packed_weight_size.<locals>.kai_get_bf16_datatype_size_in_bytes  s    r8   c                 6    |z  dk(  sJ t        | |      |z  S rc  r)  )rU  r9  r:  s     r6   rC  z:get_kai_packed_weight_size.<locals>.kai_num_blocks_per_row  s)    //A555"1b)R//r8   c                 (    | z  dk(  sJ | dz  |z   S )Nr   r   r<   )r9  r?  r:  s     r6   rD  z;get_kai_packed_weight_size.<locals>.kai_num_bytes_per_block  s'    //A555a#;;;r8   r<   )n_bitsru  K	groupsizekai_nrkai_krkai_srr7  r=  r:  rB  r;  r4  r-  r<  rC  r1  r2  rD  r3  s            @@@@@@@@@@@r6   get_kai_packed_weight_sizerN  ~  s    {>FFF$%!+,(!"8
 K1fff  ^q Q]a%7FFF$%!!"!"!#  ,0< M1fffi u &8 [ r8   c                 V    t        j                   j                  t         j                  u  fd       t         j                  j
                  j                         r||k(  r|j                  t         j                  k(  s2||k  re|dz  dk(  r]||z  dk(  rU|j                  t         j                  k(  r8t        d|||      } j                  t        |      t         j                        S  j                         |j                         z   } j                  |t         j                        S )Nc                  "    d j                    S r  r   )weightss   r6   rX   z2meta__dyn_quant_pack_4bit_weight.<locals>.<lambda>  s    .w}}o> r8   r  r   r  r   )rI   rZ   rQ   r  backendskleidiaiis_availablerM   rP  rN  r   r   r   )rQ  scales_zerosr/  
block_sizein_featuresout_featurespacked_weight_sizes   `      r6    meta__dyn_quant_pack_4bit_weightrZ    s     
LL$> ~~++-	{	"|'9'9U[['H$R1$j(A-""enn4 8|[*
   %7!8 LL <+=+=+??/u{{CCr8   c                     t        j                   j                         dk(  d        t        j                   j                  t         j                  fv  fd        j                  d      } j                  || j                        S )Nr   c                       y)Nzinput must be a 2D tensorr<   r<   r8   r6   rX   z-meta__dyn_quant_matmul_4bit.<locals>.<lambda>  r`   r8   c                  "    d j                    S )Nzexpected input to be f32, got r   )inps   r6   rX   z-meta__dyn_quant_matmul_4bit.<locals>.<lambda>  s    0< r8   r   r   )rI   rZ   ry   rQ   rN  r   r   )r^  packed_weightsrV  rW  rX  r  s   `     r6   meta__dyn_quant_matmul_4bitr`    sg     
LLa!DE	LL		emm_$< 	A==L		=::r8   c                 (    t        j                   j                         dk(  d        t        j                   j                  t         j                  t         j
                  t         j                  fv  fd       t        j                  j                         dk(  d        t        j                  j                  t         j                  u fd        j                   j                  d      j                  d       j                        S )Nr   c                       yr  r<   r<   r8   r6   rX   z*meta__weight_int8pack_mm.<locals>.<lambda>  r`   r8   c                  "    d j                    S r  r   r  s   r6   rX   z*meta__weight_int8pack_mm.<locals>.<lambda>  r  r8   c                       yr  r<   r<   r8   r6   rX   z*meta__weight_int8pack_mm.<locals>.<lambda>  r`   r8   c                  "    d j                    S )Nzexpected w to be int8, got r   r  s   r6   rX   z*meta__weight_int8pack_mm.<locals>.<lambda>  s    -aggY7 r8   r   r   )
rI   rZ   ry   rQ   rN  rO  rP  r4  r   r   )rF   r  q_scaless   `` r6   meta__weight_int8pack_mmrg    s    	LLA>?	LL	EMM5==%..AA? 
LLA>?	LL	5::7 ;;qvvay!&&)177;;;r8   c                 f    t        j                   j                         dk\   fd       t        j                  j                         dk\  fd       t        j                   j                  d      j                  d      k(   fd       t        j                  t	        j
                   j                        d        t        j                  t	        j
                  j                        d        t        j                  |dk\  d	        t        j                  d
v fd        j                  d      }j                  d      } j                  d d }j                  d d }t        t        j                  ||            }|j                  ||g        j                  |      S )Nr   c                  ,    d j                          dS )Nz1cdist only supports at least 2D tensors, X1 got: r  r   )x1s   r6   rX   z$meta_cdist_forward.<locals>.<lambda>(      CBFFH:QO r8   c                  ,    d j                          dS )Nz1cdist only supports at least 2D tensors, X2 got: r  r   )x2s   r6   rX   z$meta_cdist_forward.<locals>.<lambda>,  rk  r8   r   c                  P    d j                  d       dj                  d       S )Nz4X1 and X2 must have the same number of columns. X1: r   z X2: r   )rj  rm  s   r6   rX   z$meta_cdist_forward.<locals>.<lambda>0  s,    Frwwr{mSXY[Y`Y`acYdXef r8   c                       y)Nz=cdist only supports floating-point dtypes, X1 got: {x1.dtype}r<   r<   r8   r6   rX   z$meta_cdist_forward.<locals>.<lambda>4  r`   r8   c                       y)Nz=cdist only supports floating-point dtypes, X2 got: {x2.dtype}r<   r<   r8   r6   rX   z$meta_cdist_forward.<locals>.<lambda>8  r`   r8   r   c                       y)Nz)cdist only supports non-negative p valuesr<   r<   r8   r6   rX   z$meta_cdist_forward.<locals>.<lambda>:  r`   r8   Nr   r   c                      d  S )Nz%possible modes: None, 1, 2, but was: r<   )compute_modes   r6   rX   z$meta_cdist_forward.<locals>.<lambda>=  s    7~F r8   r  )rI   rZ   ry   r   rA   is_float_dtyperQ   r   r   broadcast_shapesextendr   )	rj  rm  r/  rt  r1r2batch_tensor1batch_tensor2rW  s	   `` `     r6   meta_cdist_forwardr|  $  sJ   	LL
AO 
LL
AO 
LL
rwwr{"f 
LLRXX&O 
LLRXX&O 
LLaLM	LL$F 
B	BHHSbMMHHSbMM..}mLMLR!<<%%r8   c                 4   |j                   d   }|j                   d   }|j                   d   }|j                   d d }|j                   d d }	t        t        j                  ||	            }
|
j	                         }|j                  ||g       t        j                  |
      }|dk(  s|dk(  s
|dk(  s|dk(  rt        j                  |      S |t        |j                         k7  r|j                  |      }t        j                  |t        j                        S )Nr   r  r   r   )r   r   rI   rv  copyrw  mathprod
zeros_likers  r   r   )r3  rj  rm  r/  cdistc1rx  ry  rz  r{  ro  tensor1_expand_sizebatch_products                r6   meta_cdist_backwardr  H  s     
"B	"B	"BHHSbMMHHSbMM 6 6}m TU.335Bx(II23M	Qw"'R1W(:##d288n,YY*+Be.E.EFFr8   c	                     t        j                  j                  t         j                  t         j                  fv fd       t        j                  j                  t         j                  t         j                  fv fd       t        j                  t        j                   j                         fd       j                  d      }	|rt        j                  |	dk\  d        |	dz  }	 j                  |	 j                  d            }
}t        j                  |t        k(  d        t        j                  j                  dk(  fd       t        j                  j                         j                         k(  fd	       fd
d fd}t              dk7  r|j                  j                  d            }j                  j                               }|t        k(  r"j                  |	 j                  d            }nj                  d      }n | |
|      }|t        t        fv s|s!j                  j                  d            }nj                  d      }j                  |	      }j                  d   }|t        k(  rA|rt        j                  |dk\  d        |dz  }j                  | j                  d         }nj                  |j                               }|
|||fS )Nc                  "    d j                    S )Nz(expected indices to be long or int, got r   )r   s   r6   rX   z$meta_embedding_bag.<locals>.<lambda>m      :7==/J r8   c                  "    d j                    S )Nz(expected offsets to be long or int, got r   )rg  s   r6   rX   z$meta_embedding_bag.<locals>.<lambda>q  r  r8   c                  "    d j                    S )Nz/expected weight to be floating point type, got r   )r-  s   r6   rX   z$meta_embedding_bag.<locals>.<lambda>u  s    A&,,P r8   r   r   c                       yNz1include_last_offset: numBags should be at least 1r<   r<   r8   r6   rX   z$meta_embedding_bag.<locals>.<lambda>|  r`   r8   c                       y)Nz@embedding_bag: per_sample_weights only supported with mode='sum'r<   r<   r8   r6   rX   z$meta_embedding_bag.<locals>.<lambda>  r`   r8   c                  $    d j                    dS )Nz1expected per_sample_weights to be 1D tensor, got r  r  )per_sample_weightss   r6   rX   z$meta_embedding_bag.<locals>.<lambda>  s    GHZH_H_G``ab r8   c                  N    dj                          d j                          dS )Nz%expected per_sample_weights.numel() (z$ to be the same as indices.numel() (rl   r   )r   r  s   r6   rX   z$meta_embedding_bag.<locals>.<lambda>  s4    78J8P8P8R7S T66=mmo5FaI r8   c                 D     | ||      xr |j                  d      dk(  S Nr   r   r   )r  re  r   padding_idxis_fast_path_index_selects       r6   is_fast_path_index_select_scalez;meta_embedding_bag.<locals>.is_fast_path_index_select_scale  s(    %c6;?XELLQROWXDX	
r8   c                     | j                   t        j                  k(  xs | j                   t        j                  k(  xr1 | j	                  d      dk(  xr |j	                  d      dk(  xr |dk  S Nr   r   )rQ   rI   rM   rK   r   )r  r   r  s      r6   r  z5meta_embedding_bag.<locals>.is_fast_path_index_select  sb    YY%++%@ejj)@  

1" a A%  a		
r8   c                 2    | | |||      S  | ||      S r3   r<   )r  re  r   r  r  r  s       r6   is_fast_pathz(meta_embedding_bag.<locals>.is_fast_path  s)    23v{SS,S&+FFr8   cpuc                       yr  r<   r<   r8   r6   rX   z$meta_embedding_bag.<locals>.<lambda>  r`   r8   )rI   rZ   rQ   r   r   rA   ru  r   r   MODE_SUMr   r   r   MODE_MAX	MODE_MEANr   )r-  r   rg  scale_grad_by_freqrL  sparser  include_last_offsetr  num_bagsr   r  
offset2bagbag_sizemax_indicesfast_path_sumnumBagsr  r  s   ```   `          @@r6   meta_embedding_bagr  _  s}    
LL%**eii00J 
LL%**eii00J 
LLV\\*P
 ||AHMG	
 	AhA7F%HV	
 	##q(b	
 	$$&'--/9	



G 7u$&&w||A7
$$W\\^48!++Hfkk!nEK!++A.K$V-?UIx(( **7<<?;J **1-J$$X.--"8"qLO 1!++GV\\!_EK!++HMMO<K:x44r8   c                     t        | ||g| \  }}}}t        |      dk(  r|j                  |j                               }||||fS )Nr  )r  r   r   r   )r-  r   rg  rC   r   r  r  r  s           r6   meta_embedding_bag_forward_onlyr    sX    0B1#'1-FJ+ 7u$$$W\\^4:x44r8   c                     |r|S | j                   j                  s| j                   j                  r| j                   S |rt        j                  S | j                   S r3   )rQ   r   r   rI   r   )r   rQ   promote_int_to_longs      r6   _get_reduction_dtyper    sD    {{$$(>(>{{	zz;;r8   r   c                    t        | |d      }t        j                  | j                  |      }t	        | ||      }| j                  ||      S )NT)r  r   )r  rA   r{  r   r|  r   )r   r{  r~  rQ   r  rW  s         r6   meta_nansumr    sI     (u$OLT2D+E4AL??<|?<<r8   c           	          t        j                  | j                  t        t	        | j                                           }| j                  |      S r3   )rA   r|  r   rY   r   ry   r   )r   rW  s     r6   meta_medianr    s<    77U5-.L ??<((r8   c                    t        |       dk(  rt        j                  d       t        j                  | j                  |f      }t        | ||      }| j                  |      | j                  |t        j                        fS )Nr   zmedian CUDA with indices outputr   )	r   rA   alert_not_deterministicr{  r   r|  r   rI   r   )r   ry   r~  rW  s       r6   meta_median_mode_dimr    sp     5V#%%&GH


u{{SF
3C+E3@L%EJJ7 r8   c                     | S r3   r<   r   s    r6   meta_logical_not_r    rX  r8   c                    t        j                  t        |      | j                         k\  d        t	        |      D ]"  \  t        j                  dk\  fd       $ t        |      | j                         z
  }d|z  t        | j                        z   }t        t        |            D cg c]  }||   ||   z   }}| j                  |      S c c}w )Nc                       y)NzZNumber of dimensions of repeat dims can not be smaller than number of dimensions of tensorr<   r<   r8   r6   rX   zmeta_repeat.<locals>.<lambda>
  r`   r8   r   c                      d d  S )Nz"Repeats cannot be negative, found r  r<   )r   reps   r6   rX   zmeta_repeat.<locals>.<lambda>  s    8ZsK r8   r  )	rI   rZ   r   ry   r  rY   r   r   r   )r   r  num_new_dimensionspadded_sizer   target_sizer  s       ` @r6   meta_repeatr    s    	LLG
"l G$ 
31HK	

 W
2++eDJJ.??K8=c'l8KL1;q>GAJ.LKL>>+&& Ms   1Cc                     | S r3   r<   r   s    r6   
meta_zero_r    rX  r8   c                 z    t        |t        j                        r t        | j                  |j                         | S r3   )rc   rI   r   r\   r   r   r   s     r6   meta_binop_inplacer    s)     %&

EKK8Kr8   c                     d }d }d } ||       r ||      rt        d       ||       r ||      st        d      t        |t        j                        r t	        | j
                  |j
                         | S )a*  
    Some checks for inplace ops.
    Checks for promotion rules for some dtypes.
    int.add/sub_(float) and bool.add/sub_(others) are rejected.
    Promoting in these in-place operations would require reallocating
    and copying over elements, hence not allowed.
    Checks for alpha param.
    c                     t        | t              rt        j                  | j                        S t        | t
              S r3   )rc   r   rA   r.  rQ   r   rf   s    r6   is_integericz.meta_binop_inplace_alpha.<locals>.is_integericB  s.    c:&))#))44c7++r8   c                     t        | t              rt        j                  | j                        S t        | t
              S r3   )rc   r   rA   ru  rQ   r   r  s    r6   
is_floaticz,meta_binop_inplace_alpha.<locals>.is_floaticH  s.    c:&''		22c9--r8   c                     t        | t              rt        j                  | j                        S t        | t
              S r3   )rc   r   rA   is_boolean_dtyperQ   r   r  s    r6   is_booleanicz.meta_binop_inplace_alpha.<locals>.is_booleanicN  s.    c:&))#))44c8,,r8   z]Promotion of int.add/sub_(float) in in-place ops are not possible due to element size change.z_Promotion of book.add/sub_(others) in in-place ops are not possible due to element size change.)r  rc   rI   r   r\   r   )r   r   r@  r  r  r  s         r6   meta_binop_inplace_alphar  0  sz    $,.- Dj/k
 	

 D,u"5m
 	
 %&

EKK8Kr8   c                 :    t        | |t        j                        S r  rG   r   rB   r   r   r@  s      r6   meta_binop_alphar  e  s     e$C$K$K r8   c                 8    t        | t        j                        S r  r  )r   r  s     r6   
meta_roundr  q  s    <DD r8   c                 l    t        j                  t        j                  j                         fd       t        t         j                        r8t        j                  t        j                  j                         fd       y t        j                  t        t               fd       y )Nc                  &      dj                    S )Nz7: Expected input tensor to have an integral dtype. Got r   )r  r   s   r6   rX   z#shift_dtype_check.<locals>.<lambda>{  s    7)RSWS]S]R^_ r8   c                  &      dj                    S )Nz6: Expected shift value to have an integral dtype. Got r   r  r  s   r6   rX   z#shift_dtype_check.<locals>.<lambda>  s    wiUVYV_V_U`a r8   c                        d S )Nz): Expected shift value to be an int. Got r<   r  s   r6   rX   z#shift_dtype_check.<locals>.<lambda>  s    wiHN r8   )rI   rZ   rA   r.  rQ   rc   r   r   )r  r   r  s   ```r6   shift_dtype_checkr  x  sp    	LLtzz*_ #u||$""399-a	

 	sG$N	
r8   c                 T    t        d| |       t        | |t        j                        S )Nrshiftr  r  rG   r   rB   r  s     r6   meta_rshiftsr    )    he,e$C$K$K r8   c                 T    t        d| |       t        | |t        j                        S )Nlshiftr  r  r  s     r6   meta_lshiftsr    r  r8   c                 8    | j                  | j                        S r3   r  r   s    r6   	meta_zeror    s    >>$**%%r8   c                     | S r3   r<   r   r  s     r6   
meta_fill_r    rX  r8   c                 ,    t        j                  |       S r3   r\  r  s     r6   	meta_fillr        D!!r8   c                     | S r3   r<   r   s    r6   
meta_relu_r    rX  r8   c                 :    t        | |t        j                        S r  r  r  s      r6   meta__add_relur    s     e$C$K$K r8   c                 ,    t        j                  |       S r3   r\  r   noiselowerr  r  r   s         r6   meta_rrelu_with_noiser    s    
 D!!r8   c                 V    t        j                  |       t        j                  |      fS r3   r\  r  s         r6    meta_rrelu_with_noise_functionalr    s%     D!5#3#3E#:::r8   c                     | S r3   r<   )r   r  r  r  r   s        r6   meta_rrelu_with_noise_r    s	     Kr8   c                 ,    t        j                  |       S r3   r\  r   r   r   
accumulates       r6   meta_index_putr    r  r8   c                 F    t        | j                  |j                         | S r3   r\   r   )r   rm  values      r6   meta_masked_fill_r    s    DJJ

3Kr8   c                     | j                  | j                               j                  t        j                  |             }|S r   )r   r   r  rA   r   )r   rm  re  masked_scales       r6   meta__masked_scaler    s<    >>$))+.1111$7 2 L r8   c                      t        j                  |j                  t         j                  t         j                  fv d        t        j                   j                  j                  k(   fd        S )Nc                       y)NzMask must be bool or uint8r<   r<   r8   r6   rX   z&meta_masked_scatter_.<locals>.<lambda>  r`   r8   c                  <    d j                    dj                    S )NzEmasked_scatter: expected self and source to have same dtypes but got r   r   )r   rZ  s   r6   rX   z&meta_masked_scatter_.<locals>.<lambda>  s"     **U6<<.: r8   )rI   rZ   rQ   r  r  )r   rm  rZ  s   ` `r6   meta_masked_scatter_r    sU    	LL

uzz5;;//1U 
LL

fll"	:
 Kr8   c                     t        | |      \  } }t        j                  | t        j                        }t	        |||      S r   )r&   rI   r   r   r  )r   rm  rZ  r   s       r6   meta_masked_scatterr
    s;     "$-JD$d%2I2IJFf55r8   c                 $    | j                  |      S r3   ru  )r   rm  r7  s      r6   meta_masked_scatter_backwardr    s    >>%  r8   c                     | S r3   r<   r  s       r6   meta_index_put_r    rX  r8   c                   
 t        j                  | j                         dk(  d        t        j                  |j                         dk(  d        | j                         }|j                         
|d   |d   |d   }
d   }||ft        j                  
d   k(  xr 
d   k(  
fd       |r| j                  t         j
                  k(  xs | j                  t         j                  k(  xr |t         j                  k(  }t        j                  || j                  k(  xs |d        |j                        j                  |      }	n|j                        }	|sUSt        j                  j                         dk(  d	        t        j                  j                         k(  fd
       |	S )Nr.   c                       yr=  r<   r<   r8   r6   rX   z)common_meta_baddbmm_bmm.<locals>.<lambda>  r`   r8   c                       yr?  r<   r<   r8   r6   rX   z)common_meta_baddbmm_bmm.<locals>.<lambda>  r`   r8   r   r   r   c            	      .    d d d d    d d    d	S rD  r<   rF  s   r6   rX   z)common_meta_baddbmm_bmm.<locals>.<lambda>  s5    RSURV
l<?*;2l1o=NbR r8   c                       y)Nzfout_dtype only supported for torch.float32 output with float16/bfloat16 inputs or same as input dtypesr<   r<   r8   r6   rX   z)common_meta_baddbmm_bmm.<locals>.<lambda>  r`   r8   c                       y)Nzself must be a 3D tensorr<   r<   r8   r6   rX   z)common_meta_baddbmm_bmm.<locals>.<lambda>  r`   r8   c                  0    d  dj                          S )Nz*Expected an input tensor shape with shape z but got shape: r   )rX  self_baddbmms   r6   rX   z)common_meta_baddbmm_bmm.<locals>.<lambda>   s!    @M]^j^o^o^q]rs r8   )
rI   rZ   ry   r   rQ   rO  rP  rN  r   r  )rA  rB  is_bmmr  r1  rP  res_rowsres_colssupported_out_dtyper   rG  rH  rI  rX  s      `      @@@@r6   common_meta_baddbmm_bmmr    s   	LL"$HI	LL"$HI;;=L;;=L	aB#AAHAHx*K	LLQ2E,q/5E"E	R
 LLEMM)KV\\U^^-K)5==( 	 	%<)<|	
 !!+.11)< !!+.l.\%%'1,.PQ;.s	

 Mr8   c                     t        | |d      S )NTr  )r   r<  s     r6   meta_bmmr  &  s    "4t44r8   c                      t        | |d|      S )NT)r1  r  )r   r<  r1  s      r6   meta_bmm_dtyper   +  s    "4tyIIr8   c                 h    | |z  }| |z  }|dk7  r"t        |dk        t        |dk        k7  r|dz  }|S r  )r  )rF   yqrP  s       r6   div_rtnr$  0  sB    	QA	AA 	Av4A;$q1u+-	QHr8   c                     t        | |z   |z   ||dz
  z  z
  dz
  |r|dz
  ndz   |      dz   }|r|dz
  |z  | |z   k\  r|dz  }|S r  )r$  )	inputSize
kernelSizer  r  r   r  r  
outputSizes           r6   pooling_output_shape_pad_lrr)  :  s     	 *q.)* 	
 'vzA/ 	
 		  Nf$	E(99!OJr8   c           	          t        j                  |dk7  d        t        j                  dk\  fd       t        j                  dz
  z  dz   dz  k  fd       t        | ||      S )Nr   c                       y)Nzstride should not be zeror<   r<   r8   r6   rX   z&pooling_output_shape.<locals>.<lambda>V  r`   r8   c                      d  S )Nz'pad must be non-negative, but got pad: r<   pads   r6   rX   z&pooling_output_shape.<locals>.<lambda>W  s    %LSE#R r8   r   r   c                      d d d  S )NzApad should be at most half of effective kernel size, but got pad=z, kernel_size=z and dilation=r<   )r  r'  r.  s   r6   rX   z&pooling_output_shape.<locals>.<lambda>Z  s'    OPSu U%,nXJ@ r8   )rI   rZ   r)  )r&  r'  r.  r   r  r  s    `` ` r6   r  r  U  ss    	LL1AB	LLRS	LLa8+a/A55	
 ':sC9 r8   c           	      >   	
  j                         }	t        j                  dkD  xr dkD  d        t        j                  |dkD  xr |dkD  d        t        j                  |dkD  xr |dkD  d         j                  d      dk7  xr  j                  d      dk7  }|t        j                  k(  r5t        j                  |dk(  xr |xr  j                  d      dk7  d	        nWt        j                  |dk(  xr  j                  d      dk7  xr |xs |dk(  xr |xr  j                  d      dk7   fd
       t        j                  dz  k\  xr dz  k\  fd       t        j                  dk\  xr dk\  
	fd       y )Nr   c                       y)NzCkernel size should be greater than zero, but got kH: {kH}, kW: {kW}r<   r<   r8   r6   rX   z$pool2d_shape_check.<locals>.<lambda>z  r`   r8   c                       y)Nz>stride should be greater than zero, but got dH: {dH}, dW: {dW}r<   r<   r8   r6   rX   z$pool2d_shape_check.<locals>.<lambda>~  r`   r8   c                       y)Nz\dilation should be greater than zero, but got dilationH: {dilationH}, dilationW: {dilationW}r<   r<   r8   r6   rX   z$pool2d_shape_check.<locals>.<lambda>  r`   r8   r   r   r  r.   c                       y)NzExpected 4D (batch mode) tensor expected for input with channels_last layout with optional 0 dim batch size for input, but got: {input.size()}r<   r<   r8   r6   rX   z$pool2d_shape_check.<locals>.<lambda>  r`   r8   c                  *    d j                          S )NzYExpected 3D or 4D (batch mode) tensor with optional 0 dim batch size for input, but got: r   r  s   r6   rX   z$pool2d_shape_check.<locals>.<lambda>  s    opupzpzp|o}~ r8   c                       d d d d  S )NzKpad should be smaller than or equal to half of kernel size, but got padW = z	, padH = z, kW = z, kH = r<   )r  r  r  r  s   r6   rX   z$pool2d_shape_check.<locals>.<lambda>  s&     ygbT> r8   c                  .    d d  d d d d dS NzGiven input size: (rF   z). Calculated output size: (z). Output size is too smallr<   )r   r!  r  r&  r  r  s   r6   rX   z$pool2d_shape_check.<locals>.<lambda>  s:    %k]!K=* N$$0><.+ O## r8   )ry   rI   rZ   r   r  )r   r  r  r  r  r  r  	dilationH	dilationWr  r   r!  r  r  r   r   
valid_dimsr&  s   ```  ``  `````   @r6   r  r  d  s   " 99;DL	LL
Q26U 
LL
Q26P 
LLA')a-n
 A!#:

1(:J+++AI;*;A!);Q	
 	QY<5::a=A-<* A	?j?UZZ]a-?~	
 
LL
a4+B!GtO	> 
LLq.\Q.	# 	#r8   rB  r?  r  r  r@  r  r  pTpHpW	dilationTr9  r:  rC  rD  rE  rF  rG  rH  r=  c           
      J   	
  j                   }t        j                  dkD  xr dkD  xr dkD  fd       t        j                  dkD  xr dkD  xr dkD  fd       t        j                  dkD  xr dkD  xr dkD  fd       t        j                  |dv  fd       t        |      D ]:  |dk(  rdk(  rt        j                   j	                        dkD   fd       < |r/t        j                  k\  xr k\  xr k\  fd	       t        j                  d
z  k\  xr d
z  
k\  xr d
z  	k\  	
fd       t        j                  dk\  xr dk\  xr dk\  fd       y )Nr   c                      d d  d S )Nz5kernel size should be greater than zero, but got kT: z, kH: z, kW: r<   )r  r?  r  s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  s#    $fRDrd, r8   c                      d d  d S )Nz0stride should be greater than zero, but got dT: z, dH: z, dW: r<   )r  r@  r  s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  s     >rd&FSURVW r8   c                      d d  d S )Nz9dilation should be greater than zero, but got dilationT: z, dilationH: z, dilationW: r<   )r9  r?  r:  s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  s$    #M)M)V r8   r9  c                  &      dj                    S )Nz/: Expected 4D or 5D tensor for input, but got: r)  )r  r   s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  s    7)J5;;-X r8   r  c                  L      dj                    dj                         dS )NzZ: Expected input's non-batch dimensions to have positive length, but input has a shape of z and non-batch dimension z has length zero!)r   r   )r  r   r   s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  s3    ) --2[[M+EJJqM?:KM r8   c                  .    d d  d d d d dS )Nzinput image (T: r  r  z ) smaller than kernel size (kT:  kH:  kW: rl   r<   )rD  rC  rE  r  r?  r  s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  s9    "5'gYd6( C$$&4uRDbT< r8   r   c                  ,    d d d  d d d S )NzHpad should be smaller than or equal to half of kernel size, but got kT: rH  rG  z padT: z padW: z padH: r<   )r  r?  r  r=  r<  r>  s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  s6    $eB4uRDt72$gbTK r8   r   c                  :    d d d  d d d d d dS r8  r<   )rD  rC  rE  rB  rG  rF  rH  s   r6   rX   z$pool3d_shape_check.<locals>.<lambda>  sI    !'!E7!G9AfX F((/y%'!F8 L'( r8   )r   rI   rZ   r   r   )r   rB  r?  r  r  r@  r  r  r<  r=  r>  r?  r9  r:  rC  rD  rE  rF  rG  rH  r  r=  r   r   s   `````````````````````  @r6   r>  r>    s   0 ::D	LL
Q$26$b1f	
 
LL
Q$26$b1f	
 
LLA9)a-9IM	
 
LLX
 4[ 
19aJJqMA	
	
 RK:GrM:fl 	
 
LL
Q"6a26"q&B,	
 	
 
LL
3v{3w!|	
 	
r8   c                 j   | j                   }t        | |||||||	|
||||||||||||       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       y )Nr  r.   r   r   r   r>  r  )r   r  r   rB  r?  r  r  r@  r  r  r<  r=  r>  r?  r9  r:  rC  rD  rE  rF  rG  rH  r  r   s                           r6   max_pool3d_backward_shape_checkrM    s    2 ::D








+0 ;dQh8;dQh6;dQh8;dQh77D$(G47D$(E27D$(G47D$(F3r8   c                     | j                   }t        | ||||||||	|
|ddd|||||||d       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       t        |||dz
  |       y )Nr   Tr  r.   r   rL  )r   r  rB  r?  r  r  r@  r  r  r<  r=  r>  rC  rD  rE  rF  rG  rH  r  r   s                       r6   rP  rP  <  s    * ::D








			-2 ;dQh8;dQh6;dQh8;dQh7r8   c                    d } |d|      \  }}t        j                  t        |      dv d        t        |      dk(  r||}
}	n |d|      \  }	}
 |d|      \  }} |d|      \  }}| j                  d	      }| j                  d
      }| j                  d      }t	        j
                  |       }|t         j                  k(  r)t        j                  | j                         dk(  d        nR|t         j                  k(  r(t        j                  | j                         dv d        nt        j                  dd        t        ||||	||      }t        ||||
||      }t        | |||	|
||||||||||       |||fS )Nc                      t        j                  t        |      dv  fd       |d   }t        |      dk(  r|n|d   }||fS )Nr   c                      d  dS )Nzmax_pool2d: r  r<   r  s   r6   rX   zEmax_pool2d_checks_and_compute_shape.<locals>.unpack.<locals>.<lambda>~  r  r8   r   r   r  r  s   `   r6   r	  z3max_pool2d_checks_and_compute_shape.<locals>.unpack{  r
  r8   r  r  c                       y)NzOmax_pool2d: stride must either be omitted, a single int, or a tuple of two intsr<   r<   r8   r6   rX   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  r`   r8   r   r   r  r  r  r  r   r  c                       y)NzMnon-empty 4D (batch mode) tensor expected for input with channels_last layoutr<   r<   r8   r6   rX   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  r`   r8   ro  c                       y)Nz9non-empty 3D or 4D (batch mode) tensor expected for inputr<   r<   r8   r6   rX   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  r`   r8   Fc                       y)NzAUnsupported memory format. Supports only ChannelsLast, Contiguousr<   r<   r8   r6   rX   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  r`   r8   )rI   rZ   r   r   rA   r   r  ry   r   r  r  )r   r  r   r  r  r  r	  r  r  r  r  r  r  r9  r:  r  r   r!  r   r  r  s                        r6   r  r  r  s    M;/FB	LLFy a 6{aRB&)B	7+JD$!*h7Iy**R.K**R.KBJ//6M+++IIK1c	
 
%11	1IIK6!O	

 	W	

 (Rr9iXL&z2tRIVK



$ k11r8   c                 |    t        |||||      \  }t        j                  j                   j                  k(   fd       |j                  fd}	 |	         |	|       t        j                        }
t        j                  j                  j                  j                  |
      S )Nc                  <    dj                    d j                    S )NzExpected dtype z  for `gradOutput` but got dtype r   rj  s   r6   rX   z7meta_max_pool2d_with_indices_backward.<locals>.<lambda>  s     /$**-MkN_N_M`a r8   c                 l    t        | dz
         t        | dz
         t        | dz
         y )Nr.   r   r   )r  )r  r&  r   r  r  s    r6   _check_dim_sizez>meta_max_pool2d_with_indices_backward.<locals>._check_dim_size  s9    q$q,7q$q,7q$q+6r8   rn  )
r  rI   rZ   rQ   r   rA   r   r~   r   rv   )r  r   r  r   r  r  r  r   r  rY  r   r&  r   r  r  s   ``         @@@@r6   %meta_max_pool2d_with_indices_backwardrZ    s     	,k67Hi		
 
LL

k'''a
 L99D7
 K G//5M;;

jj{{#	 r8   c                    t        | |||||      \  }}}| j                         dk(  r| j                  d      nd}	t        j                  |       }
| j                         dk(  r|||g}n|	|||g}t        j                  || j                  | j                  |
      t        j                  |t
        j                  | j                  |
      fS r  )
r  ry   r   rA   r   rI   r~   rQ   rv   r   r  s               r6   meta_max_pool2d_with_indicesr\    s     	,{FGXy		
  %yy{a/UZZ^QF//6Myy{a\;7\;?++<<'		
 	++<<'		
 r8   c           	         
 t        j                   j                  dv  fd        j                  }t        |dz
  |      D ]?  
t        j                   j	                  
      dkD  d j	                          d
 d       A t        j                  t              dk(  d	        t        j                  t        |      dk(  d
         j	                  d      } j	                  d       j	                  d      |dk(  r j	                  d      }nd}t        j                   j                  j                  k(  d        t        j                  j                  dk(  fd       j	                  d      }j	                  d      }j	                  d      
t        j                  ||k\  d       t        j                  ||k(  d        t        j                  
dk(  
fd       t        j                  |d   d   z   dz
  k  fd       t        j                  |d   d   z   dz
  k  fd        j                         dk(  r|||d   |d   g}	n||d   |d   g}	t        j                  |	 j                   j                        t        j                  |	t         j                   j                        fS )Nro  c                  "    d j                    S )Nz:fractional_max_pool2d: Expected 3D or 4D tensor, but got: r  r   s   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>  s    LTYYKX r8   r.   r   z_fractional_max_pool2d: Expected input to have non-zero  size for non-batch dimensions, but got r_  z emptyr   c                       y)NzNfractional_max_pool2d: kernel_size musteither be a single int or tuple of Intsr<   r<   r8   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>#  r`   r8   c                       y)NzOfractional_max_pool2d: output_size must either be a single int or tuple of Intsr<   r<   r8   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>(  r`   r8   r  r  r   r  r   c                       y)Nz6Expect _random_samples to have the same dtype as inputr<   r<   r8   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>6  r`   r8   c                  "    d j                    S )Nz1Expect _random samples to have 3 dimensions got, r  )random_sampless   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>:  s    CNDWDWCXY r8   z=Expect _random_samples.size(0) no less then input batch size.c                       y)Nz<Expect _random_samples.size(1) equals to input channel size.r<   r<   r8   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>F  r`   r8   c                      d  dS )Nz/Expect _random_samples.size(2) equals to 2 got .r<   )r   s   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>H  s    #RSTRUUV!W r8   c                      dd    d  S )Nz%fractional_max_pool2d: kernel height r   z' is too large relative to input height r<   )input_heightr  s   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>L  s    7A7GGno{n|} r8   c                      dd    d  S )Nz$fractional_max_pool2d: kernel width r   z& is too large relative to input width r<   )input_widthr  s   r6   rX   z,meta_fractional_max_pool2d.<locals>.<lambda>P  s    6{1~6FFlmxlyz r8   r  )rI   rZ   r   r   r   r   rQ   ry   r~   rv   r   )r   r  rX  rc  r   input_channelsinput_batchr   cr   r   rh  rj  s   `` `      @@@r6   meta_fractional_max_pool2drn    s   	LL		VX 99D4!8T" 
IIaL177;yy{mCSTUSVV\^	

 
LLKA	2
 
LLKA	2 YYr]N99R=L))B-Kqyiil	LL

n***H 
LLq Y
 	AAAAAA	LL	[G 
LL	^N 
LLaWX	LLAQ'!+|;} 
LLAQ'!+{:z
 xxzQ^[^[^LAA? 	**;;	

 	++;;	
 r8   c                 |   t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }t        j                  | xs t        |      dv d        |s|n|d   }	|s|nt        |      dk(  r|	n|d   }
|s|nt        |      dk(  r|	n|d   }t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }t        j                  | j                  d	v d
        | j                  dk(  r| j	                  d      nd}| j	                  d      }| j	                  d      }| j	                  d      }| j	                  d      }t        ||||	||      }t        ||||
||      }t        ||||||      }t        | |||||	|
|||||||||||||d       | j                  dk(  xr& t        j                  |       t         j                  k(  }| j                  dk(  rK| j                  d      }|j                          xr  |j                  t         j                        }||||f}n|||||f}| j                  |      }| j                  |t         j                        }|r@|j                  t         j                        }|j                  t         j                        }||fS )Nr1  c                       yNzMmax_pool3d: kernel_size must either be a single int, or a tuple of three intsr<   r<   r8   r6   rX   z.meta_max_pool3d_with_indices.<locals>.<lambda>r  r`   r8   r   r   r   c                       yNzQmax_pool3d: stride must either be omitted, a single int, or a tuple of three intsr<   r<   r8   r6   rX   z.meta_max_pool3d_with_indices.<locals>.<lambda>z  r`   r8   c                       yNzImax_pool3d: padding must either be a single int, or a tuple of three intsr<   r<   r8   r6   rX   z.meta_max_pool3d_with_indices.<locals>.<lambda>  r`   r8   c                       yNzJmax_pool3d: dilation must be either a single int, or a tuple of three intsr<   r<   r8   r6   rX   z.meta_max_pool3d_with_indices.<locals>.<lambda>  r`   r8   r9  c                       yr;  r<   r<   r8   r6   rX   z.meta_max_pool3d_with_indices.<locals>.<lambda>  r`   r8   r  r  r  r  r   zmax_pool3d_with_indices()r  r   r   )rI   rZ   r   r   r   r  r>  rA   r   r  r  rR  r   r   r  )r   r  r   r  r  r  r?  r  r  r@  r  r  r<  r=  r>  r?  r9  r:  r  rB  rC  rD  rE  rF  rG  rH  r  input_channels_last_checkr   r   r   s                                  r6   meta_max_pool3d_with_indicesr{  f  sJ    
LLKF"_ 
QB;1$+a.B;1$+a.B	LL
+c&kV+c vayBc&kQ&6F1IBc&kQ&6F1IB	LLG[ 
B7|q gajB7|q gajB	LLH\ I ]a/	Xa[I ]a/	Xa[I	LL

fK
  %zzQUZZ^AFjjnGJJrNEjjnGZZ^F BIyIE"7BB	9MG!&"b"iKF








#+2 	

aXE77>%BXBXX  zzQ$)OOA$6!)7799
'5500 6 
 	
 eWf5	WeWf=	
//)
$Cooiu{{o;Gff5#9#9f:**5+A+A*B<r8   c                    t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }	t        |      dk(  r|n|d   }
t        j                  | xs t        |      dv d        |s|n|d   }|s|	nt        |      dk(  r|n|d   }|s|
nt        |      dk(  r|n|d   }t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }t        j                  t        |      dv d        |d   }t        |      dk(  r|n|d   }t        |      dk(  r|n|d   }t        j                  |j                  d	v d
        |j	                  d      }|j	                  d      }|j	                  d      }|j	                  d      }| j	                  d      }| j	                  d      }| j	                  d      }t        || ||||	|
|||||||||||||||d       |j                  dk(  xr& t        j                  |      t         j                  k(  }|j                  dk(  rD|j                  d      }|j                          xr  |j                  t         j                        }|j                  |j                        }|r |j                  t         j                        }|S )Nr1  c                       yrq  r<   r<   r8   r6   rX   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r`   r8   r   r   r   c                       yrs  r<   r<   r8   r6   rX   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r`   r8   c                       yru  r<   r<   r8   r6   rX   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r`   r8   c                       yrw  r<   r<   r8   r6   rX   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r`   r8   r9  c                       yr;  r<   r<   r8   r6   rX   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r`   r8   r  r  r  r   z"max_pool3d_with_indices_backward()r  r  r   )rI   rZ   r   r   r   rM  rA   r   r  r  rR  r   r   r  )r  r   r  r   r  r  r  r   r?  r  r  r@  r  r  r<  r=  r>  r?  r9  r:  rB  rC  rD  rE  rF  rG  rH  r  rz  r  s                                 r6   %meta_max_pool3d_with_indices_backwardr    s    
LLKF"_ 
QB;1$+a.B;1$+a.B	LL
+c&kV+c vayBc&kQ&6F1IBc&kQ&6F1IB	LLG[ 
B7|q gajB7|q gajB	LLH\ I ]a/	Xa[I ]a/	Xa[I	LL

fK
 jjnGJJrNEjjnGZZ^FR Er"Gb!F#








,/6 	

aXE77>%BXBXX  zzQ$)OOA$6!)7799
'5500 6 
 	 -J]]1G1G]H
r8   gridc                 z    t        j                   j                  j                  k(   fd       t        j                   j                  t         j                  k(  xr j                  t         j                  k(   fd       t        j                   j
                  d   j
                  d   k(   fd       t        j                  j
                  d    j                  dz
  k(   fd       t        d j                        D ],  t        j                   j
                     dkD   fd       . y )	Nc                  <    dj                    d j                    S )NzNgrid_sampler(): expected input and grid to be on same device, but input is on z and grid is on r  r  r   s   r6   rX   z+check_grid_sampler_common.<locals>.<lambda>9  s'    \\N"24;;-A r8   c                  <    dj                    d j                    S )NzTgrid_sampler(): expected input and grid to have torch.strided layout, but input has z and grid has )ru   r  s   r6   rX   z+check_grid_sampler_common.<locals>.<lambda>@  s&    nT[[MC r8   r   c                  <    dj                    d j                    S )NzZgrid_sampler(): expected grid and input to have same batch size, but got input with sizes  and grid with sizes r)  r  s   r6   rX   z+check_grid_sampler_common.<locals>.<lambda>G  s'      %},A$**O r8   r   r   c                  B    dj                   dz
   d j                   S )Nz+grid_sampler(): expected grid to have size r   z, in last dimension, but got grid with sizes )r   r   r  s   r6   rX   z+check_grid_sampler_common.<locals>.<lambda>N  s,    9%**q.9I J226**? r8   c                  *    dj                    d  dS )NzYgrid_sampler(): expected input to have non-empty spatial dimensions, but input has sizes r_  r`  r)  rs  s   r6   rX   z+check_grid_sampler_common.<locals>.<lambda>W  rt  r8   )rI   rZ   rv   ru   r  r   r   r   )r   r  r   s   ``@r6   check_grid_sampler_commonr  6  s    	LL#	
 
LL%F$++*F	
 
LLA$**Q-'	
 
LL

2%**q.(	
 1ejj! 
KKNQ	

r8   c                       e Zd ZdZdZdZy)GridSamplerInterpolationr   r   r   N)rn   
__module____qualname__BILINEARNEARESTBICUBICr<   r8   r6   r  r  ^  s    HGGr8   r  interpolation_modec                     t        j                   j                  dk(  xr  j                  j                  k(   fd       t        j                   j                  dk(  xr |t        j                  j
                  k(   d        y )Nr  c                  <    dj                    d j                    S )Nzdgrid_sampler(): expected 5D input and grid with same number of dimensions, but got input with sizes r  r)  r  s   r6   rX   z'check_grid_sampler_3d.<locals>.<lambda>g  s&    449KK=#DJJ<1 r8   c                       y)Nz<grid_sampler(): bicubic interpolation only supports 4D inputr<   r<   r8   r6   rX   z'check_grid_sampler_3d.<locals>.<lambda>r  r`   r8   )rI   rZ   r   r  r  r   )r   r  r  s   `` r6   check_grid_sampler_3dr  d  sp    	LL

a3EJJ$))3	
 
LLJJ!O M"&>&F&F&L&LL	
 	Or8   c                     |d   }|r&t        j                  |t         j                        }nd }t        j                  |t         j                        }	||	fS Nr   r   )rI   r  r   r   
r  r   r  r  padding_modealign_cornersr  input_requires_gradr  	grad_grids
             r6   grid_sampler_2d_backward_metar  v  sQ     &a.%%e5;R;RS

  U5L5LMI	""r8   c                     t        | |       t        | ||       | j                  d   }| j                  d   }|j                  d   }|j                  d   }|j                  d   }	| j                  |||||	f      S )Nr   r   r   r.   )r  r  r   r   )
r   r  r  r  r  ru  Cout_Dout_Hout_Ws
             r6   grid_sampler_3dr    sv     eT*%'9:AAAAJJqMEJJqMEJJqME??Aq%677r8   r  c                     t        ||       t        |||       |d   }|r&t        j                  |t        j                        }nd }t        j
                  |t        j                        }	||	fS r  )r  r  rI   r  r  r   r  s
             r6   grid_sampler_3d_backwardr    sm     eT*%'9:%a.%%!?!?

 
  U5S5STIy  r8   c                     |j                  dd       }|st        j                  |      }||d<   t        j                  | g|i |S )NrQ   )rP   rA   	get_dtyperI   r~   )r   r  rC   r  rQ   s        r6   fullr    sE    JJw%E
+F7O;;t-d-f--r8   c                 N   |t         j                  k(  rt        j                  |d u d        t        j                  d|| j                  n|||| j
                  n||      }| j                  r>|j                  | j                         | j                         | j                                n/|j                  | j                         | j                         d       |j                  d       |S t        j                  j                  | |||||      }|j!                  d       |S )Nc                       y)Nz9memory format option is only supported by strided tensorsr<   r<   r8   r6   rX   zzeros_like.<locals>.<lambda>  r`   r8   r   r   Tr  )rI   
sparse_coorZ   r~   rQ   rv   	is_sparsesparse_resize_and_clear_r   
sparse_dim	dense_dimry   _coalesced_r+   r   r  fill_)r   rQ   ru   rv   rw   r   r  s          r6   r  r    s     !!!T!O	

 kk %$**5"(.4;;f!
 >>((		T__.0@ ((dhhj!D

//
!
!# " C IIaLJr8   rt   c                    |t        j                         }|t        j                         }|t         j                  }t        j                  | ||||      S r   rI   r{   get_default_devicer  r~   r   rQ   ru   rv   rw   rx   s         r6   	meta_onesr    T     }'')~))+~;;E&J r8   c                    |t        j                         }|t        j                         }|t         j                  }t        j                  | ||||      S r   r  r  s         r6   
meta_zerosr    r  r8   c                 ,    t        j                  |       S r3   rA   clone_preserve_strides)r   r  ry   r   s       r6   meta_select_scatterr        ''--r8   c                 ,    t        j                  |       S r3   r  )r   r  ry   rp   ro   steps         r6   meta_slice_scatterr    r  r8   dim_post_exprwrap_scalarc                 v    |dk  r|sJ d}| }|dz
  }| |k  s| |kD  rJ d|  d| d| d       | dk  r| |z  } | S )Nr   r   zdim z out of bounds (rk   rl   r<   )ry   r  r  r   r  s        r6   r   r   "  sm    {.C
!
Cc	S3YR4u4DSEC5PQ)RR'
Qw}Jr8   c                 J    | j                         dk(  rdS | j                  |   S r  r  )r  ry   s     r6   ensure_nonempty_sizer  .  s!    11.!''#,.r8   c                 :    t         j                         d      }t        j                         d      }t        j                  ||k(  d        t	        |      D ];  k7  s	t        j                  t              t               k   fd       = y )Nr   c                       y)NzDIndex tensor must have the same number of dimensions as input tensorr<   r<   r8   r6   rX   z$gather_shape_check.<locals>.<lambda>8  r`   r8   c                  N    d dj                    dj                    d  z   S )Nz!Size does not match at dimension z expected index  to be no larger than self  apart from dimension r)  )ry   r   r   r   s   r6   rX   z$gather_shape_check.<locals>.<lambda>>  s7    ;A3>Nu{{m\/

|;QRUQVWX r8   )r  ry   rI   rZ   r   r  )r   ry   r   	self_dims
index_dimsr   s   ```  @r6   gather_shape_checkr  3  s    DHHJ"IUYY[!$J	LLZV 9 8LL$UA.2FtQ2OOXr8   c                 p   ddl m} t        || j                               } |j	                         dk(        }|s`t        j                  j                  t
        j                  k(  xs j                  t
        j                  k(  fd       t        | |       | j                  j                        S )Nr   r  c                  "    d j                    S )Nz8gather(): Expected dtype int32/int64 for index, but got r   r   s   r6   rX   zmeta_gather.<locals>.<lambda>L  s    Nu{{m\ r8   )r  r  r   ry   r   rI   rZ   rQ   r   r   r  r   r   )r   ry   r   sparse_gradr  wrapped_dimis_index_emptys     `    r6   meta_gatherr  C  s    D dhhj1K#EKKMQ$67NKK5::%A		)A\	
 	4e4>>%++&&r8   c                     |r6| dk(  ry| dk(  ry| dk(  ry| dk(  ry| d	k(  ry
t        j                  dd        y | dk(  ry| dk(  ryt        j                  dd        y )Nr  
REDUCE_ADDr  REDUCE_MULTIPLYmeanREDUCE_MEANamaxREDUCE_MAXIMUMaminREDUCE_MINIMUMFc                       y)Nz=reduce argument must be either sum, prod, mean, amax or amin.r<   r<   r8   r6   rX   z#get_operator_enum.<locals>.<lambda>a  r`   r8   addmultiplyc                       y)Nz/reduce argument must be either add or multiply.r<   r<   r8   r6   rX   z#get_operator_enum.<locals>.<lambda>i  r`   r8   rR  )reduce_use_new_optionss     r6   get_operator_enumr  S  s{    e$ ##S	
 	e
"$UUVr8   c                 P    ddl m}  ||j                         dk7        rSt        j                  |j
                  t        j                  k(  xs |j
                  t        j                  k(   fd       |1t        j                  |j
                  |j
                  k(   fd       y y )Nr   )r:  c                        dS )Nz((): Expected dtype int32/int64 for indexr<   method_names   r6   rX   z,scatter_gather_dtype_check.<locals>.<lambda>t  s    {m#KL r8   c                        dS )Nz0(): Expected self.dtype to be equal to src.dtyper<   r  s   r6   rX   z,scatter_gather_dtype_check.<locals>.<lambda>z  s    {m#ST r8   )r  r:  r   rI   rZ   rQ   r   r   )r  r   r   src_optr:  s   `    r6   scatter_gather_dtype_checkr  n  sv    CU[[]a'(KK5::%A		)AL	

 JJ'--'T	
 r8   c                     t        | d      S r"  )r  r   s    r6   ensure_nonempty_dimr  ~  s    sA;r8   c                     ddl m}  |j                         dk(        ry t        j                  t         j                               t        j                               k(  d        d}t         j                               }t        |      D ]'  }t        |      }|k(  r|t         |      kD  s%d} n |s1/t        |      D ]!  }t        |      }|t        |      kD  sd} n ft        j                  t         j                               t        j                               k(  d        t        j                  |  fd       y t        j                  |  fd       y )	Nr   r  c                       yNzCIndex tensor must have the same number of dimensions as self tensorr<   r<   r8   r6   rX   z%scatter_shape_check.<locals>.<lambda>  r`   r8   FTc                       yr  r<   r<   r8   r6   rX   z%scatter_shape_check.<locals>.<lambda>  r`   r8   c                  b    dj                    dj                    d  dj                    z   S )NExpected index r  r  z and to be no larger than src r)  )ry   r   r   r  s   r6   rX   z%scatter_shape_check.<locals>.<lambda>  s8    oekk]2Mdjj\Z&se+I'--YZ r8   c                  H    dj                    dj                    d  z   S )Nr  r  r  r)  )ry   r   r   s   r6   rX   z%scatter_shape_check.<locals>.<lambda>  s,    oekk]2Mdjj\Z&se,- r8   )	r  r  r   rI   rZ   r  ry   r   r  )	r   ry   r   r  r  is_wrong_shaper  r   index_d_sizes	   ````     r6   scatter_shape_checkr    sP   Dekkmq()	LLDHHJ'+>uyy{+KKU
 N#DHHJ/I 9 +E158.tQ77!N g1y! 	A/q9L27A>>!%		 
+/B599;/OOY	
 	Z	
 	-	
r8   c                     t        || j                               }t        d| ||       t        | |||       |t	        ||       y y )Nscatter)r   ry   r  r  r  )r   ry   r   r  r  r  r  s          r6   scatter_meta_implr    sE     dhhj1Ky$s;k5#6'?3 r8   c                 V    t        | |||d       | j                  | j                        S Nr  r  r   r   r   ry   r   r  s       r6   meta_scatter_addr    s%    dCU3>>$**%%r8   c                 $    t        | |||d       | S r  r  r  s       r6   meta_scatter_add_r    s    dCU3Kr8   c                     t        |t        j                        r|nd }t        | ||||       | j	                  | j
                        S r3   )rc   rI   r   r  r   r   r   ry   r   src_or_valuer[  r  s         r6   meta_scatterr	    s;     %\5<<@,dCdCV4>>$**%%r8   c                 `    t        |t        j                        r|nd }t        | ||||       | S r3   )rc   rI   r   r  r  s         r6   meta_scatter_r    s-     %\5<<@,dCdCV4Kr8   queryr   r   	dropout_p	is_causalreturn_debug_maskre  c           	      V   | j                  d      }| j                  d      }| j                  d      }	| j                  d      }
|j                  d      }| j                  dd      }t        j                  |      j                  dd      }t        j                  |||	ft        j
                  | j                        }|ra|
dkD  rdnd}t        j                  |	|z        }|dk  rd}n|dk  rd}t        j                  |||	|f| j                  | j                        }n,t        j                  d| j                  | j                        }t        j                  j                  rkt        j                  j                         rMt        j                  d	t        j                  d
      }t        j                  d	t        j                  d
      }nLt        j                  dt        j                  d
      }t        j                  d	t        j                  d
      }||d d |	||||f	S )Nr   r   r   r.   r  @         r<   rs   )r   r  rI   r   r~   rM   rv   r  ceilrQ   versionhipr   rT  r   r  )r  r   r   r  r  r  re  r   	num_headsmax_seqlen_batch_qhead_dimmax_seqlen_batch_kquery_t	attention	logsumexpblocksize_cmax_seqlen_k
debug_maskseedoffsets                       r6   (meta__scaled_dot_product_flash_attentionr#    s    AJ

1IAzz!}H!ooa#G  )33Aq9I	Y 23kk||I %]cyy!3k!AB$L3&L[[$6E++<<

 [[%++ellK
 }}UZZ446{{2UZZ?Ruzz&A{{Aell6BRu||FC 	
 
r8   	res_shape.c                     t         j                        |k(  r9 j                  dd      }t        j                  |      j                  dd      }|S t        g d fdd      }|D cg c]  }||   	 }}t        t        |            D cg c]  }|j                  |       }}t        j                  | j                   j                        j                  |      }|S c c}w c c}w )Nr   r   )r   r   r   r.   c                 *    j                         |    S r3   r  )idxr  s    r6   rX   z,alloc_with_matching_layout.<locals>.<lambda>0  s    %,,.*= r8   Tr   r  )rY   r   r  rI   r   sortedr   r   r   r~   rQ   rv   r   )	r  r$  r  r  	dim_orderr'  permuted_shaper   final_permutes	   `        r6   alloc_with_matching_layoutr,  '  s     U[[Y&//!Q'w'11!Q7 J =t
	 5>>S)C.>>5:3y>5JK+KKkk%++ell

'-
  	 J ?Ks   *C%C*	attn_biascompute_log_sumexpc	           	         | j                  d      }	| j                  d      }
| j                  d      }|j                  d      }|j                  d      }|	|
||f}t        | |      }t        j                  |	|
|dft        j                  | j
                        }t        j                  dt        j                  d      }t        j                  dt        j                  d      }||d d ||||d f	S Nr   r   r   r   r  r<   rs   r   r,  rI   r~   rM   rv   r   )r  r   r   r-  r.  r  r  r  re  r  r  S_QS_KVD_Vr$  r  
logsum_expr!  r"  s                      r6   (meta__scaled_dot_product_cudnn_attentionr6  ;  s     	

1A

1A
**Q-C88A;D
**R.CAsC I
$UI
6C	
AsAkk||J ;;rF;D[[5::f=F 	
 
r8   c           	         | j                  d      }| j                  d      }	| j                  d      }
|j                  d      }|j                  d      }||	|
|f}t        | |      }t        j                  ||	|
ft        j                  | j
                        }t        j                  dt        j                  d      }t        j                  dt        j                  d      }||d d |
|||d f	S r0  r1  )r  r   r   r-  r  r  r  re  r  H_Qr2  r3  r4  r$  r  r5  r!  r"  s                     r6   5meta__scaled_dot_product_fused_attention_overrideabler9  g  s     	

1A
**Q-C
**Q-C88A;D
**R.CCc"I
$UI
6C	
Ckk||J ;;rF;D[[5::f=F 	
 
r8   ra  r  	cum_seq_q	cum_seq_kmax_qmax_kphilox_seedphilox_offsetc                 J   t        j                  |j                  dd            j                  dd      }t        j                  |j                  dd            j                  dd      }t        j                  |j                  dd            j                  dd      }|||fS rr  )rI   r   r  )ra  r  r   r   r   r  r:  r;  r<  r=  r  r  r>  r?  re  grad_qgrad_kgrad_vs                     r6   'meta__scaled_dot_product_flash_backwardrD    s    , eooa34>>q!DFcmmAq12<<QBFeooa34>>q!DF66!!r8   	attn_maskc                     | j                  d      }| j                  d      }| j                  d      }	t        j                  |       }
t        j                  ||	|ft        j                  | j
                        j                  dd      }|
|fS )Nr   r   r   r  )r   rI   r   r~   rM   rv   r  )r  r   r   r  r  rE  re  r   r  r  r  r  s               r6   0meta__scaled_dot_product_flash_attention_for_cpurG    s     AJ

1IA  'I	

 kk|| i1o  	 r8   c
                 n   t        j                  |j                         d|j                  |j                        }
t        j                  |j                         d|j                  |j                        }t        j                  |j                         d|j                  |j                        }|
||fS )Nr   r   r   r.   r  )rI   empty_permutedr   rQ   rv   )ra  r  r   r   r   r  r  r  rE  re  rA  rB  rC  s                r6   9meta__scaled_dot_product_flash_attention_for_cpu_backwardrK    s    & !!

kk||	F !!
iizz	F !!

kk||	F 66!!r8   dropout_maskc                      d } |       \   ||      \  }	}
 ||      \  }}
j                   \  |	j                   \  }
}}
 fd}fd}dk\  s
|k  rdk\  r |       S  |       S )Nc                 l   | j                         dk(  r| j                  d      dfS | j                         dkD  rxd}t        | j                         dz
        D ]  }|| j                  |   z  } | j	                  || j                  d      | j                  d      | j                  d            dfS | d	fS )
Nr.   r   Tr  r   r  r  r   F)ry   r  r   r   viewr   )rF   r   r   s      r6   	ensure_4dzBmeta__scaled_dot_product_attention_math_for_mps.<locals>.ensure_4d  s    557a<;;q>4''UUWq[J1557Q;' )aggaj(
)66*affRj!&&*affRjI4OOe8Or8   c                  X   j                  j                        } 	r| j                        } j                  f      }	raj                         dk(  r|j	                  d      }| |fS t        j                  d d       |j                  dd z   }|j                  |      }| |fS )Nr.   r   r  r   r  )r   r   view_asry   squeezer   rO  )
r   attnr   r   max_seq_lengthnum_headq_q_sizer  
unsqueezeds
      r6   sdpa_vector_fast_mpszMmeta__scaled_dot_product_attention_math_for_mps.<locals>.sdpa_vector_fast_mps  s    ll288$++e$C||Z6>JKyy{a||A Dy U[["-.Aa@yy'Dyr8   c                  r    d} j                  j                        }j                  | f      }||fS )Nr  r  )blocksr   r  r   	head_sizerV  rW  rX  s      r6   sdpa_vector_2pass_mpszNmeta__scaled_dot_product_attention_math_for_mps.<locals>.sdpa_vector_2pass_mps$  s>    ll288$||Z669$UVL  r8   i   i   r)  )r  r   r   rE  r  r  rL  re  rP  k_rD   v_k_sizerZ  r^  r   r]  rU  rV  rW  rX  rY  s   `              @@@@@@@r6   /meta__scaled_dot_product_attention_math_for_mpsrb    s    	 u%NB
cNEBeEB.0hh+J&)#%88 Av~q ! ! 	$FVO$8N$&&#%%r8   c                 @   | j                  dd      } |j                  dd      }|j                  dd      }| j                  d      }| j                  d      }	| j                  d      }
|j                  d      }t        j                  ||	|
|| j                  | j
                        }t        j                  j                  r&t        j                  j                         r	 |r|	nd}n|rt        j                  |	dz        dz  nd}t        j                  ||
|ft        j                  | j
                        }|j                  dd      }t        j                  dt        j                  d	      }t        j                  dt        j                  d	      }||||fS )
Nr   r   r   r  r   r  r  r<   rs   )r  r   rI   r~   rQ   rv   r  r  r   rT  r  r  rM   r   )r  r   r   r-  r.  r  r  re  r  r  r  Kvr  logsumexp_dimr5  r!  r"  s                    r6   ,meta__scaled_dot_product_efficient_attentionrf  0  sD    OOAq!E
--1
COOAq!E

1A

1A

2I	BB
++aIrU\\
RC}}UZZ446	 0Q2D		!b&)B.!	
I}%kk||J --1
C ;;rF;D[[5::f=F
D&((r8   grad_input_maskc                    |j                  d      }|j                  d      }|j                  d      }|j                  d      }|j                  d      }|j                  d      }t        j                  ||||fd|j                  |j                        }t        j                  ||||fd|j                  |j                        }t        j                  ||||fd|j                  |j                        }d }|~|
d   ry|j                  d      }|dz  dk(  r|n
|dz   |dz  z
  }t        |j                               }||d<   t        j                  ||j                  |j                        }|d	d |f   }||||fS )
Nr   r   r   r.   rI  r  r   rM  .)r   rI   rJ  rQ   rv   r   r~   )ra  r  r   r   r-  r   r  r>  r?  r  rg  r  re  r   r  r<  r  
head_dim_vr=  rA  rB  rC  	grad_biaslastDimlastDimAligned	new_sizess                             r6   +meta__scaled_dot_product_efficient_backwardrn  ]  s{   ( AJ

1IJJqMEzz!}HAJHHQKE!!	Yx0kk||	F !!	Yx0iizz	F !!	Yz2kk||	F I!3..$$+bLA$57R<'TV,;V)*	&	"KKY__Y5E5E
	 c8G8m,	669,,r8   c                     t        j                  |      }t        j                  |      }t        j                  |      }|||fS r3   r\  )ra  r  r   r   r   r  r>  r?  r-  r:  r;  r<  r=  r  r  re  rA  rB  rC  s                      r6   'meta__scaled_dot_product_cudnn_backwardrp    sA    . e$Fc"Fe$F66!!r8   window_size_leftwindow_size_right	seqused_kalibi_slopesc                    || j                  d      n|j                         dz
  }|| j                  d      n|}||j                  d      n|}| j                  d      }| j                  d      }t        j                  |       }|4t        j                  |||ft        j
                  | j                        }nC| j                  d      }t        j                  ||ft        j
                  | j                        }|	ra|dkD  rdnd}t        j                  ||z        }|dk  rd}n|dk  rd}t        j                  ||||f| j                  | j                        }n,t        j                  d| j                  | j                        }d	\  }}t        j                  j                  rkt        j                  j                         rMt        j                  d
t        j                  d      }t        j                  d
t        j                  d      }nLt        j                  dt        j                  d      }t        j                  d
t        j                  d      }|||||fS )Nr   r   r  r   r  r  r  r  NNr<   rs   r   )r   r   rI   r   r~   rM   rv   r  r  rQ   r  r  r   rT  r   r  )r  r   r   r:  r;  r<  r=  r  r  r  re  rq  rr  rs  rt  r   r  r  r  r  r  r  total_qr  r  r   r!  r"  s                               r6   meta__flash_attention_forwardrx    s   4 #,"3A9JQ9NJ*3*;A(1(9!u

2Izz"~H   'IKK$67++<<
	 **Q-KK ELL
	 %]cyy!3k!AB$L3&L[[$6E++<<

 [[%++ellK
 LD&}}UZZ446{{2UZZ?Ruzz&A{{Aell6BRu||FC r8   c                     t        j                  |      }t        j                  |      }t        j                  |      }|||fS r3   r\  )ra  r  r   r   r   r  r:  r;  r<  r=  r  r  r>  r?  re  rq  rr  
grad_querygrad_key
grad_values                       r6   meta__flash_attention_backwardr}    sA    0 !!%(J$H!!%(Jx++r8   cu_seqlens_qcu_seqlens_kmax_seqlen_qr  custom_mask_typecausal_diagonalseqlen_kwindow_sizec                    | j                  d      }| j                  d      }|j                  d      }| j                  d      }|j                  d      }t        j                  ||||| j                  | j                        }||j                  d      dz
  n|}|}||J |}||n|}|
rt        j                  |dz        dz  nd}t        j                  |||ft        j                  | j                        }t        j                  dt        j                  d      }t        j                  dt        j                  d      }||||||fS )	Nr   r   r  r   r  r  r<   rs   )	r   rI   r~   rQ   rv   r  r  rM   r   )r  r   r   r/  r~  r  r  r  r  r  r.  re  r  r  r  r  r  ru  r  rd  r  logsumexp_batch_dimactual_max_seqlen_qactual_max_seqlen_kre  r5  r!  r"  s                               r6   !meta__efficient_attention_forwardr  %  s9   , 	

1A

1AA

2I	BB
++aIrU\\
RC7C7O,++A.2VW'''**6*B,4F		%*+b0A  	i7kk||J ;;rF;D[[5::f=F
D&*=?RRRr8   bias_requires_gradnum_splits_keyshared_storage_dqdkdvc                    |rt        j                  |j                  d   |j                  d   k(  d        t        j                  |j                  d   |j                  d   k(  d        t        j                  g |j                  dd d|j                  d   |j                  d   |j                  |j
                        }|j                  d	d      }|j                  d	d      }|j                  d	d
      }n?t        j                  |      }t        j                  |      }t        j                  |      }|z|j                  d      }|dz  dk(  r|n
|dz   |dz  z
  }t        |j                               }||d<   t        j                  ||j                  |j
                        }|dd |f   }n!t        j                  d|j
                        }||||fS )Nr   c                       y)Nz,seqlen must match for `shared_storage_dqdkdvr<   r<   r8   r6   rX   z4meta__efficient_attention_backward.<locals>.<lambda>u  r`   r8   r.   c                       y)Nz3embedding dim must match for `shared_storage_dqdkdvr<   r<   r8   r6   rX   z4meta__efficient_attention_backward.<locals>.<lambda>y  r`   r8   r   r  r   r  r  r   rM  .r<   r  )
rI   rZ   r   r~   rQ   rv   r  r   r   r   )ra  r  r   r   r/  r~  r  r  r  r  r  r>  r?  r  r  re  r  r  chunkrz  r{  r|  rk  rl  rm  rj  s                             r6   "meta__efficient_attention_backwardr  Y  s   2 KKNciil*B	
 	KKNciil*I	
 Eekk!BEEEKKOEU[[_E++<<

 \\"a(
<<A&\\"a(
%%e,
##C(%%e,
))B-$+bLA$57R<'TV,;V%	&	"KK	DKKP	c8G8m,	KK5<<8	xY66r8   scale_ascale_bscale_resultuse_fast_accumc                 
    d }t        j                   j                         dk(  xr j                         dk(   fd       t        j                   | j                        xr  |j                         fd       t	               dk(  r%d }	d }
d }t        j                   |	 j                               xs  |        fd	       t        j                   |
j                               xs  |      fd
       t        j                   j                  d      dz  dk(   fd       t        j                  j                  d      dz  dk(  xr j                  d      dz  dk(  fd        j                  \  }j                  d      j                  t         j                  k(  xr j                  t         j                  k(  xs< j                  t         j                  k(  xr j                  t         j                  k(  }j                         dk(  rfj                         dk(  rSt        j                  j                  t         j                  k(  xr j                  t         j                  k(  d        n&|rj                  t         j                  k(  rd}|dz  }nd}d}d } |||      } ||d      dz  }| ||      z  |z  | ||      z  |z  j                         k(  r_j                         k(  rLt        j                  j                         d        t        j                  j                         d        nSt        j                  dfd       n5t        j                  j                  t         j                  k(  xr j                  t         j                  k(  d        t        j                  j                         dk(  xr j                         dk(  fd       j                  d      k(  rtj                  d      dk(  r`j                  d      dk(  rLj                  d      k(  r8t        j                  j                         xr j                         d        nt        j                  dfd       ||n j                  }t        j                   j                  d      j                  d      | j                        S )Nc                     | t         j                  t         j                  t         j                  t         j                  t         j
                  fv S r3   )rI   rQ  float8_e5m2float8_e4m3fnuzfloat8_e5m2fnuzfloat4_e2m1fn_x2r   s    r6   is_fp8_or_fp4_typez*meta_scaled_mm.<locals>.is_fp8_or_fp4_type  sA    !!!!""
 
 	
r8   r   c                  L    dj                          d j                          S )Nz%Inputs must be 2D but got self.dim()=z and mat2.dim()=r   r<  r   s   r6   rX   z meta_scaled_mm.<locals>.<lambda>  s'    7
|CSTXT\T\T^S_` r8   c                  <    dj                    d j                    S )Nz?Expected both inputs to be fp8 or fp4 types but got self.dtype=z and mat2.dtype=r   r  s   r6   rX   z meta_scaled_mm.<locals>.<lambda>  s&    QRVR\R\Q]]mnrnxnxmyz r8   r   c                 ,    | d   | d   kD  xr | d   dk(  S r  r<   r  s    r6   is_row_majorz$meta_scaled_mm.<locals>.is_row_major  s"    !9vay(;VAY!^;r8   c                 &    | d   dk(  xr | d   dkD  S r  r<   r  s    r6   is_col_majorz$meta_scaled_mm.<locals>.is_col_major  s    !9>3fQi!m3r8   c                 V    | j                  d      dk(  xs | j                  d      dk(  S r  r   )	tensor_2ds    r6   has_zero_dimz$meta_scaled_mm.<locals>.has_zero_dim  s)    >>!$)CY^^A->!-CCr8   c                  *    d j                          S )Nz#self must be row_major, got stride r  r   s   r6   rX   z meta_scaled_mm.<locals>.<lambda>      9$++-I r8   c                  *    d j                          S )Nz#mat2 must be col_major, got stride r  r<  s   r6   rX   z meta_scaled_mm.<locals>.<lambda>  r  r8   r   rM  r   c                  ,    d j                  d       S )NzBExpected self.size(1) to be divisible by 16, but got self.size(1)=r   r   r   s   r6   rX   z meta_scaled_mm.<locals>.<lambda>  s    XY]YbYbcdYeXfg r8   c                  "    d j                    S )Nz?Expected both dimensions of mat2 to be divisible by 16 but got r)  r  s   r6   rX   z meta_scaled_mm.<locals>.<lambda>  s    UVZV`V`Uab r8   c                       y)NzNFor tensorwise scaling, both scale_a and scale_b must be float (fp32) tensors.r<   r<   r8   r6   rX   z meta_scaled_mm.<locals>.<lambda>  r`   r8   r  r  c                     | |z   dz
  |z  S r"  r<   r  s     r6   ceil_divz meta_scaled_mm.<locals>.ceil_div  s    A	a''r8   r  c                       y)Nzscale_a must be contiguousr<   r<   r8   r6   rX   z meta_scaled_mm.<locals>.<lambda>  r`   r8   c                       y)Nzscale_b must be contiguousr<   r<   r8   r6   rX   z meta_scaled_mm.<locals>.<lambda>  r`   r8   Fc            	      Z    d  dj                          d dj                          d	S )NzTInvalid blockwise scaling configuration. For blockwise scaling, scale_a should have  elements, got z, scale_b should have rf  r  )expected_a_sizeexpected_b_sizer  r  s   r6   rX   z meta_scaled_mm.<locals>.<lambda>  sH    FFUEVVefmfsfsfuev w//>.?w}}N__`b r8   c                       y)NzKFor rowwise scaling, both scale_a and scale_b must be float (fp32) tensors.r<   r<   r8   r6   rX   z meta_scaled_mm.<locals>.<lambda>  r`   r8   c                  L    d j                         dj                         S )NzLFor non-tensorwise scaling, scale tensors must be 2D, but got scale_a.dim()=z and scale_b.dim()=r   r  r  s   r6   rX   z meta_scaled_mm.<locals>.<lambda>  s,    gY`YdYdYfXhh|nunynyn{m}~ r8   c                       y)Nz@Both scale_a and scale_b must be contiguous for rowwise scaling.r<   r<   r8   r6   rX   z meta_scaled_mm.<locals>.<lambda>%  r`   r8   c                      d  d dj                  d       dj                  d       dj                  d       dj                  d       dS )	Nz}Invalid scaling configuration. For tensorwise scaling, both scales should be scalar. For rowwise scaling, scale_a should be (z, 1), scale_b should be (1, z). Got scale_a.size()=(r   rk   r   z) and scale_b.size()=(rl   r   )rV  r   r  r  s   r6   rX   z meta_scaled_mm.<locals>.<lambda>+  sk    CCD#Eabcad e//6||A.?r',,q/AR S//6||A.?r',,q/ARRS	U r8   r  )rI   rZ   ry   rQ   r   r   r   r   float8_e8m0fnurQ  r   rN  rR  r~   rv   )r   r<  r  r  r/  r  r1  r  r  r  r  r  _kis_blockwise_scalingblock_size_kblock_size_mnr  num_k_blockspadded_num_k_blocks
_out_dtyper  r  rV  r   s   ````                @@@@r6   meta_scaled_mmr    s   
 
LL
a+DHHJ!O` 
LL4::&I+=djj+Iz
 4F"	<	4	D 	'=<+=I	
 	'=<+=I	
 	IIaL2"g	
 	IIaL2"=tyy|b'8A'=b	
 

2IIaL MMU111 6!5!55 
 MMU000 5!4!44 	 ==?aGMMOq$8LL.Q7==EMM3Qh " }} 3 33  "!V!M( $B5L"*<";a"? M ::=PP  M ::=PP 
 ?2MMO6))+8 ))+8
  LL.Q7==EMM3Qe
 LL"9w{{}'9~ Q1$LLOq(LLOq(LLOq( ))+G0E0E0G^ 	 (3J;;tyy|TYYq\DKKXXr8   c                 Z    t        | ||||d       | j                  | j                        S NT)r  r   r   ry   r   r  r[  rY  s         r6   meta_scatter_reduce_twor  8  s)     dCVTJ>>$**%%r8   c                 (    t        | ||||d       | S r  r  r  s         r6   meta_scatter_reduce__twor  ?  s    dCVTJKr8   c                t    t        j                  d j                         cxk  xr dk  nc  fd        j                         dk(  r0t        j                  |t         j                   j
                        S t        j                   j                  d      |t         j                   j
                        S )Nr   r   c                  *    d j                          S )NzAThe probability distributions dimensions must be 1 or 2, but got r   r  s   r6   rX   z"meta_multinomial.<locals>.<lambda>J  s    STYT]T]T_S`a r8   r   r  )rI   rZ   ry   r~   r   rv   r   )r   num_samplesreplacementr   s   `   r6   meta_multinomialr  E  s|     
LL	EIIK1a yy{a{{;ejjNN;;

1{%**U\\ r8   c                 "    d}| D ]  }||z  }	 |S r"  r<   )vsrP  vs      r6   multiply_integersr  S  s$    	A 	QHr8   c                 L    t        j                  t              k(  fd       dz   t        j                  t               k(   fd       t        j                  t        d  dd  D              xr t        d D               fd        d d \  }}||gS )Nc                  &    d  dt               S )Nz%It is expected output_size equals to , but got size r  )num_spatial_dimsrX  s   r6   rX   z'upsample_common_check.<locals>.<lambda>]  s    78H7IY\]hYiXjk r8   r   c                  &    d  dt               S )Nz$It is expected input_size equals to r  r  )expected_input_dimsr.  s   r6   rX   z'upsample_common_check.<locals>.<lambda>b  s    67J6K?[^_i[jZkl r8   c              3   &   K   | ]	  }|d kD    ywr   Nr<   re   r  s     r6   rg   z(upsample_common_check.<locals>.<genexpr>f  s     *aAE*   c              3   &   K   | ]	  }|d kD    ywr  r<   r  s     r6   rg   z(upsample_common_check.<locals>.<genexpr>f  s     2NQ1q52Nr  c                      d  d S )NzDInput and output sizes should be greater than 0, but got input size z and output size r<   )r.  rX  s   r6   rX   z'upsample_common_check.<locals>.<lambda>g  s      \!2;-A r8   )rI   rZ   r   r  )r.  rX  r  r  channelsr  s   ```  @r6   upsample_common_checkr  Z  s    	LLK,,k +Q.	LLJ..l
 
LL*:ab>**Ns2N+2N/N	A ""1~FHH+{++r8   c                 4    t        j                   j                         dk7  xs t         j	                         dd         fd       t         j	                         |d      } j                  |      j                  t        j                               S )Nr   r   c                  *    d j                          S )Nz>Non-empty 3D data tensor expected but got a tensor with sizes r   r  s   r6   rX   z$upsample_nearest1d.<locals>.<lambda>u      PQVQ[Q[Q]P^_ r8   r  r   
rI   rZ   r   r  r   r  r   r  rA   r   )r   rX  scalesfull_output_sizes   `   r6   upsample_nearest1dr  o       
LLA/

QR0@A_ -

kA ??+,//11%8 0  r8   c                     t        j                   j                         dk7  xs t         j	                         dd         fd       t         j	                         |d      } j                  |      }t        j                         } j                  \  }}}} j                  j                  dk(  r|dk  rt         j                  }|j                  |      }|S )	Nr   r   c                  *    d j                          S Nz>Non-empty 4D data tensor expected but got a tensor with sizes r   r  s   r6   rX   z$upsample_nearest2d.<locals>.<lambda>  r  r8   r   r  r   r  r   )rI   rZ   r   r  r   r  r   rA   r   r   rv   rm   r   
contiguous)	r   rX  scales_hscales_wr  r   r   rD   
n_channelss	   `        r6   upsample_nearest2dr    s     
LLA/

QR0@A_ -

kA __-.F //6M  ++Az1a||F"zA~//];FMr8   rX  r.  r  r  c                 X    t        ||d      t        j                   j                  dk(   fd       t	        d      D ]2  t        j                   j                           k(   fd       4  j                  |      j                  t        j                               S )Nr   r  r  c                  "    d j                    S NzFExpected grad_output to be a tensor of dimension 4 but got: dimension r  r~  s   r6   rX   z-upsample_nearest2d_backward.<locals>.<lambda>      XYdYiYiXjk r8   c            
      D    d d     d dj                         S )NzCExpected grad_output to have the same shape as output; output.size() = z but got grad_output.size(r   r  r  r   s   r6   rX   z-upsample_nearest2d_backward.<locals>.<lambda>  s>      !s$'7':&;,QCtK4D4DQ4G3HJ r8   r   )
r  rI   rZ   r   r   r   r   r  rA   r   )r  rX  r.  r  r  r  r   s   `    @@r6   upsample_nearest2d_backwardr    s     -K! 
LLAk 1X 
Q#3A#66	

   ,//11+> 0  r8   c                 4    t        j                   j                         dk7  xs t         j	                         dd         fd       t         j	                         |d      } j                  |      j                  t        j                               S )Nr   r   c                  *    d j                          S )Nz>Non-empty 5D data tensor expected but got a tensor with sizes r   r  s   r6   rX   z$upsample_nearest3d.<locals>.<lambda>  r  r8   r.   r  r   r  )r   rX  scales_dr  r  r  s   `     r6   upsample_nearest3dr    r  r8   c                    t        j                  |       t        j                  | t         j                        }}||t        |t              sJ t        |t              sJ |j
                  }|j                         }	t        ||      }t        ||      }|j                  ||	       |j                  ||	       t        ||       t        ||       ||fS ||fS )Nr   )r~  r  )
rI   r   r   rc   r   r   r   r!   r   r#   )
r   stablery   
descendingr   r   r  r   r   
out_strides
             r6   	meta_sortr    s     D!5#3#3D#LqAg1&*---':... GG	XXZ
"695#GY79j1Iz2F3G4wa4Kr8   c           	          t        j                   j                  dk(   fd       t        j                   j                  j                  k(   fd        j	                  d      t        j                  j                  dk(  fd       t        j                  j                         k(  fd       t        j                  j                  j                  k(  fd       t        j                  j                  dk(  fd        j	                  d	      z  z  t        j                  j                         k(   fd
       t        j                  t         fdfD              d        y )Nr   c                  "     j                    dS Nz != 2r  )input_gatess   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>      ;3C3C2DE0J r8   c                  :    j                    d j                    S N != r)  )hidden_gatesr  s   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>  s     ;$$%T,*<*<)=> r8   r   c                  "     j                    dS )Nz != 1r  )
input_biass   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>  s    joo5Fe3L r8   c                  .    j                          d  S r	  r  )
gates_sizer  s   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>  s    z'')*$zl; r8   c                  :    j                    d j                    S r	  r)  )hidden_biasr  s   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>  s     z''([->->,?@ r8   c                  "     j                    dS r  r  )prev_hiddens   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>  r  r8   r   c            
      `    j                          dj                  d       d d d  d
S )Nr
  r   z * z // z (aka rl   )r   r   )expected_prev_hidden_numelfactorr  r  r  s   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>  sB    ;$$&'tK,<,<Q,?+@J<tTZS[[ab|a}}~ r8   c              3   P   K   | ]  }|j                   j                   k(    y wr3   r  )re   rF   r  s     r6   rg   z&rnn_cell_checkSizes.<locals>.<genexpr>  s(      
 HH***
s   #&c                       y)Nz%expected all inputs to be same devicer<   r<   r8   r6   rX   z%rnn_cell_checkSizes.<locals>.<lambda>
  r`   r8   )rI   rZ   r   r   r   r   r  )r  r  r  r  r  r  r  r  s   ``````@@r6   rnn_cell_checkSizesr    s@    
LL!!Q&(JK	LL\///> !!!$JZ__)+LM*,;	
 	 1 11@	
 
LL!!Q&(JK!,!1!1!!4z!AV!K	LL99 
LL 
"J[I
 	
 	8r8   c                 
   t        | |||d|       t        j                  | t        j                        }t        j                  |t        j                        }t        j                  |t        j                        }|||fS )Nr  r   )r  rI   r   r   )r  r  cxr  r  	workspacehycys           r6   _thnn_fused_lstm_cell_metar    sk     \:{ArR  E<S<STI			"E,C,C	DB			"E,C,C	DBIr8   c                 b   t        |      dk7  }|r t        |      }|d   }| j                  d   }nB|
r| j                  d   n| j                  d   }|
r| j                  d   n| j                  d   }d}|rdnd}|dk7  r|n|}|r|||z  g}n|
r||||z  gn||||z  g}| j                  |      }|	|z  ||g}|"t        j                  d| j
                        }n|j                  |      }|j                  |	|z  ||g      }|rdnd}| j                  |t        j                        }|||||fS )Nr   r   r   r   r  r   )r   r   r   rI   r~   rv   r  )r   r-  weight_stride0
weight_bufhxr  rL  hidden_size	proj_size
num_layersbatch_firstdropouttrainbidirectionalbatch_sizesdropout_stateis_input_packed
seq_length
mini_batchbatch_sizes_sumnum_directionsout_sizer   r   
cell_shaper  r  reserve_shapereserves                                r6   
_cudnn_rnnr6    sS   & +&!+O%
 ^
++a.'2U[[^A
'2U[[^A
'QQN%NyH$h&?@	  X%>?j(^*CD 	
 __Y'F~-z;GJ	z[[5<<0\\*%	zN2JI	JB AAMoom5;;o?G2r7J..r8   c                 (   |r| j                   d   n| j                   d   }|r| j                   d   n| j                   d   }|
}|r|||gn|||g}| j                  |      }|"t        j                  d| j                        }n|j                  |j                         }|"t        j                  d| j                        }n|j                  |j                         }t        j                  d| j                  t        j
                        }||||fS )Nr   r   r  r   )r   r   rI   r~   rv   r  )r   w0w1w2w3hx_cx_r   r+  rL  r$  r&  
has_biasesr*  r'  r)  r.  r/  output_chanelsr   r   r  r  r  s                           r6   mkldnn_rnn_layerr@  U  s    & $/QEKKNJ#.QEKKNJ N  
Z0*n5 
 __Y'F
{[[5<<0]]399%
{[[5<<0]]399%Aell%++FI2r9$$r8   c                     | j                   dk(  r%t        j                  dk(  xs dk(  fd       y t        j                  | j                        dk7  fd       y )Nr   r   c                       d  S )Nz4: Expected reduction dim -1 or 0 for scalar but got r<   ry   r  s   r6   rX   z'zero_numel_check_dims.<locals>.<lambda>  s    wiSTWSXY r8   c                       d  dS )Nz: Expected reduction dim z to have non-zero size.r<   rC  s   r6   rX   z'zero_numel_check_dims.<locals>.<lambda>  s    wi8=TU r8   )r   rI   r   r   )r   ry   r  s    ``r6   zero_numel_check_dimsrE  }  sR    yyA~1H!r	Y	

 	IIcNaU	
r8   c                      |(t        ||j                               }t        ||        y t        j                  |j                         dk7   fd       y )Nr   c                        dS )Nz@: Expected reduction dim to be specified for input.numel() == 0.r<   r  s   r6   rX   z%check_argmax_argmin.<locals>.<lambda>  s    tf\] r8   )r   ry   rE  rI   rZ   r   )r  r   ry   s   `  r6   check_argmax_argminrH    sC    
S$((*-dC.JJLA]	
r8   c                     t        d| |       t        j                  | j                  ||fnd       }t	        | ||      }| j                  |t        j                        S )Nargmaxr   )rH  rA   r{  r   r|  r   rI   r   )r   ry   r~  r{  r   s        r6   argmax_argmin_metarK    sQ    $,

coSF4PD$T49E>>%u{{>33r8   c                 |    |t         j                  k(  rt         j                  }t        j                  d||||      S )Nr<   r   )rI   jaggedr  r~   )r  rQ   ru   rv   rw   s        r6   scalar_tensorrN    s5    
 ;;
%v* r8   c                    t        || j                         d      }| j                         dk(  rdn| j                  |      }t        j                  |       t        j
                  ||k  d        t        | j                        }t        |      dkD  r|||<   | j                  |      | j                  |t        j                        fS )NTr  r   r   c                       y)Nzk not in range for dimensionr<   r<   r8   r6   rX   ztopk_meta.<locals>.<lambda>  r`   r8   r   )r   ry   r   rI   r  rZ   r   r   r   r   r   )r   rU  ry   largestr(  	sliceSizetopKSizes          r6   	topk_metarU    s     dhhjd
;CXXZ1_$))C.I		LLi!GHDJJH
8}q>>(#T^^HEKK^%PPPr8   c                     |	|J d       |j                         }| j                         }	t        j                  ||	j                  |	j                  |	j
                        S )Nz;segment_reduce(): Either lengths or offsets must be defined)rQ   rv   ru   )r  rI   r   rQ   rv   ru   )
r3  r   rk  r[  rf  rg  rh  rj  data_contiggrad_contigs
             r6   meta__segment_reduce_backwardrY    sj    
 '"5 E5 //#K//#K!!!!	 r8   c                    ddl m} t        | j                         d      | j                         dkD  r| j	                        nd}t        j                   ||dk\  ||k        fd       t        | j                  d  | j                  dz   d  z         }|r%| j                         dkD  r|j                  d       | j                  |      | j                  |t
        j                        fS )Nr   )sym_andTrP  r   c                      d  S )Nz9kthvalue(): selected number k out of range for dimension r<   r   s   r6   rX   zkthvalue_meta.<locals>.<lambda>  s    KC5Q r8   r   )r  r[  r   ry   r   rI   rZ   r   r   r#  r   r   )r   rU  ry   r~  r[  dimSizer   s     `    r6   kthvalue_metar^    s     >
dhhjd
;C $
QdiinAG	LLQW%Q
 DS!DJJsQwy$99:E488:>S!>>% $..ekk."JJJr8   c                    | | n|}t        j                  |j                         dk(  d        |j                         }| (t        j                  | j                         |k(  d        |(t        j                  |j                         |k(  d        t        j                  |j                         |k(  d        t        j                  |j                         |k(  d        t        j                  |j                         dk(  d        t        j                  |j	                         |d   |d	   z  d
z  k(  d        y )Nr   c                       yN r<   r<   r8   r6   rX   z(checkLSTMBackwardSizes.<locals>.<lambda>  r`   r8   c                       yra  r<   r<   r8   r6   rX   z(checkLSTMBackwardSizes.<locals>.<lambda>  r`   r8   c                       yra  r<   r<   r8   r6   rX   z(checkLSTMBackwardSizes.<locals>.<lambda>  r`   r8   c                       yra  r<   r<   r8   r6   rX   z(checkLSTMBackwardSizes.<locals>.<lambda>  r`   r8   c                       yra  r<   r<   r8   r6   rX   z(checkLSTMBackwardSizes.<locals>.<lambda>  r`   r8   c                       yra  r<   r<   r8   r6   rX   z(checkLSTMBackwardSizes.<locals>.<lambda>  r`   r8   r   r   r  c                       yra  r<   r<   r8   r6   rX   z(checkLSTMBackwardSizes.<locals>.<lambda>  r`   r8   )rI   rZ   ry   r   r   )grad_hygrad_cyr  r  r  defined_gradexp_sizes          r6   checkLSTMBackwardSizesrm    s    %17wL	LL!!#q(*5  "HW\\^x/<W\\^x/<	LLh&
3	LLh&
3	LLA%z2	LL"hqkHQK&?!&CCZPr8   c                     | |yt        | ||||       t        j                  |t              }t        j                  |t              }|r|j	                  dd      nd }|||fS )NNNNr   r   F)r~  )rm  rI   r   legacy_contiguous_memory_formatr  )	ri  rj  r  r  r  has_bias
grad_gatesgrad_cxrj  s	            r6   #_thnn_fused_lstm_cell_backward_implrt    sl    7?7GRY?!!!@J r1PQG4<
q%0$Iw	))r8   c                    d }d }d }|d   r|j                  | j                               }|d   s|d   rQ|j                  |j                  d      | j                  d      f      }|j                  |j                  d            }|||fS )Nr   r   r   r   r  )r  r  r  r  r  grad_weightrj  s          r6   linear_backwardrw    s    JKI1~!++FKKM:
1~Q",,l.?.?.CV[[QS_-UV **<+<+<R+@A	Y//r8   c                     t         j                        dkD  r j                  d   ||z  z  dk(  sJ d j                   d|        d  fd} j                  d   ||z  z  } j                  d   |z  } j                  d	   |z  }g  j                  d d |||} j                  |      }|j                   |       
      }|S )Nr   r  r   z'Invalid input shape for pixel_shuffle: z with upscale_factor = c                 b    t         j                  j                  |       t         j                  k(  S r3   r  r  s    r6   r  z,meta_pixel_shuffle.<locals>.is_channels_last  s$    ""88=ATATTTr8   c                  2           r.t              dk(  rt        j                  S t        j                  S j	                  t        j                        rt        j                  S j	                  t        j
                        rt        j
                  S y r  )r   rI   r   r  rR  r  )r  r   s   r6   r  z.meta_pixel_shuffle.<locals>.pick_memory_format  s|    D!4 F*...***e.E.EF***e.C.CD((( Er8   r  r   r   )r   r   r   r  )	r   upscale_factorr  r  HrWrr   r   r  s	   `       @r6   meta_pixel_shuffler~  
  s     	DJJ!

2.>2Q RVW W 2$**=TUcTdeW
U	) 	

2>N:;A	B.	(B	B.	(B-$**Sb/-1-b-"-I
..
#C
&&13&
4CJr8   c                 X   | j                  | j                        }|j                  |j                        }|j                  |j                        }|j                  |j                        }|j                  |j                        }|j                  |j                        }|||||||fS r3   r  )r   weight0weight1weight2weight3r<  cx_tmpr   hy_cy_grad_output_r_optgrad_hy_r_optgrad_cy_r_optr   rL  r$  r&  r>  r)  r*  r+  r'  r  diff_xdiff_hxdiff_cxdiff_w1diff_w2diff_bs                                r6   mkldnn_rnn_layer_backwardr  *  s    4 __U[[)FmmCII&Gv||,G.G.Gw}}-F7GVVWgEEr8   )	out_int32r   c                    t        j                  | |rt         j                  nt         j                  t         j                        S )NrQ   r   )rI   r   r5  r   r   )r   
boundariesr  r   s       r6   meta_bucketizer  M  s2     &ekkEKK-- r8   c                     dt               dk(  r't        j                   j                          fd       t               dk(  r% j                         rt	        j
                  d       t        j                  t        t              fd       t        j                  dkD  fd       t        j                  t        t              fd	       t        j                  t        t              fd
       t        j                  k\  d        t        j                   j                   j                        S )Nzhistc()r  c                  $    d j                    dS )Nz%"histogram_cpu" not implemented for 'r  r   r  s   r6   rX   zmeta_histc.<locals>.<lambda>^  s    =ekk]!L r8   r   z%_histc_cuda with floating point inputc                  $     dt                S )Nz#: argument 'bins' must be int, not r  binsr  s   r6   rX   zmeta_histc.<locals>.<lambda>d  s    7)>tDzlK r8   r   c                       d  S )Nz: bins must be > 0, but got r<   r  s   r6   rX   zmeta_histc.<locals>.<lambda>f  s    gY.J4&#Q r8   c                  $      dt               S )Nz%: argument 'min' must be Number, not r  )r  r   s   r6   rX   zmeta_histc.<locals>.<lambda>i      7)@cL r8   c                  $      dt               S )Nz%: argument 'max' must be Number, not r  )r  r  s   r6   rX   zmeta_histc.<locals>.<lambda>m  r  r8   c                       y)Nz&{fn_name}: max must be larger than minr<   r<   r8   r6   rX   zmeta_histc.<locals>.<lambda>o  r`   r8   r   )r   rI   rZ   r   rA   r  rc   r   r   r~   rv   rQ   )r   r  r   r  r  s   ````@r6   
meta_histcr  W  s     G5U"##%L	
 5V#(?(?(A%%&MN	LL4!K 
LLQR	LL3L 
LL3L 
LLMN;;tELLDDr8   c                 B    t         j                         |d      }t        j                   j	                         dk7  xs# t        d  j                         dd  D               fd        j                  |      j                  t        j                               S )Nr   r  r   c              3   &   K   | ]	  }|d kD    ywr  r<   )re   r   s     r6   rg   z,meta_upsample_bimode2d_aa.<locals>.<genexpr>  s     !Ht$(!Hr  r   c                  *    d j                          S r  r   r  s   r6   rX   z+meta_upsample_bimode2d_aa.<locals>.<lambda>  r  r8   r   )
r  r   rI   rZ   r   r  r   r  rA   r   )r   rX  r  r  r  r  s   `     r6   meta_upsample_bimode2d_aar  s  s     -

kA 
LLHc!Huzz|AB7G!HH_ ??+,//11%8 0  r8   c                 T    t        ||d      t        j                   j                  dk(   fd       t	        d      D ]0  t        j                   j
                        k(   fd       2  j                  |      j                  t        j                               S )Nr   r  r  c                  "    d j                    S r  r  r~  s   r6   rX   z4meta_upsample_bimode2d_aa_backward.<locals>.<lambda>  r  r8   c            
      D    d d     d dj                         S )NzD
Expected grad_output to have the same shape as output; output.size(r  z
but got grad_output_size(r   r  s   r6   rX   z4meta_upsample_bimode2d_aa_backward.<locals>.<lambda>  s@     DDE3dK[\]K^J_ `D!1!1!!4 59 r8   r   )
r  rI   rZ   r   r   r   r   r  rA   r   )r  rX  r.  r  r  r  r  r   s   `     @@r6   "meta_upsample_bimode2d_aa_backwardr    s     -K! 
LLAk 1X 
a $4Q$779	

   ,//11+> 0  r8   c                 P   t        j                  |j                         dk(  d        t        j                  |j                         dk(  d        t        j                  |j                  j                  d        t        j                  |j                  j                  d        y )Nr   c                       y)Nz%found_inf must be a 1-element tensor.r<   r<   r8   r6   rX   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r`   r8   c                       y)Nz%inv_scale must be a 1-element tensor.r<   r<   r8   r6   rX   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r`   r8   c                       y)Nz!found_inf must be a float tensor.r<   r<   r8   r6   rX   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r`   r8   c                       y)Nz!inv_scale must be a float tensor.r<   r<   r8   r6   rX   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r`   r8   )rI   rZ   r   rQ   r   )r   r  	inv_scales      r6   *_amp_foreach_non_finite_check_and_unscale_r    s|    	LLQ O 
LLQ O 
LL))3 
LL))3r8   c                 ,    t        j                  |       S r3   r\  )r   nanposinfneginfs       r6   
nan_to_numr    r^  r8   c                    | j                   t        j                  t        j                  t        j                  t        j
                  hvsJ d| j                    d       | j                  }t        ||      }t        ||      }||k(  r| S t        | j                               }t        | j                               }||   ||   c||<   ||<   ||   ||   c||<   ||<   | j                  ||       | S )Nz>torch.transpose_: in-place transposition is not supported for z layout)ru   rI   r  
sparse_cscr  
sparse_bscr   r   r   r   r   r   )r   dim0rM  ndimsr   r   s         r6   r  r    s    ;;	   IU\]  IIE$&D$&Dt|		D$++- F!'vd|F4L&,!$ZdDJT
T6"Kr8   c                    | j                   }| j                  r8| j                         }| j                         }|dk  r|dk(  s,J d| d| d       | j	                         dk  sJ d| d       t        | d|dk  rd      S d      S )	Nr   r   zEt_ expects a tensor with <= 2 sparse and 0 dense dimensions, but got z sparse and z dense dimensionsz6t_ expects a tensor with <= 2 dimensions, but self is r  r   )r   r  r  r  ry   r  )r   r  r  r  s       r6   t_r    s    IIE~~__&
NN$	Q9> 	
!l,yk9JL	
1
 xxzQ 	
DUG1M	
 dAEAIq55155r8   )r  r   sidesorterc                \    t        j                  t         j                        dk  xs  j                  d d j                  d d k(   fd       t        j                  d u xs  j                  j                  k(   fd       t        j                  |dk7  xs | d       |rt         j                  nt         j
                  }t        t         j                        r&t        j                  |t         j                        S t        j                  d| j                  	      S )
Nr   r   c                  `    dt        j                         dt         j                         S )Nztorch.searchsorted(): boundaries tensor should be 1 dimension or the first N-1 dimensions of boundaries tensor and input value tensor must match, but we got boundaries tensor z and input value tensor r   r   )r   sorted_sequences   r6   rX   z#meta_searchsorted.<locals>.<lambda>  s8    3378M8M3N2O P""&tzz"2!35 r8   c                  l    dt         j                         dt        j                         S g  S )Nz[torch.searchsorted(): boundary and sorter must have the same size, but got boundary tensor z and got sorter tensor r  )r  r  s   r6   rX   z#meta_searchsorted.<locals>.<lambda>
  sO    ##'(=(=#>"??V%+%7tFLL!@B  >@@B r8   r   zetorch.searchsorted(): side and right can't be set to opposites, got side of left while right was Truer  r<   r  )rI   rZ   r   r   r5  r   rc   r   r   r   r~   rv   )r  r   r  r   r  r  rQ   s   ``   ` r6   meta_searchsortedr    s     
LLO!!"a' 	9  "%CR8	
	 
LL$?///6<<?	
 
LL#e)	$ %EKK%++E$%U-D-D
 	
 {{2U?3I3IJJr8   c                      t        j                   t         j                  t         j                  t         j                  fv fd       y )Nc                      d  S )Nz/Unsupported input type encountered for isin(): r<   r   s   r6   rX   z3_check_for_unsupported_isin_dtype.<locals>.<lambda>$  s    A%I r8   )rI   rZ   r  
complex128	complex64r   s   `r6   !_check_for_unsupported_isin_dtyper  !  s/    	LLejj%"2"2EOODDIr8   c                 J    | j                  || j                  d      f      }|S )Nr   r  )r  r   num_weightsr  r  rv  s         r6   meta_embedding_dense_backwardr  (  s*     ''k6F6Fr6J(KLKr8   c                 j    |	rt         j                  | ||||||||
|
      S t        | ||||||||
|
      S r3   )r+   _embedding_bag_sparse_backward!meta_embedding_bag_dense_backward)r3  r   rg  r  r  maximum_indicesr  r  rL  r  r  r  s               r6   meta_embedding_bag_backwardr  4  se     22
 	
 1
 	
r8   c
                 N    t        j                   j                  t         j                  t         j                  t         j
                  t         j                  fv  fd       |t        k(  rt        j                  |d u        j                  | j                  d      f      }
|
S )Nc                  "    d j                    S )Nz$Unsupported input type encountered: r   )r3  s   r6   rX   z3meta_embedding_bag_dense_backward.<locals>.<lambda>n  s    6tzzlC r8   r   )
rI   rZ   rQ   rO  rP  rN  float64r  r   r   )r3  r   r  r  r  r  r  rL  r  r  index_grad_weights   `          r6   r  r  _  sv     
LL

u}}ennemmU]]SSC x_D01TYYq\'BCr8   c                    | j                  d      }t        j                  |t        k(  d       t        j                  | j	                         dk(         t        j                  |j	                         dk(         |j                  d      }t        j                  |j	                         dk(         t        j                  |j                  d      |k(         | j                  |f      }	|	S )Nr   zHembedding_bag_backward: per_sample_weights only supported for mode='sum'r   r   )r   rI   rZ   r  ry   r   )
r3  r-  r   rg  r  rL  r  embedding_featuresr  r   s
             r6   .meta_embedding_bag_per_sample_weights_backwardr  v  s     1	LLR 
LLq!	LL!#$,,q/K	LL"#	LLQ#556^^[N+FMr8   )assume_uniqueinvertc                   t        j                  t        | t              xs t        |t              d        t        | t              s!t        j                  | |j
                        } t        |t              s!t        j                  || j
                        }t        | j                         t        |j                         t        j                  | t         j                        S )Nc                       y)Nz<At least one of elements and test_elements must be a Tensor.r<   r<   r8   r6   rX   zmeta_isin.<locals>.<lambda>  r`   r8   r  r   )
rI   rZ   rc   r   r$  rv   r  rQ   r   r  )elementstest_elementsr  r  s       r6   	meta_isinr    s     
LL8V$I
=&(IN h'<<1E1EFmV,]8??K%hnn5%m&9&9:HEJJ77r8   r   c                     t        j                  | dk\  d        t        |t        j                        \  }}t        j
                  ||      S )Nr   c                       y)Nz,polygamma(n, x) does not support negative n.r<   r<   r8   r6   rX   z meta_polygamma.<locals>.<lambda>  r`   r8   r  r   )rI   rZ   r   r   r  r   )r   r   rD   rE   s       r6   meta_polygammar    sF     
LLaOP(;HHOA| D55r8   c                     t        d      )Nz.Tensor.item() cannot be called on meta tensors)r  r   s    r6   meta_local_scalar_denser    s    
G
HHr8   c                 ,    t        j                  |       S r3   r\  r   s    r6   silur    r^  r8   c                 l    t        | t        j                        \  }}t        j                  | |      S r  )r   r   r  rI   r   )r   rD   rE   s      r6   sigmoidr    s3     );HHOA| D55r8   c                 R   | j                         dk(  }|j                         dk(  }|r|r4|j                  d      | j                  d      |j                  d      g}n"t        j                  |j                  d      |j                  d      k(  d        | j                  d      |j                  d      g}n|r[t        j                  |j                  d      | j                  d      k(  d        | j                  d      |j                  d      g}njt        j                  | j                  d      |j                  d      k(  d        | j                  d      | j                  d      |j                  d      g}|xs | j                  }t        j
                  j                  rZd|j                  z  }|d   |z   dz
  |z  |z  }||k(  r|d   |z  |dg}	n|dg}	t        j                  ||	|| j                  	      }
|
S t        j                  ||| j                  	      }
|
S )
Nr   r   r   c                       yNz matrix batch sizes have to matchr<   r<   r8   r6   rX   z2_create_grouped_mm_output_tensor.<locals>.<lambda>  r`   r8   r   c                       yr  r<   r<   r8   r6   rX   z2_create_grouped_mm_output_tensor.<locals>.<lambda>  r`   r8   c                       y)Nzbatched dimension has to matchr<   r<   r8   r6   rX   z2_create_grouped_mm_output_tensor.<locals>.<lambda>  r`   r8   rM  r  )ry   r   rI   rZ   rQ   r  r   itemsizer  rv   r~   )r:  r<  offsr1  
mat1_is_2d
mat2_is_2dr2  	alignmentsize_paddedr  r   s              r6    _create_grouped_mm_output_tensorr    s   qJqJ		!diilDIIaLAHLL		!		!,.X 		!diim4HLL		!		!,.X 		!diil3H LL		!		!,.V 		!diilDIIbMBH'TZZI}}),,,	|i/!3	AIM#"1+3[!DJ%q)J!!j	$++

 J kk()DKKHJr8   mat_amat_br  c	                     t        j                  d u d u k(  d        d uxr d u}	|	rst         j                  j                  rt         j                  nt         j
                  }
t        j                   j                  |
k(  xr j                  |
k(   fd       nTt        j                   j                  t         j                  k(  xr j                  t         j                  k(   fd       t        j                   j                         dv xr j                         dv  fd        j                         dk(  }j                         dk(  }|r|s7t        j                   j                  d      j                  d      k(  d	       |	rDd
 }d }t        j                   |        fd       t        j                   |      fd       d } |d         |d       # t        j                  j                  t         j                  k(  xr j                  t         j                  k(  xs< j                  t         j                  k(  xr j                  t         j                  k(  fd       j                  t         j                  k(  xr j                  t         j                  k(  d dfd	}|r|rj                  d   nd} |d d|        |dd|       t        j                  |d u d        |s|r}t        j                  d u fd       xt        j                  j                         dk(  fd       t        j                  j                  t         j                  k(  fd       nt        j                  d u d        t        j                  |d u d        t        j                  |d u xs |t         j                  k(  d        t         |      S ) Nc                       y)Nz,Either both scale factors are given, or noner<   r<   r8   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  r`   r8   c                  >    d j                    dj                    dS )Nz5Expected inputs of E4M3 FP8 type but got mat_a.dtype= and mat_b.dtype=rf  r   r  r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  s%    KEKK=Xijojujuivvwx r8   c                  >    d j                    dj                    dS )Nz1Expected inputs of BF16 type but got mat_a.dtype=r  rf  r   r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  s%    G}Tefkfqfqerrst r8   )r   r.   c                  L    d j                          dj                          S )Nz3Multiplicands must be 2D or 3D but got mat_a.dim()=z and mat_b.dim()=r   r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  s'    Eeiik]Rcdidmdmdocpq r8   r   r   r  z3contraction dimension of mat_a and mat_b must matchc                 F    | j                         }|d   dkD  xr |d   dk(  S Nr  r   r   r  mat
mat_strides     r6   r  z-_meta_grouped_mm_common.<locals>.is_row_major  s*    Jb>A%=*R.A*==r8   c                 F    | j                         }|d   dk(  xr |d   dkD  S r  r  r  s     r6   r  z-_meta_grouped_mm_common.<locals>.is_col_major  s*    Jb>Q&=:b>A+==r8   c                  0    d j                         dd   S )NzNExpected mat_a tensor to be row major in the last two dimensions, got strides r  r  )r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>%  s#    dejeqeqestvtwexdyz r8   c                  0    d j                         dd   S )NzQExpected mat_b tensor to be column major in the last two dimensions, got strides r  r  )r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>)  s#    ghmhththvwywzh{g|} r8   c                     j                         dz
  dj                         z  }j                         dz
     dk(  rG   t        dj                  dz
           k\  r%t        j                     |z  dk(   fd       y    dk(  rJdz
     t        dj                           k\  r(t        j                  dz
     |z  dk(   fd       y t        j                  dfd       y )Nr   rM  r   c                  "    d d  d     dS )Nr   stride along % dim to be multiple of 16 bytes, got rf  r<   end_dimmat_namer	  s   r6   rX   zF_meta_grouped_mm_common.<locals>.check_valid_strides.<locals>.<lambda>5  s)    )H:^G9Dijtu|j}i~~  A r8   c                  .    d d dz
   d dz
      dS )Nr  r	  r   r	  rf  r<   r		  s   r6   rX   zF_meta_grouped_mm_common.<locals>.check_valid_strides.<locals>.<lambda><  sK    )H:^GaK=Hmnx  zA  DE  zE  oF  nG  GH  I r8   Fc                  *    d d j                    dS )NzInvalid strides/sizes, got z for strides and z for sizes.r)  r  s   r6   rX   zF_meta_grouped_mm_common.<locals>.check_valid_strides.<locals>.<lambda>A  s!    5j\ARSVS\S\R]]hi r8   )ry   element_sizer   r  r   rI   rZ   )r	  r 	  r  r
	  r	  s   `` @@r6   check_valid_stridesz4_meta_grouped_mm_common.<locals>.check_valid_strides,  s    '')a-#**,,	ZZ\
gk"a'Jw,?3syy1%D
 -
 LL7#i/14 A  A%*Wq[*ASsyy!F
 +
 LL7Q;')3q8 I
 LLir8   r  r  c                  >    d j                    dj                    dS )NzhFor FP8 scales must both be float32, or for MXFP8 both scales must be float8_e8m0fnu. Got scale_a.dtype=z and scale_b.dtype=rf  r   r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>N  sT    ~  @G  @M  @M  N  Na  bi  bo  bo  ap  pq  r r8   c                     | |z   dz
  |z  |z  S )z$Rounds up x to nearest multiple of yr   r<   rF   r"  s     r6   round_upz)_meta_grouped_mm_common.<locals>.round_upU  s    UQY1$))r8   r   c                    	
 j                         dk(  rt        j                  j                          fd       r;t        j                  j                         j                         k(   fd       y t        j                  j                         dk(   fd       t        j                  j                  d   j                     z  k(   fd       y t        j                  j                  d      dk(   fd	       t        j                  j                  d   j                  d   k(   fd
       rt        j                  j                  j                  k(   fd       j                  \  }}d} ||z  d      	 |d      
t        j                  j                  d   	k(  xr j                  d   
k(  	
fd       y t        j                  j                         dk(   fd       t        j                  j                  d   j                  dz      k(   fd       y )Nr   c                      d  dS )Nr  z to be contiguous.r<   
scale_names   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>]  s    i
|3EF r8   c                  B    d d j                    dj                    S )NzKFor MXFP8, scale must have same number of dimensions as target tensor, but  has mat.ndim= and scale.ndim=r  r 	  re  r	  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>f  sZ    "mnxmy  zH  IL  IQ  IQ  HR  Rb  ch  cm  cm  bn  !o r8   r   c                  2    d d j                          dS )Nr  z to be 1D tensor, but got 	D tensor.r   re  r	  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>k  #    )J<7QRWR[R[R]Q^^g h r8   r   c                  V    d d j                      z   dj                   d    dS )Nr  z	 to have r  r   z
 elements.r)  )r 	  re  scale_multiplierr	  
scaled_dims   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>o  sW    )J<y:AVYiAi@jjyz  {F  {F  GH  {I  zJ  JT  !U r8   r   c                      d  dS )Nr  z( to be contiguous in the last dimension.r<   r	  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>t  s    i
|3[\ r8   c                  P    d d j                   d    dj                   d    dS )Nr  z batch dimension to be r   , got rf  r)  r	  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>x  s6    i
|3J399UV<.X^_d_j_jkl_m^nnop r8   c                  B    d d j                    dj                    S )NzMFor MXFP8, scale should have same number of dimensions as target tensor, but r	  r	  r  r	  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  sZ    "opzo{  |J  KN  KS  KS  JT  Td  ej  eo  eo  dp  !q r8   r  r  r  r  c            
      N    dj                    d  d d dj                    
S )NzFor MXFP8, expected mat.shape=z to have scale shape of (,z), but got r)  )G	blocked_K	blocked_Nr 	  re  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  sT    "@Kdefdgghirhsstu~t  @K  LQ  LW  LW  KX  !Y r8   c                  2    d d j                          dS )Nr  z to be 2D tensor, but got r	  r   r	  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  r	  r8   c                  V    d d j                   dz       dj                   d    dS )Nr  z non-batch dimension to be r   r%	  rf  r)  )r 	  re  r	  r"	  s   r6   rX   z>_meta_grouped_mm_common.<locals>.check_scale.<locals>.<lambda>  sT    )J<7RSVS\S\]^ak]kSlRmmstytt  AB  uC  tD  DE  !F r8   )ry   rI   rZ   rR  r   r   r   )r	  re  r 	  r"	  r!	  rI  ru  rV  r)	  r*	  r+	  is_mxfp8r	  s   `````   @@@r6   check_scalez,_meta_grouped_mm_common.<locals>.check_scaleY  s   wwyA~'')F LL		swwy0 o
 LL		q(h LLA#))J*?BR*RR U
 LL$)\ KKNciil2p LLEJJ. q $kkGAq!!#J (Z ;I (C 0ILL		"2Qsyy}	7Q Y
 LL		q(h LLA#))A
N*CC Fr8   r   r  r  c                       y)Nz:Scale result tensor provided, but it is not supported yet.r<   r<   r8   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  r`   r8   c                  N    d j                          dj                          dS )Nz/Offsets tensor not provided, but is needed for zD/zD multiplicand layouts.r   r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  s*    Eeiik]RTUZU^U^U`Taaxy r8   c                  ,    d j                          dS )Nz.Offsets tensor must be 1D, but got offs.dim()=rf  r   r  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  s    HTUV r8   c                  $    d j                    dS )Nz7Offsets tensor must be integer (int32) tensor, but got rf  r   r3	  s   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  s    QRVR\R\Q]]^_ r8   c                       y)NzJOffsets tensor provided, but is not needed for 3D/3D multiplicand layouts.r<   r<   r8   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  r`   r8   c                       y)Nz2Bias tensor provided, but it is not supported yet.r<   r<   r8   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  r`   r8   c                       y)Nz4If output dtype provided, it must be torch.bfloat16.r<   r<   r8   r6   rX   z)_meta_grouped_mm_common.<locals>.<lambda>  r`   r8   r  )rI   rZ   r  r  r  rQ  rQ   rP  ry   r   rN  r  r   r5  r  )r  r  r  r  r  r/  r  r1  r  scaled	fp8_dtypemat_a_is_2dmat_b_is_2dr  r  r	  r/	  r!	  r.	  r	  s   `````             @@r6   _meta_grouped_mm_commonr<	    sa    
LL	Dgo.> D 8WD%8F -2]]->->E))EDWDW	KK9$A	)Ax	

 	KK5>>)KekkU^^.Kt	

 
LL		v7%))+"7q
 ))+"K))+"KkJJrNejjn,A	

 	>	> 	z	
 	}	

0 ''w2]]emm+N0N !5!55 :MMU%9%99 r	
 MMU111 6!5!55 	
	*:	z "-++DJJqMST 	 	Iwq2BCIwq2BCD P	

 ky	
 LL
aV LL

ekk)_
 	DL`	

 
LLD
 
LLT8Y%..8F
 ,E5$	JJr8   c           
      (    t        | |d d ||d |      S )N)r  r  r  r/  r  r1  r<	  )r  r  r  r/  r1  s        r6   meta_grouped_mmr?	    s)     #	 	r8   c	                 *    t        | ||||||||	      S )N)r  r  r  r/  r  r1  r  r>	  )	r  r  r  r  r  r/  r  r1  r  s	            r6   meta_scaled_grouped_mmrA	    s,     #!%
 
r8   rF   half_to_floatc                    |r| j                   t        j                  k(  sJ t        j                  | t        j
                  j                        \  }}|s|n|}t        j                  | |t        j                        }|S )Nr  r  )	rQ   rI   rK   rA   r   r   rB   r   r   )rF   ry   rB	  computation_dtyperE   r  s         r6   softmaxrE	    so     ww%**$$$&+&>&>	uDDLL'#| (5<:KL


1L@W@W
XCJr8   c           	        	
 t        j                  t              dz  dk(  fd       | j                  t              
t              dz  }
|z
  	t        j                  
|k\  
fd       t	        d D              r| }t        	
      D ]t  d
z
  dz
  z     dk  r*|j                      |j                        z         }dz      dk  sL|j                  d|j                     dz      z         }v |j                         S t        d 	       }t        |      D ]^  t              dz   dz  z
  	z         z   dz      z   }t        j                  |dk\  	fd       |j                  |       ` t        j                  || j                  | j                  | j                  t        |             S )	Nr   r   c                       dt                S )Nz1Length of pad must be even but instead it equals r  r-  s   r6   rX   z'_constant_pad_nd_meta.<locals>.<lambda>  s    CCH:N r8   c                  (    dt               d  dS )Nz`Length of pad should be no more than twice the number of dimensions of the input. Pad length is z while the input has z dimensions.r  )l_inpr.  s   r6   rX   z'_constant_pad_nd_meta.<locals>.<lambda>  s"     225c(;P' r8   c              3   ^   K   | ]%  }t        |t        j                        xr |d k   ' ywr  )rc   rA   IntWithoutSymInt)re   r/  s     r6   rg   z(_constant_pad_nd_meta.<locals>.<genexpr>  s)     
I:a//0;Q!V;
Is   +-r   c            	      F    d z       d    ddz       d z    d	S )NzThe input size z, plus negative padding r   r   zG resulted in a negative output size, which is invalid. Check dimension z of your input.r<   )r   r   l_diffr.  pad_idxs   r6   rX   z'_constant_pad_nd_meta.<locals>.<lambda>&  sG    ok&1*&=%>>V7|nE#gk"2!3 4117!OM r8   )rQ   rv   rx   r   )rI   rZ   r   r   r  r   narrowr   r   r   r~   rQ   rv   rx   r   )r   r.  r   l_padc_input	new_shapenew_dimr   r   rM	  rI	  rN	  s    `     @@@@@r6   _constant_pad_nd_metarT	    s    
LLC1N
 ++KEHMEU]F	LL	 
IS
IIvu% 	TA519q=)G7|a!..G}gmmA&6W&E 7Q;!#!..Aw}}Q/?#gPQkBR/RS	T }}[&)*I5\ 	"c(q1uk*fqj)CL83w{;KKqLM	
 	!	" ;;kk||))+E2 r8   r  r  r  c                    | j                         dk(  sJ d       | j                  }|j                  }|j                  dk(  r|d   f}n$|j                  dk(  r|d   |d   f}n
g ||d   }| j                  }| j	                  ||      S )Nr   z'weight' must be 2-Dr   r   r   )ry   r   r   rQ   r   )	r-  r   r  r  r  weight_shapeindices_shaper   r1  s	            r6   	embeddingrX	  5  s     ::<1444<<LMMM||q&21o%7			"1%|A7	5m5\!_5	IIY77r8   max_lengthspadding_valuec                     t        |      dk(  sJ t        |      dk(  sJ |d   j                  d   dz
  }|d   }||g| j                  dd  }| j                  |      S r  )r   r   r   )r   rg  rY	  rZ	  r  rh  rW  s          r6   $meta__jagged_to_padded_dense_forwardr\	  M  sv     w<1{q   
aAAAq,6<<+,LL))r8   c                 B    t        |       t               d               }|S )Nc                 8    t        | t        j                        S r  rG   r   r  r  s    r6   _fz)_create_unary_float_meta_func.<locals>._f_  s      =JJ
 	
r8   r=   r$   funcr`	  s     r6   _create_unary_float_meta_funcrd	  ^  *    4]
  

 Ir8   c                 B    t        |       t               d               }|S )Nc                 :    t        | |t        j                        S r  r_	  r	  s     r6   r`	  z*_create_binary_float_meta_func.<locals>._fj  s      q!@!M!M
 	
r8   ra	  rb	  s     r6   _create_binary_float_meta_funcrh	  i  re	  r8   c                      t                fd       } j                   d}||_         t        t        t        |            |      }|S )Nc                 `     | g|i |}t        | j                  |j                         | S r3   r  )r   rC   r  r   r5   s       r6   _fnz#_register_inplace_meta.<locals>._fn  s.    '''

CII6r8   rD   )r   rn   r=   getattrr+   )r5   rk	  inplace_names   `  r6   _register_inplace_metarn	    sO    
2Y 
 kk]!$LCL
4-l3
4S
9CJr8   c                 f    t        j                   j                  j                  k(   fd        g}t        t              rQj
                  dk7  r1t        j                   j                  j                  k(   fd       |j                         t        |dt        j                  iS )Nc                  <    dj                    d j                    S )Nre  z for `end`, but got dtype r   )ro   rp   s   r6   rX   zlerp.<locals>.<lambda>  s    /%++.HT r8   r   c                  <    d j                    dj                    S )Nre  z for `weight`, but got dtype r   )rp   r-  s   r6   rX   zlerp.<locals>.<lambda>  s!    /%++6STZT`T`Sab r8   r>   )
rI   rZ   rQ   rc   r   r   r   rG   r   rB   )rp   ro   r-  rC   s   ``` r6   lerprr	    s     
LLsyy T 3<D&*%;;!LLv||+b 	F	=EE r8   )r   c                <    t        | ||t        j                        S r  r  r   tensor1tensor2r   s       r6   addcmulrw	    s!     w0O0W0W r8   c                    t        j                  t        j                  |j                        xr t        j                  |j                         d        t        | ||t        j                        S )Nc                       y)N)zFInteger division with addcdiv is no longer supported, and in a future zErelease addcdiv will perform a true division of tensor1 and tensor2. z4The historic addcdiv behavior can be implemented as zA(input + value * torch.trunc(tensor1 / tensor2)).to(input.dtype) zfor integer inputs and as z6(input + value * tensor1 / tensor2) for float inputs. z?The future addcdiv behavior is just the latter implementation: z4(input + value * tensor1 / tensor2), for all dtypes.r<   r<   r8   r6   rX   zaddcdiv.<locals>.<lambda>  r`   r8   r  )rI   rZ   rA   r.  rQ   rG   r   rB   rt	  s       r6   addcdivrz	    sb     
LL""7==1 6&&w}}5	
		
  w0O0W0W r8   c                     i } dD ]  }t         |   }|D ]  }|| vs||   | |<    ! | j                         D ]  \  }}t        |t        j                  j
                        r,t        |t              sJ  |j                  t        j                  j                  j                        |       t        j                  j                  |j                         d      r|t         d   v st        | d      |j                  r|j                         dv rd|j                         v rt        j!                  ||       	d|j                         v rt"        j!                  ||       3d|j                         v rt$        j!                  ||       ]d	|j                         v rt&        j!                  ||       t(        j!                  ||        y )
N)rs   post_autogradpre_autogradCompositeImplicitAutogradrs   z is a CompositeImplicitAutograd op, we shouldn't register meta function for it. Instead, we should let the decomposition run and write meta kernels for the base operators.>   aten::cloneaten::copy_aten::rot90aten::_to_copyaten::empty_stridedaten::constant_pad_ndaten::as_strided_scatterzmkldnn::zmkl::zonednn::zquantized::)r   itemsrc   rI   _opsHigherOrderOperatorr   py_impl_CDispatchKeyr-   %_dispatch_has_kernel_for_dispatch_keyr  r  is_view2_meta_lib_dont_use_me_use_register_meta_for_mkldnnimpl/_meta_lib_dont_use_me_use_register_meta_for_mkl2_meta_lib_dont_use_me_use_register_meta_for_onednn5_meta_lib_dont_use_me_use_register_meta_for_quantized'_meta_lib_dont_use_me_use_register_meta)activate_meta_tablerm   registryopoop_overloadr5   s         r6   activate_metar	    s    : 9-d3 	9C--+3C=#C(	99 /446 6NR
 k5::#A#AB+z2226EHH00556r:8899 ;
 8@@""m $; ; 
    	 [--//BGGUWXK,,..?DD[RTU{//11BGGUWX+"2"2"44EJJ 8<<["Mm6Nr8   )Fro  r3   )NNNFr   r   r   rv  )Tr  )r  )r  T)FF)TT)rN  )FTN)TFF)TF)r   )g      ?N)r/   str)r<   r  r  F)r<   r  FTN)Fr   FNFr   )NF)r   F)g      ?gUUUUUU?FN)NNNNN)r   NNr   )NNF)        FFN)Nr	  FFN)r	  FNN)Nr	  FNN)r	  FN)FN)FNNNN)NNNF)Nr   FNN)NNNN)r   TT)NNr   N)d   r   r   )r   )NNNNF)r   FF)r	  (  r  collections.abcr   enumr   	functoolsr   typingr   r   r   r	   typing_extensionsr
   rI   torch._prims_commonr  rA   r   r   r   torch._decompr   r   r   r   
torch._opsr   torch._primsr   r   r   r   r   r   r   r   r   r   r   r   r   torch._prims_common.wrappersr    r!   r"   r#   r$   r  r%   r&   torch.fx.experimentalr'   rK  torch.utilsr(   r9   r)   r*   opsr+   libraryLibraryr	  r   r  r  r  r=   rG   rS   r\   linspacelogspacer  r   taker  r   r   r   r   cummaxcumminr   r   r   r   r   r  r   _fft_c2cr   r   r   _fft_r2cr   randpermgenerator_outr   r   r  randintr
  r  low_outr  randr  _fft_c2rr  r  r   r(  
unsqueeze_r,  _sparse_semi_structured_linearr	  rQ   r9  _sparse_semi_structured_mmr?  _sparse_semi_structured_addmmrC  _cslt_sparse_mmrX  index_reducer_  index_reduce_ra  index_selectre  segment_reducerr  r  	unary_outrv  ry   r  r   r  r  r  r  r  _assert_asyncr  msgr  _printr  _make_dep_tokenr  r  _functional_sym_constrain_ranger  r  (_functional_sym_constrain_range_for_sizer  _functional_assert_asyncr  r   r  r   r  r  r  r  _linalg_eighr  r  _linalg_eigvalslinalg_eigvalsr  
linalg_eigr  r  r  r  r  r  r  r  linalg_inv_exr  linalg_ldl_factor_exrY   r#  linalg_ldl_solver2  	linalg_lur8  linalg_lu_factor_exr<  linalg_lu_solverE  	lu_unpackrK  rT  	linalg_qrr[  r_  r\  _linalg_svdri  r/  r  ry  r  linalg_solve_triangularr  r  r  _linalg_detr  r  r  r  reflection_pad1dr  replication_pad1dr  r  reflection_pad1d_backwardr  replication_pad1d_backwardr  r  reflection_pad2dr
  replication_pad2dr  reflection_pad2d_backwardr  replication_pad2d_backwardr  r"  reflection_pad3dr$  replication_pad3dr'  reflection_pad3d_backwardreplication_pad3d_backwardr.  _pdist_forwardrM   r2  _pdist_backwardr8  baddbmmrQ  	bernoullirT  
bernoulli_rW  r/  rZ  poissonr]  _fused_moving_avg_obs_fq_helperrn  mmrx  r|  r   r  r  miopen_batch_normr  convolutionr  r	  _has_mkldnnr	  r  _convolution_pointwiser  _linear_pointwiser  has_mklr	  r  _mkl_linearr  r	  r  qconv2d_pointwiseqconv_pointwiser  binaryr  qlinear_pointwiser$  r  binary_tensorr  linear_dynamic_fp16linear_relu_dynamic_fp16r  r	  r  
max_pool2dr  int4mm_packed_weight_cpur  r  
avg_pool2dr"  r'  avg_pool2d_backwardr/  
avg_pool3drI  avg_pool3d_backwardrT  _adaptive_avg_pool2drY  _adaptive_avg_pool3dr\  _adaptive_avg_pool2d_backwardrf  _adaptive_avg_pool3d_backwardrk  ri  adaptive_max_pool2dr{  r  r  adaptive_max_pool3dr  r  r  repeat_interleaver  rd   r  r  r  r   _unsafe_indexr  convolution_backwardr  addbmmr  randint_liker  _fused_adam__fused_adamw_r  _fused_adamr  _int_mmr  _convert_weight_to_int4packr  #_convert_weight_to_int4pack_for_cpur
  _weight_int4pack_mmr  _weight_int4pack_mm_for_cpur  r$  r&  rN  _dyn_quant_pack_4bit_weightrZ  _dyn_quant_matmul_4bitr`  _weight_int8pack_mmrg  _cdist_forwardr|  _cdist_backwardr  _embedding_bagr  _embedding_bag_forward_onlyr  r  nansumr  median	nanmedianr  
dim_valuesrL  r   r  logical_not_r  repeatr  zero_r  mul_Scalardiv_logical_and_logical_or_logical_xor_r  add_sub_r  r  subr  rounddecimalsr  r  
__rshift__r  
__lshift__r  zeror  r  r  fillr  relu_r  	_add_relur  rrelu_with_noiser  rrelu_with_noise_functionalr  rrelu_with_noise_r  	index_put_unsafe_index_putr  masked_fill_r  _masked_scaler  masked_scatter_r  masked_scatterr
  masked_scatter_backwardr  
index_put_r  r  bmmr  r   r$  r)  r  r  r>  rM  rP  r   max_pool2d_with_indices_backwardrZ  max_pool2d_with_indicesr\  fractional_max_pool2drn  max_pool3d_with_indicesr{   max_pool3d_with_indices_backwardr  r  r  r  grid_sampler_2d_backwardr  r  r  r  r  onesr  zerosr  select_scatterr  slice_scatterr  r   r  r  gatherr  r  r  r  r  r  scatter_addr  scatter_add_r  r  r  r   r[  value_reducer	  scatter_r  #_scaled_dot_product_flash_attentionr#  r,  #_scaled_dot_product_cudnn_attentionr6  0_scaled_dot_product_fused_attention_overrideabler9  ,_scaled_dot_product_flash_attention_backwardrD  +_scaled_dot_product_flash_attention_for_cpurG  4_scaled_dot_product_flash_attention_for_cpu_backwardrK  *_scaled_dot_product_attention_math_for_mpsrb  '_scaled_dot_product_efficient_attentionrf  0_scaled_dot_product_efficient_attention_backwardrn  ,_scaled_dot_product_cudnn_attention_backwardrp  _flash_attention_forwardrx  _flash_attention_backwardr}  _efficient_attention_forwardr  _efficient_attention_backwardSymIntr  
_scaled_mmr  scatter_reducetwotwo_outr  scatter_reduce_r  multinomialr  r  r  r  _upsample_nearest_exact1dr  _upsample_nearest_exact2dr  "_upsample_nearest_exact2d_backwardr  _upsample_nearest_exact3dr   r  values_stabler  r  _thnn_fused_lstm_cellr  r6  r@  rE  rH  rJ  argminrK  rN  topkrU  _segment_reduce_backwardrY  kthvaluer^  r   rp  rm  rt  rw  pixel_shuffler~  r  	bucketize
Tensor_outr  histcr  _upsample_bilinear2d_aa_upsample_bicubic2d_aar   _upsample_bilinear2d_aa_backwardr  r  r  r  r  searchsortedr  r  embedding_dense_backwardr  _embedding_bag_backwardr  _embedding_bag_dense_backwardr  *_embedding_bag_per_sample_weights_backwardr  isinr  	polygammar  _local_scalar_denser  r  r  r  r<	  _grouped_mmr?	  _scaled_grouped_mmrA	  _softmaxrE	  constant_pad_ndrT	  rX	  _jagged_to_padded_dense_forwardr\	  rd	  rh	  special_airy_aispecial_bessel_y0special_bessel_y1special_modified_bessel_i0special_modified_bessel_i1special_modified_bessel_k0special_modified_bessel_k1!special_scaled_modified_bessel_k0!special_scaled_modified_bessel_k1special_chebyshev_polynomial_tspecial_chebyshev_polynomial_uspecial_chebyshev_polynomial_vspecial_chebyshev_polynomial_w&special_shifted_chebyshev_polynomial_t&special_shifted_chebyshev_polynomial_u&special_shifted_chebyshev_polynomial_v&special_shifted_chebyshev_polynomial_wspecial_hermite_polynomial_hspecial_hermite_polynomial_hespecial_laguerre_polynomial_lspecial_legendre_polynomial_prn	  rr	  rw	  rz	  lerp_addcmul_addcdiv_torch._refs.nn.functionaltorch._refs.specialr	  r<   r8   r6   <module>r
     sH    $   5 5 '  # + +  " U     < 7 ) T]t_yy~~*/--*?*?PV*W ' %a )X
8BF#3"4hr2v6F"FG 
3(* t}}-.
 

==5  /5p 		!!499==12'  3' !!))4+<+<+@+@AB%' %  C%$ t%%&I  'I 	[[$++//4;;+>+>P Xy! " !!))4+<+<+@+@ABI  CI3lV $s) 4  %%t}}'8'89:K  ;K $s)  %%t}}'8'89:8
  ;8
v t}}**+"& 3 ,3 t}}$$% **
 &
 $$dll&6&678
 **  9&   $,,"6"678 **  9& 		!!499==12%)$tPT   3 %%t}}'8'89:$Dv $DDI $Dc $DC $D  ;$DN tzz!!" #0( t&&' ( t223
 "%)'+  6
	
 c] $ 4B t../
 (,	
  $	 00 t112 	
'+
  	 $ 3D t##$ ""'+"-<,,-<\\-< 6
-< F	-<
 $-< -< -< -< -< %-<` t  (() 	I
	I		I 	I LL		I
 	I 	I 	I *	I t!!))* 	
			 	 LL		
 	 	 	 +	 t  (()' * ' t""**+
 !% $ $ W
 W W f	 W
 f W f W  W  W  W , WF   $(("4"456  7 txx||    $(("4"456  7 txx||  tzz!!"6 #6 tzz~~( (
 t!!))* + t!!%%& ' t{{""# $ t##++, ) -) t''//0, 1, t33;;< =
 t008896 :6& t<<DDE F
 t,,001 2
F C    F  #  N (,


 !%
$V S C 
 
F 
$ 
 
"  	  	C  !!))4+<+<+H+HIJ]N+ s T  , K" $$,,d.A.A.E.EFGB B6 B  HB  !]N+	6 	 , "	Q QF Q t**+) )F )4 )F )  ,) t""#J JF J4 JF J  $J t}})6 )$ )6 )  ) t$$%)6 )$ )6 )  &) t&&../&  T  0" 	$$,,d.M.M.Q.QR .f .6 .f . .d t!!))*&   + ))1143L3L3P3PQRT8V$ 	
  	
 666!" % S& %%--t/D/D/H/HIJ ''' '
 ' '  K'T &&(:(:;<S#s/3 f  fff>T8U   =4 ((00$2J2J2N2NOPT8V$ 	  	
 666!" % QD $$,,d.B.B.F.FGH 444 4
 4 4 4  I4n t~~S#s 	$$$ $ 	$
 666!"$  $P tTz!2 * &&(:(:;<S#f C ffn8M   =4 $$,,d.B.B.G.GHIV[$1'v '%(F"G ' 2 J'$ t''(   	""" " SM	" )"J.
.
. 49d3i .".
.
. 3-. 66>	.(f V   t$$%
 ##!777 	7
 7 V7 	7 V7 6
7 6666)*7 &7t ,,44d6R6R6V6VWX   	
   
&	  Y2 t$$%S#4( +(
+(+( +( 	+(
 +( 66>+( ) &+(^ t''(
 )
 tzz
 WW	W W 	W
 W W  Wt>#;L t$$%=  &= t%%&>  '>(< t--.\S  /S t../\T  0T2Ej t$$%=  &= t%%&>  '> &&..&&11''//''22	 \& &:<G~ t$$%=  &= t%%&>  '> &&..&&11''//''22	 \$( $(N t""#

f 

 

v 

  $

 t##$Pv PV P Pf PQW P  %P $$dll&6&678/0 '  9': &&(:(:;<&* I  =I
 t$$% & t~~ I !I
 $$dll&6&678"  9" t33;;< * =*. tww	  	B
* 7;j,,jLLj $s)S.!j 49c>"	j
 DIsN#j j j U49c>23jZQ t%%--."$,,"$LL"$ 5<<
 "$ 5<<(	"$
 %,,'"$ "$ !&"$ "$ /"$J t''(),,)LL) ,,) I	)
 #Y) 3i) ) I) ) ))X 	889>9N9N&&:6 599##::BBC D, 599##55==>S ?S
 xx:?--:O:O66;
7 
uyy}}00	1	 
2	
 :?9N9N&&:6 599##55==>599##33;;<0 = ?0d 599##55<<= >6 599##55==>599##55<<= > ?8 599##55<<=599##55CCD! E >!F 599##77??@599##<<DDE	 F A	 =BMM<Q<QVV=9 599&&112 
 3
8 599&&??@@ A@( t&&' M (Mb(<X t''//0E 1EP t UJ   UJp t''(\K(  )K(\ t((001 2" t((001@ 2@ t1199:F ;F, t112\P  3P
	
6 	
S 	
 t''(UI+  )+\ t001\H  2H$ t''(UI'  )'T t001\(  2(
 t%%,,-* .* $$dll&6&678  9 ##++T-@-@-D-DEF46 @c @  G@ 		&&..		0F0F0J0JKL  M" 

!!4#5#5#<#<=> ?D ))1123H 4H: ##T[[__56./q '  7'0 !!(()*' +' !!))4+=+=+E+EFG  !
 H
2   (()*  !! +!H ~B  B* 0012 3& 889: ; (()*@ +@ 0012< 3< >>?@< A<"3 "3 "3 "jZ 0012D!)&!1D 3D0 ++,-; .;  (()*< +< t""**+ & , &F t##$G  %G* t""**+
 	
`5 ,`5F t//7785 95
 ##T[[__56=$ =  7= ##T^^%;%;<=) >) !!					 Xy! "	 t  (() * t{{""#' $'& tzz!!" # 								!!  !!

 									**Z  

""DJJ$7$789 :
" &&(>(>?@ A &&(>(>?@ A tyy  !& "& 

!!4::#4#456 7 		  $))"2"234" 5" tzz!!" # t~~$$%F   & %%&'RV"  (" 0012RV; 3; &&'(KO ) &&(>(>(F(FGH" I" t  ''( )
 t!!))* + t##$	 %	 t""#6  $6 t++,! -! t&&' (&R txx 5 !5 txx~~J J6;h #-YYY 	Y 		Y
 	Y 	Y 	Y 	Y 	Y 	Y 	Y Y Y Y Y  !Y" #Y$ %Y& 'Y( )Y* +Y, -Yx;4|383838 38 		38
 	38 	38 	38 	38 	38 	38 	38 	38 38 38 38  !38" #38$ %38& '38lI2X t44<<=( >(V t++334 # 5#L t))112Q 3Qh t++,UI d  -dN t445\b  6bJ%
V %
6 %
Pt  v 3 $ t,,445# 6#$ t##$8  %8" t,,-\;'! ( .!, 		!!"#. $. t&&' ) ()X 		!!499==12   3( 

""DJJNN34   5( t""**+. ,. t!!))*. +.
	 	C 	d 	/
  t{{""#' $'6
 
-
b4 t''(& )&
 t  ! "
 !!	 & & ""	 889:
 #!==	= = 	=
 = = E?= ;=@S#X( 889: #!((	( ( 	(
 ( ( ( ( E?( ;(V EEFG
 #'#!''	' ' 	'
 ' ' ' E?' H'T 99( """" 
" 	"
 
" " " " " " " " " " E?"
". 88 "&!	  	
   E?
: AA #'!!"!"!" 
!" 	!"
 
!" !" !" !" !" E?!"
!"H ??@A
 #'%)!3&3&	3& 3& 	3&
 3& 3& 6"3& E?3& 66>3& B3&l <<=> !))))	)) )) 	))
 )) )) E?)) ?))X ==" !4-4-4- 
4- 	4-
 4- 
4- 4- 4- 4- 4- $Z4- 4- E?4-
4-n 99* "!""" 
" 	"
 
" " " " " " " " " " "  E?!"
"0 %%  "&*'+"&%)HH	H H 	H
 H H H H H H E?H smH  }H H 6"H
HV &&( "&*'+#,,, 
, 	,
 
, , , , , , , , , , E?,  sm!,"  }#,
,4 ))   %!(,!%!%,S,S	,S ,S 6
	,S
 6",S 6",S 3-,S 3-,S ,S ,S ,S E?,S f%,S v,S #,S
,S^ *** "$("'%474747 
47 	47
 6
47 6"47 6"47 ,,47 ,,47 47 47 47 47 47 47  E?!47" SM#47$  %47
47n ''() $(+/'+ _Y
,,_Y
,,_Y \\_Y \\	_Y
 5<<
 _Y 5<<(_Y $_Y _Y *_YD ##'')<)<)D)DEF&  G&
 t##''( )
   (($*:*:*>*>?@	 	  A	,* 	$$d&D&D&L&LM

 	$$d&D&D&L&LM. ((00//77 !% $%U\\ 123 sELL012 uo	
 uo: 	$$d&D&D&L&LM

 									&$N t))112
  3 t&&'4/ (4/n t$$,,-$% .$%N


 ##T[[%8%89:4 ;4 t!!))* + tyy  !
Q "
Q t,,-LP  .  %%t}}';';<=Xy!K " >K  #("9"9 Q t77??@	* A	* t##++,	0 -	0 t!!))* +> t--556F 7FD %%t~~'@'@AB27u   C 

|E  E4 	!!))4+F+F+N+NO & 55==>?  @8 t>>FFG H$ '')<)<=>"  ?" uyy~~(() *8 uyy~~  !6 "6$ t  !
 
	-K  "-K` t,,- . t++, '
 -'
T t112  3, t>>?  @. tyy8=e 8  8  t~~6c 6 6F 6  6 t''(I& I )I tyy"v "& "  " t||6& 6V 6  6'^ "!+/'+ TKTKTK ell#TK ell#	TK
 6
TK 6
TK 5<<(TK $TK TKn t  "!'+ 6
 6
	
 $   !& ''() $(#'+/'+ <<<< \\ \\	
 5<<
  5<<
  5<<( $  *0 t}}	v 	C 	 	 	  	 t##$3  %3l t~~ $888 8 	8
 8 8  8, t33;;<
 	**&\* c* 	* =*  d22 3 d44 5 d44 5 d== > d== > d== > d== > dDD E dDD E tBB C tBB C tBB C tBB C tJJ K tJJ K tJJ K tJJ K t@@ A tAA B tAA B tAA B tyy  $ t||./    t||./   , 	tyy)!$,,/!$,,/
    BNJ r8   