
    iq                   w   U d dl mZ d dlZd dlZd dlZd dlZd dlZd dlZd dlZd dl	Z	d dl
Z
d dlZd dlmZ d dlmZmZmZmZ d dlmZmZmZmZmZmZmZ d dlmZ d dlmZ d dlZd dl Z d dl!Z d dl"Z d dl#m$c m%Z& d dl'm(Z( d d	l)m*Z* d d
l+m,Z, d dl-m.Z. d dl/m0Z0 d dl1m2Z2m3Z3m4Z4m5Z5m6Z6m7Z7m8Z8m9Z9m:Z:m;Z;m<Z< d dl=m>Z>m?Z? d dl@mAZAmBZBmCZC d dlDmEZE d dlFmGZGmHZHmIZImJZJmKZK ddlLmMZM ddlNmOZOmPZPmQZQmRZR ddlSmTZTmUZU ddlQmVZVmWZWmXZXmYZYmZZZm[Z[m\Z\m]Z]m^Z^m_Z_m`Z`maZambZbmcZcmdZdmeZemfZf ddl$mgZgmhZhmiZimjZjmkZkmlZlmmZmmnZnmoZompZpmqZqmrZr ddlsmtZtmuZu erddlvmwZw  ed      Zx ed      Zy eEddg      Zz ej                  e|      Z}i Z~ded <   i Zd!ed"<    eEe j                  j                            Ze j                  j                  Ze j                  j
                  Ze j                  j                  Z eEe j                  j                            Z eEe j                  j                     e j                  j                  g      Z eEe j                  j                            Zi Zd#ed$<   e j                  j                  Zdld%Z	 	 	 	 dmd&Zdnd'Z	 	 	 	 dod(Zdpd)Z	 	 	 	 dqd*Z	 	 	 	 	 	 drd+Z eej*                  ej,                  ej.                  ej0                  ej2                  ej4                  ej6                  ej8                  ej:                  ej<                  ej>                  ej@                  ejB                  ejD                  g       e jF                  e jH                  e jJ                  e jL                  e jN                  e jP                  e jR                  e jT                  e jV                  e jX                  e jV                  e jZ                  e j\                  d,Zdsd-Zdtd.Zdud/Z	 	 	 	 	 	 dvd0Zd1 Z	 	 	 	 	 	 dwd2Zdxd3Z	 	 	 	 	 	 	 	 	 	 	 	 dyd4Z	 	 	 	 	 	 dzd5Z	 	 	 	 	 	 	 	 	 	 d{d6Zd7e7jt                  d7e~f	 	 	 d|d8Zd9 Zd}d:Z	 	 	 	 	 d~d;Zdd<Z	 d	 	 	 	 	 dd=Z ee j                  j                  d>      d?        Z eej                  d>      dd@       Zd7dAddBZ eej                  j                  d>      ddC       Zd7d7dDddEZ eej                  d>      dddF       ZddGe7jt                  d7ddd7dfdHZdI Z e˫        	 ddJZ eej                  d7dK      dL        Z eej                  d7dK      dM        Z eej                  ej                  ej                  ej                  ej                  g      dN        Z eedO      r  eej                        eԫ        eej                  d>      ddP       Z eej                  d>      ddQ       Z eej                  g      ddR       Z eej                        dS        Z eej                        dT        Z eej                        dU        Z eej                        dV        Z eej                  j                        dW        Z eej                        dX        Z eej                  d>      dY        Z eej                  d>      dZ        Z eej                  d>      d[        Z eej                        d\        Z eej                  d>       eej                  d>       eej                  d>      dd]                     Z eej                  d>      d^        Z eej                  d>      dd_       Z eej*                  d>      dd`       Z eej                  d>      dda       Z eej,                  d>      ddb       ZddcZ eej                  d>      	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 ddd       Zde Z eej                  j                        df        Z eej                  j                        dg        Z eej                  d>      ddh	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 ddi       Z eej                  j                  d>      	 	 	 	 	 	 	 	 	 	 	 	 	 	 ddj       Z eej                  j                  d>      ddh	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 ddk       Z eej                  j                  d>      	 	 	 	 	 	 	 	 	 	 	 	 	 	 ddl       Z eej                  j                  d>      ddh	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 ddm       Z eej                        ddn       Z eej                  d>      dddo       Z eej                  d>      dddp       Z eej                  d>      dddq       Z eej                   d>      dr        Z  eej                  d>      dds       Z eej                  d>      ddt       Z eej                  d>      ddu       Z eej                  d>      dv        Z eej
                  d>      dw        Z eej                  d>      dx        ZddyZ eej                        dd{       Zdd|Z	ej                  d}        Zddd~ZdddZdddZddZd Z ee j                  j"                  j$                  d>      d        Z eej&                  d>      d        Z eej(                  d>      d        Z eej*                  j,                  d>      d        Z eej0                        d        Z ej2                  d      d        Zd Z e	ej8                  j                        Z e	ej8                  j<                        Z e	ej@                  j                        Z! e	ej@                  j<                        Z" eejF                          ee j                  jH                  jJ                  j                          ee j                  jH                  jL                  j                          eej8                        d        Z eej@                        d        Z  eePjN                  d>      d        Z( eePjR                  d>      dd       Z* eePjV                  d>      d        Z, eePjZ                  d>      d        Z. eePj^                  d>      d ddd       Z0 eePjF                  d>      d d	 	 	 	 	 	 	 	 	 dd       Z1ddZ2ddZ3 eejh                  jj                  d>      d7d7ddd	 	 	 	 	 	 	 	 	 	 	 	 	 dd       Z4 eejl                  e7jn                  >      d7d7d	 	 	 	 	 	 	 dd       Z6d Z8d Z9d Z:d Z;d Z<d Z=d Z>d Z? eej                          eej                          eej                          eej                  d7        eej                  j                  d7        eej                          eej                  d7        eej                  d7       e j                  j                         r eej                  d7       e j                  j                         r eej                  d7        eej                          eej                          eej                          eej                  j                          eej                  j                          eej                          eej                  j                          eej                  j                          eej                  j                          eej                          eej                  d7        eej                  e8        eej6                  e>        eej                  e8        eej                  e9        eej                  e9        eej                  e9        eej                          eej                          eej                          eej                          eej                          eej                          eej                  e8        eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                  e9        eej                          eej                          eej                          eej                          eej                  j                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                          eej                           eej                          eej                          eej                          eej                          eej
                          eej                          eej                          eej                          eej                          eej                          eej                          eej                  j                          eej                          eej                          eej                           eej"                          eej$                          eej&                          eej(                          eej*                  j                          eej,                  j                  d7        eej.                  e8        ee j0                  j2                  j4                          ee j0                  j2                  j6                          ee j0                  j2                  j8                          eej:                          eej<                          eej>                  e9        eej@                          eejB                          eejD                          eejF                          eejH                          eejJ                  jL                          eejN                  j                  e?d7        eejP                  j                  e?d7        eejR                  j                  e?d7        eejT                  j                  e?d7        eejV                  j                  e?d7        eejX                  j                  e?d7        eejZ                  j                  e?d7        eej\                  j                  e?d7        eej^                  j                  e?d7        eej`                  j                  e?d7        eejb                  j                  e?        eejd                  j                  e?        eejf                  j                  e?        eejh                  j                  e?        eejj                          eejl                  jj                  dG        eejn                  j                  e9        eejp                  d>      dd       Z eejr                        ddd       Zd Z eed      r  eejv                        e        eejx                        d        Z eejz                  d>      dd       Z eej|                  d>      dd       Zd Z ee j                  ej                  g      dddd7dd       Z ee j                        d}d       Z ee j                        d        Z eej                        d        Z eej                        d        Z eej                        	 ddddd       Zd Zd Zd Z ee j                  ej                  g      dddddddd       Zd Zd Z  eej                         eːeʫ            Z e ed            Z e ed             Zd Z eej                        dddddd       Z eej                        dddddd       Z eej                        dddddd       Z eej                  j                        d        Z ee j                  ej                  g      d        Z eej                  d>      dd       Z eej                  d>      ddÄ       ZdĄ Z	 ddńZdƄ ZڐddǄZ eej                  d>      dȄ        Z eej                  d>      dɄ        Z eej                  d>      ddʄ       Z eej                        dd˄       Zd̄ Zd̈́ Z eej                  d>      dd΄       Z eePj                  d>      ddτ       ZddЄZ e	ej                  j                  d7ѫ      Z e	ej                  j                  d7ѫ      Z eej                  d>      d҄        Z eej                  d>      dӄ        ZedԄ        Z eej                  d>      ddՄ       Z eej                  d>      ddք       ZddGdל	 	 	 	 	 	 	 dd؄Z eej                  d>      ddٜddڄ       Z eej                  d>      ddۄ       Z eej                  d>      dd܄       Z eej                  d>      dd݄       Z eej                  d>      dGdޜdd߄       Z	 	 d	 	 	 	 	 ddZ eej                  j                        ddd       Z eej                  j                        ddd       Z eej@                  j                        	 d}	 	 	 dd       Z eejB                  j                        	 d}	 	 	 dd       Z eej                  j                        	 	 	 d	 	 	 	 	 dd       Z eej                  j                        	 	 	 d	 	 	 	 	 dd       Zd Z eej                  j                        d        Z	 	 	 	 	 	 	 	 ddZ eej                  d>      dd       ZddZddZddZ	 ddZdddZd Z dddZd Z eej                  d>      	 dd       Z	 	 	 	 	 	 	 	 	 	 ddZ eej
                  d>      d        Zd Z eej8                  d>      	 	 	 	 dd       Z eej:                  d>      	 	 	 	 dd       Z e	ej<                  j                  d7ѫ      Z eej<                  d>      d        ZddZd Z	d Z
d Z e	ej                  j                  d7ѫ      Z eej                        d         Z e	ej                  j                  d7ѫ      Z eej                        d        Zd Z eej"                        d        Z eej$                        d        Zd Z eej(                  j                        	 dd       Z e	ej.                  j                  d7ѫ      Z e	ej,                  j                  d7ѫ      Z eej.                  d>      	 	 	 	 	 dd       Z eej,                  d>      	 	 	 	 	 dd       Zd	 Z e	ej0                  j                  d7ѫ      Z eej0                  d>      	 dd
       Z e	ej4                  j                  d7ѫ      Z eej4                  d>      	 dd       Zd ZdddZdddZd Z eej@                        dddd       Z d Z!d Z"d Z#d Z$ eejJ                  ejJ                  g      ddd7dd       Z& eejN                        ddd7dd       Z'd Z(ed        Z) e	ejT                  jV                  d7ѫ      Z, e	ejT                  jZ                  d7ѫ      Z. e	ejT                  j^                  d7ѫ      Z0 eejT                  dG      d        Z*ddZ1 eejd                        d        Z2 eejf                  d>      dd        Z3ed!        Z4ed"        Z5 eejl                  dG      dd#       Z7 eejp                  gdG      d$        Z8dd%Z9 eejl                  gdG      d&        Z: eejv                  ejl                  jj                  gdGe7jx                  K      d'        Z6 eejz                  ejz                  gdG      d(        Z= eej|                  ej|                  g      dddd)       Z? e	ej                  j                        ZA e	ej                  j                        ZC e	ej                  j                        ZE e	ej                  j                        ZG e	ej                  j                        ZI eej                        d}d*       Z@ eej                        d}d+       ZB eej                        d,        ZD eej                  d>      dd-       ZF eej                  d>      dd.       ZH eej                        dddd/       ZJ eej                        dd0       ZL eej                  d>      dd1       ZN eej                  d>      dd2       ZP  eej                         ed3               eej                         ed4            ZS  eej                         ed5            ZU  eej                         ed6e jN                  7            ZW  eej                         ed8e jN                  7            ZY eej                  dGd9:      ZZ e	ej                  j                  d7ѫ      Z[ eej                  j                  d>      ddzd7d;d<       Z\ eej                  j                  d>      dd=       Zqd}d>Z]dd?Z^ e]ej                        Z_ e^ej                        Z` e]ej                        Za e]ej                        Zb eej                        Zc e^ej                        Zd e^ej                        Ze eej                        Zf eej                  dG@      Zg e^ej                          e^ej                          eej                        Zj eej                        Zk eej                        Zl eej                  dAB      Zm eej                        Zn eej                        Zo eej                        Zp e]ej                          e]ej                        Zr  eej                  e7jx                  >      er        e]ej                          e]ej                          e]ej                          e^ej                          eej                  ddGe jZ                  C      Zw eej                  ddGe jZ                  C      Zx eej                  ddGe jZ                  C      Zy eej                  ddGe jZ                  C      Zz eej                        Z{ eej                        Z|  eej                        e{         eej                        e|        eej                        Z eej                        Zj e]ej                         Z eej                          eej                  dDB      Z eej                          eej                  e jZ                  7         eej                        e        eej
                  e jZ                  7        eej                  e jZ                  7        eej                  e jZ                  7        eej                  e jZ                  7      Z eej                  e jZ                  7        eej                  e jZ                  7        e]ej                          e]ej                          e]ej                          e]ej                          e]ej                          e]ej                           e]ej"                          e]ej$                          e]ej&                          e]ej(                          e]ej*                          e]ej,                          e]ej.                          e]ej0                          e]ej2                          e]ej4                         ddElmZmZ dF ZeD ]S  Z eee      D ]  \  ZZZ eʐeeeeG         eee      D ]  \  ZZZ eʐeeeeG        U  eejF                  jH                  eZdG@      Z eejF                  jZ                  eZdG@      Z eejF                  jj                  eZdG@        eejN                  jH                  e8      Z eejN                  jj                  e8        eejN                  jZ                  e8      Z eejT                  jH                  eg        eejT                  jZ                  eg        eejV                  j                  e        eejX                  j                  ej        eejZ                  jZ                  e*        eejZ                  jH                  e*        eejZ                  j\                  e*        eej^                  jH                  e6      Z eej^                  jj                  e6        eej^                  jZ                  e6      Z eejd                  ee        eejf                  e_        eejh                  jH                  e{        eejh                  jZ                  e{        eejj                  jH                  e|        eejj                  jZ                  e|        eejl                  jH                  e{        eejl                  jZ                  e{        eejn                  jH                  e|        eejn                  jZ                  e|        eejp                  e        eejr                  e        eejt                  e      ZdH Z eejz                  jH                  ejF                  jH                  e        eejz                  jZ                  ejF                  jZ                  e        eej|                  jH                  ejN                  jH                  e        eej|                  jZ                  ejN                  jZ                  e        eej~                  jH                  ej^                  jH                  e        eej~                  jZ                  ej^                  jZ                  e        eej                  j                  ejt                  j                  e       dI Z eej                  eZ        eej                  ek        eej                  el        eej                  em        eej                  en        eej                  eo        eej                  ep        eej                  e8        eej                  jj                  e6        eej                  j                  e7        eej                  ew        eej                  ex        eej                  ey        eej                  ez        eej                  eg        eej                  ec        eej                  ed         eej                        ek         eej                        el         eej                        en         eej                        eo         eej                        ep        eej                  ej                          eej                  ej                          eej                  ej                          eej                  ej                          eej                  ej                          eej                        d}dJ       Z eej                  j                        dK        Z eej                  j                        dL        Z eej                        dM        Z e>j                         D ]  \  ZZ  e e?e            e         ee j                        dN        Z eej0                        dO        Z ee j                  j                  j                        dP        Z ee j                  j                  j                        dQ        Z ee j                  j                  j                  j                        dR        Z ee j                  j                  dS      r; ee j                  j                  jf                  j                        dT        Z ee j                  j                  j                        dddU       Zd dVlmZ  ee        ee,      dW        Z ee j                  j                  j                  d>      	 	 ddX       Z ee j                  j                  j                  d>      ddY       Z  ee j                  j                  j                  d>       ej                  edGZ              ee j                  j                  j                  d>      dd[       Zdd\Zd d]lmZ  eed>      d^        Z ee j                  j                  d>      dd_dd`       Z  ee*d>      	 	 	 	 dda       Z ee j                  j                  j                  j                        db        Z ee j                  j                  j                  j                        dc        Z ee j                  j                  j                  d>      dd        ZddelmZ  e         eePj                  d>      df        ZddglNmZ  eMe       ddhlNm	Z	  e	j                           e	j                          ddilNmZ  ej                          ddjlNmZ  ej                          ej                   ddk       Zy(      )annotationsN)defaultdict)Callable
CollectionIterableSequence)AnycastOptionalTYPE_CHECKING	TypeGuardTypeVarUnion)	ParamSpec)patch)counters)associative_scan_op)triton_kernel_wrapper_mutation)FakeScriptObject)get_layout_constraint_tag)canonicalize_dimcanonicalize_dimscheckdtype_to_typeelementwise_dtypesELEMENTWISE_TYPE_PROMOTION_KINDget_computation_dtypeis_boolean_dtypeis_float_dtypeis_integer_dtypeNumber)magic_methodsmethod_to_operator)free_unbacked_symbolshas_free_unbacked_symbolsresolve_unbacked_bindings)
OrderedSet)CeilDivFloorDivIdentityModModularIndexing   )import_submodule   )configinductor_primsirtest_operators)decompositionsget_decompositions)BaseView	DtypeView
ExpandViewIndexingConstantIRNode	is_triton
MutableBoxOnlineSoftmaxReductionops_wrapperPermuteView	Pointwise	ReductionShapeAsConstantBufferSqueezeView	TensorBoxvalidate_irView)ceildivdecode_device
is_dynamicis_gpuis_pointwise_useis_view,needs_fallback_due_to_atomic_add_limitationspad_listlike#register_op_dtype_propagation_rules#register_op_requires_libdevice_fp64sympy_productuse_scatter_fallback)opsV)ReductionType_T_Pztorchvision::roi_alignzaten::index_add8dict[Union[Callable[..., Any], str], Callable[..., Any]]	loweringsz9dict[torch._ops.OpOverload, Optional[Callable[..., Any]]]_maybe_layout_constraintsz2dict[torch._ops.OpOverload, torch._ops.OpOverload]inplaceable_foreach_opsc                     t         j                  j                  j                  D ]7  } | j                  D ]&  }|j                  dk(  r|j
                  t        v r%  y 9 y)Ncall_functionTF)rT   graphcurrent_nodeusersoptargetforeach_ops)nodeusers     R/var/www/html/engine/venv/lib/python3.12/site-packages/torch/_inductor/lowering.pycur_node_has_non_foreach_usersrg      sS    $$** JJ 	DGG.DKK;4N	
     c                f   t        t              }d}t        |       D ]  \  }}t        |t              sd}|f}t        |  xs t        j                  }d }|D ].  }t        |t              s|j                  j                         } n |J d       |r|\  }|||f   j                  ||f        |S )NFTz.foreach op should have at least one tensor arg)r   list	enumerate
isinstancer   rI   r0   #combo_kernel_foreach_dynamic_shapesrD   data
get_deviceappend)	arg_pairsoutunpack_argsiargsuse_foreachdevicets           rf   group_foreach_argsry      s     d
CKY' 54$)K7DD!!OV%O%O 	  	A!Y'**,	 !S#SS!GTV[!"))1d)4!5" Jrh   c                    t        | t        j                  j                        syt	        | d      x}rt        |      S | t        v r	t        |    S y)zHGet layout constraints. Returns None if there are no layout constraints.NF)with_default)rl   torch_ops
OpOverloadr   tag_to_layout_constraintrZ   )fnmaybe_layout_tags     rf   maybe_layout_constraintsr      sO    b%**//04ReLLL'(899	&&(,,rh   c                |   | t         j                  j                  j                  k(  rt        S | t         j                  j                  j
                  k(  rt        S | t         j                  j                  j                  k(  rt        S | t         j                  j                  j                  k(  ry t        d|        )NzUnknown layout constraint tag: )r|   _CTagneeds_exact_stridesconstrain_to_fake_tensorsneeds_contiguous_stridesrequire_contiguous_stridesneeds_fixed_stride_orderconstrain_to_fx_stridesflexible_layoutAssertionError)tags    rf   r   r      s     ehhll...((
ehhll333))
ehhll333&&
ehhll***
:3%@
AArh   c                $    | st        d|       y )Nzinductor does not support NotImplementedErrorcondmsgs     rf   
assert_nyir      s    !$>se"DEE rh   c                    t         t        t        t        t        f      r D cg c]  }t        |       c}S t         t        j                  j                        rt        j                          y t         t        j                  j                        r,t        j                   fd j                         D               y c c}w )Nc              3  6   K   | ]  }t        |        y wN)getattr).0overloadr   s     rf   	<genexpr>z,add_needs_realized_inputs.<locals>.<genexpr>   s      %
&.GB!%
   )rl   rj   settupler'   add_needs_realized_inputsr|   r}   r~   needs_realized_inputsaddOpOverloadPacketupdate	overloads)r   xs   ` rf   r   r      s     "tS%45689)!,99"ejj++,!!"%
 	 
B

33	4$$ %
24,,.%
 	
  :s   Cc                    t        | t        j                  j                        r)| j	                         D ]  }|t
        t        | |      <    y |t
        | <   y r   )rl   r|   r}   r   r   rZ   r   )r   
constraintr   s      rf   add_layout_constraintr      sN     "ejj112 	JH?I%gb(&;<	J )3!"%rh   )r   r/   r-                     	   
         c                b    t        | t              s| S | t        v sJ d|  d       t        |    } | S )Nzid z missing from DTYPE_ID_LOOKUP)rl   intDTYPE_ID_LOOKUPdtypes    rf   decode_dtyper     s=    eS!O#Os5'1N%OO#E"ELrh   c                    t        | t              r4t        | j                               xs t	        | j                               S t        | t
        j                        r| j                  du S t        | t              S NT)	rl   rD   r    	get_dtyper   sympyExpr
is_integerr   r   s    rf   is_integer_typer     sX    !Y.Q2B1;;=2QQ	Auzz	"||t##!S!!rh   c                t    t        | t              rt        | j                               S t        | t              S r   )rl   rD   r   r   boolr   s    rf   is_boolean_typer   (  s*    !Y..!T""rh   c                ^    dd}|D cg c]
  } ||       }}t        |d| i\  }}|S c c}w )Nc                    t        | t        t        j                  f      r| S t	        | j                               }t        j                  dg|z  | j                               S )Nr/   r   )	rl   r!   r   Basiclenget_sizer|   zerosr   )inpdims     rf   construct_inputz+get_promoted_dtype.<locals>.construct_input2  sG    cFEKK01Jclln%C;;sSy@@rh   type_promotion_kind)r   r	   returnr	   )r   )r   ru   r   arginps_r   s          rf   get_promoted_dtyper   /  s@    A -11SOC 1D1!4Q=PQHAuL 2s   *c                0   t        | t        t        f      s| g} nt        |       } t        |       D ]b  }t        |t        j                  j
                        s(|j                         D ](  }t        ||      }|t        vs| j                  |       * d | S r   )
rl   rj   r   r|   r}   r   r   r   rY   rp   )aten_fnr   r   other_fns       rf   get_overloadsr   ?  s    ge}-)w-7m -b%**556LLN -"2x09,NN8,-- Nrh   c                    t        | t        j                  j                        r|| j                  v S t        | t        j                  j
                        r|| j                         v S yNF)rl   r|   r}   r   _qualified_op_namer~   name)ra   	namespaces     rf   in_namespacer   O  sQ     "ejj112B1111	B

--	.BGGI%%rh   c           	     $   t        | j                  t        j                        rt	        | j                               r| S | j                         D cg c]+  }t        j                  j                  j                  |      - }}| j                         }|u|j                  dk(  rf||k7  rat        |      dk(  st        |      dk(  rE|d   dk(  r=t        t        j                  t        j                  j!                  | |d                  S | S c c}w )zB
    Copy cpu scalar if doesn't not match with given `device`
    cpur   r/   F)rl   rn   r2   ReinterpretViewr%   r   rT   r^   sizevarssize_hint_or_throwro   typer   rD   
StorageBox
DeviceCopycreate)r   rw   ssize
cur_devices        rf   maybe_copy_cpu_scalarr   Y  s     affb0015N	

6 <=JJLIqAGG//2IDIJOOu$& Y!^D	Q47a<r}}';';Az5'QRSSH Js   0Dc           	         t               D cg c]  \  }}t        |t              s| }}}j                         D 	cg c]  \  }}	t        |	t              s| }
}}	|s|
s fS |s|r|rt        j
                  nl D cg c]0  }t        |t        t        j                  f      st        |d      r|2 }}|j                  d j                         D               t        |d|i|r |d      n|
d      j                         |D ]  }t         |          |<    |
D ]  }t        |         |<    d	fd} D cg c]
  } ||       c} j                         D 	ci c]  \  }}	| ||	       c}	}|r1t        t!        t#        j$                   fd|D        fd|
D                     }t!        |d   j'                               }t)        ||dt+        |             D ]
  \  }}| |<    t)        |
|t+        |      d       D ]
  \  }}||<    t-        t+                     D ]<  }t         |   t.        j0                        s!t3        j4                   |   |       |<   > D ]<  }t        |   t.        j0                        s!t3        j4                  |   |      |<   >  fS c c}}w c c}	}w c c}w c c}w c c}	}w )
zB
    Transforms arguments for broadcasting and type promotion
    r   c              3  :   K   | ]  }t        |d       s|  yw)r   N)hasattr)r   as     rf   r   z!transform_args.<locals>.<genexpr>  s     !T7@S!!Ts   r   r   c                    t        | t              rt        |       S t        | t        j                        r"t        j                  | j
                        S | S )Nvaluer   rw   )rl   rD   to_dtyper2   Constantr   )r   rw   r   s    rf   promoteztransform_args.<locals>.promote  sD    #y)U++C-{{%OO
rh   c              3  (   K   | ]	  }|     y wr    )r   rt   ru   s     rf   r   z!transform_args.<locals>.<genexpr>  s     3T!W3   c              3  (   K   | ]	  }|     y wr   r   )r   kkwargss     rf   r   z!transform_args.<locals>.<genexpr>  s     71VAY7r   N)r   r	   r   r	   )rk   rl   rD   itemsr|   r   r!   r   r   r   extendvaluesr   ro   r   broadcast_tensorsrj   	itertoolschainr   zipr   ranger2   r   r8   r   )ru   r   	broadcastr   convert_input_to_boolrt   r   args_indicesr   vkwargs_indicesr   promoting_argsr   broadcastedr   rw   r   s   ``              @@rf   transform_argsr
  m  s    #,D/N$!QZ95MANLN$*LLNODAqjI6NaONOV|3 JJE
 a&%++!6771g;N N  !!!TV]]_!TT&$7E &2Da!vnQ>O7P
*, 	  	=A+DGV<DG	=   	AA-fQi@F1I	A	 %))q
),2LLN;DAq!WQZ-;'3l377
 KN++-.k2EC4E&FG 	DAqDG	C4E4G(HI 	DAqF1I	 s4y! 	;A$q'2;;/$++DGT:Q	;  	?A&)R[[1&--fQi>q		? <I OO> *;s(   KKK	K	
5KK4Kc                    t        j                        dfd       }t        |       }t        j	                  |       t
        j	                  t        j                  ||             |S )a  
    Add a foreach lowering to lowerings dict.

    Arguments:
        aten_fn: torch.ops.aten.* fn we are lowering
        decomp_fn: alternate implementation on our IR
        broadcast: True to apply broadcasting to tensor inputs
        type_promotion_kind: kind of type promotion applied to tensor inputs, `None` means no type promotion
        convert_input_to_bool: some logical ops require inputs are converted to bool
    c                 N    t        |       dk  sJ  | i |}t        |       |S )Nr-   )r   rE   )ru   r   rr   	decomp_fns      rf   wrappedz+_register_foreach_lowering.<locals>.wrapped  s/    4yA~~((C
rh   )ru   r	   r   r	   r   r	   )	functoolswrapsr   rc   r   rY   dictfromkeys)r   r  r  aten_fnss    `  rf   _register_foreach_loweringr    sX     __Y   W%Hx T]]8W56Nrh   c                     t        j                         fd       }t                |j                  t        j                   |             |S )a  
    Add a lowering to lowerings dict

    Arguments:
        aten_fn: torch.ops.aten.* fn we are lowering
        decomp_fn: alternate implementation on our IR
        broadcast: True to apply broadcasting to tensor inputs
        type_promotion_kind: kind of type promotion applied to tensor inputs, `None` means no type promotion
        convert_input_to_bool: some logical ops require inputs are converted to bool
    c                 P   t        |       } t        |      }d}t        |       dk(  r)t        | d   t         t        f      rd}t        | d         } t        d D              st        d |D              rJ d       t        | |      \  } }|r| g}  | i |}t        |       |S )NFr/   r   Tc              3  H   K   | ]  }|t         v xs t        |d         yw)_c10d_functionalN)	fallbacksr   )r   r   s     rf   r   z6_register_lowering.<locals>.wrapped.<locals>.<genexpr>  s)      
JLR9_DR1C DD
s    "c              3  &   K   | ]	  }|d k(    yw)rr   Nr   r   r   s     rf   r   z6_register_lowering.<locals>.wrapped.<locals>.<genexpr>  s     6!1:6   zout= ops aren't yet supported)	rj   r  r   rl   r   allanyr
  rE   )	ru   r   unpackedrr   r   r  r  r  r   s	       rf   r  z#_register_lowering.<locals>.wrapped  s    t*!%ft9>ja4-@HQ=D 
PW
 
 6v66W8WW6%&)%8:O
f 6D((C
rh   )r  r  r   r   r  r  )r   r  r  r   r  lowering_dictr  s   `````  rf   _register_loweringr!    sJ    & __Y  6 G$Gw89Nrh   Fc                @    t        j                  t        | ||||      S )z+
    Shim to support decorator syntax.
    )r  r   r  r   )r  partialr!  )r   r  r   r  r   s        rf   register_loweringr$    s)     /3# rh   c                   g }t        j                  t        |       t        |      t        j                  j
                        D ]  \  }}t        j                  j                  j                  |      r|j                  |       Bt        j                  j                  j                  |      r|j                  |       }t        j                  j                  j                  ||       t        t        j                  |      j                        t        t        j                  |      j                        k  r|j                  |       |j                  |        t        t        |            S )z
    Broadcasting logic based on symbolic shapes.

    We give the shapes 0 and 1 concrete values, while all other shapes
    are symbolic sympy formulas.
    )	fillvalue)r   zip_longestreversedr   SOnerT   r^   r   is_size_one_or_falserp   check_equalsr   expandfree_symbolsr   )r   boutputr   ys        rf   broadcast_symbolic_shapesr2  %  s     F%%hqk8A;%''++V 
!177003MM!WW2215MM!GG))!Q/5<<?//03u||A7S7S3TTa a 
! &!""rh   c                V   |	|J d       ||t         j                  }t        d | D              s| S t        d | D              r*|xs t	        | d|ifd}| D cg c]
  } ||       c}S t        d | D              }g }| D ]  }t        |t        t        f      ro|j                  t        j                  t        j                  ||j                         |j                               t!        |j#                                            t        |t$        j&                        rf|j                  t        j                  t)        ||j                         |j                               t!        |j#                                            	|j                  |        |S c c}w )	NzEonly one of override_return_dtype or type_promotion_kind may be givenc              3  f   K   | ])  }t        |t        j                  t        t        f       + y wr   )rl   r   r   r   floatr  s     rf   r   z$promote_constants.<locals>.<genexpr>C  s"     HAz!ekk367H   /1c              3  f   K   | ])  }t        |t        t        t        j                  f       + y wr   )rl   r   r5  r   r   r  s     rf   r   z$promote_constants.<locals>.<genexpr>E  s"     
D:a#uekk23
Dr6  r   c                    t        | t        j                        r!t        j                  | t        d             S t        j                  | t        d             S )Nindexr   rw   r   )rl   r   r   r2   r9   rH   r   r   r   s    rf   
const_funcz%promote_constants.<locals>.const_funcL  sJ    !U[[)**5t1D  {{%d@STTrh   c              3  l   K   | ],  }t        |t        t        t        j                  f      s)| . y wr   )rl   rD   r8   r2   r   r  s     rf   r   z$promote_constants.<locals>.<genexpr>U  s$     WA:a)Z1U#VaWs   *44r   r9  )r   DEFAULTr  r  r   nextrl   r   r5  rp   r8   r   r2   r   r   get_device_or_errorrj   r   r   r   r9   )inputsoverride_return_dtyper   r<  r   exrr   r   s          @rf   promote_constantsrD  ;  s    (,?,G OG $)<)D=EEHHH

DV
DD% 
);*
 !4*
	U (..!
1..	WW	WB
C a#u&JJ!!KKr||~b>T>T>V '	 5;;'JJ!!$r||~b>T>T>V '	 JJqM), J3 /s   F&c                .     d dd fd}|S )Nalphac           	       	
 t        d D              r	rJ   S t              r%| '| dk7  r"t              t        d   |       d<   n| J D cg c]  }|j	                          c}
d   j                         xs d   j                         dd  D ]`  }t        |t        j                        rt              t        |j                               k(  rDJ d d d|j                                  t        j                  t        j                  ft        j                  d uxr{ t!        t        j                  dd       d uxr\ t        j                  j"                  j$                  d uxr4 t        j                  j"                  j$                  j'                  dd	      		xr v 	
f	d
}sSd }D ]7  }t)        |j+                         j,                        s'|j+                         } n |sd   j+                         }xs }t/        j0                  ||      S c c}w )Nc              3  V   K   | ]!  }t        |t              xr t        |       # y wr   rl   r:   r;   )r   r   s     rf   r   z0make_pointwise.<locals>.inner.<locals>.<genexpr>y  s(      /
;>JsF#6	#6/
   ')r/   r   zndim mismatch  r_   low_precision_pointwise_barrierFc                
  	 t        |       t              k(  sJ d|  d        t        j                  k(  r D cg c]
  } ||        c} S g }t              D ]e  \  }} ||       }|   j	                         }	r2|v r.t        j                  ||d      }t        j                  ||      }|j                  |       g  
| }r.t        j                  |d      }t        j                  |      S |S c c}w )Nzwrong ndim rM  F)use_compute_types)r   r|   r   rk   r   rS   r   rp   )r:  loadinputs_loaded	inp_indexrr   	inp_dtypedowncastr   emulate_output_castemulate_precision_castsr   rA  loaders	low_pr_fpoverride_fn_when_input_boolrangess          rf   inner_fnz/make_pointwise.<locals>.inner.<locals>.inner_fn  s
   u:V,LE7!F8.LL,

"'B'N2W4UTT%[4UVV "'0'9 .OItu+C &y 1 ; ; =I.9	3I#&<<YRW#X!ll8Y?!((-. -(&  #||C%PH<<%88
# 5Vs   D rw   r   r\  r[  )r  rD  rj   mulmake_loaderr   r   rl   r2   BaseConstantr   r|   bfloat16float16rT   r^   r   r_   metagetrJ   ro   r   r@   r   )rG  rA  r   otherr\  rw   rt   r   rV  rW  rX  rY  r[  allow_alphar   override_devicerZ  rB  triton_fallbacks    `     @@@@@@rf   innerzmake_pointwise.<locals>.innerx  sL   &3 /
BH/
 ,
 #"?"F++"6+@A UaZf U3r
= =,23q1==?3##%%>)<)<)>ABZ 	AEeR__5V I : At1VHAenn.>-?@A 	A ^^U]]3	GG4 X6dBX$$))5X $$))--.OQVW	 	  6L%9:L	 	, F !,,.--.\\^F	
 --/ !*F	
 	
q 4s   !I)rA  rD   r   )r   rB  rg  rZ  rf  rh  ri  s   `````` rf   make_pointwiserj  p  s     )- M
 M
^ Lrh   c                     ddd fd}|S )Nr/   rF  c                   t        t        j                  j                  j                        dk(  xs6 t        j                  j                  j
                  t        v xs
 t               }d }|D ]  }t        |t        t        f      s|} n |J d       g }|D ]H  }t        |t        t        f      s|j                  |gt        |      z         8|j                  |       J t        t        |       }d gt        |      z  }|j                         D ]  \  \  }}	}
g }|
D ]~  \  }}r	 |d| i}n | }|||<   t        j                  j                  |t         j"                        sJ|	sM|sP|j%                          |j                  |j'                                 |st        j                  j)                  |        t+        d |D              sJ |S )Nr   z1at least one input must be a list to a foreach oprG  c              3  $   K   | ]  }|d u 
 y wr   r   r  s     rf   r   z8make_foreach_pointwise.<locals>.inner.<locals>.<genexpr>   s     2Q1D=2   )r   rT   r^   r_   r`   rb   inplace_foreach_opsrg   rl   rj   r   rp   ry   r  r   has_featureBackendFeatureFOREACHrealizeget_operation_nameregister_operation_listr  )rG  rA  realize_outputsa_list_inputinputbroadcast_inputsgroupsoutputsrw   rv   groupoperation_list
output_indru   r0  rf  pw_fns                  rf   ri  z%make_foreach_pointwise.<locals>.inner  s   $$**+q0 0ww##**.AA0-/ 	  	E%$/$	 ' 	
?	
'
  	/EedE]3 ''#l2C(CD ''.		/ $C)9$:;&3|,,,2LLN 	@(!V[5(*N G "D66F"D\F&,
# GG''0F0FG#'NN$"))&*C*C*EF%G( //?1	@4 2'2222rh   )rA  zlist[list[TensorBox]]r   )r  rf  ri  s   `` rf   make_foreach_pointwiser    s    45 6 6p Lrh   c                    | j                         k(  r|rt        |       S | S fd} t        |      |       S )Nc                4    t        j                  |       S )N)	src_dtype)rS   r   )r   r   r  s    rf   	_to_dtypezto_dtype.<locals>._to_dtype  s    ||Au	::rh   rB  )r   clonerj  )r   r   copyr  r  s    `  @rf   r   r     sE     IEuQx&Q&; B>)5A!DDrh   r   c                   ddl m} |}| j                  } ||t        j                        }t        j
                  |      5   |j                  |  ddd       |j                  }|sJ t        |      }dgt        |      z  }	|j                         D ]  \  \  }
}}g }|D ]k  \  }}||	|<   t        j                  j                  |
t        j                        s:|s=|j                          |j                  |j!                                m |s~t        j                  j#                  |        t%        d |	D              sJ |	S # 1 sw Y   xY w)aI  
    This lowers an invocation of foreach_map
    The way this works is that an arbitrary N-arg func is provided by the user, looped over by the
    polyfill with the same semantics as a foreach op (a loop applying an n-ary function to n args)
    and then traced into a subgraph by dynamo.
    This code allows us to inline the subgraph into the main graph lowering using the PontwiseSubgraphLowering.
    The graph outputs represent the vertically fused sequence of ops, and then register_operation_list
    below registers the buffers as horizontally fuseable in the scheduler.
    r/   )PointwiseSubgraphLowering)root_graph_loweringNc              3  $   K   | ]  }|d u 
 y wr   r   r  s     rf   r   z_foreach_map.<locals>.<genexpr><  s     .q}.rn  )subgraph_loweringr  graph_modulerT   r^   set_graph_handlerrungraph_outputsry   r   r   rp  rq  rr  rs  rp   rt  ru  r  )subgraphru   r   r  rA  gmpw_subgraphsub_outputsrz  r{  rw   rv   r|  r}  r~  r0  s                   rf   _foreach_mapr    sA    =F			B+BAGGLK	
		[	) ! ! ++K;,Ffs;''G(. <$u$& 	C 
"(GJww""6>+A+AB{ %%f&?&?&AB	C GG++N;< .g....N3! !s   E		Ec                @   |j                   s| j                         j                   rk| j                         r/t        | |      }t        j
                  j                  ||        |S  t        t        j                  j                  d      | |      S t        | |d      S )Nr   Fadd_to_fallback_setTr  )
is_complexr   r   
empty_liker2   InplaceCopyFallbackr   fallback_handlerprimsconvert_element_typedefaultr   )r   r   dsts      rf   _convert_element_typer  @  s    1;;=33::< Qe,C""))#q1J#**22  Au4((rh   r  c                  | j                         }||k(  r|rt        |       S | S d } ||      } ||      }||k7  r* t        t        j                  j
                        | |      S t        t        j                  | |            S )Nc                    | j                   rt        j                  |       j                  S t        j                  |       j                  S r   )is_floating_pointr|   finfobitsiinfor   s    rf   _get_primitive_bitwidthz1to_dtype_bitcast.<locals>._get_primitive_bitwidthU  s5    "";;u%***;;u%***rh   )	r   r  r  atenviewr   rD   r7   r   )r   r   r  x_dtyper  src_bitsdst_bitss          rf   to_dtype_bitcastr  P  s~    kkmG%uQx&Q&+ 'w/H&u-H80		0E::))!U344rh   c                &   |j                   s| j                         j                   r`t        j                  t        j
                  j                  t        j                  j                  j                  j                  | |            S t        | |      S r   )r  r   rD   r   r2   ComplexViewr|   rS   r  r  r   r  r;  s     rf   _view_dtyper  d  se    1;;=33NN!!%)).."5"5";";QF
 	
 Au%%rh   r  non_blockingc                   t        |      }| j                         |k(  r|rt        |       S | S t        j                  t
        j                  j	                  | ||            S r   )rH   ro   r  rD   r   r2   r   )r   rw   r  r  s       rf   	to_devicer  m  sO    6"F||~uQx&Q&BMM00FLIJJrh   c                     t        | |d|      S )NTr  )r  )r   rw   r  s      rf   _device_putr  t  s    QTEErh   Tc	                   |xs | j                   }t        |      }	t        |||       |t        |      }t        |	||||      }	 t	        | |||      |	      }	t        t        |      r" t	        t        t        |      d|      |	       |	S )z3A pointwise function that maps ops.{name} to inputsN)rB  rZ  rf  rh  )r  r   r  )r   r  )__name__r>   rO   rj  r$  r   r  r   )
r   r   r  r   r  rB  rZ  rf  rh  r   s
             rf   register_pointwiser  y  s     #7##D	T	B'!#8 #.&12M&N#	
3$?'
B
	/3	

 	

B ud	
E4  $"7	
 		
 Irh   c                 .   d} t        d      fd}fd}t        |      t        |t        j                        gfd} t	        t
        j                        |      }t        t        |       r! t	        t        t        |       d      |       |S )z2A pointwise function that maps ops.frexp to inputsfrexpc                      | i |d   S Nr   r   ru   r   r  s     rf   frexp0zregister_frexp.<locals>.frexp0      d%f%a((rh   c                      | i |d   S Nr/   r   r  s     rf   frexp1zregister_frexp.<locals>.frexp1  r  rh   r  c                 0     d   | i | d   | i |fS Nr   r/   r   )ru   r   pw_fnss     rf   r   zregister_frexp.<locals>.fn  s.    vay$)&)96!9d+Ef+EEErh   Nr  )
r>   rj  r|   int32r$  r  r  r   r  r   )r   r  r  r   r  r  s       @@rf   register_frexpr    s    D E)) 	vvU[[AF
F
	




B ud	
E4  $	
 	 Irh   c                8    t        ||      }t        | |      }|S )Nrf  )r  r  )r   pointwise_lowering_fnrf  r   s       rf   register_foreach_pointwiser    s"    
 
  5;	OB	#GR	0BIrh   )r  r   c           
        d }t        |t        t        f      r t        |      |      }t        |t        t        f      r t        |      |      }| ||g}t	        |d   |d   t
        j                        }t        |      D cg c]  \  }}t        |t              s| }}}t        |t        |D cg c]  }||   	 c}       D ]
  \  }}|||<    t        t        |            D ]Y  }t        ||   t        j                        s!t        j                   ||   t#        ||d      j%                                     ||<   [  t'        ||      |d   t)        |d   |      t)        |d   |            S c c}}w c c}w )Nc                 &    t        j                  |  S r   )rS   whereru   s    rf   r   zwhere.<locals>.fn  s    yy$rh   r/   r-   r  r   r  )rl   r5  r   constant_liker   r   r>  rk   rD   r  r   r  r   r2   r   r8   r   rj   r   rj  r   )	r   r   r/  r   ru   r   rt   r   indicess	            rf   r  r    so     !eS\"M!Q!eS\"M!Q!Q<DQa.M.U.UE 't_ITQ
1i0HqIGIG.'0JQa0JKL 1Q3t9 Td1gr{{+ ''Qd71:6F6O6O6Q1RSDGT ;>"E:Q$q'5)8DGU+C  J0Js   E9 E96E?c                    t        |       dk(  r$t        | d   t        t        f      rt	        | d    S t        j                  t        | D cg c]  }|j                          c}g       }g }| D ]b  }|j                         }t        |      t        |      k7  st        d t        ||      D              rt        ||      }|j                  |       d |S c c}w )Nr/   r   c              3     K   | ]Z  \  }}t         j                  j                  j                  |      t         j                  j                  j                  |      k7   \ y wr   )rT   r^   r   r+  r   r   r/  s      rf   r   z$broadcast_tensors.<locals>.<genexpr>  sP      ,
 1 GG11!4ww44Q78,
s   A A")r   rl   rj   r   r   r  reducer2  r   r  r  r-  rp   )rA  r   rb   r{  sizess        rf   r   r     s    
6{aJvay4-@ &),,(//!&#AQAJJL#A2 F G 	

u:V$ ,
 E6*,
 )

 q&!Aq	 N $Bs   C
c                    | S r   r   r   s    rf   nopr     s    Hrh   
lift_freshc                   t        | t              sJ |(t        t        j                  | j                              S t        |t
        t        j                  f      r)t        j                  j                  j                  |      nt        d |D              }t        t        | j                               |      }t!        t        |t              s|fn|      }g }t#        | j                               D ]X  \  }}||v r>t        j                  j                  j%                  t        j&                  |d            rH|j)                  |       Z || j                         k7  rt+        | |      S | S )Nc              3  n   K   | ]-  }t         j                  j                  j                  |       / y wr   )rT   r^   r   	guard_intr   ds     rf   r   zsqueeze.<locals>.<genexpr>  s%     >Q177##--a0>s   35r/   )rl   rD   rC   r   rn   r   r   r   rT   r^   r   r  r   r   r   r   r'   rk   guard_or_falseEqrp   r  )r   r   dims	new_shaper  r   s         rf   squeezer  	  s   a###
{++AFF344 cC,- 	
""3'>#>> 
 C

-s
3CJsE$:sfDDI!**,'  1T	agg..==ehhq!nMQ 
 "+ajjl!:49AArh   c                ,    t        t        | |            S r   )r  r  )r   r   s     rf   squeeze_copyr     s    C!!rh   c                    t        | |      }t        | t              sJ t        |t              sJ |j                  | _        | S r   )r  rl   rD   rn   r   r   vals      rf   squeeze_r  %  s=    
!S/Ca###c9%%%XXAFHrh   c                    t        |       rt        | dt        j                        S t	        d      } t        |t        j                        |       S )NFr   isinfr  r   	full_liker|   r   r>   rj  r   r   s     rf   r  r  .  ?    qE44	W	B?>"EJJ?BBrh   c                    t        |       rt        | dt        j                        S t	        d      } t        |t        j                        |       S )NFr   isnanr  r  r  s     rf   r  r  6  r  rh   c                f    t        |       rt        |       S t        d      } t        |      |       S )Nceilr   r  r>   rj  r  s     rf   r  r  >  s/    qQx	V	B>"a  rh   c                f    t        |       rt        |       S t        d      } t        |      |       S )Nfloorr  r  s     rf   r  r  F  /    qQx	W	B>"a  rh   c                f    t        |       rt        |       S t        d      } t        |      |       S )Nroundr  r  s     rf   r  r  N  s/    qQx!!~b!!$$rh   c                f    t        |       rt        |       S t        d      } t        |      |       S )Ntruncr  r  s     rf   r  r  W  r  rh   c                   t        | g      \  } t        | t        j                        rt	        j
                  | t        |            S t        | t              sJ t        |t        t        f      sJ t        | j                               t        |      k(  r| S t        | j                               st        j                  j                  j                  t        | j                                     }|dkD  rOt        |      sD| j!                  t        j                  j                  j                  t        |            |z         t        t	        j
                  | j"                  t        |                  S r  )rD  rl   r2   r`  r8   r   r   rD   rj   r   r$   rT   r^   r   r   rQ   
mark_reusern   )r   r  x_size_products      rf   r-  r-  _  s   aS!DQ!R__%  E%L11a###edE]+++QZZ\eEl* .))<<!**,'
 A&;E&BLL  33M%4HI!" Z&&qvvuU|<==rh   c                    t        |      }|D ]  }d||<   	 | }t        |      D ]  \  }}|dk7  st        ||      } t        ||      S NrL  )rj   rk   	unsqueezer-  )r   shapebroadcast_dimensionsr   broadcast_dimensionr  idxr   s           rf   broadcast_in_dimr
  z  sf    UA3 $!#
$ 	
AA, "Q7!S!A" !Urh   c                6    t        | |j                               S r   )r-  r   )r   r1  s     rf   	expand_asr    s    !QZZ\""rh   c                j   t        | j                               t              t              kD  rKt        j                  j
                  gt              t              z
  z  z   t        | t                    } t              t        | j                               k(  sJ t        | j                               }d}t        t                    D ]  }|   dk(  rd}||   |   z  ||<    |r*t        || j                         | j                               S t        d t              D              rt        t        | |            S fd}t              st        |      s{t         j"                  j$                  j'                  t)                    }|dkD  rD| j+                  t         j"                  j$                  j'                  t)        |            |z         | j-                         t/        j0                  | j                         | j                         |t        |            S )NFr   Tr   rw   c              3  :   K   | ]  \  }}|d k(  xs |d k(    ywr/   Nr   r  s      rf   r   zrepeat.<locals>.<genexpr>  s$     
A$!QAFa1f
A   c                   t        |       t              k(  sJ t        |       } t        t                    D ]G  }|   dk7  s|   dk(  rt        j                  j
                  | |<   2t        | |   d|         | |<   I  |       S r  )r   rj   r  r   r)  Zeror,   )r:  rt   old_sizerepeatsx_loaders     rf   r\  zrepeat.<locals>.inner_fn  s    5zS\)))Us7|$ 	IAqzQA;!#$ww||E!H.uQxHQKHE!H	I rh   r]  )rj   r   r   r   r)  r*  r  r  emptyr   ro   r  r  r  r-  r$   rT   r^   r   r   rQ   r  r_  r@   r   )	r   r  new_sizezero_tensorrt   r\  old_size_productr  r  s	    `     @@rf   repeatr    s   AJJL!H
7|c(m#GGKK=CL3x=$@AHLDN#w<3qzz|,,,,AJJL!HK3w<  /1:?KqkGAJ./
 XQ[[]1<<>JJ

A#gx*@
AAVAx())	 !*3H3R77++>>}X?VWa LL  33M(4KL#$
 }}H||~kkmH~	 rh   c                T    t        t        j                  | j                  |            S r   )rD   rF   r   rn   )r   r  s     rf   r  r    s     T[[/00rh   c                    t        | t              sJ t        |t        t        f      sJ t        t	        j
                  | j                  t        |                  S r   )rl   rD   rj   r   r?   r   rn   )r   r  s     rf   permuter    sF    a###dT5M***[''d<==rh   c           
     .   ddl m}m} t        | t              sJ t        | |d      }| j                         |   }t        j                  |      }t        |t        j                        s|dkD  sJ |       	 |dk(  r1t        j                  j                  j                  ||      r|dk(  r| S dd}	d\  }
}|}|r |	||d      }
 |	|||      }|
||
|}}d}|s8t	        t        j                   j#                  | j$                  |||||            S |sJ  |t        j                  j                  j&                  t        j                  j(                  j*                  d         }|J t-        |      d	k  sJ |       d\  }}|j/                         D ]7  \  }}| |d
      t1        j2                  |      fk(  r|})| |d      fk(  s6|}9 |
|J t        j4                  ||||| j                         |         }t        j                  j7                  |      |_        t        j                  j;                  |       |}| j=                         | j?                          |
6|J | jA                         jB                  |
| jE                         |   z  z   }nt        jF                  ||| jA                         jB                  | jE                         |   | j                         |   d      }t        j                  j7                  |      |_        t        j                  j;                  |       |}tI        | j                               }tI        | jE                               }|||<   ||xx   |z  cc<   tK        | |||      S # t        $ r Y w xY w)a	  
    Lowers a slice call, creating ExternKernels for the output size & storage offset symbols,
    if the indices are unbacked and appropriate semantics aren't known.
    If they are known (indices are static/backed/unbacked with info), a SliceView is created.
    r   )CallMethodKeyr&   r/   c                   | |S d }t        j                  |       } t        j                  |      } |t        j                  | d            r |t        j                  | |            r| S  |t        j                  | d            r" |t        j                  | |             r| |z   S  |t        j
                  | |            r|S  |t        j                  | |             ryy )Nc                T    t         j                  j                  j                  |       S r   )rT   r^   r   r  r   s    rf   <lambda>z5slice_.<locals>.compute_slice_index.<locals>.<lambda>  s    qww''66q9 rh   r   )r   r-  GeLeLtGt)r:  r   r  r   s       rf   compute_slice_indexz#slice_.<locals>.compute_slice_index  s    =N9U#||D!ehhua !b%)>&?L"#588ED5+A(B4<%&K&'rh   NNFclampunbacked_bindingsr-   r   storage_offsetTr   )&%torch.fx.experimental.symbolic_shapesr   r&   rl   rD   _validate_dimr   r   r-  r   rT   r^   r   statically_known_leq	TypeErrorr2   	SliceViewr   rn   	shape_envr_   rc  r   r   pytreeSequenceKeyDynamicSliceSizeregister_bufferr   register_operationmaybe_get_layoutrs  
get_layoutoffset
get_strideDynamicSelectStorageOffsetrj   
as_strided)r   r   startendstepr+  r   r&   r   r(  start_index	end_indexambiguous_slicer,  sym_sizesym_storagesymkeypathb_sizer  new_storage_offset	b_storage	new_sizesnew_stridess                           rf   slice_rN    sp   
 a###
3
"C::<D<<DdEJJ'4!89T93QJ  55dC@	H
" (KO)%q9'T48	"y'<$i3E#O LLUCUK
 	
 L51	""AGG$8$8$=$=>Q$R ((( !Q&9(99&&Hk)//1 W}V,f.@.@.EFFH'78::K	 )"333  	

SF ''))&1FKGGv&H#			"""\\^22[1<<>RUCV5VV11LLN!!LLN3JJL
	 00;		""9-(QZZ\"Iq||~&KIcNaK1CDDw  s   75N 	NNc                   d }d }t        | t              rZt        | j                  t        j                        r6| j                         }| j                  }| j                  j                         } | j                          t        j                  |       st        d|  d      t        j                  |       \  }}t        j                  |r|n|j                  |r|n|j                  |D cg c]  }t        j                  |       c}|D cg c]  }t        j                  |       c}t        j                  |xs d            }	t        t        j                   ||	            S c c}w c c}w )Nzunrealized as_strided(z, ...)r   rn   layout)rl   rD   rn   r2   r6   ro   r   unwrap_viewrs  is_storage_and_layoutr   as_storage_and_layoutFixedLayoutrw   r   r-  r   )
r   r   strider-  
new_device	new_dtypestorage
old_layoutr   
new_layouts
             rf   r>  r>  I  s   JI!YJqvvr{{$C \\^
GG	FF IIK##A&!$:1#V"DEE2215GZ 
j&7&7	J$4$4"&'Qa'"()Qa)^(q)J R''WZHII	 	()s   ,E)E.c                d    t        | t              sJ t        | |||      j                  | _        | S r   )rl   rD   r>  rn   )r   r   rV  r-  s       rf   as_strided_r]  f  s/    a###48==AFHrh   c                4    t        | |||      }t        |      S r   )r>  r  )r   r   rV  r-  results        rf   as_strided_copyr`  m  s    48F=rh   c                    g d} D ]1  }j                  |||j                            z   f       d   d   }3  D cg c]  }|j                          c} fd}t         d   j                               }d   d   |<   t	        j
                   d   j                          d   j                         ||      S c c}w )Nr   rL  c           	       
 t        j                  |    t        j                        }g }g }t	        t                    D ]?  

dk(  r$t        j                  dt        j                        n)t        j                  
   d   t        j                        }t        j                  
   d   t        j                        }t        j                  ||      }t        j                  ||      }
dk(  r|}n*
t              dz
  k(  r|}nt        j                  ||      }|j                  |       t        |       t           
   d   z
        <   |j                  t        j                  |
fdd             B |d   }	t	        t              dz
  dd      D ]  
t        j                  |
   |
   |	      }	! |	S )Nr   r/   c                                S r   r   )rt   idx_loadinputs_loaderss   rf   r#  z1pointwise_cat.<locals>.inner_fn.<locals>.<lambda>  s    -N1-h7 rh           rL  r-   )rS   
index_exprr|   int64r  r   constantgeltand_rp   rj   r*   maskedr  )r	  idx_dimmasksmasked_loadsr?  r@  
start_condend_condmasknext_valrt   rd  r   rA  re  inputs_rangess             @@rf   r\  zpointwise_cat.<locals>.inner_fn}  s   ..S5;;7s6{#  	A 6 Q,^^M!$4Q$7E 
 ..q!1!!4ekkBC/Jvvgs+HAvc&kAo%!xx
H5LLCyH %Xc]]15Ea5H%HIHSM

75 	D  #Fq("b1 	AyyaQH	 rh   r]  )rp   r   r_  rj   r@   r   ro   r   )rA  r   prev_endr   r\  r  re  ru  s   ``    @@rf   pointwise_catrw  s  s    9;MH )h3<<>#3F(FGH $R() 4::Ccoo':N.` F1I&&()H!"%b)HSMay##%Qi!!#	 k ;s   Cc           	       	
 t        j                               dk(  sJ d       t        j                               dk(  sJ d       | j                         t        j                  k(  rt        | t        j                        } | j                         t        j                  k(  sJ d| j                                 t        | j                               k  s!J dt        | j                                       | j                         j                         	j                         
	
f	d}t        j                  | j                         || j                               S )Nr/   expect scales 1 dimexpect zero_points 1 dim<Expecting input to have dtype torch.float32, but got dtype: Expecting axis to be < c                b  	 | 
   f} |       } |      } |      }t        t        j                        \  }}j                  t        j                  k7  r$t	        j
                  |t        j                        }j                  t        j                  k7  r$t	        j
                  |t        j                        }t	        j                  |      }t	        j                  ||z        |z   }t	        j                  |t	        j                  ||            }	t	        j
                  |	      S Nr   )_create_constantsr|   float32r   rS   r   r  
reciprocalr  maximumminimum)r	  channel_idxrx  scale
zero_pointqminqmax	inv_scaler  clampedaxisr   input_loader	quant_max	quant_minscalesscales_loaderzero_pointszero_points_loaders             rf   r\  z;quantized_decomposed_quantize_per_channel.<locals>.inner_fn  s    4ylS!k*'4
&y)5==Q
d<<5==(LL6E+j%++>JNN5)	ii	)*Z7++dCKKc$:;||GU++rh   r]  )r   r   r   r|   ra  r   r  r_  r@   r   ro   )rx  r  r  r  r  r  r   r\  r  r  r  s    `````` @@@rf   )quantized_decomposed_quantize_per_channelr    s?    v !Q&=(==&{##%&!+G-GG+ENN*.??- 
FuGXFYZ- #enn&'' 
!#enn&6"7!89' $$&L&&(M$002, ," !~~	 rh   c           	     "     j                          t         t        j                          fd}t	        j
                   j                          j                         |t         j                                     }|j                          |S )Nc                    t         j                  j                         5  t        j                   j                         |             cd d d        S # 1 sw Y   y xY wr   )r2   ComputedBufferforce_realizerS   device_assert_asyncr_  )r:  r   r   s    rf   r\  z_assert_async.<locals>.inner_fn  sO    ,,. 	K**+=4+;+;+=e+DcJ	K 	K 	Ks   *AAr]  )
rs  r   r|   r   r@   r   ro   r   rj   r   )r   r   r\  assertion_ops   ``  rf   _assert_asyncr    sm    LLND%**%DK ## nnDMMO$	L rh   c                    t        | |      S r   r  r   s     rf   lower_assert_asyncr        s##rh   c                    t        | |      S r   r  r   s     rf   lower_assert_functional_asyncr    r  rh   )	out_dtypec               v  	
 t        j                               dk(  sJ d       t        j                               dk(  sJ d       | j                         |k(  sJ d| d| j                                 t        | j                               k  s!J dt        | j                                       t        j                  | j                         	j                         
j                         	
fd}t        j                  | j                         || j                               S )	Nr/   ry  rz  Expecting input to have dtype , but got dtype: r|  c                   |    f} |       } 
|      } |      }	j                   t        j                  k7  r$t        j                  |t        j                        }j                   t        j                  k7  r$t        j                  |t        j                        }t        j
                  t        j                  |t        j                        |      |z  }t        j                  |      }|S r   )r   r|   r  rS   r   sub)r	  r  rx  r  r  r  r  r  r  r  r  r  r  s         rf   r\  z=quantized_decomposed_dequantize_per_channel.<locals>.inner_fn%  s    4ylS!k*'4
<<5==(LL6E-j%--@Jggcll5%--8*EMll3	*
rh   r]  	r   r   r   r|   r  r_  r@   r   ro   )rx  r  r  r  r  r  r   r  r\  r  r  r  s    ```   ` @@@rf   +quantized_decomposed_dequantize_per_channelr    s-    v !Q&=(==&{##%&!+G-GG+??% 
(/@AR@ST% #enn&'' 
!#enn&6"7!89' MM	$$&L&&(M$002  !~~	 rh   c                   | j                         t        j                  k(  rt        | t        j                        } | j                         t        j                  k(  sJ d| j                                 | j                         fd}t        j                  | j                         t        j                  |t        |      t        |            | j                               S )Nr{  c                L    
|       }t        d|z  |t        j                        \  }}t        j                  ||z        |z   }t        t        j                        \  }}t        j
                  t        j                  ||      |      }t        j                  |	      S )N      ?r   )r  r|   r  rS   r  r  r  r   )r	  r  r  rx  r  r  r  r  r  r   r  r  r  s            rf   r\  zBquantized_decomposed_quantize_per_tensor_default.<locals>.inner_fnO  s    S! 1%K5==!
	: ii	)*Z7&y)5==Q
d++ckk#t4d;||GU++rh   r  r  r]  )r   r|   ra  r   r  r_  r@   r   ro   r  r#  r5  r   r   )rx  r  r  r  r  r   r\  r  s      ``` @rf   0quantized_decomposed_quantize_per_tensor_defaultr  <  s     ENN*.??- 
FuGXFYZ- $$&L, !""E%LS_
 ~~ rh   c               l   | j                         |k(  sJ d| d| j                                 t        j                  | j                         fd}t	        j
                  | j                         t        j                  |t        |      t        |            | j                               S )Nr  r  c                     |       }t        ||t        j                        \  }}t        j                  t        j
                  |t        j                        |      |z  }t        j
                  |      }|S r~  )r  r|   r  rS   r  r   )r	  r  r  rx  r  r  r  s        rf   r\  zDquantized_decomposed_dequantize_per_tensor_default.<locals>.inner_fny  s]    S!-eZu}}Uzggcll5%--8*EMll3	*
rh   r  r]  )r   r|   r  r_  r@   r   ro   r  r#  r5  r   r   )	rx  r  r  r  r  r   r  r\  r  s	         ` @rf   2quantized_decomposed_dequantize_per_tensor_defaultr  c  s     ??% 
(/@AR@ST% MM	$$&L !""E%LS_
 ~~ rh   c                0  	 | j                         t        j                  k(  rt        | t        j                        } | j                         t        j                  k(  sJ d| j                                 t        j                               dk(  s9t        j                               dk(  rj                         d   dk(  sJ d       t        j                               dk(  s9t        j                               dk(  rj                         d   dk(  sJ d       | j                         j                         j                         		fd}t        j                  | j                         || j                               S )Nr{  r   r/   expect scale as scalar tensor"expect zero_point as scalar tensorc                    	|       } t        j                               dk(  rdnd      } t        j                               dk(  rdnd      }j                  t        j                  k7  r$t        j                  |t        j                        }j                  t        j                  k7  r$t        j                  |t        j                        }t        j                  |t        j                  |      z        |z   }t        
t        j                        \  }}t        j                  t        j                  ||      |      }t        j                  |      S )Nr/   r   r   r   )r   r   r   r|   r  rS   r   r  r  r  r  r  )r	  rx  _scale_zero_pointr  r  r  r  r   r  r  r  r  scale_loaderr  zero_point_loaders           rf   r\  zAquantized_decomposed_quantize_per_tensor_tensor.<locals>.inner_fn  s    S!c%..*:&;q&@dbI'ENN4D0E0JPRS;;%--'\\&%--8Fu}},,,{EMMBKiiv 667+E&y)5==Q
d++ckk#t4d;||GU++rh   r]  )r   r|   ra  r   r  r   r   r_  r@   r   ro   )
rx  r  r  r  r  r   r\  r  r  r  s
    ````` @@@rf   /quantized_decomposed_quantize_per_tensor_tensorr    sf    ENN*.??- 
FuGXFYZ- u~~ A%ENN"u~~'7':a'?'&'  z""$%*J!"a'J,?,?,A!,D,I,+,  $$&L$$&L"..0, , !~~	 rh   c                 	
 t        j                               dk(  s9t        j                               dk(  rj                         d   dk(  sJ d       t        j                               dk(  s9t        j                               dk(  rj                         d   dk(  sJ d       | j                         |k(  sJ d| d| j                                 t        j                  | j                         j                         	j                         
	
fd}t        j                  | j                         || j                               S )	Nr   r/   r  r  r  r  c                V    |       } t        j                               dk(  rdnd      } 
t        j                               dk(  rdnd      }j                  t        j                  k7  r$t        j                  |t        j                        }	j                  t        j                  k7  r$t        j                  |t        j                        }t        j                  t        j                  |t        j                        |      |z  }t        j                  |      }|S )Nr/   r  r   )r   r   r   r|   r  rS   r   r  )r	  rx  r  r  r  r  r  r  r  r  r  s        rf   r\  zCquantized_decomposed_dequantize_per_tensor_tensor.<locals>.inner_fn  s    S!c%..*:&;q&@dbI'ENN4D0E0JPRS;;%--'\\&%--8Fu}},,,{EMMBKggcll5%--8+FOll3	*
rh   r]  r  )rx  r  r  r  r  r   r  r\  r  r  r  s    ``   ` @@@rf   1quantized_decomposed_dequantize_per_tensor_tensorr    sU    u~~ A%ENN"u~~'7':a'?'&'  z""$%*J!"a'J,?,?,A!,D,I,+,  ??% 
(/@AR@ST% MM	$$&L$$&L"..0
 
 !~~	 rh   c                   | d   j                         j                  dk(  }|rt        d | D              rp| D ]  }|j                           t        d | D              rt	        t
        j                  g|  \  } } t        t
        j                  j                        | |      S t        |       dk(  rt        | d         S t        | d   |d      }t        | dt        j                  i}| D cg c]  }t        ||       } }ddd fd	t!        fd
| D              }dfdt"        j$                  rt'        | |      S |r)t)        t*        j,                  j/                  | |            S fdd}dddt        |       |k  s1t        |       t"        j0                  k  rt        fd| D              rt        fdt2        j4                  j6                  D              }	t!        fd| D              xr |	}
t        fd| D              xr t!        fd| D               }|
s|r|st'        | |      S t)        t*        j,                  j/                  | |            S c c}w )Nr   r   c              3  |   K   | ]4  }|j                         t        j                  t        j                  fv  6 y wr   )r   r|   int8uint8r   rx  s     rf   r   zcat.<locals>.<genexpr>  s.      ;@ejj%++66s   :<c              3  T   K   | ]   }t        |j                               d k(   " yw)r   N)r   r   r  s     rf   r   zcat.<locals>.<genexpr>  s!     >es5>>#$)>s   &(r/   r   c                   t        | t              rJt        | j                  t        j                        r| j                  j                         S | j                  S t        | t        j                        r| j                  S | S r   )rl   rD   rn   r2   r6   rR  r   r   s    rf   unwrap_tensorzcat.<locals>.unwrap_tensor  sV    a#!&&"++.vv))++vva'66Mrh   c                    t        | t        j                        xr$ t        | j                  t        j                        S r   )rl   r2   r  rn   rA   )rx   s    rf   is_reductionzcat.<locals>.is_reduction  s)    !R../TJqvvr||4TTrh   c                    t        | t        t        j                  f      r  |             S  |       xs> t        | t        j                        xr" t        fd| j                         D              S )Nc              3  h   K   | ])  } t         j                  j                  |             + y wr   )rT   r^   
get_buffer)r   readcan_fuse_reductions     rf   r   z2cat.<locals>.can_fuse_reduction.<locals>.<genexpr>  s-       #177#5#5d#;<s   /2)rl   rD   r2   r   r@   r  get_read_names)rx   r  r  r  s    rf   r  zcat.<locals>.can_fuse_reduction  sh    a)R]]34%mA&677O !R\\*  ,,. 	
rh   c              3  .   K   | ]  } |        y wr   r   r   rx   r  s     rf   r   zcat.<locals>.<genexpr>  s     Ba.q1B   c                6   t        j                  |       r:t        j                  | d      \  }}t         j                  j	                  |       S t        | t        t         j                  f      r  |             S t        | t         j                        ryy)NF)freezeT)	r2   rS  rT  ConcatKernelcan_realize_into_without_copyrl   rD   r   r@   )r   rY  r   should_lower_cat_inputr  s      rf   r  z#cat.<locals>.should_lower_cat_input!  sx     ##A&11!EBJGQDDWMMMa)R]]34)-*:;;a&rh   c                H   t        | t        t        j                  f      r  |             S t        | t        j                        sy| j                         j                  }| j                         D ]*  }| t        j                  j                  |            z  }, |S r  )rl   rD   r2   r   r@   inner_fn_opcountnum_opsr  rT   r^   r  )r   countr  op_countr  s      rf   r  zcat.<locals>.op_count8  s    a)R]]34M!,-- !R\\*""$,,$$& 	8DXagg00677E	8 rh   r   r-   c                n    | t         j                  j                  t         j                  j                  fv S r   )r  catr  constant_pad_ndra   s    rf   additional_pointwise_opsz%cat.<locals>.additional_pointwise_opsM  s(    dhh&&(<(<(D(DEEErh   c              3  4   K   | ]  } |      k    y wr   r   )r   rx   MAX_SIMPLE_OP_COUNTr  s     rf   r   zcat.<locals>.<genexpr>R  s     Cq22Cs   c              3  6   K   | ]  }t        |        y wr   )rK   )r   user  s     rf   r   zcat.<locals>.<genexpr>T  s!      
 S":;
r   c              3  .   K   | ]  } |        y wr   r   r   r   r  s     rf   r   zcat.<locals>.<genexpr>[  s     >&s+>r  c              3  .   K   | ]  } |        y wr   r   r  s     rf   r   zcat.<locals>.<genexpr>`  s      "
,/"3'"
r  c              3  .   K   | ]  } |        y wr   r   r  s     rf   r   zcat.<locals>.<genexpr>b  s     <(+<r  )r   zUnion[TensorBox, ir.StorageBox]r   	ir.IRNoder   r   ra   torch._ops.OpOverload)ro   r   r  rs  require_channels_lastr  r  r  r  r   r  r/  r   r   r>  r   r  r0   force_pointwise_catrw  rD   r2   r  r   max_pointwise_cat_inputsrT   r_   r`   )rA  r   
cpu_devicerx  r   r   r   fusable_reductionMAX_COMPLEX_POINTWISE_CATpointwise_usesfuse_pointwise_usehorizontal_fuse_catr  r  r  r  r  r  r  s               @@@@@@@rf   r  r    sI   %%',,5Jc DJ 
  	EMMO	>v>>-dhh@@IFA1 0 01&#>>
6{aVAY
q	3
*C	%D%L%LE /55shsE"5F5
U

 B6BB !!VS)) //<==$ !"F 6{//	V77	7CFCC 
~~++
 
 >v>>Q> 	 " "
39"
 
 =<V<<< 	 "5>O --R__++FC899K 6s   5I1c                  
 | j                         t              t              t              t        k7  fd       t        j
                  j                  j                  t        j                  |d            }|r\t        j
                  j                  j                  t        j
                  j                  j                     |z            d      }n[t        j
                  j                  j                  t        j
                  j                  j                        |z
        d      }d
|r| df
nd|f
t              D cg c]  \  }}|fvs| }}}|j                  |       
fd}	t        t        j                   j#                  | ||	            S c c}}w )N)r	  rankc                     d  d S )Nz(diagonal dimensions cannot be identical z, r   )dim1dim2s   rf   r#  zdiagonal.<locals>.<lambda>q  s     HbQUPVW rh   r   )r   r   c                    | d   }dgt        	      z  }d}t              D ]1  }|k(  r|d   z   ||<   |k(  r|d   z   ||<   %| |   ||<   |dz  }3 |t        	      dz
  k(  sJ |S )NrL  r   r/   r-   )r   r  )
r	  diag_idxoriginal_idxcur_dimr  base_idxr  r  num_dimsoriginal_shapes
        rf   	reindexerzdiagonal.<locals>.reindexer  s    r7sS00x 	ADy"*Xa["8Qd"*Xa["8Q"%g,Q1	 #n-1111rh   )r   r   r   r   rT   r^   r   evaluate_exprr   r&  evaluate_maxevaluate_minrk   rp   rD   r2   GenericViewr   )rx  r;  r  r  offset_negative	diag_sizert   r   r  r  r  r  r   s     ``      @@@rf   diagonalr  i  s   ^^%N>"H84D84D	W gg&&44UXXfa5HIOGG$$11GG))t$v-~d/C 	
	 GG$$11GG))t$nT&:V&C 	
	 HGQ<v;$^4N41a$8MQNEN	LL   R^^**5%CDD' Os    F:/F:c                0    t        t        | |||            S r   )r  r  )rx  r;  r  r  s       rf   diagonal_copyr
    s    %t455rh   c                P    t        |       }t        ||||      }t        ||       |S r   )r  r  	mutate_to)rx  srcr;  r  r  r0  rb   s          rf   diagonal_scatterr    s*    5\FffdD1FfcMrh   c                h   t        j                  |      }t        j                  | j                         |         }d }t        j                  j
                  j                  t        j                  |d            r||z   }n?t        j                  j
                  j                  t        j                  |d            r|}|t        |      rg| j                          | j                         }| j                         }| j                         j                  ||   |z  z   }||= ||= t        | |||      S t        | |||dz   d      }t!        ||      S t#        t        j                  j
                  j$                  t        j                  j&                  j(                  d         }	|	J t+        |	      dk(  sJ |	       t-        t/        |	j1                                     \  }
}| j                          | j                         }| j                         }|
}t3        j4                  |
|| j                         j                  ||   | j                         |   d      }t        j                  j7                  |      |_        t        j                  j;                  |       ||= ||= t        | |||      S )Nr   r/   Fr*  r,  )r   r-  r   rT   r^   r   r  r&  r$  r%   rs  r<  r:  r;  r>  rN  r  r&   r3  r_   rc  r   r?  iterr   r2   r=  r7  r   r8  )r   r   r	  r   actual_indexr  
new_striderJ  slice_resultr,  unbacked_offset_symr   buffers                rf   selectr    sE   
,,s
C<<

S)*DLww&&uxxQ'78Tz	
			(	(#q)9	:$S) IIKzz|HJ!"!6!6C<9W!W3a:7IJJ "!S,q8HPUVL<-- 2	""AGG$8$8$=$=>Q$R ((( !Q&9(99&!$'8'>'>'@"AB IIKzz|HJ,**	3	

SF ''))&1FKGGv&3a:/ABBrh   c           
     r   t        | |d      }|}t        |t        t        f      s`| j	                         |   }t
        j                  j                  j                  t        ||z   dz
  |            }|g|z  }||dz
  |z  z
  |d<   g }d}|D ](  }||z   }	|j                  t        | |||	d             |	}* |S )Nr   r/   rL  Fr*  )r/  rl   rj   r   r   rT   r^   r   r  r)   rp   rN  )
r   r  r   sizes_x_sizechunksr_  r?  r   r@  s
             rf   splitr    s    
3
"CF edE]+c"!!++HVe^a5G,OP6!vzU22r
 FE dl 	fQUCu=> Mrh   c                    t        | ||      S r   )r  )r   r  r   s      rf   split_with_sizesr  	  s    E3rh   c                    t        | |d      }t        j                  j                  j	                  | j                         |         }t        |      D cg c]  }t        | ||       }}|S c c}w r  )r/  rT   r^   r   r  r   r  r  )r   r   r  rt   r_  s        rf   unbindr    s_    
3
"CWW''

S(9:F).v7AfQQ7F7M 8s   A-c                @   | j                         }t        |      }t        ||      |dk(  rt        t	        | d      |d      S |   }t
        j                  j                  }|j                  ||       |j                  d       t        ||z
        dz   }|j                  |      dkD  r-| j                  |j                  t        ||z  |                   g |d  ||dz   d  |}	fd}
t        t        j                   j#                  | |	|
            S )Nr   F)r@  r+  r/   c                D    | d   |    z  z   }g | d  || dz   d S )NrL  r/   r   )r	  dim_idxr   rA  s     rf   r  zunfold.<locals>.reindexer,  s?    b'CHtO+8Tc8G8c#'B&788rh   )r   r   r   rN  r  rT   r^   r   	check_leqcheck_ltr)   r   r  r(   rD   r2   r  r   )r   	dimensionr   rA  r  ndimdim_sizer   new_dim_sizeout_sizer  r   s      `       @rf   unfoldr*    s   JJLEu:D
4
+Cqyi1o4u==SzHwwHtX&aHtOT2Q6L""8,q0	''t0CX(NO	
 EtDlDU379-=DtDH9 R^^**1h	BCCrh   c                    t        | |d      }t        | j                               }|j                  |t        j
                  j                         t        | |      S r  )r/  rj   r   insertr   r)  r*  r  )r   r   r  s      rf   r  r  3  sE    
3
"CQZZ\"IS%''++&9rh   c                    t        | |      }t        | t              sJ t        |t              sJ |j                  | _        | S r   )r  rl   rD   rn   r  s      rf   
unsqueeze_r.  ;  s>    
As
Ca###c9%%%XXAFHrh   c                   t         j                  j                  j                  j	                  t        j                  |            }t        | j                               }|dk  r|||z   z  }d|cxk  r	||z   k  sJ  J |S r  )	rT   r^   r   r3  r  r   sympifyr   r   )r   r   r;  r&  s       rf   r/  r/  D  sr    
''


$
$
2
25==3E
FCqzz|D
Qwtf}#dVm#####Jrh   rL  c                   t        | |d      }t        j                  j                  j	                  | j                         |         dz  }t        | |d|d      }t        | |||dz  d      }t        |t        |            S )Nr   r-   Fr*  )	r/  rT   r^   r   r  r   rN  r^  sigmoid)r   r   new_lenr   r/  s        rf   glur4  M  ss    
3
"Cgg((c):;q@Gq#q'/Aq#w!59Aq'!*rh   c                N     |rt         j                           fd}d|_        |S )Nc                 x    d }t        j                  |t        j                  j                  g| i |      S )Nc                d    t        | t        j                        rt        j                  |       S | S r   rl   r2   r:   rD   r   r   s    rf   wrap_tensorsz7fallback_handler.<locals>.handler.<locals>.wrap_tensors]  %    *4Q		*B9##A&IIrh   )r4  tree_mapr2   FallbackKernelr   )ru   r   r9  kernels      rf   handlerz!fallback_handler.<locals>.handler\  s<    	J "++226KDKFK
 	
rh   T)r  r   _is_fallback_handler)r=  r  r>  s   `  rf   r  r  X  s&    f
 $(G Nrh   c                 .    t        j                  d       y )NzjTorchinductor does not support code generation for complex operators. Performance may be worse than eager.)warningswarnr   rh   rf   _warn_complex_not_supportedrC  j  s    MMtrh   c                   | j                         rt                y| j                  ry| j                  ry| j                  t
        j                  k(  r|syt        |j                  t
        j                  j                        xrr |j                  t        j                  j                  t        j                  j                  t        j                  j                  t        j                   j                  fv xsE t        |j                  t
        j                  j                        xr t#        |j                         S y)z0Do not support reading or writing to this tensorTF)r  rC  is_meta	is_sparser   r|   float8_e8m0fnurl   rb   r}   r~   r  r  r  r  r  
_scaled_mmrL   )rx   rd   s     rf   unsupported_input_tensorrI  s  s    ||~#%yy{{ww%&&&
 t{{EJJ$9$9: 		  

""''	Y 4;;

(=(=>W74;;CW

 
	
 rh   c                *   t         j                  j                  t        j                  j
                  j                  j                  f}||j                  |v r| j                         ryt        | |      ry| j                  xr t        j                  S )z2Do not support writing tensor but can read from itFT)r  r  r   r|   rS   r  r  r  rb   r  rI  is_cpur0   disable_cpp_codegen)rx   rd   supported_complex_viewss      rf   unsupported_output_tensorrN    so     					,,44 DKK+BBq||~4(882222rh   c                R     j                   t        j                  j                  u ry j                  dk(  ry j                   t        j
                  j                  u ry fd}t        j                   j                  i  j                  D ]  } ||d      s y  | d      S )NFplaceholderc                D   t        | t        j                  j                        syd| j                  vryt        j                  | j                  d         D ]F  }t        |t        j                  j                        s(|rt        |      s7 yt        |      sF y y)NFr  T)rl   r|   fxNoderc  r4  tree_leaves_subclasses
FakeTensorrN  rI  )inp_out_node	is_outputrc  rd   s      rf   check_skip_conditionzCfallback_node_due_to_unsupported_type.<locals>.check_skip_condition  s    ,6)))&&|'8'8'?@ 		 DdE$5$5$@$@A,T48+D$7		  rh   )rX  T)
rb   r  view_as_complexr  ra   lift_fresh_copyr4  arg_tree_leavesru   r   )rd   allow_cpu_inputsrY  r   s   `   rf   %fallback_node_due_to_unsupported_typer^    s    {{d**222ww- {{d**222* %%tyy@DKK@ u5  55rh   c                *   | t         vs|s
J d|         |rt        t        j                  d            rt	        | g      rt
        j                  r&| t        j                  j                  j                  v si|sgt        j                  j
                  j                  r4dt        j                  j
                  _        t        j                  d       t        d|  d      fd}t!        | t        j"                  j$                        r*| j'                         D ]  }t)        | |      } ||        y t!        | t        j"                  j*                  t        j"                  j,                  f      r	 ||        y t/        d|  d	t1        |              )
Nz*both a fallback and a decomp for same op: CIFznA make_fallback error occurred in suppress_errors config, and suppress_errors is being disabled to surface it.zmake_fallback(a.  ): a decomposition exists, we should switch to it. To fix this error, either add a decomposition to core_aten_decompositions (preferred) or inductor_decompositions, and delete the corresponding `make_fallback` line. Get help from the inductor team if unsure, don't pick arbitrarily to unblock yourself.c                n    t        |        t        |         t        | d       t        |             S Nr  )r   r   r$  r  )op_overloadlayout_constraints    rf   register_fallbackz(make_fallback.<locals>.register_fallback  s;    !+.(!+/@AG $G[)
 	
rh   zUnsupported fallback z with type )r4   r   osgetenvr5   r0   fallback_randomr|   _decompdecompositions_for_rngextra_random_decomps_dynamosuppress_errorslogwarningr   rl   r}   r   r   r   r~   HigherOrderOperatorRuntimeErrorr   )ra   rd  rB  override_decompre  olrc  s    `     rf   make_fallbackrt    sQ   ^# 
4RD96 	4!t$ ""emm::OOO ==//38EMM  0KKH RD !f f
 	

 "ejj112,,. 	+B!"b/Kk*	+ 
B..

0N0NO	P"22$k$r(LMMrh   c                T    d}| D ]  }||z  }	 t        |t        j                        S )z
    TorchInductor offset calculation differs from PyTorch eager offset
    calculation for random ops (tl.rand vs torch.rand). In future, we should
    strive for same impl for tl.rand and torch.rand.
    r/   r   tensorr|   rh  )r  numelr   s      rf   philox_rand_offsetry    s2     E 	%u{{++rh   c                B  	
 t        j                  || t         j                  j                  |             j	                         
|j                         |j                         		
fd}t        j                  ||t        |             }t        |       }||fS )Nc                   t        j                   g       t        j                        }t        j                   g       t        j                        }t        j                  t        j
                   |       t        j                        |      }t        j                  ||      }t        j                  |      S r   )rS   r   r|   r  r   rg  rand)	r:  seed_index_exproffset_index_exprrand_index_exprr_  r   offset_loader
random_posseed_loaders	        rf   r\  zphilox_rand.<locals>.inner_fn	  s     ,,{2DLLr):EKKH''NN:e,ekk:<M
 
 ||FE**rh   r]  )
r2   rU  FlexibleLayoutcontiguous_stridesmake_indexerr_  r@   r   rj   ry  )r   seedr;  rV  rw   r   r\  random_values_nodeoffset_noder  r  r  s        `   @@@rf   philox_randr  	  s     
,,T2	
 ln  ""$K&&(M+ #))Dz	 %T*K{**rh   c           	         t         j                  r\t        j                  t        j
                  t        j                  j                  t        j                  j                  | ||            S t        d      )Nz&should be handled in replace_random.py)r0   rh  r4  r;  rD   r   r2   r<  r  native_dropoutr  r   )r   ptrains      rf   r  r  /	  sY    $$T%8%8%@%@!QN
 	

 EFFrh   c                x   t         j                  s-| j                         t        j                  d      k(  sJ d       | j                          t        |      dk(  st        |d   t              rt        j                  j                  nt        j                  j                  }t        j                  || g|  | S )Nr   Tthis should be handled in decomps unless config.fallback_random or the device is CPUr   )r0   rh  ro   r|   rw   rs  r   rl   r5  r  
bernoulli_Tensorr2   InplaceBernoulliFallback)r   ru   rc  s      rf   r  r  :	  s    !!Q\\^u||E7J%J ^J IIK t9>ZQ7 	__## 
 Q66Hrh   c                    t         j                  s-| j                         t        j                  d      k(  sJ d       t        t        |       g| S )Nr   r  )r0   rh  ro   r|   rw   r  r  )r   ru   s     rf   bernoulli_pr  I	  sG    !!Q\\^u||E7J%J ^J eAh&&&rh   c                    t         r   r   )r   s    rf   _foobarr  R	  s    
rh   c                .    t         j                  d       y )Nz1using triton random, expect difference from eager)rn  info)salts    rf   _warn_triton_randomr  W	  s    HH@Arh   c                 J    t        t        j                  j                         y r   )r  rT   r^   creation_timer   rh   rf   warn_triton_randomr  \	  s    --.rh   c                     |j                  d      t        | i |S t        j                  r|j	                  dd        t        | i |S t        d      N	generatorz-should have been handled in replace_random.py)rd  fallback_rand_generatorr0   rh  popfallback_rand_defaultr   ru   r   s     rf   r|  r|  l	  sT    zz+*&777			

;%$d5f55
H
IIrh   c                     |j                  d      t        | i |S t        j                  r|j	                  dd        t        | i |S t        d      r  )rd  fallback_randn_generatorr0   rh  r  fallback_randn_defaultr   r  s     rf   randnr  v	  sT    zz+*'888			

;%%t6v66
H
IIrh   c                l    t        j                  |      }t         j                  j                  | |      S r   )r2   get_stride_orderExternKernelrequire_stride_order)input_tensorrV  stride_orders      rf   inductor_force_stride_orderr  	  s)    &&v.L??//lKKrh   c                    t        d      )Nz.should be handled in fuse_seed_creation_pass()r  )rw   s    rf   inductor_seedr  	  s    
I
JJrh   c                z    t                t        j                  t        j                  | t        |                  S r   )r  rD   r   r2   RandomSeedsrH   )r  rw   s     rf   inductor_seedsr  	  s)    BNN5-2GHIIrh   c                |      fd}t        j                   j                          j                         |g       S )Nc                L    t        j                  j                               S r   )rS   	load_seedget_name)r   r:  seedss    rf   r\  z&inductor_lookup_seed.<locals>.inner_fn	  s    }}U^^-u55rh   r]  )r@   r   ro   r   )r  r:  r\  s   `` rf   inductor_lookup_seedr  	  s9    6 !oo	 rh   r;  c                 	 t         j                  rJ dv sJ g | } t        j                  }|j	                         }t        j                  ||| t
        j                  j                  |       |      j                         |j                         		fd}t        j                  |||g |       }|j                          |S )N)r|  r  r  c                     t        t               g       t        j                   |       t        j                              S r   )r   rS   rg  r|   r  )r:  moder  r  s    rf   r\  z!inductor_random.<locals>.inner_fn	  s7    !wsD!ONN:e,ekk:
 	
rh   r]  )r0   rh  r|   r  r@  r2   rU  r  r  r  r_  r@   r   rs  )
r   r  r  r;  r   rw   r\  r_  r  r  s
     `     @@rf   inductor_randomr  	  s    %%%%$$$$T7DMME%%'FtR..AA$GPVln  ""$K
 w	F NNMrh   c               f   	 t         j                  rJ g |}t        j                  }|j	                         }t        j                  |||t
        j                  j                  |      |      j                         |j                         	 	fd}t        j                  |||g |      S )Nr  c           	        t        j                   g       t        j                   |       t        j                        t        j                  t        j
                        t        j                  t        j
                              S r   )rS   	randint64rg  r|   r  rh  )r:  highlowr  r  s    rf   r\  z"inductor_randint.<locals>.inner_fn	  sV    }}ONN:e,ekk:NN3,NN4-	
 	
rh   r]  )r0   rh  r|   rh  r@  r2   rU  r  r  r  r_  r@   r   )
r  r  r   r  r;  r   rw   r\  r  r  s
   ``      @@rf   inductor_randintr  	  s     %%%%T7DKKE%%'FtR..AA$GPVln  ""$K
 w	 rh   c                    | j                         }| j                         }t        d t        ||      D              |d   z   }| j	                         |d   ||d   fS )Nc              3  2   K   | ]  \  }}|d z
  |z    ywr  r   )r   r   sts      rf   r   z%_boundaries_helper.<locals>.<genexpr>	  s     Aeaa!er\As   rL  )r   r<  sumr  r  )tbr   rV  
max_offsets       rf   _boundaries_helperr  	  s^    
 ;;=D]]_FAs4/@AAF2JNJ
Rr
	 rh   c                H    | j                         | j                         d   fS r  )r  r<  r  s    rf   _sorter_helperr  	  s    ;;="--/"---rh   	out_int32rightsidesorterc               J   
 d } |       r ||      r
9 |      s1 t        t        j                  j                  d       |||      S ||dk(  rd|rt        j
                  nt        j                  
|j                          j                          j                          t         j                               dk(  r

 fd}n	
 fd	}|j                         }t        j                  |
||j                  
      }	|	j                          |	S )Nc                ^    t         j                  j                  | t        j                        S r   )rT   r^   rp  rq  	BUCKETIZEr  s    rf   r#  zsearchsorted.<locals>.<lambda>	  s     AGG$7$7
N$$% rh   Fr  r  r  Tr/   c           
          |       }t        j                  |t              dd n
t              d       S d      S )Nr   r  sorter_indicesrS   	bucketizer  r  )r	  r  index_dtyper  sorted_sequencer  values_loaders     rf   r\  zsearchsorted.<locals>.inner_fn
  sV    $C=="?3%~t>&3I'-~t  <= rh   c                             }d fd}t        j                  |t               |      d n
t              d       S  |            S )Nc                    | j                         }t        j                  t        j                  t
        j                  d t        |d d d d       D                    S )Nc              3  ,   K   | ]  \  }}||z    y wr   r   )r   r   rt   s      rf   r   zNsearchsorted.<locals>.inner_fn.<locals>.get_flattened_index.<locals>.<genexpr>1
  s     &UAq1u&U   rL  )r<  rS   rg  r  r  operatorr   r  )r  stridesr	  r  s     rf   get_flattened_indexz;searchsorted.<locals>.inner_fn.<locals>.get_flattened_index-
  sU    --/~~$$ &UWSb\3sPR89T&U  	 rh   r  )r  rD   r  )r	  r  r  r  r  r  r  r  s   `  rf   r\  zsearchsorted.<locals>.inner_fn(
  sh    $C =="?3#O4%~t>&3I'-~t  <Ov;V rh   r]  )r  r  searchsortedr  r|   r  rh  r_  rs  r   r   ro   r@   r   r  )r  selfr  r  r  r  validate_bucketizer\  rw   r_  r  r  s   `  ` `    @@rf   r  r  	  s    /!$''9&'AT 1 1 8 8eT
 	
 DGO!*%++K$$&M 
?##%&!+
	 
		 	0 __Fzz	F NNMrh   r  r  c                  t        j                               dk(  sJ t        j                  j	                  | t
        j                        r.t        j                  j	                  t
        j                        s/ t        t        j                  j                  d      | |      S j                          | j                         }| j                         |rt        j                  nt        j                   fd}t#        j$                  ||| j                               }|j                          |S )Nr/   Fr  r  c                \     |       }t        j                  |t              d      }|S r  )rS   r  r  )r:  r  r  
boundariesr  r  r  s      rf   r\  zbucketize.<locals>.inner_fnk
  s6    5!--z*
 rh   r]  )r   r   rT   r^   rp  rq  r  r  r  r  r  rs  ro   r_  r|   r  rh  r@   r   )	rx  r  r  r  rw   r\  r_  r  r  s	    ` `   @@rf   r  r  M
  s     z""$%*** 	
E>#;#;<GG
N,D,DEQ 5 55Q:%
 	
 F$$&L!*%++K
 ~~	F NNMrh   c                    t        j                  t        j                  t        j                  j
                  ||f      \  }}||fS r   )r4  tree_map_onlyr2   r:   r  require_stride1r   ru   r   s      rf   require_denser  
  s;    ''
		2??22T6NLD& <rh   c                    t        j                  t        j                  t        j                  j
                  ||f      \  }}||fS r   )r4  r  r2   r:   r  require_contiguousr  s      rf   r  r  
  s;    ''
		2??55f~LD& <rh   c                    t        j                  t        j                  t        j                  j
                  ||f      \  }}||fS r   )r4  r  r2   r:   r  r   r  s      rf   r   r   
  s=     ''
		2??==f~LD& <rh   c                    t        j                  t        j                  t        j                  j
                  ||f      \  }}||fS r   )r4  r  r2   r:   r  r  r  s      rf   r  r  
  s;    ''
		2??884.LD& <rh   c           	     &   || S t        |t              r| S t        | t        j                        rm|j	                         D cg c]4  }t        |t
        j                        r|j                  j                  n|6 }}t        j                  j                  | |      S t        | t              r!| D ci c]  }|t        | |   ||          c}S t        | t        t        f      r" t        |       d t!        | |      D              S | S c c}w c c}w )Nc              3  :   K   | ]  \  }}t        ||        y wr   constrain_to_fake_tensor)r   r   f_as      rf   r   z+constrain_to_fake_tensor.<locals>.<genexpr>
  s!      
19!S$Q,
r  )rl   r   r2   r:   rV  r|   SymIntrd   exprr  require_exact_stridesr  r  r   rj   r   r  )r   fake_argr   meta_stride_exprkeys        rf   r  r  
  s    
(,-
#ryy!GOGX
BC:a6AFFKKA=
 
 44S:JKK#tRUV3-c#hFFVV	C%	'tCy 
=@h=O
 
 	
 J

 Ws   9D	2Dc           
         t        d t        | |      D              } |j                         D ci c]  \  }}|t        |||          }}}| |fS c c}}w )Nc              3  :   K   | ]  \  }}t        ||        y wr   r  )r   r   r   s      rf   r   z,constrain_to_fake_tensors.<locals>.<genexpr>
  s#      C 	!h/r  )r   r  r   r  )ru   r   	fake_argsfake_kwargsr   r  s         rf   r   r   
  sb      y1 D JPXAa)![^<<XFX< Ys   Ac           
         fdt        fdt        || j                        D              }|j                         D ci c]  \  }}| || j                  |          }}}||fS c c}}w )Nc           	        t        | t        j                        rst        j                  |j                  d   j                         t        j                  j                  j                        }t        j                  j                  | |      S t        | t              r| D ci c]  }| | |   ||          c}S | S c c}w Nr  )rl   r2   r:   r  rc  rV  rT   r^   r   r3  r  r  r  )r   fx_argr  r  apply_constraints       rf   r  z1constrain_to_fx_strides.<locals>.apply_constraint
  s    c299%..E"))+QWW-=-=-G-GL ??77\JJc4 LOPSC)#c(F3K@@PP
 Qs   #B?c              3  6   K   | ]  \  }} ||        y wr   r   )r   r   r
  r  s      rf   r   z*constrain_to_fx_strides.<locals>.<genexpr>
  s"      *5#vf%r   )r   r  ru   r   r   fx_noderu   r   r   r  r  s        @rf   r   r   
  ss      9<T7<<9P D EKLLNSDAqa!!W^^A%677SFS< Ts   !A*c                      fdt        fdt        t        | j                              D              }|j	                         D ci c]  \  }}| d| j
                  |          }}}||fS c c}}w )Nc           
     
   t        |t        j                        s|S |j                  d   }|j	                         D cg c]4  }t        |t
        j                        r|j                  j                  n|6 }}t        j                  j                  j                  }t        j                  |j	                         |      }|r<|d   dk7  r4t        t        t!        t#        |j%                                                 }j&                  t(        j*                  j,                  k(  r| dv rt#        |      dk(  sJ d}|j.                  s t        j0                  j3                  ||      S dj&                  t
        j4                  j(                  j6                  j,                  k(  xr | dk(  }t        |t8              sJ t#        |j%                               d	vr|S t        j:                  |      }	|	r3t        j<                  t        j0                  j?                  |      |      S t        |t              rE|jA                         5|	r3t        j<                  t        j0                  j?                  |      |      S |rt        |j%                               }
g }|jA                         }t!        t#        |j%                               d
z
        D ]q  }t        j                  j                  jC                  ||   d      s1|3t        j                  j                  jC                  ||   d      sa|jE                  |       s dgt#        |
      z  }d
|d<   d
}t!        t#        |
      dz
  dd      D ]  }||d
z      dk7  r||
|d
z      z  }||v r9t        j                  j                  jC                  ||d
z      z  d      rd||<   Vt        j                  j                  jC                  |z  d      stG        |      z  }|||<    t        j0                  jI                  ||      S |	r3t        j<                  t        j0                  j?                  |      |      S t        |t              rE|jA                         5|	r3t        j<                  t        j0                  j?                  |      |      S fd}t        |jJ                  t        jL                        rQ ||      sI ||jO                               r3t        j<                  t        j0                  j?                  |      |      S t        j0                  j3                  ||      S c c}w )Nr  rL  r   )r   r   r   )r   r/   r-   r   r   r   r   r   r/   r-   c                    t         j                  j                  j                  t	        j
                  t        | j                         d         d            S )NrL  r   )rT   r^   r   r  r   r  r+   r   )r   	ALIGNMENTs    rf   
is_alignedz=sdpa_constraint.<locals>.apply_constraint.<locals>.is_alignedI  s@    77##22QZZ\"-y91= rh   )(rl   r2   r:   rc  rV  r|   r  rd   r  rT   r^   r   r3  r  rj   r(  r  r   r   rb   r  0_scaled_dot_product_efficient_attention_backwardr  is_cudar  r  rS   '_scaled_dot_product_efficient_attentionrD   is_aligned_realized_tensortry_match_insignificant_stridesrealize_inputmaybe_get_stridestatically_known_equalsrp   rG   r  rn   r6   rR  )r	  r   r
  meta_valr   r  r3  r  effn_attn_fwd_biasis_aligned_tensorr)  expanded_dimsmaybe_stridert   out_stridesrV  r  r  r  s                    @rf   r  z)sdpa_constraint.<locals>.apply_constraint
  s   #ryy)J;;u%GOGX
BC:a6AFFKKA=
 
 GG$$..	**8??+<iHL,1s3<<>/B)C DEL NNDDLLMv|$)))
 (L??77\JJ 	 NNyy~~EEMMN q 	 #y)))s||~f,J99#yI55--c24D 
 sF#$$&2!55--c24D  CLLN+HM //1L3s||~.23 ,77##;;<LQ<OQRS ,((@@aRST!((+, $X.KKOF3x=1,b"5 (q1u%*#hq1uo5F %ww''??#AE*Y6 *+A ww''??@RTUV$VY7)CF!'A!($ ??88kJJ55--c24D 
 sF#$$&2!55--c24D 	
 chh,c?coo/0==55c:<L  33CFFo
s    9Uc              3  >   K   | ]  \  }\  }} |||        y wr   r   )r   r	  r   r
  r  s       rf   r   z"sdpa_constraint.<locals>.<genexpr>W  s*      C#v 	c6*s   rL  )r   rk   r  ru   r   r   r  s   `    @rf   sdpa_constraintr$  
  s}    |G|  "+Cgll,C"D D IOW1a!"a):;;WFW< Xs   "A5)rB  )rr  c                   t        |t        j                        s*t        || j	                         | j                               }|}| j                         |j                         k7  rt        || j                               }| j	                         |j	                         k7  rt        || j	                               }| j                         |j                         k7  r%t        || j                               }t        |      S t        |      S Nr  )rl   r2   r:   rw  r   ro   r  r   r   r-  r  )r  r  r  r   rr   s        rf   r  r  :  s    c299%S 09JKACNN,,a*+~~3==?*Q()}}#,,.(Q(Sz8Orh   )memory_formatc          	         t        j                  | j                         | j                         | j	                         t        | j                                     S Nr]  )r@   r   ro   r   r_  rj   r   )r   r'  s     rf   r  r  L  s@     ||~kkmAJJL!	 rh   c                   g }t        | t              rt        | j                  t        j                        rw| j                  } t        | t        j                        rF|j                  | j                                | j                  } t        | t        j                        rFt        |       } t        |       } |r;| j                  } |d d d   D ]  }t        j                  | |      }  t        |       } | S )NrL  rP  )rl   rD   rn   r2   r   rp   r:  r  )r   reinterpret_view_layoutsrQ  s      rf   clone_preserve_reinterpret_viewr,  W  s    !!YJqvvr7I7I$JFFB../$++ALLN;A B../ aLaAFF.tt4 	:F""&9A	:aLHrh   r[  c               \    fd}t        j                  t        |      || g      S )Nc                D    t        j                  | d   z  z         S )Nr   r   rS   rg  )r:  r   r?  rA  s    rf   r   ziota.<locals>.fnz  s!    ~~dU1Xo5UCCrh   r]  )r@   r   rH   )lengthr?  rA  r   rw   requires_gradr   s    ```   rf   iotar2  p  s1    D V$x	 rh   c           	        | j                         |j                         k(  sJ | j                         t        | d      t        j                  j
                  j                  t        j                  d            r| j                            z   njt        j                  j
                  j                  t        j                  d            rn, t        t        j                  j                        | |      S t        j                  j
                  j                  d       t        j                  j
                  j!                  | j                                   t#        t%        |      | j                               }|j                         fd}t'        j(                  | j+                         | j                         |t-        | j                                     S )Nr   c           	         t        j                  t        j                  t        j                  |    t        j
                        t        j                  t        j
                               |        |             S r   )rS   r  eqrg  r|   r  )r	  r   r:  
src_loaderr  s    rf   r\  z select_scatter.<locals>.inner_fn  sX    yyFFs3x5uekk2 sOSM
 	
rh   r]  )r   r_  r/  rT   r^   r   r  r   r&  r   r$  r  r  select_scatterr  r#  r$  r-  r  r@   r   ro   rj   )r   r  r   r:  r\  r6  r  s     `` @@rf   r7  r7    s\   ;;=CMMO+++}}H
3
"Cww&&uxxq'9:

S))	
			(	(%);	< = 3 3 ; ;<QS%PPGGq%(GGeQZZ\#%67
3$ajjl
3C"J
 ||~kkmAJJL!	 rh   c           	     :   	
 t        | j                               } j                         
t         d       j	                            t
        j                  j                         \  t         j	                               }t        z
  dz
  z         |<   t        ||      }|j                         		 
fd}t        j                   j                          j                         |t         j	                                     S )Nr   r/   c                   	dk(  rk(  r
dk(  r |       S t        j                  |    t        j                        }t	        |       t        |    	z
  
      <   g }	dk7  rZ|j                  t        j                  |t        j                  t        j                  	      t        j                                     k7  rZ|j                  t        j                  |t        j                  t        j                        t        j                                     
dk7  rz|j                  t        j                  t        j                  t        |    	z
  d
      t        j                        t        j                  dt        j                                     |sJ t        j                  t         j                   |      }t        j"                  |fdt%              rdnd      }t        j&                  || |             S )Nr   r/   c                             S r   r   )src_idxr6  s   rf   r#  z1slice_scatter.<locals>.inner_fn.<locals>.<lambda>  s    Jw' rh   rf  )rS   rg  r|   rh  rj   r)   rp   rj  r   r-  rk  r5  r,   ri  r  r  rl  rm  r   r  )r	  rn  rs  src_valr;  r   r'  r@  r6  r?  rA  r   r  s       @rf   r\  zslice_scatter.<locals>.inner_fn  s   A:#/daic?"..S5;;7s)C5 0$7A:KKNN5<<#6D (?KKNN5<<#4ekkB 19KKNN'C5(8!TBEKK LLEKK0	 t$/**' #A

 yySM
 	
rh   r]  )r   r   r_  r/  r   r2   r2  normalize_start_endrj   r)   r-  r@   r   ro   )r   r  r   r?  r@  rA  src_sizer\  r'  r6  r  s   ` ````  @@@rf   slice_scatterr?    s    
3
&C}}H
3
"Czz|C H 11!S%EJE3AJJL!HS5[D1H5t<HSM
h
C"J,
 ,
\ ||~kkmAJJL!	 rh   c                j    t        | t        t        f      rt        |       dkD  rt	        | d         S | S r  )rl   rj   r   r   _unwrapr   s    rf   rA  rA    s-    !dE]#A
qt}Hrh   r   rw   rQ  
pin_memoryc                   t        |d t        j                  fv d|        t        | d       t        t	               t
              rxs t        j                  nxs t        j                         g }t         t        j                        r fd}nt         t        t
        f      r fd}nt               dk(  s't         d   t        t
        f      rBt               dk  r4|j                  t        j                  t                             fd}n5t        j                  j!                  t        j"                   |            S t%        j&                  t)        |      ||	      S )
Nlayout=rC  c                0    t        j                        S r   r/  r:  rn   r   s    rf   r\  ztensor.<locals>.inner_fn  s    >>$..rh   c                0    t        j                        S r   rS   ri  rG  s    rf   r\  ztensor.<locals>.inner_fn  s    <<e,,rh   r   r   c                      fdt              dk(  rt        j                  d      S  dt                    S )Nc           	     j   | |k  sJ || z
  dk(  rt        j                  |          S || z
  dz  | z   }t        j                  t        j                  t        j                  d   t
        j                        t        j                  |t
        j                               | |       ||            S )Nr/   r-   r   )rS   ri  r  rk  rg  r|   rh  )r?  r@  midbinary_searchrn   r   r:  s      rf   rM  z/tensor.<locals>.inner_fn.<locals>.binary_search  s    s{"{;!#<<UU;;U{q(50yyFFuQx=S%++6 "%-!#s+ rh   r   )r   rS   ri  )r:  rM  rn   r   s   `@rf   r\  ztensor.<locals>.inner_fn  s7     4yA~||Au-- CI..rh   r  r]  )r   r|   stridedrl   rA  r   rh  get_default_dtyper   r   r5  r   rp   IntegerrT   r^   add_tensor_constantrw  r@   r   rH   )rn   r   rw   rQ  rC  r[  r\  s   ``     rf   rw  rw    s   v$..'&0BC:~|,'$-%$2002!F$$	/ 
D5#,	'	- 
Ta:d1gs|<TaemmCI./	/( ww**LLU6:
 	
 V$	 rh   c                z    t        | t              r|t        | |      } |t        | |      } | S t	        | ||      S r&  )rl   rD   r   r  rw  )rn   r   rw   s      rf   	as_tensorrS  -  sD    $	"D%(DT6*D$eF33rh   c                8    t        | t        j                        S r~  rv  )rn   s    rf   long_tensorrU  8  s    $ekk**rh   c                   t        t        j                  j                  j                  t        j                  j
                  j                  d         }|J t        |      dk(  sJ |       t        t        |j                                     \  }}t        j                  |||       }t        j                  j                  |      |_        t        j                  j                  |       t        j                  j
                  j                  d   }t!        |t"        j$                  t"        j&                  t"        j(                  f      r|j*                  j,                  S t/        j0                  |      S )Nr,  r/   r  )r&   rT   r^   r   r3  r_   rc  r   r?  r  r   r2   DynamicScalarr7  r   r8  rl   r|   r  SymFloatSymBoolrd   r  r   r0  )rn   r,  binding_symrH  r  r  s         rf   _local_scalar_denser[  =  s    2	""AGG$8$8$=$=>Q$R ((( !Q&9(99&   %6%<%<%> ?@Kk7D9F''))&1FKGGv& ''


#
#E
*C#ennemmDExx}}}}S!!rh   c                     y r   r   )rn   r   s     rf   _assert_scalarr]  j  s     rh   )rw   rQ  c                    y r   r   )r   r   rV  r   rw   rQ  s         rf   _assert_tensor_metadatar_  t  s     rh   c                   | t        | t        t        f      st        d      rj                  t        t        t        f      rfd}nTt        t
        j                        rfd}n3t        j                               dk(  sJ j                         fd}t        j                  ||t        |            S )Nr   c                0    t        j                        S r   rI  r:  r   r   s    rf   r\  z_full.<locals>.inner_fn  s    <<u--rh   c                0    t        j                        S r   r/  rb  s    rf   r\  z_full.<locals>.inner_fn  s    >>%//rh   r   c                     g       S r   r   )r:  value_loaders    rf   r\  z_full.<locals>.inner_fn  s    ##rh   r]  )rl   r   r5  r   r   r   r   r   r   r_  r@   r   rj   )
fill_valuerw   r   r   r\  r   re  s     `  @@rf   _fullrg  {  s    Ej3,/GE74K%#u&	. 
E5;;	'	0 5>>#$)))((*	$ Dz	 rh   c                8     t        t        |            | fi |S r   create_tensor_liketensor_constructor)r   rf  r   s      rf   r  r    s     =0<=aJ6JJrh   c                "     d d d d dd d fd
}|S )NF)namesr   rw   rQ  rC  r'  c                   t        | d u d       t        |d t        j                  fv d|        t        | d       t        |      }|xs t        j                         }t        |      dk(  r6t        |d   t        t        t        j                  f      rt        |d         }|D ]  }t        |t        j                        sJ  |D cg c]  }t        j                  |       }}t        |||      S c c}w )Nnamed tensorsrE  rC  r/   r   )r   r|   rN  rH   rO  r   rl   rj   r   Sizer  r   r-  rg  )	rm  r   rw   rQ  rC  r'  r   r   rf  s	           rf   ri  z!tensor_constructor.<locals>.inner  s     	5D=/26dEMM22gfX4FGz><0v&2002t9>ja4

2KLa>D  	3A!!U\\222	3)-.AQ..Z55 /s   C8r   )rf  ri  s   ` rf   rk  rk    s#     6. Lrh   )rm  r   rQ  rw   rC  r'  c                    t        | d u d       t        |      }t        |      dk(  r6t        |d   t        t
        t        j                  f      rt        |d         }t        |d ||||      S )Nro  r/   r   r   rQ  rw   rC  )	r   rH   r   rl   rj   r   r|   rp  empty_strided)rm  r   rQ  rw   rC  r'  r   s          rf   r  r    sf     u}o.6"F
4yA~*T!WtUEJJ.GHT!W~d%v* rh   c                      dddddd fd
}|S )zZ
    Shim to convert X_like(...) into X(...).  For example zeros_like() into zeros().
    NF)r   rw   rQ  rC  r'  c                  t        | d       t        |d t        j                  fv d|        || j                         }nt	        |      }|xs | j                         }t        | j                               } |||||      S )NrC  rE  rB  )r   r|   rN  r   r   ro   rj   r   )r   r   rw   rQ  rC  r'  r   creation_fns          rf   _constant_likez*create_tensor_like.<locals>._constant_like  s     	z><06dEMM22gfX4FG=KKME 'E)1<<>AJJL!fV

 	
rh   r   )rv  rw  s   ` rf   rj  rj    s     duTX
 rh   c                *    t        t        |             S r   ri  )rf  s    rf   r  r    s    0<==rh   c                     d d d d d fd
}|S )Nrr  c               n   t        |t        t        f      sJ t        | d       t        |d t        j
                  fv d|        t        |      xs | j                         }|xs | j                         }|D cg c]  }t        j                  |       }}t        t        |      ||      S c c}w )NrC  rE  )rl   rj   r   r   r|   rN  r   r   ro   r   rP  rg  rH   )r   r   r   rQ  rw   rC  r   rf  s          rf   _new_constantz#new_constant.<locals>._new_constant  s     $u...z><06dEMM22gfX4FGU#4q{{})1<<>*./Qa //Zv!6tDD 0s   =B2r   )rf  r{  s   ` rf   new_constantr|    s    tDT
E rh   rr  c          	     ~    || j                         }|| j                         }t        |d ||t        |      |      S Nrr  r   ro   rs  rH   )r   r   r   rQ  rw   rC  s         rf   	new_emptyr    sF    }~V$ rh   c                  t        | t        t        f      sJ t        |t        t        t        d       f      sJ t	        | d       t	        |d t
        j                  fv d|        t        |      xs t        j                         }|xs t        j                  d      j                  }t        |      }t        d|||       }|j                          |j                  j                  }t        j                   |j                  dgt#        |       z        |_        t        |t$        j&                        sJ | D cg c]  }t)        j*                  |       } }|r#|D cg c]  }t)        j*                  |       c}nt$        j,                  j/                  |       }t%        j0                  ||| |      |_        |S c c}w c c}w )NrC  rE  rf  r   )rf  rw   r   r   )r[  )rw   r   r   rV  )rl   rj   r   r   r   r|   rN  r   rO  rw  rw   rH   rg  rs  rn   dataclassesreplacer   r2   r  r   r-  r  r  rU  rQ  )	r   rV  r   rQ  rw   rC  	pointwiser  r   s	            rf   rs  rs    s{    dT5M***ftUDJ7888:~|,v$..'&0BC<5#:#:#<E/u||C(//F6"F6TJI^^  F%%fkk1#D	/JFKfb//000%)*ELLO*D*  #))Qa)11$7 
 NN	FM  +)s    G$Gc          	     ~    || j                         }|| j                         }t        ||||t        |      |      S r~  r  )r   r   rV  r   rQ  rw   rC  s          rf   new_empty_stridedr  -  sH     }~V$ rh   c                
   |D cg c]+  }t         j                  j                  j                  |      - }}t	        t        t        |            |j                        }t        j                  j                  | |      S c c}w )N)r  )rT   r^   r   r   sortedr  r   __getitem__r2   r  r  )r   rV  r   r  s       rf   copy_stridedr  ?  sb    >DEagg11!4EFE%F,&2D2DEL??//<@@ Fs   0B c                V    |j                  d      J d        t        |      | fi |S )Nr   z(dtype should be handled by decomposition)rd  rk  )r   rf  r   s      rf   fullr  F  s5    ::g*V,VV*)j)$9&99rh   c                   t        | t              sJ |j                         dk(  rt        | |j	                               S | j	                         t              dk(  }t        | |      |rt        | dg      } dg| j                         |j                         fd}t        j                  | j                         | j                         ||j	                               S )Nr   r/   c                    t        |       } t        j                   |                }t        |       dk(  r|g} n|| <    |       S r  )rj   rS   indirect_indexingr   )r	  
gather_idxr   index_loaderr   r  s     rf   r   zgather.<locals>.fn`  sL    3i**<+<d3iH
s8q=,C!CH}rh   r]  )rl   rD   	get_numelr  r   r   r/  r-  r_  r@   r   ro   r   )	r   r   r:  sparse_gradr;  r   r  r   r  s	    `    @@@rf   gatherr  L  s     a###AENN,--::<DY!^F
3
'C1qcNs}}H$$&L ||~kkm~~	 rh   c                0  	
 |r- t        t        j                  j                        | ||||      S |rJ t	        | t
              sJ t	        |t
              sJ dt        |j                               v sJ | j                         	|j                         t        |j                               | j                         
g |j                         
dd  	
fd}t        j                  | j                         | j                         |      S )Nr   r/   c                    t        |       t              k(  sJ |  d         | d        }t        j                  |d         gg | d  z   } |      S )Nz != r   )r   rS   r  )r	  	var_index
weight_idxindices_loaderindices_ndimr  weight_loaderweight_sizes      rf   r   zembedding.<locals>.fn  s{    3x3x=(@SEhZ*@@("3}#56	++I{1~FG K
K
 

 Z((rh   r]  )r  r  	embeddingr  rl   rD   strr   r_  r   r   r@   r   ro   )weightr  padding_idxscale_grad_by_freqsparser   r  r  r  r  r  s         @@@@@rf   r  r  q  s   7 6 67G[*<f
 	
 :fi(((gy)))C))+,,,,&&(M((*Nw'')*L//#K6!!#6k!"o6H) )   " 	 rh   c           
        t        d | D              s)J d| D cg c]  }||j                          c}        t        d | D              rt        d      t	        |       D cg c]  \  }}t        |t              s| }}}t        |      dkD  sJ d       d gt        |       z  }t        |t        |D cg c]  }| |   	 c}       D ](  \  }}|j                         |k7  rt        d      |||<   * ||fS c c}w c c}}w c c}w )Nc              3     K   | ]T  }|P|j                         t        j                  t        j                  t        j                  t        j
                  fv  V y wr   )r   r|   rh  r  r   r  r   rt   s     rf   r   z.check_and_broadcast_indices.<locals>.<genexpr>  sA      = 	
%++u{{EJJLLs   AAz)indices must be int64, byte or bool. Got c              3     K   | ]7  }||j                         t        j                  t        j                  fv  9 y wr   )r   r|   r   r  r  s     rf   r   z.check_and_broadcast_indices.<locals>.<genexpr>  s1      78q}%**ekk22s   ?5?zFallback for bool indicesr   z"requires at least 1 non-None indexz.Fallback when indices is on a different device)r  r   r  r   rk   rl   rD   r   r  r   ro   )r  rw   rt   r   
valid_idxsnew_indicess         rf   check_and_broadcast_indicesr    s/      
 4G4eqWXWdQ[[]4e3fg   <C  ""=>> )' 2O1jI6N!OJOz?QD DD&3w<'KJ 1
3S1GAJ3S TU 1 <<>V#%&VWWA 
""# 5f P 4Ts   C=C='D D:Dc	           
     T   dt        j                        D ]  \  }	}
|
|	z
  dk7  sd t              D cg c]  \  }}|	| |    c}}g | t              t              z   d  d   }rz   nd | z   |d  z   f
d}|fS c c}}w )NFr/   Tr   c                  
 t        |       t              k(  sJ t              t        
      k(  sJ t              }g }d   }rdn|}d}t        d   dz         D ]}  }||k(  r||z  }|   *|t        |       k  sJ |j                  | |          |dz  }<|   }|J 
|   }|j                  t        j                   || |||z          |	              g || |d  }|S  |      S )Nr   rL  r/   r   wrap_neg)r   r  rp   rS   r  )r	  r  	new_indexfirst_tensor_indexstart_offsetnext_idxrt   loaderr   r   indexed_sizer  indices_loadersnon_consecutive_tensorsoutput_sizetensor_indicestensor_sizer  r  s            rf   r   z*index_output_size_and_inner_fn.<locals>.fn  sN   3x3{++++?#s<'8888;	+A.3q9K~b)A-. 	AL D qz!#c(***  X/A(+)))#A  ))s<,2EFG#!)		&

^
	 %,yE(92EErh   )r   pairwiserk   r   )r  r  r  r  r  r  r  r   r  previouscurrentrt   r  r  r   r  r  s    ````````      @@rf   index_output_size_and_inner_fnr    s    $ $&//? +'X"&*#+ ,5W+=MC6!9MKSKS&[)9C<O)O)Q"RSK'*!K/ ++,,-./ 	 F  FD ?_ Ns   
B$B$c                    t        | ||      \  }}}t        j                  | j                         | j	                         ||      S r)  )index_impl_helperr@   r   ro   r   )r   r  r   r  r\  r   s         rf   
index_implr    sB    0GUCK1||~kkm	 rh   c                   t        |t        t        f      sJ | j                         t	        || j                               \  }}t        |      dkD  sJ d       |D cg c]  }||j                         nd  }}t        ||d      j                               }| j                         }t        t        |            D cg c]  }||   	||    }	}|rd|	v rd|vrt        d      t        t        |            D cg c]  }||   	 }	}t        ||||||	d ||	      \  }
fd}|
|fS c c}w c c}w c c}w )Nr   z Must have at least one valid idxz0index is out of bounds for dimension with size 0r  c                       |             S r   r   )r	  index_inner_fnr  s    rf   r\  z#index_impl_helper.<locals>.inner_fn   s    s+,,rh   )rl   rj   r   r_  r  ro   r   r   r  
IndexErrorr  )r   r  r   r  r  rt   r  r  r  r  r  r\  r  r  s               @@rf   r  r    sS   ge}---}}H9'1<<>RG^~"F$FF"KRSa!-q}}TASOS w~a01::<=KZZ\F',S\':U!gaj>TF1IULUl"q';KLL',S\':;!F1I;L;"@
#K- .009 T V <s   !D2
D7D7D<c                    	 t        | |d      S # t        $ r? | j                           t        t        j
                  j                  d      | |      cY S w xY w)NTr   Fr  )r  r   rs  r  r  r:  r  r   r  s     rf   r:  r:  &  sT    
!WD11 
			M

 1 1uMw
 	

s    AAAc                    t        | |d      S )NFr  )r  r  s     rf   _unsafe_indexr  2  s    a..rh   c                6    t        t        |       |||dd      S )NTFr   may_realizeindex_put_impl_r  r   r  r   
accumulates       rf   	index_putr  ?  s     a'6:Tu rh   c                6    t        t        |       |||dd      S )NFr  r  r  s       rf   _unsafe_index_putr  F  s     a'6:U rh   c                    |j                         | j                         k7  rt        || j                               }|rt        | |      }t        | t	        |d   ||             S r  )ro   r  r   r  r  )r  r  r   r  s       rf   index_put_as_masked_fillr  M  sV    T__..%!23D% T5UD9::rh   c                   ddl m} t        t        j                  t
        j                  j                  j                  j                        }t
        j                  j                  }t        j                  sF|D ||      r<d}|j                  j                  dd       x}r| d| }|t
        j                  _        t        j                   || |||       | S )Nr/   ),_fx_node_is_input_dependent_cudagraph_unsafezLindex_put_ fallback with boolean indexing is not compatible with CUDA graphsstack_trace Found from : 
 )utilsr  r   r  
index_put_rT   r^   r_   rb   _overloadnamer0   graph_partitionrc  rd  disable_cudagraphs_reasonr2   IndexPutFallback)	r  r  r   r  r  rc  r  r   r  s	            rf   index_put_fallbackr  U  s    C$//177+?+?+F+F+T+TUK
 gg""G""8A\!,,**=$??;?E*;-8C,/)T7FJGKrh   c                $    t        | |||dd      S )NTr  r  r  r  r   r  s       rf   r  r  l  s    gvz4 rh   c                $    t        | |||dd      S )NFTr  r  r  s       rf   _unsafe_index_put_r  s  s    gvzD rh   c           
        |rLd t        j                  |       |j                         v r$t        fd|D              s|j	                          |j                         dk(  rt        |      dk(  r|d   j                         t        j                  t        j                  fv r_|d   }t        t        |j                               t        | j                                     D ]  }t        |d      } t        | |g||      S t        j                         rt!        | |||      S |D ]E  }||j                         t        j                  t        j                  fv s7t!        | |||      c S  | j                         }	t        |	      }
|rLt#        | j                               r3|
dk(  rt%        | dg      } t!        | |||      } |
dk(  rt%        | g       } | S t'        || j                               }	 t)        || j+                               \  }}|D cg c]  }||j/                         nd  }}t1        | t2              sJ | j	                          |
dk(  rt%        | dg      } t5        ||d      j                               }t        t        |            D cg c]  }|	|   	 }}t7        |	|||||d |      \  }}t9        ||      }| j+                         }|J t        j:                  || j                         |j/                         |||rdnd       }t        j<                  d t        j>                  |       |	      }t@        jB                  jE                  |      |_#        t@        jB                  jI                  |       |
dk(  rt%        | g       } | S # t,        $ r t!        | |||      cY S w xY wc c}w c c}w )
Nc                   t        | t              rt        | j                  t        j                        r| j                  j                         } t        | t        j                        xr t        | j                  t        j                        xrg t        | j                  dd       xrN | j                  j                  j                  t        j                  j                  j                  j                  u S y)Nr  F)rl   rD   rn   r2   r6   rR  r   r  r   r  rb   r|   rS   r  randpermr  )indices    rf   indice_slice_from_randpermz3index_put_impl_.<locals>.indice_slice_from_randperm}  s     &),FKK1U002vr}}5 V"6;;@VY=V ++22eiinn6M6M6U6UU	 rh   c              3  .   K   | ]  } |        y wr   r   )r   r  r  s     rf   r   z"index_put_impl_.<locals>.<genexpr>  s      H
39&v.H
r  r/   r   rL  r  
atomic_addrw   r   r\  r[  output_indexerscatter_moder   rQ  rn   )%r2   try_get_namer  r  rs  r  r   r   r|   r   r  r  r   r  r  $are_deterministic_algorithms_enabledr  rM   r  r   r  ro   r   r_  rl   rD   rj   r  r-  Scatterr  MutationLayoutSHOULDREMOVErT   r^   r7  r   r8  )r  r  r   r  r   r  rs  r   r:  r  x_ndimr  rt   r  r  r  expected_vals_sizer\  rw   scatterr  r  s                        @rf   r  r  z  s   	 ??4 F$9$9$;;C H
=DH
 E
 NN 	aLAAJ  "uzz5;;&??qzs4==?+S-AB 	'AT2&D	''tffjII 113!$DD  I!2uzz5;;6O!O%dGVZHHI ]]_F[FB4>>CSTQ;qc?D!$DQ;b>Dfdnn./FE"=T__&#
 LSSa!-q}}TASOSdI&&&LLN {D1# w~a01::<=K',S\':;!F1I;L;#A	$  F./F __Fjjnn##%!%/\TG ,,T2F
 ''))&1FKGGv&{D"~Ke  E!$DDE T <s   3N N7N<N43N4r  c                   	 t        | |dd      \  }}j                         | j                         		fd}t        j                  | j	                         | j                         ||      S )NFr  c                     j                   t        j                  k7  r+t        j                          t        j                        }n        }t        j
                  | fd      S )Nc                                     S r   r   )_unsafe_index_fnr	  self_loaders   rf   r#  z8_unsafe_masked_index.<locals>.inner_fn.<locals>.<lambda>  s    K8H8M,N rh   )r   r|   r   rS   r   rm  )r	  mask_valr  fillrs  mask_loaderr  s   ` rf   r\  z&_unsafe_masked_index.<locals>.inner_fn  sK    ::#||K$4ejjAH"3'Hzz($NPTUUrh   r]  )r  r_  r@   r   ro   r   )
r  rs  r  r  r[  r   r\  r  r  r  s
    ` `   @@@rf   _unsafe_masked_indexr    sw    "3gUU#FA ""$K""$KV V  nn	 rh   c           	         t        ||d      }| j                         }t        t        |            D cg c]#  }||   rt	        ||   ||    ||   dz
        nd % }}t        | ||d      S c c}w )Nr   r/   T)r  )r  r   r  r   r+  r  )r   rs  r  r   masked_valuer  rt   clamped_indicess           rf   #_unsafe_masked_index_put_accumulater    s    vq)LJJLE s7|$ 7>ajgaj58)U1X\2dJO  Q$OOs   (A-c                V    t        j                  |t        j                  ||             S r   )rS   r  r  r   minmaxs      rf   r+  r+    s    ;;sCKKQ/00rh   c                P    t        |       }t        ||||      }t        ||       |S r   )r  r>  copy_)r  r  r   rV  r-  r0  output_views          rf   as_strided_scatterr  #  s*    4[FVT6>BK	+sMrh   c                2    t        t        |       |||fi |S r   )scatter_r  )r   r   r:  r  r   s        rf   r  r  +  s    E!Hc5#888rh   r  include_selfc               :   t        |t              }t        | ||j                         t	        t
        j                  |r|j                         n
t        |            |r|j                         j                  nd|      rt        j                  | ||||||       |S y )Nznot implr  )rl   rD   rR   r   r
   r|   r   r   ro   r2   ScatterFallback)rc  r  r   r:  r  r  r  src_is_tensors           rf   scatter_fallbackr  0  s     sI.MU[[]#--/S	J!.J 	%	
 rh   r  c                  |dv sJ |[t        t        j                  t        j                  j
                  j                  j                        }t        || ||||      }||S |dk(  rd}n|dk(  rd}t        | ||||      S )N)Nr   multiplyr  r   r  r  prod)
r   r  r  rT   r^   r_   rb   r  r  scatter_reduce_)r  r   r:  r  r  rc  fallback_results          rf   r  r  Q  s    ....~dmmQWW-A-A-H-H-V-VW*sE3v
 &""	:	4eS&99rh   c                0    t        t        |       |||      S r   )scatter_add_r  r   r   r:  r  s       rf   scatter_addr  c  s    a#uc22rh   c                     t        | |||d      S )Nr  )r  r  s       rf   r  r  h  s    1c5#u55rh   c                4    t        t        |       ||||fi |S r   )r  r  )r   r   r:  r  reduction_typer   s         rf   scatter_reducer!  m  s    58S%nOOOrh   )r  c          
         |dv sJ t        t        j                  j                               dk(  r dt        j                  j                         v sJ d       t	        t
              rt               t        t        j                  j                   |||      }|r|S t	         t              sJ dt        |j                               v sJ t         j                               }|dk(  rt         dg       t	        t              r)t        j                               dk(  rt        dg      t	        |t              r)t        |j                               dk(  rt        |dg      }|j                         dk(  r S t                j!                          |j#                         t	        t              rj#                         nd  fd} fd	}	d
 }
 j%                         }|J |st'        j(                  | j                          fd|j                         |d       }t'        j*                  d t'        j,                         |      }t.        j0                  j3                  |      |_        t.        j0                  j7                  |       t'        j(                  | j                         |	|j                         | |
|            }t'        j*                  d t'        j,                         |      }t.        j0                  j3                  |      |_        t.        j0                  j7                  |       |dk(  rt         g         S )N)Nr  r  meanamaxaminr/   twozKaten.scatter_reduce_.two is not the unique overload of aten.scatter_reduce_r  r   r   c                    j                         }t        |      }t        |       }t        j                   |       |dk(  rdn|   d      |<   |S )Nr   r/   F)r  )r   r   rj   rS   r  )r	  r  r&  indirect_idxr   r  r  s       rf   r  z'scatter_reduce_.<locals>.output_indexer  sU    5zCy11DAIq5:
S rh   c                `    r |       S t        j                  j                               S r   rS   ri  r   )r	  r  r  r6  s    rf   r   zscatter_reduce_.<locals>.fn  s*    c?" <<T^^%566rh   c                    | dk(  ry| J y )Nr  r  r   r  s    rf   backend_reduce_strz+scatter_reduce_.<locals>.backend_reduce_str  s    U? >!>rh   c                L    t        j                  dj                               S r  r*  )r:  r  s    rf   r#  z!scatter_reduce_.<locals>.<lambda>  s    3<<4>>3C#D rh   r  r  )r   r  r  r   rl   r!   r  r  r&  rD   r  r   r   r  r  r/  rs  r_  ro   r2   r  r  r  rT   r^   r7  r   r8  )r  r   r:  r  r  r  r  r&  r  r   r,  rw   zero_outr  r  r  r6  s   `` `           @@rf   r  r  r  s   BBBBD  **,-2T))3355U UU	6 #vc"&  !O dI&&&C)****t}}DqyD1##y!c#,,.&9Q&>3n%#ENN,<(=(BUQC A
c
"CLLN$$&L&0i&@"dJ7 __F::.."D>>#)
 ""006

 gg--f5	""6*
 jjnn~~%'/G ,,T2F
 ''))&1FKGGv&qyD"~Krh   c                T   | j                          | j                         | j                          d  | j                         d   }D cg c]+  }t        j                  j
                  j                  |      - c}t        |      k(  sJ |}t        |      D cg c]
  \  }}||z   c}}t        |      D ]  \  }}	|		d|	z  |<    fdfd}
t        j                  | j                         | j                         |
g ||      S c c}w c c}}w )Nr  c                   t        j                  | t        j                        } r8t        j                  | t        j
                  dt        j                              } t        j                  | t        j
                  |t        j                              } t        j                  | t        j                        } t        j                  | |d      S )N      ?Fr  )
rS   rg  r|   r  r   ri  r^  r   r  r  )r   r  r   exacts      rf   scale_fnz$upsample_nearestnd.<locals>.scale_fn  s     NN1emm,3<<U]];<AGGAs||E5==9:LLEKK($$QE::rh   c                    |  d  }| d   } 
g |t        |      D cg c]  \  }}} 	|||       c}}}      S c c}}}w r   )r  )r	  r   r/  rt   r   r   i_sizes
inv_scalesnr3  r  s         rf   r   zupsample_nearestnd.<locals>.fn  s`    H!HWaW#aW:UVVJAq$8Aq$'VW
 	
Vs   Ar]  )realize_hintr_  r   rT   r^   r   r  r   r  rk   r@   r   ro   r   )r   r  scales_xr7  r2  batchrt   o_sizesor  r   r5  r6  r3  r  s      ``      @@@@rf   upsample_nearestndr=    s    NN}}HjjlA23GJJL1"E6=>qww))!,>Gx=AG$'$9:DAq!a%:Jh' (5%KJqM(	;
 
 ||~kkm!!!	 9 ?
 ;s   0D'D$c                "    t        | ||fd      S )Nr/   r7  r=  r   r  r  s      rf   upsample_nearest1drB    s    avi1==rh   c                $    t        | ||fdd      S )Nr/   Tr7  r2  r@  rA  s      rf   _upsample_nearest_exact1drE    s    avi1DIIrh   c                $    t        | |||fd      S )Nr-   r?  r@  r   r  scales_hscales_ws       rf   upsample_nearest2drJ     s     ax.BaHHrh   c                &    t        | |||fdd      S )Nr-   TrD  r@  rG  s       rf   _upsample_nearest_exact2drL  '  s     ax.BatTTrh   c                &    t        | ||||fd      S )Nr   r?  r@  r   r  scales_drH  rI  s        rf   upsample_nearest3drP  .  s     ax8.LPQRRrh   c                (    t        | ||||fdd      S )Nr   TrD  r@  rN  s        rf   _upsample_nearest_exact3drR  9  s#     	;8X6!4 rh   c                ,     t         fd|D              S )Nc              3  J   K   | ]  }t        j                  |        y wr   rI  )r   r   r   s     rf   r   z$_create_constants.<locals>.<genexpr>G  s     6Aa'6s    #)r   )r   ru   s   ` rf   r  r  F  s    6666rh   c                    | j                         | j                         fd}t        j                  | j	                         | j                         |      S )Nc                    t        |       } t        |       t              k(  sJ D ]  }|   dz
  | |   z
  | |<     |       S r  )rj   r   )r	  r   r  r  r  s     rf   r  zrev.<locals>.loaderP  sW    3i3x3u:%%% 	3Cc
Q#c(2CH	3 }rh   r]  )r_  r   r@   r   ro   r   )r   r  r  r  r  s    ` @@rf   revrW  J  sM     }}HJJLE ||~kkm	 rh   c                   d } |       ryt        |      dk7  st        | j                               dk7  ry| j                          t        | t        j
                        rt        | j                  t        j                        rt        | j                  j                  t        j                        s>t        j                  rNt        | j                  j                  t        j                        r | j                  j                  j                  sy| j                          t	        j                  |       \  }}|j                  }|d   dk7  ry|d   dk7  s|d   dk7  s|d   dk7  ry|d   }|dk(  ry|d   }|j                   d   }	||	|z   k  ry| j                  j                  j                  }
|j                   d   |j                   d   |z   g}|t"        j$                  j&                  |
<   t)        | ||j                  |j*                        }t-        |d|	|	|z   d	      }t/        ||       t0        d
   dxx   dz  cc<   |S )z
    This optimization changes the semantics of padding from 'clone'
    style to 'view' style.

    Thanks to functionalization, this change can still maintain numerical
    correctness.
    c                    t         j                  j                  } | yt        | j                        }t        |      dk(  rD|d   j                  t        j                  j                  t        j                  j                  fv ryy)a  
        Conservatively check if padding can be fused with downstream op.
        1. if the downstream op is a sum, then there is little benefit to
           do inplace padding
        2. if the downstream op is a matmul, doing inplace padding can
           save membw.
        Tr/   r   F)rT   r^   r_   r   r`   r   rb   r  mmr  addmm)r_   r`   s     rf   _padding_can_be_fusedz6inplace_constant_pad_nd.<locals>._padding_can_be_fusedk  sk     ww++l(()u:?uQxGGOOJJ3
  
 rh   Nr   r-   r/   r   r   F)r   r?  r@  r+  inductorinplace_padding)r   r   rs  rl   r2   rD   rn   r   r  r0   can_inplace_pad_graph_inputInputBufferr   freeze_layoutrT  rV  r   rT   r^   buffer_to_padded_sizer>  r;  rN  fill_r   )r   paddingrf  r\  r   rQ  r  npadstride0rowsizebufnamepadded_size	resized_xsliced_xs                 rf   inplace_constant_pad_ndrl  `  s   (  7|qC

-2
 IIK
 q",,'!&&"--0qvv{{B$5$5622qvv{{BNN; vv{{OO((+IAvmmGqzQqzQ'!*/WQZ1_1:DqyajGkk!nG4ffkkG;;q>6;;q>D#89K-8AGG!!'*		I iQg7T>QVWH	(JZ*+q0+rh   c                   t        |      dz  dk(  sJ t        d |D              rt        |       S t        j                  rt        | |      }|r|S | j                         }t        t        t        t        |d d d   |dd d                           t        |      t              z
  g D ]?  \  }}j                  t        j                  j                  j                  |      |f       A t        |d        }g t        |d        D ]C  \  \  }}	}
j                  |
       |j                  t        j                   |
|z   |	z                E t        |      t        |      k(  sJ  t#        | j%                                     fdfd}| j'                         t)        j*                  | j-                         | j%                         ||      S )Nr-   r   c              3  &   K   | ]	  }|d k(    ywr   Nr   r   r  s     rf   r   z"constant_pad_nd.<locals>.<genexpr>  s     
#a16
#r  r/   c                B    g }t         	d        D ]J  \  }\  }}}|dk7  r|j                  t        |d             |dk7  s0|j                  t        ||             L t	        j
                  t        j                  |      }t        j                  | 
fd      S )Nr   c                             S r   r   )r:  r  s   rf   r#  z/constant_pad_nd.<locals>.mask.<locals>.<lambda>  s     rh   )	r  rp   range_mask_lowrange_mask_highr  r  rS   rl  rm  )r:  rs  r	  r  r  r0  boundsrf  
mask_sizesr7  r  s   `     rf   rs  zconstant_pad_nd.<locals>.mask  s    (+E!"Ivz(J 	:$C#tfaxN323qyOC89		:
 $/zz$ 7DDrh   c                    t        | d        }t        | d        D ]  \  }\  }}|j                  ||z
          t        |      t        |       k(  sJ  |      S r   )rj   r  rp   r   )r:  r  r	  r  _highbounds_precomprs  r7  s        rf   	offset_fnz"constant_pad_nd.<locals>.offset_fn  si    rO	!$U12Y!? 	(C#uS3Y'	(9~U+++Irh   r]  )r   r  r  r0   r^  rl  r   rj   r(  r  rp   rT   r^   r   lookup_precomputed_sizer   r-  r   r   r_  r@   r   ro   )r   rd  rf  rr   r  lhr  r  r  r   rz  ru  ry  rs  rv  r7  r  s     `         @@@@@@rf   r  r    s   L1"""

#7
##Qx%a*=J JJLE(4GCaCL'!$Q$- @ABCFE
S[ A 68N P1qww//GGJANOP uRay/KJ qr3 <dT$5<<s
T(9:;< {s5z)))-q{{}-j9JE E }}H||~kkm	 rh   c                    t        j                  t        j                  | t        j                        t        j                  t        j                  |      t        j                              S r   )rS   rj  rg  r|   rh  r   rP  )rt   r  s     rf   rs  rs    s@    66q%++&u}}S)5;;7 rh   c                    t        j                  t        j                  | t        j                        t        j                  |t        j                              S r   )rS   rk  rg  r|   rh  )rt   r  s     rf   rt  rt    s7    66q%++&tU[[) rh   c                V    t        j                  t        | |      t        | |            S r   )rS   rl  rs  rt  )rt   r  r  s      rf   
range_maskr    s'    88q#4  rh   c                      j                          d   j                         xs dgz   fd}|S )Nr   c                N   | d   |  d  t        j                  t        j                  t	              D cg c]  }t        |   |   
|   z   
|          ! c}      }	rt        j                  |fd      S t        j                  |fd      S c c}w )Nc                 4     t               g       S )Nr   )constant_boundary_condition)r   ihpad_fill_valueprefixr   s   rf   r#  z;constant_boundary_condition.<locals>.load.<locals>.<lambda>   s$    O3A~3O"fNrN rh   c                      g        S r   r   )r  r  r  s   rf   r#  z;constant_boundary_condition.<locals>.load.<locals>.<lambda>&  s    (>V>b>*B rh   )r  r  rS   rl  r  r  rm  )r:  rt   rs  r  r  r   rf  r}  r  rd  	padding_hr   r  s      @@rf   rQ  z)constant_boundary_condition.<locals>.load  s    uC45\HHLQRUJWqZ1qtil2Yq\MBW
  JJ 
	
 D"BJO
	
 Xs   $B"
)r   r_  )	r   rf  rd  r  r   rQ  r}  r  r  s	   ````` @@@rf   r  r    sE     	


cTUA}}H$A39I
 
* Krh   dilationc                  |dgt        |      z  }t        | d||   z  z   ||   ||   dz
  z  z
  ||   dz
  z   ||         }|rt        | d||   z  z   ||   ||   dz
  z  z
  d||   dz
  z  z   ||         }t        j                  j                  j                  |dz
  ||   z  | z
  ||   z
        dk\  r>|dz  }t        j                  j                  j                  d|||   z  | z
  ||   z
         t        j                  j                  j                  ||z
        dk(  r0t        j                  j                  j                  ||       d}||fS |}||fS )Nr/   r-   r   F)r   r)   rT   r^   r   	size_hintr#  r,  )	r   rt   kernel_sizerV  rd  	ceil_moder  x_outx_alts	            rf   pooling_sizer  ,  s   3W%	A
NXa[KNQ,>??6!9q=Qq	E
 '!*nqk[^a/01 6!9q=!" 1I
 77%%uqyF1I&=&AGAJ&NOSTTQJEGG&&q%&)*;a*?'!**LM77%%eem49GG))%7I ) E)rh   c               l    t        | |      } t        j                  t        j                  |       }|dkD  S )N   )rN   r  r  r  r^  )r  n_dimwindow_sizes      rf   %should_fallback_max_pool_with_indicesr  J  s/    {E2K""8<<=Krh   assert_fallbackc                  |dk(  rdg|z  }|dk(  rdg|z  }|s|}t        ||      }t        ||      }t        ||      }t        ||      }t        | t              sJ t        |      |k(  sJ t        |      |k(  sJ t        |      |k(  sJ t        |      |k(  sJ t        | j	                               |dz   |dz   fv sJ t        ||      }|||k(  sJ |||||fS )Nr   r/   r-   r  )rN   rl   rD   r   r   r  )r   r  rV  rd  r  r  r  use_fallbacks           rf   max_pool_checksr  P  s    !|#+1}3;{E2K&%(F7E*GHe,Ha###{u$$$v;%w<5   x=E!!!qzz|EAI 66668ERL"...<??rh   c                  | j                          | j                  d   }| j                   d  }t        t              D 	cg c]  }	t	        ||	   |	||       c}	 \  }
}| j
                  }|t        j                  u rdn5|j                  rt        d      nt        j                  |      j                  }t        |      t        |
      z   }t              st        |      st        d D              rt        | |      n| j                         fd}t!        j"                  d| | j%                         |||||      }t!        j"                  d	| | j%                         t        j&                  ||||      }t)        |j*                  j*                  t               r|j-                          t)        |j*                  j*                  t               r|j-                          ||fS c c}	w )
Nr  F-infc              3  &   K   | ]	  }|d kD    ywr  r   r  s     rf   r   z)_max_pool_with_offsets.<locals>.<genexpr>  s     ,EqQU,Er  r  c                    | d   }|  d  }t              D cg c]  }||   	|   z  ||   |   z  z   |   z
  ! }} 
g ||      S c c}w r   r  )r	  reduction_idxr  bhrt   r  r  r  rd  rV  r  s         rf   fn_innerz(_max_pool_with_offsets.<locals>.fn_inner  s    Wuf%\ 5\
 UVAY=#3hqk#ABWQZO
 
 &2''	
s   $Ar  r   
input_noderw   	dst_dtyper  r\  r[  reduction_rangesargmax)r8  r  r  r  r  r   r|   r   r  r5  r  r  rj   r  r  r_  rA   r   ro   rh  rl   rn   rs  )r   r  rV  rd  r  r  r  r:  dhwr  dhw_outr   	min_valuer  r  r_  offsetsr  s     ``` `          @rf   _max_pool_with_offsetsr  m  s    NNGGGeVE
''5&'
C
 5\	

  A;X

GY GGE EJJ 	$66eFmEKK<N<R<R  E{T']*H
7|s9~,EH,E)E.q)G==?( ( ||~$	F ||~++$	G &++""I.',,##Y/7?s

s   G8c           
        t        |      }t        | |||||d      \  }}}}}t        j                  d      5  t	        | ||||||      \  }}	|t        |	t        j                        fcd d d        S # 1 sw Y   y xY w)NFr  r  unroll_reductions_thresholdr  )r   r  r0   r   r  r   r|   r  )
r   r  rV  rd  r  r  r  r   r_  r  s
             rf   !_low_memory_max_pool_with_offsetsr    s     E 1@	1-K(A 
"	5 
50
 x44
5 
5 
5s   1A55A>c                P   t              | j                         t        j                  t	        j
                  t        j                              fd}t        j                  | j                         t        j                  || j                               }|S )Nc                     	|       }t        j                  |
      }t        j                  |      } | |      }t        j                  t        j
                  | d        t        j                        S r   )rS   r  r1   _flattened_index_to_ndrg  _flatten_indexr|   rh  )r	  r;  offset_sympyr  idhwincrements_to_index
input_sizer  r  offsets_loaderr  s        rf   offsets_to_indicesz4_pool_offsets_to_indices.<locals>.offsets_to_indices  sm    $,,V[A&==lKX"36~~))$
E670CDekk
 	
rh   r]  )r   r_  r   r0  r  r  r  r^  r@   r   ro   r|   rh  r   )	r  r  r  r  r  r  r  r  r  s	    ```  @@@rf   _pool_offsets_to_indicesr    s     E((*N--	 0 0{ KLK
 
 !!#kk#!	G Nrh   c                L    t        |      fd}t        | |||      S )Nc                    |  d  }t              D cg c]  }||   |   z  ||   |   z  z   |   z
  ! c}S c c}w r   r  )r	  r  r  rt   r  r  rd  rV  s       rf   r  zD_low_memory_max_pool_offsets_to_indices.<locals>.increments_to_index  s_    %\ 5\
 UVAY=#3hqk#ABWQZO
 	
 
s   $<)r   r  )r  r  r  rV  rd  r  r  r  s      ``` @rf   '_low_memory_max_pool_offsets_to_indicesr    s-     E
 $j*= rh   c           	         t        | |||||      \  }}}}}t        | ||||||      \  }}	t        |	|| j                  | d  |||      }
||
fS )Nr  )r  r  r  r  )r   r  rV  rd  r  r  r  r   rr   r  r  s              rf   _max_pool_with_indicesr    s}     1@	;1-K(A *	;9ELC 6	G <rh   c           	     &    t        | |||||d      S Nr-   r  r  r   r  rV  rd  r  r  s         rf   max_pool2d_with_indicesr  -        "	;9A rh   c           	     &    t        | |||||d      S Nr   r  r  r  s         rf   max_pool3d_with_indicesr  <  r  rh   c                   dk(  rddg|dk(  rddg}st        |t              sJ t              dk(  sJ t              dk(  sJ t              dk(  sJ t        |      dk(  sJ t        |j                               dv sJ | j	                          | j                         }t        |t              rt        |j                  j                  t              r|j                  j                  }	|	j                         }
|
J t        j                  d t        j                  |
|	j                         |	j                               |	      }|j                          |j                         }n|j                         }|d uxr |d   dk(  xs |d uxr |d   dk(  }t        d |D              rt!        | ||||      S |j                         ^ }}| j                         ^ }|j#                         | j#                         t%        |j                               }t'        fdt)        d   dz        D              t'        fd	t)        d   dz        D              z  }|d
kD  rt!        | ||||      S |j                         fd}t        j*                  | j                         | j                         ||      }|rt        j,                  j/                  |      S |S )Nr   r/   r-   r  )rw   r   r   r  c              3  &   K   | ]	  }|d k7    ywr  r   r  s     rf   r   z3max_pool2d_with_indices_backward.<locals>.<genexpr>{  s     
$a16
$r  c              3     K   | ]=  }t        t        |d          t        d t        |d    z
  d                z
  d       ? ywr   r/   Nr  r)   r   r}  r  rV  s     rf   r   z3max_pool2d_with_indices_backward.<locals>.<genexpr>  M       	HQq	"SHQQ5GPQ,S%TTVWX   AAc              3     K   | ]=  }t        t        |d          t        dt        |d    z
  d                z
  d        ? ywr/   r   Nr  r   wr  rV  s     rf   r   z3max_pool2d_with_indices_backward.<locals>.<genexpr>  r  r  r  c                   | ^ }}}t        j                  |z  |z   t        j                        }|d   z   }|d   z   }t        j                  t	        |d   z
  d   z   d         t        j                        }t        j                  t	        |d   z
  d   z   d         t        j                        }t        j                  t	        |d         dz   t        j                        }t        j                  t	        |d         dz   t        j                        }t        j
                  |t        j                  dt        j                              }t        j
                  |t        j                  dt        j                              }t        j                  |t        j                  t        j                              }t        j                  |t        j                  t        j                              }d }	t              D ]3  }
t              D ]!  }t        j                  |t        j                  |
t        j                              }t        j                  |t        j                  |t        j                              }g |t        j                  t        j                  |t        j                  |t        j                  dt        j                                    d   d      t        j                  t        j                  |t        j                  |t        j                  dt        j                                    d   d      } |      } |      }t        j                  ||      }|	;t        j                  ||t        j                  dt        j                              }	t        j                  t        j                  t        j                   ||      t        j                   ||            |      }t        j                  |t        j                  |	|      |	      }	$ 6 |	J |	S )Nr   r/   Fr  rL  rf  )rS   rg  r|   r  r)   r  ri  r  r  r   r  r  r5  r  r  rl  rk  )r	  r  r}  r  
index_testphstartpwstartphendpwendgradientph_pw_phpw
grad_indexindex_actual	grad_partr   rs  grad_loaderh_window_sizer  indices_sizer  rd  pooled_heightpooled_widthrV  w_window_sizewidths                      rf   r   z,max_pool2d_with_indices_backward.<locals>.fn  sA   A^^AIM5;;?

N
N..QQ'&)3VAY?
 ..QQ'&)3VAY?
 x6!9595;;Gx6!9595;;G++gs||Au{{'CD++gs||Au{{'CDE3>>-#MNE3>>,#LM' #	WC]+ "WWWWcll3&DEWWWcll3&DE))Bs||Au{{7S(TU$R(# ))Bs||Au{{7S(TU$R(#
  .j9'
3	|Z8#"yyy#,,sEMM*J H 88FF2u-FF2u- D  #yyswwx/KXVHE"W#	WH ###rh   r]  )rl   rD   r   r   r8  r  rn   r@   ro   r2   r  r  r   decide_layoutr<  r  )fallback_max_pool2d_with_indices_backwardr_  rj   r  r  r   r  r  )grad_outputr   r  rV  rd  r  r  r  	gO_stridern   rw   x_bufferx_strideis_channels_last_batch_heightr   r  r  r   rr   r  r  r  r  r  r  r  r  s     ```                @@@@@@@@rf    max_pool2d_with_indices_backwardr  P  s    !|a&1}q6a###{q   v;!w<1x=Aqzz|&&& ,,.I!YJqvv{{I$Fvv{{"!!!$$$$nn&]]_
 
 	 &&(%%' ,A!1A 3)A,!"3  
$8
$$8K(Iw
 	
  jjlVWe&1&:&:&<#Q|((*N))+KAJJL!H {1~)* M  {1~)* M
  -/KR8K(Iw
 	
 ##%L9 9v 

%%'##%	C 44S99
rh   c                6    | j                         fd}|S )Nc           
        	
 |\  	|\  
|\  }}t        j                  t        j                  t        j                  z   t        j
                        t        j                  |t        j
                              t        j                  t        j                  
	z   t        j
                        t        j                  |t        j
                                    }t        j                  |	 
fd      S )Nc                 ,     g  z   z         S r   r   )h_start_indexr  iwr  w_start_indexr  s   rf   r#  z3pad_adaptive_loader.<locals>.load.<locals>.<lambda>  s&    HNvN}r'9N=2;MNO rh   )rS   rl  rk  rg  r|   rh  rm  )r  
incrementsstart_indicesend_indicesh_end_indexw_end_indexrs  r  r  r  r   pad_valr  s   `      @@@@rf   rQ  z!pad_adaptive_loader.<locals>.load  s    B'4$}#. [xxFF}r15;;?{EKK8 FF}r15;;?{EKK8	
 zzOO
 	
rh   r_  )r   r  rQ  r  s    ` @rf   pad_adaptive_loaderr    s    }}H
, Krh   c                    t        j                  | ||      }t        j                  |||      }t        j                  | ||      }t        j                  |||      }	||||	fS )N)out_diminp_dim)r  r#  )
rB  rC  h_inw_inh_outw_outr  r  r   r  s
             rf    compute_indices_adaptive_poolingr    sa    %%k5$OM##IudKK%%k5$OM##IudKK+}kAArh   c                d    |\  }}|\  }}	t        | |||||	      \  fd}
|
S )Nc                   | ^ }}} |      } |      } |      } |      }d }	t        j                  t        d         t        d               D ]$  \  }
} |||
|g||g||g      }|	|}	 ||	      }	& |	S r  )r   productr  )r	  r  r  r  bwr  r  r   r  r_  r  r  r  h_end_index_fnh_start_index_fnkernel_maxes
pooling_fnw_end_index_fnw_start_index_fns                rf   r   z _adaptive_pooling_fn.<locals>.fn  s    R(,$R((,$R(''l1o(>lSTo@VW 
	1FBR.k*	C ~#C0
	1 rh   r  )rB  rC  r  in_sizes	out_sizesr  r  r  r  r  r   r  r  r  r  s     `  `     @@@@rf   _adaptive_pooling_fnr    sP     JD$LE5 	)YdE5	
 . Irh   c                h   
 |\  }|\  }}t        | ||||      \  

fd}	|	S )Nc                   | ^ }}} |      } |      } |      } |      }d }	d }
t        j                  t        d         t        d               D ]  \  }} ||||g||g||g      }t        j                  ||z   z  |z   |z   t
        j                        }|
|}
n+t        j                  t        j                  ||	      ||
      }
|	|}	| ||	      }	 |
S r  )	r   r  r  rS   rg  r|   rh  r  gt)r	  r  r  r  r  r  r  r   r  maxvalmaxindexr  r  r  r:  r  r  r  r  r  r  r  s                  rf   r   z)_adaptive_pooling_fn_with_idx.<locals>.fn@  s   R(,$R((,$R(''l1o(>lSTo@VW 	1FBR.k*	C NN#t+m;b@%++E  99SVVC%8%J~#C0)	1, rh   r  )rB  rC  r  r  r  r  r  r  r  r   r  r  r  r  r  s     `  `    @@@@@rf   _adaptive_pooling_fn_with_idxr$  1  sQ     JD$LE5 	)YdE5	
! !F Irh   c                6     j                         t        j                  k(  rt        d      t	         t
              sJ t        |      dk(  sJ  j                           j                         ^ }}}t        j                  j                  j                  |      }t        j                  j                  j                  |      }|\  }}||k(  r||k(  rt               S |dk(  s|dk(  r2g |||}t        | j                          j                               S ||z  dk(  r,||z  dk(  r$t!        ||      t!        ||      g}t#         |      S t%        ||z   dz
  |      }	t%        ||z   dz
  |      }
t'        |      ||gz   } j                         }|	|
z  }|dkD  rt)         |      S d }d }t+        |||	|
g||g||gt,        j.                  	      t1        t3                      fd
}t5        j6                   j                         |||      }|S )Nz0'adaptive_avg_pool2d' not implemented for 'Long'r-   r   r  r/   r  c                     t        | |z  |      S r   r)   r:  r
  r  s      rf   rB  z)_adaptive_avg_pool2d.<locals>.start_index      733rh   c                2    t        | dz   |z  |z   dz
  |      S r  r'  r(  s      rf   rC  z'_adaptive_avg_pool2d.<locals>.end_index  "    g-7!;WEErh   rB  rC  r  r  r  r  c                ^    t        j                   | t                     |             S r   )rS   truedivr  )r	  fn_sumones_loaderr   s    rf   r   z _adaptive_avg_pool2d.<locals>.fn  s-    {{3+A./[1I
 	
rh   r]  )r   r|   rh  rq  rl   rD   r   r8  r   rT   r^   r   r  r  r  ro   r)   
avg_pool2drG   rj   fallback_adaptive_avg_pool2dr  rS   r   r  	ones_liker@   r   )r   r  r:  r  r  r  r  o_sizer  h_kernel_maxw_kernel_maxr  r   r  rB  rC  r   rvr/  r0  s   `                 @@rf   _adaptive_avg_pool2dr8  k  s   {{}#MNNa###{q   NNUD$77%%d+D77%%d+DLE5 u}QxzUaZ'5'%''V1;;=HHe|qTE\Q.e,htU.CD![))D5L1,u5LD5L1,u5LE{eU^+HKKME-KR+A{;;4F ""L1%.77F &il3K

 
		||~	
B Irh   c                     j                         t        j                  k(  rt        d      t	         t
              sJ t        |      dk(  sJ  j                           j                         ^ }}}t        j                  j                  j                  |      }t        j                  j                  j                  |      }|\  }}|dk(  s|dk(  r\g |||}t        | j                          j                               t        |t        j                   j                               fS ||z  dk(  r||z  dk(  rt        t!        ||z   dz
  |      }t!        ||z   dz
  |      }	t#        |      ||gz   }
 j                         }||	z  }|dkD  rt%         |      S d }d }t'        ||||	g||g||gt(        j*                  	      t-        ||||	g||g||gt(        j*                  	       fd
} fd}t/        j0                   j                         |||
      }t/        j0                   j                         t        j                  ||
      }||fS )Nz,adaptive_max_pool2d not implemented for Longr-   r   r  r/   r  c                     t        | |z  |      S r   r'  r(  s      rf   rB  z(adaptive_max_pool2d.<locals>.start_index  r)  rh   c                2    t        | dz   |z  |z   dz
  |      S r  r'  r(  s      rf   rC  z&adaptive_max_pool2d.<locals>.end_index  r+  rh   r,  c           	     <     | t        t        d                  S Nr  r  r5  )r	  inner_func_max_valr   s    rf   inner_fn_max_valz-adaptive_max_pool2d.<locals>.inner_fn_max_val      !#':1eFm'LMMrh   c           	     <     | t        t        d                  S r=  r>  )r	  inner_func_max_idxr   s    rf   inner_fn_max_idxz-adaptive_max_pool2d.<locals>.inner_fn_max_idx  rA  rh   r]  )r   r|   rh  rq  rl   rD   r   r8  r   rT   r^   r   r  r  ro   
ValueErrorrG   rj   fallback_adaptive_max_pool2dr  rS   r  r$  r@   r   )r   r  r:  r  r  r  r  r4  r5  r6  r  r   r  rB  rC  r@  rD  r7  rirC  r?  s   `                  @@rf   adaptive_max_pool2drH    sL   {{}#IJJa###{q   NNUD$77%%d+D77%%d+DLE5zUaZ'5'%''V1;;=H%%++allnK
 
 	
 e|qTE\Q.D5L1,u5LD5L1,u5LE{eU^+HKKME-KR+A{;;4F ."L1%.;; 7"L1%.;;NN 
		||~!	
B 
		||~kk!	
B r6Mrh   c                h               j                          fd}|S )Nc                `   j                         }t        |      dk(  r[t        |       dk(  r d| d   dz
  
z
  g      }nJt        |       dk\  r | d   | d   dz
  
z
  g      }n$ dddz
  
z
  g      }n g | dz
  
z
        }t        j                  |j	                               }t        j                  z
  t
        j                        }t        j                  dz
  t
        j                        }t        j                  t        j                  |t
        j                        t        j                  |t
        j                              }t        j                  t        j                  |d      d|      }t        j                  ||z   |z        t        j                  ||z        z
  }t        j                  |t
        j                        }t        j                  ||      }	t        j                  t        j                  |	||      t        j                               S )Nr   r/   r   r-   )r   r   rS   rg  r   r|   rh  r.  r   float64r  r5  r  rk  r  r   r0  )r  rt   samples_shapesamplei_exprdiffout_sz_exprrG  seq_irs  r   in_sz	kernel_szndimsout_szsamplessamples_loaders             rf   rQ  z)_fractional_pooling_offsets.<locals>.load  s     ((*}"6{a (F1Iuqy3(GHV! (F1Iuqy3(OP (Auqy3(?@ $$>f$>eai#o$>?F7#4#4#67~~ei/=nnVaZ=LLu}}-s||K/W
 		#&&a0!U;		6F?e34syy%7PPUEKK0vvfk*$$SYYtUD%A5==QVCWXXrh   r  )rV  rR  rU  rS  r   rT  rQ  rW  s   `````` @rf   _fractional_pooling_offsetsrX    s@    C[F#JE#I((*N!Y !YF Krh   c                "    t        | |||d      S r  _fractional_max_poolr   r  r  random_sampless       rf   fractional_max_pool2dr^  4      ;^STUUrh   c                "    t        | |||d      S r  rZ  r\  s       rf   fractional_max_pool3dra  9  r_  rh   c                   | j                          | j                  d   | j                   d  }}t        j                  d      5  t	              D cg c]  }t        |||||       c}| j                         fd}fdt        |      t        |      z   }	| j                         }
t        j                  d| | j                         |
|
||	|      }t        j                  d| | j                         t        j                  |
||	|      }t        |t              sJ |       t        |j                   j                   t              r|j#                          t        |t              sJ |       t        |j                   j                   t              r|j#                          t%        ||| j                        }||fcd d d        S c c}w # 1 sw Y   y xY w)	Nr  r  )rV  rR  rU  rS  rT  r   c                6    | d   } g | | |            S r   r   )r	  r  r  r  r  r  s      rf   r  z&_fractional_max_pool.<locals>.fn_innerQ  s.    5&\FOfO':3'NOPPrh   c                ~    | d   }|  d  }t              D cg c]  } |   |||         ||   z    c}S c c}w r   r  )r	  r  r  bdhwr  dhw_index_fnr  s        rf   r  z1_fractional_max_pool.<locals>.increments_to_indexU  s^    5&\Fvw<D u  QQ0=3CC  s   :r  r  r  )r8  r  r0   r   r  rX  r_  rj   r   rA   r   ro   r|   rh  rl   rD   rn   rs  r  )r   r  r  r]  r  r:  inp_dhwr  r  r  r   r_  r  r  rf  r  r  s       `         @@@rf   r[  r[  >  s   NNWWWuf%qwwvw'77E	"	5 = 5\

  (&"%

 ==?	Q	 ;k!22!! <<>(	
 ""#<<>kk(	
 &),4f4,fkk&&	2NN'9-6w6-gll''3OO*[!''+>
 w{= =

= =s   GG1EGGG%c                ~     j                           j                         ^ }}}t        j                  j                  j                  |      }t        j                  j                  j                  |      }|^ }}}	||z  dk(  r,||	z  dk(  r$t         t        ||      t        ||	      gd      S t        ||      }
t        ||	      }d fd}t        ||
|g||g||	gt        j                         fd}t        j                   j                          j                         |t!        |            }|S )	Nr   r/   )divisor_overridec                F    t        | |z  t        j                  |            S r   )r(   r   r0  r(  s      rf   rB  z0upsample_nearest2d_backward.<locals>.start_index  s    uwg(>??rh   c                     | dz   ||      S r  r   )r:  r
  r  rB  s      rf   rC  z.upsample_nearest2d_backward.<locals>.end_index  s    EAI99rh   r,  c                (     | t                    S r   )r  )r	  r/  r   s    rf   r   z'upsample_nearest2d_backward.<locals>.fn  s    c.q122rh   r]  )r8  r   rT   r^   r   r  r1  r)   rG   r  rS   r   r@   r   ro   r   rj   )r   r  r  rH  rI  r  inp_hinp_wout_hout_wr5  r6  rC  r   r7  r/  rB  s   `              @@rf   upsample_nearest2d_backwardrq    s1    NNJJLVUEGG&&u-EGG&&u-E 'VUEu}eemq0&(>?RS
 	
 5%(L5%(L@: ""L1%.77F3 
		||~kkmJ
B Irh   c           
     (    t        | ||||||d      S )Nr-   r  _avg_poolndr   r  rV  rd  r  count_include_padri  s          rf   r1  r1    )     		 	rh   c           
     (    t        | ||||||d      S )Nr   r  rs  ru  s          rf   
avg_pool3dry    rw  rh   c                v   ssdgz  t              t              t              t        | t              sJ t              k(  sJ t              k(  sJ t              k(  sJ t        | j	                               dz   dz   fv sJ | j                          | j	                         d   }| j	                          d  t        t              D 	cg c]  }	t        |	   |	|       c}	 \  }
}t              st        |      rt        | d      d}n| j                         d}t        |      t        |
      z   }| j                         t        j                  t         j"                        }|dkD  r4dk(  rt$        }nd	k(  rt&        }nt)        d
        || ||      S fd|r|r)|r|n|j*                  rdz  fd}nfd}nf	d}t-        j.                  | j1                         ||      }|S c c}	w )Nr   r/   r-   rf  r  TFr  r   zUnknown dim: c           	     T   | d 	  }| 	 d  }d }t        j                  t        	      D cg c]  }t        
|          c} D ]V  }t        	      D cg c]  }||   |   z  ||   z   |   z
   }} |g ||      }||}At        j                  ||      }X |S c c}w c c}w r   )r   r  r  rS   r   )r	  r  r  r/  totalrt   r  r   r  r   r  rd  rV  s            rf   r/  z_avg_poolnd.<locals>.fn_sum)  s    UsdJ##U3Z%PeKN&;%PQ 	,BBG*MQ1Q4&)#be+gaj8MCM6C)C}U+	,  &QMs   B B%c                f    t        j                   |       t        j                              S r   )rS   r^  ri  )r	  r   r/  r  r  s    rf   r   z_avg_poolnd.<locals>.fn;  s&    wwvc84cll5%6PQQrh   c                f    t        j                   |       t        j                              S r   )rS   truncdivri  )r	  divisorr   r/  r  s    rf   r   z_avg_poolnd.<locals>.fn@  s'    ||F3$93<<QV;WXXrh   c                F  	 | 	 d  }g }t        	      D ]  }||   |   z  |   z
  }t        j                  ||   z   |   |   z         }s/t        j                  |d      }t        j                  ||         }t	        j
                  ||z
  t        j                        }|j                  |        t        j                  t        j                  |      }
j                  rt	        j                   |       |      S t	        j                   |       |      S r  )r  r   MinMaxrS   rg  r|   r  rp   r  r  r^  r  r.  r  )r	  r  divide_factorsrt   hstarthendfactordivide_factorrv  r   r   r/  r}  r  rd  rV  r  s           rf   r   z_avg_poolnd.<locals>.fnF  s   cTUBN3Z .A*WQZ7yy+a.!8!A$:KL("YYvq1F 99T1Q40Dvu{{C%%f-. &,,SWWnEM&&{{6#x#8-HH<<sH 5}EErh   r]  )rN   rl   rD   r   r   r8  r  r  r  r  r  r_  rj   r   r  r  r  r^  fallback_avg_pool2dfallback_avg_pool3drE  r  r@   r   ro   )r   r  rV  rd  r  rv  ri  r   r:  rt   r  
ceil_modeshad_paddingr  r  fallbackr   r7  r  r   r/  r}  r  r  s    ``` ` `          @@@@@@rf   rt  rt    sT    #){C0K&#&F7C(Ga###{s"""v;#w<3qzz|q#' 2222NNJJL3$E	

cTUA 3Z

 1q+vw	J

E: 7|s:.q#3?==?E{T%[(HKKME""8<<=KR!8*HAX*H}SE233
 	
 *&6"K""KER R
Y Y	F 	F$ 
		||~	
B Iu

s   -H6c                   dk7  sJ d       ssddgt        | t              sJ t        |t              sJ t              dk(  sJ t              dk(  sJ t              dk(  sJ t        |j                               dv sJ | j	                          |j                         ^ }t        d|      \  }	}
t        d|      \  }}| j                         d   xs d   xs |
xs || j                         ^ }t        |j                               }|j                         }t        fdt        d   dz        D              t        fdt        d   dz        D              z  }|dkD  rt        | ||      S fd	fd
}t        j                  | j                         |||      }|S )Nr   divisor must be not zeror-   r  r/   c              3     K   | ]=  }t        t        |d          t        d t        |d    z
  d                z
  d       ? ywr  r  r  s     rf   r   z&avg_pool2d_backward.<locals>.<genexpr>  r  r  c              3     K   | ]=  }t        t        |d          t        dt        |d    z
  d                z
  d        ? ywr  r  r  s     rf   r   z&avg_pool2d_backward.<locals>.<genexpr>  r  r  r  c           	     @   t        j                  d   t        j                        }t        j                  d   t        j                        }t        j                  d   t        j                        }t        j                  d   t        j                        }t        j                  d   t        j                        }t        j                  d   t        j                        }t        j                  t        j
                  | |      |      }t        j                  t        j
                  ||      |      }	t        j                  t        j                  ||      t        j                  t        j                  t        j                        |            }
t        j                  t        j                  |	|      t        j                  t        j                  t        j                        |            }t        j                  |t        j                  dt        j                              }t        j                  |	t        j                  dt        j                              }	t        j                  |
t        j                  t        j                              }
t        j                  |t        j                  t        j                              }t        j
                  t        j                  |
|      t        j                  ||	            }|S )z{
        This computes the scaling factor that we will divide an element
        by when `count_include_pad=False`
        r   r/   )
rS   ri  r|   r  r  r^  r  r   rg  r  )r  r  stride_hstride_wpad_hpad_wkernel_hkernel_wr  wstartr  wendr  heightr  rd  rV  r  s                rf   !compute_pool_size_without_paddingz>avg_pool2d_backward.<locals>.compute_pool_size_without_padding  s   
 <<q	5;;7<<q	5;;7WQZ5WQZ5<<A<<<A<X.6X.6{{GGFH%GGCNN65;;7?
 {{GGFH%GGCNN5%++6>
 VS\\!U[[%ABVS\\!U[[%AB{{4!DE{{4u{{!CDf 5swwtV7LMrh   c                8   | ^ }}}|d   z   }|d   z   }t        j                  t        |d   z
  d   z   d         t        j                        }t        j                  t        |d   z
  d   z   d         t        j                        }t        j                  t        |d         dz   t        j                        }t        j                  t        |d         dz   t        j                        }t        j
                  |t        j                  dt        j                              }t        j
                  |t        j                  dt        j                              }t        j                  |t        j                  t        j                              }t        j                  |t        j                  t        j                              }d }t              D ]+  }	t              D ]  }
t        j                  |t        j                  |	t        j                              }t        j                  |t        j                  |
t        j                              }}nssd   d   z  }n	 ||      }t        j                   g |t        j                  t        j                  |t        j                  |t        j                  dt        j                                    d      t        j                  t        j                  |t        j                  |t        j                  dt        j                                    d            |      }t        j                  t        j                  ||      t        j                  ||            }|;t        j                  ||t        j                  dt        j                               }t        j                  |t        j                  ||      |      } . |J |S )Nr   r/   Fr  rf  )rS   rg  r)   r|   r  r  ri  r  r  r   r.  r  r  rl  rk  r  r  )r	  r  r}  r  r  r  r  r  r  r  r  r  r  r  partrs  r  rv  ri  r  r  r  r  rd  r  r  rV  r  s                   rf   r   zavg_pool2d_backward.<locals>.fn  s2   A
N
N..QQ'&)3VAY?
 ..QQ'&)3VAY?
 x6!9595;;Gx6!9595;;G++gs||Au{{'CD++gs||Au{{'CDE3>>-#MNE3>>,#LM' *	RC]+ )RWWWcll3&DEWWWcll3&DE#/,E&k'N[^;E=b"EE{{#11 #$&s||Au{{7S(T!" !.&+  11 #$&s||Au{{7S(T!" !-&+& ). xxFF2u%FF2u% #"yytS\\#u}}5UVH"yyswwx/FQHS)R*	RV ###rh   r]  )rl   rD   r   r   r8  r  r_  rj   r   r  r  fallback_avg_pool2d_backwardr@   r   ro   )r  r   r  rV  rd  r  rv  ri  r   _h_out
ceil_mode1_w_out
ceil_mode2r  r   r  r   r7  r  r  r  r  r  r  r  r  r  s     ``` ``          @@@@@@@@@rf   avg_pool2d_backwardr  g  s    #'71'<X>XX<a&k9---a###{q   v;!w<1qzz|&&&

Q%;FJ &eQVWiXFJ))+K!*F
FjFJK&1&:&:&<#Q|AJJL!HKKME {1~)* M  {1~)* M
  -/KR+	
 		
 8? ?B 
		%%'	
B Irh   c                    dk7  sJ d       ssg dt        | t              sJ t        |t              sJ t              dk(  sJ t              dk(  sJ t              dk(  sJ t        |j                               dv sJ | j	                          |j                         ^ } t        d|      \  }	}
t        d|      \  }}t         d|      \  }}| j                         t              xs
 |
xs |xs || j                         ^ }t        |j                               }|j                         }fdt        d      D        \  z  z  }|d	kD  rt        | ||      S  fd
fd}t        j                  | j                         |||      }|S )Nr   r  )r   r   r   r   )r   r   r/   r-   c              3  j   K   | ])  t        fd t           dz        D               + yw)c           
   3  t   K   | ]/  }t        |   z  t        d |   z
     z        z
  d       1 ywr  )r  )r   r  rt   r  rV  s     rf   r   z0avg_pool3d_backward.<locals>.<genexpr>.<genexpr>@  sF      
 VAYQ[^);q	(I!JJAN
s   58r-   N)r  r  )r   rt   r  rV  s    @rf   r   z&avg_pool3d_backward.<locals>.<genexpr>?  s;      3
 	 	 
;q>A-.
 	
3s   /3}   c           	        d D        \  }}}d D        \  }}}d D        \  }	}
}d t        | ||g|||g|||g      D        \  }}}d t        |||g|	|
|gg|||g      D        \  }}}d |||fD        \  }}}d t        |||gg      D        \  }}}t        j                  t        j                  t        j                  ||      t        j                  ||            t        j                  ||            }|S )Nc              3  d   K   | ](  }t        j                  |t        j                         * y wr   rS   ri  r|   r  )r   r   s     rf   r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>V  s     'UQ(D'U   .0c              3  d   K   | ](  }t        j                  |t        j                         * y wr   r  rp  s     rf   r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>W  s     Ms||Au{{;Mr  c              3  d   K   | ](  }t        j                  |t        j                         * y wr   r  )r   r   s     rf   r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>X  s$      (
-.CLLEKK((
r  c              3  x   K   | ]2  \  }}}t        j                  t        j                  ||      |       4 y wr   )rS   r  r^  )r   r  r   pads       rf   r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>\  s3      "
1c GGCGGAqM3'"
s   8:c           
   3     K   | ]i  \  }}}}t        j                  t        j                  ||      t        j                  t        j                  |t        j
                        |             k y wr   )rS   r  r   rg  r|   r  )r   r?  r   r   r  s        rf   r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>b  sU      

 #q#s KKq!3773>>#u{{+KS#Q

s   A/A1c              3     K   | ]<  }t        j                  |t        j                  d t        j                               > ywro  rS   r  ri  r|   r  )r   r?  s     rf   r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>m  s2      "
 KKs||Au{{;<"
   AAc              3     K   | ]?  \  }}t        j                  |t        j                  |t        j                               A y wr   rS   r  rg  r|   r  )r   r@  r   s      rf   r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>q  s6      
S KKS^^C=>
   AA)r  rS   r^  r  )pdr  r  stride_dr  r  pad_dr  r  kernel_dr  r  dstartr  r  dendr  r  r  depthr  r  rd  rV  r  s                      rf   r  z>avg_pool3d_backward.<locals>.compute_pool_size_without_paddingU  sH   'Uf'U$(HMWMue(
2=(
$(H"
 Rx8<ueU>S"


 '*(8X.&u%	'	

dD"
 &&1"

tT 2UFE4JK
dD GGCGGD&)3774+@A3774QWCX
 rh   c                .   | ^ }}}}d t        |||g      D        \  }}}d t        |||g!      D        \  }}}d t        |||g!      D        \  }}	}
d |||fD        \  }}}d t        ||	|
g g      D        \  }}	}
d }t              D ]{  }t              D ]i  }t        "      D ]W  }d t        |||g|||g      D        \  }}}}n ssd   d   z  d	   z  }n
 |||      }t        j                   g |t        j                  t        j
                  |t        j                  |t        j                  dt        j                                    d
      t        j                  t        j
                  |t        j                  |	t        j                  dt        j                                    d
      t        j                  t        j
                  |t        j                  |
t        j                  dt        j                                     d
            |      }t        j                  t        j                  t        j                  ||      t        j                  ||	            t        j                  ||
            }|;t        j                  ||t        j                  dt        j                              }-t        j                  |t        j                  ||      |      }Z l ~ |J |S )Nc              3  ,   K   | ]  \  }}||z     y wr   r   )r   r  r  s      rf   r   z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>|  s     Avq#1s7Ar  c              3     K   | ]<  \  }}}t        j                  t        ||z
  |z   |      t        j                         > y wr   rS   rg  r)   r|   r  )r   r  r   r   s       rf   r   z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>~  s<      %
1a NN8AEAIq15;;?%
r  c              3     K   | ]8  \  }}t        j                  t        ||      d z   t        j                         : ywr  r  )r   r  r   s      rf   r   z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>  s6      
1 NN8Aq>A-u{{;
s   >A c              3     K   | ]<  }t        j                  |t        j                  d t        j                               > ywro  r  )r   pstarts     rf   r   z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>  s2      %
 KKQ <=%
r  c              3     K   | ]?  \  }}t        j                  |t        j                  |t        j                               A y wr   r  )r   pend
pooled_dims      rf   r   z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>  s6      
 j KKcnnZEF
r  c              3     K   | ]?  \  }}t        j                  |t        j                  |t        j                               A y wr   )rS   r   ri  r|   r  )r   r  p_s      rf   r   z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>  s6      "&FB R(EF"r  r   r/   r-   Fr  rf  )r  r  rS   r.  r  r  r  ri  r|   r  rl  rk  r  r  r   )#r	  r  r  r}  r  pdstartr  r  pdendr  r  r  pd_r  r  r  r  r  r  r  rs  r  rv  d_window_sizeri  r  r  r  r  rd  pooled_depthr  r  rV  r  s#                        rf   r   zavg_pool3d_backward.<locals>.fnz  s   AqAaAY)@A1a%
1ay+v>%
!'

Q1Iv.
ue
%
"GW5%
!'
$'u%m\'R%
ue ' 8	VC]+ 7V / 6VC"*-$gw7#sC+"JBB (3 0*+ +AQ ?+a. P A"b" M;;#!' # 5 5$'KK(*CGGE3<<5;;;W,X%& %1*/!" !$ 5 5$'KK(*CGGE3<<5;;;W,X%& %2*/!"  !$ 5 5$'KK(*CGGE3<<5;;;W,X%& %1*/!"!4 7D< 88E!2CFF2u4EFr5)D  '#&99 $S%--(H$ $'99T3778T3JH#Um6V7V8	Vr ###rh   r]  )rl   rD   r   r   r8  r  r_  r  rj   r   r  fallback_avg_pool3d_backwardr@   r   ro   )!r  r   r  rV  rd  r  rv  ri  r  _d_outceil_mode_dr  ceil_mode_hr  ceil_mode_wr   r  r   r  r   r7  r  r  r  r  r  r  r  r  r  r  r  r  s!     ``` ``             @@@@@@@@@@@@rf   avg_pool3d_backwardr    s    #'71'<X>XX<k9---a###{q   v;!w<1qzz|&&&$%JJL!VUFE&q+vw	FK ';FK 'q+vw	FK ))+Kg,K+KKK4?4H4H4J1Qm\AJJL!HKKME3
 q3/M=-  -/-?KS+	
 		
# #JU U Un 
		%%'	
B Irh   c                   | j                         }t        |t              r|g}n|st        t	        |            }t	        |      dk(  rt        |      dv s
J d|        g S t        |      }t        t	        |            D ]_  }||   dk  r#||xx   t	        |      rt	        |      ndz  cc<   d||   cxk  rt	        |      k  rFn t	        |      dk(  r	||   dk(  r_J  t	        t        |            t	        |      k(  sJ d       |S )Nr   )r   r  rL  zinvalid axis: r/   zreduction axis not unique)r   rl   r   r  r   r   rj   r'   )r   r  r   rt   s       rf   _validate_reduction_axisr    s    ::<D$vSY
4yA~T{//H>$1HH/	:D3t9 M7Q;GCIs4y14GDG'c$i'CINtAw!|LLM z$ CI-J/JJ-Krh   )r   c               J   |t        | |      } | j                         t        t           t	        | |            }g }g g g t        t                    D ]Q  }||v r&j                  |       j                  |          -j                  |       |j                  |          S d|dv rt              dkD  rt        |       rt        | j                  t              rdnt        | j                  t        j                        sRt        | j                  t        j                        rat        | j                  j                  t        j                        r3| j!                         }|j#                         xs |j%                          fd}	r0t'              }
D ]  }t(        j*                  j,                  |
|<   ! n|}
| j/                         t1        | j3                         |xs | j5                         | j5                         |	|
      S )NF)r  argminr/   Tc                |   t        |      t              k(  sJ 
r+t        |       t              k(  sJ D cg c]  }| |   	 } }t        |       t              k(  sJ d gt        |       t        |      z   z  }t        j                  t        |       t        |            D ]
  \  }}|||<     	|      }ru|D cg c]  }t	        j
                  |       }}|d   }t        dt        |            D ]  }||   z  ||   z   } |t        j                  |t        j                        fS |S c c}w c c}w r  )r   r   r   r  r   r-  r  rS   rg  r|   rh  )r:  reduction_indexrt   r  r	  varr   rindex
linear_idxinner_loaderkeepdimskept_idxreduced_idxreduced_sizesshould_compute_logical_indexr   s            rf   r  z%_make_reduction_inner.<locals>.loader  sL   ?#s;'7777u:T***'/0!U1X0E05zS]***Fc%j3+??@	!% #k?"C
 	!HC !IcN	! Y' (/>?!ell1o?F?  J1c&k* G'-*::VAYF
G 3>>*ekkBCC+ 1 @s   D4D9)rw   r  r  r\  r[  r  )r   r   r'   r   r  r  r   rp   r;   rl   rn   r?   r2   r   r   Bufferr:  is_transposedis_contiguousrj   r   r)  r*  r_  r  ro   r   )r   r  r  r   rB  r   
kept_sizesrt   rQ  r  r  r  r  r  r  r  r   s     `        @@@@@@rf   _make_reduction_innerr    s    Q::<Dc?3At<=DJHMK3t9 '9q!  a)OOAd1g&' $) .."aLaffk*+/( 2 23qvvr}}-*QVV[["))2T\\^F$$&Df.B.B.D*D ) 6 : 	&A''++HQK	& ==?L||~'81;;=++-& rh   c                     dd d fd}|S )Nr   c                   t        | |||      }t        j                  d| d|}t        |j                  j                  t              r|j                          |S )N)r  r  r   rB  r   )r   r  r   )r  rA   r   rl   rn   rs  )r   r  r  r   r   r_  rB  r   s         rf   ri  zmake_reduction.<locals>.inner?  se    &"7)
 !!XAXQWXKK
 NNrh   r   r   )r   rB  ri  s   `` rf   make_reductionr  >  s    T " Lrh   c                   |t        | |      } t        | |      }t        | j                         | j	                         f| j                         f| j                         |      S )N)rw   dtypes	inner_fnsr   r  )r   r/  r  ro   r   r_  r   )r   r  r   s      rf   _make_scan_innerr  S  sZ    QD!D||~==?$ZZ\ rh   r   c                   |t        | |      } | j                         t        | |      }| j                         }|t        j
                  t        j                  fv rt        | t        j                        } t        | ||      }t        fd|D              }t        j                  || j                         | j                               }t        j                  |t        |j                                     }t        t!        ||      |      S )Nc              3  (   K   | ]	  }|     y wr   r   r   rt   r   s     rf   r   zmean.<locals>.<genexpr>l       0a$q'0r   r9  )r   r   r  r   r|   rb  ra  r5  sum_rQ   r2   r9   ro   r8   r   rj   div)r   r  keepdimr   output_dtype
sum_resultdenomr   s          @rf   r#  r#  a  s    Q::<D#At,D;;=Lu~~66Q$aw'J0400Ee1;;=XEeT**=*=*?%@AEC
E*L99rh   c                2  
 |d}| j                         
t        | |      }t        | |d      }|r|j                          t	        t        | |            }t        |||      }t        
fd|D              }|rt        j                  ||z
  d      }t        j                  || j                         | j                               }t        j                  |t!        |j                                     }t#        ||      }	|s|	fS |r|nt%        ||      }|	|fS )Nr/   T)r  c              3  (   K   | ]	  }|     y wr   r   r  s     rf   r   z var_mean_sum_.<locals>.<genexpr>  r  r   r   r9  )r   r  r#  rs  squarer  r  rQ   r   r  r2   r9   r   ro   r8   r   rj   r  r  )r   r  
correctionr  return_meanx_meandiffsr  r  x_varr   s             @rf   var_mean_sum_r  r  s    
::<D#At,D!T4(F3q&>"EeT7+J0400E		%*,a0e1;;=XEeT**=*=*?%@AE
E"ExVGFD$9F&=rh   c                    t        | |      }t        | ||d d       }|d   }t        |d         }t        |t        j
                        xr, t        |      t        j                  k  xr t        |      dk7  S )Nr  r  r   rB  r[  r  r/   )	r  r  rQ   rl   r   rP  r   r0   r  )r   r  r  r   r[  reduction_numels         rf   use_two_step_variancer    s|    #At,D"	wd$F HF#F+=$>?O?EMM2 	' 6#E#EE	'&!Q&rh   c                  dt        | ||d d       }|j                  d      }|j                  d       |j                  d       t        j                  j                  d|fd| j                         d|\  }}}	|j                          | j                         | j                         t        | |      }t        fd|D              d	 fd
}
 t        |
      |      }|r|j                          ||fS |fS )Nr/   r  r\  r  r  welford_reduce)r  r   r   c              3  (   K   | ]	  }|     y wr   r   r  s     rf   r   z$var_mean_welford_.<locals>.<genexpr>  s     1q471r   c                    t        | t        j                        rD| j                  s8t	        j
                  t	        j                  | t        j                        |      S t	        j                  | |      S r   )
rl   r   r   	is_numberrS   r   rg  r|   rh  ri  r;  s     rf   get_constant_or_index_exprz5var_mean_welford_.<locals>.get_constant_or_index_expr  sG    a$Q[[<<q%++ >FF||Au%%rh   c                           }       }t        j                  d      }| t        j                  |||z
        z  S r  )rS   ri  r  )rn   cNzeror  r   r  rnumels       rf   r3  z#var_mean_welford_.<locals>.scale_fn  sE    &z59&vu5||Au%ckk$A...rh   r   )r  r  r2   WelfordReductionr   r   rs  r   r  rQ   rj  )r   r  r  r  r   r   r  r#  m2r   r3  r  r   r  r  r   s     `         @@@@rf   var_mean_welford_r    s    
"	wd$F ZZ
#F
JJ{
JJ{%%,, )'kkm 	KD"a JJLKKME::<D#At,D1D11F&
/ #.
"2
&CDy6Mrh   c                   | j                         t              }t        | |d      } t        | ||||      }t	        | ||      rt        di |n
t        di |}t        fd|D              }|s|d   S |S )NFr  )r   r  r  r  r   )r  r  c              3  :   K   | ]  }t        |d         yw)Fr  N)r   )r   r   r  s     rf   r   z#var_mean_helper_.<locals>.<genexpr>  s     F!8Ayu55Fs   r   r   )r   r   r   r  r  r  r  r   )	r   r  r  r  r   compute_dtyper   r0  r  s	           @rf   var_mean_helper_r    s    I))4MM.A
F !w? 	(( 
 FvFFF'6!93V3rh   )r  r  c               "    t        | |||d      S )NFr  r  r  r   r  r   r  r  r  s       rf   var_r    s    	W% rh   c               "    t        | |||d      S )NTr  r  r  s       rf   var_meanr!    s    	W$ rh   c                   |dk  r!t        t        j                  |       | |      S |dk(  rt        j                  d|      S |dk(  r| S t        | |dz  |      }t        j                  ||      }|dz  dk(  rt        j                  ||       }|S )Nr   r/   r-   )pow_recursiverS   r  ri  r^  )r   r1  r   r_  s       rf   r#  r#    s    1uS^^A.E::Av||Au%%Av1a1fe,FWWVV$F	A!|#Mrh   c                .    t        j                  | |      S r   )rS   powr   r/  s     rf   
pow_nativer'    s    771a=rh   )r  c                v    t        t              r#t              k(  rt         t                    S t        t              rdk(  rt	               S t        t              rdk(  rt               S t        d  fD              }t        |      }t        t              xr dcxk  xr dk  nc xs	 |xr dk\  }|rZ j                          fd}t        j                   j                          j                         | j                               S t         t              r: dk(  rt        d      S  d	k(  r$t!        j                               rt#              S |rDt         t              rt%               S t        t              rt'               S t)               S t+               S )
Nr1  r/   c              3  r   K   | ]/  }t        |t        j                        s|j                          1 y wr   )rl   r2   rD   r   r  s     rf   r   zpow.<locals>.<genexpr>  s#     N1*Q2MNs   77i    r   c                F    t         |       j                               S r   )r#  r   )r	  r   r/  r  s    rf   r   zpow.<locals>.fn  s     a??rh   r]  r-   )rl   r5  r   r%  sqrtr  r?  r    r_  r@   r   ro   r   r   r!   r  r   exp2fallback_pow_scalarfallback_pow_tensor_scalarfallback_pow_tensor_tensorr'  )r   r/  r   is_integer_powembed_exponentr   r  s   ``    @rf   r%  r%    sx   !USV1c!f~	Au	!s(Aw	As	QQx NANNE%e,N  3' a"32AF  	@ <<>++-::<	
 	
 !V6Q?"6nQ[[]37Na &q!,,6"-a33-a33arh   c                   t        | t              r| j                  }n| }t        |t              r|j                  }t        |t        j                        st        j                  | j                         | j                         |j                         | j                               }t        |t        t        f      sJ |j                  }t        |t        j                        sJ t        |t        j                        rg|j                         sW|j                         sGt        |j                  t        j                        s#|j!                          |j                  |_        | S t        j"                  j%                  |||       | S )Nr]  unsafe_alias)rl   rD   rn   r2   r   r@   r   ro   r   r_  r   r6   r<   is_input_bufferis_module_buffer	NopKernelrs  r  realize_into)changedr  r5  changed_datard   s        rf   r  r  7  s*   '9%||#y!hhc2==)%%'##%__&##%	
 $: 6777ii#r}}---,.$$&((*l''6 	HH!!..\ /  Nrh   c                .    t        | t        | |            S r   )r  r  )r   rf  s     rf   rc  rc  \  s    Q	!Z011rh   c                    | |u r| S t        || j                               }t        || j                               }t	        || j                               }t        | |      S r   r  ro   r   r   r-  r   r  )r  r  r  s      rf   r	  r	  a  sR    
cz

C)
*C
3
(C
clln
%CS#rh   c                .    t        j                  | |      S r   )rS   floordivr&  s     rf   r@  r@  l      <<1rh   c                .    t        j                  | |      S r   )rS   r  r&  s     rf   r  r  q  rA  rh   c                >   t        |       xr t        |      }t        |       xr t        |      }|dk(  r,|rJ d       |rt        | |      S t        t	        | |            S |dk(  r,|rJ d       |rt        | |      S t        t	        | |            S t	        | |      S )Nr  z5floordiv operands can not be boolean at the same timer  z5truncdiv operands can not be boolean at the same time)r   r   r@  r  r  r  r  )r   r/  rounding_modeboth_integerboth_booleans        rf   div_moderG  v  s    "1%</!*<L"1%</!*<L X!XX!-x1~C5Q3CCX!XX!-x1~C5Q3CCq!9rh   c                    t        |       xr t        |      }|rt        | |      S t        t        j                  j
                        } t        |      | |      S r   )r   logical_andr>   r  r^  r  rj  )r   r/  	both_boolr   s       rf   r^  r^    sN    "9q'9I1a  **+!~b!!Q''rh   c                f   t        | t        j                        rt        | j                        S t        | t        j
                        rt        | j                               S t        | t        j                        r| S t        | t        j                        syt        j                  j                  j                  | j                               }t        j                  |      5  t!        j"                  t        j$                  dd      5   | j&                  | j)                          }ddd       ddd       t        t        j                  j*                  j,                        sJ t        |j.                  t        j                        r|j.                  S y# 1 sw Y   rxY w# 1 sw Y   vxY w)z:Try convert an arbitrary IR node into an ir.Constant valueNallow_indexingT)rl   r2   r<   get_constant_valuern   r6   rR  r   Loopsr|   	_inductorops_handlerExtractConstantsHandlerro   rT   set_ops_handlerr   objectr  r\  inner_fn_argsvirtualizedOpsValuer   )r   r>  rr   s      rf   rM  rM    s/   
 !R]]#!!&&))!R[[!!!--/22!R[[! a"oo))AA!,,.QG	'"-R&&(8$?- ajj!//+,	- - c5??66??@@@#))R[[)yy- - - -s$   &&F'F*F'F$	 F''F0c                l   t        d | |fD              }|rt        | |      S t        |      x}q| j                         j                  dk7  rT|j
                  dk(  r*t        j                  t        d      |j
                        }nd|j
                  z  }t        | |      S d } t        |      | |      S )Nc              3  L   K   | ]  }t        |      xs t        |        y wr   )r   r   r  s     rf   r   zdiv_prim.<locals>.<genexpr>  s"     O1oa(>OA,>>Os   "$r   r   infr  c                 &    t        j                  |  S r   )rS   r.  r  s    rf   r   zdiv_prim.<locals>.fn  s    {{D!!rh   )r  r  rM  ro   r   r   mathcopysignr5  r^  rj  )r   r/  is_integralr  r  r   s         rf   div_primr^    s    OAOOK1~ &a((5!,,.:M:MQV:V ==AuU|W]]CJw}},J1j!!" >"a##rh   c                Z    t        | |ft        j                        \  } }t        | |      S rb  )rD  r   INT_TO_FLOATr^  r&  s     rf   r  r    s0     	
A$C$P$PDAq Aq>rh   c                h    t        |       xs t        |       }|rd }nd } t        |      | |      S )Nc                .    t        j                  | |      S r   )rS   modr&  s     rf   r   zfmod.<locals>.fn  s    771a= rh   c                .    t        j                  | |      S r   )rS   fmodr&  s     rf   r   zfmod.<locals>.fn  s    88Aq>!rh   )r   r   rj  )r   r/  r]  r   s       rf   re  re    s9    !!$:(:K	!
	" >"a##rh   c                   t        | j                               st        | j                               r|t        j                  }t        d|      } || |||      S )Nr  r  r   r    r   r   r|   rh  r  r   r  r  r   r   s        rf   r  r    sK     	'+;AKKM+J
-	U	;Baxu--rh   c                   t        | j                               st        | j                               r|t        j                  }t        | j                               dk(  r(|dv sJ |xs | j                         }t        | |d      S d }t        | ||      }t        j                  j                  d	i |d|i\  }|t        | ||      S |S )
Nr   r   rL  Tr  c                @    | \  }|\  }t        j                  ||      fS r   )rS   r   a_tupleb_tupler   r/  s       rf   
combine_fnzcumsum.<locals>.combine_fn  #    1rh   r  r   ro  r   r   r   )r    r   r   r|   rh  r   r   r   r  r2   Scanr   fallback_cumsumr   r  r   ro  r   r_  s         rf   cumsumrv    s     	'+;AKKM+J
-
1::<Aw&5t,, 
 ad%8F??J?IV~qd%88Mrh   c                   t        | j                               st        | j                               r|t        j                  }t        | j                               dk(  r(|dv sJ |xs | j                         }t        | |d      S d }t        | ||      }t        j                  j                  d	i |d|i\  }|t        | ||      S |S )
Nr   rj  Tr  c                @    | \  }|\  }t        j                  ||      fS r   )rS   r^  rl  s       rf   ro  zcumprod.<locals>.combine_fn  rp  rh   rq  ro  rr  r   )r    r   r   r|   rh  r   r   r   r  r2   rs  r   fallback_cumprodru  s         rf   cumprodrz    s     	'+;AKKM+J
-
1::<Aw&5t,, 
 ad%8F??J?IV~t599Mrh   c                
   d }| j                         }t        | j                               dk(  r|dv sJ t        |       S t	        | ||      }t        j                  j                  di |d|i\  }|t        | |      S |S )Nc           	     *   | \  }|\  }t        j                  ||      }t        j                  ||      }||k7  t        j                  |       z  }t        j                  |t        j
                  t        j                  ||z
              |z   |      fS r   )rS   r  r  r  r  log1pexp)rm  rn  r   r/  min_vmax_vrs  s          rf   log_add_exp_helperz(logcumsumexp.<locals>.log_add_exp_helper)  s{    Aq!Aq!CIIe$4#45		$		#''%%-*@ AE I1MOOrh   r   rj  rq  ro  r  r   )	r   r   r   r  r  r2   rs  r   fallback_logcumsumexp)r   r   r  r   r   r_  s         rf   logcumsumexpr  '  s    P KKME
1::<Ag~~Qxac7FGG4FGIV~$QC00Mrh   c                   t        | j                               dk(  r,dv sJ t        |       t        | t        j
                        fS | j                         }t        j                  d|d      }t        | |      }|t        j
                  f|d<   | j                         fd	f|d
<   t        j                  j                  di |d|i\  }}|t        |       S ||fS )Nr   rj  r   r  Fr   arg_break_ties_leftrq  r  c                R    t        j                  |    t        j                        S r   rS   rg  r|   rh  r	  r  s    rf   r#  zcummax.<locals>.<lambda>L      CNN3t9ekk: rh   r  ro  r  r   )r   r   r  r  r|   rh  r   r2   get_reduction_combine_fnr  r_  rs  r   fallback_cummaxr   r  r   ro  r   r   r  s    `     rf   cummaxr  =      
1::<AwQxAU[[999KKME,,5J ad%8Fu{{+F8	:F; ggnnEvE*EOFG~qd++7?rh   c                   t        | j                               dk(  r,dv sJ t        |       t        | t        j
                        fS | j                         }t        j                  d|d      }t        | |      }|t        j
                  f|d<   | j                         fd	f|d
<   t        j                  j                  di |d|i\  }}|t        |       S ||fS )Nr   rj  r   r  Fr  rq  r  c                R    t        j                  |    t        j                        S r   r  r  s    rf   r#  zcummin.<locals>.<lambda>c  r  rh   r  ro  r  r   )r   r   r  r  r|   rh  r   r2   r  r  r_  rs  r   fallback_cumminr  s    `     rf   cumminr  T  r  rh   c                   t        | j                               st        | j                               r|t        j                  }t        d|      } || |||      S )Nr  r  r   rg  rh  s        rf   r  r  k  sK     	'+;AKKM+J
-	e	<Baxu--rh   c                ^    t        | t        j                        }  t        d      | ||      S )Nr  r  r  )r   r|   r   r  r   r   r  s      rf   
reduce_anyr  v  s(    EJJA >% w??rh   c                Z    |t        | ||      t        | ||      fS t        | d |      S Nr  )reduce_amaxreduce_argmaxr  s      rf   
reduce_maxr  |  :    
g6!#8
 	

 qtg66rh   c                Z    |t        | ||      t        | ||      fS t        | d |      S r  )reduce_aminreduce_argminr  s      rf   
reduce_minr    r  rh   xor_sumr  r  r  r  r  
logical_or)rf  rZ  stabler   
descendingc          	        |d}| j                         }| j                         }t        t        |      |      }t        |      dk(  r't	        |       t        d|t        j                  |      fS t        |      r||   nd}t        j                  j                  j                  |t        j                  t        j                        j                        st        | |||      S t!        |ddt        j                  |d      }dgt        |      z  }t        |      r|||<   t#        ||      }t%        ||      }t&        j(                  j+                  || j,                  |j,                  f| j/                         |j/                         f||||      \  }	}|	t        | |||      S |J |	t1        |t        j                        fS )NFr   r/   r  )r?  rA  r   rw   r1  )rw   r  r  r   r  r  r  )r   ro   r   r   r  rg  r|   rh  rT   r^   r   statically_known_ltr  int16r  sort_fallbackr2  r  r-  r2   Sortr   r   r_  r   )
r   r  r   r  r  rw   r'  r  
view_shaper   s
             rf   sort_stabler    s   ~JJLE\\^F
3u:s
+C
5zQQxq&%++u=== ZuSzQH77//%++ekk:R:V:VWQv3:NNVSXG s5z!J
5z"
37J'GWe$Gggnn'==?G$7$7$9: % OFG ~Qv3:NN8GU[[111rh   c                     t        | d||      S )NFr  )r  )r   r   r  s      rf   sortr    s    qCJGGrh   c                <    t        | |t        j                  |      S )Nr   r   rh  )r  r   r`  )ra   r   rh  s      rf   register_pointwise_numericr    s!    
;HH'	 rh   c                b    t        | j                         t        | t        j                        S rb  )rP   r  r  r   r`  r  s    rf    register_pointwise_numeric_ldf64r    s'    '4
;HH rh   r  logical_not)rZ  )r   r  rB  identity)rq  pointwise_overrides_datac              #  F  K   t         |   t        | j                  d       }|y fd}t        |t        j
                  j                        r9|j                         D ]%  }t        ||      }|j                   ||      f ' y |j                   ||      f y w)Nc                4    j                   t        |       S y r   )tritonr  )ra   rn   s    rf   make_triton_fallbackz6_get_pointwise_overrides.<locals>.make_triton_fallbackF  s    ;;#B'' rh   )	r  r   r   rl   r|   r}   r   r   r   )nsr   ra   r  olnamers  rn   s         @rf   _get_pointwise_overridesr  @  s     #D)D	TYY	%B	z( "ejj112lln 	IFV$Bd..0DR0HHH	I $**,@,DDDs   BB!r  c                d    | t         |<   t        j                  |        fd}t        | |       y )Nc                      | i |}g }t        | d   |      D ]"  \  }}|j                  t        ||d             $ |S )Nr   Tr4  )r  rp   r  )ru   r   resultsmut_resultsr   r_  outplace_ops         rf   r   z$register_foreach_inplace.<locals>.fn  sV    t.v.tAw0 	JKCyf4HI	J rh   )r[   ro  r   r  )aten_opoutplace_aten_opr  r   s     ` rf   register_foreach_inplacer    s.    07,-G$ w+rh   c                2    t        | d       fd       }|S )Nr  c                 l     | i |}t        || d   j                               }t        | d   |      S r  )r   r   r  )ru   r   r_  r  s      rf   r   zregister_inplace.<locals>.fn  s<    d-f-&$q'"3"3"56a&))rh   )r$  )r  r  r   s    ` rf   register_inplacer    s#    wD9* :*
 Irh   c                     y r   r   r  s      rf   sym_constrain_ranger        rh   c                    t         j                  j                  j                  d   }t	        |t
        j                        r|j                  j                  S t        |      S r	  
rT   r^   r_   rc  rl   r|   r  rd   r  r   r   r   r  s      rf   rE  rE    A    
''


#
#E
*C#u||$xx}}3xrh   c                    t         j                  j                  j                  d   }t	        |t
        j                        r|j                  j                  S t        |      S r	  r  r  s      rf   
sym_strider    r  rh   c                "    | j                         S r   )r  r   s    rf   	sym_numelr    s    ;;=rh   c                &    t        j                  |  S r   )r   Addr  s    rf   sym_sumr    s    99drh   c                    t        d      )NzHelpful for debuggingr   )r  ru   r   s      rf   foobarr    s    
5
66rh   c                8    | j                          t        |       S r   )rs  r  r   s    rf   _realizer    s    IIK8Orh   c                R    | j                          t        j                  | |       | S r   )rs  r2   ResizeStorageBytes)variabler  s     rf   resize_storage_bytes_r    s#    (H-Orh   c                    | j                          |j                          t        j                  t        j                  | |            S r   )rs  rD   r   r2   SetSourceTensorKernel)r  source_tensors     rf   set__source_tensorr    s5    LLNB44T=IJJrh   r	  c                    | |u r| S t        || j                               }t        || j                               }t	        || j                               }t        | |      S r   r>  )r  r  s     rf   
fsdp_copy_r    sR    #:JS^^-.sCMMO,S#,,.)c""rh   c                 	
 t        | t              sJ t        |t        t        f      sJ |t        j
                  }|t        j                  k(  rt        d|       |t        j                  k(  rt        |      dk(  sJ |t        j                  k(  rt        |      dk(  sJ | j                         
| j                         }| j                         }t        | j                  t        j                         r| j                  j#                         | _        t	        j$                         rit        j&                  j(                  j*                  rEt-        |      rt/        d      n0t1        |      r t	        j2                  |      j4                  ndndt6        j8                  j:                  j=                  
d      rt?        |||      S tA        | 
gd	g      }|jC                         	t        jD                  jG                  ||      }t        jH                  ||||      jK                         	
fd
}tM        jN                  |||t        |            }|S )Nzunsupported memory format: r   r   nanTrf  r   r  r/   c                    |       t        j                  t        j                        }t        j                  t        j                        }t        j                  ||      }t        j
                  |fd      S )Nc                       g      S r   r   )
flat_indexflat_loaders   rf   r#  z*resize.<locals>.inner_fn.<locals>.<lambda>b  s    ZL(A rh   )rS   rg  r|   rh  rk  rm  )	r	  flat_index_exprlimitrs  r  r  	old_numelout_indexeruninitialized_vals	       @rf   r\  zresize.<locals>.inner_fn]  sZ     %
..U[[Ay%++6vvou-zz$ ACTUUrh   r]  )(rl   rD   rj   r   r|   contiguous_formatpreserve_formatrq  channels_lastr   channels_last_3dr  r   r@  rn   r2   r6   rR  r  r  deterministicfill_uninitialized_memoryr   r5  r    r  r  rT   r^   r   r  r  r>  r_  r   stride_ordered_for_memory_formatrU  r  r@   r   )r   r   r'  r   rw   x_flat
out_strider\  rr   r  r  r  r  s            @@@@rf   resizer  )  s   a###dT5M***//---8HII+++4yA~~...4yA~~IKKME""$F!&&"++&##% 	224KK%%??%  %ee$ %E 2 6 6 $  ww//	1=D+5HH		
 	
F $$&K""CCD-XJ..jANNPKV 

UXd4jC Jrh   )auto_functionalizedc                    ddl m} |j                  |      }t        j                  | ||i ||       |j                         D ci c]  \  }}t        |t              s|| c}}S c c}}w )Nr   )kernel_side_table)
kernel_idxgridtma_descriptor_metadatakernel_args)*torch._higher_order_ops.triton_kernel_wrapr  get_constant_argsr2   UserDefinedTritonKernelr   rl   rD   )	r   constant_args_idxr  r  r   r  constant_argsr  r  s	            rf   triton_kernel_wrap_r	  p  sj     M%778IJM 7/v//	 &,\\^Rcz#y7QCHRRRs   A*!A*c                ^   t        d | g|D              rTd}t        j                  j                  j                  j                  dd       x}r| d| }|t        j                  _        t        j                  j                  | |||      }t        t        t        j                  |            S )Nc              3  V   K   | ]!  }t        |t              xr t        |       # y wr   rJ  r  s     rf   r   zcond.<locals>.<genexpr>  s#     
Ma:a 1Yq\1
MrK  z"control flow operator: torch.cond.r  r  )r  rT   r^   r_   rc  rd  r  r2   Conditionalr   rj   maprD   )predtrue_fnfalse_fnoperandsr   r  r_  s          rf   r   r     s     
MD;L8;L
MM2''..3377tLL;LE*;-8C,/)^^""4(HEFI$$f-..rh   c                   t         j                  sit        d ||z   D              rTd}t        j                  j
                  j                  j                  dd       x}r| d| }|t        j                  _        t        j                  j                  | ||||      }t        |t              sJ t        t        t        j                  j                   |            S )Nc              3  V   K   | ]!  }t        |t              xr t        |       # y wr   rJ  r  s     rf   r   zwhile_loop.<locals>.<genexpr>  s+      * 	1f.)A,.*rK  z(control flow operator: torch.while_loop.r  r  )r0   r  r  rT   r^   r_   rc  rd  r  r2   	WhileLoopr   rl   r   rj   r  _maybe_wrap_as_tensor_box)cond_fnbody_fncarried_inputsadditional_inputsstack_outputr   r  r_  s           rf   
while_loopr    s     !!c *"33* ' 9''..3377tLL;LE*;-8C,/)\\  .*;\F fh'''BLL::FCDDrh   )r  c                    t        j                  j                  | g| }t        t	        t
        j                  |            S r   )r2   InvokeSubgraphr   rj   r  rD   )subgraph_fn
identifierr  r_  s       rf   invoke_subgraphr     s5    %%k=H=FI$$f-..rh   c                   d}t        | j                  j                        D ]  \  }}|j                  dk(  r?|t        j                  j
                  vsJ ||   t        j                  j
                  |<   U|j                  dk(  r]t        j                  j                  |      \  }}t        j                  j                  j                  t        j                  |||      }|t        j                  j
                  vsJ t        j                  j                  |      t        j                  j
                  |<    |t        d      |S )a  Process nodes from a FX graph by executing them through V.graph.

    This is a common pattern for executing a subgraph's nodes:
    - Placeholder nodes are mapped to the provided args
    - Output nodes return their result
    - Other nodes are executed via V.graph.run_node

    NrP  r0  zNo output node found in graph)rk   r^   nodesra   rT   envfetch_args_kwargs_from_envr|   rR  Interpreterr0  run_noderq  )r  ru   r0  rt   rd   output_argsr   s          rf   process_subgraph_nodesr(    s     F\//556 
7477m#qww{{*** $QAGGKKWW "#''"D"DT"JKXX))00$VTFqww{{*** ! 0 0 6AGGKK
7 ~:;;Mrh   )control_depsc                   g }| D ]B  }t        |t              s|j                          |j                  |j	                                D t
        j                  j                  j                  }d}t        |      |z   t        |      k(  sJ t        t
        j                  j                        }t        |j                  j                  j                  d            t        |      k(  sJ t        |j                  t        |            }|| sJ t
        j                  j                  |d D ]E  }	|D ]>  }
|	j                  }|J t
        j                  j                   |   j#                  |
       @ G |S )aS  
    Lower control_deps_op by ensuring dependencies are realized and tracking them.

    The control_deps_op HOP makes dependencies explicit in the graph. During lowering:
    1. Realize all additional dependencies to ensure they're computed
    2. Execute the target operation normally
    3. Track the dependencies for the scheduler
    r-   rP  r  N)rl   r:   rs  rp   r  rT   r^   r_   ru   r   
operationsr  
find_nodesr(  rj   operation_nameadditional_buffer_depsr   )additional_depsr  ru   	dep_namesdeporiginal_args
arg_offsetoperation_lenr0  ra   dep_nameop_names               rf   control_deps_op_loweringr7    sV    I )#v&() GG((--MJt9z!S%7777**+M{''--88M8JKsSWyXXX $K$<$<d4jIF/11 gg  0 B! 	BH''G&&&GG**7377A	BB Mrh   )schemec                  d }t         j                  j                  j                  j	                  dd       }|J t        | j                  j                  j                        D ]  \  }}|j                  dk(  r!||   t         j                  j                  |<   7|j                  dk(  rt         j                  j                  |      \  }}t        j                  ||j                               D ]  }	|	j                          |j                  r7t         j                  j                   j#                  |	j%                                t         j                  j&                  j#                  |	j%                                 t(        j*                  j,                  j/                  t         j                  |||      }Xt         j                  j1                  |      t         j                  j                  |<    |S )Nquant_optionsrP  r0  )rT   r^   r_   rc  rd  rk   r  r"  ra   r#  r$  r   r   r   rs  codegen_low_precisionlow_precision_codegen_opsr   rt  invoke_quant_opsr|   rR  r%  r0  r&  )
r  r8  r  r0  r:  rt   rd   ru   r   r  s
             rf   invoke_quant_tracerr>    sl   FGG((--11/4HM$$$[55;;AAB 7477m# (AGGKKWW 77==dCLD&__T6==?; E		 66GG5599!:N:N:PQ((,,Q-A-A-CDE XX))00$fMF ! 0 0 6AGGKK%7( Mrh   c                  
 ddl m}m} t        |      dkD  rt	        d      t        j                  ||      D cg c](  } ||j                         |j                               * }} || |      

fd}t        |d   dd       }t        d |D              |d	<   t        d
 |D              |d<   t        j                  j                  d|dd|}	|	d   t	        d      |	S c c}w )Nr/   )InputDescriptorlower_pointwise_subgraphr   zSUnable to generate code for associative_scan op, because there are lifted argumentsr  c                d     g t        j                  |       t        j                  |       S r   )r4  rT  )lhsrhslowered_combine_fns     rf   wrapped_combine_fnz,associative_scan.<locals>.wrapped_combine_fn2  s6    ! 
$
$
 	
rh   rq  c              3  <   K   | ]  }|j                           y wr   )r   r  s     rf   r   z#associative_scan.<locals>.<genexpr>9  s     7qQ[[]7   r  c              3  <   K   | ]  }|j                           y wr   r  r  s     rf   r   z#associative_scan.<locals>.<genexpr>:  s     <A<rH  r  F)ro  can_fallback_to_atenz/Unable to generate code for associative_scan opr   )r  r@  rA  r   rq  r   r   r   ro   r  r   r2   rs  r   )ro  xsr  r@  rA  r   subgraph_inputsrF  r   r_  rE  s             @rf   associative_scanrM  !  s     M
!a
 	
 R( 	akkmALLNCO  2*oN
 be!48F7B77F8<<<F;WW^^ %" F
 ayLMMM-s   -C'c                     y r   r   )tokenss    rf   _sink_tokensrP  E  r  rh   c                      y r   r   r   rh   rf   _make_tokenrR  J  r  rh   c                (   ddl m}m}  ||      }||t        j                  j
                  j                  u rddlm}m	} |j                         }	|	r|	j                  j                  t        j                  j
                  j                        }
|
rMt        |
|      sJ |
j                  |d         }|r)t        |      dk(  sJ d       t!        t#        |            }t        t$        j&                  j(                        }|t*        v r+t+        |   |i |}t-        j.                  t0        d |       n:d }t-        j2                  |t5        j6                  j8                  |g|i |      }t        t$        j&                  j(                  |d       dkD  sJ d	| d
       |rt$        j&                  j:                  j=                  |      }t$        j&                  j(                  |d D ]W  }d |_        |s|jA                         }t$        j&                  jB                  |   jE                  |jA                                Y t$        j&                  j:                  |<   	 t-        j.                  t4        jF                  d ||f      \  }} ||||      }t        |jP                        dk(  r| |fS t        |jP                        dk(  r| |fS | g|S # tH        $ r2}tK        |      }tL        jO                  d||       | g|cY d}~S d}~ww xY w)z
    We lower the operator directly, and then we add StarDep dependencies to all
    the newly created nodes in the graph.
    r   )_get_effect_get_schemaN)InvokeSubgraphCacheTracingContextr/   zMultiple effects NYIc                "    | j                         S r   )rs  r  s    rf   r#  zwith_effects.<locals>.<lambda>p  s    !))+ rh   c                d    t        | t        j                        rt        j                  |       S | S r   r8  r   s    rf   r9  z"with_effects.<locals>.wrap_tensorss  r:  rh   zCNo operation nodes were generated when lowering effectful operator .c                      yr   r   r   rh   rf   r#  zwith_effects.<locals>.<lambda>  s    rh   c                "    | j                         S r   )	get_valuer  s    rf   r#  zwith_effects.<locals>.<lambda>  s    !++- rh   z5Failed to get schema for %s: %s. Assuming list output))torch._higher_order_ops.effectsrT  rU  r|   rS   higher_orderr   torch._guardsrV  rW  try_gethop_dispatch_set_cache	get_cacherl   get_effectsr   r?  r  rT   r^   r+  rY   r4  r  rD   r;  r2   r<  r   effectful_opsrd  has_side_effectsr  additional_star_depsr   TorchBindObjectrq  r  rn  ro  returns)tokenra   ru   r   rT  rU  effect_typerV  rW  tracing_ctxinvoke_subgraph_cacheeffectsr4  r_  r9  prev_effect_buffernew_opr6  schemae	error_msgs                        rf   with_effectsrt  O  s    I b/KrUYY%;%;%K%KKE$,,.$/$F$F$P$P		&&66%! %!"79LMMM/;;DGDw<1,D.DD,"&tG}"5K **+M 
Y2//Y(=vF	J "++222GGG
 qww!!-.12Q6 
MbTQRS6 WW2266{Cgg((8 	YF&2F#! //+,,W599:L:U:U:WX	Y  	
k*
 ++ 7$
f Rv. 6>>av	V^^		!v   F	CR	
  s   #5K 	L'LLL)register_comm_loweringsc                t   t        | |ddd      }|d   }t        j                  j                  j	                  t        |            }t        j                  j                  di |d|d\  }}|dk(  rXt        j                  j                  j                  |t        j                        r t        j                  d| d|d	|\  }}||fS t        j                  t!        j"                  d
             t%        | |d      }	t'        t(        j*                     t-        | |	            }
t/        |
|d      }|	|fS )zn
    Lowering inductor_prims.prepare_softmax_online to compute max/sum in one pass if no split is needed.
    TNr  r  online_softmax_reduce)r   r  r/   r-   )r  
num_outputreduction_hintz
            Online softmax is disabled on the fly since Inductor decides to
            split the reduction. Cut an issue to PyTorch if this is an
            important use case and you want to speed it up with online
            softmax.
            )r  r   )r  rT   r^   r   simplifyrQ   r2   rA   
num_splitsstatically_known_geqr0   r  r=   r   rA  rB  textwrapdedentr  rY   r  r~  r  r  )r   r   r   r  r  hint	num_split
max_tensor
sum_tensorr$  r~  xsums               rf   prepare_softmax_onliner    s5   
 #	d$dF 01WW&&}5E'FGFll-- 
.OD) A~!''**??22 "8!>!> "
Qt"
?E"

J :%% 	OO		
 1cD1!#a,/Ct,Tzrh   )r=  )quantized_lowerings)mkldnn_lowerings)jagged_loweringsc              #  X  K   t        | t        j                  j                        sJ d       t        j                  |       }	  t        |       t        |              d |r
|t        | <   yt        j                  |        y# |r
|t        | <   w t        j                  |        w xY ww)z^
    A context manager to force fallback an op. Used in unit test
    for FallbackKernel.
    z+Only OpOverload to make the clean up easierN)	rl   r|   r}   r~   rY   rd  r$  r  r  )ra   old_handlers     rf   force_fallbackr    s      b%**//0 50 --#K".r23'IbMMM" 'IbMMM"s   AB*B ""B*#B''B*r  )rq   zIterable[Any]r   z4defaultdict[tuple[Any, bool], list[tuple[int, Any]]])r   Callable[..., Any]r   zOptional[Callable[..., Any]])r   ztorch._C.Tagr   z(Optional[Callable[..., tuple[Any, Any]]])r   r   r   r  r   None)r   zUnion[Collection[Union[torch._ops.OpOverload, torch._ops.OpOverloadPacket]], torch._ops.OpOverload, torch._ops.OpOverloadPacket]r   zOptional[list[Any]])r   z9Union[torch._ops.OpOverloadPacket, torch._ops.OpOverload]r   zCallable[..., tuple[Any, Any]]r   r  )r   zUnion[int, torch.dtype]r   torch.dtype)r   r	   r   z,TypeGuard[Union[TensorBox, sympy.Expr, int]])r   r	   r   z!TypeGuard[Union[TensorBox, bool]])ru   r	   r   r   r   r  )ra   z>Union[Any, torch._ops.OpOverloadPacket, torch._ops.OpOverload]r   r  r   r   )r   rD   rw   torch.devicer   rD   )ru   	list[Any]r   zdict[str, Any]r  r   r   )Optional[ELEMENTWISE_TYPE_PROMOTION_KIND]r  r   r   z tuple[list[Any], dict[str, Any]])r   r  r  r  r   r  )
r  r  r  r   r   r  r  r   r   rX   )r   r  r   z.Callable[[Callable[_P, _T]], Callable[_P, _T]]r)  )NNNFN)F)r   'Union[TensorBox, ShapeAsConstantBuffer]r   r  r  r   )r   rD   r   r  )r   rD   rw   r  r   )r   rD   r  zSequence[sympy.Expr]r   rD   )r   r   l            r/   Tr  )rx  rD   r  rD   r  rD   r  r   r  r   r  r   r   r  r   r  )rx  rD   r  rD   r  rD   r  r   r  r   r  r   r   r  r  Optional[torch.dtype]r   r  )rx  rD   r  r5  r  r   r  r   r  r   r   r  r   r  )rx  rD   r  r5  r  r   r  r   r  r   r   r  r  r  r   r  )rx  rD   r  rD   r  rD   r  r   r  r   r   r  r   r  )rx  rD   r  rD   r  rD   r  r   r  r   r   r  r  r  r   r  )r   r   r/   )r;  r   r  r   r  r   r  )T)rx   ztorch.Tensor)rd   ztorch.fx.Node)NTF)rw   r  )r   	list[int]r  rD   r  r  r;  r   )
r  r   r  r   r   r  r  rD   r;  r   )r  rD   r   z.tuple[str, sympy.Expr, sympy.Expr, sympy.Expr])r  rD   r   ztuple[str, sympy.Expr])r  rD   r  rD   r  r   r  r   r  Optional[str]r  Optional[TensorBox]r   r  )rx  rD   r  rD   r  r   r  r   )r   r   r:  r   )r   NNr/   )NNN)rL  FF)r   r   )rc  r  r   r   r  r  r  r   )r   r   r  r  )r   r   r  r   )r-   F)r9  ztuple[Optional[float], ...]r7  r   r2  r   )r  Optional[float])rH  r  rI  r  )rO  r  rH  r  rI  r  )r   rD   rd  zSequence[int]rf  r5  r   r  )rt   
sympy.Exprr  zUnion[sympy.Expr, int])rt   r  r  r  )rt   r  r  r  r  r  )Nr  N)
r  rD   r  "Sequence[Union[int, torch.SymInt]]r  r  r  zxCallable[[Sequence[Union[int, torch.SymInt]], Sequence[Union[int, torch.SymInt]]], torch._inductor.virtualized.OpsValue]r   r  )Nr   r/   F)rf  )NNNN)r   r   FTN)r   rU   r   )r   r  r   zOptional[ir.Constant])rL  F)ra   ztorch._ops.OpOverloadPacket)r   z3list[Union[ir.TensorBox, ir.ShapeAsConstantBuffer]])r  ir.Subgraphr  r  )r  ztorch.fx.GraphModuleru   r  )r  r  )ro  r  r  ztuple[torch.Tensor]r  (  
__future__r   
contextlibr  r  r   loggingr[  r  rf  r}  rA  collectionsr   collections.abcr   r   r   r   typingr	   r
   r   r   r   r   r   typing_extensionsr   unittest.mockr   r   r|   $torch.ao.quantization.fx._decomposedtorch.fxtorch.utils._pytreer  _pytreer4  torch._dynamo.utilsr   (torch._higher_order_ops.associative_scanr   r  r   "torch._library.fake_class_registryr   torch._library.utilsr   torch._prims_commonr   r   r   r   r   r   r   r   r   r    r!   torch.fx.experimental.sym_noder"   r#   r.  r$   r%   r&   torch.utils._ordered_setr'   torch.utils._sympy.functionsr(   r)   r*   r+   r,   _dynamo.utilsr.    r0   r1   r2   r3   decompositionr4   r5   r6   r7   r8   r9   r:   r;   r<   r=   r>   r?   r@   rA   rB   rC   rD   rE   rF   rG   rH   rI   rJ   rK   rL   rM   rN   rO   rP   rQ   rR   rU  rS   rT   rP  rU   rV   rW   FALLBACK_ALLOW_LIST	getLoggerr  rn  rY   __annotations__rZ   r}   r~   r  r  tr_c10dr  r   _higher_order_opsr  rc   ro  r[   quantized_decomposedrg   ry   r   r   r   r   r   r>  r`  r1  r  bmmconvolutionconvolution_backwardr  r  r  rZ  rJ  rL  _int_mmr  r  r  r  rh  rb  r  rK  	complex32	complex64r   ra  r   r   r   r   r   r   r   r   r
  r  r!  r>  r$  r2  rD  rj  r  r   r  r  r  r  r   r  r  
device_putr  r  r  r  r  r   aliasdetachdetach_liftview_ofr  r   r  r  r  r  r  r  r  r  r  r  r  r-  r
  r  r  _unsafe_viewreshaper  slicerN  r]  rw  quantize_per_channelr  r  r   r  _functional_assert_asyncr  dequantize_per_channelr  quantize_per_tensorr  dequantize_per_tensorr  rw  r  r  r  r  r
  r  r  r  r  r  r*  r  r.  r/  r4  r  cacherC  rI  rN  r^  rt  ry  rngprimsr  r  r  	bernoullir  r  r  	lru_cacher  r  r|  r  r  r  r  r  r  randintstreamsrecord_event
wait_eventforce_stride_orderr  r  r  r  r  lookup_seedr  randomr  r  r  r  r  r  r  	NO_OPMATHr  r  r   r  r  r   r   r$  _adaptive_avg_pool3dadaptive_max_pool3d*_scaled_dot_product_attention_math_for_mpsuniformexponential_pdist_forwardsoft_margin_loss_backward_fused_rms_normxpuis_availableembedding_dense_backwardmtia_is_compilednative_layer_norm_cdist_forward_cdist_backward
_trilinearsegment_reduce_segment_reduce_backwardhistc	histogrambin_ct_histogramdd_bin_edges_histogramdd_from_bin_ctsaddbmm_addmm_activation_grouped_mm
_cudnn_rnn_cudnn_rnn_backward_embedding_bag_embedding_bag_forward_only_embedding_bag_backward*_embedding_bag_per_sample_weights_backward_fused_moving_avg_obs_fq_helper*_fused_moving_avg_obs_fq_helper_functional max_pool3d_with_indices_backward_adaptive_avg_pool2d_backward_adaptive_avg_pool3d_backwardadaptive_max_pool2d_backwardadaptive_max_pool3d_backwardfractional_max_pool2d_backwardfractional_max_pool3d_backwardreplication_pad1d_backwardreplication_pad2d_backwardupsample_linear1d_backwardupsample_bicubic2d_backwardupsample_trilinear3d_backwardgrid_sampler_2d_backward_pdist_backwardr  r  kthvaluetopkr  median	nanmedianr  resize_
resize_as__linalg_detlinalg_householder_productlinalg_inv_exlinalg_ldl_factor_exlinalg_ldl_solve	linalg_lulinalg_lu_factor_exlinalg_lu_solvelinalg_matrix_exp	linalg_qr_linalg_slogdet_linalg_solve_exlinalg_solve_triangular_linalg_svd	lu_unpackormqr_linalg_check_errorslinalg_pinvatol_rtol_tensor_linalg_eightriangular_solvelinalg_cholesky_excholesky_inversecholesky_solvegeqrf_fft_r2cnonzerogcd_thnn_fused_lstm_cell_prims	rng_primsrun_and_save_rng_staterun_with_rng_stategraphsafe_run_with_rng_statemasked_scattermasked_scatter_backwardrZ  angle_efficientzerotensor(_sparse_coo_tensor_with_dims_and_tensors	to_sparse
_to_sparser   rm  r  r  #_scaled_dot_product_flash_attention,_scaled_dot_product_flash_attention_backward#_scaled_dot_product_cudnn_attention,_scaled_dot_product_cudnn_attention_backward+_scaled_dot_product_flash_attention_for_cpu4_scaled_dot_product_flash_attention_for_cpu_backward0_scaled_dot_product_fused_attention_overrideable9_scaled_dot_product_fused_attention_overrideable_backward_flash_attention_forward_flash_attention_backward_efficient_attention_forward_efficient_attention_backwardindex_reducerepeat_interleave_weight_norm_interface_backwardr  r  r,  r[  r2  r7  r?  rA  scalar_tensorrS  
LongTensorrU  r[  r]  r_  rg  r  rk  r  rj  r  r  r3  
zeros_liker|  r  rs  r  r  r  r  r  r  r  r  r  r:  r  r  r  r  r  r  r  r  r  fallback__unsafe_masked_indexr  ,fallback__unsafe_masked_index_put_accumulater+  r  r  r  r  r  r  r!  r  r=  rB  rE  rP  rR  r  rW  rl  r  rs  rt  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r$  r8  r2  rH  rF  rX  r^  ra  r[  rq  r  ry  r  rt  r  r  r  r  r  r  r  r#  r  r  r  r  r  r  r!  r#  r'  r%  Tensor_Tensorr0  Scalarr.  Tensor_Scalarr/  r  rc  r	  r@  r  r  rG  r^  rM  r^  true_divider`  re  r  r  rv  rt  rz  ry  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r$  r  r%  r  r  r  r  r  r   r  r  r  r  rsqrtr~  r-  expm1relur2  r,  r  r  cossinabsbitwise_andbitwise_left_shiftbitwise_not
bitwise_orbitwise_right_shiftbitwise_xorlgammaerfspecial_erfr}  tantanhrI  r  r  logical_xorr  r  	clamp_min	clamp_maxnegr  	remaindersignsignbit	_neg_viewlerk  rj  r!  r5  necoshsinhacosacoshasinasinhatan2atanatanhr\  erfcerfinvhypotlog10log2	nextaftercodegen.commonrq  r  r  r   ra   r   rh  _foreach_addListforeach_add_listforeach_add_scalar_foreach_mulforeach_mul_listforeach_mul_scalar_foreach_sub_foreach_neg_foreach_abs_foreach_powScalarAndTensor_foreach_divforeach_div_listforeach_div_scalar_foreach_sqrt_foreach_rsqrt_foreach_maximum_foreach_minimum_foreach_clamp_min_foreach_clamp_max_foreach_reciprocal_foreach_sign_foreach_copyforeach_copyr  _foreach_add__foreach_mul__foreach_div__foreach_copy_r  add_bitwise_and_bitwise_left_shift_bitwise_not_bitwise_or_bitwise_right_shift_bitwise_xor_mul_div_Tensor_modelogical_and_logical_not_logical_or_logical_xor_sub_relu_sigmoid___and__
__lshift____or__
__rshift____xor____iand____ilshift____ior____irshift____ixor__r  rE  r   r  r  r   methodfuncr  r  _inductor_testrs  r  r]  r  set_source_Tensorr  fsdpr  r  *torch._higher_order_ops.auto_functionalizer  r	  r_  r   r  while_loop_stack_outputr#  r   r(  .torch._inductor.fx_passes.control_dependenciesr)  r7  invoke_quantr>  rM  rP  rR  rt  comm_loweringru  r  r=  r  register_quantized_opsregister_woq_mm_opsr  register_onednn_fusion_opsr  register_jagged_opscontextmanagerr  r   rh   rf   <module>r	     s8   "        	   # D D P P P '    +  $ $ ( H U ? :    M 
 0  - 8 8 =    &      * T]t_ !   g!FH	C H     .Juzz,,-/	yy~~
))

		9
5::#8#89; /j../
))*
 8j!6!679 NP K Pyy55 92B	B-BF
	 $3A3.3 
3   !!$$$$--&&, {{zz{{{{{{}}}}}}

."#%D  FSV	(O
OO O C	O
  O &Od"/A62!2 2 C	2
  2 L2n  	(// 4,#,2n  $Wt9z RW
E.
E7B
EJN
E 5**77TR) S)X 5--4H) I) @E 5( 499??=& >& ;@e K 5##>F ?F 
7?? $*Z>    4::DI J0 4))UPTU V( DJJT\\499emmTU V 4&doo&s+ 4<<T:B ;B, 4$$$?" @" DMM?# $ 4::C C 4::C C 499! ! 4::! ! 4::%%&% '% 4::! ! 4;;D9> :>4 5))tD
 E
 4>>t<# =# 4;;1  1h 4$$$?499$74<<T:1 ; 8 @1 4<<T:> ;> 4::48uE 9uEp 4??=J >J8 4##> ? 4''TB C
BJ '<<RVW/// / 	/
 / / / -/ X/d$ 4%%))*$ +$ 400445$ 6$ //T (,/// / 	/
 / / / %/ -//d ,,44$!!! ! 	!
 ! ! -!!H ..66D (,!!! ! 	!
 ! ! %! -!!H ,,33*** * 	*
 * * -**Z ..554 (,,,, , 	,
 , , %, -,,^ 488y: y:x 4==d;3E <3El 4%%4@6 A6 4((dC D 4;;D9=C :=C@ 4::48 94 4((dC  D  4;;D9 : 4;;D9D :D8 4>>t< = 4??= > 488 $  D
3%6P/Nd	, 599%%11tL#+ M#+L 4&&DAG BG 4??= > 4>>##>' ?' 4<<  ! QB B/
 )):):; *499+>+>? )$***<*<= +DJJ,@,@A  dll  eii,,44 5 eii**22 3 499J J 4::J J >44$OL PL
 >&&DAK BK >''TBJ CJ
 >--4H	 I	 >((dCRS  D6 >))tDLM	(09FI E8 . 4$$++F
 "&^^
^ 	^
 ^ ^  ^ -^ G^B NN(G(Q(Q 777 	7
 77t&$FT d'' ( d&& ' d== > dll ' d&&U 3 d!! " d,,5 9 d"" /99%%E 	::U
 d!! " d"" # doo  d!!)) * d++33 4 djj  dnn## $ d))11 2 d,,44 5 dkk  d$$5 1 d . d'')@ A doo} - d&&(: ; d!!#5 6 d..0B C d** + d== > d== > d22 3 d== > d33 4 d00- @ d00 1 d// 0 d// 0 d11 2 d11 2 d-- . d-- . d-- . d..0B C d00 1 d++ , d"" # dii  dii  dmm  dii  dii  dkk  dnn  dmm  dll  doo  d  d-- . d   ! d'' ( d## $ dnn  d&& ' d"" # d$$ % dnn  d"" # d## $ d** + d  dnn  djj  d'' ( d// 0 d   d## $ d%% & d## $ d!! " djj  dmm  dll"" # dhhU + d((- 8 ell$$;; < ell$$77 8 ell$$AA B
 d!! " d** + d""$6 7 djj  d'' ( d;; < dnn  doo  djj  0088	
 99AA	
 ,,44	
 55==	
 ,,44	
 55==	
 44<<	
 ==EE	
 99AA	
 BBJJ	
 d++33_ E d,,44o F d//77 I d0088/ J d   d$$++T B d22::<N O
 499$7 8" 4::"  * 4"#+d**+E2 5:: ( 4&&DA  B F 4%%4@A AAH ELL$"4"456tDU 5 75p 5??#4 $4 5##$+ %+ 4++,)" -)"X 4&&' ( 4//0%)59$ 1<K8 EKK,- 
 .$.> 0t/0B50IJ
1!45	 21 56
  4>>" $T$4  # 4%%&Td '@ 4))*"4 +" 5%%--.A /A EJJ		*+: ,:
 4;;D9! :!H 4>>t< =B#F FR"1J 4::48
 9
 4%%4@/ A/ 4>>t< = 4))* +;. 4??= > >44$O Ptn !1%%5!  0@,,44%0 ,
 4,,$G H, 4;;QUV	P W	P 1 1 4**E F 4<<T:9 ;9 !& 
  B 4==d;DH : <:" 4##>3 ?3 4$$$?6 @6 4&&DAP BP 4''TBPT q Cqp , *, 	,
 ,^ 4**223> 4> 41199:J ;J 4**223RVI-I@OI 4I 41199:RVU-U@OU ;U 4**223 !% $ $S S 	S
 S 4S 41199: !% $ $	 	 		
 	 ;	7 599$$% &*\\(\6;\\~ 4''TB4 C4n :>@ MQ < JN@:HV 5::PTU 5 V5D3 3	 -> 	11t$< 4//TJ 
 K
 4//TJ 
 K
 -=))11- ) 488dSN TNb8B&R2j  0%%5  
 4,,-B .BJ  0$$%  
 4++,N -Nb)X 4--.V /V 4--.V /VAH 433;;<BF. =.b 'OO  'OO 
 4??=  >* 4??=  >*wt  0$$%  
 4++F b GbJ  0$$%  
 4++F D GDN& HLOd * 499:t : : 4 (V4( DHHeii()T5  * 4==!u  "   .HH  'txxER -HH 
 488t,- --`"J 4::2 2 4::48 9     488t, - DHH:.( /(@ EII;$/$ 0$. 	txx'7DD

 DIIuzz*d;$ <$  DHHeii(). . *. #4;;#6#67#DLL$8$89 ():):)B)BC "4;;#6#67"4;;#6#67 4;;  . 4<<  !. 4$$% &* 4;;D9 :, 4;;D9 :, 499. . . 488@ @
 48867 77 48867 77 ! %--  	!: ;*		*>%+@A*		*>%+@A.!$++.85;;? /!$++.85;;? HH$L !!1!1uM 499##>!re $2 ?$2N 499$$$?H @H 	#4::.&txx0!$)),"4::.$))$
*4<<
8'		2	DKK	(t4   *   *" !1!12'(?(?@  -  0
()A)AB  !1!12 4;; ' * *I*V*V 4:: & 488 $ 499 %   * **	 !**	  OO**	
 !**	 T\\
*
T\\
* ! $.. !' * ! $.. !' *""'8
 4>> "$))L 499  4<<uzz B ! $.. !# & 477%** = 477%** = 477%** =uzzB 477%** = 477%** = 499 % 499 % 499 % 4:: & 499 % 4:: & 4:: & 499 % 4:: & 4== ) 499 % 4;; ' 4:: & 4:: & 499 % 4>> * D DE$ % 
D4Ld5 
0 	 3+		

 5Mt5 
0 	 3+		


, .CT  0ct  4,,33Sd K-d.?.?.D.DcJ  4,,33S 9/0A0A0H0H#N  4,,113 7 4,,33S 9 4,,44c : 4,,44c : 4,,33S 9 4,,113 7 4,,<<c B-d.?.?.D.DcJ  4,,33S 9/0A0A0H0H#N  4--t 4 4.. 6 40055w ? 40077 A 40055w ? 40077 A 42277 A 422997 C 42277 A 422997 C 433Z @ 4--t 4)$*<*<dC, T..335E t00779K T..335E t00779K T..335E t00779K !3!3!;!;\
 C   ""K 0 ))+= > ""K 0 !!: . **,? @ ""K 0 C   !!3 ' && 1 ""K 0 ""K 0 !!: . ""K 0 C   T "  (   $,,  , " $// "#5 6  $++ z * " $// "#6 7  $,,  ,  - !!4?? 3 t{{ + !!4?? 3  - 4++, - 4==$$% & 4??&&' ( 4>>" # (M'') 8LFD1(01$78 5==! " 4<< 7 !7 599++334 5
 599%%;;< = 599>>&&445K 6K 599>>7#uyy~~++334# 5# 599>>(()%) = *=@ K J ! " 12S 3S( 599))..DI/8/ J/ 599))44$OE PE& 	II22)JT24
 599))99tT/ U/
< H G <T:+ ;+\ 5**77TRDH  S8 &DA  4G  B F 599??//778 9 599??..667 8 599))66DQM  RM ` 3 2   >88dS1 T1j     ! ! +  * * , '  ' ' )   ,  + + -   %  $ $ &  rh   