o
    Ih                    @  sZ0  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 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"  m#Z$ d dl%m&Z& d d	l'm(Z( d d
l)m*Z* d dl+m,Z,m-Z-m.Z.m/Z/m0Z0m1Z1m2Z2m3Z3m4Z4m5Z5m6Z6 d dl7m8Z8m9Z9 d dl:m;Z; d dl<m=Z=m>Z>m?Z?m@Z@mAZA ddlBmCZC ddlDmEZEmFZFmGZGmHZH ddlImJZJmKZK ddlGmLZLmMZMmNZNmOZOmPZPmQZQmRZRmSZSmTZTmUZUmVZVmWZWmXZXmYZY ddl"mZZZm[Z[m\Z\m]Z]m^Z^m_Z_m`Z`maZambZbmcZcmdZd ddlemfZfmgZg er!ddlhmiZi edZjedZke;ddgZlemenZoi Zpdeqd< i Zrdeqd< e;ejsjt  ZuejfjvZvejfjwZwejfjxZxe;ejsjt  Zye;ejsjt ejzj{gZ|e;ejsjt  Z}i Z~d eqd!< ejfjZd"d# Zdd&d'Zdd,d-Zd.d/ Zd0d1 Zd2d3 Zd4d5 Zeevjevjevjevjevjevjevjevjevjevjevjevjevjg ejejejejejejejejejejejejejd6Zdd9d:Zd;d< Zd=d> ZddAdBZdCdD ZdEdF ZddPdQZdRdS ZddTdUZdVe1jdVfddXdYZdZd[ Zdd\d]Z					V	dd^d_Zdd`daZdddedfZeejzj{ddgdhdi Z{eexjddgddjdkZdVdlddmdnZeevjjddgddodpZdVdVdqddtduZeexjddgdddvdwZddxe1jdVdddVdVdf	dydzZd{d| Ze  	Vdd}d~ZeevjdVdddd ZeevjdVdddd Zeevjevjevjevjexjgdd Zeevdreevjƃeă eevjddgdddZeevjddgdddZeevjgdddZeevjʃdd Zeevj˃dd Zeevj̃dd Zeevj̓dd Zeevjjσdd ZeevjЃdd Zeevjddgdd Zeexjddgdd Zeevjddgdd Zeevjԃdd Zeevjddgeevjddgeevjddgdd Zeevjddgdd ZeevjddgdddZeevjddgdddZeevjddgdddZeevjddgdddZdddZeejddgdddZeejddgdddddZeejjddgdddZeejjddgdddddÄZeejjddgdddńZeejjddgdddddǄZeevjdddɄZeevjddgdʐddd΄ZeevjddgdʐdddЄZeevjddgdʐddd҄ZeevjddgddԄ ZeevjddgdddքZeevjddgddd؄ZeevjddgdddڄZeevjddgdd܄ Zeevjddgddބ Zeevjddgdd ZdddZeevjdddZdddZeddd ZddddZddddZd͐dddZdddZdd Zeejfjjddgdd Zeevjddgdd Zeevjddgdd Zeevj jddgdd Zeevjd d Zeddd Zdd ZeevjjσZeevjjZ	eevj
jσZeevj
jZeevj eevjdd Zeevj
dd	 Z
eeFjddgd
d ZeeFjddgdѐddZeeFjddgdd ZeeFjddgdd ZeeFjddgd ddҐddZeeFjddgd ddӐddZdԐd d!ZdՐd#d$ZeevjjddgdVdVddd%d֐d.d/ZeevjddgdVdVd0dאd2d3Zd4d5 Zd6d7 Zd8d9 Z d:d; Z!d<d= Z"d>d? Z#eevj$ eevj% eevj& eevj' eevj(dVd@ eevj)jdVd@ eevj* eevj+dVd@ eevj, eevj- eevj. eevj/jσ eevj0jσ eevj1 eevj2j3 eevj4jσ eevj5jσ eevj6 eevj7dVd@ eevje" eevj8e eevj9e eevj:e eevj;e eevj< eevj= eevj= eevj> eevj? eevj@ eevjAe eevjB eevjC eevjD eevjE eevjF eevjG eevjH eevjI eevjJe eevjK eevjLe eevjM eevjN eevjNjO eevjP eevjQ eevjR eevjS eevjT eevjU eevjV eevjW eevjX eevjY eevjZ eevj[ eevj\ eevj] eevj^ eevj_ eevj` eevja eevjb eevjc eevjd eevje eevjf eevjg eevjh eevjijj eevjk eevjl eevjm eevjn eevjo eevjp eevjq eevjrjσ eevjsjdVd@ eevjte eejujvjw eejujvjx eejujvjy eevjz eevj{ eevj|e eevj} eevj~ eevj eevj eevj eevjj eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe#dVd@ eevjjϐe# eevjjϐe# eevjjϐe# eevjjϐe# eevj eevjddgddAdBZeevjddCdDdEZdFdG ZeevdH	reevje eexjdIdJ ZeevjddgdؐdMdNZeevjddgdِdOdPZdQdR ZeejevjgddddVdSdTdUZeejddVdWZeejdXdY ZeevjdZd[ Zeevjd\d] Zeevj	dddd^d_d`Zdadb Zeevjddgdcdd Zdedf ZeejevjgdddddddgdhdiZdjdk Zdldm ZeevjeeZeedZeed Zdndo ZeevjdddddpdqdrZeevjdddddpdsdtZeevjdddddpdudvZeexjjσdwdx Zeejevjgdydz Zeevjddgdd{d|Zeevjddgdېd}d~Zdd Z	xd͐ddZdd Zd͐ddZeevjddgddL Zeevjddgdd ZeevjdddZeevjdddZdd Zdd ZeevjddgdddZeeFjddgdddZdddZeevjjdVdZeevjjdVdZeevjddgdd Zeevjddgdd Zedd ZeevjddgdddZeevjddgdܐddZddxddݐddZeevjddgdddސddZeevjddgdܐddZeevjddgdܐddZeevjddgdܐddZeevjddgdxddߐddZ		VddddZeevjjσddddZeevjjσddddZeevjjσ	dddÐdĄZeevjjσ	dddŐdƄZeevjjσ			dڐddȐdɄZeevjjσ			dڐddʐd˄Zѐd̐d̈́ Zeexjjσdΐdτ ZӐddӐdԄZeevjddgdÐdՐdքZՐddڐdۄZ֐ddܐd݄Zאddސd߄Z	dddZِdd Zڐdd ZddddZ	VdddZeexjddg	VdddZeexjddgdd ZeevjjdVdZeevjddg		 		VdddZeevjjdVdZeevjddgdd ZdddZdd Zdd Zdd ZeevjjdVdZeevjdd ZeevjjdVdZeevjdd  ZeevjjdVdZdd Zeevjdd Zeevjjσ	dddZeevjjdVdZeevjjdVdZeevjddg		 	V	x	ddd	Zeevjddg		 	V	x	dd
dZdd ZeevjjdVdZeevjddg	dddZeevjjdVdZeevjddg	dddZdd Zdd ZddddZdd ZeevjdddddZdd  Zd!d" Zd#d$ Zd%d& ZeevjexjgdddVd'd(d)Zeevj dddVd'd*d+Z d,d- Zed.d/ ZeevjjdVdZeevjjdVdZeevjjdVdZ	eevjdxd0d1d2 Zdd3d4Z
eevjd5d6 Zeevjddgdd7d8Zed9d: Zed;d< Zeevjdxd0dd=d>Zeevjgdxd0d?d@ ZddCdDZeexjgdxd0dEdF Zeevjevjjgdxe1jddGdH Zeevjexjgdxd0dIdJ ZeevjexjgddddKdLZeevjjσZeevjjσZeevjjσZeevjjσZ eevj!jσZ"eevjddMdNZeevjddOdPZeevjdQdR ZeevjddgddSdTZeevj!ddgddUdVZ!eevj#ddddWdXZ#eevj$ddYdZZ%eevj&ddgdd[d\Z'eevj(ddgdd]d^Z)eexj*ed_ eevj+ed`Z,eevj-edaZ.eevj/edbejdcZ0eevj1eddejdcZ2eevj3dxdedfZ3eevjNjOdVdZ4eevjNjOddgdddVdgdhdiZ5eevjNjddgddjdkZNddldmZ6dndo Z7e6evj8Z8e7evj9Z9e6evj:Z:e6evj;Z;eevj<Z<e7evj=Z=e7evj>Z>eevj?Z?eevj@dxdpZ@e7evjA e7evjB eevjCZCeevjDZDeevjEZEeevjFdqdrZFeevjGZGeevjHZHeevjIZIe6evjJ e6evjKZKeevjLe1jdgeK e6evjM e6evjN e6evjO e7evjo eevjPddxejdsZPeevjQddxejdsZQeevjRddxejdsZReevjSddxejdsZSeevjTZTeevjUZUeevjVeT eevjWeU eevjXZXeevjCZCe6evjYZYeevjZ eevj[dtdrZ[eevj̃ eevj\ejdc eevj]eX eevj^ejdc eevj_ejdc eevj`ejdc eevjaejdcZaeevjbejdc eevjcejdc e6evjd e6evje e6evjf e6evjg e6evjh e6evji e6evjj e6evjk e6evjl e6evjm e6evjn e6evjo e6evjp e6evjq e6evjr e6evjs ddultmuZumvZv dvdw ZwevD ]@ZxewevexD ]\ZyZzZ{eeyexeze{dx qewexexD ]\ZyZzZ{eeyexeze{dx qqeevj|j}e3dxdpZ~eevj|je3dxdpZeevj|je3dxdp eevjj}eZeevjje eevjjeZeevjj}e@ eevjje@ eevjjϐeX eevjjϐeC eevjje eevjj}e eevjje eevjj}eZeevjje eevjjeZeevje> eevje8 eevjj}eT eevjjeT eevjj}eU eevjjeU eevjj}eT eevjjeT eevjj}eU eevjjeU eevjeY eevje[ eevje dydz Zeevjj}evj|j}e~ eevjjevj|je eevjj}evjj}e eevjjevjje eevjj}evjj}e eevjjevjje d{d| Zeevje3 eevjeD eevjeE eevjeF eevjeG eevjeH eevjeI eevje eevjje eevjje eevjeP eevjeQ eevjeR eevjeS eevje@ eevje< eevje= eevjeD eevjeE eevjeG eevjeH eevjeI eevjevj eevjevj eevjevj eevjevj eevjevj eevjdd}d~Zeevjjdd Zeevjjdd Zeevjdd Ze8 D ]\ZZee9ee qeejdd Zeevjdd Zeejfjjdd Zeejfjjdd ZeejfjvjÐjădd ZeejfjƐd8reejfjƐjjσdd ZeejfjvjȃddCddZd dlɐmʐZ eeʃ ee*dd Zeejfj̐jddgdd Zeejfj̐jddgdd Zeejfj̐jddgdddZeejzjddgdddddZee(ddgdddZeejfjxjjσdd Zeejfj̐jddgdd ZddlՐm֐Z eփ  eeFjddgdd ZddlDmؐZ eCe؃ ddlDmِZ eِڡ  eِۡ  ddlDmܐZ eܐݡ  ddlDmސZ eސߡ  ejdddZdS (      )annotationsN)defaultdict)IterableSequence)AnyCallablecastOptionalTYPE_CHECKINGTypeVarUnion)	ParamSpec)patch)counters)associative_scan_op)triton_kernel_wrapper_mutation)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)
OrderedSet)CeilDivFloorDivIdentity
IntTrueDivModularIndexing   )import_submodule   )configinductor_primsirtest_operators)decompositionsget_decompositions)	DtypeView
ExpandViewIndexingConstantIRNode	is_tritonOnlineSoftmaxReductionops_wrapperPermuteView	Pointwise	Reduction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sympy_productuse_scatter_fallback)opsV)ReductionType_T_Pztorchvision::roi_alignzaten::index_addz8dict[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                  C  s<   t jjjD ]} | jD ]}|jdkr|jtv s  dS q
qdS )Ncall_functionTF)rH   graphcurrent_nodeusersoptargetforeach_ops)nodeuser rX   L/var/www/vscode/kcb/lib/python3.10/site-packages/torch/_inductor/lowering.pycur_node_has_non_foreach_users|   s   
rZ   	arg_pairs%Iterable[Union[tuple[Any, Any], Any]]c                 C  s   t t}d}t| D ]A\}}t|tsd}|f}t|  ptj}d }|D ]}t|tr2|j	
 } nq$|d us;J d|r@|\}|||f ||f q
|S )NFTz.foreach op should have at least one tensor arg)r   list	enumerate
isinstancer   r>   r(   #combo_kernel_foreach_dynamic_shapesr9   data
get_deviceappend)r[   outunpack_argsiargsuse_foreachdevicetrX   rX   rY   group_foreach_args   s&   


rk   fnCallable[..., Any]returnOptional[Callable[..., Any]]c                   sR   t  tjjs	dS  tv rt  S  tv rdt < dS  fdd}t }||S )zHGet layout constraints. Returns None if there are no layout constraints.Nc                   sF   | t jjju rtt < t  S | t jjju rd t < d S td|  )NzUnknown layout constraint tag: )torch_CTagneeds_fixed_stride_orderconstrain_to_fx_stridesrM   flexible_layoutAssertionError)tagrl   rX   rY   handle_layout_constraint_tag   s   z>maybe_layout_constraints.<locals>.handle_layout_constraint_tag)r_   rp   _ops
OpOverloadrM   rL   get_layout_constraint_tag)rl   ry   rw   rX   rx   rY   maybe_layout_constraints   s   
r}   c                 C  sX   t jjjt jjjg}|D ]}|| jv r|  S qt jj| r$t jjjS t	t jjt
jS N)rp   rq   rr   rs   ru   tags_libraryutils
is_builtingetattrr(   #custom_op_default_layout_constraint)rl   tags_by_priorityrw   rX   rX   rY   r|      s   

r|   c                 C  s   | s	t d| d S )Nzinductor does not support NotImplementedError)condmsgrX   rX   rY   
assert_nyi   s   r   c                   s\   t  ttttfrdd  D S t  t  tjj	r,t
 fdd  D  d S d S )Nc                 S  s   g | ]}t |qS rX   )add_needs_realized_inputs.0xrX   rX   rY   
<listcomp>       z-add_needs_realized_inputs.<locals>.<listcomp>c                 3  s    | ]}t  |V  qd S r~   )r   )r   overloadrx   rX   rY   	<genexpr>   s    

z,add_needs_realized_inputs.<locals>.<genexpr>)r_   r]   settupler   needs_realized_inputsaddrp   rz   OpOverloadPacketupdate	overloadsrx   rX   rx   rY   r      s   
r   c                 C  s:   t | tjjr|  D ]	}|tt| |< qd S |t| < d S r~   )r_   rp   rz   r   r   rM   r   )rl   
constraintr   rX   rX   rY   add_layout_constraint   s
   r   )r   r'   r%                     	   
         dtypeintc                 C  s2   t | ts| S | tv sJ d|  dt|  } | S )Nzid z missing from DTYPE_ID_LOOKUP)r_   r   DTYPE_ID_LOOKUPr   rX   rX   rY   decode_dtype  s
   
r   c                 C  sB   t | trt|  pt|  S t | tjr| jdu S t | tS )NT)	r_   r9   r   	get_dtyper   sympyExpr
is_integerr   r   rX   rX   rY   is_integer_type  s
   


r   c                 C  s    t | trt|  S t | tS r~   )r_   r9   r   r   boolr   rX   rX   rY   is_boolean_type  s   

r   type_promotion_kindr   c                   s0   dd   fdd|D }t |d| i\}}|S )Nc                 S  s8   t | ttjfr
| S t|  }tjdg| |  dS )Nr'   r   )	r_   r   r   Basiclenget_sizerp   zerosr   )inpdimrX   rX   rY   construct_input!  s   z+get_promoted_dtype.<locals>.construct_inputc                      g | ]} |qS rX   rX   )r   argr   rX   rY   r   )  r   z&get_promoted_dtype.<locals>.<listcomp>r   )r   )r   rg   inps_r   rX   r   rY   get_promoted_dtype   s   r   c                 C  sh   t | ttfs| g} nt| } t| D ]}t |tjjr1| D ]}t||}|tvr0| 	| q q| S r~   )
r_   r]   r   rp   rz   r   r   r   rL   rc   )aten_fnrl   r   other_fnrX   rX   rY   get_overloads.  s   

r   c                 C  s6   t | tjjr|| jv S t | tjjr||  v S dS NF)r_   rp   rz   r   _qualified_op_namer{   name)rS   	namespacerX   rX   rY   in_namespace>  s
   
r   rg   	list[Any]kwargsdict[str, Any]	broadcastr   )Optional[ELEMENTWISE_TYPE_PROMOTION_KIND]convert_input_to_bool tuple[list[Any], dict[str, Any]]c                   s  dd t  D }dd  D }|s|s fS |s|ri|r$tjndd  D }|dd  D  t|d|i|rF |d  n|d   fd	d
fdd D  fdd D |rtt	t
 fdd|D fdd|D  }t	|d  }	t||d t| D ]\}
}| |
< qt||t|d  D ]\}}||< qtt D ]}
t |
 tjrt |
 |	 |
< qD ]}t| tjrt| |	|< qЈ fS )Nc                 S     g | ]\}}t |tr|qS rX   r_   r9   r   rf   r   rX   rX   rY   r   M      z"transform_args.<locals>.<listcomp>c                 S  r   rX   r   r   kvrX   rX   rY   r   N  r   c                 S  s*   g | ]}t |ttjfst|d r|qS r   )r_   r   r   r   hasattrr   arX   rX   rY   r   X  s    c                 s  s    | ]
}t |d r|V  qdS )r   N)r   r   rX   rX   rY   r   ^      z!transform_args.<locals>.<genexpr>r   r   c                   s6   t | tr
t| S t | tjrtj| j dS | S )Nvaluer   ri   )r_   r9   to_dtyper*   Constantr   )r   )ri   r   rX   rY   promotei  s
   

ztransform_args.<locals>.promotec                   r   rX   rX   r   r   rX   rY   r   q  r   c                   s   i | ]	\}}| |qS rX   rX   r   r   rX   rY   
<dictcomp>r      z"transform_args.<locals>.<dictcomp>c                 3      | ]} | V  qd S r~   rX   r   rf   rg   rX   rY   r   x      c                 3  r   r~   rX   r   r   r   rX   rY   r   y  r   )r^   itemsrp   r   extendvaluesr   rb   broadcast_tensorsr]   	itertoolschainr   zipr   ranger_   r*   r   r/   create)rg   r   r   r   r   args_indiceskwargs_indicespromoting_argsbroadcastedsizerf   r   r   rX   )rg   ri   r   r   r   rY   transform_argsF  sZ   

r   c                   s>   t   fdd}t| }t| tt|| |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                    s*   t | dksJ  | i |}t| |S )Nr%   )r   r:   )rg   r   rd   	decomp_fnrX   rY   wrapped  s   z+_register_foreach_lowering.<locals>.wrapped)	functoolswrapsr   rU   r   rL   dictfromkeys)r   r   r   aten_fnsrX   r   rY   _register_foreach_lowering  s   
r   c                   s<   t  fdd}t  tt | |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                    s   t | } t|}d}t| dkr!t| d t tfr!d}t | d } tdd  D s9tdd | D r9J dt| |\} }|rH| g} | i |}t	| |S )	NFr'   r   Tc                 s  s"    | ]}|t v pt|d V  qdS )_c10d_functionalN)	fallbacksr   )r   rl   rX   rX   rY   r         
z6_register_lowering.<locals>.wrapped.<locals>.<genexpr>c                 s      | ]}|d kV  qdS )rd   NrX   r   rX   rX   rY   r     r   zout= ops aren't yet supported)
r]   r   r   r_   r   allanykeysr   r:   )rg   r   unpackedrd   r   r   r   r   r   rX   rY   r     s(   
z#_register_lowering.<locals>.wrapped)r   r   r   rL   r   r   r   )r   r   r   r   r   r   rX   r	  rY   _register_lowering  s
   r
  F.Callable[[Callable[_P, _T]], Callable[_P, _T]]c                 C  s   t jt| |||dS )z+
    Shim to support decorator syntax.
    r   r   r   )r   partialr
  )r   r   r   r   rX   rX   rY   register_lowering  s   r  c                 C  s   g }t jt| t|tjjdD ]O\}}tjjj	j
t|dddr(|| qtjjj	j
t|dddr<|| qtjj|| tt|jtt|jk rZ|| q|| q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.
    )	fillvaluer'   Tsize_oblivious)r   zip_longestreversedr   SOnerH   rP   sizevars	shape_envevaluate_exprEqrc   guard_equalsr   expandfree_symbolsr   )r   boutputr   yrX   rX   rY   broadcast_symbolic_shapes  s   $

 r   c              
     s,  |d u s|d u sJ d|d u r|d u rt j}tdd | D s"| S tdd | D rC|p3t| d|ifdd  fdd	| D S td
d | D }g }| D ]C}t|ttfrq|	t
tj|| | dt|  qPt|tjr|	t
t|| | dt|  qP|	| qP|S )NzEonly one of override_return_dtype or type_promotion_kind may be givenc                 s  s"    | ]}t |tjttfV  qd S r~   )r_   r   r   r   floatr   rX   rX   rY   r          z$promote_constants.<locals>.<genexpr>c                 s  s"    | ]}t |tttjfV  qd S r~   )r_   r   r!  r   r   r   rX   rX   rY   r     r"  r   c                   s4   t | tjrtj|  td dS tj|  td dS )Nindexr   ri   r   )r_   r   r   r*   r0   r=   r   r   r   rX   rY   
const_func  s
   
z%promote_constants.<locals>.const_funcc                   r   rX   rX   r   )r%  rX   rY   r   "  r   z%promote_constants.<locals>.<listcomp>c                 s  s&    | ]}t |tttjfr|V  qd S r~   )r_   r9   r/   r*   r   r   rX   rX   rY   r   #  s   $ r   r#  )r   DEFAULTr  r  r   nextr_   r   r!  rc   r/   r   r*   r   r   get_device_or_errorr]   r   r   r   r0   )inputsoverride_return_dtyper   exrd   r   rX   )r%  r   rY   promote_constants  sL   

	r,  c              	     s$   d dd fdd}|S )Nalphar)  r9   c              
     s  d urt dd D rrJ  S tr2| d ur1| dkr1ttd | d< n| d u s8J dd D d  pLd   ttd  j	dd  D ]!}t
|tjstt| ksJ d d	 d	|  q^tjtjftjd uottjd
d d uotjjjd uotjjjddo v  
f
dd}	sd }D ]}t| j	r| } nq|sՈd  }	p|}tj| |dS )Nc                 s  "    | ]}t |tot|V  qd S r~   r_   r1   r2   r   r   rX   rX   rY   r   H  r  z0make_pointwise.<locals>.inner.<locals>.<genexpr>r'   c                 S     g | ]}|  qS rX   make_loaderr   rX   rX   rY   r   U  r   z1make_pointwise.<locals>.inner.<locals>.<listcomp>r   zndim mismatch  rQ   low_precision_pointwise_barrierFc                   s   t  t 
ksJ d  d
 tjkr&	d ur&	 fddD  S r:r:tjkr: fddD  S g }tD ]'\}}| }|  }rb|v rbtj||dd}t||}|| q@| }r|tj|dd}t|S |S )Nzwrong ndim r6  c                      g | ]}| qS rX   rX   r   loadr$  rX   rY   r   n  r   zCmake_pointwise.<locals>.inner.<locals>.inner_fn.<locals>.<listcomp>c                   r8  rX   rX   r9  r;  rX   rY   r   t  r   F)use_compute_types)	r   rp   r   float64r^   r   rG   r   rc   )r$  inputs_loaded	inp_indexr:  rd   	inp_dtypedowncast)
r   emulate_precision_castsrl   r)  is_gpu_deviceloaders	low_pr_fpoverride_fn_when_gpu_float64override_fn_when_input_boolrangesr;  rY   inner_fnk  s,   $
z/make_pointwise.<locals>.inner.<locals>.inner_fnri   r   rI  rH  )r  r,  r]   mulr   r   r?   r=   rb   typer_   r*   BaseConstantr   rp   bfloat16float16rH   rP   r   rQ   metagetr6   r   )r.  r)  otherrI  ri   rf   allow_alpharl   override_devicerF  rG  r*  triton_fallback)r   rB  r)  rC  rD  rE  rH  rY   innerG  sd   

zmake_pointwise.<locals>.inner)r)  r9   rX   )rl   r*  rU  rG  rF  rT  rV  rW  rX   rS  rY   make_pointwise>  s    	RrX  c                   s   ddd fdd}|S )Nr'   r-  r)  list[list[TensorBox]]c                   sT  t tjjjdkptjjjtv pt }d }|D ]}t|t	t
fr$|} nq|d us-J dg }|D ]}t|t	t
fsE||gt |  q1|| q1tt| }d gt | }| D ]@\\}}	}
g }|
D ]-\}} rt|d| i}n| }|||< tj|tjr|	r|r|  ||  qf|rtj| q\tdd |D sJ |S )Nr   z1at least one input must be a list to a foreach opr.  c                 s      | ]}|d uV  qd S r~   rX   r   rX   rX   rY   r     r   z8make_foreach_pointwise.<locals>.inner.<locals>.<genexpr>)r   rH   rP   rQ   rR   rT   inplace_foreach_opsrZ   r_   r]   r   rc   rk   r   r   has_featureBackendFeatureFOREACHrealizeget_operation_nameregister_operation_listr  )r.  r)  realize_outputsa_list_inputinputbroadcast_inputsgroupsoutputsri   rh   groupoperation_list
output_indrg   r  rT  pw_fnrX   rY   rW    sZ   
z%make_foreach_pointwise.<locals>.inner)r)  rY  rX   )rl  rT  rW  rX   rk  rY   make_foreach_pointwise  s   6rm  r   r9   torch.dtypec                   s>   |    kr|rt| S | S  fdd}t| d| S )Nc                   s   t j|  dS )N)	src_dtype)rG   r   r   r   ro  rX   rY   	_to_dtype     zto_dtype.<locals>._to_dtyper*  )r   clonerX  )r   r   copyrq  rX   rp  rY   r     s
   r   r   c                 O  s   ddl m} |}| j}||tjd}t| |j|  W d   n1 s'w   Y  |j}|s3J t|}dgt	| }	|
 D ]0\\}
}}g }|D ]\}}||	|< tj|
tjri|ri|  ||  qL|rrtj| qBtdd |	D s~J |	S )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                 s  rZ  r~   rX   r   rX   rX   rY   r   
  r   z_foreach_map.<locals>.<genexpr>)subgraph_loweringrw  graph_modulerH   rP   set_graph_handlerrungraph_outputsrk   r   r   r\  r]  r^  r_  rc   r`  ra  r  )subgraphrg   r   rw  r)  gmpw_subgraphsub_outputsrf  rg  ri   rh   rh  ri  rj  r  rX   rX   rY   _foreach_map  s6   r  c                 C  sZ   |j s|  j r&|  rt| |d}tj||  |S ttj	j
dd| |S t| |ddS )Nr   Fadd_to_fallback_setTru  )
is_complexr   r   
empty_liker*   InplaceCopyFallbackr   fallback_handlerprimsconvert_element_typedefaultr   )r   r   dstrX   rX   rY   _convert_element_type  s   r  r  c                C  sb   |   }||kr|rt| S | S dd }||}||}||kr)ttjj| |S tt| |S )Nc                 S  s   | j r	t| jS t| jS r~   )is_floating_pointrp   finfobitsiinfor   rX   rX   rY   _get_primitive_bitwidth#  s   z1to_dtype_bitcast.<locals>._get_primitive_bitwidth)	r   rt  r  atenviewr   r9   r.   r   )r   r   ru  x_dtyper  src_bitsdst_bitsrX   rX   rY   to_dtype_bitcast  s   r  c                 C  s8   |j s|  j rttjtjjj	j
| |S t| |S r~   )r  r   r9   r   r*   ComplexViewrp   rG   r  r  r   r  r   r   rX   rX   rY   _view_dtype2  s
   
r  ru  non_blockingri   torch.devicec                C  s:   t |}|  |kr|rt| S | S ttj| ||S r~   )r=   rb   rt  r9   r   r*   
DeviceCopy)r   ri   ru  r  rX   rX   rY   	to_device;  s   r  c                 C  s   t | |d|dS )NTr  )r  )r   ri   r  rX   rX   rY   _device_putB     r  Tc
                 C  s   |p| j }t|}
|rtd| }td| || t||| |dur't|}t|
|||r/|nd||	d}
t| |||d|
}
tt|rPttt|d|d|
 |
S )z3A pointwise function that maps ops.{name} to inputs
libdevice_N)r*  rG  rF  rT  rV  r  )r   r   )__name__r4   rD   rX  r  r   r  r   )r   r   r   r   r   r*  rG  rT  use_libdevice_for_f64rV  rl   fn_libdevicerX   rX   rY   register_pointwiseG  sJ   



r  c                    sx   d} t d  fdd} fdd}t|t|tjdgfdd}ttj|}tt| r:tt	t| d	d
| |S )z2A pointwise function that maps ops.frexp to inputsfrexpc                        | i |d S Nr   rX   rg   r   r  rX   rY   frexp0     zregister_frexp.<locals>.frexp0c                    r  Nr'   rX   r  r  rX   rY   frexp1  r  zregister_frexp.<locals>.frexp1rs  c                    s$    d | i | d | i |fS Nr   r'   rX   r  )pw_fnsrX   rY   rl     s   $zregister_frexp.<locals>.fnNrv  )
r4   rX  rp   int32r  r  r  r   r  r   )r   r  r  rl   rX   )r  r  rY   register_frexp{  s*   
r  c                 C  s   t ||d}t| |}|S )NrT  )rm  r   )r   pointwise_lowering_fnrT  rl   rX   rX   rY   register_foreach_pointwise  s   
r  )r   r   c                   s  dd }t |ttfrt||}t |ttfrt||}| ||g t d  d tjd}dd t D }t|t	 fdd|D  D ]\}}| |< qFt
t D ]}t  | tjrqt | t |d	    |< qUt||d
 d	 t d |t d |S )Nc                  W  
   t j|  S r~   )rG   wherer   rX   rX   rY   rl        
zwhere.<locals>.fnr'   r%   rv  c                 S  r   rX   r   r   rX   rX   rY   r     r   zwhere.<locals>.<listcomp>c                      g | ]} | qS rX   rX   r   r   rX   rY   r     r   r   rs  )r_   r!  r   constant_liker   r   r&  r^   r   r   r   r   r*   r   r/   r   r]   r   rX  r   )r   r   r  rl   r   indicesrf   r   rX   r   rY   r    s&   
$
$
r  c                  G  s   t | dkrt| d ttfrt| d  S ttdd | D g }g }| D ]$}| }t |t |ks?t	dd t
||D rDt||}|| q%|S )Nr'   r   c                 S  r3  rX   )r   r   rX   rX   rY   r     r   z%broadcast_tensors.<locals>.<listcomp>c                 s  s    | ]?\}}t jjjjt|d ddo"t jjjjt|d dd p?t jjjjt|d dd o?t jjjjt|d ddV  qdS )r'   Tr  N)rH   rP   r  r  r  r   r  r   r   r  rX   rX   rY   r     s"    


	

z$broadcast_tensors.<locals>.<genexpr>)r   r_   r]   r   r   r   reducer   r   r  r   r  rc   )r)  rT   rg  r   sizesrX   rX   rY   r     s   
r   c                 C  s   | S r~   rX   r   rX   rX   rY   nop     r  
lift_freshc                 C  s   t | tsJ |d u rtt| jS t |ttjfr"tj	j
|ntdd |D }tt|  |}tt |ts=|fn|}g }t|  D ]\}}||v r]tj	j
jt|dddsb|| qH||  krnt| |S | S )Nc                 s  s    | ]
}t jj|V  qd S r~   rH   rP   r  evaluate_static_shaper   drX   rX   rY   r     r   zsqueeze.<locals>.<genexpr>r'   Tr  )r_   r9   r8   r   ra   r   r   r   rH   rP   r  r  r   r   r   r   r   r^   r  r  rc   r  )r   r   dims	new_shaper  srX   rX   rY   squeeze  s"   
r  c                 C  s   t t| |S r~   )rt  r  )r   r   rX   rX   rY   squeeze_copy     r  c                 C  2   t | |}t| tsJ t|tsJ |j| _| S r~   )r  r_   r9   ra   r   r   valrX   rX   rY   squeeze_  
   
r  c                 C  2   t | rt| dtjdS td}t|tjd| S )NFr   isinfrs  r   	full_likerp   r   r4   rX  r   rl   rX   rX   rY   r       r  c                 C  r  )NFr   isnanrs  r  r  rX   rX   rY   r    r  r  c                 C  $   t | rt| S td}t|| S )Nceilr   rt  r4   rX  r  rX   rX   rY   r  %     r  c                 C  r  )Nfloorr  r  rX   rX   rY   r  -  r  r  c                 C  r  )Nroundr  r  rX   rX   rY   r  5  s   r  c                 C  r  )Ntruncr  r  rX   rX   rY   r  >  r  r  c                 C  s   ddl m} t| g\} t| tjrt| t|S t| t	s!J t|t
tfs*J t|  t|kr6| S ||  s]tjjt|  }|dkr]||s]| tjjt||  t	t| jt|S )Nr   )free_unbacked_symbols)%torch.fx.experimental.symbolic_shapesr  r,  r_   r*   rM  r/   r   r   r9   r]   r   rH   rP   r  	size_hintrE   
mark_reusera   )r   r  r  x_size_productrX   rX   rY   r  F  s   r  c                 C  sL   t |}|D ]}d||< q| }t|D ]\}}|dkr t||}qt||S Nr2  )r]   r^   	unsqueezer  )r   shapebroadcast_dimensionsr  broadcast_dimensionr   idxr   rX   rX   rY   broadcast_in_dim`  s   


r  c                 C  s   t | | S r~   )r  r   )r   r  rX   rX   rY   	expand_asn  r  r  c                   sR  t |   tt kr$tjjgtt      t| t  } tt|  ks0J t |  }d}ttD ]}| dkrHd}|| |  ||< q>|r`t|| 	 | 
 dS tdd t D rstt| |S  fdd}tjjt }|dkr| tjjt||  |  tj| 
 | 	 |t |d	S )
NFr   Tr   ri   c                 s  s$    | ]\}}|d kp|d kV  qdS r'   NrX   r  rX   rX   rY   r        " zrepeat.<locals>.<genexpr>c                   st   t | t ks
J t| } tt D ]!}| dkr5 | dkr)tjj| |< qt| | d | | |< q| S r  )r   r]   r   r   r  Zeror$   )r$  rf   old_sizerepeatsx_loaderrX   rY   rI    s   zrepeat.<locals>.inner_fnrJ  )r]   r   r   r   r  r  r  r   emptyr   rb   r  r   rt  r  rH   rP   r  r  rE   r  r5  r6   r   )r   r  new_sizezero_tensorrf   rI  old_size_productrX   r  rY   repeats  s8   r  c                 C  s2   t | tsJ t |ttfsJ tt| j|S r~   )r_   r9   r]   r   r;   r   ra   )r   r  rX   rX   rY   r    s   r  c                 C  s6   t | tsJ t |ttfsJ tt| jt|S r~   )r_   r9   r]   r   r5   r   ra   )r   r  rX   rX   rY   permute  s   r              c              	   C  s8   t | tsJ t| |d}ttjj| j|||||dS )Nr   clamp)r_   r9   _validate_dimr*   	SliceViewr   ra   )r   r   startendstepr  rX   rX   rY   slice_  s   r  c              	   C  s   t | trt | jtjr| j } |   t| s"td|  dt	| \}}t
|j|jdd |D dd |D t|p@d}ttj||dS )Nzunrealized as_strided(z, ...)c                 S     g | ]}t |qS rX   r   r  r   r  rX   rX   rY   r         zas_strided.<locals>.<listcomp>c                 S  r  rX   r  r  rX   rX   rY   r     r	  r   ra   layout)r_   r9   ra   r*   BaseViewunwrap_viewr_  is_storage_and_layoutr   as_storage_and_layoutFixedLayoutri   r   r   r  ReinterpretView)r   r   stridestorage_offsetstorage
old_layout
new_layoutrX   rX   rY   
as_strided  s   

r  c                 C  s$   t | tsJ t| |||j| _| S r~   )r_   r9   r  ra   )r   r   r  r  rX   rX   rY   as_strided_  s   r  c                 C  s   t | |||}t|S r~   )r  rt  )r   r   r  r  resultrX   rX   rY   as_strided_copy  s   r  c                   s   g d}D ]} |||    f d d }qdd D  fdd}td  }d d | < tjd  d  ||dS )Nr   r2  c                 S  r3  rX   r4  r1  rX   rX   rY   r     r   z!pointwise_cat.<locals>.<listcomp>c           
   	     s@  t |  tj}g }g }ttD ]n  dkr t dtjn
t   d tj}t   d tj}t ||}t ||} dkrI|}n td krT|}nt 	||}|
| t| t   d  < |
t | fddd q|d }	ttd ddD ] t |  |  |	}	q|	S )Nr   r'   c                     s     S r~   rX   rX   )rf   idx_loadinputs_loadersrX   rY   <lambda>      z1pointwise_cat.<locals>.inner_fn.<locals>.<lambda>        r2  r%   )rG   
index_exprrp   int64r   r   constantgeltand_rc   r]   r"   maskedr  )
r  idx_dimmasksmasked_loadsr  r  
start_condend_condmasknext_valr   r)  r  inputs_ranges)rf   r  rY   rI    sD   
zpointwise_cat.<locals>.inner_fnrJ  )rc   r   r]   r6   r   rb   r   )r)  r   prev_endr   rI  r  rX   r.  rY   pointwise_cat  s   0

r1  rd  scaleszero_pointsaxis	quant_min	quant_maxc              	     s   t  dksJ dt  dksJ d|  tjkr%t| tj} |  tjks5J d|    t |  k sHJ dt |   |     f	dd}tj	| 
 ||  d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           
        s   |   f}| }|}|}t tjd\}}jtjkr(t|tj}jtjkr5t|tj}t|}t|| | }t	|t
||}	t|	S Nr   )_create_constantsrp   float32r   rG   r   r  
reciprocalr  maximumminimum)
r  channel_idxrd  scale
zero_pointqminqmax	inv_scaler  clamped	r4  r   input_loaderr6  r5  r2  scales_loaderr3  zero_points_loaderrX   rY   rI  ;  s   

z;quantized_decomposed_quantize_per_channel.<locals>.inner_fnrJ  )r   r   r   rp   rN  r   r=  r5  r6   r   rb   )rd  r2  r3  r4  r5  r6  r   rI  rX   rH  rY   )quantized_decomposed_quantize_per_channel!  s(   
rL  	out_dtyperN  Optional[torch.dtype]c          	        s   t  dksJ dt  dksJ d|  |ks*J d| d|    t |  k s=J dt |   d u rDtj|     fdd}tj|  ||  d	S )
Nr'   r7  r8  Expecting input to have dtype , but got dtype: r:  c                   s   |   f}| }|}|}j tjkrt|tj}j tjkr+t|tj}tt|tj|| }t|}|S r~   )r   rp   r=  rG   r   sub)r  rA  rd  rB  rC  r  r4  rI  rN  r2  rJ  r3  rK  rX   rY   rI  r  s   
z=quantized_decomposed_dequantize_per_channel.<locals>.inner_fnrJ  	r   r   r   rp   r=  r5  r6   r   rb   )	rd  r2  r3  r4  r5  r6  r   rN  rI  rX   rS  rY   +quantized_decomposed_dequantize_per_channelT  s(   rU  rB  r!  rC  c                   s   |   tjkrt| tj} |   tjksJ d|    |   fdd}tj|   t	j
|t|t|d|  dS )Nr9  c           	        sf   | }t d| |tjd\}}t|| | }t tjd\}}tt|||}t| S )N      ?r   )r<  rp   r=  rG   r  r@  r?  r   )	r  rB  rC  rd  rF  r  rD  rE  rG  r   rI  r6  r5  rX   rY   rI    s   
zBquantized_decomposed_quantize_per_tensor_default.<locals>.inner_fnrB  rC  rJ  )r   rp   rN  r   r=  r5  r6   r   rb   r   r  r!  r   r   rd  rB  rC  r5  r6  r   rI  rX   rW  rY   0quantized_decomposed_quantize_per_tensor_default  s   
rZ  c                  sv   |   |ksJ d| d|    d u rtj|    fdd}tj|  tj|t	|t
|d|  dS )NrP  rQ  c                   sF    | }t ||tjd\}}tt|tj|| }t|}|S r;  )r<  rp   r=  rG   rR  r   )r  rB  rC  rd  r  rI  rN  rX   rY   rI    s
   zDquantized_decomposed_dequantize_per_tensor_default.<locals>.inner_fnrX  rJ  )r   rp   r=  r5  r6   r   rb   r   r  r!  r   r   rd  rB  rC  r5  r6  r   rN  rI  rX   r[  rY   2quantized_decomposed_dequantize_per_tensor_default  s   r]  c                   s   |   tjkrt| tj} |   tjksJ d|    t dks9t dkr5 d dks9J dt dksUt dkrQ d dksUJ d|     fdd}tj	| 
  ||  dS )	Nr9  r   r'   expect scale as scalar tensor"expect zero_point as scalar tensorc                   s   | }t  dkrdnd}t  dkrdnd}jtjkr-t|tj}jtjkr:t|tj}t|t| | }t	tjd\}}t
t|||}t| S )Nr'   r   rX   r   )r   r   r   rp   r=  rG   r   r  r>  r<  r@  r?  )r  rd  _scale_zero_pointr  rD  rE  rG  r   rI  r6  r5  rB  scale_loaderrC  zero_point_loaderrX   rY   rI    s   zAquantized_decomposed_quantize_per_tensor_tensor.<locals>.inner_fnrJ  )r   rp   rN  r   r=  r   r   r5  r6   r   rb   rY  rX   rc  rY   /quantized_decomposed_quantize_per_tensor_tensor  s.   ""rf  c                  s   t  dkst  dkr d dksJ dt  dks8t  dkr4 d dks8J d|  |ksJJ d| d|   d u rQtj|      fdd}tj|  ||  d	S )
Nr   r'   r^  r_  rP  rQ  c                   s    | }t  dkrdnd}t  dkrdnd}jtjkr-t|tj}jtjkr:t|tj}tt|tj|| }t|}|S )Nr'   r`  rX   )r   r   r   rp   r=  rG   r   rR  )r  rd  ra  rb  r  rI  rN  rB  rd  rC  re  rX   rY   rI  %  s   zCquantized_decomposed_dequantize_per_tensor_tensor.<locals>.inner_fnrJ  rT  r\  rX   rg  rY   1quantized_decomposed_dequantize_per_tensor_tensor  s.   ""rh  c           
        s  | d   jdk}|r:tdd | D r:| D ]}|  qtdd | D r1ttjg| R  \} }ttjj| |S t	| dkrFt
| d S t| d |d}t| dtjifdd	| D } d%dddd fddtfdd| D }d&fddtjrt| |S |rttj| |S fddd}d d'ddt	| |kst	| tjkrt fd d| D rtfd!dtjjD }tfd"d| D o|}tfd#d| D otfd$d| D  }	|s|	r|st| |S ttj| |S )(Nr   cpuc                 s  s$    | ]}|  tjtjfv V  qd S r~   )r   rp   int8uint8r   rd  rX   rX   rY   r   <  s    
zcat.<locals>.<genexpr>c                 s  s     | ]}t | d kV  qdS )r   N)r   r   rl  rX   rX   rY   r   C      r'   r   c                   s   g | ]}t | qS rX   r   r1  r   rX   rY   r   N  r	  zcat.<locals>.<listcomp>r   Union[TensorBox, ir.StorageBox]rn   	ir.IRNodec                 S  s>   t | trt | jtjr| j S | jS t | tjr| jS | S r~   )r_   r9   ra   r*   r  r  
StorageBoxr   rX   rX   rY   unwrap_tensorP  s   

zcat.<locals>.unwrap_tensorc                 S  s   t | tjot | jtjS r~   )r_   r*   ComputedBufferra   r7   rj   rX   rX   rY   is_reduction\     zcat.<locals>.is_reductionc                   sJ   t | ttjfr | S | p$t | tjo$t fdd|  D S )Nc                 3  s     | ]} t j|V  qd S r~   )rH   rP   
get_buffer)r   readcan_fuse_reductionrX   rY   r   e  s
    
z2cat.<locals>.can_fuse_reduction.<locals>.<genexpr>)r_   r9   r*   rq  r6   r  get_read_namesrt  )rz  ru  rr  rX   rY   rz  _  s   zcat.<locals>.can_fuse_reductionc                 3      | ]} |V  qd S r~   rX   r   rj   ry  rX   rY   r   l  r   r   c                   sZ   t | rt j| dd\}}t j| S t| tt jfr# | S t| t jr+dS dS )NF)freezeT)	r*   r  r  ConcatKernelcan_realize_into_without_copyr_   r9   rq  r6   )r   r  r   )should_lower_cat_inputrr  rX   rY   r  n  s   
z#cat.<locals>.should_lower_cat_inputc                   s\   t | ttjfr | S t | tjsdS |  j}|  D ]}| tj	
|7 }q|S r  )r_   r9   r*   rq  r6   inner_fn_opcountnum_opsr{  rH   rP   rw  )r   countrx  )op_countrr  rX   rY   r    s   
zcat.<locals>.op_countr   r%   rS   torch._ops.OpOverloadc                 S  s   | t jjt jjfv S r~   )r  catr  constant_pad_ndrS   rX   rX   rY   additional_pointwise_ops     z%cat.<locals>.additional_pointwise_opsc                 3  s    | ]	}| kV  qd S r~   rX   r}  )MAX_SIMPLE_OP_COUNTr  rX   rY   r         c                 3  s    | ]}t | V  qd S r~   )r@   )r   use)r  rX   rY   r     s
    
c                 3  r|  r~   rX   r1  r  rX   rY   r     r   c                 3  r|  r~   rX   r1  r  rX   rY   r         
c                 3  r|  r~   rX   r}  ry  rX   rY   r     r   )r   ro  rn   rp  )rn   r   rS   r  )rb   rL  r  r_  require_channels_lastr  r  r  r  r   rt  r   r   r   r&  r  r(   force_pointwise_catr1  r9   r*   r  r   max_pointwise_cat_inputsrH   rQ   rR   )
r)  r   
cpu_devicerd  r   fusable_reductionMAX_COMPLEX_POINTWISE_CATpointwise_usesfuse_pointwise_usehorizontal_fuse_catrX   )r  r  rz  r   ru  r  r  rr  rY   r  9  s`   




r  offsetdim1dim2c                   s  |   ttdtdtkfdd tjjt	|d}|rBtjj
tjj |  d}ntjj
tjj  | d}d |r`| df nd|f fddtD }||  fdd	}ttj| ||S )
N)r  rankc                     s   d  d S )Nz(diagonal dimensions cannot be identical z, rX   rX   r  r  rX   rY   r        zdiagonal.<locals>.<lambda>r   )r   r   c                   s    g | ]\}}| fvr|qS rX   rX   )r   rf   r  r  rX   rY   r          zdiagonal.<locals>.<listcomp>c                   s   | d }dgt  }d}tD ]&}|kr | d  ||< q|kr-| d  ||< q| | ||< |d7 }q|t d ksBJ |S )Nr2  r   r'   r%   )r   r   )r  diag_idxoriginal_idxcur_dimr  base_idxr  r  num_dimsoriginal_shaperX   rY   	reindexer  s   
zdiagonal.<locals>.reindexer)r   r   r   r   rH   rP   r  r  r   Ltevaluate_maxevaluate_minr^   rc   r9   r*   GenericViewr   )rd  r  r  r  offset_negative	diag_sizer  r  rX   r  rY   diagonal  s:   
r  c                 C  s   t t| |||S r~   )rt  r  )rd  r  r  r  rX   rX   rY   diagonal_copy     r  c                 C  $   t | }t||||}t|| |S r~   )rt  r  	mutate_to)rd  srcr  r  r  r  rT   rX   rX   rY   diagonal_scatter     
r  c                 C  s,   t ||  | }tt| |||d |S r  )r;   handle_negative_indexr   r  r  )r   r   r  rX   rX   rY   select  s   r  c           
   
   C  s   t | |d}|}t|ttfs2|  | }tjjt	|| d |}|g| }||d |  |d< g }d}|D ]}|| }	|
t| |||	dd |	}q8|S )Nr   r'   r2  Fr  )r   r_   r]   r   r   rH   rP   r  r  r!   rc   r  )
r   r  r   sizes_x_sizechunksr  r  r   r  rX   rX   rY   split   s    
r  c                 C  s   t | ||S r~   )r  )r   r  r   rX   rX   rY   split_with_sizes     r  c                   s>   t  d tjj   } fddt|D }|S )Nr   c                   s   g | ]}t  |qS rX   )r  r   r   r   rX   rY   r   &      zunbind.<locals>.<listcomp>)r   rH   rP   r  r  r   r   )r   r   r  r  rX   r  rY   unbind"  s   r  c                   s   |   }t|}t|| |dkrtt| d|dS |  }tjj}||| |	d t
|| d }||dkrK| |t|| | g |d   || d d  |}	 fdd}
ttj| |	|
S )Nr   )r  r'   c                   s:   | d |     }g | d   ||  d d R S )Nr2  r'   rX   )r  dim_idxr   r  rX   rY   r  >  s   &zunfold.<locals>.reindexer)r   r   r   r  r  rH   rP   r  	guard_leqguard_ltr!   r  r  r    r9   r*   r  r   )r   	dimensionr   r  r  ndimdim_sizer  new_dim_sizeout_sizer  rX   r  rY   unfold*  s   
(r  c                 C  s2   t | |d}t|  }||tjj t| |S r  )r   r]   r   insertr   r  r  r  )r   r   r  rX   rX   rY   r  E  s   
r  c                 C  r  r~   )r  r_   r9   ra   r  rX   rX   rY   
unsqueeze_M  r  r  c                 C  sZ   t jjjt|}t|  }|dk r||| 7 }d|  kr(|| k s+J  J |S r  )	rH   rP   r  r  r  r   sympifyr   r   )r   r   r  r  rX   rX   rY   r   V  s    r   r2  c                 C  sT   t | |d}tjj|  | d }t| |d|}t| |||d }t|t|S )Nr   r%   )	r   rH   rP   r  r  r   r  rK  sigmoid)r   r   new_lenr   r  rX   rX   rY   glu_  s
   r  c                   s$   |rt    fdd}d|_|S )Nc                    s*   dd }t |tjj g| R i |S )Nc                 S  s   t | tjrt| S | S r~   )r_   r*   r1   r9   r   r   rX   rX   rY   wrap_tensorsn  rv  z7fallback_handler.<locals>.handler.<locals>.wrap_tensors)pytreetree_mapr*   FallbackKernelr   )rg   r   r  kernelrX   rY   handlerm  s   z!fallback_handler.<locals>.handlerT)r  r   _is_fallback_handler)r  r  r  rX   r  rY   r  i  s
   
	r  c                   C     t d d S )NzjTorchinductor does not support code generation for complex operators. Performance may be worse than eager.)warningswarnrX   rX   rX   rY   _warn_complex_not_supported{  s   r  rj   torch.Tensorc                 C  sr   |   r|r|jtjjjjtjjjj	fv rdS t
  dS | jtjkr7|s&dS |jtjjtjj	fv p5t|j S dS )z0Do not support reading or writing to this tensorFT)r  rT   rp   rG   r  r  r   r  r  r  r  float8_e8m0fnur  rA   rj   parentrV   rX   rX   rY   unsupported_input_tensor  s&   

	r  c                 C  s   t | |rdS | jotjS )z2Do not support writing tensor but can read from itT)r  is_cpur(   disable_cpp_codegenr  rX   rX   rY   unsupported_output_tensor  s   
r  rV   torch.fx.Nodec                 C  sh   | j tjju r	dS | j tjju rdS dd }tj| ji | jD ]}||| ddr, dS q || | ddS )NFc                 S  st   t | tjjs	dS d| jvrdS t| jd D ]}t |tjjs"q|r.t	||| r- dS qt
||| r7 dS qdS )NFr  T)r_   rp   fxNoderP  r  tree_leaves_subclasses
FakeTensorr  r  )rV   r  	is_outputrP  rX   rX   rY   check_skip_condition  s   
zCfallback_node_due_to_unsupported_type.<locals>.check_skip_condition)r  T)	rT   r  view_as_complexr  lift_fresh_copyr  arg_tree_leavesrg   r   )rV   allow_cpu_inputsr  r   rX   rX   rY   %fallback_node_due_to_unsupported_type  s   r  c                   s   | t vs|sJ d|  |r>ttdr>t| gr>tjr%| tjj	j
v s>|s>tjjjr6dtjj_td td|  d fdd}t| tjjr]|  D ]}t| |}|| qOd S t| tjjtjjfrn||  d S td	|  d
t|  )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                   s.   t |   d urt|   t| d dt| S Nrv  )r   r   r  r  )op_overloadlayout_constraintrX   rY   register_fallback  s   

z(make_fallback.<locals>.register_fallbackzUnsupported fallback z with type )r,   r   osgetenvr-   r(   fallback_randomrp   _decompdecompositions_for_rngextra_random_decomps_dynamosuppress_errorslogwarningrv   r_   rz   r   r   r   r{   HigherOrderOperatorRuntimeErrorrL  )rS   r  r  override_decompr  olr  rX   r  rY   make_fallback  s>   




r  c                 C  s$   d}| D ]}|| }qt |tjdS )z
    TorchInductor offset calculation differs from PyTorch eager offset
    calculation for random ops (tl.rand vs torch.rand). In future, we should
    strive for same impl for tl.rand and torch.rand.
    r'   r   tensorrp   r!  )r  numelr  rX   rX   rY   philox_rand_offset   s   
r  c           	        sd   t | | t j|  | |  fdd}tj| |t| d}t	| }||fS )Nc                   sV   t g tj}t g tj}t t | tj|}t ||}t | S r~   )rG   r   rp   r  r   r   rand)r$  seed_index_exproffset_index_exprrand_index_exprr  r   offset_loader
random_posseed_loaderrX   rY   rI    s   zphilox_rand.<locals>.inner_fnrJ  )
r*   r  FlexibleLayoutcontiguous_stridesmake_indexerr5  r6   r   r]   r  )	r   seedr  r  ri   r   rI  random_values_nodeoffset_noderX   r  rY   philox_rand  s&   
r  c              	   C  s.   t jrttjtjtj	j
| ||S td)Nz&should be handled in replace_random.py)r(   r  r  r  r9   r   r*   r  r  native_dropoutr  rv   )r   ptrainrX   rX   rY   r  3  s   r  c                 G  sj   t js|  tdksJ d|   t|dks!t|d tr%t	j
jnt	j
j}tj|| g|R   | S )Nri  Tthis should be handled in decomps unless config.fallback_random or the device is CPUr   )r(   r  rb   rp   ri   r_  r   r_   r!  r  
bernoulli_Tensorr*   InplaceBernoulliFallback)r   rg   r  rX   rX   rY   r   >  s   r   c                 G  s4   t js|  tdksJ dtt| g|R  S )Nri  r  )r(   r  rb   rp   ri   r   rt  )r   rg   rX   rX   rY   bernoulli_pM  s   r#  c                 C  s   t r~   rv   r   rX   rX   rY   _foobarV  r  r&  c                 C  r  )Nz1using triton random, expect difference from eager)r  info)saltrX   rX   rY   _warn_triton_random[  r  r)  c                   C  s   t tjj d S r~   )r)  rH   rP   creation_timerX   rX   rX   rY   warn_triton_random`  r  r+  c                  O  F   | dd d urt| i |S tjr|dd  t| i |S tdN	generatorz-should have been handled in replace_random.py)rQ  fallback_rand_generatorr(   r  popfallback_rand_defaultrv   r  rX   rX   rY   r  l     r  c                  O  r,  r-  )rQ  fallback_randn_generatorr(   r  r0  fallback_randn_defaultrv   r  rX   rX   rY   randnv  r2  r5  c                 C  s   t |}t j| |S r~   )r*   get_stride_orderExternKernelrequire_stride_order)input_tensorr  stride_orderrX   rX   rY   inductor_force_stride_order  s   
r;  c                 C     t d)Nz.should be handled in fuse_seed_creation_pass()r$  )ri   rX   rX   rY   inductor_seed     r=  c                 C  s   t   tt| t|S r~   )r+  r9   r   r*   RandomSeedsr=   )r  ri   rX   rX   rY   inductor_seeds  s   r@  c                   s(    fdd}t j  |g dS )Nc                   s   t   S r~   )rG   	load_seedget_namer%  r$  seedsrX   rY   rI    rr  z&inductor_lookup_seed.<locals>.inner_fnrJ  )r6   r   rb   r   )rD  r$  rI  rX   rC  rY   inductor_lookup_seed  s   rE  r  r   	list[int]r  modestrc                  s   t jrJ  dv sJ g | } tj}| }tj||| tj| |d	 |
  fdd}tj|||g | d}|  |S )N)r  r5  rF  c                   s"   t t g t| tjS r~   )r   rG   r   rp   r  r;  rH  r  r  rX   rY   rI    s   z!inductor_random.<locals>.inner_fnrJ  )r(   r  rp   r=  r(  r*   r  r  r  r  r5  r6   r   r_  )r   r  rH  r  r   ri   rI  r  rX   rJ  rY   inductor_random  s(   
rK  lowhighc                  sp   t jrJ g |}tj}| }tj|||tj||d	 |
  fdd}tj|||g |dS )NrF  c              	     s6   t g t | tjt tjt  tjS r~   )rG   	randint64r   rp   r  r!  r;  rM  rL  r  r  rX   rY   rI    s   z"inductor_randint.<locals>.inner_fnrJ  )r(   r  rp   r!  r(  r*   r  r  r  r  r5  r6   r   )rL  rM  r   r  r  r   ri   rI  rX   rO  rY   inductor_randint  s"   
rP  tb.tuple[str, sympy.Expr, sympy.Expr, sympy.Expr]c                 C  s4   |   |  d |  d |  d  |  d fS Nr2  r   )rB  r   
get_striderQ  rX   rX   rY   _boundaries_helper  s
   

rV  tuple[str, sympy.Expr]c                 C  s   |   |  d fS r  )rB  rT  rU  rX   rX   rY   _sorter_helper  r  rX  	out_int32rightsidesortersorted_sequenceselfrZ  r[  r\  Optional[str]r]  Optional[TensorBox]c          	        s   dd }|r||rd ur$|s$t tjjdd|||dS |d ur.|dkr.d|r3tjntj |   d urF  t	
 dkrY fd	d
}n
 fdd
}| }tj| ||jdS )Nc                 S  s   t j| tjS r~   )rH   rP   r\  r]  	BUCKETIZErU  rX   rX   rY   r    s    zsearchsorted.<locals>.<lambda>Fr  rY  r[  Tr'   c              	     sD   | }t j|td d u rd ntd u rd dS ddS )Nr   r]  sorter_indicesrG   	bucketizerV  rX  )r  r  index_dtyper[  r^  r]  values_loaderrX   rY   rI  	  s   
zsearchsorted.<locals>.inner_fnc              	     s\    }d fdd}t j|t|d u rd ntd u r(d dS |dS )NrQ  r9   c                   s>   |   }tttjdd t|d d  d d D S )Nc                 s  s    | ]	\}}|| V  qd S r~   rX   )r   r  rf   rX   rX   rY   r   *	  r  zNsearchsorted.<locals>.inner_fn.<locals>.get_flattened_index.<locals>.<genexpr>r2  )rT  rG   r   r   r  operatorr   r   )rQ  strides)r  rh  rX   rY   get_flattened_index&	  s   &z;searchsorted.<locals>.inner_fn.<locals>.get_flattened_indexrc  )rQ  r9   re  )r  r  rl  rg  r  rY   rI  !	  s   	
rJ  )r  r  searchsortedr!  rp   r  r!  r5  r_  r   r   rb   r6   r   r  )	r^  r_  rZ  r[  r\  r]  validate_bucketizerI  ri   rX   rg  rY   rn    s>   
rn  rZ  r[  
boundariesc                  s   t   dks
J tj| tjrtj tjs(ttj	j
dd|  |dS    |  }|  |r9tjntj fdd}tj|||  dS )Nr'   Fr  rp  c                   s"   | }t |t d}|S r  )rG   rf  rV  )r$  r  r  rq  rh  rI  r[  rX   rY   rI  ^	  s   zbucketize.<locals>.inner_fnrJ  )r   r   rH   rP   r\  r]  rb  r  r  rf  r!  r_  rb   r5  rp   r  r!  r6   r   )rd  rq  rZ  r[  ri   rI  rX   rr  rY   rf  B	  s&   rf  c                 O  $   t tjtjj||f\}}||fS r~   )r  tree_map_onlyr*   r1   r7  require_stride1r   rg   r   rX   rX   rY   require_denser	     rw  c                 O  rs  r~   )r  rt  r*   r1   r7  require_contiguousrv  rX   rX   rY   ry  y	  rx  ry  c                 O  rs  r~   )r  rt  r*   r1   r7  r  rv  rX   rX   rY   r  	  rx  r  c                   sH    fdd t  fddt| |D }  fdd| D }| |fS )Nc                   s   t  tjrdd  D }tj |S t  tr( fdd  D S t  tt	fr?t
 fddt D S  S )Nc                 S  $   g | ]}t |tjr|jjn|qS rX   r_   rp   SymIntrV   exprr  rX   rX   rY   r   	  s    zGconstrain_to_fake_tensors.<locals>.apply_constraint.<locals>.<listcomp>c                       i | ]}| | | qS rX   rX   r   key)apply_constraintr   fake_argrX   rY   r   	  s    zGconstrain_to_fake_tensors.<locals>.apply_constraint.<locals>.<dictcomp>c                 3      | ]
\}} ||V  qd S r~   rX   )r   r   f_ar  rX   rY   r   	      
zFconstrain_to_fake_tensors.<locals>.apply_constraint.<locals>.<genexpr>)r_   r*   r1   r  r7  require_exact_stridesr   r  r   r]   rL  r   )r   r  meta_stride_exprr  )r   r  rY   r  	  s   
z3constrain_to_fake_tensors.<locals>.apply_constraintc                 3  r  r~   rX   )r   r   r  r  rX   rY   r   	  r  z,constrain_to_fake_tensors.<locals>.<genexpr>c                   s    i | ]\}}| || qS rX   rX   r   r  fake_kwargsrX   rY   r   	  r  z-constrain_to_fake_tensors.<locals>.<dictcomp>)r   r   r   )rg   r   	fake_argsr  rX   r  rY   constrain_to_fake_tensors	  s   r  c                   sJ    fdd t  fddt|jD } fdd| D }||fS )Nc                   s^   t  tjrtjd  tjjj	}tj
 |S t  tr- fdd  D S  S )Nr  c                   r~  rX   rX   r  )r  r   fx_argrX   rY   r   	  r  zEconstrain_to_fx_strides.<locals>.apply_constraint.<locals>.<dictcomp>)r_   r*   r1   r6  rP  r  rH   rP   r  r  r7  r8  r   r  )r   r  r:  r  )r   r  rY   r  	  s   
z1constrain_to_fx_strides.<locals>.apply_constraintc                 3  r  r~   rX   )r   r   r  r  rX   rY   r   	  r  z*constrain_to_fx_strides.<locals>.<genexpr>c                   s"   i | ]\}}| |j | qS rX   r   r   r  fx_noderX   rY   r   	  s   " z+constrain_to_fx_strides.<locals>.<dictcomp>)r   r   rg   r   r  rg   r   rX   r  rY   rt   	  s   

rt   c                   sN   fdd t  fddtt|jD } fdd| D }||fS )Nc                   s  t |tjs|S |jd }dd | D }t| }|r1|d dkr1tttt	|
 }jtjjkrF| dv rFt	|dksDJ d}|jsPtj||S d	 jtjjjjko^| d
k}t |tsfJ t	|
 dvrp|S t| rttj||S t |tr| d urt| rttj||S |r4t|
 }g }| }	tt	|
 d D ]}
tjj||
 ds|	d urtjj|	|
 dr||
 qdgt	| }d|d< d}tt	|d ddD ]A}
||
d  dkr|||
d   }|
|v rtjj||
d    drd||
< qtjj|  ds(t|   }|||
< qtj ||S t| rEttj||S t |trc| d urct| rcttj||S  fdd}t |j!tj"r||s||# rttj||S tj||S )Nr  c                 S  rz  rX   r{  r  rX   rX   rY   r   	  s    z=sdpa_constraint.<locals>.apply_constraint.<locals>.<listcomp>r2  r   )r   r   r   )r   r'   r%   r   r   r   r   r   r'   r%   c                   s   t jj|  d   dkS rS  )rH   rP   r  r  r   r   	ALIGNMENTrX   rY   
is_aligned$
  s   z=sdpa_constraint.<locals>.apply_constraint.<locals>.is_aligned)$r_   r*   r1   rP  r  r6  r]   r  r   r   r   rT   r  0_scaled_dot_product_efficient_attention_backwardr  is_cudar7  r8  rp   rG   '_scaled_dot_product_efficient_attentionr9   is_aligned_realized_tensortry_match_insignificant_stridesrealize_inputmaybe_get_striderH   rP   r  statically_known_equalsrc   r<   r  ra   r  r  )r  r   r  meta_valr  r:  effn_attn_fwd_biasr  expanded_dimsmaybe_striderf   out_stridesr  r  )r  r  rY   r  	  s   







z)sdpa_constraint.<locals>.apply_constraintc                 3  s$    | ]\}\}} |||V  qd S r~   rX   )r   r  r   r  r  rX   rY   r   0
  s
    


z"sdpa_constraint.<locals>.<genexpr>c                   s$   i | ]\}}| d |j | qS r2  r   r   r  rX   rY   r   4
  s   $ z#sdpa_constraint.<locals>.<dictcomp>)r   r^   r   rg   r   r  rX   r  rY   sdpa_constraint	  s   {r  )r  c                 C  sn   |}|   |  krt||   }|  | kr t||  }|  | kr3t||  }t|S t|S r~   )rb   r  r   r   r   r  rt  )r_  r  r  r   rd   rX   rX   rY   ru    s   ru  )memory_formatc                C  s&   t j|  |  |  t|  dS NrJ  )r6   r   rb   r   r5  r]   r   )r   r  rX   rX   rY   rt    s   
rt  c                 C  s   g }t | tr+t | jtjr+| j} t | tjr'||   | j} t | tjst| } t| } |rI| j} |d d d D ]	}tj| |d} q;t| } | S )Nr2  r
  )r_   r9   ra   r*   r  rc   
get_layoutrt  )r   reinterpret_view_layoutsr  rX   rX   rY   clone_preserve_reinterpret_view  s   r  r  c                  s(    fdd}t jt| || gdS )Nc                   s   t j| d    dS )Nr   r   rG   r   r;  r   r  r  rX   rY   rl   A  rv  ziota.<locals>.fnrJ  )r6   r   r=   )lengthr  r  r   ri   requires_gradrl   rX   r  rY   iota7  s   
r  r   r$  c                   s   |   |  ks
J |  t|  d tjjtdr'| 	    tjj
d tjj| 	    tt| | 	 }|  fdd}tj|  |   |t| 	 dS )Nr   c              	     s6   t t t |   tjt tj| | S r~   )rG   r  eqr   rp   r  rm  r   r$  
src_loaderr  rX   rY   rI  X  s   z select_scatter.<locals>.inner_fnrJ  )r   r5  r   rH   rP   r  r  r   r  r   r  r  r  r  r6   r   rb   r]   )r   r  r   r$  rI  rX   r  rY   select_scatterL  s    

r  c                   s     |  ks
J  t d    tj \t }t d  | < t	||}|  fdd}t
j   |t dS )Nr   r'   c              
     s2  dkrkrdkr| S t |  tj}t|  t|    < g }dkr?|t |t t	tj krT|t 
|t t	tj dkrs|t t t|   dtjt dtj |swJ tt j|}t | fddtrdnd}t ||| S )Nr   r'   c                         S r~   rX   rX   )src_idxr  rX   rY   r        z1slice_scatter.<locals>.inner_fn.<locals>.<lambda>r  )rG   r   rp   r!  r]   r!   rc   r#  r   r  r$  r  r$   r"  r   r  r%  r&  r   r  )r  r'  r,  src_valr   r  r  r  r  r  r   r  )r  rY   rI  x  sR   zslice_scatter.<locals>.inner_fnrJ  )r   r5  r   r   r*   r  normalize_start_endr]   r!   r  r6   r   rb   )r   r  r   r  r  r  src_sizerI  rX   r  rY   slice_scatterj  s    
.
r  c                 C  s*   t | ttfrt| dkrt| d S | S r  )r_   r]   r   r   _unwrapr   rX   rX   rY   r    s   r  r   ri   r  
pin_memoryc                  s  t |d tjfv d|  t | d tt tr ptjnp%t g }t tj	r6 fdd}nBt t
tfrE fdd}n3t dksZt d t
tfrlt dkrl|tt   fdd}ntjtj |d	S tjt|||d
S )Nlayout=r  c                      t  S r~   r  r;  ra   r   rX   rY   rI       ztensor.<locals>.inner_fnc                   r  r~   rG   r"  r;  r  rX   rY   rI    r  r   r   c                   s8    fdd t dkrtdS  dt S )Nc              	     sr   | |k sJ ||  dkrt |  S ||  d |  }t t t d tjt |tj | | ||S )Nr'   r%   r   )rG   r"  r  r$  r   rp   r!  )r  r  mid)binary_searchra   r   r$  rX   rY   r    s   z/tensor.<locals>.inner_fn.<locals>.binary_searchr   )r   rG   r"  r;  r  )r  r$  rY   rI    s   r  rJ  )r   rp   stridedr_   r  r   r!  get_default_dtyper   r   r!  r   rc   IntegerrH   rP   add_tensor_constantr
  r6   r   r=   )ra   r   ri   r  r  rH  rI  rX   r  rY   r
    s,   *r
  c                 C  s@   t | tr|d urt| |} |d urt| |} | S t| ||dS )Nr  )r_   r9   r   r  r
  )ra   r   ri   rX   rX   rY   	as_tensor  s   


r  c                 C  s   t | tjdS r;  r	  ra   rX   rX   rY   long_tensor  r  r  c                 C  s   ddl m} |tjjjtjjjd }|d usJ t|dks#J |t	t
| \}}t||| }tj||_tj| tjjjd }t|tjtjtjfrW|jjS t|S )Nr   )resolve_unbacked_bindingsunbacked_bindingsr'   r  )r  r  rH   rP   r  r  rQ   rP  r   r'  iterr   r*   DynamicScalarregister_bufferr   register_operationr_   rp   r|  SymFloatSymBoolrV   r}  r   r  )ra   r  r  binding_symkeypathbufferr  rX   rX   rY   _local_scalar_dense  s   
r  c                 C     d S r~   rX   )ra   r   rX   rX   rY   _assert_scalar,  s   r  )ri   r  c                C  r  r~   rX   )r   r   r  r   ri   r  rX   rX   rY   _assert_tensor_metadata5  s   r  c                   s   | t | ttfstdrjt ttfr  fdd}n"t tjr. fdd}nt dks8J 	 fdd}t
j| |t|dS )Nr   c                      t  S r~   r  r;  r   r   rX   rY   rI  C  r  z_full.<locals>.inner_fnc                   r  r~   r  r;  r  rX   rY   rI  H  r  r   c                   s    g S r~   rX   r;  )value_loaderrX   rY   rI  O  s   rJ  )r_   r   r!  r   r   r   r   r   r   r5  r6   r   r]   )
fill_valueri   r   r   rI  rX   )r   r   r  rY   _full<  s    r  c                 K  s   t t|| fi |S r~   create_tensor_liketensor_constructor)r   r  r   rX   rX   rY   r  Z  s   r  c                   s    d d d d dd d fdd
}|S )NF)namesr   ri   r  r  r  c                   s   t | d u d t |d tjfv d|  t | d t|}|p#t }t|dkr;t|d tttj	fr;t|d }|D ]
}t|tj
rGJ q=dd |D }t |||S )Nnamed tensorsr  r  r'   r   c                 S  r  rX   r  r  rX   rX   rY   r   u  r	  z5tensor_constructor.<locals>.inner.<locals>.<listcomp>)r   rp   r  r=   r  r   r_   r]   r   Sizer|  r  )r  r   ri   r  r  r  r   r  r  rX   rY   rW  a  s   	"z!tensor_constructor.<locals>.innerrX   )r  rW  rX   r  rY   r  _  s   r  )r  r   r  ri   r  r  c                 G  sX   t | d u d t|}t|dkr"t|d tttjfr"t|d }t|d ||||dS )Nr  r'   r   r   r  ri   r  )	r   r=   r   r_   r]   r   rp   r  empty_strided)r  r   r  ri   r  r  r   rX   rX   rY   r  {  s   
"r  c                   s   dddddd fdd
}|S )zZ
    Shim to convert X_like(...) into X(...).  For example zeros_like() into zeros().
    NF)r   ri   r  r  r  c                  sj   t | d t |d tjfv d|  |d u r|  }nt|}|p%|  }t|  } |||||dS )Nr  r  r  )r   rp   r  r   r   rb   r]   r   )r   r   ri   r  r  r  r   creation_fnrX   rY   _constant_like  s   

z*create_tensor_like.<locals>._constant_likerX   )r  r  rX   r  rY   r    s   
r  c                 C  s   t t| S r~   r  r  rX   rX   rY   r    r  r  c                   s   d d d d d fdd
}|S )Nr  c                  st   t |ttfs	J t| d t|d tjfv d|  t|p#|  }|p)|  }dd |D }t	 t
|||S )Nr  r  c                 S  r  rX   )r   r  r  rX   rX   rY   r     r	  z7new_constant.<locals>._new_constant.<locals>.<listcomp>)r_   r]   r   r   rp   r  r   r   rb   r  r=   r   r   r   r  ri   r  r  rX   rY   _new_constant  s   z#new_constant.<locals>._new_constantrX   )r  r  rX   r  rY   new_constant  s   r  r  c                C  s8   |d u r|   }|d u r|  }t|d ||t||dS Nr  r   rb   r  r=   r  rX   rX   rY   	new_empty  s   r  c                C  s  t | ttfs	J t |tttd fsJ t| d t|d tjfv d|  t|p/t }|p7t	dj
}t|}td||| d}|  |jj}tj|jdgt|  d|_t |tjsbJ dd | D } |rrd	d |D ntj| }tj||| |d
|_|S )Nr  r  r  r   )r  ri   r   r   )rH  c                 S  r  rX   r  r  rX   rX   rY   r     r	  z!empty_strided.<locals>.<listcomp>c                 S  r  rX   r  r  rX   rX   rY   r     r	  )ri   r   r   r  )r_   r]   r   rL  r   rp   r  r   r  r
  ri   r=   r  r_  ra   dataclassesreplacer   r*   rs  r  r  r  r  )r   r  r   r  ri   r  	pointwiser  rX   rX   rY   r    s0   
r  c                C  s8   |d u r|   }|d u r|  }t||||t||dS r  r  )r   r   r  r   r  ri   r  rX   rX   rY   new_empty_strided  s   r  c                 C  s2   dd |D }t tt||jd}tj| |S )Nc                 S     g | ]	}t jj|qS rX   )rH   rP   r  r  r  rX   rX   rY   r     r   z copy_strided.<locals>.<listcomp>)r  )sortedr   r   __getitem__r*   r7  r8  )r   r  r:  rX   rX   rY   copy_strided  s   r  c                 K  s*   | dd usJ dt|| fi |S )Nr   z(dtype should be handled by decomposition)rQ  r  )r   r  r   rX   rX   rY   full  s   r  c                   s   t | tsJ | dkrt| | S | tjksJ |  tdk}t	|  | |r8t
| dg} dg|  |  fdd}tj|  |  || dS )Nr   r'   c                   sF   t | } t|   }t| dkr|g} | S ||  < | S r  )r]   rG   indirect_indexingr   )r  
gather_idxr   index_loaderr   r  rX   rY   rl   !  s   zgather.<locals>.fnrJ  )r_   r9   	get_numelr  r   r   rp   r!  r   r   r  r5  r6   r   rb   )r   r   r$  sparse_gradr  rl   rX   r  rY   gather  s&   	r   c                   s   |rJ t | tsJ t |tsJ dt| v sJ |  |  t| |  g | dd   fdd}tj| 	 |  |dS )Nr   r'   c                   s\   t | t ksJ |  d  | d  }t|d gg | d   }|S )Nz != r   )r   rG   r  )r  	var_index
weight_idxindices_loaderindices_ndimr  weight_loaderweight_sizerX   rY   rl   ?  s   "
zembedding.<locals>.fnrJ  )
r_   r9   rI  r   r5  r   r   r6   r   rb   )weightr  padding_idxscale_grad_by_freqsparserl   rX   r  rY   	embedding2  s    r  c                   s   t dd  D sJ ddd  D  tdd  D r"tddd t D }t|d	ks5J d
d gt  }t|t fdd|D  D ]\}}| |krXtd|||< qJ||fS )Nc                 s  s4    | ]}|d ur|  tjtjtjtjfv V  qd S r~   )r   rp   r!  r  r   rk  r   rX   rX   rY   r   P  s    z.check_and_broadcast_indices.<locals>.<genexpr>z)indices must be int64, byte or bool. Got c                 S  s   g | ]
}|d ur|  qS r~   r   r   rX   rX   rY   r   U      z/check_and_broadcast_indices.<locals>.<listcomp>c                 s  s,    | ]}|d ur|  tjtjfv V  qd S r~   )r   rp   r   rk  r   rX   rX   rY   r   W  s    "zFallback for bool indicesc                 S  r   rX   r   r   rX   rX   rY   r   \  r   r   z"requires at least 1 non-None indexc                   r  rX   rX   r   r  rX   rY   r   _  r   z.Fallback when indices is on a different device)r  r  r   r^   r   r   r   rb   )r  ri   
valid_idxsnew_indicesrf   r   rX   r  rY   check_and_broadcast_indicesO  s"   
$
r  c	              
     s   dt dd  D ]\}	}
|
|	 dkrdq
fddtD g 
tt d  d }r> nd |  |d    	f
dd}|fS )	NFr'   Tc                   s    g | ]\}}|d u r | qS r~   rX   )r   rf   r  r  rX   rY   r     r  z2index_output_size_and_inner_fn.<locals>.<listcomp>r   c           	   	     s  t | t ks
J t t ksJ t }g }d }r"dn|}d}td d D ]F}||kr8||7 }| d u rR|t | k sFJ || |  |d7 }q.| }|d us\J | }|tj|| |||  | d q.g || |d  }	d u r|S 	|S )Nr   r2  r'   r   wrap_neg)r   r   rc   rG   r  )	r  r  	new_indexfirst_tensor_indexstart_offsetnext_idxrf   loaderr   )
r   indexed_sizer  indices_loadersnon_consecutive_tensorsoutput_sizetensor_indicestensor_sizer  r  rX   rY   rl     s>   

z*index_output_size_and_inner_fn.<locals>.fn)r   r^   r   )r  r  r  r   r  r  r  r   r  previouscurrentr  rl   rX   )r   r  r  r  r  r  r  r   r  r  r  rY   index_output_size_and_inner_fni  s$    


"r#  c                 C  s,   t | ||\}}}tj|  |  ||dS r  )index_impl_helperr6   r   rb   r   )r   r  r   r  rI  r   rX   rX   rY   
index_impl  s   r%  c           
        s   t ttfs	J |  t|  \}t|dks J ddd D }t|d   }|  fddttD }|rQd|v rQd|vrQt	dfddttD }t
||||d ||d	\}  fd	d
}	||	 fS )Nr   z Must have at least one valid idxc                 S      g | ]}|d ur|  nd qS r~   r4  r   rX   rX   rY   r     r  z%index_impl_helper.<locals>.<listcomp>c                   s    g | ]} | d ur| qS r~   rX   r   )r  r  rX   rY   r     r  z0index is out of bounds for dimension with size 0c                   r  rX   rX   r   r  rX   rY   r     r   r  c                   s    | S r~   rX   rm  )index_inner_fnr  rX   rY   rI    r  z#index_impl_helper.<locals>.inner_fn)r_   r]   r   r5  r  rb   r   r   r   
IndexErrorr#  )
r   r  r   r  r  r  r   r  r  rI  rX   )r'  r  r  r  rY   r$    s0   

r$  c                 C  sB   zt | |ddW S  ty    |   ttjjdd| | Y S w )NTr   Fr  )r%  r   r_  r  r  r$  r!  r   r  rX   rX   rY   r$    s   c                 C  s   t | |ddS )NFr)  )r%  r*  rX   rX   rY   _unsafe_index  r  r+  c                 C  s   t t| |||dddS )NTFr   may_realizeindex_put_impl_rt  r   r  r   
accumulaterX   rX   rY   	index_put     r2  c                 C  s   t t| |||dddS )NFr,  r.  r0  rX   rX   rY   _unsafe_index_put  r3  r4  c                 C  sB   |  |   krt||   }|rt| |}t| t|d || S r  )rb   r  r   r  r  )r_  r  r   r1  rX   rX   rY   index_put_as_masked_fill	  s
   
r5  c                 C  sl   t  }t|r(|s|r(|sdnd}tjjjdd  }r$| d| }|tj_t	
tjjj| ||| | S )Nzindex put with accumulate.zdeterministic index put.stack_trace Found from : 
 )rp   $are_deterministic_algorithms_enabledr2   rH   rP   rQ   rP  rQ  disable_cudagraphs_reasonr*   IndexPutFallbackrT   )r_  r  r   r1  deterministicr   r6  rX   rX   rY   index_put_fallback  s   r<  c                 C  s   t | |||dddS )NTr,  r/  r_  r  r   r1  rX   rX   rY   
index_put_!     r?  c                 C  s   t | |||dddS )NFTr,  r=  r>  rX   rX   rY   _unsafe_index_put_(  r@  rA  c              
     s  |r!dd }dd  || |  v r!t fdd|D s!|  | dkrZt|dkrZ|d  tjtjfv rZ|d }t	t|
 t| 
 D ]}t|d	}qJt| |g||S t ret| |||S |D ]}	|	d ur|	 tjtjfv rt| |||  S qg| 
 t}
|rt|  r|
dkrt| dg} t| |||} |
dkrt| g } | S t||  }zt||  \}}W n ty   t| ||| Y S w d
d |D }t| tsJ |   |
dkrt| dg} t||d  
 }fddt	t|D }t|||||d |d\}}t||}tj|  |  | |||r(dnd d}tjd t| |d}tj !||_"tj #| |
dkrNt| g } | S )Nc                 S  sP   t | tjr	| j} t | tjr|  } t | tjr| j} t | tjr&|  S d S r~   )	r_   r*   r9   ra   r  r  rq  BufferrB  r   rX   rX   rY   try_get_name2  s   z%index_put_impl_.<locals>.try_get_namec                 S  sd   t | tr0t | jtjr0| j } t | tjo/t | jtjo/t| jdd o/| jj	j
tjjjjkS dS )Nr  F)r_   r9   ra   r*   r  r  rq  r7  r   r  rT   rp   rG   r  randpermr  )indicerX   rX   rY   indice_slice_from_randperm;  s   
z3index_put_impl_.<locals>.indice_slice_from_randpermc                 3  r|  r~   rX   )r   rE  )rF  rX   rY   r   I  r  z"index_put_impl_.<locals>.<genexpr>r'   r   r2  c                 S  r&  r~   r4  r   rX   rX   rY   r   }  r  z#index_put_impl_.<locals>.<listcomp>c                   r  rX   rX   r   r  rX   rY   r     r   r)  
atomic_addri   r   rI  rH  output_indexerscatter_moder   r  ra   )$r{  r  r_  r  r   r   rp   r   rk  r   r   r  r5  r8  r<  rB   r  r   r  rb   r   r_   r9   r]   r#  r  r*   Scatterr5  rs  MutationLayoutSHOULDREMOVErH   rP   r  r   r  )r_  r  r   r1  r   r-  rC  r,  r   r$  x_ndimr  r  r   r  expected_vals_sizerI  scatterr  rX   )rF  r  rY   r/  /  s   	




r/  r  c                   sT   t | |ddd\}}  |   fdd}tj|  |  ||dS )NFr  c                   sB   j tjkrt tj}n }t| fddS )Nc                     s    S r~   rX   rX   )_unsafe_index_fnr  self_loaderrX   rY   r    r  z8_unsafe_masked_index.<locals>.inner_fn.<locals>.<lambda>)r   rp   r   rG   r   r&  )r  mask_valrQ  fillr,  mask_loaderrR  rm  rY   rI    s   z&_unsafe_masked_index.<locals>.inner_fnrJ  )r$  r5  r6   r   rb   r   )r_  r,  r  rU  rH  r   rI  rX   rT  rY   _unsafe_masked_index  s   rW  c                   s@   t ||d}|   fddtt D }t| ||ddS )Nr   c                   s6   g | ]} | rt  | |  | d  ndqS r  r  r   r  r  rX   rY   r     s    (z7_unsafe_masked_index_put_accumulate.<locals>.<listcomp>T)r1  )r  r   r   r   r4  )r   r,  r  r   masked_valueclamped_indicesrX   rX  rY   #_unsafe_masked_index_put_accumulate  s   
r[  c                 C  s   t |t || S r~   )rG   r?  r@  r   minmaxrX   rX   rY   r       r  c                 C  r  r~   )rt  r  copy_)r_  r  r   r  r  r  output_viewrX   rX   rY   as_strided_scatter  r  rb  c                 K  s   t t| |||fi |S r~   )scatter_rt  )r   r   r$  r  r   rX   rX   rY   rP       rP  r  include_selfr  r  r  rf  c             	   C  sf   t |t}t| || ttj|r| nt||r| jnd|r1t	j
| ||||||d |S d S )Nznot implre  )r_   r9   rF   r   r   rp   r   rL  rb   r*   ScatterFallback)r  r_  r   r$  r  r  rf  src_is_tensorrX   rX   rY   scatter_fallback  s(   

	ri  r  c                C  sr   |dv sJ |d u r$t tjtjjjj}t|| ||||d}|d ur$|S |dkr+d}n|dkr1d}t	| ||||S )N)Nr   multiplyrj  r   sumrk  prod)
r   r  rc  rH   rP   rQ   rT   _overloadnameri  scatter_reduce_)r_  r   r$  r  r  r  fallback_resultrX   rX   rY   rc    s   rc  c                 C  s   t t| |||S r~   )scatter_add_rt  r   r   r$  r  rX   rX   rY   scatter_add  r  rs  c                 C  s   t | |||dS )Nrl  )ro  rr  rX   rX   rY   rq  $  r  rq  c                 K  s   t t| ||||fi |S r~   )ro  rt  )r   r   r$  r  reduction_typer   rX   rX   rY   scatter_reduce)  s   ru  )rf  c             	     s,  |dv sJ t tj dkrdtj v sJ dttr$tttjj |||d}|r5|S tt	s<J dt
| v sFJ t  }|dkrVtdgtt	rit  dkritdgt|t	r|t | dkr|t|dg}| dkrS t    | tt	r nd  fdd	}fd
d}	dd }
|stj  fdd| |d d}tjd t|d}tj||_tj| tj  |	| ||
|d}tjd t|d}tj||_tj| |dkrtg S )N)Nrl  rm  meanamaxaminr'   twozKaten.scatter_reduce_.two is not the unique overload of aten.scatter_reduce_re  r   r   c                   sD     }t|}t| }tj| |dkrdn|  dd| < |S )Nr   r'   F)r  )r   r   r]   rG   r  )r  r  r  indirect_idx)r   r  r_  rX   rY   rI  \  s   
z'scatter_reduce_.<locals>.output_indexerc                   s   r| S t   S r~   rG   r"  r   rm  )r_  r  r  rX   rY   rl   f  s   zscatter_reduce_.<locals>.fnc                 S  s   | dkrdS | d u sJ d S )Nrl  rG  rX   rj  rX   rX   rY   backend_reduce_strm  s   z+scatter_reduce_.<locals>.backend_reduce_strc                   s   t d  S r  r{  r;  )r_  rX   rY   r  z  r  z!scatter_reduce_.<locals>.<lambda>rH  rK  )r   r  ro  r   r_   r   r  ri  ry  r9   rI  r   r   r  r  r   r_  r5  r*   rL  rb   rs  rM  rH   rP   r  r   r  )r_  r   r$  r  r  rf  rp  r  rI  rl   r|  zero_outr  rP  rX   )r   r  r_  r  r  rY   ro  .  s   







ro  scales_xtuple[Optional[float], ...]nexactc           
        s   |    |  |   d  |  d   }dd D t|ks)J |}dd t|D t|D ]\}}|d urGd| |< q9 fddfdd}	tj|  | 	 |	g ||d	S )
Nc                 S  r  rX   r  r   rX   rX   rY   r     r   z&upsample_nearestnd.<locals>.<listcomp>c                 S  s   g | ]\}}|| qS rX   rX   )r   rf   orX   rX   rY   r     r  rV  c                   s\   t | tj}  rt | t dtj} t | t |tj} t | tj} t j	| |ddS )N      ?Fr)  )
rG   r   rp   r=  r   r"  rK  r   r  r  )r   rB  r   )r  rX   rY   scale_fn  s   z$upsample_nearestnd.<locals>.scale_fnc                   sB   |  d  }| d   }g |fddt | D S )Nc                   s   g | ]\}}} |||qS rX   rX   )r   rf   r  r   )r  rX   rY   r     r   z2upsample_nearestnd.<locals>.fn.<locals>.<listcomp>)r   )r  r   r  )i_sizes
inv_scalesr  r  r  rX   rY   rl     s
    zupsample_nearestnd.<locals>.fnrJ  )
realize_hintr5  r   r   r   r^   r6   r   rb   r   )
r   r  r~  r  r  batcho_sizesrf   rB  rl   rX   )r  r  r  r  r  r  rY   upsample_nearestnd  s(   
r  Optional[float]c                 C  s   t | ||fddS )Nr'   r  r  r   r  r2  rX   rX   rY   upsample_nearest1d  r  r  c                 C  s   t | ||fdddS )Nr'   Tr  r  r  r  rX   rX   rY   _upsample_nearest_exact1d  r_  r  scales_hscales_wc                 C  s   t | |||fddS )Nr%   r  r  r   r  r  r  rX   rX   rY   upsample_nearest2d  s   r  c                 C  s   t | |||fdddS )Nr%   Tr  r  r  rX   rX   rY   _upsample_nearest_exact2d  s   r  scales_dc                 C  s   t | ||||fddS )Nr   r  r  r   r  r  r  r  rX   rX   rY   upsample_nearest3d  s   r  c                 C  s   t | ||||fdddS )Nr   Tr  r  r  rX   rX   rY   _upsample_nearest_exact3d  s   r  c                   s   t  fdd|D S )Nc                 3  s    | ]	}t | V  qd S r~   r  r   r   rX   rY   r     r  z$_create_constants.<locals>.<genexpr>)r   )r   rg   rX   r   rY   r<    s   r<  c                   s:   |   |   fdd}tj|  |  |dS )Nc                   sF   t | } t| tksJ  D ]}| d | |  | |< q| S r  )r]   r   )r  r   r  r  r  rX   rY   r    s
   zrev.<locals>.loaderrJ  )r5  r   r6   r   rb   r   )r   r  r  rX   r  rY   rev  s   r  paddingSequence[int]r  c                 C  s  dd }| r	dS t |dkst |  dkrdS |   t| tjrBt| jtjrBt| jjtjs=t	j
rBt| jjtjrB| jjjsDdS |   t| \}}|j}|d dkrZdS |d dksl|d dksl|d dkrndS |d }|dkrxdS |d }|jd }	||	| k rdS | jjj}
|jd |jd | g}|tjj|
< t| ||j|j}t|d|	|	| d	}t|| td
 d  d7  < |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                  S  sL   t jj} | du r
dS t| j}t|dkr$|d jtjj	tj
j	fv r$dS dS )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.
        NTr'   r   F)rH   rP   rQ   r   rR   r   rT   r  mmr  addmm)rQ   rR   rX   rX   rY   _padding_can_be_fused#  s   
z6inplace_constant_pad_nd.<locals>._padding_can_be_fusedNr   r%   r'   r   r   )r   r  r  inductorinplace_padding)r   r   r_  r_   r*   r9   ra   rq  rs  r(   can_inplace_pad_graph_inputInputBufferr   freeze_layoutr  r  r   rH   rP   buffer_to_padded_sizer  r  r  fill_r   )r   r  r  r  r   r  rk  npadstride0rowsizebufnamepadded_size	resized_xsliced_xrX   rX   rY   inplace_constant_pad_nd  s\   

$


r  c              	     st  t |d dks
J tdd |D rt| S tjr$t| |}|r$|S |  }tttt	|d d d |dd d  t |t   g  D ]\}}
tjj||f qIt|d  }g t	 |d  D ]\\}}	}

|
 |
t|
| |	  qmt |t |ksJ t|   fddfdd	}|  tj|  |  ||d
S )Nr%   r   c                 s  r  r   NrX   r   r  rX   rX   rY   r   z  r   z"constant_pad_nd.<locals>.<genexpr>r'   c                   s~   g }t  d  D ]\}\}}}|dkr|t|d |dkr+|t|| qttj|}t| fddS )Nr   c                     r  r~   rX   rX   )r$  r  rX   rY   r    r  z/constant_pad_nd.<locals>.mask.<locals>.<lambda>)	r   rc   range_mask_lowrange_mask_highr   r  rG   r%  r&  )r$  r,  r  rL  rM  r  )boundsr  
mask_sizesr  r  r;  rY   r,    s   "zconstant_pad_nd.<locals>.maskc                   sZ   t | d  }t| d   D ]\}\}}|||  qt|t| ks)J |S r~   )r]   r   rc   r   )r$  r  r  rL  _high)bounds_precompr,  r  rX   rY   	offset_fn  s
   z"constant_pad_nd.<locals>.offset_fnrJ  )r   r  rt  r(   r  r  r   r]   r  r   rc   rH   rP   r  lookup_precomputed_sizer   r  r   r   r5  r6   r   rb   )r   r  r  rd   r  lhr  rL  rM  r   r  rX   )r  r  r  r,  r  r  r  rY   r  w  s:   *

r  rf   
sympy.ExprUnion[sympy.Expr, int]c                 C  s&   t t | tjt t|tjS r~   )rG   r#  r   rp   r!  r   r  )rf   rL  rX   rX   rY   r    s   r  c                 C  s    t t | tjt |tjS r~   )rG   r$  r   rp   r!  )rf   rM  rX   rX   rY   r    s   r  c                 C  s   t t| |t| |S r~   )rG   r%  r  r  )rf   rM  rL  rX   rX   rY   
range_mask  s   r  rV  c                   sF       d   pdg   fdd}|S )Nr   c                   s|   | d   |  d   t tj fddtD }r1t| fddS t| 	fddS )Nc                   s.   g | ]}t |  | |  |  qS rX   )r  r   )r  ih	padding_hrX   rY   r     s   . z=constant_boundary_condition.<locals>.load.<locals>.<listcomp>c                     s   t  dg S )Nr   )constant_boundary_conditionrX   )r   r  pad_fill_valueprefixr   rX   rY   r    s    
z;constant_boundary_condition.<locals>.load.<locals>.<lambda>c                     s   g  S r~   rX   rX   )r  r  r  rX   rY   r    r  )r   r  rG   r%  r   r&  )r$  r,  r   r  r  r  r  r  r   r  )r  r  rY   r:    s   	z)constant_boundary_condition.<locals>.load)r   r5  )r   r  r  r  r   r:  rX   r  rY   r    s
   r  c                 C  s   t | d||   || d  || d  || }|r|t | d||   || d  d|| d   || }tjj|d ||  |  ||  dkra|d8 }tjjd|||  |  ||   tjj|| dkrztjj|| d}||fS |}||fS )Nr%   r'   r   F)r!   rH   rP   r  r  r  r  )r   rf   kernel_sizer  r  	ceil_modex_outx_altrX   rX   rY   pooling_size  s    ,0*$r  c                 C  s4   t | d} | d | d  }|dkptdd |D S )Nr%   r   r'      c                 s      | ]}|d kV  qdS r  rX   r  rX   rX   rY   r     r   z:should_fallback_max_pool2d_with_indices.<locals>.<genexpr>)rC   r  )r  dilationwindow_sizerX   rX   rY   'should_fallback_max_pool2d_with_indices  s   
r  assert_fallbackc                C  s   |dkrddg}|dkrddg}|s|}t |d}t |d}t |d}t |d}t| ts/J t|dks7J t|dks?J t|dksGJ t|dksOJ t|  dv sYJ t||}|d urh||kshJ |||||fS )Nr   r'   r%   r  )rC   r_   r9   r   r   r  )r   r  r  r  r  r  use_fallbackrX   rX   rY   max_pool2d_checks  s(   




r  c              
     s:  |    |  ^ }}}t|d||\}	}
t|d||\}}| j}|tju r,dn|jr3tdnt|j	}t
||	|g }d sMd sM|
sM|rUt| |ddn|  d  fdd}tjd	| |  |||||d
}tjd| |  tj||||d
}t|jjtr|  t|jjtr|  ||fS )Nr   r'   F-infr%   r  c                   sH   | d   }|  d    fddt D }g ||S )Nc                   s,   g | ]} | |  |  |  qS rX   rX   r   )bhr  reduction_idxr  rX   rY   r   ;     , z>_max_pool2d_with_offsets.<locals>.fn_inner.<locals>.<listcomp>r   )r  r  r  r  r   r  r  r  )r  r  rY   fn_inner8  s   z*_max_pool2d_with_offsets.<locals>.fn_innerr^  )rt  
input_noderi   	dst_dtypero  rI  rH  reduction_rangesargmax)r  r   r  r   rp   r   r  r!  r  r]  r]   r  r5  r7   r   rb   r!  r_   ra   r_  )r   r  r  r  r  r  r  r  wh_out
ceil_mode1w_out
ceil_mode2r   	min_valuer  r  r  offsetsrX   r  rY   _max_pool2d_with_offsets  sP   


r  c           	      C  st   t | ||||dd\}}}}}tjdd t| |||||\}}|t|tjfW  d    S 1 s3w   Y  d S )NFr  r  )unroll_reductions_threshold)r  r(   r   r  r   rp   rj  )	r   r  r  r  r  r  r   r  r  rX   rX   rY   #_low_memory_max_pool2d_with_offsets\  s&   
	$r  c                   sH   |   fdd  fdd}tj|  tj||  d}|S )Nc           	        sf   t  tj}t |d  d  tj}t |d  d  tj}||  }|| }|| | S r  )rG   r   rp   r!  )	h_incw_incr  bww_inhbasewbaser  iw)input_widthr  r  rX   rY   increments_to_index  s   zF_low_memory_max_pool2d_offsets_to_indices.<locals>.increments_to_indexc                   sP   | ^ }}}g |||}t tj}|| }|||  } ||||S r~   rG   r"  rp   r  )r  r  r  r  r  kw_constr  r  )r  kernel_widthoffsets_loaderrX   rY   offsets_to_indices  s   zE_low_memory_max_pool2d_offsets_to_indices.<locals>.offsets_to_indicesrJ  )r5  r6   r   rb   rp   r!  r   )r  r  r  r  r  r  r  rX   )r  r  r  r  r  r  rY   )_low_memory_max_pool2d_offsets_to_indices{  s   r  c           
      C  sx   t | ||||\}}}}}tdd |D r t| |||||dS t| |||||\}}t||d | jd ||}	||	fS )Nc                 s  r  r  rX   r  rX   rX   rY   r     r   z*max_pool2d_with_indices.<locals>.<genexpr>)r  r2  )r  r   fallback_max_pool2d_with_indicesr  r  r  )
r   r  r  r  r  r  r   rd   r  r  rX   rX   rY   max_pool2d_with_indices  s   	
r  c                   sN  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s3J t|dks;J t| dv sEJ |   |  }t |trzt |jjtrz|jj}	tj	d tj
|	 |	 |	 d|	d}
|
  |
 }n| }|d ur|d dkp|d uo|d dk}tdd |D rt| ||||S | ^ }}
|  ^ }| |   t| }tfd	dtd d D tfd
dtd d D 		 }|dkrt| ||||S |  	
fdd}tj|  |  ||d}|r%tj|S |S )Nr   r'   r%   r  )ri   r   r   rK  c                 s  s    | ]}|d kV  qdS r  rX   r  rX   rX   rY   r     r   z3max_pool2d_with_indices_backward.<locals>.<genexpr>c                 3  <    | ]}t |d   t d | d   d    dV  qdS r   r'   Nr^  r   r  r  r  rX   rY   r     
    *
c                 3  <    | ]}t |d   t d| d   d    d V  qdS r'   r   Nr  r   r  r  rX   rY   r     r   r  c                   sV  | ^ }}}t |
 | tj}|d  }|d  }t t|d  d  d tj}t t|d  d  d tj}t t|d d tj}t t|d d tj}t |t dtj}t |t dtj}t |t tj}t |t tj}d }	tD ]}
t	D ]}t 	|t |
tj}t 	|t |tj}g |t j
t |t |t dtjd ddt j
t |t |t dtjd dd}|} |}t ||}|	d u rt ||t dtj}	qt t t ||t |||}t |t 	|	||	}	qq|	d us)J |	S )Nr   r'   Fr)  r2  r  )rG   r   rp   r  r!   r?  r"  r@  r   r   r  rR  r  r  r=  r%  r$  )r  r  r  r  
index_testphstartpwstartphendpwendgradientph_pw_phpw
grad_indexindex_actual	grad_partr   r,  grad_loaderh_window_sizer  indices_sizer  r  pooled_heightpooled_widthr  w_window_sizewidthrX   rY   rl     sl     


#z,max_pool2d_with_indices_backward.<locals>.fnrJ  )r_   r9   r   r   r  r  ra   r6   r*   rs  r  rb   r   decide_layoutrT  r  )fallback_max_pool2d_with_indices_backwardr5  r]   r^  r   r   r7  r  )grad_outputr   r  r  r  r  r  r  	gO_stridera   x_bufferx_strideis_channels_last_batch_heightr   r  r  rl   rd   rX   r  rY    max_pool2d_with_indices_backward  s|   	
 ;r#  r  c                   s   |    fdd}|S )Nc              
     s   |\|\ |\}}t t t   tjt |tjt t  tjt |tj}t | fddS )Nc                     s   g    S r~   rX   rX   )h_start_indexr  r  r  w_start_indexr  rX   rY   r  m  r  z3pad_adaptive_loader.<locals>.load.<locals>.<lambda>)rG   r%  r$  r   rp   r!  r&  )r  
incrementsstart_indicesend_indicesh_end_indexw_end_indexr,  pad_valr  )r$  r  r  r  r%  rY   r:  [  s$   z!pad_adaptive_loader.<locals>.loadr4  )r   r,  r:  rX   r+  rY   pad_adaptive_loaderX  s   r-  c           
      C  sL   t j| ||d}t j|||d}t j| ||d}t j|||d}	||||	fS )N)out_diminp_dim)r   r  )
start_index	end_indexh_inr  r  r  r$  r)  r%  r*  rX   rX   rY    compute_indices_adaptive_poolingt  s
   r3  c                   sD   |\}}|\}}	t | |||||	\  fdd}
|
S )Nc                   s   | ^ }}}|} |}|}|}d }	t td td D ]\}
}|||
|g||g||g}|	d u r>|}	q&||	}	q&|	S r  )r   productr   )r  r  r  r  r  r$  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_fnrX   rY   rl     s"   $z _adaptive_pooling_fn.<locals>.fnr3  )r0  r1  r8  in_sizes	out_sizesr9  r2  r  r  r  rl   rX   r5  rY   _adaptive_pooling_fn~  s   r?  c           
        sF   |\}|\}}t | ||||\  fdd}	|	S )Nc                   s   | ^ }}}|} |}|}|}d }	d }
t td td D ]>\}}||||g||g||g}t||  | | tj}|
d u rO|}
ntt||	||
}
|	d u ra|}	q(||	}	q(|
S r  )	r   r4  r   rG   r   rp   r!  r  gt)r  r  r  r  r  r$  r)  r%  r*  maxvalmaxindexr  r  r  r$  r6  r7  r8  r9  r:  r  r;  rX   rY   rl     s0   $z)_adaptive_pooling_fn_with_idx.<locals>.fnr<  )
r0  r1  r8  r=  r>  r9  r2  r  r  rl   rX   rC  rY   _adaptive_pooling_fn_with_idx  s   #rD  c                   s    tjkrtdttsJ t|dksJ    ^ }}}t	j
j|}t	j
j|}|\}}||krD||krDtS |dksL|dkr_g |||}t|   dS || dkrx|| dkrx|| || g}t|S t|| d |}	t|| d |}
t|||g }  }|	|
 }|dkrt|S dd }d	d
 }t|||	|
g||g||gtjd tt fdd}tj |||d}|S )Nz0'adaptive_avg_pool2d' not implemented for 'Long'r%   r   r  r'   r  c                 S     t | | |S r~   r!   r$  r.  r/  rX   rX   rY   r0       z)_adaptive_avg_pool2d.<locals>.start_indexc                 S     t | d | | d |S r  rF  rG  rX   rX   rY   r1  
  rv  z'_adaptive_avg_pool2d.<locals>.end_indexr0  r1  r8  r=  r>  r9  c                   s   t  | t | S r~   )rG   truedivr-  rm  fn_sumones_loaderr   rX   rY   rl     s   z _adaptive_avg_pool2d.<locals>.fnrJ  )r   rp   r!  r  r_   r9   r   r  r   rH   rP   r  r  rt  r  rb   
avg_pool2dr<   r]   fallback_adaptive_avg_pool2dr?  rG   r   r-  	ones_liker6   r   )r   r  r  r2  r  r  r  o_sizer  h_kernel_maxw_kernel_maxr  r   r  r0  r1  rl   rvrX   rL  rY   _adaptive_avg_pool2d  sV   

	rV  c                   s    tjkrtdttsJ t|dksJ    ^ }}}t	j
j|}t	j
j|}|\}}|dks@|dkr]g |||}t|   dt|tj dfS || dkrk|| dkrktt|| d |}t|| d |}	t|||g }
  }||	 }|dkrt|S dd }d	d
 }t||||	g||g||gtjdt||||	g||g||gtjd fdd} fdd}tj |||
d}tj tj||
d}||fS )Nz,adaptive_max_pool2d not implemented for Longr%   r   r  r'   r  c                 S  rE  r~   rF  rG  rX   rX   rY   r0  Q  rH  z(adaptive_max_pool2d.<locals>.start_indexc                 S  rI  r  rF  rG  rX   rX   rY   r1  T  rv  z&adaptive_max_pool2d.<locals>.end_indexrJ  c                       | t tdS Nr  r-  r!  rm  )inner_func_max_valr   rX   rY   inner_fn_max_vali  r  z-adaptive_max_pool2d.<locals>.inner_fn_max_valc                   rW  rX  rY  rm  )inner_func_max_idxr   rX   rY   inner_fn_max_idxl  r  z-adaptive_max_pool2d.<locals>.inner_fn_max_idxrJ  )r   rp   r!  r  r_   r9   r   r  r   rH   rP   r  r  r  rb   
ValueErrorr<   r]   fallback_adaptive_max_pool2dr?  rG   r?  rD  r6   r   )r   r  r  r2  r  r  r  rR  rS  rT  r  r   r  r0  r1  r[  r]  rU  rirX   )r\  rZ  r   rY   adaptive_max_pool2d,  sp   
		ra  c                   sP      t  d    fdd}|S )Nr'   c                   s   g | d  }t | }t   }t || | t ||  }t |tj}t |t d tj}t ||t  tjS r  )	rG   r   r   r  r   rp   r!  r$  r  )r  rf   samplei_expr
alpha_exprseq_ir,  r.  r   in_sz	kernel_szndimsout_szsamplessamples_loaderrX   rY   r:    s   z)_fractional_pooling_offsets.<locals>.load)r#   r5  )rk  rg  rj  rh  r   ri  r:  rX   rf  rY   _fractional_pooling_offsets  s   rm  c                   s   |    |  ^ }\}}|\}}|| dkr!t| ||S tjt|g|dd}	|	dd |	dd|   fdd}
t|||g }tj	| 
 |  tj|
d	d
|d}tj	| 
 tjtj|
dd
|d}||fS )Nr  r%   )rk  rg  rj  rh  ri  r   r  r'   c              	     s   | ^ }}}t  ||}t ||}d }d }ttd td D ]I\}	}
g |||	 ||
 }|rft ||	  | |
 tj}|d u rT|}nt t 	t 
||t |||}|d u rm|}q*t ||}q*|rx|S |S r  )rG   r  r   r4  r   r   rp   r!  r  or_r@  r  r?  )r  return_indexr  r  r  r$  r%  rA  rB  r  r  r  r$  
h_index_fninp_hinp_wr  
w_index_fnr  rX   rY   rl     s,   $z!fractional_max_pool2d.<locals>.fnF)ro  rJ  T)r  r   fallback_fractional_max_pool2dr   r  rm  r5  r]   r6   r   rb   r   rp   r!  )r   r  r  random_samplesr  kernel_hkernel_wr  r  gen_offsets_for_dimrl   r  rU  r`  rX   rp  rY   fractional_max_pool2d  sD   
	
rz  c                   s       ^ }}}tjj|}tjj|}|^ }}}	|| dkr9||	 dkr9t|| ||	 gddS t||}
t||	}dd fdd}t||
|g||g||	gt	j
d  fd	d
}tj  |t|d}|S )Nr   r'   )divisor_overridec                 S  s   t | | t|S r~   )r    r   r  rG  rX   rX   rY   r0    r  z0upsample_nearest2d_backward.<locals>.start_indexc                   s    | d ||S r  rX   rG  )r0  rX   rY   r1    rr  z.upsample_nearest2d_backward.<locals>.end_indexrJ  c                   s    | t S r~   )r-  rm  )rM  r   rX   rY   rl     rH  z'upsample_nearest2d_backward.<locals>.fnrJ  )r  r   rH   rP   r  r  rO  r<   r?  rG   r   r6   r   rb   r   r]   )r   r  
input_sizer  r  r!  rr  rs  out_hout_wrS  rT  r1  rl   rU  rX   )rM  r0  r   rY   upsample_nearest2d_backward  s6   

	r  rX   c              
   C     t | ||||||ddS )Nr%   r  _avg_poolndr   r  r  r  r  count_include_padr{  rX   rX   rY   rO       
rO  c              
   C  r  )Nr   r  r  r  rX   rX   rY   
avg_pool3d/  r  r  c              	     s  
s
sdg t t 

t t| ts!J tks)J t
ks1J tks9J t|  d d fv sIJ |   |  d   }|   d  t 
fddtD  \}	}
tszt|
rt	| ddd}n| 
 d	}t|t|	 }|  ttj}|d
krdkrt}ndkrt}ntd || 
 |S 
fdd|r|r|r|n|jrd 		fdd}nfdd}n
f	dd}tj|  ||d}|S )Nr   r'   r%   c              	     s"   g | ]}t | | qS rX   )r  r   )r  r  r  r  r  rX   rY   r   b  s    z_avg_poolnd.<locals>.<listcomp>r  r  TFr  r   zUnknown dim: c                   s   | d   }|  d   d }t jfddtD  D ]% fddtD }|g ||}|d u r=|}qt||}q|S )Nc                   s   g | ]}t  | qS rX   r  r   )r  rX   rY   r     r  z/_avg_poolnd.<locals>.fn_sum.<locals>.<listcomp>c                   s,   g | ]} | |  |  |  qS rX   rX   r   )r  r  r  r  rX   rY   r     r  )r   r4  r   rG   r   )r  r  r  totalr   r  )r   r  r  r  )r  r  rY   rM    s    z_avg_poolnd.<locals>.fn_sumc                   s   t | t  S r~   )rG   rK  r"  rm  )r   rM  rB  r  rX   rY   rl     rv  z_avg_poolnd.<locals>.fnc                   s   t | t  S r~   )rG   truncdivr"  rm  )divisorr   rM  r  rX   rY   rl     rd  c                   s   |  d  }g }t D ]<}|| |  |  }t||  | |  } s;t|d}t|| }t|| tj}|| qt	
tj|}jr]t| |S t| |S r  )r   r   MinMaxrG   r   rp   r  rc   r   r  rK  r  rK  r  )r  r  divide_factorsrf   hstarthendfactordivide_factor)	r  r   r   rM  r  r  r  r  r  rX   rY   rl     s    rJ  )rC   r_   r9   r   r   r  r   r   r  r  r5  r]   r   r   r  rj  rK  fallback_avg_pool2dfallback_avg_pool3dr^  r  r6   r   rb   )r   r  r  r  r  r  r{  r   r  r  
ceil_modeshad_paddingr  r  fallbackrl   rU  rX   )r  r  r   r  r   rM  r  r  r  rB  r  r  rY   r  E  st   




 
r  c                   s  d u sdksJ dssddgt | tsJ t |ts$J tdks,J tdks4J tdks<J t| dv sFJ |   | ^ }td|\}	}
td|\}}|  d pwd pw|
pw||  ^ }	
t| }| }t	fddt
d d D t	fddt
d d D  }|d	krt| ||S fd
d  	
fdd}tj|  |||d}|S )Nr   divisor must be not zeror%   r  r'   c                 3  r  r  r  r  r  rX   rY   r     r   z&avg_pool2d_backward.<locals>.<genexpr>c                 3  r  r  r  r  r  rX   rY   r     r   r  c              	     sX  t d tj}t d tj}t d tj}t d tj}t d tj}t d tj}t t | ||}t t |||}	t t ||t t  tj|}
t t |	|t t tj|}t 	|t dtj}t 	|	t dtj}	t |
t  tj}
t |t tj}t t |
|t ||	}|S )z{
        This computes the scaling factor that we will divide an element
        by when `count_include_pad=False`
        r   r'   )
rG   r"  rp   r  rR  rK  r@  r   r   r?  )r  r  stride_hstride_wpad_hpad_wrw  rx  r  wstartr  wendr  )heightr  r  r  r  rX   rY   !compute_pool_size_without_padding  s,   

z>avg_pool2d_backward.<locals>.compute_pool_size_without_paddingc                   sR  | ^ }}}|d  }|d  }t t|d  
d  
d tj}t t|d  
d  
d tj}t t|
d d tj}t t|
d d tj}t |t dtj}t |t dtj}t |t tj}t |t 	tj}d }tD ]}	tD ]}
t 	|t |	tj}t 	|t |
tj}d ur}nssd d  }n ||}t 
g |t jt |t |t dtjddt jt |t |t dtj	dd|}t t ||t ||}|d u rt ||t dtj}qt |t 	|||}qq|d us'J |S )Nr   r'   Fr)  r  )rG   r   r!   rp   r  r?  r"  r@  r   r   rK  r  rR  r%  r$  r  r=  )r  r  r  r  r  r  r  r	  r
  r  r  r  r  rB  partr,  )r  r  r{  r  r  r  r  r  r  r  r  r  rX   rY   rl      sv     
	


*zavg_pool2d_backward.<locals>.fnrJ  )r_   r9   r   r   r  r  r5  r]   r   r^  r   fallback_avg_pool2d_backwardr6   r   rb   )r  r   r  r  r  r  r  r{  r   _h_outr  _w_outr  r  r   r  rl   rU  rX   )r  r  r{  r  r  r  r  r  r  r  r  r  r  r  rY   avg_pool2d_backward  sb   "Ar  c                   s  d u sdksJ ds	
sg d
t | tsJ t |ts$J t	dks,J tdks4J t
dks<J t| dv sFJ |   | ^ }td	
|\}	}
td	
|\}}td	
|\}}|  t
p|
p|p||  ^ }t| }|	 }	fdd	t
dD \  }|d
krt| |	
|S 	
fdd  	
fdd}tj|  |||d}|S )Nr   r  )r   r   r   r   )r   r   r'   r%   c                 3  s6    | ] t  fd dt  d D V  qdS )c                 3  s<    | ]}t |   t d |       dV  qdS r  r  r  )rf   r  r  rX   rY   r     r   z0avg_pool3d_backward.<locals>.<genexpr>.<genexpr>r%   N)r^  r   )r   r  )rf   rY   r     s    
z&avg_pool3d_backward.<locals>.<genexpr>}   c              	     s  dd D \}}}dd D \}}}dd D \}	}
}dd t | ||g|||g|||gD \}}}dd t |||g|	|
|g g|||gD \}}}dd |||fD \}}}dd t |||g gD \}}}ttt||t||t||}|S )	Nc                 s      | ]
}t |tjV  qd S r~   r  r  rX   rX   rY   r     r   zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>c                 s  r  r~   r  r  rX   rX   rY   r     r   c                 s  r  r~   r  r   rX   rX   rY   r     r  c                 s  s*    | ]\}}}t t |||V  qd S r~   )rG   rR  rK  )r   r  r  padrX   rX   rY   r     s
    
c              
   s  s>    | ]\}}}}t t ||t t |tj|V  qd S r~   )rG   r@  r   r   rp   r  )r   r  r   r   r  rX   rX   rY   r     s    

c                 s  &    | ]}t |t d tjV  qdS r  rG   r?  r"  rp   r  )r   r  rX   rX   rY   r     
    
c                 s  *    | ]\}}t |t |tjV  qd S r~   rG   r@  r   rp   r  )r   r  r   rX   rX   rY   r     
    
)r   rG   rK  rR  )pdr  r  stride_dr  r  pad_dr  r  kernel_drw  rx  dstartr  r  dendr  r  r  )depthr  r  r  r  r  rX   rY   r    s8   $z>avg_pool3d_backward.<locals>.compute_pool_size_without_paddingc                   sJ  | ^ }}}}dd t |||gD \}}}dd t |||gD \}}}dd t |||gD \}}	}
dd |||fD \}}}dd t ||	|
g	
gD \}}	}
d }tD ]}tD ]}tD ]}dd t |||g|||gD \}}}d ur}nssd d	  d
  }n |||}tg |tjt|t|td	tj		ddtjt|t|	td	tj	
ddtjt|t|
td	tj	dd|}t
t
t||t||	t||
}|d u rt||tdtj}qjt|t|||}qjqdq^|d us#J |S )Nc                 s  s    | ]	\}}|| V  qd S r~   rX   )r   r   r  rX   rX   rY   r     r  z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>c                 s  s2    | ]\}}}t t|| | |tjV  qd S r~   rG   r   r!   rp   r  )r   r   r   r  rX   rX   rY   r     s
    
c                 s  s,    | ]\}}t t||d  tjV  qdS r  r  )r   r   r  rX   rX   rY   r     s
    
c                 s  r  r  r  )r   pstartrX   rX   rY   r     r  c                 s  r  r~   r  )r   pend
pooled_dimrX   rX   rY   r     r  c                 s  r  r~   )rG   r   r"  rp   r  )r   r  p_rX   rX   rY   r     r  r   r'   r%   Fr)  r  )r   r   rG   rK  r  r@  rR  r"  rp   r  r%  r$  r  r=  r   )r  r  r  r  r  pdstartr  r  pdendr  r	  r
  pd_r  r  r  r  r  rB  r  r,  )r  r  d_window_sizer{  r  r  r  r  r  pooled_depthr  r  r  r  rX   rY   rl     s    	

8zavg_pool3d_backward.<locals>.fnrJ  )r_   r9   r   r   r  r  r5  r  r]   r   r   fallback_avg_pool3d_backwardr6   r   rb   )r  r   r  r  r  r  r  r{  r!  _d_outceil_mode_dr  ceil_mode_hr  ceil_mode_wr   r  r   r  rl   rU  rX   )r  r  r  r  r{  r  r  r  r  r  r  r  r  r  r  r  r  rY   avg_pool3d_backwardo  sf   &%Wr  c                 C  s   |   }t|tr|g}n|stt|}t|dkr*t|dv s(J d| g S t|}tt|D ]5}|| dk rL||  t|rHt|nd7  < d||   krZt|k sin t|dkrg|| dksiJ q4tt|t|ksxJ d|S )Nr   )rX   r`  r  zinvalid axis: r'   zreduction axis not unique)r   r_   r   r   r   r   r]   r   )r   r4  r   rf   rX   rX   rY   _validate_reduction_axis7  s    
 :r  c          
        s   |d ur	t | |} |  tt t| |}g }g g }g ttD ]}||v r7| ||  q$| ||  q$ fdd}r`t}	D ]}t	j
j|	|< qVn|}	|   t|  |po|  |  ||	|dS )Nc                   s   t |t ks
J rt  t ksJ  fddD  t  t ks)J d gt  t |  }tt t|D ]\}}|||< q@|S )Nc                   r  rX   rX   r   r;  rX   rY   r   _  r   z9_make_reduction_inner.<locals>.loader.<locals>.<listcomp>)r   r   r   r   )r$  reduction_indexr  r  varinner_loaderkeepdimskept_idxreduced_idxr   r;  rY   r  [  s   
z%_make_reduction_inner.<locals>.loader)ri   r  ro  rI  rH  r  )r   r   r   r   r  r   r   rc   r]   r   r  r  r5  r   rb   r   )
r   r4  r  r   r*  
kept_sizesreduced_sizesrf   r  r  rX   r  rY   _make_reduction_innerI  s<   



r  rt  rI   c                   s   dd d fdd}|S )NFr   c                  sB   t | ||| d}tjd| d|}t|jjtr|  |S )Nr4  r  r   r*  )rt  r  rX   )r  r7   r   r_   ra   r_  )r   r4  r  r   r   r  r*  rt  rX   rY   rW  {  s   zmake_reduction.<locals>.innerr   rX   )rt  r*  rW  rX   r  rY   make_reductionz  s   r  c                C  sB   |d ur	t | |} t| |}t|  |  f|  f|  |dS )N)ri   dtypes	inner_fnsr   r4  )r   r   r   rb   r   r5  r   )r   r4  r   rX   rX   rY   _make_scan_inner  s   

r  r   c                  s   |d ur	t | |} |   t| |}|  }|tjtjfv r$t | tj} t| ||}t	 fdd|D }t
j||  |  d}t|t| }t t|||S )Nc                 3  r   r~   rX   r   r   rX   rY   r     r   zmean.<locals>.<genexpr>r#  )r   r   r  r   rp   rO  rN  r!  sum_rE   r*   r0   rb   r/   r   r]   div)r   r4  keepdimr   output_dtype
sum_resultdenomrX   r  rY   rv    s   

rv  c           
        s   |d u rd}|    t| |}t| |dd}|r|  tt| |}t|||}t fdd|D }|r>t	|| d}t
j||  |  d}t|t|  }t||}	|s^|	fS |rb|nt||}|	|fS )Nr'   T)r  c                 3  r   r~   rX   r   r  rX   rY   r     r   z var_mean_sum_.<locals>.<genexpr>r   r#  )r   r  rv  r_  squarerR  r  rE   r   r  r*   r0   r   rb   r/   r   r]   r  r  )
r   r4  
correctionr  return_meanx_meandiffsr  r  x_varrX   r  rY   var_mean_sum_  s&   

r  c                 C  sV   t | |}t| ||d d d}|d }t|d }t|tjo*t|tjk o*t|dkS )Nr  rH  r  r'   )	r  r  rE   r_   r   r  r   r(   r  )r   r4  r  r   rH  reduction_numelrX   rX   rY   use_two_step_variance  s   


r  c                  s    d u rd t | ||d d d}|d}|d |d tjjd|fd|  d|\}}}	|  |  |  t| |}t	fdd	|D d
d  fdd}
t
|
|}|rj|  ||fS |fS )Nr'   r  rI  r  ro  welford_reduce)r  rt  r   c                 3  r   r~   rX   r   r  rX   rY   r     r   z$var_mean_welford_.<locals>.<genexpr>c                 S  s4   t | tjr| jstt| tj|S t	| |S r~   )
r_   r   r   	is_numberrG   r   r   rp   r!  r"  r  rX   rX   rY   get_constant_or_index_expr  s   z5var_mean_welford_.<locals>.get_constant_or_index_exprc                   s4    }}t d}| t |||  S r  )rG   r"  r?  )ra   cNzero)r  r   r  rnumelrX   rY   r    s   

z#var_mean_welford_.<locals>.scale_fnrX   )r  r0  r*   WelfordReductionr   r   r_  r   r  rE   rX  )r   r4  r  r  r  r   r  rv  m2r   r  r  rX   )r  r   r  r  r   rY   var_mean_welford_  s6   




r  c                  s   |    t }t| |dd} t| ||||d}t| ||dr&tdi |ntdi |}t fdd|D }|s>|d S |S )	NFr  )r   r4  r  r  r  )r4  r  c                 3  s    | ]
}t | d dV  qdS )Fr  Nrn  r   rM  rX   rY   r     r   z#var_mean_helper_.<locals>.<genexpr>r   rX   )r   r   r   r   r  r  r  r   )r   r4  r  r  r  compute_dtyper   r  rX   rM  rY   var_mean_helper_  s    	r  )r  r  c                C     t | |||ddS )NFr4  r  r  r  r  r   r4  r  r  rX   rX   rY   var_     
r  c                C  r  )NTr  r  r  rX   rX   rY   var_mean  r  r  c                 C  st   |dk rt t| | |S |dkrtd|S |dkr| S t | |d |}t||}|d dkr8t|| }|S )Nr   r'   r%   )pow_recursiverG   r>  r"  rK  )r   r  r   r  rX   rX   rY   r  $  s   r  c                 C     t | |S r~   )rG   powr   r  rX   rX   rY   
pow_native3  r  r  )r   c                   sV  t trtkrt tS t trdkrt S t tr,dkr,t S tdd  fD }t|}t toQd  k oIdk n  pQ|oQdk}|ro   fdd	}t	j
    |  d
S t  tr dkr}tdS  dkrt rtS |rt  trt S t trt S t S t S )Nr  r'   c                 s  s$    | ]}t |tjr| V  qd S r~   )r_   r*   r9   r   r   rX   rX   rY   r   K  r  zpow.<locals>.<genexpr>i    r   c                   s   t |   S r~   )r  r   rm  r   r  r  rX   rY   rl   U  r  zpow.<locals>.fnrJ  r%   )r_   r!  r   r  sqrtrt  r'  r   r5  r6   r   rb   r   r   r   r  r   exp2fallback_pow_scalarfallback_pow_tensor_scalarfallback_pow_tensor_tensorr  )r   r  r   is_integer_powembed_exponentrl   rX   r  rY   r  A  s@   
"







r  c                 C  s   t | tr	| j}n| }t |tr|j}t |tjs3tj|  |  |	 | 
 dj}t |tjs3J t |tjrR| sR| sRt |jtjsR|  |j|_| S tjj|||d | S )NrJ  unsafe_alias)r_   r9   ra   r*   rq  r6   r   rb   r   r5  r   is_input_bufferis_module_buffer	NopKernelr_  rM  realize_into)changedr  r  changed_datarX   rX   rY   r  q  s:   

r  c                 C  s   t | t| |S r~   )r  r  )r   r  rX   rX   rY   r    r  r  c                 C  @   | |u r| S t ||  }t||  }t||  }t| |S r~   r  rb   r   r   r  r   r  )r  r  r  rX   rX   rY   r`       
r`  c                 C  r  r~   )rG   floordivr  rX   rX   rY   r    r  r  c                 C  r  r~   )rG   r  r  rX   rX   rY   r    r  r  c                 C  s   t | ot |}t| ot|}|dkr(|rJ d|r!t| |S tt| |S |dkr@|r2J d|r9t| |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_booleanrX   rX   rY   div_mode  s   
r  c                 C  s8   t | ot |}|rt| |S ttjj}t|| |S r~   )r   logical_andr4   r  rK  r  rX  )r   r  	both_boolrl   rX   rX   rY   rK    s
   
rK  rp  Optional[ir.Constant]c              	   C  s   t | tjrt| jS t | tjrt|  S t | tjr| S t | tjs'dS t	j
j|  }t|' ttjdd | j|   }W d   n1 sPw   Y  W d   n1 s_w   Y  t |t	j
jjsnJ t |jtjrx|jS dS )z:Try convert an arbitrary IR node into an ir.Constant valueNallow_indexingT)r_   r*   
MutableBoxget_constant_valuera   r  r  r   Loopsrp   	_inductorops_handlerExtractConstantsHandlerrb   rH   set_ops_handlerr   objectr  rI  inner_fn_argsvirtualizedOpsValuer   )r   r  rd   rX   rX   rY   r    s*   
 r  c                 C  s|   t dd | |fD }|rt| |S t| }d ur3|jdkr)ttd|j}nd|j }t| |S dd }t|| |S )Nc                 s  s     | ]}t |pt|V  qd S r~   )r   r   r   rX   rX   rY   r     rm  zdiv_prim.<locals>.<genexpr>r   infrV  c                  W  r  r~   )rG   rK  r   rX   rX   rY   rl     r  zdiv_prim.<locals>.fn)	r  r  r  r   mathcopysignr!  rK  rX  )r   r  is_integralr  r>  rl   rX   rX   rY   div_prim  s   



r)  c                 C  s    t | |ftjd\} }t| |S r  )r,  r   INT_TO_FLOATr)  r  rX   rX   rY   r    s   


r  c                 C  s4   t | pt| }|rdd }ndd }t|| |S )Nc                 S  r  r~   )rG   modr  rX   rX   rY   rl     r  zfmod.<locals>.fnc                 S  r  r~   )rG   fmodr  rX   rX   rY   rl     r  )r   r   rX  )r   r  r(  rl   rX   rX   rY   r,  	  s
   
r,  c                C  B   t |  st|  r|d u rtj}td|d}|| |||dS )Nrl  rs  r   r   r   r   rp   r!  r  r   r4  r  r   rl   rX   rX   rY   r       

r  c                 C     t |  st|  r|d u rtj}t|  dkr.|dv s!J |p&|  }t| |ddS dd }t| ||d}t	j
jd
i |d|i\}|d u rRt| ||d	S |S )Nr   r   r2  Tr  c                 S     | \}|\}t ||fS r~   )rG   r   a_tupleb_tupler   r  rX   rX   rY   
combine_fn8     zcumsum.<locals>.combine_fnr4  r   r7  r   r   rX   )r   r   r   rp   r!  r   r   r   r  r*   Scanr   fallback_cumsumr   r4  r   r7  r   r  rX   rX   rY   cumsum,      

r>  c                 C  r1  )Nr   r2  Tr  c                 S  r3  r~   )rG   rK  r4  rX   rX   rY   r7  P  r8  zcumprod.<locals>.combine_fnr9  r7  r:  rX   )r   r   r   rp   r!  r   r   r   r  r*   r;  r   fallback_cumprodr=  rX   rX   rY   cumprodD  r?  rA  c                 C  sv   dd }|   }t|  dkr|dv sJ t| S t| ||d}tjjdi |d|i\}|d u r9t| |dS |S )	Nc              	   S  s\   | \}|\}t ||}t ||}||kt | B }t |t t || | |fS r~   )rG   r@  r?  r  r  log1pexp)r5  r6  r   r  min_vmax_vr,  rX   rX   rY   log_add_exp_helper^  s   $z(logcumsumexp.<locals>.log_add_exp_helperr   r2  r9  r7  r  rX   )	r   r   r   rt  r  r*   r;  r   fallback_logcumsumexp)r   r   rF  r   r   r  rX   rX   rY   logcumsumexp\  s   rH  c                 C     t |  dkr|dv sJ t| t| tjdfS |  }tjd|dd}t	| ||d}|tjf|d< | 
 d	d
 f|d< tjjdi |d|i\}}|d u rVt| |dS ||fS )Nr   r2  r   r  Fr   arg_break_ties_leftr9  r  c                 S     dS NrindexrX   r%  rX   rX   rY   r        zcummax.<locals>.<lambda>r  r7  r  rX   )r   r   rt  r  rp   r!  r   r*   get_reduction_combine_fnr  r5  r;  r   fallback_cummaxr   r4  r   r7  r   r   r  rX   rX   rY   cummaxr     rS  c                 C  rI  )Nr   r2  r   argminFrJ  r9  r  c                 S  rL  rM  rX   r%  rX   rX   rY   r    rO  zcummin.<locals>.<lambda>r  r7  r  rX   )r   r   rt  r  rp   r!  r   r*   rP  r  r5  r;  r   fallback_cumminrR  rX   rX   rY   cummin  rT  rW  c                C  r-  )Nrm  rs  r   r.  r/  rX   rX   rY   rm    r0  rm  c                 C  s   t | tj} td| ||dS )Nr  r4  r  )r   rp   r   r  r   r   r  rX   rX   rY   
reduce_any  s   rZ  c                 C  2   |d urt | ||dt| ||dfS t | d |dS NrX  )reduce_amaxreduce_argmaxrY  rX   rX   rY   
reduce_max  
   r_  c                 C  r[  r\  )reduce_aminreduce_argminrY  rX   rX   rY   
reduce_min  r`  rc  xor_sumr^  r]  r  rs  rU  
logical_or)rT  rG  stabler   
descendingc          
   	   C  s:  |d u rd}|   }|  }tt||}t|dkr't| td|tj|fS t|r/|| nd}tj	j
|ttjjsFt| |||dS t|ddtj|dd}dgt| }t|r`|||< t||}t||}tjj|| j|jf|  | f||||d\}	}|	d u rt| |||dS |d usJ |	t|tjfS )NFr   r'   rf  )r  r  r   ri   r  )ri   r  r  r   r4  rg  rh  )r   rb   r   r   rt  r  rp   r!  rH   rP   r  statically_known_ltr  int16r^  sort_fallbackr  r  r  r*   Sortr   r   r5  r   )
r   rg  r   rh  r  ri   r  r  
view_shaper   rX   rX   rY   sort_stable  s>   



	rn  c                 C  s   t | d||dS )NFrf  )rn  )r   r   rh  rX   rX   rY   sort  r  ro  c                 C  s   t | |tj|dS )Nr   r   rV  r  r   r*  )rS   r   rV  rX   rX   rY   register_pointwise_numeric  s   rr  c                 C  s   t | tjddS )NT)r   r  rq  r  rX   rX   rY    register_pointwise_numeric_ldf64  s
   rs  r  logical_not)rG  )r   r   r*  identity)r]  pointwise_overrides_datac                 #  s    t |  t|  jd }|d u rd S  fdd}t|tjjr6| D ]}t||}| j||fV  q#d S | j||fV  d S )Nc                   s    j d u r	t| S d S r~   )tritonr  r  r  rX   rY   make_triton_fallbacku  s   
z6_get_pointwise_overrides.<locals>.make_triton_fallback)	rv  r   r   r_   rp   rz   r   r   r   )nsr   rS   rx  olnamer  rX   r  rY   _get_pointwise_overrideso  s   
r{  rp  c                   s,   | t |< t|   fdd}t| | d S )Nc                    sB    | i |}g }t | d |D ]\}}|t||dd q|S )Nr   Tr  )r   rc   r  )rg   r   resultsmut_resultsr   r  outplace_oprX   rY   rl     s
   z$register_foreach_inplace.<locals>.fn)rN   r[  r   r   )aten_opoutplace_aten_opr  rl   rX   r~  rY   register_foreach_inplace  s   
r  c                   s   t | d d fdd}|S )Nrv  c                    s.    | i |}t || d  }t| d |S r  )r   r   r  )rg   r   r  r~  rX   rY   rl     s   zregister_inplace.<locals>.fn)r  )r  r  rl   rX   r~  rY   register_inplace  s   
r  c                 C  r  r~   rX   r\  rX   rX   rY   sym_constrain_range
  r  r  c                 C  &   t jjjd }t|tjsJ |jjS Nr  	rH   rP   rQ   rP  r_   rp   r|  rV   r}  r   r   r  rX   rX   rY   sym_size  s   r  c                 C  r  r  r  r  rX   rX   rY   
sym_stride   s   r  c                 C  s   |   S r~   )r  )r   rX   rX   rY   	sym_numel(  r>  r  c                 C  r  r~   )r   Addr   rX   rX   rY   sym_sum1  s   
r  c                 O  r<  )NzHelpful for debuggingr   )r_  rg   r   rX   rX   rY   foobar6  r>  r  c                 C  s   |    t| S r~   )r_  rt  r   rX   rX   rY   _realize;  s   r  c                 C  s   |    t| | | S r~   )r_  r*   ResizeStorageBytes)variabler  rX   rX   rY   resize_storage_bytes_A  s   r  c                 C  s"   |    |   tt| |S r~   )r_  r9   r   r*   SetSourceTensorKernel)r_  source_tensorrX   rX   rY   set__source_tensorH  s   r  c                 C  r  r~   r  )r  r  rX   rX   rY   
fsdp_copy_Q  r  r  c          	        sv  t | tsJ t |ttfsJ |d u rtj}|tjkr#td| |tjkr0t	|dks0J |tj
kr=t	|dks=J |  |  }|  }t | jtjrV| j | _t rvtjjjrvt|rhtdnt|rst|jndndtjjdrt|||dS t | gd	g}|!  tj"#||}t$||||%  fd
d}t&j'|||t|d}|S )Nzunsupported memory format: r   r   nanTr  r   r  r'   c                   sH   |  t  tj}t tj}t ||}t | fddS )Nc                     s
    gS r~   rX   rX   )
flat_indexflat_loaderrX   rY   r    s   
 z*resize.<locals>.inner_fn.<locals>.<lambda>)rG   r   rp   r!  r$  r&  )r  flat_index_exprlimitr,  r  	old_numelout_indexeruninitalized_val)r  rY   rI    s
   zresize.<locals>.inner_fnrJ  )(r_   r9   r]   r   rp   contiguous_formatpreserve_formatr  channels_lastr   channels_last_3dr  r   r(  ra   r*   r  r  r8  r   r;  fill_uninitialized_memoryr   r!  r   r  r^  rH   rP   r  r  r  r  r5  r   stride_ordered_for_memory_formatr  r  r6   r   )	r   r   r  r   ri   x_flat
out_striderI  rd   rX   r  rY   resize\  sT   



	r  )auto_functionalizedc                 C  sB   ddl m} ||}tj| ||i ||d dd | D S )Nr   )kernel_side_table)
kernel_idxgridtma_descriptor_metadatakernel_argsc                 S  s    i | ]\}}t |tr||qS rX   r   )r   r  r  rX   rX   rY   r     r  z'triton_kernel_wrap_.<locals>.<dictcomp>)*torch._higher_order_ops.triton_kernel_wrapr  get_constant_argsr*   UserDefinedTritonKernelr   )r  constant_args_idxr  r  r   r  constant_argsrX   rX   rY   triton_kernel_wrap_  s   	

r  c                 C  sj   t dd | g|D r$d}tjjjdd  }r | d| }|tj_tj	| |||}t
ttj	|S )Nc                 s  r/  r~   r0  r   rX   rX   rY   r     r"  zcond.<locals>.<genexpr>z"control flow operator: torch.cond.r6  r7  )r  rH   rP   rQ   rP  rQ  r9  r*   Conditionalr   r]   mapr9   )predtrue_fnfalse_fnoperandsr   r6  r  rX   rX   rY   r     s   r   c                 C  sh   t dd || D r#d}tjjjdd  }r| d| }|tj_tj	| |||}t
ttj	|S )Nc                 s  r/  r~   r0  r   rX   rX   rY   r     s
    
zwhile_loop.<locals>.<genexpr>z(control flow operator: torch.while_loop.r6  r7  )r  rH   rP   rQ   rP  rQ  r9  r*   	WhileLoopr   r]   r  r9   )cond_fnbody_fncarried_inputsadditional_inputsr   r6  r  rX   rX   rY   
while_loop  s   r  subgraph_fnir.Subgraph
identifierc                 C  s   t j| |}tttj|S r~   )r*   InvokeSubgraphr   r]   r  r9   )r  r  r  r  rX   rX   rY   invoke_subgraph  s   r  )schemec          
      G  s   d }t jjjdd }|d usJ t| jjjD ]Y\}}|jdkr*|| t jj	|< q|jdkrgt j
|\}}t|| D ]}	|	  |jrQt jj|	  t jj|	  q?tjjt j|||}qt j|t jj	|< q|S )Nquant_optionsplaceholderr  )rH   rP   rQ   rP  rQ  r^   rz  nodesrS   envfetch_args_kwargs_from_envr   r   r   r_  codegen_low_precisionlow_precision_codegen_opsr   r`  invoke_quant_opsrp   r  Interpreterr  run_node)
r  r  r  r  r  rf   rV   rg   r   r   rX   rX   rY   invoke_quant_tracer  s"   

r  r7  r  tuple[torch.Tensor]c                   s   ddl m m} t|dkrtd fddt||D }|| |fdd}t|d dd d	}td
d |D |d< tdd |D |d< t	j
jd|dd|}|d d u r_td|S )Nr'   )InputDescriptorlower_pointwise_subgraphr   zSUnable to generate code for associative_scan op, because there are lifted argumentsc                   s    g | ]} |  | d qS )r  )r   rb   r   )r  rX   rY   r     s    z$associative_scan.<locals>.<listcomp>c                   s    g t | t |R  S r~   )r  r  )lhsrhs)lowered_combine_fnrX   rY   wrapped_combine_fn  s
   z,associative_scan.<locals>.wrapped_combine_fnr9  c                 s      | ]}|  V  qd S r~   r  r   rX   rX   rY   r     r   z#associative_scan.<locals>.<genexpr>r  c                 s  r  r~   r4  r   rX   rX   rY   r     r   r  F)r7  can_fallback_to_atenz/Unable to generate code for associative_scan oprX   )ry  r  r  r   r  r   r   r  r   r*   r;  r   )r7  xsr  r  subgraph_inputsr  r   r  rX   )r  r  rY   associative_scan  s,   


r  c                 C  r  r~   rX   )tokensrX   rX   rY   _sink_tokens  r  r  c                 O  s   t jj|g|R i |}ddlm} ||||}|d usJ tjj| }|d u r,|fS t	t j
tj|}t|ttfs@||fS |g|R S )Nr   )get_effect_key)r*   EffectfulKernelr   torch._higher_order_ops.effectsr  rH   rP   effectful_opsr  rt  MultiOutputr9   r_   r]   r   )tokenrS   rg   r   r  r  effect_typeeffectful_kernelrX   rX   rY   with_effects  s   r  )register_comm_loweringsc                 C  s   t | |dddd}|d }tjjt|}tjjdi |d|d\}}|dkrFtjj	|t
jkrFtjd| d|d	|\}}||fS ttd
 t| |dd}	ttj t| |	}
t|
|d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)rt  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  rX   )r  rH   rP   r  simplifyrE   r*   r7   
num_splitsr  r(   r  r3   r   r  r  textwrapdedentr]  rL   r  rC  rR  r  )r   r   r   r  r  hint	num_split
max_tensor
sum_tensorrw  rC  xsumrX   rX   rY   prepare_softmax_online9  s8   



r  r  )quantized_lowerings)mkldnn_lowerings)jagged_loweringsrS   c              	   c  st    t | tjjsJ dt| }zt| t|  dV  W |r&|t| < dS t|  dS |r4|t| < w t|  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)	r_   rp   rz   r{   rL   rQ  r  r  r0  )rS   old_handlerrX   rX   rY   force_fallback  s   

r  )r[   r\   )rl   rm   rn   ro   )r   r   )r   r   )rg   r   r   r   r   r   r   r   r   r   rn   r   )r   r   )r   r   rn   r  )NN)NNNNFN)F)r   r9   r   rn  )r   r9   ri   r  r~   )r   r   r  r'   Tr`  )rd  r9   r2  r9   r3  r9   r4  r   r5  r   r6  r   r   rn  rn   r9   )rd  r9   r2  r9   r3  r9   r4  r   r5  r   r6  r   r   rn  rN  rO  rn   r9   )rd  r9   rB  r!  rC  r   r5  r   r6  r   r   rn  rn   r9   )rd  r9   rB  r!  rC  r   r5  r   r6  r   r   rn  rN  rO  rn   r9   )rd  r9   rB  r9   rC  r9   r5  r   r6  r   r   rn  rn   r9   )rd  r9   rB  r9   rC  r9   r5  r   r6  r   r   rn  rN  rO  rn   r9   )r   r   r'   )r  r   r  r   r  r   r  )T)rj   r  )rV   r  )NTF)ri   r  )r   rG  r  r9   rH  rI  r  r   )
rL  r   rM  r   r   rG  r  r9   r  r   )rQ  r9   rn   rR  )rQ  r9   rn   rW  )r^  r9   r_  r9   rZ  r   r[  r   r\  r`  r]  ra  rn   r9   )rd  r9   rq  r9   rZ  r   r[  r   )r   r   r$  r   )r   NNr'   )NNN)r2  FF)r   r   )r  r  r   r   r  r`  rf  r   )r   r   r  r`  )r   r   rf  r   )r%   F)r~  r  r  r   r  r   )r2  r  )r  r  r  r  )r  r  r  r  r  r  )r   r9   r  r  r  r!  rn   ra  )rf   r  rL  r  )rf   r  rM  r  )rf   r  rM  r  rL  r  )NrV  N)Nr   r'   F)r  )NNNN)rX   r   FTN)rt  rI   r   )r   rp  rn   r  )r2  F)r  r  r  rI  )r  r  )r7  r  r  r  r  (  
__future__r   
contextlibr  r   r   loggingr&  rj  r  r  r  collectionsr   collections.abcr   r   typingr   r   r   r	   r
   r   r   typing_extensionsr   unittest.mockr   r   rp   $torch.ao.quantization.fx._decomposedtorch.fxtorch.utils._pytreer   _pytreer  torch._dynamo.utilsr   (torch._higher_order_ops.associative_scanr   r  r   torch._prims_commonr   r   r   r   r   r   r   r   r   r   r   torch.fx.experimental.sym_noder   r   torch.utils._ordered_setr   torch.utils._sympy.functionsr    r!   r"   r#   r$   _dynamo.utilsr&    r(   r)   r*   r+   decompositionr,   r-   r.   r/   r0   r1   r2   r3   r4   r5   r6   r7   r8   r9   r:   r;   r<   r=   r>   r?   r@   rA   rB   rC   rD   rE   rF   r#  rG   rH   r  rI   rJ   rK   FALLBACK_ALLOW_LIST	getLoggerr  r  rL   __annotations__rM   rz   r{   r  r  tr_c10dr  r   _higher_order_opsr  rU   r[  rN   quantized_decomposedrZ   rk   r}   r|   r   r   r   r  r  rO  r  bmmconvolutionconvolution_backwardr  r#  r  r  r  _int_mmrk  rj  rj  r  r!  rO  r=  r=  	complex32	complex64r   rN  r   r   r   r   r   r   r   r   r   r
  r&  r  r   r,  rX  rm  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  slicer  r  r1  quantize_per_channelrL  dequantize_per_channelrU  quantize_per_tensorrZ  dequantize_per_tensorr]  r
  rf  rh  r  r  r  r  r  r  r  r  r  r  r  r   r  r  	lru_cacher  r  r  r  r  r  rngprimsr  r  r   	bernoullir  r#  r&  r)  r+  r  r1  r.  r/  r5  r4  r3  randintforce_stride_orderr;  r  r=  rD  r@  lookup_seedrE  randomrK  rP  rV  rX  rn  r!  rf  rw  ry  r  r  rt   r  _adaptive_avg_pool3dadaptive_max_pool3dfractional_max_pool3dmax_pool3d_with_indicesuniformexponential_pdist_forwardsoft_margin_loss_backward_cdist_forward_cdist_backward
_trilinearsegment_reduce_segment_reduce_backwardhistc	histogrambin_ct_histogramdd_bin_edges_histogramdd_from_bin_ctsaddbmm_addmm_activation
_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_backwardro  rg  kthvaluetopkrH  median	nanmedianrD  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_backwardr  angle_efficientzerotensor(_sparse_coo_tensor_with_dims_and_tensors	to_sparse
_to_sparser   r  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_reduceru  rt  r  r  r  r  r  r  scalar_tensorr  
LongTensorr  r  r  r  r  r  r  r  r  r  r  rQ  
zeros_liker  r  r  r  r  r  r   r  r  r#  r%  r$  r$  r+  r2  r4  r5  r<  r?  rA  r/  rW  fallback__unsafe_masked_indexr[  ,fallback__unsafe_masked_index_put_accumulater  rb  rP  ri  rc  rs  rq  ru  ro  r  r  r  r  r  r<  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r-  r3  r?  rD  rV  rP  ra  r_  rz  ru  rm  r  r  r  r  r  r  r  r  r  r  r  r  rv  r  r  r  r  r  r  r  r  r  r  Tensor_Tensorr  Scalarr  Tensor_Scalarr  r  r  r`  r  r  r  r  rK  r  r)  true_divider*  r,  rl  r  r>  r<  rA  r@  rH  rG  rS  rQ  rW  rV  rm  r  rZ  r^  r_  r]  rc  rd  rw  r]  rx  ra  r  r^  rU  rb  r   rk  rn  rr  rs  rsqrtrC  r   expm1relur  r  r  rR  cossinabsbitwise_andbitwise_left_shiftbitwise_not
bitwise_orbitwise_right_shiftbitwise_xorlgammaerfspecial_erfrB  tantanhr  rt  re  logical_xorr?  r@  	clamp_min	clamp_maxnegr>  	remaindersignsignbit	_neg_viewler$  r#  r@  r  necoshsinhacosacoshasinasinhatan2atanatanhr'  erfcerfinvhypotlog10log2	nextaftercodegen.commonr]  rv  r{  r   rS   r   rV  _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_copyr  _foreach_add__foreach_mul__foreach_div_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  r  r   r  r  r   methodfuncr  r  _inductor_testr_  r  r  r  set_source_Tensorr  fsdpr  r  *torch._higher_order_ops.auto_functionalizer  r  higher_orderr   r  r  invoke_quantr  r  r  r  comm_loweringr  r  r  r  register_quantized_opsregister_woq_mm_opsr  register_onednn_fusion_opsr  register_jagged_opscontextmanagerr  rX   rX   rX   rY   <module>   s
  $4@4



	H85^:
,
4


#












1
E22$$-/|6



		
%2
&





			]/ 


C
8



.



%#I%

~



!t/


_7#A 

 
)5


E


Q


F-


z
	 &

	 H1
+


/
#
	 











'	














@

#6
