o
    ei	                    @  s5  U d dl mZ d dlZd dlZd dlZd dlZd dlZd dlZd dlZd dl	Z	d dl
Z
d dlZd dlmZ d dlmZmZmZmZ d dlmZmZmZmZmZmZmZ d dlmZ d dlmZ d dlZd dl Z d dl!Z d dl"Z d dl#m$  m%Z& d dl'm(Z( d d	l)m*Z* d d
l+m,Z, d dl-m.Z. d dl/m0Z0 d dl1m2Z2m3Z3m4Z4m5Z5m6Z6m7Z7m8Z8m9Z9m:Z:m;Z;m<Z< d dl=m>Z>m?Z? d dl@mAZAmBZBmCZC d dlDmEZE d dlFmGZGmHZHmIZImJZJmKZK ddlLmMZM ddlNmOZOmPZPmQZQmRZR ddlSmTZTmUZU ddlQmVZVmWZWmXZXmYZYmZZZm[Z[m\Z\m]Z]m^Z^m_Z_m`Z`maZambZbmcZcmdZdmeZe ddl$mfZfmgZgmhZhmiZimjZjmkZkmlZlmmZmmnZnmoZompZpmqZq ddlrmsZsmtZt erAddlumvZv edZwedZxeEddgZyeze{Z|i Z}de~d < i Zd!e~d"< eEe jj  Ze jsjZe jsjZe jsjZeEe jj  ZeEe jj e jjgZeEe jj  Zi Zd#e~d$< e jsjZd
d'd(Zdd,d-Zdd1d2Zdd6d7Zdd<d=Zdd@dAZddEdFZeejejejejejejejejejejejejejejg e je je je je je je je je je je je je jdGZddKdLZddPdQZddSdTZdUdVdd[d\Zd]d^ ZddbdcZddgdhZddpdqZddudvZddxdyZdUe7jdUe}fdd{d|Zd}d~ ZdddZ				U	dddZdddZdd ZddddZee jjdddd ZeejdddddZdUddddZeejjdddddZdUdUdd ddZeejdddd ddZdde7jdUdddUdfddZende7jdd eejdddd!ddZdd Ze΃  	UdddZeejdUdddd ZeejdUdddd Zeejejejejejgdd Zeedreejكe׃ eejddd"ddZeejddd"ddZeejgd"ddZeej݃dd Zeejރdd Zeej߃dd Zeejdd Zeejjdd Zeejdd Zeejdddd Zeejdddd ZeejddddĄ ZeejddƄ Zeejddeejddeejddd#ddʄZeejdddd̄ Zeejddd$ddτZeejddd"ddфZeejddd"ddӄZeejddd"ddՄZd%ddׄZeejddd&ddZdd Zeejjdd Zeejjdd Zeejddddd'ddZeejjddd(ddZeejjddddd)ddZeejjddd*ddZeejjddddd+ddZeejd%ddZeej ddd,d-ddZ eejddd,d-ddZeejddd,d-d dZeejdddd Zeejddd%ddZeejddd%ddZeejddd%dd	Zeejddd
d Zeejdddd Zeej	dddd Z	d%ddZ
eejd.ddZd/ddZejdd Zd"d0ddZd"d0ddZd/d1d!d"Zd2d#d$Zd%d& Zee jsjjddd'd( Zeejddd)d* Zeejddd+d, Zeejjddd-d. Zeejd/d0 Zedd1d2 Zd3d4 ZeejjZ eejj!Z"eej#jZ$eej#j!Z%eej& ee jsj'j(j ee jsj'j)j eejd5d6 Zeej#d7d8 Z#eePj*ddd9d: Z+eePj,ddd3d;d<Z-eePj.ddd=d> Z/eePj0ddd?d@ Z1eePj2ddd dAd4dFdGZ3eePj&ddd dAd5dJdKZ4d6dNdOZ5d7dQdRZ6eej7j8dddUdUdddSd8d\d]Z7eej9j8e7j:ddUdUd^d9d`daZ9dbdc Z;ddde Z<dfdg Z=dhdi Z>djdk Z?dldm Z@dndo ZAdpdq ZBeejC eejD eejE eejFdUdr eejGjdUdr eejHe< eejIdUdr eejJdUdr e jKL reejMdUdr e jNO reejPdUdr eejQ eejR eejS eejTj eejUj eejV eejWjX eejYj eejZj eej[ eej\dUdr eej]e; eejeA eej^e; eej_e< eej`e; eejae< eejbe< eejce< eejd eeje eeje eejf eejg eejh eejie; eejj eejk eejl eejm eejn eejo eejp eejq eejre< eejs eejt eejue< eejv eejvjw eejx eejy eejz eej{ eej| eej} eej~ eej eej eej eej eej eej eej eej eej eej eej eej eej eej eej eej eej eej eejj eej eej eej eej eej eej eej eejj eejjdUdr eeje; ee jjj ee jjj ee jjj eej eej eeje< eej eej eej eej eej eejj eejjeBdUdr eejjeBdUdr eejjeBdUdr eejjdUdr eejjeBdUdr eejjeBdUdr eejjeBdUdr eejjeBdUdr eejjeBdUdr eejjeBdUdr eejjeBdUdr eejjeB eejj eejjeB eejjeB eejjeB eej eejj8dds eejje< eejddddtdZeejddudvdwZdxdy ZeedzrJeeje eejd{d| Zeejddd:ddZeejddd;ddZĐdd Zee jejgddddUdddZee jǃdddZee jȃdd Zeejʃdd Zeej˃dd Zeej̃	d<dddddZ̐dd Z͐dd Zΐdd Zee jejgdddddddddZАdd Zѐdd ZeejӃeѐeЃZӐeѐedZԐeѐed ZՐdd Zeej׃dddddddZeej؃dddddddZeejكdddddddZeejjdd Zee jejgdd ZeejdddddZeejddd=ddZݐdd Z	d/ddZߐdd Zd/ddZeejdddd~ Zeejdddd ZeejdddddZeejdddZdd ZddÄ ZeejddddĐdńZeePjddddƐdǄZddȐdɄZeejjdUdʍZeejjdUdʍZeejdddːd̄ Zeejddd͐d΄ ZedϐdЄ Zeejddd"dѐd҄Zeejddd>dӐdԄZddd՜d?dِdڄZeejddddۜd@dܐd݄Zeejddd>dސd߄Zeejddd>ddZeejddd>ddZeejdddddAddZ		UdBdCddZeejjd"dDddZeejjd"dDddZeejj	ddEddZeejj	ddEddZeejj			d<dFddZeejj			d<dFddZdd Zeejjdd ZdGddZeej ddd%ddZ dHd
dZdIddZdJddZ	dKddZddddZdd ZddddZdd Zeej	dd	UdddZ	dLd%d&Z
eejddd'd( Zd)d* Zeejdd		 		UdMd+d,Zeejdd		 		UdMd-d.ZeejjdUdʍZeejddd/d0 ZdNd2d3Zd4d5 Zd6d7 Zd8d9 ZeejjdUdʍZeejd:d; ZeejjdUdʍZeejd<d= Zd>d? Zeejd@dA ZeejdBdC ZdDdE Zeejj	dOdFdGZeejdd	H	 	U		dPdIdJZeejdd	H	 	U		dPdKdLZeejjdUdʍeejjdUdʍeejjdUdʍgZdMdN ZeejjdUdʍZeejdd	d"dOdPZeej jdUdʍZ!eej dd	d"dQdRZ dSdT Z"ddUdVdWZ#d"dQdZd[Z$d\d] Z%eej&dRdd^d_d`Z&dadb Z'dcdd Z(dedf Z)dgdh Z*eej+ej+gd"ddUdidjdkZ,eej-d"ddUdidldmZ-dndo Z.edpdq Z/eej0j1dUdʍZ2eej0j3dUdʍZ4eej0j5dUdʍZ6eej0ddrdsdt Z0ddudvZ7eej8dwdx Z8eej9ddddydzZ9ed{d| Z:ed}d~ Z;eej<ddrd"ddZ=eej>gddrdd Z>dSddZ?eej<gddrdd Z@eejAej<j8gde7jddd Z<eejBejBgddrdd ZBeejCejCgdRdd^ddZDeejEjZFeejGjZHeejIjZJeejKjZLeejMjZNeejEdddZEeejGdddZGeejIdd ZIeejKddd"ddZKeejMddd"ddZMeejOdRdd^ddZOeejPdRddZQeejRdddRddZSeejTdddRddZUeejVe$d eejWe$dZXeejYe$dZZeej[e$de jdZ\eej]e$de jdZ^eej_dddZ_eejvjwdUdʍZ`eejvjwdddddUdddZaeejvjdddTddZvdddZbdUddZcebejdZdecejeZeebejfZfebejgZgeejhZhecejiZiecejjZjeejkZkeejlddZleejmddrddddZmdVddZneejoj3en ecejp ecejq eejrZreejsZseejtZteejuddZueejvZveejwZweejxZxebejy ebejzZzeej{e7jdez ebej| ebej} ebej~ ecej| eejdde jdZeejdde jdZeejdde jdZeejdde jdZeejZeejZeeje eeje eejZeejrZrebejZeej eejddZeej߃ eeje jd eeje eeje jd eeje jd eeje jd eeje jdZeeje jd eeje jd ebej ebej ebej ebej ebej ebej ebej ebej ebej ebej ebej ebej ebej ebej ebej ebej ddlmZmZ dd ZeD ]@ZeeeD ]\ZZZeʐeeeed qeeeD ]\ZZZeʐeeeed qߐqeejje_ddZeejj3e_ddZeejj8e_dd eejje>Zeejj8e> eejj3e>Zeejjel eejj3el eejje eejjer eejj3e0 eejje0 eejje0 eejje<Zeejj8e< eejj3e<Zeejej eejed eejje eejj3e eejje eejj3e eejje eejj3e eejje eejj3e eeje eeje eejeZÐdd ZĐeejŐjejje eejŐj3ejj3e eejƐjejje eejƐj3ejj3e eejǐjejje eejǐj3ejj3e eejjejjeÃ dÐdĄ Zɐeejʐe_ eejːes eej̐et eej͐eu eejΐev eejϐew eejАex eejѐe> eejҐj8e< eejҐjӐe= eejԐe eejՐe eej֐e eejאe eejؐel eejِeh eejڐei eejۃes eej܃et eej݃ev eejރew eej߃ex eejejۃ eejej܃ eejej݃ eejejރ eejej߃ eejddŐdƄZeejjdǐdȄ Zeejjdɐdʄ Zeejdːd̄ Ze> D ]\ZZee?ee qee jd͐d΄ ZeejdϐdЄ Zee jsjjdѐd҄ Zee jsjjdӐdԄ Zee jsjjjdՐdք Zee jsjdzrMee jsjj9jdאd؄ Zee jsjjddudِdڄZd dېlmZ ee ee,dܐd݄ Zee jsjjdddWdd8Zee jsjjdddddZee jsjj ddejedd ee jsjjdddXddZdYddZd dlmZ eedddd Zee jjdddddZddZee*ddd[ddZ	ee jsjj
jdd Z
ee jsjjjdd Zee jsjjdddd ZddlmZmZ e  e  eePjdddd Zd d ZeePjdddd ZddlNmZ eMe ddlNmZ e  e  ddlNmZ e  ddlNmZ e  ejd\dd	ZdS (]      )annotationsN)defaultdict)Callable
CollectionIterableSequence)AnycastOptionalTYPE_CHECKING	TypeGuardTypeVarUnion)	ParamSpec)patch)counters)associative_scan_op)triton_kernel_wrapper_mutation)FakeScriptObject)get_layout_constraint_tag)canonicalize_dimcanonicalize_dimscheckdtype_to_typeelementwise_dtypesELEMENTWISE_TYPE_PROMOTION_KINDget_computation_dtypeis_boolean_dtypeis_float_dtypeis_integer_dtypeNumber)magic_methodsmethod_to_operator)free_unbacked_symbolshas_free_unbacked_symbolsresolve_unbacked_bindings)
OrderedSet)CeilDivFloorDivIdentityModModularIndexing   )import_submodule   )configinductor_primsirtest_operators)decompositionsget_decompositions)BaseView	DtypeView
ExpandViewIndexingConstantIRNode	is_triton
MutableBoxOnlineSoftmaxReductionops_wrapperPermuteView	Pointwise	ReductionSqueezeView	TensorBoxvalidate_irView)ceildivdecode_device
is_dynamicis_gpuis_pointwise_useis_view,needs_fallback_due_to_atomic_add_limitationspad_listlike#register_op_dtype_propagation_rules#register_op_requires_libdevice_fp64sympy_productuse_scatter_fallback)opsV)ReductionType_T_Pztorchvision::roi_alignzaten::index_add8dict[Union[Callable[..., Any], str], Callable[..., Any]]	loweringsz9dict[torch._ops.OpOverload, Optional[Callable[..., Any]]]_maybe_layout_constraintsz2dict[torch._ops.OpOverload, torch._ops.OpOverload]inplaceable_foreach_opsreturnboolc                  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)rR   graphcurrent_nodeusersoptargetforeach_ops)nodeuser re   b/var/www/addictedbytheproject.nl/epg/venv/lib/python3.10/site-packages/torch/_inductor/lowering.pycur_node_has_non_foreach_users   s   
rg   	arg_pairsIterable[Any]4defaultdict[tuple[Any, bool], list[tuple[int, 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$|d us;J d|r@|\}|||f ||f q
|S )NFTz.foreach op should have at least one tensor arg)r   list	enumerate
isinstancer   rG   r/   #combo_kernel_foreach_dynamic_shapesrB   data
get_deviceappend)rh   outunpack_argsiargsuse_foreachdevicetre   re   rf   group_foreach_args   s&   


ry   fnCallable[..., Any]Optional[Callable[..., Any]]c                 C  s>   t | tjjs	dS t| dd }rt|S | tv rt|  S dS )zHGet layout constraints. Returns None if there are no layout constraints.NF)with_default)rm   torch_ops
OpOverloadr   tag_to_layout_constraintrX   )rz   maybe_layout_tagre   re   rf   maybe_layout_constraints   s   r   tagtorch._C.Tag(Optional[Callable[..., tuple[Any, Any]]]c                 C  sV   | t jjjkr	tS | t jjjkrtS | t jjjkrtS | t jjj	kr$d S t
d|  )NzUnknown layout constraint tag: )r~   _CTagneeds_exact_stridesconstrain_to_fake_tensorsneeds_contiguous_stridesrequire_contiguous_stridesneeds_fixed_stride_orderconstrain_to_fx_stridesflexible_layoutAssertionError)r   re   re   rf   r      s   r   condmsgstrNonec                 C  s   | s	t d| d S )Nzinductor does not support NotImplementedErrorr   r   re   re   rf   
assert_nyi   s   r   Union[Collection[Union[torch._ops.OpOverload, torch._ops.OpOverloadPacket]], torch._ops.OpOverload, torch._ops.OpOverloadPacket]Optional[list[Any]]c                   sj   t  ttttfrdd  D S t  tjjrt	  d S t  tjj
r3t fdd  D  d S )Nc                 S  s   g | ]}t |qS re   )add_needs_realized_inputs.0xre   re   rf   
<listcomp>       z-add_needs_realized_inputs.<locals>.<listcomp>c                 3  s    | ]}t  |V  qd S N)getattr)r   overloadrz   re   rf   	<genexpr>   s    

z,add_needs_realized_inputs.<locals>.<genexpr>)rm   rk   settupler&   r~   r   r   needs_realized_inputsaddOpOverloadPacketupdate	overloadsr   re   r   rf   r      s   
r   9Union[torch._ops.OpOverloadPacket, torch._ops.OpOverload]
constraintCallable[..., tuple[Any, Any]]c                 C  s:   t | tjjr|  D ]	}|tt| |< qd S |t| < d S r   )rm   r~   r   r   r   rX   r   )rz   r   r   re   re   rf   add_layout_constraint   s
   r   )r   r.   r,                     	   
         dtypeUnion[int, torch.dtype]torch.dtypec                 C  s2   t | ts| S | tv sJ d|  dt|  } | S )Nzid z missing from DTYPE_ID_LOOKUP)rm   intDTYPE_ID_LOOKUPr   re   re   rf   decode_dtype  s
   
r   r   r   ,TypeGuard[Union[TensorBox, sympy.Expr, int]]c                 C  sB   t | trt|  pt|  S t | tjr| jdu S t | tS NT)	rm   rB   r   	get_dtyper   sympyExpr
is_integerr   r   re   re   rf   is_integer_type  s
   


r   !TypeGuard[Union[TensorBox, bool]]c                 C  s    t | trt|  S t | tS r   )rm   rB   r   r   r[   r   re   re   rf   is_boolean_type'  s   

r   F)return_compute_dtyperu   type_promotion_kindr   r   c                   s:   d	dd  fdd|D }t |d| i\}}|r|S |S )
Ninpr   rZ   c                 S  s8   t | ttjfr
| S t|  }tjdg| |  dS )Nr.   r   )	rm   r    r   Basiclenget_sizer~   zerosr   )r   dimre   re   rf   construct_input3  s   z+get_promoted_dtype.<locals>.construct_inputc                      g | ]} |qS re   re   )r   argr   re   rf   r   ;  r   z&get_promoted_dtype.<locals>.<listcomp>r   )r   r   rZ   r   )r   )r   r   ru   inpscompute_dtyperesult_dtypere   r   rf   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   )
rm   rk   r   r~   r   r   r   r   rW   rq   )aten_fnrz   r   other_fnre   re   rf   get_overloadsB  s   

r   r`   >Union[Any, torch._ops.OpOverloadPacket, torch._ops.OpOverload]	namespacec                 C  s6   t | tjjr|| jv S t | tjjr||  v S dS NF)rm   r~   r   r   _qualified_op_namer   name)r`   r   re   re   rf   in_namespaceR  s
   
r   rB   rw   torch.devicec                 C  s   t | jtjrt|  r| S dd |  D }|  }|durH|jdkrH||krHt|dks;t|dkrH|d dkrHt	t
tj| |dS | S )zB
    Copy cpu scalar if doesn't not match with given `device`
    c                 S     g | ]	}t jj|qS re   rR   r]   sizevarssize_hint_or_throwr   sre   re   rf   r   d      z)maybe_copy_cpu_scalar.<locals>.<listcomp>Ncpur   r.   F)rm   ro   r1   ReinterpretViewr$   r   rp   typer   rB   
StorageBox
DeviceCopycreate)r   rw   size
cur_devicere   re   rf   maybe_copy_cpu_scalar\  s   
$r   	list[Any]kwargsdict[str, Any]	broadcast)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|r|r$tjndd  D }|dd  D  t|d|i|rF |d  n|d   |D ]}t |  |< qP|D ]}	t|	 |	< q^dfddfdd D  fdd D |rt	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rt|	 ||	< q fS )zB
    Transforms arguments for broadcasting and type promotion
    c                 S     g | ]\}}t |tr|qS re   rm   rB   r   rt   r   re   re   rf   r   {      z"transform_args.<locals>.<listcomp>c                 S  r   re   r   r   kvre   re   rf   r   |  r   c                 S  s*   g | ]}t |ttjfst|d r|qS r   )rm   r    r   r   hasattrr   are   re   rf   r     s    c                 s  s    | ]
}t |d r|V  qdS )r   N)r  r  re   re   rf   r         z!transform_args.<locals>.<genexpr>r   r   r   r   rZ   c                   s6   t | tr
t| S t | tjrtj| j dS | S )Nvaluer   rw   )rm   rB   to_dtyper1   Constantr	  )r   )rw   r   re   rf   promote  s
   

ztransform_args.<locals>.promotec                   r   re   re   r  r  re   rf   r     r   c                   s   i | ]	\}}| |qS re   re   r  r  re   rf   
<dictcomp>  r   z"transform_args.<locals>.<dictcomp>c                 3      | ]} | V  qd S r   re   r   rt   ru   re   rf   r         c                 3  r  r   re   r   r  r   re   rf   r     r  N)r   r   rZ   r   )rl   itemsr~   r[   extendvaluesr   rp   r   broadcast_tensorsrk   	itertoolschainr   zipr   rangerm   r1   r  r7   r   )ru   r   r   r   r   args_indiceskwargs_indicespromoting_argsrt   r  broadcastedr   r   re   )ru   rw   r   r   r  rf   transform_argsp  sb   

r!  r   torch._ops.OpOverload	decomp_fnc                   s@   t  d 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
    ru   r   r   rZ   c                    s    | i |}t | |S r   )rC   )ru   r   rr   r#  re   rf   wrapped  s   z+_register_foreach_lowering.<locals>.wrappedN)ru   r   r   r   rZ   r   )	functoolswrapsr   rb   r   rW   dictfromkeys)r   r#  r%  aten_fnsre   r$  rf   _register_foreach_lowering  s   
r+  lowering_dictc                   s<   t  fdd}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 s7tdd |D r7J dt| |\} }|rF| g} | i |}t| |S )	NFr.   r   Tc                 s  s"    | ]}|t v pt|d V  qdS )_c10d_functionalN)	fallbacksr   )r   rz   re   re   rf   r         
z6_register_lowering.<locals>.wrapped.<locals>.<genexpr>c                 s      | ]}|d kV  qdS )rr   Nre   r   re   re   rf   r     r  zout= ops aren't yet supported)	rk   r(  r   rm   r   allanyr!  rC   )ru   r   unpackedrr   r   r   r   r#  r   re   rf   r%    s$   
z#_register_lowering.<locals>.wrapped)r&  r'  r   r   r(  r)  )r   r#  r   r   r   r,  r%  re   r4  rf   _register_lowering  s
   r5  .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,  )r&  partialr5  )r   r   r   r   r,  re   re   rf   register_lowering  s   r8  c                 C  s   t |}| r
| |kr|S g }tjt| t|tjjdD ]A\}}tjj	
|r-|| qtjj	
|r:|| qtjj	|| tt|jtt|jk rX|| 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.
    )	fillvalue)r   r  zip_longestreversedr   SOnerR   r]   r   is_size_one_or_falserq   check_equalsr   expandfree_symbols)r  boutputr   yre   re   rf   broadcast_symbolic_shapes'  s   $ rE  c              
     s,  |d u s|d u sJ d|d u r|d u rt j}tdd | D s"| S tdd | D rC|p3t| d|ifdd  fdd	| D S td
d | D }g }| D ]C}t|ttfrq|	t
tj|| | dt|  qPt|tjr|	t
t|| | dt|  qP|	| qP|S )NzEonly one of override_return_dtype or type_promotion_kind may be givenc                 s  s"    | ]}t |tjttfV  qd S r   )rm   r   r   r   floatr   re   re   rf   r   I       z$promote_constants.<locals>.<genexpr>c                 s  s"    | ]}t |tttjfV  qd S r   )rm   r   rF  r   r   r   re   re   rf   r   K  rG  r   c                   s4   t | tjrtj|  td dS tj|  td dS )Nindexr   rw   r  )rm   r   r   r1   r8   rF   r  r   r   re   rf   
const_funcR  s
   
z%promote_constants.<locals>.const_funcc                   r   re   re   r   )rJ  re   rf   r   Z  r   z%promote_constants.<locals>.<listcomp>c                 s  s&    | ]}t |tttjfr|V  qd S r   )rm   rB   r7   r1   r  r   re   re   rf   r   [  s   $ r  rH  )r   DEFAULTr2  r1  r   nextrm   r   rF  rq   r7   r   r1   r  r   get_device_or_errorrk   r   r   r   r8   )inputsoverride_return_dtyper   exrr   r   re   )rJ  r   rf   promote_constantsA  sL   

	rQ  c                   s"   d dd fdd}|S )NalpharN  rB   c              	     s  d urt dd D rrJ  S tr2| d ur1| dkr1ttd | d< n| d u s8J dd D d  pLd   dd  D ]!}t|tjstt	t	| kstJ d d	 d	|  qSt
jt
jftjd uottjd
d d uotjjjd uotjjjddo v  
f	dd}	sd }D ]}t| jr| } nq|sˈd  }	p|}tj| |dS )Nc                 s  "    | ]}t |tot|V  qd S r   rm   r9   r:   r   r   re   re   rf   r     r/  z0make_pointwise.<locals>.inner.<locals>.<genexpr>r.   c                 S     g | ]}|  qS re   make_loaderr   re   re   rf   r     r   z1make_pointwise.<locals>.inner.<locals>.<listcomp>r   zndim mismatch  r^   low_precision_pointwise_barrierFc                   s   t  t 	ksJ d  d	 tjkr&d ur& fddD  S g }tD ]'\}}| }|  }rN|v rNtj||dd}t||}|| q,| }rhtj|dd}t|S |S )Nzwrong ndim r[  c                   s   g | ]}| qS re   re   )r   loadrI  re   rf   r     r   zCmake_pointwise.<locals>.inner.<locals>.inner_fn.<locals>.<listcomp>F)use_compute_types)r   r~   r[   rl   r   rQ   r
  rq   )rI  inputs_loaded	inp_indexr]  rr   	inp_dtypedowncast)	r   emulate_output_castemulate_precision_castsrz   rN  loaders	low_pr_fpoverride_fn_when_input_boolrangesr^  rf   inner_fn  s    $z/make_pointwise.<locals>.inner.<locals>.inner_fnrw   r   rj  ri  )r2  rQ  rk   mulr   r   rm   r1   BaseConstantr   r~   bfloat16float16rR   r]   r   r^   metagetrH   rp   r   r?   r   )rS  rN  otherrj  rw   rt   allow_alpharz   override_devicerh  rO  triton_fallback)r   rd  re  rN  rf  rg  ri  rf   inner~  s`   

zmake_pointwise.<locals>.inner)rN  rB   re   )rz   rO  ru  rh  rt  rv  rw  re   rs  rf   make_pointwisev  s   Nrx  c                   s   ddd fdd}|S )Nr.   rR  rN  list[list[TensorBox]]c                   s   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| } fdd}t|t |||S )Nr   z1at least one input must be a list to a foreach opc                   s    r	| diS |  S )NrS  re   r  )rt  rS  pw_fnre   rf   apply_fn  s   z7make_foreach_pointwise.<locals>.inner.<locals>.apply_fn)r   rR   r]   r^   r_   ra   inplace_foreach_opsrg   rm   rk   r   rq   ry   r  foreach_group_loop)rS  rN  realize_outputsa_list_inputinputbroadcast_inputsgroupsr{  rt  rz  rR  rf   rw    s,   
z%make_foreach_pointwise.<locals>.inner)rN  ry  re   )rz  rt  rw  re   r  rf   make_foreach_pointwise  s   #r  c                 C  s   dg| }|   D ]6\\}}}g }|D ]#\}	}
||
}|||	< tj|tjr6|r6|r6|  ||  q|r?tj	| q	t
dd |D sKJ |S )aa  
    Common loop over grouped foreach arguments.

    Args:
        groups: Result of group_foreach_args - dict mapping (device, use_foreach) to groups
        num_outputs: Number of outputs to produce
        apply_fn: Function to apply to each set of args, returns the output
        realize_outputs: Whether to realize outputs for foreach fusion
    Nc                 s      | ]}|d uV  qd S r   re   r   re   re   rf   r     r  z%foreach_group_loop.<locals>.<genexpr>)r  rR   r]   has_featureBackendFeatureFOREACHrealizerq   get_operation_nameregister_operation_listr1  )r  num_outputsr{  r~  outputsrw   rv   groupoperation_list
output_indru   rC  re   re   rf   r}    s(   

r}  copyc                   s>   |    kr|rt| S | S  fdd}t| d| S )Nc                   s   t j|  dS )N)	src_dtype)rQ   r
  r   r   r  re   rf   	_to_dtype     zto_dtype.<locals>._to_dtyperO  )r   clonerx  )r   r   r  r  re   r  rf   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  r  r   re   r   re   re   rf   r   J  r  z_foreach_map.<locals>.<genexpr>)subgraph_loweringr  graph_modulerR   r]   set_graph_handlerrungraph_outputsry   r   r  r  r  r  r  rq   r  r  r1  )subgraphru   r   r  rN  gmpw_subgraphsub_outputsr  r  rw   rv   r  r  r  rC  re   re   rf   _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r  )
is_complexr   r   
empty_liker1   InplaceCopyFallbackr   fallback_handlerprimsconvert_element_typedefaultr
  )r   r   dstre   re   rf   _convert_element_typeN  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_pointr~   finfobitsiinfor   re   re   rf   _get_primitive_bitwidthc  s   z1to_dtype_bitcast.<locals>._get_primitive_bitwidth)	r   r  r  atenviewr   rB   r6   r   )r   r   r  x_dtyper  src_bitsdst_bitsre   re   rf   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   rB   r   r1   ComplexViewr~   rQ   r  r  r   r  r   r   re   re   rf   _view_dtyper  s
   
r  r  non_blockingc                C  s:   t |}|  |kr|rt| S | S ttj| ||S r   )rF   rp   r  rB   r   r1   r   )r   rw   r  r  re   re   rf   	to_device{  s   r  c                 C  s   t | |d|dS )NTr  )r  )r   rw   r  re   re   rf   _device_put     r  Tc	           
      C  sz   |p| j }t|}	t||| |durt|}t|	||||d}	t| |||d|	}	tt|r;ttt|d|d|	 |	S )z3A pointwise function that maps ops.{name} to inputsN)rO  rh  rt  rv  )r   r   r   )r   r   )__name__r=   rM   rx  r8  r  r  r   )
r   r   r   r   r   rO  rh  rt  rv  rz   re   re   rf   register_pointwise  s>   

r  ldexp)r   rO  )r   r   nc                   s   t d |  }| }|j}|j o|tjk}|r)|r) fdd}t|| |S t| r0tjn|fdd}t|d| |S )Nr  c                   s
    | |S r   re   )r   r  )ldexp_fnre   rf   compute_ldexp     
z%ldexp_lowering.<locals>.compute_ldexpc                   s0   t | }t d }t ||}t | |S )Ng       @)rQ   r
  constantpowrl  )r   r  
n_out_typetwo
pow_result	out_dtypere   rf   compute_fallback  s   z(ldexp_lowering.<locals>.compute_fallbackr  )r=   r   r  r~   r[   rx  r   float32)r   r  r  n_dtype
x_is_floatn_is_intr  r  re   )r  r  rf   ldexp_lowering  s    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   re   ru   r   r  re   rf   frexp0     zregister_frexp.<locals>.frexp0c                    r  Nr.   re   r  r  re   rf   frexp1  r  zregister_frexp.<locals>.frexp1r  c                    s$    d | i | d | i |fS Nr   r.   re   r  )pw_fnsre   rf   rz     s   $zregister_frexp.<locals>.fnNr  )
r=   rx  r~   int32r8  r  r  r  r  r   )r   r  r  rz   re   )r  r  rf   register_frexp  s*   
r  c                 C  s   t ||d}t| |}|S )Nrt  )r  r+  )r   pointwise_lowering_fnrt  rz   re   re   rf   register_foreach_pointwise  s   
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   )rQ   wherer  re   re   rf   rz   	  r  zwhere.<locals>.fnr.   r,   r  c                 S  r   re   r   r   re   re   rf   r     r   zwhere.<locals>.<listcomp>c                      g | ]} | qS re   re   r  r  re   rf   r     r   r   r  )rm   rF  r   constant_liker   r   rK  rl   r  r  r  r   r1   r  r7   r   rk   r   rx  r
  )r   r  rB  rz   r   indicesrt   r   re   r  rf   r    s&   
$
$
r  c                  G  s   t | dkrt| d ttfrt| d  S | S ttdd | D d}g }| D ]+}t|  }|kr4nt |t |ksHt	dd t
||D rMt||}|| q'|S )Nr.   r   c                 s      | ]}|  V  qd S r   )r   r   re   re   rf   r   '  r  z$broadcast_tensors.<locals>.<genexpr>re   c                 s  s0    | ]\}}t jj|t jj|kV  qd S r   )rR   r]   r   r>  r   r  rB  re   re   rf   r   .  s    
)r   rm   rk   r   r  r&  reducerE  r   r2  r  r@  rq   )rN  ra   r  r   sizesre   re   rf   r     s"   
r  c                 C  s   | S r   re   r   re   re   rf   nop8     r  
lift_freshc                 C  s   t | tsJ |d u rtt| jS t |ttjfr"tj	j
|ntdd |D }tt|  |}tt |ts=|fn|}g }t|  D ]\}}||v r[tj	j
t|ds`|| qH||  krlt| |S | S )Nc                 s  s    | ]
}t jj|V  qd S r   rR   r]   r   	guard_intr   dre   re   rf   r   J  r  zsqueeze.<locals>.<genexpr>r.   )rm   rB   rA   r   ro   r   r   r   rR   r]   r   r  r   r   r   r   r&   rl   guard_or_falseEqrq   r  )r   r   dims	new_shaper  r   re   re   rf   squeezeA  s   
r  c                 C  s   t t| |S r   )r  r  )r   r   re   re   rf   squeeze_copyX     r  c                 C  2   t | |}t| tsJ t|tsJ |j| _| S r   )r  rm   rB   ro   r   r   valre   re   rf   squeeze_]  
   
r  c                 C  2   t | rt| dtjdS td}t|tjd| S )NFr   isinfr  r   	full_liker~   r[   r=   rx  r   rz   re   re   rf   r  f     r  c                 C  r  )NFr   isnanr  r  r  re   re   rf   r
  n  r	  r
  c                 C  $   t | rt| S td}t|| S )Nceilr   r  r=   rx  r  re   re   rf   r  v     r  c                 C  r  )Nfloorr  r  re   re   rf   r  ~  r  r  c                 C  r  )Nroundr  r  re   re   rf   r    s   r  c                 C  r  )Ntruncr  r  re   re   rf   r    r  r  c                 C  s   t | g\} t| tjrt| t|S t| tsJ t|ttfs$J t| 	 t|kr0| S t
| 	 sWtjjt| 	 }|dkrWt
|sW| tjjt||  tt| jt|S r  )rQ  rm   r1   rm  r7   r   r   rB   rk   r   r#   rR   r]   r   r   rO   
mark_reusero   )r   r  x_size_productre   re   rf   r@    s$   
r@  c                 C  sL   t |}|D ]}d||< q| }t|D ]\}}|dkr t||}qt||S NrW  )rk   rl   	unsqueezer@  )r  shapebroadcast_dimensionsr   broadcast_dimensionr  idxr   re   re   rf   broadcast_in_dim  s   


r  c                 C  s   t | | S r   )r@  r   )r   rD  re   re   rf   	expand_as  r  r  c                   sb  t |   tt kr$tjjgtt      t| t  } tt|  ks0J t |  }d}ttD ]}| dkrHd}|| |  ||< q>|r`t|| 	 | 
 dS tdd t D rstt| |S  fdd}t st|stjjt }|dkr| tjjt||  |  tj| 
 | 	 |t |d	S )
NFr   Tr   rw   c                 s  s$    | ]\}}|d kp|d kV  qdS r.   Nre   r  re   re   rf   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   rk   r  r   r<  Zeror+   )rI  rt   old_sizerepeatsx_loaderre   rf   rj    s   zrepeat.<locals>.inner_fnrk  )rk   r   r   r   r<  r=  r  r  emptyr   rp   r1  r  r  r@  r#   rR   r]   r   r   rO   r  rZ  r?   r   )r   r"  new_sizezero_tensorrt   rj  old_size_productre   r   rf   repeat  s>   r(  r  Sequence[sympy.Expr]c                 C  s   t t| j|S r   )rB   rD   r   ro   )r   r  re   re   rf   r    s   r  c                 C  s6   t | tsJ t |ttfsJ tt| jt|S r   )rm   rB   rk   r   r>   r   ro   )r   r  re   re   rf   permute  s   r*              c              	   C  s  ddl m}m} t| tsJ t| |d}|  | }t|}t|tj	s.|dks.J |z|dkrBt
jj||rB|dkrB| W S W n	 tyL   Y nw ddd}	d\}
}|}|ru|	||d}
|	|||}|
duru|duru|
|}}d}|sttjj| j|||||d	S |sJ |t
jjjt
jjjd
 }|dusJ t|dksJ |d\}}| D ]\}}||dt|fkr|}q||dfkr|}q|
du s|du sJ t|||||  | }t
j||_t
j| |}|  du r|   |
dur|du sJ |   j!|
| " |   }n%tj#|||   j!| " | |  | dd	}t
j||_t
j| |}t$|  }t$| " }|||< ||  |9  < t%| |||S )a	  
    Lowers a slice call, creating ExternKernels for the output size & storage offset symbols,
    if the indices are unbacked and appropriate semantics aren't known.
    If they are known (indices are static/backed/unbacked with info), a SliceView is created.
    r   )CallMethodKeyr%   r.   Nc                 S  s   | d u r|S dd }t | } t |}|t | dr&|t | |r&| S |t | dr;|t | | r;| | S |t | |rE|S |t | | rPdS d S )Nc                 S  s   t jj| S r   )rR   r]   r   r  r   re   re   rf   <lambda>+  s    z5slice_.<locals>.compute_slice_index.<locals>.<lambda>r   )r   r@  GeLeLtGt)rI  r   r  rz   re   re   rf   compute_slice_index'  s   

 "z#slice_.<locals>.compute_slice_indexNNFclampunbacked_bindingsr,   r   storage_offsetTr   )&%torch.fx.experimental.symbolic_shapesr,  r%   rm   rB   _validate_dimr   r   r@  r   rR   r]   r   statically_known_leq	TypeErrorr1   	SliceViewr   ro   	shape_envr^   rp  r   r  pytreeSequenceKeyDynamicSliceSizeregister_bufferr   register_operationmaybe_get_layoutr  
get_layoutoffset
get_strideDynamicSelectStorageOffsetrk   
as_strided)r   r   startendstepr5  r,  r%   r   r2  start_index	end_indexambiguous_slicer6  sym_sizesym_storagesymkeypathb_sizer%  new_storage_offset	b_storage	new_sizesnew_stridesre   re   rf   slice_  s   






rX  c           	   	   C  s   d }d }t | trt | jtjr|  }| j}| j } |   t	| s-t
d|  dt| \}}t|r:|n|j|r@|n|jdd |D dd |D t|pSd}ttj||dS )Nzunrealized as_strided(z, ...)c                 S     g | ]}t |qS re   r   r@  r   re   re   rf   r         zas_strided.<locals>.<listcomp>c                 S  rY  re   rZ  r   re   re   rf   r     r[  r   ro   layout)rm   rB   ro   r1   r5   rp   r   unwrap_viewr  is_storage_and_layoutr   as_storage_and_layoutFixedLayoutrw   r   r@  r   )	r   r   strider7  
new_device	new_dtypestorage
old_layout
new_layoutre   re   rf   rH    s$   

rH  c                 C  s$   t | tsJ t| |||j| _| S r   )rm   rB   rH  ro   )r   r   rb  r7  re   re   rf   as_strided_  s   rh  c                 C  s   t | |||}t|S r   )rH  r  )r   r   rb  r7  resultre   re   rf   as_strided_copy  s   rj  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   rW  c                 S  rX  re   rY  rV  re   re   rf   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   re   re   )rt   idx_loadinputs_loadersre   rf   r-        z1pointwise_cat.<locals>.inner_fn.<locals>.<lambda>        rW  r,   )rQ   
index_exprr~   int64r  r   r  geltand_rq   rk   r)   maskedr  )
r  idx_dimmasksmasked_loadsrI  rJ  
start_condend_condmasknext_valr   rN  rl  inputs_ranges)rt   rk  rf   rj    sD   
zpointwise_cat.<locals>.inner_fnrk  )rq   r   rk   r?   r   rp   r   )rN  r   prev_endr   rj  r%  re   r|  rf   pointwise_cat  s   0

r  r  scaleszero_pointsaxisr   	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_constantsr~   r  r   rQ   r
  r  
reciprocalr  maximumminimum)
r  channel_idxr  scale
zero_pointqminqmax	inv_scaler  clamped	r  r   input_loaderr  r  r  scales_loaderr  zero_points_loaderre   rf   rj  
  s   

z;quantized_decomposed_quantize_per_channel.<locals>.inner_fnrk  )r   r   r   r~   rn  r
  r  rZ  r?   r   rp   )r  r  r  r  r  r  r   rj  re   r  rf   )quantized_decomposed_quantize_per_channel  s(   
r  c                   sP       t tj  fdd}tj    |t 	 d}|   |S )Nc                   sB   t j  t  | W  d    S 1 sw   Y  d S r   )r1   ComputedBufferforce_realizerQ   device_assert_asyncrZ  r^  r   re   rf   rj  '  s   $z_assert_async.<locals>.inner_fnrk  )
r  r
  r~   r[   r?   r   rp   r   rk   r   )r   r   rj  assertion_opre   r   rf   _assert_async#  s   
r  c                 C  
   t | |S r   r  r   re   re   rf   lower_assert_async5     
r  c                 C  r  r   r  r   re   re   rf   lower_assert_functional_async:  r  r  r  r  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.   r  r  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   r~   r  rQ   r
  sub)r  r  r  r  r  r  r  r  r  r  r  r  r  re   rf   rj  ]  s   
z=quantized_decomposed_dequantize_per_channel.<locals>.inner_fnrk  	r   r   r   r~   r  rZ  r?   r   rp   )	r  r  r  r  r  r  r   r  rj  re   r  rf   +quantized_decomposed_dequantize_per_channel?  s(   r  r  rF  r  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 )Nr  c           	        sf   | }t d| |tjd\}}t|| | }t tjd\}}tt|||}t| S )N      ?r   )r  r~   r  rQ   r  r  r  r
  )	r  r  r  r  r  r  r  r  r  r   r  r  r  re   rf   rj    s   
zBquantized_decomposed_quantize_per_tensor_default.<locals>.inner_fnr  r  rk  )r   r~   rn  r
  r  rZ  r?   r   rp   r&  r7  rF  r   r   r  r  r  r  r  r   rj  re   r  rf   0quantized_decomposed_quantize_per_tensor_defaultt  s   
r  c                  sv   |   |ksJ d| d|    d u rtj|    fdd}tj|  tj|t	|t
|d|  dS )Nr  r  c                   sF    | }t ||tjd\}}tt|tj|| }t|}|S r  )r  r~   r  rQ   r  r
  )r  r  r  r  r  r  r  re   rf   rj    s
   zDquantized_decomposed_dequantize_per_tensor_default.<locals>.inner_fnr  rk  )r   r~   r  rZ  r?   r   rp   r&  r7  rF  r   r   r  r  r  r  r  r   r  rj  re   r  rf   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 )	Nr  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   re   r   )r   r   r   r~   r  rQ   r
  r  r  r  r  r  )r  r  _scale_zero_pointr  r  r  r  r   r  r  r  r  scale_loaderr  zero_point_loaderre   rf   rj    s   zAquantized_decomposed_quantize_per_tensor_tensor.<locals>.inner_fnrk  )r   r~   rn  r
  r  r   r   rZ  r?   r   rp   r  re   r  rf   /quantized_decomposed_quantize_per_tensor_tensor  s.   ""r  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  r  r  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  re   )r   r   r   r~   r  rQ   r
  r  )r  r  r  r  r  r  r  r  r  r  r  re   rf   rj    s   zCquantized_decomposed_dequantize_per_tensor_tensor.<locals>.inner_fnrk  r  r  re   r  rf   1quantized_decomposed_dequantize_per_tensor_tensor  s.   ""r  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   r   c                 s  s$    | ]}|  tjtjfv V  qd S r   )r   r~   int8uint8r   r  re   re   rf   r   '  s    
zcat.<locals>.<genexpr>c                 s  s     | ]}t | d kV  qdS )r   N)r   r   r  re   re   rf   r   .      r.   r   c                   s   g | ]}t | qS re   r
  rV  r   re   rf   r   9  r[  zcat.<locals>.<listcomp>r   Union[TensorBox, ir.StorageBox]rZ   	ir.IRNodec                 S  s>   t | trt | jtjr| j S | jS t | tjr| jS | S r   )rm   rB   ro   r1   r5   r^  r   r   re   re   rf   unwrap_tensor;  s   

zcat.<locals>.unwrap_tensorc                 S  s   t | tjot | jtjS r   )rm   r1   r  ro   r@   rx   re   re   rf   is_reductionG     zcat.<locals>.is_reductionc                   sJ   t | ttjfr | S | p$t | tjo$t fdd|  D S )Nc                 3  s     | ]} t j|V  qd S r   )rR   r]   
get_buffer)r   readcan_fuse_reductionre   rf   r   P  s
    
z2cat.<locals>.can_fuse_reduction.<locals>.<genexpr>)rm   rB   r1   r   r?   r2  get_read_namesr  )r  r  r  re   rf   r  J  s   zcat.<locals>.can_fuse_reductionc                 3      | ]} |V  qd S r   re   r   rx   r  re   rf   r   W  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)	r1   r_  r`  ConcatKernelcan_realize_into_without_copyrm   rB   r   r?   )r   re  _)should_lower_cat_inputr  re   rf   r  Y  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  )rm   rB   r1   r   r?   inner_fn_opcountnum_opsr  rR   r]   r  )r   countr  )op_countr  re   rf   r  p  s   
zcat.<locals>.op_countr   r,   r`   r"  c                 S  s   | t jjt jjfv S r   )r  catr  constant_pad_ndr`   re   re   rf   additional_pointwise_ops     z%cat.<locals>.additional_pointwise_opsc                 3  s    | ]	}| kV  qd S r   re   r  )MAX_SIMPLE_OP_COUNTr  re   rf   r         c                 3  s    | ]}t | V  qd S r   )rI   )r   use)r  re   rf   r     s
    
c                 3  r  r   re   rV  r  re   rf   r     r  c                 3  r  r   re   rV  r  re   rf   r         
c                 3  r  r   re   r  r  re   rf   r     r  )r   r  rZ   r  rZ   r[   r`   r"  )rp   r   r1  r  require_channels_lastr  r  r  r  r   r  r9  r   r   rK  r2  r/   force_pointwise_catr  rB   r1   r  r   max_pointwise_cat_inputsrR   r^   r_   )
rN  r   
cpu_devicer  r  fusable_reductionMAX_COMPLEX_POINTWISE_CATpointwise_usesfuse_pointwise_usehorizontal_fuse_catre   )r  r  r  r   r  r  r  r  rf   r  $  s`   




r  rE  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, re   re   r  r  re   rf   r-        zdiagonal.<locals>.<lambda>r   )r   r   c                   s    g | ]\}}| fvr|qS re   re   )r   rt   r   r  re   rf   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 )NrW  r   r.   r,   )r   r  )r  diag_idxoriginal_idxcur_dimr  base_idxr  r  num_dimsoriginal_shapere   rf   	reindexer  s   
zdiagonal.<locals>.reindexer)r   r   r   r   rR   r]   r   evaluate_exprr   r0  evaluate_maxevaluate_minrl   rq   rB   r1   GenericViewr   )r  rE  r  r  offset_negative	diag_sizer  r  re   r  rf   diagonal  s:   
r  c                 C  s   t t| |||S r   )r  r  )r  rE  r  r  re   re   rf   diagonal_copy     r  c                 C  $   t | }t||||}t|| |S r   )r  r  	mutate_to)r  srcrE  r  r  rC  ra   re   re   rf   diagonal_scatter     
r  c                 C  s  t |}t |  | }d }tjjt |dr || }ntjjt |dr-|}|d urit	|rY| 
  |  }|  }|  j|| |  }||= ||= t| |||S t| |||d dd}t||S ttjjjtjjjd }	|	d us|J t|	dksJ |	tt|	 \}
}| 
  |  }|  }|
}tj|
||  j|| |  | dd}tj||_tj| ||= ||= t| |||S )Nr   r.   Fr4  r6  )r   r@  r   rR   r]   r   r  r0  r.  r$   r  rF  rD  rE  rH  rX  r  r%   r=  r^   rp  r   rL  iterr  r1   rG  rA  r   rB  )r   r   r  r   actual_indexr%  
new_striderT  slice_resultr6  unbacked_offset_symr  bufferre   re   rf   select  sR   



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.   rW  Fr4  )r9  rm   rk   r   r   rR   r]   r   r  r(   rq   rX  )
r   r  r   sizes_x_sizechunksri  rI  r   rJ  re   re   rf   split&  s   
r  c                 C  s   t | ||S r   )r  )r   r  r   re   re   rf   split_with_sizesA     r  c                   s>   t  d tjj   } fddt|D }|S )Nr   c                   s   g | ]}t  |qS re   )r  r  r   r   re   rf   r   J      zunbind.<locals>.<listcomp>)r9  rR   r]   r   r  r   r  )r   r   r  ri  re   r  rf   unbindF  s   r  c                   s   |   }t|}t|| |dkrtt| d|ddS |  }tjj}||| |	d t
|| d }||dkrL| |t|| | g |d   || d d  |}	 fdd}
ttj| |	|
S )Nr   F)rJ  r5  r.   c                   s:   | d |     }g | d   ||  d d R S )NrW  r.   re   )r  dim_idxr   rK  re   rf   r  d  s   &zunfold.<locals>.reindexer)r   r   r   rX  r  rR   r]   r   	check_leqcheck_ltr(   r   r  r'   rB   r1   r  r   )r   	dimensionr   rK  r  ndimdim_sizer   new_dim_sizeout_sizer  re   r  rf   unfoldN  s"   
(r#  c                 C  s2   t | |d}t|  }||tjj t| |S r  )r9  rk   r   insertr   r<  r=  r  )r   r   r  re   re   rf   r  k  s   
r  c                 C  r  r   )r  rm   rB   ro   r   re   re   rf   
unsqueeze_s  r  r%  c                 C  sZ   t jjjt|}t|  }|dk r||| 7 }d|  kr(|| k s+J  J |S r  )	rR   r]   r   r=  r  r   sympifyr   r   )r   r   rE  r  re   re   rf   r9  |  s    r9  rW  c                 C  s\   t | |d}tjj|  | d }t| |d|dd}t| |||d dd}t|t|S )Nr   r,   Fr4  )	r9  rR   r]   r   r  r   rX  rl  sigmoid)r   r   new_lenr  rB  re   re   rf   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     t | tjrt| S | S r   rm   r1   r9   rB   r   r   re   re   rf   wrap_tensors  r  z7fallback_handler.<locals>.handler.<locals>.wrap_tensors)r>  tree_mapr1   FallbackKernelr   )ru   r   r,  kernelre   rf   handler  s   z!fallback_handler.<locals>.handlerT)r.  r   _is_fallback_handler)r0  r  r1  re   r/  rf   r    s
   
	r  c                   C     t d d S )NzjTorchinductor does not support code generation for complex operators. Performance may be worse than eager.)warningswarnre   re   re   rf   _warn_complex_not_supported  s   r6  rx   torch.Tensorc                 C  s   |   r	t  dS | jrdS | jrdS | jtjkrD|sdS t|jtj	j
r6|jtjjtjjtjjtjjfv pBt|jtj	j
oBt|j S dS )z0Do not support reading or writing to this tensorTF)r  r6  is_meta	is_sparser   r~   float8_e8m0fnurm   ra   r   r   r  r  r  r  r  
_scaled_mmrJ   )rx   rc   re   re   rf   unsupported_input_tensor  s,   r<  c                 C  sL   t jjtjjjjf}|dur|j|v r| 	 rdS t
| |r dS | jo%tjS )z2Do not support writing tensor but can read from itNFT)r  r  r   r~   rQ   r  r  r  ra   r  r<  is_cpur/   disable_cpp_codegen)rx   rc   supported_complex_viewsre   re   rf   unsupported_output_tensor  s   

r@  rc   torch.fx.Nodec                   sv    j tjju r	dS  jdkrdS  j tjju rdS  fdd}tj ji  j	D ]}||ddr4 dS q)| ddS )NFplaceholderc                   sp   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
| r5 dS qdS )NFr  T)rm   r~   fxNoderp  r>  tree_leaves_subclasses
FakeTensorr@  r<  )inp_out_node	is_outputrp  rc   re   rf   check_skip_condition  s   


zCfallback_node_due_to_unsupported_type.<locals>.check_skip_condition)rI  T)
ra   r  view_as_complexr  r`   lift_fresh_copyr>  arg_tree_leavesru   r   )rc   allow_cpu_inputsrK  r   re   rJ  rf   %fallback_node_due_to_unsupported_type  s   
rP  c                   s  t jo| tjtjjtjjhv }| tvs|s|sJ d|  |rOtt	
drOt| grOt jr6| tjjjv sO|sOtjj jrGdtjj _td td|  d fdd}t| tjjrn|  D ]}t| |}|| q`d S t| tjjtjjfr||  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 Nr  )r   r   r8  r  )op_overloadlayout_constraintre   rf   register_fallback*	  s   

z(make_fallback.<locals>.register_fallbackzUnsupported fallback z with type )r/   re  r  addcmul_foreach_addcmulScalar_foreach_addcdivr3   r[   osgetenvr4   fallback_randomr~   _decompdecompositions_for_rngextra_random_decomps_dynamosuppress_errorslogwarningr   rm   r   r   r   r   r   HigherOrderOperatorRuntimeErrorr   )r`   rU  r5  override_decompskip_decomp_for_precisionrV  olrS  re   rT  rf   make_fallback	  sH   




rj  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   tensorr~   rp  )r  numelr   re   re   rf   philox_rand_offset<	  s   
rn  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   )rQ   r
  r~   r  r   ro  rand)rI  seed_index_exproffset_index_exprrand_index_exprri  r   offset_loader
random_posseed_loaderre   rf   rj  U	  s   zphilox_rand.<locals>.inner_fnrk  )
r1   ra  FlexibleLayoutcontiguous_stridesmake_indexerrZ  r?   r   rk   rn  )	r   seedrE  rb  rw   r   rj  random_values_nodeoffset_nodere   rs  rf   philox_randH	  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-  rB   r   r1   r.  r  native_dropoutr  r   )r   ptrainre   re   rf   r~  o	  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 )Nr   Tthis should be handled in decomps unless config.fallback_random or the device is CPUr   )r/   r]  rp   r~   rw   r  r   rm   rF  r  
bernoulli_Tensorr1   InplaceBernoulliFallback)r   ru   rS  re   re   rf   r  z	  s   r  c                 G  s4   t js|  tdksJ dtt| g|R  S )Nr   r  )r/   r]  rp   r~   rw   r  r  )r   ru   re   re   rf   bernoulli_p	  s   r  c                 C  s   t r   r   r  re   re   rf   _foobar	  r  r  c                 C  r3  )Nz1using triton random, expect difference from eager)rc  info)saltre   re   rf   _warn_triton_random	  r  r  c                   C  s   t tjj d S r   )r  rR   r]   creation_timere   re   re   rf   warn_triton_random	  r  r  c                  O  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_defaultr   r  re   re   rf   ro  	     ro  c                  O  r  r  )rq  fallback_randn_generatorr/   r]  r  fallback_randn_defaultr   r  re   re   rf   randn	  r  r  c                 C  s   t |}t j| |S r   )r1   get_stride_orderExternKernelrequire_stride_order)input_tensorrb  stride_orderre   re   rf   inductor_force_stride_order	  s   
r  c                 C     t d)Nz.should be handled in fuse_seed_creation_pass()r  )rw   re   re   rf   inductor_seed	     r  c                 C  s   t   tt| t|S r   )r  rB   r   r1   RandomSeedsrF   )r  rw   re   re   rf   inductor_seeds	  s   r  c                   s(    fdd}t j  |g dS )Nc                   s   t   S r   )rQ   	load_seedget_namer  rI  seedsre   rf   rj  	  r  z&inductor_lookup_seed.<locals>.inner_fnrk  )r?   r   rp   r   )r  rI  rj  re   r  rf   inductor_lookup_seed	  s   r  rE  r   	list[int]rz  modec                  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)ro  r  r  c                   s"   t t g t| tjS r   )r   rQ   ro  r~   r  r^  r  ru  rv  re   rf   rj  	  s   z!inductor_random.<locals>.inner_fnrk  )r/   r]  r~   r  rM  r1   ra  rw  rx  ry  rZ  r?   r   r  )r   rz  r  rE  r   rw   rj  ri  re   r  rf   inductor_random	  s(   
r  lowhighc                  sp   t jrJ g |}tj}| }tj|||tj||d	 |
  fdd}tj|||g |dS )Nr  c              	     s6   t g t | tjt tjt  tjS r   )rQ   	randint64ro  r~   r  rp  r^  r  r  ru  rv  re   rf   rj  
  s   z"inductor_randint.<locals>.inner_fnrk  )r/   r]  r~   rp  rM  r1   ra  rw  rx  ry  rZ  r?   r   )r  r  r   rz  rE  r   rw   rj  re   r  rf   inductor_randint	  s"   
r  tb.tuple[str, sympy.Expr, sympy.Expr, sympy.Expr]c                 C  sH   |   }|  }tdd t||D |d  }|  |d ||d fS )Nc                 s  s     | ]\}}|d  | V  qdS r  re   )r   r   stre   re   rf   r   
  r  z%_boundaries_helper.<locals>.<genexpr>rW  )r   rF  sumr  r  )r  r   rb  
max_offsetre   re   rf   _boundaries_helper
  s    r  tuple[str, sympy.Expr]c                 C  s   |   |  d fS r  )r  rF  r  re   re   rf   _sorter_helper'
  r  r  	out_int32rightsidesortersorted_sequenceselfr  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   )rR   r]   r  r  	BUCKETIZEr  re   re   rf   r-  5
  s    zsearchsorted.<locals>.<lambda>Fr  r  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rQ   	bucketizer  r  )r  r  index_dtyper  r  r  values_loaderre   rf   rj  Z
  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 )Nr  rB   c                   s>   |   }tttjdd t|d d  d d D S )Nc                 s  s    | ]	\}}|| V  qd S r   re   )r   r   rt   re   re   rf   r   q
  r  zNsearchsorted.<locals>.inner_fn.<locals>.get_flattened_index.<locals>.<genexpr>rW  )rF  rQ   ro  r&  r  operatorr   r  )r  strides)r  r  re   rf   get_flattened_indexm
  s   &z;searchsorted.<locals>.inner_fn.<locals>.get_flattened_indexr  )r  rB   r  )r  r  r  r  r  rf   rj  h
  s   	
rk  )r  r  searchsortedr  r~   r  rp  rZ  r  r   r   rp   r?   r   r  )
r  r  r  r  r  r  validate_bucketizerj  rw   ri  re   r  rf   r  +
  sB   
r  r  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  r  c                   s"   | }t |t d}|S r  )rQ   r  r  )rI  r  r  r  r  r  r  re   rf   rj  
  s   zbucketize.<locals>.inner_fnrk  )r   r   rR   r]   r  r  r  r  r  r  r  r  rp   rZ  r~   r  rp  r?   r   )r  r  r  r  rw   rj  ri  re   r  rf   r  
  s*   
r  c                 O  $   t tjtjj||f\}}||fS r   )r>  tree_map_onlyr1   r9   r  require_stride1r  ru   r   re   re   rf   require_dense
     r  c                 O  r  r   )r>  r  r1   r9   r  require_contiguousr  re   re   rf   r  
  r  r  c                 O  r  r   )r>  r  r1   r9   r  r   r  re   re   rf   r   
  s   r   c                 O  r  r   )r>  r  r1   r9   r  r  r  re   re   rf   r  
  r  r  c                   s   d u r S t tr S t  tjrtj  S t  tr+ fdd D S t  tt	fr@t
 dd t D S  S )Nc                   s    i | ]}|t  | | qS re   constrain_to_fake_tensorr   keyr   fake_argre   rf   r  
  r  z,constrain_to_fake_tensor.<locals>.<dictcomp>c                 s      | ]
\}}t ||V  qd S r   r  )r   r  f_are   re   rf   r   
      
z+constrain_to_fake_tensor.<locals>.<genexpr>)rm   r   r1   r9   r  require_exact_stridesrb  r(  r   rk   r   r  r  re   r  rf   r  
  s   

r  c                   s6   t dd t| |D }  fdd| D }| |fS )Nc                 s  r  r   r  )r   r   r  re   re   rf   r   
  s
    
z,constrain_to_fake_tensors.<locals>.<genexpr>c                   s    i | ]\}}|t | | qS re   r  r  fake_kwargsre   rf   r  
  r  z-constrain_to_fake_tensors.<locals>.<dictcomp>)r   r  r  )ru   r   	fake_argsr  re   r  rf   r   
  s
   r   c                   sJ    fdd t  fddt|jD } fdd| D }||fS )Nc                   sZ   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                   s    i | ]}| | | qS re   re   r  )apply_constraintr   fx_argre   rf   r  	  r  zEconstrain_to_fx_strides.<locals>.apply_constraint.<locals>.<dictcomp>)rm   r1   r9   r  rp  rb  rR   r]   r   r=  r  r  r(  )r   r  r  r  )r   r  rf   r    s   
z1constrain_to_fx_strides.<locals>.apply_constraintc                 3  s    | ]
\}} ||V  qd S r   re   )r   r   r  r  re   rf   r     r  z*constrain_to_fx_strides.<locals>.<genexpr>c                   s"   i | ]\}}| |j | qS re   r  r  r  fx_nodere   rf   r    s   " z+constrain_to_fx_strides.<locals>.<dictcomp>)r   r  ru   r  r  ru   r   re   r  rf   r     s   

r   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jjj}t	| |}|r7|d dkr7t
ttt| }jtjjkrL| dv rLt|dksJJ d}|jsVtj||S d	 jtjjjjkod| d
k}t |tslJ t| dvrv|S t| }|rttj||S t |tr| d ur|rttj||S |r9t
| }	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 ]B}||d  dkr||	|d   }||
v rtjj||d    drd||< qtjj|  ds-t |   }|||< qtj!||S |rFttj||S t |tr`| d ur`|r`ttj||S  fdd}t |j"tj#r||s||$ rttj||S tj||S )Nr  c                 S  s$   g | ]}t |tjr|jjn|qS re   )rm   r~   SymIntrc   exprr   re   re   rf   r     s    z=sdpa_constraint.<locals>.apply_constraint.<locals>.<listcomp>rW  r   )r   r   r   )r   r.   r,   r   r   r   r   r   r.   r,   c                   s$   t jjtt|  d  dS )NrW  r   )rR   r]   r   r  r   r  r*   r   r   	ALIGNMENTre   rf   
is_aligned  s   z=sdpa_constraint.<locals>.apply_constraint.<locals>.is_aligned)%rm   r1   r9   rp  rb  rR   r]   r   r=  r  rk   r;  r  r   r   ra   r  0_scaled_dot_product_efficient_attention_backwardr  is_cudar  r  r~   rQ   '_scaled_dot_product_efficient_attentionrB   is_aligned_realized_tensortry_match_insignificant_stridesrealize_inputmaybe_get_stridestatically_known_equalsrq   rE   r  ro   r5   r^  )r  r   r  meta_valmeta_stride_exprr=  r  effn_attn_fwd_biasis_aligned_tensorr"  expanded_dimsmaybe_stridert   out_stridesrb  r  )r  r  rf   r    s   






z)sdpa_constraint.<locals>.apply_constraintc                 3  s$    | ]\}\}} |||V  qd S r   re   )r   r  r   r  r  re   rf   r     s
    


z"sdpa_constraint.<locals>.<genexpr>c                   s$   i | ]\}}| d |j | qS rW  r  r  r  re   rf   r    s   $ z#sdpa_constraint.<locals>.<dictcomp>)r   rl   r  ru   r  r  re   r  rf   sdpa_constraint  s   ~r  )r5  )rg  c                 C  s   t |tjst||  |  d}|}|  | kr"t||  }|  | kr1t||  }|  | krDt	||  }t
|S t
|S Nr  )rm   r1   r9   rl  r   rp   r  r
  r   r@  r  )r  r  r  r   rr   re   re   rf   r  ~  s   )memory_formatc                C  s&   t j|  |  |  t|  dS Nrk  )r?   r   rp   r   rZ  rk   r   )r   r  re   re   rf   r    s   
r  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 )NrW  r\  )rm   rB   ro   r1   r   rq   rD  r  )r   reinterpret_view_layoutsr]  re   re   rf   clone_preserve_reinterpret_view  s   r
  rM  c                  s(    fdd}t jt| || gdS )Nc                   s   t j| d    dS )Nr   r   rQ   ro  r^  r   rI  rK  re   rf   rz     r  ziota.<locals>.fnrk  )r?   r   rF   )lengthrI  rK  r   rw   requires_gradrz   re   r  rf   iota  s   
r  r   rI  c                   s   t ||  }|  t|  d tjjt	dr%| 
    ntjjtdr1nttjj| | S 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   )rQ   r  eqro  r~   r  r  r   rI  
src_loaderr#  re   rf   rj    s   z select_scatter.<locals>.inner_fnrk  )r
  r   rZ  r9  rR   r]   r   r  r   r0  r   r.  r  r  select_scatterr  r  r  r@  r  r?   r   rp   rk   )r   r  r   rI  rj  re   r  rf   r    s&   

r  c                   s   t | } 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   re   re   )src_idxr  re   rf   r-        z1slice_scatter.<locals>.inner_fn.<locals>.<lambda>rn  )rQ   ro  r~   rp  rk   r(   rq   rq  r   r@  rr  r  r+   r  r&  r  rs  rt  r   r  )r  ru  rz  src_valr   r   rJ  r  rI  rK  r   r#  )r  rf   rj    sR   zslice_scatter.<locals>.inner_fnrk  )r
  r   rZ  r9  r   r1   r<  normalize_start_endrk   r(   r@  r?   r   rp   )r   r  r   rI  rJ  rK  src_sizerj  re   r  rf   slice_scatter  s    
.
r  c                 C  s*   t | ttfrt| dkrt| d S | S r  )rm   rk   r   r   _unwrapr   re   re   rf   r  0  s   r  r   rw   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^  ro   r   re   rf   rj  C     ztensor.<locals>.inner_fnc                   r   r   rQ   r  r^  r!  re   rf   rj  H  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   )rQ   r  r  rr  ro  r~   rp  )rI  rJ  mid)binary_searchro   r   rI  re   rf   r%  P  s   z/tensor.<locals>.inner_fn.<locals>.binary_searchr   )r   rQ   r  r^  r!  )r%  rI  rf   rj  O  s   r  rk  )r   r~   stridedrm   r  r   rp  get_default_dtyper   r   rF  r   rq   IntegerrR   r]   add_tensor_constantrl  r?   r   rF   )ro   r   rw   r]  r  ri  rj  re   r!  rf   rl  6  s,   *rl  c                 C  s@   t | tr|d urt| |} |d urt| |} | S t| ||dS r  )rm   rB   r
  r  rl  )ro   r   rw   re   re   rf   	as_tensoro  s   


r*  c                 C  s   t | tjdS r  rk  ro   re   re   rf   long_tensorz  r  r,  c                 C  s   t 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rQ|jjS t|S )Nr6  r.   r  )r%   rR   r]   r   r=  r^   rp  r   rL  r
  r  r1   DynamicScalarrA  r   rB  rm   r~   r  SymFloatSymBoolrc   r  r   r&  )ro   r6  binding_symrR  r  r  re   re   rf   _local_scalar_dense  s   
r1  c                 C     d S r   re   )ro   r   re   re   rf   _assert_scalar  s   r3  )rw   r]  c                C  r2  r   re   )r  r   rb  r   rw   r]  re   re   rf   _assert_tensor_metadata  s   r4  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	  re   rf   rj    r"  z_full.<locals>.inner_fnc                   r5  r   r  r^  r6  re   rf   rj    r"  r   c                   s    g S r   re   r^  )value_loaderre   rf   rj    s   rk  )rm   r   rF  r  r	  r   r   r   r   rZ  r?   r   rk   )
fill_valuerw   r   r   rj  re   )r   r	  r7  rf   _full  s    r9  c                 K  s   t t|| fi |S r   create_tensor_liketensor_constructor)r   r8  r   re   re   rf   r    s   r  c                   s    d d d d dd d fdd
}|S )NF)namesr   rw   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  rY  re   rZ  r   re   re   rf   r     r[  z5tensor_constructor.<locals>.inner.<locals>.<listcomp>)r   r~   r&  rF   r'  r   rm   rk   r   Sizer  r9  )r=  r   rw   r]  r  r  r   r   r8  re   rf   rw    s   	"z!tensor_constructor.<locals>.innerre   )r8  rw  re   r@  rf   r<    s   r<  )r=  r   r]  rw   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]  rw   r  )	r   rF   r   rm   rk   r   r~   r?  empty_strided)r=  r   r]  rw   r  r  r   re   re   rf   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   rw   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   r~   r&  r   r   rp   rk   r   )r   r   rw   r]  r  r  r   creation_fnre   rf   _constant_like  s   

z*create_tensor_like.<locals>._constant_likere   )rD  rE  re   rC  rf   r;    s   
r;  c                 C  s   t t| S r   r:  r@  re   re   rf   r  %  r"  r  c                   s   d d d d d fdd
}|S )NrA  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  rY  re   )r   r(  r   re   re   rf   r   8  r[  z7new_constant.<locals>._new_constant.<locals>.<listcomp>)rm   rk   r   r   r~   r&  r   r   rp   r9  rF   r   r   r   r]  rw   r  r@  re   rf   _new_constant/  s   z#new_constant.<locals>._new_constantre   )r8  rG  re   r@  rf   new_constant.  s   rH  rA  c                C  s8   |d u r|   }|d u r|  }t|d ||t||dS NrA  r   rp   rB  rF   rF  re   re   rf   	new_empty>  s   rK  c                C  s  t | ttfs	J t |tttd fsJ t|d tjfv d|  t|p)t }|p1t	dj
}t|}td||| d}|  |jj}tj|jdgt|  d|_t |tjs\J dd | D } |rldd |D ntj| }tj||| ||p{d	d
|_|S )Nr  rn  r   )r8  rw   r   r   )ri  c                 S  rY  re   rZ  r   re   re   rf   r   _  r[  z!empty_strided.<locals>.<listcomp>c                 S  rY  re   rZ  r   re   re   rf   r   a  r[  F)rw   r   r   rb  	is_pinned)rm   rk   r   r   r   r~   r&  r   r'  rl  rw   rF   r9  r  ro   dataclassesreplacer   r1   r  rw  rx  ra  r]  )r   rb  r   r]  rw   r  	pointwiser  re   re   rf   rB  N  s0   
rB  c                C  s8   |d u r|   }|d u r|  }t||||t||dS rI  rJ  )r   r   rb  r   r]  rw   r  re   re   rf   new_empty_stridedo  s   rP  c                 C  s2   dd |D }t tt||jd}tj| |S )Nc                 S  r   re   r   r   re   re   rf   r     r   z copy_strided.<locals>.<listcomp>)r  )sortedr  r   __getitem__r1   r  r  )r   rb  r  re   re   rf   copy_strided  s   rS  c                 K  s*   | dd usJ dt|| fi |S )Nr   z(dtype should be handled by decomposition)rq  r<  )r   r8  r   re   re   rf   full  s   rT  c                   s   t | tsJ | dkrt| | S |  tdk}t|  | |r/t| 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  )rk   rQ   indirect_indexingr   )r  
gather_idxr   index_loaderr   r#  re   rf   rz     s   zgather.<locals>.fnrk  )rm   rB   	get_numelrK  r   r   r9  r@  rZ  r?   r   rp   r   )r   r   rI  sparse_gradrE  rz   re   rW  rf   gather  s$   	r[  c                   s   |rt tjj| ||||S |rJ t| tsJ t|ts J dt| v s*J |  |  t	|
 | 
 g |
 dd   fdd}tj|  |  |dS )Nr   r.   c                   s\   t | t ksJ |  d  | d  }t|d gg | d   }|S )Nz != r   )r   rQ   rU  )r  	var_index
weight_idxindices_loaderindices_ndimr%  weight_loaderweight_sizere   rf   rz     s   "
zembedding.<locals>.fnrk  )r  r  	embeddingr  rm   rB   r   r   rZ  r   r   r?   r   rp   )weightr  padding_idxscale_grad_by_freqsparserz   re   r^  rf   rc    s(   

rc  c                   s   t dd  D sJ ddd  D  tdd  D r"tddd t D }t|d	ks5J d
d gt  }t|t fdd|D  D ]\}}| |krXtd|||< qJ||fS )Nc                 s  s4    | ]}|d ur|  tjtjtjtjfv V  qd S r   )r   r~   rp  r  r[   r  r  re   re   rf   r     s    z.check_and_broadcast_indices.<locals>.<genexpr>z)indices must be int64, byte or bool. Got c                 S  s   g | ]
}|d ur|  qS r   r   r  re   re   rf   r         z/check_and_broadcast_indices.<locals>.<listcomp>c                 s  s,    | ]}|d ur|  tjtjfv V  qd S r   )r   r~   r[   r  r  re   re   rf   r     s    "zFallback for bool indicesc                 S  r   re   r   r   re   re   rf   r     r   r   z"requires at least 1 non-None indexc                   r  re   re   r  r  re   rf   r     r   z.Fallback when indices is on a different device)r1  r2  r   rl   r   r  r  rp   )r  rw   
valid_idxsnew_indicesrt   r   re   rj  rf   check_and_broadcast_indices  s"   
$
rm  c	              
     s   dt 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   re   )r   rt   r  r  re   rf   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   rW  r.   r   wrap_neg)r   r  rq   rQ   rU  )	r  r  	new_indexfirst_tensor_indexstart_offsetnext_idxrt   loaderr   )
r   indexed_sizer  indices_loadersnon_consecutive_tensorsoutput_sizetensor_indicestensor_sizerp  r#  re   rf   rz     s>   

z*index_output_size_and_inner_fn.<locals>.fn)r  pairwiserl   r   )r  r  rz  r{  rw  rv  r#  r   rp  previouscurrentrr  rz   re   )r   rv  r  rw  rx  ry  rz  r{  rp  r#  r  rf   index_output_size_and_inner_fn  s$    


"r  c                 C  s,   t | ||\}}}tj|  |  ||dS r  )index_impl_helperr?   r   rp   r   )r   r  r   ry  rj  r  re   re   rf   
index_impl8  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   rY  r  re   re   rf   r   I  r  z%index_impl_helper.<locals>.<listcomp>c                   s    g | ]} | d ur| qS r   re   r  )r  r  re   rf   r   Q  r  z0index is out of bounds for dimension with size 0c                   r  re   re   r  rn  re   rf   r   U  r   ro  c                   s    | S r   re   r  )index_inner_fnr#  re   rf   rj  b  r"  z#index_impl_helper.<locals>.inner_fn)rm   rk   r   rZ  rm  rp   r   r   r  
IndexErrorr  )
r   r  r   rp  rz  rw  r{  rv  ry  rj  re   )r  r  r#  r  rf   r  C  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  rI  r  r   r  re   re   rf   rI  h  s   c                 C  s   t | |ddS )NFr  )r  r  re   re   rf   _unsafe_indext  r  r  c                 C  s   t t| |||dddS )NTFr   may_realizeindex_put_impl_r  r   r  r  
accumulatere   re   rf   	index_put     r  c                 C  s   t t| |||dddS )NFr  r  r  re   re   rf   _unsafe_index_put  r  r  c                 C  sB   |  |   krt||   }|rt| |}t| t|d || S r  )rp   r  r   r  r  )r  r  r	  r  re   re   rf   index_put_as_masked_fill  s
   
r  c           	      C  s   ddl m} ttjtjjjj	}tjj}t
js5|d ur5||r5d}|jdd  }r1| d| }|tj_t|| ||| | S )Nr.   ),_fx_node_is_input_dependent_cudagraph_unsafezLindex_put_ fallback with boolean indexing is not compatible with CUDA graphsstack_trace Found from : 
 )utilsr  r   r  
index_put_rR   r]   r^   ra   _overloadnamer/   graph_partitionrp  rq  disable_cudagraphs_reasonr1   IndexPutFallback)	r  r  r  r  r  rS  r  r   r  re   re   rf   index_put_fallback  s   r  c                 C  s   t | |||dddS )NTr  r  r  r  r  r  re   re   rf   r       r  c                 C  s   t | |||dddS )NFTr  r  r  re   re   rf   _unsafe_index_put_  r  r  c              
     s  |rdd  t | | v rt fdd|D s|  | dkrWt|dkrW|d  tj	tj
fv rW|d }tt| t|  D ]}t|d}qGt| |g||S t rbt| |||S |D ]}|d ur}| tj	tj
fv r}t| |||  S qd|  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||}|  }|d usJ t j||  | |||r.dnd d}t jd t | |d}t j!"||_#t j!$| |	dkrTt| g } | S )Nc                 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u S dS )Nr  F)rm   rB   ro   r1   r5   r^  r   r  r   r  ra   r~   rQ   r  randpermr  )indicere   re   rf   indice_slice_from_randperm  s   
z3index_put_impl_.<locals>.indice_slice_from_randpermc                 3  r  r   re   )r   r  )r  re   rf   r     r  z"index_put_impl_.<locals>.<genexpr>r.   r   rW  c                 S  r  r   rY  r  re   re   rf   r     r  z#index_put_impl_.<locals>.<listcomp>c                   r  re   re   r  rn  re   rf   r     r   r  
atomic_addrw   r   rj  ri  output_indexerscatter_moder   r]  ro   )%r1   try_get_namer  r1  r  rY  r   r   r~   r[   r  r  r   r  r  $are_deterministic_algorithms_enabledr  rK   r  r
  rm  rp   r   rm   rB   rk   r  r@  ScatterrZ  r  MutationLayoutSHOULDREMOVErR   r]   rA  r   rB  )r  r  r  r  r   r  rz  r  rI  x_ndimrz  rw  r{  rv  expected_vals_sizerj  rw   scatterr  re   )r  r  rf   r    s   




r  r  c                   sT   t | |ddd\}}  |   fdd}tj|  |  ||dS )NFro  c                   sB   j tjkrt tj}n }t| fddS )Nc                     s    S r   re   re   )_unsafe_index_fnr  self_loaderre   rf   r-  I  rm  z8_unsafe_masked_index.<locals>.inner_fn.<locals>.<lambda>)r   r~   r[   rQ   r
  rt  )r  mask_valr  fillrz  mask_loaderr  r  rf   rj  D  s   z&_unsafe_masked_index.<locals>.inner_fnrk  )r  rZ  r?   r   rp   r   )r  rz  r  r  ri  r  rj  re   r  rf   _unsafe_masked_index<  s   r  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  r4  r  r  r  re   rf   r   W  s    (z7_unsafe_masked_index_put_accumulate.<locals>.<listcomp>T)r  )r  r   r  r   r  )r   rz  r  r  masked_valueclamped_indicesre   r  rf   #_unsafe_masked_index_put_accumulateS  s   
r  c                 C  s   t |t || S r   )rQ   r  r  r  minmaxre   re   rf   r5  `     r5  c                 C  r  r   )r  rH  copy_)r  r  r   rb  r7  rC  output_viewre   re   rf   as_strided_scattere  r	  r  c                 K  s   t t| |||fi |S r   )scatter_r  )r   r   rI  r  r   re   re   rf   r  m  s   r  r  include_selfrS  r  r  c             	   C  sf   t |t}t| || ttj|r| nt||r| jnd|r1t	j
| ||||||d |S d S )Nznot implr  )rm   rB   rP   r   r	   r~   r   r   rp   r1   ScatterFallback)rS  r  r   rI  r  r  r  src_is_tensorre   re   rf   scatter_fallbackr  s(   

		r  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   multiplyr  r   r  r  prod)
r   r  r  rR   r]   r^   ra   r  r  scatter_reduce_)r  r   rI  r  r  rS  fallback_resultre   re   rf   r    s   r  c                 C  s   t t| |||S r   )scatter_add_r  r   r   rI  r  re   re   rf   scatter_add  r  r  c                 C  s   t | |||dS )Nr  )r  r  re   re   rf   r    r  r  c                 K  s   t t| ||||fi |S r   )r  r  )r   r   rI  r  reduction_typer   re   re   rf   scatter_reduce  s   r  )r  c             	     s8  |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 }
 }|d usJ |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)Nr  r  meanamaxaminr.   r  zKaten.scatter_reduce_.two is not the unique overload of aten.scatter_reduce_r  r   r   c                   sD     }t|}t| }tj| |dkrdn|  dd| < |S )Nr   r.   F)rp  )r   r   rk   rQ   rU  )r  r  r  indirect_idx)r   rX  r  re   rf   r    s   
z'scatter_reduce_.<locals>.output_indexerc                   s   r| S t   S r   rQ   r  r   r  )r  r  r  re   rf   rz     s   zscatter_reduce_.<locals>.fnc                 S  s   | dkrdS | d u sJ d S )Nr  r  re   r  re   re   rf   backend_reduce_str  s   z+scatter_reduce_.<locals>.backend_reduce_strc                   s   t d  S r  r  r^  )r  re   rf   r-    r  z!scatter_reduce_.<locals>.<lambda>r  r  )r   r  r  r   rm   r    r  r  r  rB   r   r   r   r  rY  r9  r  rZ  rp   r1   r  r  r  rR   r]   rA  r   rB  )r  r   rI  r  r  r  r  r  r  rz   r  rw   zero_outr  r  re   )r   rX  r  r  r  rf   r    s   







r  scales_xtuple[Optional[float], ...]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   re   r  r  re   re   rf   r   5  r   z&upsample_nearestnd.<locals>.<listcomp>c                 S  s   g | ]\}}|| qS re   re   )r   rt   ore   re   rf   r   :  r  r  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  )
rQ   ro  r~   r  r   r  rl  r
  r  rU  )r   r  r   )r  re   rf   scale_fn?  s   z$upsample_nearestnd.<locals>.scale_fnc                   sB   |  d  }| d   }g |fddt | D S )Nc                   s   g | ]\}}} |||qS re   re   )r   rt   r   r   )r  re   rf   r   N  r   z2upsample_nearestnd.<locals>.fn.<locals>.<listcomp>)r  )r  r   rB  )i_sizes
inv_scalesr  r  r#  re   rf   rz   J  s
    zupsample_nearestnd.<locals>.fnrk  )
realize_hintrZ  r   r   r  rl   r?   r   rp   r   )
r   ry  r  r  r  batcho_sizesrt   r  rz   re   )r  r  r  r  r  r#  rf   upsample_nearestnd*  s(   
r  Optional[float]c                 C  s   t | ||fddS )Nr.   r  r  r   ry  r  re   re   rf   upsample_nearest1dY  r  r  c                 C  s   t | ||fdddS )Nr.   Tr  r  r  r  re   re   rf   _upsample_nearest_exact1d^  r  r  scales_hscales_wc                 C  s   t | |||fddS )Nr,   r  r  r   ry  r  r  re   re   rf   upsample_nearest2dc  s   r  c                 C  s   t | |||fdddS )Nr,   Tr  r  r  re   re   rf   _upsample_nearest_exact2dj  s   r  scales_dc                 C  s   t | ||||fddS )Nr   r  r  r   ry  r  r  r  re   re   rf   upsample_nearest3dq  s   r  c                 C  s   t | ||||fdddS )Nr   Tr  r  r  re   re   rf   _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   re   rf   r     r  z$_create_constants.<locals>.<genexpr>)r   )r   ru   re   r   rf   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  )rk   r   )r  r   r  r  r#  re   rf   ru    s
   zrev.<locals>.loaderrk  )rZ  r   r?   r   rp   r   )r   r  ru  re   r  rf   rev  s   r  paddingSequence[int]r8  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	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)rR   r]   r^   r   r_   r   ra   r  mmr  addmm)r^   r_   re   re   rf   _padding_can_be_fused  s   
z6inplace_constant_pad_nd.<locals>._padding_can_be_fusedNr   r,   r.   r   r   F)r   rI  rJ  r5  inductorinplace_padding)r   r   r  rm   r1   rB   ro   r   r  r/   can_inplace_pad_graph_inputInputBufferr   freeze_layoutr`  rb  r   rR   r]   buffer_to_padded_sizerH  rE  rX  fill_r   )r   r  r8  r  r  r]  r  npadstride0rowsizebufnamepadded_size	resized_xsliced_xre   re   rf   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  r0  r   Nre   r   r  re   re   rf   r     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   re   re   )rI  r#  re   rf   r-  (  r  z/constant_pad_nd.<locals>.mask.<locals>.<lambda>)	r  rq   range_mask_lowrange_mask_highr&  r  rQ   rs  rt  )rI  rz  r  r  r  r  )boundsr8  
mask_sizesr  r#  r^  rf   rz     s   "zconstant_pad_nd.<locals>.maskc                   sZ   t | d  }t| d   D ]\}\}}|||  qt|t| ks)J |S r   )rk   r  rq   r   )rI  rq  r  r  _high)bounds_precomprz  r  re   rf   	offset_fn*  s
   z"constant_pad_nd.<locals>.offset_fnrk  )r   r1  r  r/   r  r	  r   rk   r;  r  rq   rR   r]   r   lookup_precomputed_sizer   r@  r   r   rZ  r?   r   rp   )r   r  r8  rr   r  lhry  r  r  r   r  re   )r  r  r8  rz  r  r  r#  rf   r    s:   *

r  rt   
sympy.ExprUnion[sympy.Expr, int]c                 C  s&   t t | tjt t|tjS r   )rQ   rq  ro  r~   rp  r   r(  )rt   r  re   re   rf   r  :  s   r  c                 C  s    t t | tjt |tjS r   )rQ   rr  ro  r~   rp  )rt   r  re   re   rf   r  A  s   r  c                 C  s   t t| |t| |S r   )rQ   rs  r  r  )rt   r  r  re   re   rf   
range_maskH  s   r  r  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 re   )r  r  )r  ih	padding_hre   rf   r   ^  s   . z=constant_boundary_condition.<locals>.load.<locals>.<listcomp>c                     s   t  dg S )Nr   )constant_boundary_conditionre   )r   r  pad_fill_valueprefixr   re   rf   r-  c  s    
z;constant_boundary_condition.<locals>.load.<locals>.<lambda>c                     s   g  S r   re   re   )r  r  r#  re   rf   r-  i  r  )r&  r  rQ   rs  r  rt  )rI  rz  r   r8  r  r  r  r  r   r#  )r  r  rf   r]  W  s   	z)constant_boundary_condition.<locals>.load)r   rZ  )r   r8  r  r  r   r]  re   r  rf   r  O  s
   r  dilationc          	      C  s&  |d u rdgt | }t| d||   || || d   || d  || }|rt| d||   || || d   d|| d   || }tjj|d ||  |  ||  dkrt|d8 }tjjd|||  |  ||   tjj|| dkrtjj|| d}||fS |}||fS )Nr.   r,   r   F)r   r(   rR   r]   r   	size_hintr  r?  )	r   rt   kernel_sizerb  r  	ceil_moder!  x_outx_altre   re   rf   pooling_sizeo  s4   .
*$r'  c                C  s    t | |} ttj| }|dkS )N   )rL   r&  r  r  rl  )r#  n_dimwindow_sizere   re   rf   %should_fallback_max_pool_with_indices  s   
r+  assert_fallbackc                C  s   |dkr	dg| }|dkrdg| }|s|}t ||}t ||}t ||}t ||}t| ts1J t||ks9J t||ksAJ t||ksIJ t||ksQJ t|  |d |d fv saJ t||d}|d urq||ksqJ |||||fS )Nr   r.   r,   r)  )rL   rm   rB   r   r   r+  )r   r#  rb  r  r!  r)  r-  use_fallbackre   re   rf   max_pool_checks  s(   





 r0  c             
     sN  |    | jd   }| j d  t fddtD  \} | j}	|	tju r2dn|	jr9tdnt	|	j
}
t|t| }tsXt sXtdd D r`t| |
dn|  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 )Nc                   s&   g | ]}t | | d qS )r   r'  r  )r$  dhwr!  r#  r  rb  re   rf   r     s    z*_max_pool_with_offsets.<locals>.<listcomp>F-infc                 s  s    | ]}|d kV  qdS r  re   r  re   re   rf   r     r  z)_max_pool_with_offsets.<locals>.<genexpr>r  c                   sJ   | d   }|  d    fddt D }g ||S )Nc                   4   g | ]} | |  | |   |  qS re   re   r  bhr!  r  reduction_idxrb  re   rf   r         &z<_max_pool_with_offsets.<locals>.fn_inner.<locals>.<listcomp>r  )r  r7  r  r  )r!  r)  r  rb  r#  r6  r7  rf   fn_inner  s   z(_max_pool_with_offsets.<locals>.fn_innerr  r  
input_noderw   	dst_dtyper  rj  ri  reduction_rangesargmax)r  r  r  r  r   r~   r[   r  rF  r  r  rk   r2  r  rZ  r@   r   rp   rp  rm   ro   r  )r   r#  rb  r  r!  r$  r)  r  dhw_outr   	min_valuer%  r;  ri  offsetsre   )r$  r2  r!  r#  r)  r  rb  r#  rf   _max_pool_with_offsets  sV   
	
"	

rD  c           
   
   C  s   t |}t| |||||dd\}}}}}tjdd t| ||||||d\}}	|t|	tjfW  d    S 1 s:w   Y  d S )NFr,  r(  unroll_reductions_thresholdr.  )r   r0  r/   r   rD  r
  r~   r  )
r   r#  rb  r  r!  r$  r)  r  ri  rC  re   re   rf   !_low_memory_max_pool_with_offsets  s,   	

	$rG  rC  r#  "Sequence[Union[int, torch.SymInt]]
input_sizeincrements_to_indexxCallable[[Sequence[Union[int, torch.SymInt]], Sequence[Union[int, torch.SymInt]]], torch._inductor.virtualized.OpsValue]c                   sZ   t |  tttj fdd}tj	| 
 tj||  d}|S )Nc                   sJ   | }t |}t|} | |}t t| d  tjS r   )rQ   rU  r0   _flattened_index_to_ndro  _flatten_indexr~   rp  )r  rE  offset_sympyr7  idhwrJ  rI  r#  r)  offsets_loaderr*  re   rf   offsets_to_indices+  s   
z4_pool_offsets_to_indices.<locals>.offsets_to_indicesrk  )r   rZ  r   r&  r&  r  r  rl  r?   r   rp   r~   rp  r   )rC  r#  rI  rJ  rR  r  re   rP  rf   _pool_offsets_to_indices  s   		rS  c                   s(   t | fdd}t| |||S )Nc                   s,   |  d    fddt D S )Nc                   r4  re   re   r  r5  re   rf   r   H  r8  zX_low_memory_max_pool_offsets_to_indices.<locals>.increments_to_index.<locals>.<listcomp>r9  r  r7  r!  r)  r  rb  r:  rf   rJ  F  s   zD_low_memory_max_pool_offsets_to_indices.<locals>.increments_to_index)r   rS  )rC  r#  rI  rb  r  r!  rJ  re   rU  rf   '_low_memory_max_pool_offsets_to_indices=  s
   rV  c              	   C  s^   t | |||||d\}}}}}t| ||||||d\}}	t|	|| j| d  |||}
||
fS )Nr.  )r0  rD  rV  r  )r   r#  rb  r  r!  r$  r)  r  rr   rC  r  re   re   rf   _max_pool_with_indicesR  s   	
	rW  c              	   C     t | |||||ddS Nr,   r.  rW  r   r#  rb  r  r!  r$  re   re   rf   max_pool2d_with_indicesp     	r\  c              	   C  rX  Nr   r.  rZ  r[  re   re   rf   max_pool3d_with_indices  r]  r_  c                   s`  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rt |jjtr|jj}	|	 }
|
d usgJ 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rt| ||||S |  	
fdd}tj|  |  ||d}|r.t	j|S |S )Nr   r.   r,   r  )rw   r   r   r  c                 s  s    | ]}|d kV  qdS r  re   r  re   re   rf   r     r  z3max_pool2d_with_indices_backward.<locals>.<genexpr>c              	   3  @    | ]}t t|d  t d t| d   d   dV  qdS r   r.   Nr  r(   r   r  r#  rb  re   rf   r     
    .
c              	   3  @    | ]}t t|d  t dt| d   d   d V  qdS r.   r   Nrb  r   wrd  re   rf   r     re  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  rW  rn  )rQ   ro  r~   r  r(   r  r  r  r  r   rU  r  r  r  r  rs  rr  )r  r  r  ri  
index_testphstartpwstartphendpwendgradientph_pw_phpw
grad_indexindex_actual	grad_partr   rz  grad_loaderh_window_sizer_  indices_sizer#  r  pooled_heightpooled_widthrb  w_window_sizewidthre   rf   rz     sl     


#z,max_pool2d_with_indices_backward.<locals>.fnrk  )rm   rB   r   r   r  r  ro   r?   rp   r1   r  rw  r   decide_layoutrF  r2  )fallback_max_pool2d_with_indices_backwardrZ  rk   r  r  r   r  r  )grad_outputr   r#  rb  r  r!  r$  r  	gO_stridero   rw   x_bufferx_strideis_channels_last_batch_heightr  r%  r*  rz   rr   re   rx  rf    max_pool2d_with_indices_backward  s   	

 ;r  rn  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   re   re   )h_start_indexr  iwr  w_start_indexr#  re   rf   r-  :  ri  z3pad_adaptive_loader.<locals>.load.<locals>.<lambda>)rQ   rs  rr  ro  r~   rp  rt  )r  
incrementsstart_indicesend_indicesh_end_indexw_end_indexrz  pad_valr#  )r  r  r  r  r  rf   r]  (  s$   z!pad_adaptive_loader.<locals>.loadrY  )r   r  r]  re   r  rf   pad_adaptive_loader%  s   r  c           
      C  sL   t j| ||d}t j|||d}t j| ||d}t j|||d}	||||	fS )N)out_diminp_dim)r&  r7  )
rL  rM  h_inw_inh_outw_outr  r  r  r  re   re   rf    compute_indices_adaptive_poolingA  s
   r  c                   sD   |\}}|\}}	t | |||||	\  fdd}
|
S )Nc                   s   | ^ }}}|} |}|}|}d }	t td td D ]\}
}|||
|g||g||g}|	d u r>|}	q&||	}	q&|	S r  )r  productr  )r  ru  r  r6  bwr  r  r  r  ri  r  r  r  h_end_index_fnh_start_index_fnkernel_maxes
pooling_fnw_end_index_fnw_start_index_fnre   rf   rz   Z  s"   $z _adaptive_pooling_fn.<locals>.fnr  )rL  rM  r  in_sizes	out_sizesr  r  r  r  r  rz   re   r  rf   _adaptive_pooling_fnK  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  r  r  rQ   ro  r~   rp  r  gt)r  ru  r  r6  r  r  r  r  r  maxvalmaxindexr  r  r  rI  r  r  r  r  r  r  r  re   rf   rz     s0   $z)_adaptive_pooling_fn_with_idx.<locals>.fnr  )
rL  rM  r  r  r  r  r  r  r  rz   re   r  rf   _adaptive_pooling_fn_with_idxt  s   #r  c                   s    tjkrtdttsJ t|dksJ    ^ }}}t	j
j|}t	j
j|}|\}}||krD||krDtS |dksL|dkr_g |||}t|   dS || dkrz|| dkrzt||t||g}t|S t|| d |}	t|| d |}
t|||g }  }|	|
 }|dkrt|S dd }d	d
 }t|||	|
g||g||gtjd tt fdd}tj |||d}|S )Nz0'adaptive_avg_pool2d' not implemented for 'Long'r,   r   r  r.   r(  c                 S     t | | |S r   r(   rI  r  r  re   re   rf   rL       z)_adaptive_avg_pool2d.<locals>.start_indexc                 S     t | d | | d |S r  r  r  re   re   rf   rM    r  z'_adaptive_avg_pool2d.<locals>.end_indexrL  rM  r  r  r  r  c                   s   t  | t | S r   )rQ   truedivr  r  fn_sumones_loaderr   re   rf   rz     s   z _adaptive_avg_pool2d.<locals>.fnrk  )r   r~   rp  rf  rm   rB   r   r  r   rR   r]   r   r  r  r$  rp   r(   
avg_pool2drE   rk   fallback_adaptive_avg_pool2dr  rQ   r   r  	ones_liker?   r   )r   ry  r  r  r  r  r  o_sizer#  h_kernel_maxw_kernel_maxr%  r   r*  rL  rM  rz   rvre   r  rf   _adaptive_avg_pool2d  sV   

	r  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  r  r   r  r  re   re   rf   rL    r  z(adaptive_max_pool2d.<locals>.start_indexc                 S  r  r  r  r  re   re   rf   rM  !  r  z&adaptive_max_pool2d.<locals>.end_indexr  c                       | t tdS Nr3  r  rF  r  )inner_func_max_valr   re   rf   inner_fn_max_val6  r  z-adaptive_max_pool2d.<locals>.inner_fn_max_valc                   r  r  r  r  )inner_func_max_idxr   re   rf   inner_fn_max_idx9  r  z-adaptive_max_pool2d.<locals>.inner_fn_max_idxrk  )r   r~   rp  rf  rm   rB   r   r  r   rR   r]   r   r  r$  rp   
ValueErrorrE   rk   fallback_adaptive_max_pool2dr  rQ   r  r  r?   r   )r   ry  r  r  r  r  r  r  r  r  r%  r   r*  rL  rM  r  r  r  rire   )r  r  r   rf   adaptive_max_pool2d  sp   
		r  c                   s<            fdd}|S )Nc           
        sX    }t|dkr@t| dkrd| d d   g}n.t| dkr4| d | d d   g}nddd   g}ng | d   }t| }t tj}td tj}tt|tj	t|tj	}t
t|dd|}t|| | t||  }t|tj}t||}	tt
|	||tS )Nr   r.   r   r,   )r   r   rQ   ro  r   r~   rp  r  r
  float64r  r  r  rr  rU  r   r&  )
r  rt   samples_shapesamplei_exprdiffout_sz_exprrS  seq_irz  r   in_sz	kernel_szndimsout_szsamplessamples_loaderre   rf   r]  Q  s&     z)_fractional_pooling_offsets.<locals>.loadrY  )r  r  r  r  r   r  r]  re   r  rf   _fractional_pooling_offsetsK  s   #r  c                 C     t | |||ddS rY  _fractional_max_poolr   r#  ry  random_samplesre   re   rf   fractional_max_pool2dw  r  r  c                 C  r  r^  r  r  re   re   rf   fractional_max_pool3d|  r  r  c                   s\  |    | jd   | j d  }tjdd fddtD  |  fdd} fddt|t }|  }tj	d	| | 
 ||||d
}	tj	d| | 
 tj|||d
}
t|	tsrJ |	t|	jjtr}|	  t|
tsJ |
t|
jjtr|
  t|
| j}|	|fW  d    S 1 sw   Y  d S )Nr(  rE  c              
     s    g | ]}t  |d qS ))r  r  r  r  r  r   )r  r  )inp_dhwr#  r)  ry  r  re   rf   r     s    	z(_fractional_max_pool.<locals>.<listcomp>c                   s$   | d   }g | | |S r   re   r  r7  r  )rJ  r)  r#  re   rf   r;    s   z&_fractional_max_pool.<locals>.fn_innerc                   s8   | d   |  d    fddt D S )Nc                   s&   g | ]}|  | |  qS re   re   r  )bdhwdhw_index_fnr  r7  re   rf   r     s    zE_fractional_max_pool.<locals>.increments_to_index.<locals>.<listcomp>r9  rT  )r  r)  )r  r  r7  rf   rJ    s
   z1_fractional_max_pool.<locals>.increments_to_indexr  r<  r@  )r  r  r/   r   r  rZ  rk   r   r@   r   rp   r~   rp  rm   rB   ro   r  rS  )r   r#  ry  r  r)  r  r;  r%  r   ri  rC  r  re   )r  rJ  r  r#  r)  ry  r  r#  rf   r    sT   "	


$r  c                   s       ^ }}}tjj|}tjj|}|^ }}}	|| dkr;||	 dkr;tt||t||	gddS t||}
t||	}dd fdd}t	||
|g||g||	gt
jd  fd	d
}tj  |t|d}|S )Nr   r.   )divisor_overridec                 S  s   t | | t|S r   )r'   r   r&  r  re   re   rf   rL    r  z0upsample_nearest2d_backward.<locals>.start_indexc                   s    | d ||S r  re   r  )rL  re   rf   rM    r  z.upsample_nearest2d_backward.<locals>.end_indexr  c                   s    | t S r   )r  r  )r  r   re   rf   rz     r  z'upsample_nearest2d_backward.<locals>.fnrk  )r  r   rR   r]   r   r  r  r(   rE   r  rQ   r   r?   r   rp   r   rk   )r   ry  rI  r  r  r  inp_hinp_wout_hout_wr  r  rM  rz   r  re   )r  rL  r   rf   upsample_nearest2d_backward  s:   

	r  re   c              
   C     t | ||||||ddS )Nr,   r  _avg_poolndr   r#  rb  r  r$  count_include_padr  re   re   rf   r       
r  c              
   C  r  )Nr   r  r  r  re   re   rf   
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d
}fdd}ttj}|dkrtdd tD rtd  }||  |S |dkrtjddnt }|  }|d usJ | tjd| |||||d}W d    n	1 sw   Y  t|jdrt|jjtr|  |r|r,|r$|n|}t||}nfdd}t j|  |||d}t||}t!||S )Nr   r.   r,   c              	     s"   g | ]}t | | qS re   r1  r  )r$  r  r#  r  rb  re   rf   r   G  s    z_avg_poolnd.<locals>.<listcomp>rn  r  TF)r   r   c                   sL   | d   }|  d   | fddt D g |S )Nc                   s,   g | ]} | |  |  |  qS re   re   r  )r6  r  r  rb  re   rf   r   a  s   , z1_avg_poolnd.<locals>.fn_inner.<locals>.<listcomp>r9  r  )r   r  rb  r#  )r6  r  rf   r;  ]  s
   z_avg_poolnd.<locals>.fn_innerr(  c                 s  s*    | ]\}}t jjt||V  qd S r   )rR   r]   r   statically_known_truer   Ne)r   r  r   re   re   rf   r   f  
    
z_avg_poolnd.<locals>.<genexpr>rE  r  r<  ro   c                   s   |  d  }g }t D ]<}|| |  |  }t||  | |  } s;t|d}t|| }t|| tj}|| qt	
tj|S r  )r  r   MinMaxrQ   ro  r~   r  rq   r&  r  rl  )r  r6  divide_factorsrt   hstarthendfactor)r  r   r  r#  r  rb  re   rf   fn_count  s    z_avg_poolnd.<locals>.fn_countrk  )"rL   rm   rB   r   r   r  r  r  r2  r  rZ  rk   r   r   r   rK  r&  r  r  rl  fallbacks_avg_poolndr/   r   
contextlibnullcontextrp   r@   r   r  ro   r  div_primr?   r
  )r   r#  rb  r  r$  r  r  r   r  r  
ceil_modeshad_paddingr%  r   output_dtyper;  r*  fallbackcontextrw   r  divisorri  r  divide_factorre   )r$  r  r   r  r#  r  rb  r#  rf   r  *  s   




 

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`  ra  rb  rc  rd  re   rf   r     re  z&avg_pool2d_backward.<locals>.<genexpr>c              	   3  rf  rg  rb  rh  rd  re   rf   r     re  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.   )
rQ   r  r~   r  r  rl  r  r   ro  r  )rs  rt  stride_hstride_wpad_hpad_wkernel_hkernel_wr  wstartr  wendr  )heightr#  r  rb  r  re   rf   !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  rn  )rQ   ro  r(   r~   r  r  r  r  r  r   r  rU  r  rs  rr  r  r  )r  r  r  ri  rl  rm  rn  ro  rp  rq  rr  rs  rt  r  partrz  )r  r  r  ry  rz  r  r#  r  r|  r}  rb  r~  re   rf   rz     sv     
	


*zavg_pool2d_backward.<locals>.fnrk  )rm   rB   r   r   r  r'  rZ  rk   r   r  r  fallback_avg_pool2d_backwardr?   r   rp   )r  r   r#  rb  r  r$  r  r  r  _h_out
ceil_mode1_w_out
ceil_mode2r%  r   r*  rz   r  re   )r  r  r  ry  rz  r  r  r#  r  r|  r}  rb  r~  r  rf   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 ra  )r  r  )rt   r#  rb  re   rf   r     s
    *
z0avg_pool3d_backward.<locals>.<genexpr>.<genexpr>r,   N)r  r  )r   rd  )rt   rf   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   rQ   r  r~   r  r   re   re   rf   r     r  zQavg_pool3d_backward.<locals>.compute_pool_size_without_padding.<locals>.<genexpr>c                 s  r  r   r  r  re   re   rf   r     r  c                 s  r  r   r  r  re   re   rf   r     r  c                 s  s*    | ]\}}}t t |||V  qd S r   )rQ   r  rl  )r   r  r   padre   re   rf   r     s
    
c              
   s  s>    | ]\}}}}t t ||t t |tj|V  qd S r   )rQ   r  r   ro  r~   r  )r   rI  r  r   r  re   re   rf   r     s    

c                 s  &    | ]}t |t d tjV  qdS r
  rQ   r  r  r~   r  )r   rI  re   re   rf   r     
    
c                 s  *    | ]\}}t |t |tjV  qd S r   rQ   r  ro  r~   r  )r   rJ  r   re   re   rf   r     r  )r  rQ   rl  r  )pdrs  rt  stride_dr  r	  pad_dr
  r  kernel_dr  r  dstartr  r  dendr  r  r  )depthr  r#  r  rb  r  re   rf   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   re   )r   r  r  re   re   rf   r     r  z2avg_pool3d_backward.<locals>.fn.<locals>.<genexpr>c                 s  s2    | ]\}}}t t|| | |tjV  qd S r   rQ   ro  r(   r~   r  )r   r  r  r   re   re   rf   r     s
    
c                 s  s,    | ]\}}t t||d  tjV  qdS r  r)  )r   r  r   re   re   rf   r     s
    
c                 s  r  r
  r  )r   pstartre   re   rf   r     r  c                 s  r   r   r!  )r   pend
pooled_dimre   re   rf   r     r  c                 s  r   r   )rQ   r   r  r~   r  )r   r*  p_re   re   rf   r     r  r   r.   r,   Fr  rn  )r  r  rQ   r  rU  r  r  r  r~   r  rs  rr  r  r  r   )r  r  r  r  ri  pdstartrl  rm  pdendrn  ro  rp  pd_rq  rr  r"  rs  rt  r  r  rz  )r  r  d_window_sizer  ry  rz  r  r#  r  pooled_depthr|  r}  rb  r~  re   rf   rz     s    	

8zavg_pool3d_backward.<locals>.fnrk  )rm   rB   r   r   r  r'  rZ  r2  rk   r   r  fallback_avg_pool3d_backwardr?   r   rp   )r  r   r#  rb  r  r$  r  r  r  _d_outceil_mode_dr  ceil_mode_hr  ceil_mode_wr  r%  r   r*  rz   r  re   )r  r  r1  r(  r  ry  rz  r  r  r#  r  r2  r|  r}  rb  r~  r  rf   avg_pool3d_backward^  sf   &%Wr8  c                 C  s   |   }t|tr|g}n|stt|}t|dkr*t|dv s(J d| g S t|}tt|D ]5}|| dk rL||  t|rHt|nd7  < d||   krZt|k sin t|dkrg|| dksiJ q4tt|t|ksxJ d|S )Nr   )re   r  r  zinvalid axis: r.   zreduction axis not unique)r   rm   r   r  r   r   rk   r&   )r   r  r   rt   re   re   rf   _validate_reduction_axis&  s    
 :r9  )r  c                  sl  |d ur	t | |} |  tt t| |}g }g g g ttD ]}||v r7| |  q$| ||  q$d|dv rtdkrt| rt	| j
tr]dn#t	| j
tjsst	| j
tjrt	| j
j
tjr|  }| p|   fdd}	rt}
D ]}tjj|
|< qn|}
|   t|  |p|  |  |	|
dS )NF)r@  argminr.   Tc           	        s   t |t ks
J rt  t ksJ  fddD  t  t ks)J d gt  t |  }tt t|D ]\}}|||< q@|}rwdd |D }|d }tdt |D ]}||  ||  }qa|t|tjfS |S )Nc                   r  re   re   r  r^  re   rf   r   a  r   z9_make_reduction_inner.<locals>.loader.<locals>.<listcomp>c                 S  rY  re   rZ  r  re   re   rf   r   l  r[  r   r.   )	r   r  r  r  r  rQ   ro  r~   rp  )	rI  reduction_indexrq  r  varr	  rindex
linear_idxrt   inner_loaderkeepdimskept_idxreduced_idxreduced_sizesshould_compute_logical_indexr   r^  rf   ru  ]  s$   
z%_make_reduction_inner.<locals>.loader)rw   r>  r  rj  ri  r?  )r
  r   r&   r   r9  r  r   rq   r:   rm   ro   r>   r1   r   r   BufferrD  is_transposedis_contiguousrk   r   r<  r=  rZ  r(  rp   r   )r   r  rA  r   rO  r  
kept_sizesrt   r]  ru  r%  re   r?  rf   _make_reduction_inner8  sZ   



rJ  r  rS   c                   s   dd d fdd}|S )NFr   c                  sD   t | ||| d}tjd| d|}t|jjtr |  |S )N)r  rA  r   rO  r  )r  r=  re   )rJ  r@   r   rm   ro   r  )r   r  rA  r   r   ri  rO  r  re   rf   rw    s   zmake_reduction.<locals>.innerr   re   )r  rO  rw  re   rK  rf   make_reduction  s   rL  c                C  sB   |d ur	t | |} t| |}t|  |  f|  f|  |dS )N)rw   dtypes	inner_fnsr   r  )r
  r9  r(  rp   r   rZ  r   )r   r  r   re   re   rf   _make_scan_inner  s   

rO  r   c                  s   |d ur	t | |} |   t| |}|  }|tjtjfv r$t | tj} t| ||}t	 fdd|D }t
j||  |  d}t|t| }t t|||S )Nc                 3  r  r   re   r  r   re   rf   r     r  zmean.<locals>.<genexpr>rH  )r
  r   r9  r   r~   ro  rn  rF  sum_rO   r1   r8   rp   r7   r   rk   div)r   r  keepdimr   r  
sum_resultdenomre   rP  rf   r    s   

r  c           
        s   |d u rd}|    t| |}t| |dd}|r|  tt| |}t|||}t fdd|D }|r>t	|| d}t
j||  |  d}t|t|  }t||}	|s^|	fS |rb|nt||}|	|fS )Nr.   T)rS  c                 3  r  r   re   r  rP  re   rf   r     r  z var_mean_sum_.<locals>.<genexpr>r   rH  )r   r9  r  r  squarer  rQ  rO   r   r  r1   r8   r   rp   r7   r   rk   rR  r  )
r   r  
correctionrS  return_meanx_meandiffsrT  rU  x_varre   rP  rf   var_mean_sum_  s&   

r\  c                 C  sV   t | |}t| ||d d d}|d }t|d }t|tjo*t|tjk o*t|dkS )Nr  rA  r   rO  ri  r?  r.   )	r9  rJ  rO   rm   r   r(  r   r/   rF  )r   r  rS  r   ri  reduction_numelre   re   rf   use_two_step_variance  s   


r_  c                  s    d u rd t | ||d d d}|d}|d |d tjjd|fd|  d|\}}}	|  |  |  t| |}t	fdd	|D d
d  fdd}
t
|
|}|rj|  ||fS |fS )Nr.   r]  rj  r>  r  welford_reduce)rN  r  r   c                 3  r  r   re   r  rP  re   rf   r     r  z$var_mean_welford_.<locals>.<genexpr>c                 S  s4   t | tjr| jstt| tj|S t	| |S r   )
rm   r   r   	is_numberrQ   r
  ro  r~   rp  r  r  re   re   rf   get_constant_or_index_expr   s   z5var_mean_welford_.<locals>.get_constant_or_index_exprc                   s4    }}t d}| t |||  S r  )rQ   r  r  )ro   cNzero)rW  r   rb  rnumelre   rf   r    s   

z#var_mean_welford_.<locals>.scale_fnre   )rJ  r  r1   WelfordReductionr   r   r  r   r9  rO   rx  )r   r  rW  rS  rX  r   ru  r  m2r  r  r<  re   )rW  r   rb  rf  r   rf   var_mean_welford_  s6   




ri  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   r  rW  rS  rX  )r  rS  c                 3  s    | ]
}t | d dV  qdS )Fr  Nr  r   r  re   rf   r   #  r  z#var_mean_helper_.<locals>.<genexpr>r   re   )r   r   r
  r(  r_  r\  ri  r   )r   r  rW  rS  rX  r   r   rC  re   r  rf   var_mean_helper_  s    	rj  )rW  rS  c                C  r  )NFr  rW  rS  rX  rj  r   r  rW  rS  re   re   rf   var_'     
rn  c                C  r  )NTrk  rl  rm  re   re   rf   var_mean.  ro  rp  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_recursiverQ   r  r  rl  )r   rD  r   ri  re   re   rf   rq  5  s   rq  c                 C     t | |S r   )rQ   r  r  rB  re   re   rf   
pow_nativeD  r  rt  )r   c                   sR  t tr 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oOd  k oGdk n  pO|oOdk}|rm 	  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   )rm   r1   rB   r   r   re   re   rf   r   \  r  zpow.<locals>.<genexpr>i    r   c                   s   t |   S r   )rq  r   r  r  rB  ru  re   rf   rz   f  r  zpow.<locals>.fnrk  r,   )rm   rF  r   r  r   sqrtr  rL  r   rZ  r?   r   rp   r   r   r    r  r   exp2fallback_pow_scalarfallback_pow_tensor_scalarfallback_pow_tensor_tensorrt  )r  rB  r   is_integer_powembed_exponentrz   re   rv  rf   r  R  s@   
"







r  c                 C  s   t | tr	| j}n| }t |tr|j}t |tjs>tj|  |  |	 | 
 d}t |ttfs3J |j}t |tjs>J t |tjr]| s]| s]t |jtjs]|  |j|_| S tjj|||d | S )Nrk  unsafe_alias)rm   rB   ro   r1   r   r?   r   rp   r   rZ  r   r5   r;   is_input_bufferis_module_buffer	NopKernelr  r  realize_into)changedr  r  changed_datarc   re   re   rf   r    s:   

r  c                 C  s   t | t| |S r   )r  r  )r   r8  re   re   rf   r    r  r  c                 C  @   | |u r| S t ||  }t||  }t||  }t| |S r   r  rp   r
  r   r@  r   r  )r  r  r  re   re   rf   r       
r  c                 C  rr  r   )rQ   floordivrs  re   re   rf   r    r  r  c                 C  rr  r   )rQ   truncdivrs  re   re   rf   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  rR  r  r  )r  rB  rounding_modeboth_integerboth_booleanre   re   rf   div_mode  s   
r  c                 C  s8   t | ot |}|rt| |S ttjj}t|| |S r   )r   logical_andr=   r  rl  r  rx  )r  rB  	both_boolrz   re   re   rf   rl    s
   
rl  r  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)rm   r1   r;   get_constant_valuero   r5   r^  r  Loopsr~   	_inductorops_handlerExtractConstantsHandlerrp   rR   set_ops_handlerr   objectrw  rj  inner_fn_argsvirtualizedOpsValuer	  )r   r1  rr   re   re   rf   r    s*   
 r  c                 C  s   t dd | |fD }|rt| |S t| }d ur:|  jdkr:|jdkr0t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   re   re   rf   r     r  zdiv_prim.<locals>.<genexpr>r   r   infr  c                  W  r  r   )rQ   r  r  re   re   rf   rz     r  zdiv_prim.<locals>.fn)r1  r  r  rp   r   r	  mathcopysignrF  rl  rx  )r  rB  is_integralr  r  rz   re   re   rf   r    s   



r  c                 C  s    t | |ftjd\} }t| |S rR  )rQ  r   INT_TO_FLOATr  rs  re   re   rf   rR    s   


rR  c                 C  s4   t | pt| }|rdd }ndd }t|| |S )Nc                 S  rr  r   )rQ   modrs  re   re   rf   rz   &  r"  zfmod.<locals>.fnc                 S  rr  r   )rQ   fmodrs  re   re   rf   rz   +  r"  )r   r   rx  )r  rB  r  rz   re   re   rf   r     s
   
r  c                C  B   t |  st|  r|d u rtj}td|d}|| |||dS )Nr  r  r   r   r   r   r~   rp  rL  r   r  rA  r   rz   re   re   rf   rQ  1     

rQ  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   rW  Tr  c                 S     | \}|\}t ||fS r   )rQ   r   a_tupleb_tupler  rB  re   re   rf   
combine_fnO     zcumsum.<locals>.combine_fnr  r   r  r   r   re   )r   r   r   r~   rp  r   r   r
  rO  r1   Scanr   fallback_cumsumr   r  r   r  r   ri  re   re   rf   cumsumC      

r  c                 C  r  )Nr   r  Tr  c                 S  r  r   )rQ   rl  r  re   re   rf   r  g  r  zcumprod.<locals>.combine_fnr  r  r  re   )r   r   r   r~   rp  r   r   r
  rO  r1   r  r   fallback_cumprodr  re   re   rf   cumprod[  r  r  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   )rQ   r  r  r  r  log1pexp)r  r  r  rB  min_vmax_vrz  re   re   rf   log_add_exp_helperu  s   $z(logcumsumexp.<locals>.log_add_exp_helperr   r  r  r  r  re   )	r   r   r   r  rO  r1   r  r   fallback_logcumsumexp)r   r   r  r   r   ri  re   re   rf   logcumsumexps  s   r  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< | 
  fd	d
f|d< tjjdi |d|i\}}|d u rXt|  dS ||fS )Nr   r  r   r@  Fr   arg_break_ties_leftr  rM  c                      t |   tjS r   rQ   ro  r~   rp  r  r  re   rf   r-        zcummax.<locals>.<lambda>rN  r  r  re   )r   r   r  r  r~   rp  r   r1   get_reduction_combine_fnrO  rZ  r  r   fallback_cummaxr   r  r   r  r   r  r  re   r  rf   cummax      
r  c                   r  )Nr   r  r   r:  Fr  r  rM  c                   r  r   r  r  r  re   rf   r-    r  zcummin.<locals>.<lambda>rN  r  r  re   )r   r   r  r  r~   rp  r   r1   r  rO  rZ  r  r   fallback_cumminr  re   r  rf   cummin  r  r  c                C  r  )Nr  r  r   r  r  re   re   rf   r    r  r  c                 C  s   t | tj} td| ||dS )Nr2  r  rA  )r
  r~   r[   rL  r   r   rS  re   re   rf   
reduce_any  s   r  c                 C  2   |d urt | ||dt| ||dfS t | d |dS Nr  )reduce_amaxreduce_argmaxr  re   re   rf   
reduce_max  
   r  c                 C  r  r  )reduce_aminreduce_argminr  re   re   rf   
reduce_min  r  r  xor_sumr  r  r@  r  r:  
logical_or)rt  rh  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.   r  )rI  rK  r   rw   r  )rw   rM  rN  r   r  r  r  )r   rp   r   r   r  r9  r~   rp  rR   r]   r   statically_known_ltr  int16r  sort_fallbackr  r  r@  r1   Sortr   r   rZ  r
  )
r   r  r   r  r  rw   r   r  
view_shaper  re   re   rf   sort_stable  s>   



	r  c                 C  s   t | d||dS )NFr  )r  )r   r   r  re   re   rf   sort  r  r  c                 C  s   t | |tj|dS )Nr   r   rv  )r  r   r  )r`   r   rv  re   re   rf   register_pointwise_numeric  s   r  torch._ops.OpOverloadPacketc                 C  s   t | j t| tjdS rR  )rN   r  r  r   r  r  re   re   rf    register_pointwise_numeric_ldf64%  s
   
r  r  r	  c                  st   t jstS t| ||tjd |  | |  jo!tj	j
  fdd}tj|   ||  dS )a  
    Computes self + value * tensor1 * tensor2 using FMA for better precision.

    Matches eager CUDA kernel order: self + value * (tensor1 * tensor2)
    This is computed as: fma(value, tensor1 * tensor2, self)

    Note: FMA is only used for floating-point types on non-AMD GPUs. For integer types,
    we fall back to regular arithmetic since FMA doesn't support integers.

    For floating-point types, we use mul_rn (round-to-nearest multiplication)
    to force rounding of the product before the FMA. This prevents Triton's
    compiler from fusing the multiplication with the FMA, matching eager's
    rounding behavior.

    When emulate_precision_casts is False, we return NotImplemented to use the
    decomposition instead.
    r  c                   s   | }| }| }dkrrt |||S r"t ||}nt ||}ttjr5t  }nt  }rDt |||S t 	|t ||S r  )
rQ   fmamul_rnrl  rm   r   r   ro  r  r   )r  self_valt1_valt2_valt1_times_t2
value_exprr   r  	t1_loader	t2_loaderuse_fmar	  re   rf   rj  \  s   zaddcmul.<locals>.inner_fnrk  )r/   re  NotImplementedr   r   rK  rZ  r  r~   versionhipr?   r   rp   r   )r  tensor1tensor2r	  rj  re   r  rf   rW  8  s&   rW  c                   s^   t jstS ttjjjdkptjjjt	v pt
 }tt| ||} fdd}t|t| ||S )z
    Foreach version of addcmul with scalar value parameter.
    Uses foreach_group_loop for consistent grouping behavior.

    When emulate_precision_casts is False, we return NotImplemented to use the
    decomposition instead.
    r   c                   s   t | d iS )Nr	  )rW  r  r  re   rf   r{    r  z)_foreach_addcmul_scalar.<locals>.apply_fn)r/   re  r  r   rR   r]   r^   r_   ra   r|  rg   ry   r  r}  )r  r  r  r	  r~  r  r{  re   r  rf   _foreach_addcmul_scalar  s   r  logical_not)rh  )r   r   rO  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+  re   rf   make_triton_fallback  s   
z6_get_pointwise_overrides.<locals>.make_triton_fallback)	r  r   r   rm   r~   r   r   r   r   )nsr   r`   r   olnameri  re   r+  rf   _get_pointwise_overrides  s   
r  r  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  rq   r  )ru   r   resultsmut_resultsr   ri  outplace_opre   rf   rz   F  s
   z$register_foreach_inplace.<locals>.fn)rY   r|  r   r+  )aten_opoutplace_aten_opr  rz   re   r  rf   register_foreach_inplaceB  s   
r
  c                   s   t | d d fdd}|S )Nr  c                    s.    | i |}t || d  }t| d |S r  )r
  r   r  )ru   r   ri  r  re   rf   rz   i  s   zregister_inplace.<locals>.fn)r8  )r  r  rz   re   r  rf   register_inplaceh  s   
r  c                 C  r2  r   re   r  re   re   rf   sym_constrain_range  r  r  c                 C  *   t jjjd }t|tjr|jjS t	|S Nr  
rR   r]   r^   rp  rm   r~   r  rc   r  r   r  r   r  re   re   rf   rO       rO  c                 C  r  r  r  r  re   re   rf   
sym_stride  r  r  c                 C     |   S r   )rY  r  re   re   rf   	sym_numel  r  r  c                 C  r  r   )r   Addr  re   re   rf   sym_sum  r  r  c                 O  r  )NzHelpful for debuggingr   )r  ru   r   re   re   rf   foobar  r  r  c                 C  s   |    t| S r   )r  r  r   re   re   rf   _realize  s   r  c                 C  s   |    t| | | S r   )r  r1   ResizeStorageBytes)variabler%  re   re   rf   resize_storage_bytes_  s   r  c                 C  s"   |    |   tt| |S r   )r  rB   r   r1   SetSourceTensorKernel)r  source_tensorre   re   rf   set__source_tensor  s   r  c                 C  r  r   r  )r  r  re   re   rf   
fsdp_copy_  r  r   c          	        sv  t | tsJ t |ttfsJ |d u rtj}|tjkr#td| |tjkr0t	|dks0J |tj
kr=t	|dks=J |  |  }|  }t | jtjrV| j | _t rvtjjjrvt|rhtdnt|rst|jndndtjjdrt|||dS t | gd	g}|!  tj"#||}t$||||%  fd
d}t&j'|||t|d}|S )Nzunsupported memory format: r   r   nanTrn  r   r  r.   c                   sH   |  t  tj}t tj}t ||}t | fddS )Nc                     s
    gS r   re   re   )
flat_indexflat_loaderre   rf   r-    s   
 z*resize.<locals>.inner_fn.<locals>.<lambda>)rQ   ro  r~   rp  rr  rt  )r  flat_index_exprlimitrz  r#  	old_numelout_indexeruninitialized_val)r"  rf   rj    s
   zresize.<locals>.inner_fnrk  )(rm   rB   rk   r   r~   contiguous_formatpreserve_formatrf  channels_lastr   channels_last_3drY  r   rM  ro   r1   r5   r^  r  r  deterministicfill_uninitialized_memoryr   rF  r   r  r  rR   r]   r   r  rT  rH  rZ  rw   stride_ordered_for_memory_formatra  ry  r?   r   )	r   r   r  r   rw   x_flat
out_striderj  rr   re   r&  rf   resize  sT   



	r3  )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 re   r   )r   r  r  re   re   rf   r  6  r  z'triton_kernel_wrap_.<locals>.<dictcomp>)*torch._higher_order_ops.triton_kernel_wrapr5  get_constant_argsr1   UserDefinedTritonKernelr  )r6  constant_args_idxr7  r8  r   r5  constant_argsre   re   rf   triton_kernel_wrap_$  s   	

r?  3list[Union[ir.TensorBox, ir.ShapeAsConstantBuffer]]c                 C  sj   t dd | g|D r$d}tjjjdd  }r | d| }|tj_tj	| |||}t
ttj	|S )Nc                 s  rT  r   rU  r   re   re   rf   r   ?  rG  zcond.<locals>.<genexpr>z"control flow operator: torch.cond.r  r  )r2  rR   r]   r^   rp  rq  r  r1   Conditionalr   rk   maprB   )predtrue_fnfalse_fnoperandsr   r  ri  re   re   rf   r   9  s   c                 C  s   t js&tdd || D r&d}tjjjdd  }r"| d| }|tj_t	j
| ||||}t|ts7J ttt	j
j|S )Nc                 s  rT  r   rU  r   re   re   rf   r   M  s
    
zwhile_loop.<locals>.<genexpr>z(control flow operator: torch.while_loop.r  r  )r/   r  r2  rR   r]   r^   rp  rq  r  r1   	WhileLoopr   rm   r   rk   rB  _maybe_wrap_as_tensor_box)cond_fnbody_fncarried_inputsadditional_inputsstack_outputr   r  ri  re   re   rf   
while_loopI  s   
rN  )rM  subgraph_fnir.Subgraph
identifierc                 G  s$   t jj| g|R  }tttj|S r   )r1   InvokeSubgraphr   rk   rB  rB   )rO  rQ  rF  ri  re   re   rf   invoke_subgraphb  s   rS  r  torch.fx.GraphModulec              	   C  s   d}t | jjD ]X\}}|jdkr"|tjjvsJ || tjj|< q|jdkr;tj|\}}tjj	
tj|||}q|tjjvsCJ tjj}z|tj_tj|tjj|< W |tj_q|tj_w |du ritd|S )a  Process nodes from a FX graph by executing them through V.graph.

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

    NrB  rC  zNo output node found in graph)rl   r]   nodesr`   rR   envfetch_args_kwargs_from_envr~   rC  InterpreterrC  r^   run_noderf  )r  ru   rC  rt   rc   output_argsr   saved_current_nodere   re   rf   process_subgraph_nodesh  s$   	

r\  )control_depsc                 G  s   g }| D ]}t |tsq|  ||  qtjjj}d}t	|| t	|ks+J t	tjj
}t	|jjjddt	|ksAJ t|jt|}|durO| sQJ tjj
|d D ]}	|D ]}
|	j}|dushJ tjj| |
 q]qY|S )aS  
    Lower control_deps_op by ensuring dependencies are realized and tracking them.

    The control_deps_op HOP makes dependencies explicit in the graph. During lowering:
    1. Realize all additional dependencies to ensure they're computed
    2. Execute the target operation normally
    3. Track the dependencies for the scheduler
    r,   rB  r  N)rm   r9   r  rq   r  rR   r]   r^   ru   r   
operationsr  
find_nodesr\  rk   operation_nameadditional_buffer_depsr   )additional_depsrO  ru   	dep_namesdeporiginal_args
arg_offsetoperation_lenrC  r`   dep_nameop_namere   re   rf   control_deps_op_lowering  s(   

 	rj  )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_optionsrB  rC  )rR   r]   r^   rp  rq  rl   r  rU  r`   rV  rW  r  r  r  r  codegen_low_precisionlow_precision_codegen_opsr   r  invoke_quant_opsr~   rC  rX  rC  rY  )
rO  rk  rF  rC  rl  rt   rc   ru   r   r  re   re   rf   invoke_quant_tracer  s"   

rp  r  rL  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   rp   r   )rr  re   rf   r     s    z$associative_scan.<locals>.<listcomp>c                   s    g t | t |R  S r   )r>  rE  )lhsrhs)lowered_combine_fnre   rf   wrapped_combine_fn  s
   z,associative_scan.<locals>.wrapped_combine_fnr  c                 s  r  r   rh  r   re   re   rf   r     r  z#associative_scan.<locals>.<genexpr>rM  c                 s  r  r   rY  r   re   re   rf   r     r  rN  F)r  can_fallback_to_atenz/Unable to generate code for associative_scan opre   )r  rr  rs  r   rf  r  r  rO  r   r1   r  r   )r  xsrL  rs  subgraph_inputsrw  r   ri  re   )rr  rv  rf   associative_scan  s,   


r{  c                 C  r2  r   re   )tokensre   re   rf   _sink_tokens  r  r}  c                   C  r2  r   re   re   re   re   rf   _make_token  r  r~  c              
   O  sb  ddl m}m} ||}|du rP|tjjju rPddlm}m	} |
 }	|	rP|	jtjjj}
|
rPt|
|s7J |
|d }|rPt|dksJJ dtt|}ttjj}|tv rmt| |i |}ttdd | nd	d
 }t|tjj|g|R i |}ttjj|d dksJ d| d|rtjj|}tjj|d D ]}dd |_|r|  }tjj!| "|   q|tjj|< zdd }t|||f\}}||||}W n6 t#y } z)t$|}t%&d|| t|t't(fr| g|R W  Y d}~S | |fW  Y d}~S d}~ww t|j)dkr| |fS t|j)dkr+| |fS | g|R S )z
    We lower the operator directly, and then we add StarDep dependencies to all
    the newly created nodes in the graph.
    r   )_get_effect_get_schemaN)InvokeSubgraphCacheTracingContextr.   zMultiple effects NYIc                 S  r  r   )r  r  re   re   rf   r-  *  r  zwith_effects.<locals>.<lambda>c                 S  r*  r   r+  r   re   re   rf   r,  -  r  z"with_effects.<locals>.wrap_tensorszCNo operation nodes were generated when lowering effectful operator .c                   S  s   dS r   re   re   re   re   rf   r-  =  s    c              	   S  sp   t | tjr
|  S t | tr6z| j}t|dr$t|jdr'|j W S W | S W | S  tt	fy5   Y | S w | S )Nro   get_example)
rm   r1   TorchBindObject	get_valuerB   ro   r  r  AttributeErrorr   )r  re  re   re   rf   convert_ir_to_valueH  s$   
z)with_effects.<locals>.convert_ir_to_valuez5Failed to get schema for %s: %s. Assuming list output)*torch._higher_order_ops.effectsr  r  r~   rQ   higher_orderrS  torch._guardsr  r  try_gethop_dispatch_set_cache	get_cacherm   get_effectsr   rL  r
  rR   r]   r^  rW   r>  r  rB   r-  r1   r.  r   effectful_opsrq  has_side_effectsr  additional_star_depsr   rf  r   rc  rd  r   rk   returns)tokenr`   ru   r   r  r  effect_typer  r  tracing_ctxinvoke_subgraph_cacheeffectsrg  ri  r,  prev_effect_buffernew_opri  r  schema_argsschema_kwargsschemae	error_msgre   re   rf   with_effects	  sr   



r  )register_comm_loweringsregister_symm_mem_loweringsc                 C  s   t | |dddd}|d }tjjt|}tjjdi |d|d\}}|dkrEtjj	|t
jrEt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)r  r^  r.   r,   )r=  
num_outputreduction_hintz
            Online softmax is disabled on the fly since Inductor decides to
            split the reduction. Cut an issue to PyTorch if this is an
            important use case and you want to speed it up with online
            softmax.
            )rA  re   )rJ  rR   r]   r   simplifyrO   r1   r@   
num_splitsstatically_known_geqr/   rF  r<   r   r4  r5  textwrapdedentr  rW   r  r  r  rQ  )r   r   r   r?  rf  hint	num_split
max_tensor
sum_tensorr  r  xsumre   re   rf   prepare_softmax_onlinew  s:   



r  c                   C  s   t j ot j dkS )z.Check if we're on SM100+ hardware (Blackwell).)r   r   )r~   cudais_availableget_device_capabilityre   re   re   rf   _is_sm100_or_later  s   r  c                 C  s   t  std|  }|tjtjtjfvrtd| |tjkr't| tj} t	j
tjddtjddd}t|| }t|tjS )z
    Lowering for cvt_e8m0_rceil. Uses PTX cvt.rp.satfinite.ue8m0x2.f32 on SM100+.

    The PTX instruction takes 2 float32 and outputs 2 e8m0 packed in uint16.
    Currently we pass 0.0 as the second input and only use the low byte result.
    zFcvt_e8m0_rceil requires SM100+ (Blackwell) for PTX instruction supportzAcvt_e8m0_rceil requires float32, float16, or bfloat16 input, got z)cvt.rp.satfinite.ue8m0x2.f32 $0, 0.0, $1;z=h,rTr.   )asmconstraintsr   is_purepack)r  r   r   r~   r  ro  rn  r  r
  r&  r7  rQ   inline_asm_elementwiseuint16rx  r  )r   r   rz   ri  re   re   rf   cvt_e8m0_rceil_lowering  s*   
r  r/  )quantized_lowerings)mkldnn_lowerings)jagged_loweringsc              	   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)	rm   r~   r   r   rW   rq  r8  r  r  )r`   old_handlerre   re   rf   force_fallback  s   

r  r  )rh   ri   rZ   rj   )rz   r{   rZ   r|   )r   r   rZ   r   )r   r[   r   r   rZ   r   )rz   r   rZ   r   )rz   r   r   r   rZ   r   )r   r   rZ   r   )r   r   rZ   r   )r   r   rZ   r   )ru   r   r   r   r   r[   rZ   r   )r`   r   r   r   rZ   r[   )r   rB   rw   r   rZ   rB   )ru   r   r   r   r   r[   r   r   r   r[   rZ   r   )r   r"  r#  r{   rZ   r{   )
r#  r{   r   r[   r   r   r   r[   r,  rV   )r   r   rZ   r6  r3  )NNNFN)F)r   rB   r   r   r  r[   )r   rB   r   r   )r   rB   rw   r   )r   rB   r  rB   r   )r   rB   r  r)  rZ   rB   )r   r   r+  r.   Tr  )r  rB   r  rB   r  rB   r  r   r  r   r  r   r   r   rZ   rB   )r  rB   r  rB   r  rB   r  r   r  r   r  r   r   r   r  r  rZ   rB   )r  rB   r  rF  r  r   r  r   r  r   r   r   rZ   rB   )r  rB   r  rF  r  r   r  r   r  r   r   r   r  r  rZ   rB   )r  rB   r  rB   r  rB   r  r   r  r   r   r   rZ   rB   )r  rB   r  rB   r  rB   r  r   r  r   r   r   r  r  rZ   rB   )r   r   r.   )rE  r   r  r   r  r   r  )T)rx   r7  )rc   rA  )NTF)rw   r   )r   r  rz  rB   r  r   rE  r   )
r  r   r  r   r   r  rz  rB   rE  r   )r  rB   rZ   r  )r  rB   rZ   r  )r  rB   r  rB   r  r[   r  r[   r  r  r  r  rZ   rB   )r  rB   r  rB   r  r[   r  r[   )r   r   rI  r   )r   NNr.   )NNN)rW  FF)r   r   )rS  r"  r   r   r  r  r  r[   )r   r   r  r  )r   r   r  r[   )r,   F)r  r  r  r   r  r[   )r  r  )r  r  r  r  )r  r  r  r  r  r  )r   rB   r  r  r8  rF  rZ   r  )rt   r  r  r  )rt   r  r  r  )rt   r  r  r  r  r  )Nr  N)
rC  rB   r#  rH  rI  rH  rJ  rK  rZ   rB   )Nr   r.   F)rn  )NNNN)re   r   FTN)r  rS   r   )r   r  rZ   r  )rW  F)r`   r  )r.   )rZ   r@  )rO  rP  rQ  r   )r  rT  ru   r   )rO  rP  )r  rP  rL  rq  r  (  
__future__r   r  rM  r&  r  loggingr  r  r[  r  r4  collectionsr   collections.abcr   r   r   r   typingr   r	   r
   r   r   r   r   typing_extensionsr   unittest.mockr   r   r~   $torch.ao.quantization.fx._decomposedtorch.fxtorch.utils._pytreer  _pytreer>  torch._dynamo.utilsr   (torch._higher_order_ops.associative_scanr   r:  r   "torch._library.fake_class_registryr   torch._library.utilsr   torch._prims_commonr   r   r   r   r   r   r   r   r   r   r    torch.fx.experimental.sym_noder!   r"   r8  r#   r$   r%   torch.utils._ordered_setr&   torch.utils._sympy.functionsr'   r(   r)   r*   r+   _dynamo.utilsr-    r/   r0   r1   r2   decompositionr3   r4   r5   r6   r7   r8   r9   r:   r;   r<   r=   r>   r?   r@   rA   rB   rC   rD   rE   rF   rG   rH   rI   rJ   rK   rL   rM   rN   rO   rP   r  rQ   rR   r  rS   rT   rU   FALLBACK_ALLOW_LIST	getLoggerr  rc  rW   __annotations__rX   r   r   r.  r  tr_c10dr  r   _higher_order_opsr  rb   r|  rY   quantized_decomposedrg   ry   r   r   r   r   r   rH  rj  r  r  bmmconvolutionconvolution_backwardr\  r_  r  r  r  r  _int_mmr  r  r  r  rp  ro  r  r  	complex32	complex64r[   rn  r   r   r   r   r   r   r   r   r!  r+  r5  rK  r8  rE  rQ  rx  r  r}  r
  r  r  r  r  r   r  r  
device_putr  r  r  r  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*  slicerX  rh  r  quantize_per_channelr  r  r   r  _functional_assert_asyncr  dequantize_per_channelr  quantize_per_tensorr  dequantize_per_tensorr  rl  r  r  r  r  r  r  r  r  r  r  r#  r  r%  r9  r)  r  cacher6  r<  r@  rP  rj  rn  rngprimsr}  r~  r  	bernoullir  r  r  	lru_cacher  r  ro  r  r  r  r  r  r  randintstreamsrecord_event
wait_eventforce_stride_orderr  rz  r  r  r  lookup_seedr  randomr  r  r  r  r  r  r  	NO_OPMATHr  r  r   r  r  r   r   r  _adaptive_avg_pool3dadaptive_max_pool3d*_scaled_dot_product_attention_math_for_mpsuniformexponential_pdist_forwardsoft_margin_loss_backward_fused_rms_normxpur  embedding_dense_backwardmtia_is_compilednative_layer_norm_cdist_forward_cdist_backward
_trilinearsegment_reduce_segment_reduce_backwardhistc	histogrambin_ct_histogramdd_bin_edges_histogramdd_from_bin_ctsaddbmm_addmm_activation_grouped_mm
_cudnn_rnn_cudnn_rnn_backward
miopen_rnnmiopen_rnn_backward_embedding_bag_embedding_bag_forward_only_embedding_bag_backward*_embedding_bag_per_sample_weights_backward_fused_moving_avg_obs_fq_helper*_fused_moving_avg_obs_fq_helper_functional max_pool3d_with_indices_backward_adaptive_avg_pool2d_backward_adaptive_avg_pool3d_backwardadaptive_max_pool2d_backwardadaptive_max_pool3d_backwardfractional_max_pool2d_backwardfractional_max_pool3d_backwardreplication_pad1d_backwardreplication_pad2d_backwardupsample_linear1d_backwardupsample_bicubic2d_backwardupsample_trilinear3d_backwardgrid_sampler_2d_backward_pdist_backwardr  r  kthvaluetopkr  median	nanmedianr  resize_
resize_as__linalg_detlinalg_householder_productlinalg_inv_exlinalg_ldl_factor_exlinalg_ldl_solve	linalg_lulinalg_lu_factor_exlinalg_lu_solvelinalg_matrix_exp	linalg_qr_linalg_slogdet_linalg_solve_exlinalg_solve_triangular_linalg_svd	lu_unpackormqr_linalg_check_errorslinalg_pinvatol_rtol_tensor_linalg_eightriangular_solvelinalg_cholesky_excholesky_inversecholesky_solvegeqrf_fft_r2cnonzerogcd_thnn_fused_lstm_cell_prims	rng_primsrun_and_save_rng_staterun_with_rng_stategraphsafe_run_with_rng_statemasked_scattermasked_scatter_backwardrL  angle_efficientzerotensor(_sparse_coo_tensor_with_dims_and_tensors	to_sparse
_to_sparser   r=  r  r  #_scaled_dot_product_flash_attention	quantized,_scaled_dot_product_flash_attention_backward#_scaled_dot_product_cudnn_attention,_scaled_dot_product_cudnn_attention_backward+_scaled_dot_product_flash_attention_for_cpu4_scaled_dot_product_flash_attention_for_cpu_backward0_scaled_dot_product_fused_attention_overrideable9_scaled_dot_product_fused_attention_overrideable_backward_flash_attention_forward_flash_attention_backward_efficient_attention_forward_efficient_attention_backwardindex_reducerepeat_interleave_weight_norm_interface_backwardr  r  r
  rM  r  r  r  r  scalar_tensorr*  
LongTensorr,  r1  r3  r4  r9  r  r<  r$  r;  r  r  r  
zeros_likerH  rK  rB  rP  rS  rT  r[  rc  rm  r  r  r  rI  r  r  r  r  r  r  r  r  r  fallback__unsafe_masked_indexr  ,fallback__unsafe_masked_index_put_accumulater5  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r	  r  r  r  r  r  r'  r+  r0  rD  rG  rS  rV  rW  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  
avg_pool1dr  r  r  r8  r3  r9  rJ  rL  rO  r  r\  r_  ri  rj  r<  rn  rp  rq  rt  r  Tensor_Tensorr{  rY  ry  Tensor_Scalarrz  r  r  r  r  r  rR  r  rl  r  r  true_divider  r  rQ  r  r  r  r  r  r  r  r  r  r  r  r2  r  r  r  r  r  r  r  r  r  r  r@  r  r:  r  r   r  r  r  r  rsqrtr  rx  expm1relur'  rw  rV  r  rW  r  rX  cossinabsbitwise_andbitwise_left_shiftbitwise_not
bitwise_orbitwise_right_shiftbitwise_xorlgammaerfspecial_erfr  tantanhr  r  r  logical_xorr  r  	clamp_min	clamp_maxnegr  	remaindersignsignbit	_neg_viewlerr  rq  r  r  necoshsinhacosacoshasinasinhatan2atanatanhr  erfcerfinvhypotlog10log2	nextaftercodegen.commonr  r  r  r   r`   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_copyforeach_copyr
  _foreach_add__foreach_mul__foreach_div__foreach_copy_r  add_bitwise_and_bitwise_left_shift_bitwise_not_bitwise_or_bitwise_right_shift_bitwise_xor_mul_div_Tensor_modelogical_and_logical_not_logical_or_logical_xor_sub_relu_sigmoid___and__
__lshift____or__
__rshift____xor____iand____ilshift____ior____irshift____ixor__r  rO  r   r  r  r  methodfuncr  r  _inductor_testr  r  r  r  set_source_Tensorr  fsdpr   r3  *torch._higher_order_ops.auto_functionalizer4  r?  r  r   rN  while_loop_stack_outputr7  rS  r\  .torch._inductor.fx_passes.control_dependenciesr]  rj  invoke_quantrp  r{  r}  r~  r  comm_loweringr  r  r  r  cvt_e8m0_rceilr  r0  r  register_quantized_opsregister_woq_mm_opsr  register_onednn_fusion_opsr  register_jagged_opscontextmanagerr  re   re   re   rf   <module>   s  $4H8


		

R77Y' 
,
-















4
xE2



2$$-/|6@
	
"(:&




		a:		 



#D
8



,
	


 $!#I%
w



"x/


_7 K"
 
)5


E


Q
,
D1 
	 &

	 HR
+


0
%
	 











'	I















@$.
#f4&
