
    wi1             -       fb   d dl Z d dlmZ d dlmZ d dlmZmZmZm	Z	m
Z
 d dlZd dlmZ d dlmZmZmZ d dlmZmZmZmZ d dlmZ d dlmZmZ d d	lmZmZmZmZm Z m!Z!m"Z"m#Z#m$Z$m%Z% d d
l&m'Z'm(Z(m)Z)m*Z*m+Z+ d dl,m-Z-m.Z. d dl/m0Z1 ej2        j3        Z3ej4        5                    ddd          Z6d Z7de fdZ8d Z9d Z: e7e3j;        e3j<        g           e+            dddej=        ddfd                        Z> e7e3j?        j@        e3j?        jA        g           e+            d                         ZB e7e3jC        j@        e3jC        jA        g           e+            ddd                        ZC e7e3jD                   e+            d                         ZD e7e3jE        j@        e3jE        jA        e3jF        j@        e3jF        jA        g           e+dd          d                         ZG e7e3jH        j@        e3jH        jA        g           e+            d                         ZHd  ZI e7e3jJ        j@        e3jJ        jA        g           e+            d!                         ZK e7e3jL        j@        e3jL        jA        g           e+            d"                         ZM e7e3jN        jO                  dd#d$            ZP e7e3jN        j@                  ejQ        dddd%d&            ZR e7e3jS        j@        e3jS        jA        g           e+            ejQ        dddd%d'                        ZT e7e3jS        jU        e3jS        jV        g           e+            ejQ        dddd%d(                        ZW e7e3jX        j@        e3jX        jA        g           e+            ddddd%d)                        ZY e7e3jZ        j@        e3jZ        jA        g           e+            d*                         Z[ e7e3j\        j@                  dd+            Z]d, Z^ e7e3j_        j@                  d-             Z` e7e3ja                  	 	 	 dd.ed/ed0ed1ee         d2eeb         d3eejc                 fd4            Zd e7e3je                  	 dd5ed6ed7ed3eejc                 fd8            Zf e7e3jg                  d9d9dd:d.ed5ed6ed7ed3eejc                 f
d;            Zh e7e3ji                  	 	 	 	 	 	 	 dd<ej        d=ej        d1ee         d>ee         d3eejc                 d?ejd@ekdAekdBejfdC            Zl e7e3jm        j@                  dDdEdFedGekdHedIej        dJebdKejdLefdM            Zn e7e3jo        j@                  dDdEdFedGekdHedIej        dJebdKejdLefdN            Zp e+             e7e3jq        j@                  dO                         Zr e7e3js        j@                  dddd dddPdQedJebdRee         dee         dSee         dTekdUejdLefdV            Zt e7e3ju        j@        e3ju        jv        g           e+            dW                         Zw e7e3ju        jx                  ddX            Zy e7e3jz        j@        e3jz        jv        g           e+            dY                         Z{ e7e3jz        jx                  ddZ            Z| e7e3j}        j@                  d[             Z~ e7e3j}        jA                  d\             Z e7e3j        j@                  d]             Z e7e3j        j                  d^             Z e7e3j        j@                  d_             Z e7e3j        j@                  dddddd`da            Z e7e3j        j@                  ddb            Z e7e3j        j@                  ddc            Z e7e3j        j@                  ddd            Z e7e3j        j@                  de             Z e7e3j        j                  df             ZdFedgebfdhZdFediedjebfdkZ	 d dledgebdmejfdnZddiedgebdoebfdpZdiedqedrejdgebfdsZ	 dduebdted.edvebfdwZdxebfdyZ e7e3j        j@        e3j        j        g           e+dzd{          ddiedxebd}ejfd~                        Z e7e3j        j@        e3j        jA        g           e+            d.edLefd                        Z e7e3j        g           e+dzd{          d.efd                        ZdedLefdZ e7e3j                   e+            dFediedejdLefd                        Z e7e3j                   e+            ddFediedejdLefd                        Z e7e3j                   e+            ddFedejdLefd                        Z e7e3j                   e+            ddFedejdLefd                        Z e7e3j        j@                  ddiedejdejfd            Z e7e3j        j@        e3j        jA        g           e+            d.ededLefd                        Z e7e3j        j@                  ddiedejfd            Z e7e3j        j@        e3j        jA        g           e+ddd          ddddFedejdejdLe	eeef         fd                        Z e7e3j        j@        e3j        jA        g           e+            dddededqedejdLef
d                        Z e7e3j        j@        e3j        jA        g           e+dd|d          dDddiedejdLe	eeef         fd                        Z e7e3j        j@        e3j        jA        g           e+ddd          dDdddiedejdejdLe	eeef         fd                        Z e7e3j        j@        e3j        jA        g           e+            dDdddededqedrejdejdLefd                        Z e7e3j                   e+dd|d          	 	 ddededejdejdLe	eeef         f
d                        ZdebdLe	ejejf         fdZ e7e3j        j@        e3j        jA        g           e+dd          ddiedebdLe	eef         fd                        Z e7e3j        j@        e3j        j        g           e+dddd          diedLe	eeeef         fd                        Z e7e3j        j@                  	 	 	 ddiedejdejdeeb         fd            ZdededLe	eek         eek         f         fdZdededjeeb         dLe	eef         fdZd.ededLejfdZ e7e3j                  dDdddddddiedqedrejdejdtee         dee         dee         dee         dLe	eeeef         fd            Z e7e3j        j@        e3j        jA        g          dDddddiedqedejdrejdejdee         dLefd            Z e7e3j                   e+dddD          	 	 	 ddFediedejdejdejdLe	eef         fd                        Z e7e3j        j@                  d             Z e7e3j                   e+            	 	 d	d.edededrejdejdLefd                        ZdÄ ZdĄ Z e7e3j                   e+            dń                         Z e7e3j                   e+            dƄ                         ZdǄ Z e7e3j                   e+dȦ          dɄ                         Z e7e3j                   e+dȦ          dʄ                         Zd˄ Z e7e3j                   e+            d̄                         Z e7e3j                   e+            d̈́                         Z e7e3j        j@        e3j        j        e3j        j@        e3j        j        g           e+dȦ          d΄                         Zdτ Z e7e3j                   e+            dЄ                         Z e7e3j                   e+            dф                         Z e7e3j        j@        e3j        j        e3j        j@        e3j        j        g           e+dȦ          d҄                         Z e7e3j                   e+            d
dFededLefdՄ                        Z e7e3j                   e+            dedFedededLef
d؄                        Z e7e3j        j@        e3j        jA        g           e+            d9d9dٜdڄ                        Z e7e3j        j@        e3j        jA        g           e+            dd#dۄ                        Z e7e3j        j                  dd݄            Z e7e3j        j                  ddބ            Z e7e3j        j@        e3j        jA        g           e+            dd߄                        Z e7e3j        j@                  	 	 dd            Z e7e3j                   e+            d                         Zd ZddZ	 ddej        d/ej        de
eek         ekf         de
eek         ekf         de
eek         ekf         dejdekdee
eek         ekf                  fdZd Z e7e3j        j@                  dej        d/ej        d1eej                 deej                 deej                 dejdedefd            Z e7e3j        j@                  dej        d/ej        d1ej        deek         deek         deek         dejdeek         dekfd            Zej        j        rej4        5                    ddd          Z e7ej2        j         j        j@                  d             Z e7ej2        j         j        j@                  d             Zej        j        rCej4        5                    ddd          Z e7ej2        j        j                  d             Z	ej4        5                    ddd          Z
 e7ej2        j        j        j@                  d             Z e7ej2        j        j        j@                   e7ej2        j        j        j                  d                         Z e7ej2        j        j        j@                   e7ej2        j        j        j@                  d                         Zej4        5                    ddd          Z e7ej2        j        j                  	 	 	 	 dd            Zd Z e7e3j        j@                  	 	 	 	 	 dd            Zd Z e7e3j        j@                  d             Z e7e3j                   e+            	 	 	 	 	 dd                        Z e7e3j                    e+dȦ          d                         Z! e7e3j"        j@                  d	             Z# e7e3j$        j@                  d
             Z% e7e3j&        j@                  d             Z' e7e3j(                   e+dȦ          d                         Z)dedoebfdZ* e7e3j+                   e+dd          d                         Z, e7e3j-                   e+dȦ          d                         Z. e7e3j/                   e+dd          d                         Z0 e7e3j1                   e+dȦ          d                         Z2 e7e3j3        j                  dd            Z4 e7e3j5        j@        e3j5        jA        g           e+            d                         Z6 e7e3j7        j@        e3j7        jA        g           e+            dddekdekfd                        Z7 e7e3j8        j        e3j9        j        g          d             Z: e7e3j;        j@        g          d             Z< e7e3j=        j@        e3j=        jA        g           e+            d9d9dٜd                        Z> e7e3j?        j@        e3j@        j@        g          dddd            ZA e7e3jB        j@        g          dddd            ZC e7e3jD        g           e+            d                         ZE e7e3jF        g          d              ZG e7e3jH        g          d!             ZI e7e3jJ        g          d"             ZK e7e3jL        g          d#             ZM e7e3jN        g          d$             ZO e7e3jP        j@                  d%             ZQ e7e3jR                   e+            d&                         ZS e7e3jT        j@                  	 	 	 	 	 	 dd'            ZU e7e3jV        j@                  d(             ZWd d)ZX e7e3jY        j@        e3jY        jA        g           e+            ddd*d+                        ZZ e7e3j[        j@        e3j\        j@        g          d,             Z] e7e3j[        jx        e3j[        j^        e3j\        jx        e3j\        j^        e3j_        j@        e3j_        j`        g           e+dd          dd-                        Za e7e3jb        j@                  d.             Zc e7e3jd        j@                  d/             Ze e7e3jf        j@                  d0             Zg e7e3jh        ji        e3jj        ji        e3jh        j        e3jj        j        e3jk        j@        e3jl        j@        e3jm        j@        g          d1             Zn e7e3jo        ji        e3jp        ji        e3jo        j        e3jp        j        g          dd2            Zq e7e3jr        j@        e3jr        js        g          d3             Ztd4 Zu e7e3jv        j        e3jv        ji        g          d5             Zw e7e3jx        j        e3jx        ji        g          d6             Zy e7e3jz        j@                  d7             Z{ e7e3j|        j        e3j|        ji        g          d8             Z} e7e3j~        j        e3j~        ji        g          d9             Z e7e3j        j@                  d:             Z e7e3j        j                   e+            ddLefd;                        Z e7e3j        g           e+            	 dd>                        Z e7e3j        g          	 dd?            Z e7e3j        g          	 dd@            Z e7e3j        j@        e3j        j@        g          ddA            Z e7e3j        ji                  dB             Z e7e3j        j@                  dC             Z e7e3j                  dD             Z e7e3j                   e+            dE                         Z e7e3j                  dF             Z e7e3j        j@                  ddG            Z e7e3j        j@                  dH             ZddIZ e7e3j        j@                  dJ             ZdK ZdL ZdM ZdN Z	 dd.edOekdPekdQekdRekdSekdTekdUekdVekdWekdXekdYekdZekd[ekd\ekd]ekd^ekd_ekd`ekdaekduebdbejf,dcZdd Zd.ededOekdPekdQekdRekdSekdTekdUekdVekdWekdXekd\ekd]ekd^ekd_ekd`ekdaekduebf&deZdf Z e7e3j        j@                  dg             Z e7e3j        j@                  	 	 	 	 ddh            Z e7e3j        j@                  di             Z e7e3j                   e+dd          	 	 	 	 ddj                        Z e7e3j                   e+dȦ          dk                         Zd.edlefdmZ G dn doe          Zd.edledpekfdqZ e7e3j        j@                  dr             Z e7e3j                   e+            ds                         Z e7e3j                   e+dȐdt          du                         Z e7e3j        j@        g          dv             Z e7e3j        j@                  	 	 	 	 	 ddw            Z e7e3j        jk                  dx             Z e7e3j        j@                  dy             Z e7e3j        j@                  ddz            Zd dGekd{ekd|ejfd}Zd~ Zd Z e7e3j        j@                  dd            ZÐddZĐddZŐd ZƐddZǐddZ e7e3j        j@                  d             Z e7e3j                  d             Z e7e3j        j        e3j        j        e3j        j        e3j        j        g           e+            dd                        Z e7e3j        j        e3j        j        e3j        j        e3j        j        g          dd            Z e7e3j        g          	 	 	 	 ddededededejdejdee         fd            Z e7e3j        g          	 	 	 	 ddedededee         dejdedejdejdee         fd            Z e7e3j        g          	 ddededededededededekdekdedejdededee         fd            Z e7e3j        g          	 	 	 	 ddededededejdee         dee         fd            Z e7e3j        g          	 	 ddedededededededejdee         dee         fd            Z e7e3j        g          	 	 	 ddedededee         dejdejdee         fd            Z e7e3j        g          	 	 ddededededee         dedededededeej         dejdee         fd            Z e7e3j        g          	 ddedededededededededededekdekdedejdee         f d            Z e7e3j        g          	 	 	 	 	 ddedededee         dee         dekdekdedejdejdee         deek         deek         dee         dee         fd            Z e7e3j        g          	 	 	 ddededededededededekdekdedejdededee         deek         deek         f"d            Z e7e3j        g          	 	 	 	 	 ddededed1ee         dee         dee         deek         deek         dedekdejdee         dee         dee         deek         fd            Z e7e3j        g          	 	 	 ddedededed1ee         dee         dee         dej        dej        dededededekdejdee         deek         dejf$d            Z e7e3j        j@        g          	 	 	 	 ddFej        d7ej        dej        dej        d1eej                 deej                 d3eejc                 dejfd            Z e7e3j        j        e3j        j        g           e+            d d                        Z e7e3j        j                  d d            Z e7e3j        j@        e3j        jA        g           e+            ddd#d                        Zd Zd Z e7e3j        j@        e3j        j@        g          ddÄ            Z e7e3j        j@        e3j        j@        g          ddĄ            Z e7e3j        j@        e3j        j@        g          	 	 ddedee
ekej        f                  dee
ekej        f                  dee         dee         f
dɄ            Z e7e3j         j@        e3j        j@        g          ddʄ            Z  e7e3j        j@        e3j        j        e3j        j`        e3j        j        g          dd˄            Zd̄ Z e7e3j        j@                  	 	 dd̈́            Z e7e3j	        j@                  d΄             Z	 e7e3j
        j@                  dτ             Z
dЄ Zdф Z e7e3j        j@        e3j        j@        g          dd҄            Z e7e3j        j@                  ddӄ            Z e7e3j        j@                  ddԄ            Z e7e3j                   e+            	 ddՄ                        Z e7e3j        j@        e3j        j`        g           e+dd          ddք                        Zej        Zdׄ Z e7e3j        j@                  d؄             Z e7e3j        j@                  dل             Z e7e3j        j@                  dڄ             Z e7e3j        j@                  dۄ             Z e7e3j        j        e3j        j         g           e+            dddܜd݄                        Z! e7e3j"        g           e+            d d߄                        Z# e7e3j$        j@        e3j%        j@        g          	 	 dd            Z& e7e3j'        j@                  d             Z' e7e3j(        j@        e3j(        jA        g           e+            dd                        Z( e7ej2        j3        j)                  d             Z) e7ej2        j3        j*                  d             Z* e7e3j+                   e+            dddddd                        Z,d Z- e7e3j.                  	 d!d            Z/ e7e3j0                  	 d!d            Z1 e7e3j2                  	 d!d            Z3 e7e3j4                   e+            dddd                        Z5 e7e3j6                   e+            dekdFedLefd                        Z7 e7e3j8                  dFefd            Z9 e7e3j:        j@                  	 d"dedSee         deek         defd            Z;d Z<d Z= e<e3j>                    e<e3j?                    e<e3j@                    e<e3jA                    e<e3jB                    e<e3jC                    e<e3jD                    e<e3jE                    e<e3jF                    e=e3jG                    e=e3jH                    e=e3jI                    e=e3jJ                    e=e3jK                    e=e3jL                    e=e3jM                    e=e3jN                    e=e3jO                    e=e3jP                    e=e3jQ                    e=e3jR                   d ZS e7e3jT                   e+            d                         ZT e7e3jU                   e+            d9dd                        ZU e7e3jV                   e+            d9dd                        ZV eSe3jT                  ZW eSe3jU                  ZX eSe3jV                  ZYd dl,Zd dlZZd dl[Zd Z\ e\             dS (#      N)Enum)wraps)ListOptionalSequenceTupleUnion)SymBoolSymFloatTensor)_add_op_to_registry_convert_out_paramsglobal_decomposition_table
meta_table)
OpOverload)_prim_elementwise_meta$ELEMENTWISE_PRIM_TYPE_PROMOTION_KIND)
BoolLikecorresponding_complex_dtypecorresponding_real_dtypeelementwise_dtypesELEMENTWISE_TYPE_PROMOTION_KIND	FloatLikeIntLikemake_contiguous_strides_forNumber
TensorLike)_maybe_convert_to_dtype_maybe_resize_out_resize_output_check_safe_copy_outout_wrapper)_broadcast_shapes_maybe_broadcast)_pytreeatenIMPLMetac                       fd}|S )Nc                 \     t                       fd}t          j        |            S )Nc                 4    t          t          |            d S N)r   r   )opfns    i/root/.openclaw/workspace/chatterbox_venv_py311/lib/python3.11/site-packages/torch/_meta_registrations.pyregisterz0register_meta.<locals>.wrapper.<locals>.register3   s    
B33333    )r   pytree	tree_map_)r.   r0   r-   s   ` r/   wrapperzregister_meta.<locals>.wrapper0   sD     $$	4 	4 	4 	4 	4 	2&&&	r1    )r-   r4   s   ` r/   register_metar6   /   s#         Nr1   type_promotionc                     t          j        |d| i\  }fd|D             }t          | }t          |dt          j        iS )Ntype_promotion_kindc                 0    g | ]}t          |          S r5   )r   ).0xresult_dtypes     r/   
<listcomp>z$elementwise_meta.<locals>.<listcomp>E   s$    CCC#A|44CCCr1   r7   )utilsr   r$   r   r   DEFAULT)r7   args_r=   s      @r/   elementwise_metarC   <   sp    
 .	* OA| DCCCdCCCD T"D "	BJ  r1   c                     t           j        t           j        t           j        t           j        t           j        t           j        i}|                    | |           S r,   )torch	complex32halfcfloatfloatcdoubledoubleget)dtypefrom_complexs     r/   toRealValueTyperO   P   s<    eku|L
 E5)))r1   c                 x     t          t           g|R            t          j         k     fd           d S )Nc                      d d  S )Nzoutput with shape z# doesn't match the broadcast shape r5   )broadcasted_shape
self_shapes   r/   <lambda>z)check_inplace_broadcast.<locals>.<lambda>]   s    gZggTegg r1   )tupler#   rE   _check)rS   
args_shaperR   s   ` @r/   check_inplace_broadcastrX   Y   sX    /
HZHHHII	LZ'ggggg    r1   Fc	                 \   	 t           t          j                  r,t          j                                         dk    d            t          t          j                  r,t          j                                        dk    d            t          d  fD                       rVt          j        t          j                              		nAt          j        t          j	                  	fd           npt          j                    t          t          j
                  sJ t          j        t          t                     fd           t          t                    sJ t          j        dk    d            t          j        f|d||	          S )
Nr   c                      dS Nz:linspace only supports 0-dimensional start and end tensorsr5   r5   r1   r/   rT   z(meta_linspace_logspace.<locals>.<lambda>q       P r1   c                      dS r[   r5   r5   r1   r/   rT   z(meta_linspace_logspace.<locals>.<lambda>v   r\   r1   c              3   @   K   | ]}t          |t                    V  d S r,   )
isinstancecomplex)r;   args     r/   	<genexpr>z)meta_linspace_logspace.<locals>.<genexpr>y   s,      
C
C:c7##
C
C
C
C
C
Cr1   c                      d  d S )Nzlinspace(): inferred dtype z& can't be safely cast to passed dtype r5   )default_complex_dtyperM   s   r/   rT   z(meta_linspace_logspace.<locals>.<lambda>   s    z6Kzzsxzz r1   c                      dt                    j         dt                     j         dt                    j         dS )Nz4received an invalid combination of arguments - got (, ))type__name__)endstartstepss   r/   rT   z(meta_linspace_logspace.<locals>.<lambda>   sY     Hu++H Hs)),H H04U0DH H H r1   c                      dS )Nz$number of steps must be non-negativer5   r5   r1   r/   rT   z(meta_linspace_logspace.<locals>.<lambda>   s    %K r1   meta)rM   layoutdevice
pin_memoryrequires_grad)r_   rE   r   rV   dimanyr?   r   get_default_dtypeis_complex_dtyperM   _check_typer   empty)
rk   rj   rl   baserM   rp   ro   rq   rr   rd   s
   ``` `    @r/   meta_linspace_logspacerz   a   s    %&& 
IIKK1PP	
 	
 	
 #u|$$ 
GGIINPP	
 	
 	

 
C
CsE/B
C
C
CCC 3 % A#%%!
 !
 =)EEL&u--zzzzz   
 2022eU[))))) 
5'""	H 	H 	H 	H 	H 	H  
 eW%%%%%	L!KKLLL;	#   r1   c                    t          j        j        t           j        k    fd           t          j        |                                 dk    o                                dk     d            |                     j                  S )Nc                      d j          S )Nz2take(): Expected a long tensor for index, but got rM   indexs   r/   rT   zmeta_take.<locals>.<lambda>   s    RU[RR r1   r   c                      dS )Nz*take(): tried to take from an empty tensorr5   r5   r1   r/   rT   zmeta_take.<locals>.<lambda>   s    < r1   )rE   rV   rM   long_check_indexnumel	new_emptyshape)selfr   s    `r/   	meta_taker      s     
Luz!RRRR  
 
ZZ\\Q55;;==A#56<<   >>%+&&&r1   rs   c                J     j         }j         }t          j        ||k    d            t          j                                       dk    o                              dk     fd           t	           j        j                  }                     |          S )Nc                      dS )Nz=linalg.cross: inputs must have the same number of dimensions.r5   r5   r1   r/   rT   zlinalg_cross.<locals>.<lambda>       O r1      c                  f    d  d                                 d                                 S )Nzlinalg.cross: inputs dimension z must have length 3. Got  and size)rs   otherr   s   r/   rT   zlinalg_cross.<locals>.<lambda>   sG    :c : :99S>>: :(-

3: : r1   )ndimrE   rV   r   r#   r   r   )r   r   rs   x_dy_d	out_shapes   ```   r/   linalg_crossr      s     )C
*C	Ls
OO   
L		#!4

31 4	
 	
 	
 	
 	
 	
   "$*ek::I>>)$$$r1   c                     t          | d           t          | d           t          j        | t          j                  S )Nzlinalg.matrix_expmemory_format)squareCheckInputscheckFloatingOrComplexrE   
empty_likecontiguous_formatr   s    r/   linalg_matrix_expr      s?     d/0004!4555D0GHHHHr1   valuesindicesc                    t          j        | j        | j        | j                  }t          j        | j        | j        t           j                  }|                                 dk    r | j        dk    rt          || j                   ||fS )Nrp   rM   r   )	rE   rx   r   rp   rM   int64r   r   maybe_wrap_dim)r   rs   r   r   s       r/   	cummaxminr      su    
 [DKtzJJJFk$*T[LLLGzz||qTY!^^sDI&&&7?r1   c                 x    t          || j                   t          j        |                                           S r,   )r   r   rE   r   
contiguous)r   rs   s     r/   logcumsumexpr      s3     3	"""D!!,,...r1   c                    |j         }t          |          }||z
  }t          t          |                    }d t          |          D             }	|D ]}
d|	|
<   g g }}|D ]5}
|	|
         s|                    |
            |                    |
           6||z   }t          |          }|                                |d |         }|                    fdd           |||d          z   }|                    |          }dgt          |j        |d                    z   }|	                    |          }|
                    d          }||d<   |}t          t          |                    D ]}|||                  ||dz   <   | 	                    |          } d t          |          D             }d}|dz
  }|dk    r=||                     d          z  |||         <   ||||                  z  }|dz  }|dk    =t          ||          D ]&}|                     d||z
  z             |||         <   '|                     |||                                           S )	Nc                     g | ]}d S Fr5   r;   rB   s     r/   r>   z_exec_fft.<locals>.<listcomp>   s    555A%555r1   Tc                     |          S r,   r5   r<   self_stridess    r/   rT   z_exec_fft.<locals>.<lambda>   s    <? r1   keyreverser   r      c                     g | ]}d S r   r5   r   s     r/   r>   z_exec_fft.<locals>.<listcomp>  s    ***1***r1   )r   lenlistrangeappendstridesortpermuter   reshaper   
as_stridedstorage_offset)outr   	out_sizesrs   forwardr   signal_ndim
batch_dimsdim_permuteis_transformed_dimdleftright	batch_endtmpinputbatched_sizes
batch_sizebatched_out_sizesiout_stridesbatch_numelr   s                         @r/   	_exec_fftr      s   9Dc((K#J uT{{##K55t555 % % $1 b%D  !!$ 	KKNNNNLLOOOO,KD		I;;==L
jyj
!CHH****DH999IJJ//KLL%%E D4JKK 8999MMM-((EAJ!M!%3s88__ 5 5#,SV#4!a%  
++'
(
(C +*eDkk***KKQA
q&&&1CJJqMM&AKN#yQ00	Q q&& :t$$ G G&)jja*n1E&F&FKN##>>)[#2D2D2F2FGGGr1   c                     | j         j        sJ | j        }|                     |          }|s|S |d d          }|                                 |                    fdd           t          || |||          }|S )Nc                     |          S r,   r5   r   s    r/   rT   zmeta_fft_c2c.<locals>.<lambda>   s    <? r1   Tr   )rM   
is_complexr   r   r   r   r   )r   rs   normalizationr   r   outputsorted_dimsr   s          @r/   meta_fft_c2cr     s     :    
I^^I&&F aaa&K;;==L2222DAAAvtYWEEFMr1   c                     | j         j        sJ t          |                                           }|r|d         }||         dz  dz   }|||<   |                     |t          j        | j                             S )Nr      r   r}   )rM   is_floating_pointr   r   r   r?   r   )r   rs   r   onesidedoutput_sizeslast_dimlast_dim_halfsizes          r/   meta_fft_r2cr   &  s     :''''		$$L 3r7)(3q8A=!2X>>E=djII    r1   )	generatorc                H    t          |t          j        | g                    S r,   )r   rE   Size)nr   r   s      r/   meta_randpermr   6  s    S%*aS//222r1   rM   ro   rp   rq   c                4    t          j        | ||||          S Nr   rE   rx   )r   rM   ro   rp   rq   s        r/   meta_randperm_defaultr   ;  s(     ;	vf   r1   c                4    t          j        |||||          S r   r   )highr   rM   ro   rp   rq   s         r/   meta_randintr   I  s(     ;E&J   r1   c                4    t          j        |||||          S r   r   )lowr   r   rM   ro   rp   rq   s          r/   meta_randint_lowr   Y  s(     ;E&J   r1   c                4    t          j        | ||||          S r   r   )r   rM   ro   rp   rq   s        r/   meta_rand_defaultr   j  s(     ;E&J   r1   c                     | j         j        sJ t          |                                           }|||d         <   |                     |t          | j                             S )Nr   r}   )rM   r   r   r   r   rO   )r   rs   r   lastdimr   s        r/   meta_fft_c2rr   r  sX     :    		$$L#LR>>,odj.I.I>JJJr1   c                    ddl m}  ||           s't          j        |           dk    rt	          d          t          |t                    rr|                    | |          }|                                 |                                k    r2t          j
                            ||                                            | S )Nr   )free_unbacked_symbolsr   zQmore than one element of the written-to tensor refers to a single memory location)%torch.fx.experimental.symbolic_shapesr   rE   _debug_has_internal_overlapRuntimeErrorr_   r   tor   r&   expand_copydefault)r   srcnon_blockingr   intermediates        r/   
meta_copy_r  {  s     LKKKKK "!$''
,1,Md,S,SWX,X,X_
 
 	
 #v @vvdL1199;;,++----$$\499;;???Kr1   c                 <   t          |                                           }t          |                                           }||                                 k    rdn||         ||         z  }|                    |d           |                    ||           ||fS Nr   )r   r   r   rs   insert)tensorrs   result_sizesresult_strides
new_strides        r/   inferUnsqueezeGeometryr    s    &&L&--//**NVZZ\\))|C/@>RUCV/VJQ#z***''r1   c                     t          ||                                 dz             }t          | |          \  }}|                     ||           | S r  )r   rs   r  as_strided_)r   rs   g_sizes	g_stridess       r/   meta_unsqueeze_r    sL    
dhhjj1n
-
-C/c::GYWi(((Kr1   r   weight_metabias_activation_opt	out_dtypec                 l   t          | j                  }|6|                    d          |                    d          k    s
J d            |                    d          |                     d          dz  k    sJ |                    d          |d<   t          | j                  dk    s
J d            d|                     d          f}|/| j        t
          j        k    r|t
          j        k    s
J d            |                     ||| j        n|          	                    ||          }|S )	Nr   zoutput size mismatchr   r   r   z*we can only handle the squashed input case9out_dtype is only supported for i8i8->i32 linear operatorr}   )
r   r   r   r   rM   rE   int8int32r   r   )	r   r  r  r  r  r  r   transposed_stridesr   s	            r/   meta_sparse_structured_linearr    s8    $$L{{1~~1---/E---;;q>>UZZ^^a/////{{1~~L u{q   "N   UZZ]]+K5:%%)u{*B*B*BF +C*BB__&.ekkI    j122 
 Mr1   mat1	mat1_metamat2c                    t          | j                  dk    sJ t          |j                  dk    sJ t          |j                  dk    sJ |                     d          |                    d          dz  k    sJ |                     d          |                    d          g}|/|j        t          j        k    r|t          j        k    s
J d            |                    |||j        n|          }|S )Nr   r   r   r  r}   r   r   r   rM   rE   r  r  r   )r  r  r   r  r   r   s         r/   meta_sparse_structured_mmr#    s     tz??ay1$$$$tz??a99Q<<499Q<<!+++++IIaLL$))A,,/LJ%*$$ek)A)A)AF *B)AA^^%-djj9   F
 Mr1   r   )alphabetar  c                   t          | j                  dk    s
J d            t          |j                  dk    sJ t          |j                  dk    sJ t          |j                  dk    sJ |                     d          |                    d          k    s
J d            |                    d          |                    d          dz  k    sJ |                    d          |                    d          g}|/|j        t          j        k    r|t          j        k    s
J d            |                    |||j        n|          }|S )Nr   zEonly input broadcasted to columns of mat1 * mat2 product is supportedr   r   r  r}   r"  )	r   r  r  r   r$  r%  r  r   r   s	            r/   meta_sparse_structured_addmmr'    se    	EKAN 	tz??ay1$$$$tz??a::a==DII	    N   99Q<<499Q<<!+++++IIaLL$))A,,/LJ%*$$ek)A)A)AF *B)AA^^%-djj9   F
 Mr1   compressed_Adense_Br$  transpose_resultalg_idsplit_ksplit_k_one_kernelc	                    |j         t          j        t          j        t          j        t          j        t          j        hv s
J d            | j         |j         k    s
J d            t          |j                  dk    s
J d            | j         t          j        t          j        fv }	|	rdnd}
|	r|	                                r
J d            |
                    d          }|
                    d	          }|                                 d
z  |
|z  z  }|||
                    d          k    sJ |<|	r0|t          j        t          j        t          j        t          j        hv s
J d            |r||fn||f}|                    ||          S )Nz;_cslt_sparse_mm only supports fp16, bf16, int8, and fp8e4m3zinputs must have the same dtyper   z'_cslt_sparse_mm only supports 2d inputs
   	   z.dense input must be transposed for 8bit dtypesr   r      z\out_dtype is not supported for {compressed_A.dtype} x {dense_B.dtype} -> {out_dtype} matmul!r}   )rM   rE   float32float16bfloat16r  float8_e4m3fnr   r   is_contiguousr   r   r  r   )r(  r)  r  r$  r  r*  r+  r,  r-  is_8bit_input_typecompression_factorkr   moutput_shapes                  r/   meta__cslt_sparse_mmr<    s    =
    E   ...0Q...w}"""$M"""%+
E<O/PP18q <%%''	< 	<;	< 	<' 	QAQA						"(:Q(>?ADIIaLL    		j#	   j  .9Aq66Aq6L\;;;r1   T)include_selfr   rs   r   sourcereducer=  returnc                B    t          j        | t           j                  S Nr   )rE   r   r   r   rs   r   r>  r?  r=  s         r/   meta_index_reducerD  6  s     D0GHHHHr1   c                    | S r,   r5   rC  s         r/   meta_index_reduce_rF  C  s	     Kr1   c                     t          |                                           }|                                 dk    r|                                ||<   |                     |          S )Nr   )r   r   rs   r   r   )r   rs   r   result_sizes       r/   meta_index_selectrI  Q  sN     tyy{{##KxxzzA~~ ;;==C>>+&&&r1   )lengthsr   offsetsaxisunsafeinitialdatarJ  rK  rL  rM  c                     |t          d           fd}| ||j                  S |,|j        d d         |j        d         dz
  fz   }	 ||	          S t          d          )Nz?segment_reduce(): indices based reduction is not supported yet.c                 x    t          j        | j        dz   d          z   j        dt           j                  S )Nr   rn   rM   rp   r   )rE   rx   r   rM   r   )lengths_shaperL  rO  s    r/   segment_reduce_lengths_tensorz:meta_segment_reduce.<locals>.segment_reduce_lengths_tensork  sA    {DJtaxzz22*1	
 
 
 	
r1   r   r   z<segment_reduce(): Either lengths or offsets must be defined.)NotImplementedErrorr   r   )
rO  r?  rJ  r   rK  rL  rM  rN  rT  rS  s
   `    `    r/   meta_segment_reducerV  Z  s     !M
 
 	

 
 
 
 
 
 ,,W];;; crc*gmB.?!.C-EE,,];;;
U
V
VVr1   c                 ,    |                      d          S Nr5   r   r   s    r/   meta_maxrZ  ~       >>"r1   c                     t          j        | j        |f          }t          | ||          }|                     |          |                     |t
          j                  fS Nr}   r?   reduction_dimsr   _compute_reduction_shaper   rE   r   r   rs   keepdimr;  s       r/   meta_max_dimrc    W    

tzC6
2
2C+D#w??L|$$|5:66 r1   c                 ,    |                      d          S rX  rY  r   s    r/   meta_minrf    r[  r1   c                     t          j        | j        |f          }t          | ||          }|                     |          |                     |t
          j                  fS r]  r^  ra  s       r/   meta_min_dimrh    rd  r1   c                     |                                  rt          | j                  }nt          | t          j                  \  }}t          j        | |          S )Nr9   r}   )r   r   rM   r   r   INT_TO_FLOATrE   r   )r   r=   rB   s      r/   
meta_anglerl    s^     
/
;;, ? L
 
 
< D5555r1   c                     t          j        ||                                 | j                   |                    t          j        |                     S r,   )rE   _resize_output_r   rp   copy_angle)r   r   s     r/   meta_angle_outrq    s=    	#tyy{{DK88899U[&&'''r1   c                     d S r,   r5   )vals    r/   assert_asyncrt        
Fr1   c                     d S r,   r5   )rs  
assert_msgs     r/   assert_async_metarx    ru  r1   c                     d S r,   r5   )ss    r/   
print_metar{    ru  r1   rM   ro   rp   rq   r   c                 .    t          j        dd          S )Nr   rn   rp   r   r|  s        r/   make_dep_tokenr    s     ;q((((r1   c                     ddl m} t          | t          t          f          rt          d           || ||           d S )Nr   )constrain_range'Constraining SymFloat or Symbool is nyiminmax)r   r  r_   r   r
   
ValueError)r   r  r  r  s       r/   sym_constrain_ranger    s[     FEEEEE$7+,, DBCCCODcs++++++r1   c                 @    t                               | ||           |S Nr  )r&   r  r   r  r  	dep_tokens       r/   functional_sym_constrain_ranger    s#    Ts444r1   c                 P   ddl m} ||t          j        |            d S t	          | t
          t          f          rt          d          t          |           t          u r6|t          j
        | |k               |t          j
        | |k               d S  || ||           d S )Nr   )_constrain_range_for_sizer  r  )r   r  rE   _check_is_sizer_   r   r
   r  rh   intrV   )r   r  r  r  s       r/   sym_constrain_range_for_sizer    s     POOOOO
{s{T"""$7+,, DBCCCDzzS?L%%%?L%%%d555555r1   c                 @    t                               | ||           |S r  )r&   r  r  s       r/   'functional_sym_constrain_range_for_sizer    s#    %%d%===r1   c                     |S r,   r5   )rs  rw  r  s      r/   functional_assert_async_metar    s    r1   f_namec                    |                                  dk    sJ | d            |                     d          |                     d          k    s9J | d|                     d           d|                     d           d            d S )Nr   z3: The input tensor must have at least 2 dimensions.r   z5: A must be batches of square matrices, but they are  by 	 matrices)rs   r   )r   r  s     r/   r   r     s    

a
EEE 	 			"2&&&
rrtyyQS}}rrZ^ZcZcdfZgZgrrr 	'&&&&r1   Anamec                     t          j         j        j        k     fd           t          j         j        j        k     fd           t          j                            d                              d          k    fd           t          j                            d                               d          k     fd           d S )Nc                  *    dj          d j          dS )Nz:Expected b and A to be on the same device, but found b on z
 and A on 	 instead.r~  r  r   s   r/   rT   z(linearSolveCheckInputs.<locals>.<lambda>
  s0    :{: :&'h: : : r1   c                  *    dj          d j          dS )Nz=Expected b and A to have the same dtype, but found b of type z and A of type r  r}   r  s   r/   rT   z(linearSolveCheckInputs.<locals>.<lambda>  s0    =z= =*+'= = = r1   r   r  c                  b    d                      d           d                      d           dS )Nz3A must be batches of square matrices, but they are r  r  r   r  r   r  s   r/   rT   z(linearSolveCheckInputs.<locals>.<lambda>  sF    BFF2JJB B,-FF2JJB B B r1   c                      d d                      d           d                      d           d                     d           d                     d           
S )NzIncompatible matrix sizes for z: each A matrix is r   r  z but each b matrix is r  r   )r  r  r   s   r/   rT   z(linearSolveCheckInputs.<locals>.<lambda>"  s    HT H HH H)*H H%)YYr]]H H8<		"H H r1   )rE   rV   rp   rM   r   )r   r  r  s   ```r/   linearSolveCheckInputsr    s   	Lqx	
 	
 	
 	
 	
   
L
ag	
 	
 	
 	
 	
   
L	r

affRjj 	
 	
 	
 	
   
L	r

diimm#	
 	
 	
 	
 	
 	
    r1   tallow_low_precision_dtypesc                 ,   | j         t          j        |                                 p|                                 fd           |sIt          j        t          j        t          j        t          j        t          j        fv fd           d S d S )Nc                       d  S )Nz<: Expected a floating point or complex tensor as input. Got r5   rM   r  s   r/   rT   z(checkFloatingOrComplex.<locals>.<lambda>3  s    6^^W\^^ r1   c                       d  S )Nz*: Low precision dtypes not supported. Got r5   r  s   r/   rT   z(checkFloatingOrComplex.<locals>.<lambda>8  s    vPPPP r1   )	rM   rE   rV   r   r   rI   rK   rH   rJ   )r  r  r  rM   s    ` @r/   r   r   +  s    
 GE	L	/^^^^^   & 
ek5<u}MMPPPPP	
 	
 	
 	
 	

 
r1   arg_namec                 h    t          j        |                                 dk    fd           d S )Nr   c                       d  dS )Nz: The input tensor z! must have at least 2 dimensions.r5   )r  r  s   r/   rT   zcheckIsMatrix.<locals>.<lambda>@  s    6YYhYYY r1   )rE   rV   rs   )r  r  r  s    ``r/   checkIsMatrixr  =  s@    	L	1YYYYY    r1   Br   c                 4    t                      t                     t          j        r,                     d                              d          k    n+                     d                              d          k     fd           d S )Nr  r   c                       drdnd d                      d           d                      d           d                     d           d                     d           d	S )
Nz2: Incompatible shapes of A and B for the equation zAX = BzXA = Bz (r  r<   r   r   rg   r   )r  r  r  r   s   r/   rT   z#checkInputsSolver.<locals>.<lambda>I  s     J J-xxXJ JJ JffRjjJ J/0vvbzzJ J<=FF2JJJ J J r1   )r   r  rE   rV   r   )r  r  r   r  s   ````r/   checkInputsSolverr  D  s    a   !V	L$(Fr

affRjj  affRjjAFF2JJ.F	
 	
 	
 	
 	
 	
 	
    r1   resultfn_nameresult_namec                 `     t          j        j        j        k     fd           d S )Nc            	      8      d d dj          dj          	S )Nz: Expected z5 and input tensors to be on the same device, but got z on z and input on r~  )r  r   r  r  s   r/   rT   z!checkSameDevice.<locals>.<lambda>Y  sV     L L; L LL L &L L=B\L L r1   )rE   rV   rp   )r  r  r   r  s   ````r/   checkSameDevicer  Q  sU     
L%	
 	
 	
 	
 	
 	
 	
    r1   UPLOc                                                        }t          j        t                     dk    o|dk    p|dk     fd           d S )Nr   ULc                      d  S )Nz1Expected UPLO argument to be 'L' or 'U', but got r5   )r  s   r/   rT   zcheckUplo.<locals>.<lambda>d  s    JDJJ r1   )upperrE   rV   r   )r  UPLO_uppercases   ` r/   	checkUplor  `  s[    ZZ\\N	LD		QKNc1J^s5JJJJJ    r1   eigenvalueseigenvectorsr  	compute_vc                    t          | d           t          |           t          | j                  }|r;|                     |          }|                    |t          |d                     n|                     dg          }|                                 |                     |t          | j	                            }||fS )Nzlinalg.eighF	row_majorr   r}   )
r   r  r   r   r   r  r   poprO   rM   )r  r  r  r   vecsvalss         r/   meta__linalg_eighr  h  s     a'''dOOOMME  {{5!! ;EU S S STTTT{{A3	IIKKK;;uOAG$<$<;==D:r1   c                     t          | d           t          j        | j                  r| j        nt          j        | j                  }|                     | j        d d         |          S )Nzlinalg.eigvalsr   r}   r   r?   rv   rM   r   r   r   )r   complex_dtypes     r/   meta__linalg_eigvalsr  {  sj     e-... !%+..	<.u{;; 
 ??5;ss+=?AAAr1   c                    t          | d           t          j        | j                  r| j        nt          j        | j                  }|                     | j        d d         |          }|                     | j        |          }||fS )Nz
linalg.eigr   r}   r  )r   r  r   vectorss       r/   meta_linalg_eigr    s     e\*** !%+..	<.u{;; 
 __U["-]_CCFooeko??G7?r1   r  c                 t    | j                             t          j                                      dd          S )Nr   r  r   )mTclonerE   r   	transpose)r  s    r/   cloneBatchedColumnMajorr    s+    6<<e&=<>>HHRPPPr1   r  c                      t          |           S r,   )r  )r   r  r  s      r/   _cholesky_solve_helperr    s     #4(((r1   c                      t          j         j        dk     fd           t          j        j        dk    fd           t           d          \  }}t	          |||          S )Nr   c                      d j          dS )Nz-b should have at least 2 dimensions, but has  dimensions insteadr   r   s   r/   rT   z cholesky_solve.<locals>.<lambda>  s    ^	^^^ r1   c                      d j          dS )Nz-u should have at least 2 dimensions, but has r  r  r  s   r/   rT   z cholesky_solve.<locals>.<lambda>  s    [[[[ r1   cholesky_solve)rE   rV   r   !_linalg_broadcast_batch_dims_namer  )r   r  r  self_broadcastedA_broadcasteds   ``   r/   r  r    s     
L	Q^^^^   
L	![[[[   'Ha!' '#m ""2M5IIIr1   c                     |                                  dk    r t          j        | t          j                  S t	          | d           t          |           S )Nr   r   cholesky)r   rE   r   legacy_contiguous_formatr   r  r   r  s     r/   r  r    sN     zz||qE4RSSSSdJ'''"4(((r1   c                 @    t          | d           t          |           S )Ncholesky_inverse)r   r  r  s     r/   r  r    s#     d.///"4(((r1   check_errorsc                 D   t          | d           t          | d           | j        }t          |          }t	          |d          }|                     |          }|                    ||           |                     |d|dz
           t          j                  }||fS )Nzlinalg.choleskyFr   r   r}   )	r   r   r   r   r   r   r  rE   r  )r  r  r  A_shaper   	L_stridesr  infoss           r/   linalg_cholesky_exr    s    a*+++1/000gGw<<D ,GU;;I	GAMM'9%%% KKD1H-U[KAAEe8Or1   tauc                     t          j         j        dk    d            t          j                             d                               d          k    d            t          j                             d                              d          k    d            t          j         j        j        z
  dk     fd            j        dk    r: j        d d         }j        d d         t          j        |k    fd	           t          j        j         j        k     fd
           t          d d           t          j         j        t           j        d           j         j	                  S )Nr   c                      dS )NzHtorch.linalg.householder_product: input must have at least 2 dimensions.r5   r5   r1   r/   rT   z,linalg_householder_product.<locals>.<lambda>      Z r1   r  r   c                      dS )Nzbtorch.linalg.householder_product: input.shape[-2] must be greater than or equal to input.shape[-1]r5   r5   r1   r/   rT   z,linalg_householder_product.<locals>.<lambda>  s    t r1   c                      dS )Nz`torch.linalg.householder_product: input.shape[-1] must be greater than or equal to tau.shape[-1]r5   r5   r1   r/   rT   z,linalg_householder_product.<locals>.<lambda>  s    r r1   r   c                  (    dj          d j          S )Nzptorch.linalg.householder_product: Expected tau to have one dimension less than input, but got tau.ndim equal to  and input.ndim is equal to r  r   r  s   r/   rT   z,linalg_householder_product.<locals>.<lambda>  0    \),\ \OTz\ \ r1   c                      d  S )Nzltorch.linalg.householder_product: Expected batch dimensions of tau to be equal to input.shape[:-2], but got r5   actual_batch_tau_shapes   r/   rT   z,linalg_householder_product.<locals>.<lambda>      O6LO O r1   c                  (    dj          d j          S )Nz,torch.linalg.householder_product: tau dtype z does not match input dtype r}   r  s   r/   rT   z,linalg_householder_product.<locals>.<lambda>  s*    939 9 9+0;9 9 r1   z torch.linalg.householder_productr  Fr  r   r   rM   rp   )
rE   rV   r   r   r   rM   r  empty_stridedr   rp   )r   r  expected_batch_tau_shaper  s   `` @r/   linalg_householder_productr    s   
 
L
aZZ   
L

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

2#((2,,&rr  
 
L
SX"	
 	
 	
 	
 	
   zA~~#(;ss#3 !$3B3"&>>   	
 	
 	
 
L	U[ 	
 	
 	
 	
 	
   6UEJJJ[*5;%HHHk|	   r1   c                 <   t          | d           t          | dd           |                     | j                  }|                    | j        t          | j        d                     |                     | j        d d         t          j                  }||fS )Nzlinalg.inv_exF)r  r  r  r}   r   r   r   r   r  r   rE   r  )r  r  r  r  s       r/   linalg_inv_ex_metar  	  s    a)))1o%PPPP	AGAMM!'6qw%PPPQQQKKEKK88Ee8Or1   LDpivotsinfo)	hermitianr  r  c                z   t          | d           t          | d           t          j        | j        t          | j        d          | j        | j                  }|                     | j        d d         t          j	                  }|                     | j        d d         t          j	                  }|||fS )Nztorch.linalg.ldl_factor_exFr  r  r   r}   r  )
r   r   rE   r  r   r   rM   rp   r   r  )r   r  r  r  r  r  s         r/   linalg_ldl_factor_ex_metar
    s     d89994!=>>>		Z*4:GGGj{	
 
 
B ^^DJssO59^==F>>$*SbS/>;;Dvtr1   )r  c                @    t           d           t           d           t           d           t          j        j        dk    fd            j        d d         }t          j        |j        k    fd           t          j        t          j        j	                  fd           t          j         j	        j	        k     fd           t                     \  }}t          j        |t          |d	          j	        j        
          S )Nztorch.linalg.ldl_solver   c                      d j          dS )NzMtorch.linalg.ldl_solve: Expected B to have at least 2 dimensions, but it has r  r  )r  s   r/   rT   z'linalg_ldl_solve_meta.<locals>.<lambda>8  !    6&6 6 6 r1   r   c                      d j          dS )Nzjtorch.linalg.ldl_solve: Expected LD.shape[:-1] and pivots.shape to be the same, but got pivots with shape  insteadr   r  s   r/   rT   z'linalg_ldl_solve_meta.<locals>.<lambda>@  %    @)/@ @ @ r1   c                      d j          S )Nz<torch.linalg.ldl_solve: Expected pivots to be integers. Got r}   r  s   r/   rT   z'linalg_ldl_solve_meta.<locals>.<lambda>G  s    ]v|]] r1   c                  (    dj          d j          S )Nz!torch.linalg.ldl_solve: LD dtype z does not match b dtype r}   )r  r  s   r/   rT   z'linalg_ldl_solve_meta.<locals>.<lambda>K  s    _BH__VWV]__ r1   Fr  r  )r   r   r  rE   rV   r   r   r?   is_integer_dtyperM   _linalg_broadcast_batch_dimsr  r   rp   )r  r  r  r  expected_pivots_shapeB_broadcast_sizerB   s   ```    r/   linalg_ldl_solve_metar  *  sc    b2333278881b":;;;	L	!	
 	
 	
 	
   HSbSM	L-	
 	
 	
 	
   
Lv|,,]]]]   
L
AG_____   7q"==a*+;uMMMgx	   r1   Pr  )pivotr  c                    t          j         j        dk     fd           t           j                  }|d         }|d         }t          ||          }||d<   |r                     |          }n                     dg          }||d<                        |          }||d<   ||d<                        |          }|||fS )Nr   c                      d j          dS )Nz@linalg.lu: Expected tensor with 2 or more dimensions. Got size: r  r  r  s   r/   rT   z linalg_lu_meta.<locals>.<lambda>[  s    dSTSZddd r1   r  r   r   )rE   rV   r   r   r   r  r   )	r  r  sizesr:  r   r9  r  r  r  s	   `        r/   linalg_lu_metar  V  s     
L	!dddd  
 MMEb	Ab	AAq		AE"I KKKKE"I	EAE"IE"I	EAa7Nr1   LU)r  r  c                    t          j         j        dk     fd           t           j                  }|d         }|d         }t          j        |t          |d           j         j                  }|	                                 t          ||          |d<                        |t           j                  }|	                                                      |t           j                  }|||fS )	Nr   c                      d j          dS )NzFtorch.lu_factor: Expected tensor with 2 or more dimensions. Got size: r  r  r  s   r/   rT   z*linalg_lu_factor_ex_meta.<locals>.<lambda>|  s    jYZY`jjj r1   r  r   Fr  r  r}   )rE   rV   r   r   r   r  r   rM   rp   r  r  r   r  )	r  r  r  r  r:  r   r   r  r  s	   `        r/   linalg_lu_factor_ex_metar#  r  s     
L	!jjjj  
 MMEb	Ab	A		*5EBBBgx	
 
 
B 
IIKKKAq		E"I[[ei[00F 
IIKKK;;uEI;..Dvtr1   )r   adjointr$  c                    t           d           t          j         j        j        k     fd           t          j        j        t          j        k    d            t           d           t           |d           t          j                             d                              d          k    d            t          j         j        d d         j        k    fd           t                     \  }}t          j
        |t          ||           j        j        	          }|                                d
k    r*|s(|                                r|                                }|S )Nztorch.linalg.lu_solvec                  *    dj          d j          dS )NzPlinalg.lu_solve: Expected LU and B to have the same dtype, but found LU of type  and B of type r  r}   )r  r   s   r/   rT   z&linalg_lu_solve_meta.<locals>.<lambda>  s6    O$&HO O=>WO O O r1   c                      dS )NzElinalg.lu_solve: pivots should be a Tensor of scalar type torch.int32r5   r5   r1   r/   rT   z&linalg_lu_solve_meta.<locals>.<lambda>  s    W r1   zlinalg.lu_solver   c                      dS )NzYlinalg.lu_solve: Number of pivots per batch should be same as the dimension of the matrixr5   r5   r1   r/   rT   z&linalg_lu_solve_meta.<locals>.<lambda>  s    k r1   c                      d j          dS )Nzclinalg.lu_solve: Expected LU.shape[:-1] and pivots.shape to be the same, but got pivots with shape r  r  r  s   r/   rT   z&linalg_lu_solve_meta.<locals>.<lambda>  r  r1   r  r  r   )r   rE   rV   rM   r  r   r  r   r   r  r  r   rp   r   r   conj)r   r  r  r   r$  r  rB   r  s   ```     r/   linalg_lu_solve_metar,    s    26777	L
AG	
 	
 	
 	
 	
   
L	!WW   b1222b!T#4555	L
v{{2&kk   
L
"%	
 	
 	
 	
   7q"==a *+;4xPPPgx	  F ||~~4 	#[[]]FMr1   unpack_dataunpack_pivotsc                 D    t          j         j        dk     fd           |r)t          j        |j        t           j        k    d            t           j                  }|d         }|d         }t          ||          }||d<   |r                     |          }n                     dg          }|r:||d<                        |          }	||d<   ||d<                        |          }
n,                     dg          }	                     dg          }
||	|
fS )Nr   c                      d j          dS )NzFtorch.lu_unpack: Expected tensor with 2 or more dimensions. Got size: r  r  )r   s   r/   rT   z lu_unpack_meta.<locals>.<lambda>  s    kY[Yakkk r1   c                      dS )Nztorch.lu_unpack: LU_pivots is expected to be a contiguous tensor of torch.int32 dtype.
Note: this function is intended to be used with the output produced by torch.linalg.lu_factorr5   r5   r1   r/   rT   z lu_unpack_meta.<locals>.<lambda>  s    p r1   r  r   r   )	rE   rV   r   rM   r  r   r   r  r   )r   r  r-  r.  r  r:  r   r9  r  r  r  s   `          r/   lu_unpack_metar2    s?    
L
1kkkk    
LEK' 	
 	
 	
 NNEb	Ab	AAq		AE"I LLLL! b	LLb	b	LLLL!LL!a7Nr1   modec                 ~      dk    rd}d}n. dk    rd}d}n# dk    rd}d}nt          j        d fd           ||fS )NreducedTcompleteFrc                      d  dS )Nzqr received unrecognized mode 'z=' but expected one of 'reduced' (default), 'r', or 'complete'r5   )r3  s   r/   rT   z _parse_qr_mode.<locals>.<lambda>  s"    O$ O O O r1   rE   rV   )r3  	compute_qr5  s   `  r/   _parse_qr_moder;    s~    y							   	
 	
 	
 gr1   QRr5  c                 T   t          | d           t          | d           t          |          \  }}| j        d         }| j        d         }t	          ||          }|rXt          | j                  }|r|n||d<   |                     |          }|                    |t          |d                     n|                     dg          }t          | j                  }	|s|s|n||	d<   |                     |	          }
|
                    |	t          |	d                     ||
fS )Nz	linalg.qrr  r   Fr  r   )	r  r   r;  r   r  r   r   r  r   )r  r3  r:  reduced_moder:  r   r9  Q_shaper<  R_shaper=  s              r/   linalg_qr_metarB    s+    ![!!!1k***,T22I|	A	AAq		A qw--'.aaQKK  	g:7eTTTUUUUKK 17mmG#;9;!!!GBK	GAMM'6w%PPPQQQa4Kr1   sign	logabsdetc                    t          | d           t          | dd           | j        }|                     |d d                   }|                     |d d         t	          | j                            }t          j        |t          |d          | j        | j	                  }|                     |d d         t          j
                  }||||fS )Nzlinalg.slogdetFr  r}   r  r   )r   r   r   r   rO   rM   rE   r  r   rp   r  )r  r   rC  rD  r   r  s         r/   _linalg_slogdetrF  *  s     a)***1.666GE;;uSbSz""DE#2#Joag.F.FGGI		*5%88gx	
 
 
B [[ss5;[77FB&&r1   full_matrices
compute_uvdriverc                    t          | d           t          | d           t          | j        d d                   }| j        d         }| j        d         }t	          ||          }|r|||r|n|gz   }|                     |          }	|	                    |t          |d                     ||r|n||gz   }
|                     |
          }t          |           dk    }|                    |
t          |
|                     n,|                     dg          }	|                     dg          }|                     ||gz   t          | j
                            }|	||fS )	Nz
linalg.svdr  r   Fr  cudar   r}   )r  r   r   r   r  r   r  r   device_hintrO   rM   )r  rG  rH  rI  r   r:  r   r9  U_shaper  V_shapeVis_cudaSs                 r/   _linalg_svd_metarR  >  sl    !\"""1l+++agcrcl##J	A	AAq		A #<111==KK  	g:7eTTTUUU] 91==KK  
 a..F*	g:7gVVVWWWW KKKK 	
J!$OAG,D,DEEAa7Nr1   arg1arg2c                 V   | j         d d         }|j         d d         }t          ||          }t          |          }||                     d          |                     d          gz  }t          |          }||                    d          |                    d          gz  }||fS )Nr  r   )r   r#   r   r   )rS  rT  arg1_batch_sizesarg2_batch_sizesexpand_batch_portionarg1_expand_sizearg2_expand_sizes          r/   r  r  d  s    
 z#2#z#2#,-=?OPP0112		"660112		"66---r1   c                     |rt          | ||           t          | |          \  }}|| j        k    r| n|                     |          }||j        k    r|n|                    |          }||fS r,   )r  r  r   expand)rS  rT  r  rY  rZ  arg1_broadcastedarg2_broadcasteds          r/   r  r  u  s      1tT4000)EdD)Q)Q&& !DJ..DKK@P4Q4Q  !DJ..DKK@P4Q4Q  ---r1   r   c                 v    | j         d d         }|j        dk    p| j        dz
  |j        k    o
|j         |k    }|S )Nr   r   )r   r   )r   r   expected_batched_rhs_shapevector_cases       r/   linalg_solve_is_vector_rhsrb    sK    !&SbS!1*/ 
Q%*$R8R)R  r1   )r   r  r  r   r  r  c                    t           d           t          j         j        j        k     fd           t	                     }|r                    d          n}	t           |	|d           t          |	           \  }
}t          j        |p| d            |r
|
d d         n|
}t          j        |t          ||           j        j
                  } j        } j        }t          j        |t          |d           j         j
                  }                     |d d         t          j                  }                     |d d         t          j                  }||||f}||||f}t          d	 |D                       rjt!          ||          D ]Y\  }}t#          ||j                   |                    |j        |                                           t)          ||d
           Z|S )Nzlinalg.solvec                  *    d j          dj          dS )NzKlinalg.solve: Expected A and B to have the same dtype, but found A of type r'  r  r}   )r  r  s   r/   rT   z"_linalg_solve_ex.<locals>.<lambda>  s0    9w9 9'(w9 9 9 r1   r   c                      dS )Nzlinalg.solve: Vector broadcasting of the left hand side is not supported for left=False. In this case linalg.solve is equivalent to B / A.squeeze(-1)r5   r5   r1   r/   rT   z"_linalg_solve_ex.<locals>.<lambda>  s    K r1   r  Fr}   r  c              3      K   | ]}|d uV  	d S r,   r5   )r;   r<   s     r/   rb   z#_linalg_solve_ex.<locals>.<genexpr>  s&      
&
&Q1D=
&
&
&
&
&
&r1   )	copy_fromcopy_toexact_dtype)r   rE   rV   rM   rb  	unsqueezer  r  r  r   rp   r   r   r   r  allzipr   r  r   r!   )r  r  r   r  r  r   r  r  ra  B_B_broad_shaperB   result_shaperesult_r   r   LU_pivots_info_r   resr7  os   ``                     r/   _linalg_solve_exrv    s<    1n---	L	17	
 	
 	
 	
 	
   -Q22K'	.RQBaT>2223B::M1	LK	
 	
   *5G="%%-L!*<TBBgx	  G GE6D

*5%88gx	  C kk%*EKk88GKKcrc
%+K66E2vt
$CC%
(C

&
&#
&
&
&&& FSMM 	F 	FDAqa)))MM!'188::...QuEEEEEJr1   )r   unitriangularr   rw  r   c                
   ||                      dg          }t          |t                    sJ t          | ||d           t	          || d           \  }}|                    dd                                          o|                                }|rt          ||j	                  }nYt          ||j	                  rD|                    |                    dd          j	                   |                    dd           |S )Nr   zlinalg.solve_triangularr  r   )r   r_   r   r  r  r  r6  is_conjr   r   r    resize_
transpose_)	r  r  r  r   rw  r   rm  A_avoid_copy_As	            r/   linalg_solve_triangular_metar~    s     {kk1#c:&&&&&aD";<<<.q!T::FB<<B''5577HBJJLLL #RX..  RX.. 	#KKR,,2333NN2r"""Jr1   XM)ri  r  c                     t          j         j        dk     fd           t          j        j        dk    fd           t           d           j        t           j        k    rvt                     \  }}t          j        |t          |d           j	         j
                  }t          j        |t          |d          j	        j
                  }nkj        t           j        k    sj        t           j        k    r+t          j                   }                     dg          }nt          j        dd	            ||fS )
Nr   c                      d j          dS )NzMtorch.triangular_solve: Expected b to have at least 2 dimensions, but it has r  r  r   s   r/   rT   z'triangular_solve_meta.<locals>.<lambda>  s!    9)9 9 9 r1   c                      d j          dS )NzMtorch.triangular_solve: Expected A to have at least 2 dimensions, but it has r  r  r  s   r/   rT   z'triangular_solve_meta.<locals>.<lambda>  r  r1   triangular_solveFr  r  r   c                      dS )Nz+triangular_solve: Got an unexpected layout.r5   r5   r1   r/   rT   z'triangular_solve_meta.<locals>.<lambda>  s    $Q r1   )rE   rV   r   r  ro   stridedr  r  r   rM   rp   
sparse_csr
sparse_bsrr   r   )	r   r  r  r  rw  self_broadcast_sizeA_broadcast_sizesolutioncloned_coefficients	   ``       r/   triangular_solve_metar    s    
L	Q	
 	
 	
 	
   
L	!	
 	
 	
 	
   4$6777x5=  0LTST0U0U--&$./BeTTT*;	
 
 
 #0!./?5QQQ'8	
 
 
 
U%	%	%U5E)E)E#D))!^^QC00UQQRRR'''r1   c                 ~   t          | d           t          | d           |                     | j        d d                   }|                     | j                  }|                    | j        t          | j        d                     |                     | j        d d         t          j                  }|||fS )Nz
linalg.detr  Fr  r   r}   r  )r  detr   r  s       r/   _linalg_det_metar    s    a&&&1l+++
++agcrcl
#
#C	
QW		BNN1775QQQRRR[["U[[99FF?r1   c                     t          j         j        dk    d            t          j        j        dk    d            |rdndt          j        j                 j        d         k    fd           t          j        j                  j        d         k    fd           t          j        j        d          j        d         k    d            t          j         j        j        z
  d	k     fd
           t          j         j        j        k     fd            j        dk    re j        d d         }j        d d         t          j        |k    fd           j        d d         t          j        |k    fd           t          j        j         j        k     fd           t          j        j         j        k     fd           t          d d           t          d d           t          j        j        t          j        d          j        j                  S )Nr   c                      dS )Nz3torch.ormqr: input must have at least 2 dimensions.r5   r5   r1   r/   rT   zormqr.<locals>.<lambda>0      !V r1   c                      dS )Nz3torch.ormqr: other must have at least 2 dimensions.r5   r5   r1   r/   rT   zormqr.<locals>.<lambda>3  r  r1   r  r   c                      d  dS )Ntorch.ormqr: other.shape[z0] must be greater than or equal to tau.shape[-1]r5   left_size_conditions   r/   rT   zormqr.<locals>.<lambda>9  s    q,?qqq r1   c                      d  dS )Nr  z"] must be equal to input.shape[-2]r5   r  s   r/   rT   zormqr.<locals>.<lambda>=  s    c,?ccc r1   c                      dS )NzHtorch.ormqr: tau.shape[-1] must be less than or equal to input.shape[-1]r5   r5   r1   r/   rT   zormqr.<locals>.<lambda>B  r  r1   r   c                  (    dj          d j          S )Nz[torch.ormqr: Expected tau to have one dimension less than input, but got tau.ndim equal to r  r  r  s   r/   rT   zormqr.<locals>.<lambda>G  r  r1   c                  (    dj          d j          S )Nzhtorch.ormqr: Expected other to have the same number of dimensions as input, but got other.ndim equal to r  r  r   r   s   r/   rT   zormqr.<locals>.<lambda>N  s1    `+0:` `SXS]` ` r1   c                      d  S )NzWtorch.ormqr: Expected batch dimensions of tau to be equal to input.shape[:-2], but got r5   r  s   r/   rT   zormqr.<locals>.<lambda>Y  r  r1   c                      d  S )NzYtorch.ormqr: Expected batch dimensions of other to be equal to input.shape[:-2], but got r5   )actual_batch_other_shapes   r/   rT   zormqr.<locals>.<lambda>b  s    Q6NQ Q r1   c                  (    d j          dj          S )NzPtorch.ormqr: Expected input and tau to have the same dtype, but input has dtype z and tau has dtype r}   r  s   r/   rT   zormqr.<locals>.<lambda>j  s0    O#(;O OCF9O O r1   c                  (    d j          dj          S )NzRtorch.ormqr: Expected input and other to have the same dtype, but input has dtype z and other has dtype r}   r  s   r/   rT   zormqr.<locals>.<lambda>q  s0    S#(;S SEJ[S S r1   ztorch.ormqrr  r   Fr  r  )	rE   rV   r   r   rM   r  r  r   rp   )	r   r  r   r   r  expected_batch_shaper  r  r  s	   ```   @@@r/   ormqrr  &  s    
L
aVV   
L
aVV   !%,"""	L'(CIbM9qqqq   
L'(EKO;cccc  
 
L	"R(ZZ  
 
L
SX"	
 	
 	
 	
 	
   
L
ej 	
 	
 	
 	
 	
   zA~~${3B3/!$3B3"&::   	
 	
 	
 $);ss#3 $(<<   	
 	
 	
 
L	U[ 	
 	
 	
 	
 	
   
Lu{"	
 	
 	
 	
 	
   M3u555M5%999[*5;%HHHk|	   r1   c                    t          j        t                    dz  k    fd            j        }|dz   k    }|}| }|r/t	          d|          D ]}|o                     |          dk    }n.t	          d|          D ]}|o                     |          dk    }t          j        |p| fd           d S )Nr   c                  4    dd z   dt                     S )Nzpadding size is expected to be r   z, but got: r   )rs   paddings   r/   rT   z,_padding_check_valid_input.<locals>.<lambda>  s"    T!c'TTc'llTT r1   r   r   c                  0    d dz    d dz    dj          S )Nz	Expected r   zD or r   zcD (batch mode) tensor with possibly 0 batch size and other non-zero dimensions for input, but got: r  )rs   r   s   r/   rT   z,_padding_check_valid_input.<locals>.<lambda>  sC    Oa O OcAg O OAFO O r1   )rE   rV   r   r   r   r   )r   r  rs   	input_dimis_batch_modevalid_batch_modevalid_non_batch_moder   s   ```     r/   _padding_check_valid_inputr    s#   	LGCTTTTT  
 
I#'*M$,, Oq)$$ 	G 	GA/FEJJqMMQ4F	G q)$$ 	O 	OA#7#NEJJqMMQ<N   
L00	
 	
 	
 	
 	
    r1   c                   	
 d}dd} j         dk    r                     d          }dz  |dz  }t           |d           |\  	
                     |          }                               	z   
z   |r%t          j        	k     o
k      	
fd           t          j        dk    fd            j         dk    r                     |f          S                      ||f          S )Nr   r   r   r   c                  *    d d d  dj          S NzcArgument #4: Padding size should be less than the corresponding input dimension, but got: padding (rf   ) at dimension 
 of input r  dim_wr   pad_lpad_rs   r/   rT   z_pad1d_common.<locals>.<lambda>  M    c%*c c.3c cDIc cUZU`c c r1   c                      d  d S )Nz
input (W: z%) is too small. Calculated output W: r5   )input_woutput_ws   r/   rT   z_pad1d_common.<locals>.<lambda>  s    UWUU8UU r1   r   )r   r   r  rE   rV   r   )r   r  is_reflection	dim_planenbatchnplaner  r  r  r  r  s   `     @@@@@r/   _pad1d_commonr    sI   IEFzQA
Q	ug15555LE5ZZ	""FjjG&H 
GO/      	
 	
 	
 
LAUUUUU  
 zQ12229:::r1   c                 &    t          | |d          S NTr  r  r   r  s     r/   meta_reflection_pad1dr         t<<<<r1   c                 &    t          | |d          S NFr  r  r  s     r/   meta_replication_pad1dr         u====r1   c                    d|s't          j        t          |          dk    d            j        dk    rdz  |\                                }|z   z   |r%t          j        |k     o|k     fd           t          j                                       k     fd                               j                  S )Nr   r   c                      dS )Nz padding size is expected to be 2r5   r5   r1   r/   rT   z(_pad1d_backward_common.<locals>.<lambda>  s    0R r1   r   c                  *    d d d  dj          S r  r  r  s   r/   rT   z(_pad1d_backward_common.<locals>.<lambda>  r  r1   c                  :    d d                                 S Nz(grad_output width unexpected. Expected: , Got: r   r  grad_outputr  s   r/   rT   z(_pad1d_backward_common.<locals>.<lambda>  '    e8eeKL\L\]bLcLcee r1   rE   rV   r   r   r   r   r   )	r  r   r  r  r  r  r  r  r  s	   ``   @@@@r/   _pad1d_backward_commonr    s   E TS\\Q&(R(RSSSzQ
LE5jjG&H 
GO/      	
 	
 	
 
LK$$U+++eeeeee  
 ??5;'''r1   
grad_inputc                 (    t          | ||d          S r  r  r  r   r  s      r/   meta_reflection_pad1d_backwardr    s     "+ugTRRRRr1   c                 (    t          | ||d          S r  r  r  s      r/   meta_replication_pad1d_backwardr    s     "+ugUSSSSr1   c                   	
 ddd}d}t           |d            j        }|dk    r$                     d          }dz  dz  |dz  }|\                       |          }                               	                               
	z   z   
z   z   |rJt          j        
k     o
k      fd           t          j        	k     o	k      fd           t          j        dk    pdk    	
fd            j        d	k    r                     |f          S                      ||f          S )
Nr   r   r   r      c                  *    d d d  dj          S r  r  r  s   r/   rT   z_pad2d_common.<locals>.<lambda>  r  r1   c                  *    d d d  dj          S NzcArgument #6: Padding size should be less than the corresponding input dimension, but got: padding (rf   r  r  r  dim_hr   pad_bpad_ts   r/   rT   z_pad2d_common.<locals>.<lambda>  r  r1   c                       d  d d d S )Nz
input (H:  W: z%) is too small. Calculated output H: r5   )input_hr  output_hr  s   r/   rT   z_pad2d_common.<locals>.<lambda>&  s?    = = =g = =$,= =2:= = r1   r   r  r   r   rE   rV   r   )r   r  r  
dim_slicesr  r   r  r  r  r  r  r  r  r  r  r  r  s   `      @@@@@@@@@@r/   _pad2d_commonr    s   EEJFug15555:DqyyA

a
!(E5%ZZ
##FjjGjjG&H&H 
GO/      	
 	
 	
 	GO/      	
 	
 	
 
LA&Q	
 	
 	
 	
 	
 	
 	
   zQ(;<<<(CDDDr1   c                 &    t          | |d          S r  r  r  s     r/   meta_reflection_pad2dr  2  r  r1   c                 &    t          | |d          S r  r  r  s     r/   meta_replication_pad2dr  8  r  r1   c                     ddd}d}|j         }|                                dk    r|d         }dz  dz  |dz  }|\  }}}}	||         }
|         }|         }||z   |	z   ||z   |z   t          j                                       k     fd           t          j                                       k     fd           |                    |j                   S )Nr   r   r   r  c                  :    d d                                 S r  r   r  s   r/   rT   z%meta_pad2d_backward.<locals>.<lambda>^  r  r1   c                  :    d d                                 S Nz)grad_output height unexpected. Expected: r  r   r  r  r  s   r/   rT   z%meta_pad2d_backward.<locals>.<lambda>b  '    fHff[M]M]^cMdMdff r1   )r   rs   rE   rV   r   r   )r  r   r  r  r  rS   r  r  r  r  r  r  r  r  r  r  r  s   `            @@@@r/   meta_pad2d_backwardr  >  s1    EEIFJxxzzQA

Q	!(E5%	"FGG&H&H	LK$$U+++eeeeee   
LK$$U+++ffffff   >>$*%%%r1   c          	      R   	
 d	ddd}t           |d            j        dk    }|r)                     d          }	dz  	dz  dz  |dz  }|\                       |          }                               
                                                    	          
z   z   z   z   z   z   |rot          j        k     ok     	 fd           t          j        k     ok      fd           t          j        
k     o
k      fd	           t          j        dk    pdk    pdk    
fd
           |r                     ||f          S                      |f          S )Nr   r   r   r   r      c                  *    d d d  dj          S r  r  r  s   r/   rT   z_pad3d_common.<locals>.<lambda>  r  r1   c                  *    d d d  dj          S r  r  r  s   r/   rT   z_pad3d_common.<locals>.<lambda>  r  r1   c                  *    d d d  dj          S )NzcArgument #8: Padding size should be less than the corresponding input dimension, but got: padding (rf   r  r  r  )dim_dr   pad_bkpad_fs   r/   rT   z_pad3d_common.<locals>.<lambda>  sM    d%*d d.4d dEJd dV[Vad d r1   c                  ,    d  d d d d d S )Nz
input (D:  H: r  z%) is too small. Calculated output D: r5   )input_dr  r  output_dr  r  s   r/   rT   z_pad3d_common.<locals>.<lambda>  se    K K Kg K K7 K K$,K K2:K K@HK K r1   r  )r   r  r  r  
batch_moder  r  r  r  r  r  r  r  r  r  r  r  r   r  r  r  r  s   `      @@@@@@@@@@@@@@@r/   _pad3d_commonr  g  s   EEEIug15555qJ A


Q	07-E5%vZZ	""FjjGjjGjjG'H&H&H 
GO/      	
 	
 	
 	GO/      	
 	
 	
 	GO0 0      	
 	
 	
 
LA7Q7(a-	
 	
 	
 	
 	
 	
 	
 	
 	
    G(HMNNN(HEFFFr1   c                 &    t          | |d          S r  r  r  s     r/   meta_reflection_pad3dr
    r  r1   c                 &    t          | |d          S r  r	  r  s     r/   meta_replication_pad3dr    r  r1   c                     t          j        t          |          dk    d            |j        dk    sJ  j        |j        k    sJ ddd|j        dk    rdz  dz  dz  |\  }}}}}}|                              }	|                              }
|                              }|	|z   |z   |
|z   |z   ||z   |z   t          j                                       k     fd           t          j                                       k     fd           t          j                                       k     fd	           |                    |j                  S )
N   c                      dS )Nz padding size is expected to be 6r5   r5   r1   r/   rT   z%meta_pad3d_backward.<locals>.<lambda>  s    ,N r1   r   r   r   r  c                  :    d d                                 S r  r   r  s   r/   rT   z%meta_pad3d_backward.<locals>.<lambda>  r  r1   c                  :    d d                                 S r  r   r  s   r/   rT   z%meta_pad3d_backward.<locals>.<lambda>  r  r1   c                  :    d d                                 S )Nz(grad_output depth unexpected. Expected: r  r   )r  r  r  s   r/   rT   z%meta_pad3d_backward.<locals>.<lambda>  r  r1   r  )r  r   r  r  r  r  r  r  r   r  r  r  r  r  r  r  r  r  s   `           @@@@@@r/   meta_pad3d_backwardr    s    
LW"$N$NOOO:>>>>uz))))EEEzQ


07-E5%vjjGjjGjjG'H&H&H	LK$$U+++eeeeee   
LK$$U+++ffffff   
LK$$U+++eeeeee  
 ??5;'''r1   r   pc                 j   t          j        |                                 d            |                     d          }|dk    r4|                     dg                              t           j                  S |                     ||dz
  z  dz  f                              t           j                  S )Nc                      dS )Nz(_pdist_forward requires contiguous inputr5   r5   r1   r/   rT   z%meta__pdist_forward.<locals>.<lambda>  s    &P r1   r   r   r   r   )rE   rV   r6  r   r   r   r  )r   r  r   s      r/   meta__pdist_forwardr    s     
LPP   			!AAvv~~qc""%%E4R%SSS~~qAE{a/122558 6 
 
 	
r1   gradpdistc                     t          j        |                                d            t          j        |                                d            t          j        |t           j                  S )Nc                      dS )Nz._pdist_backward requires self to be contiguousr5   r5   r1   r/   rT   z&meta__pdist_backward.<locals>.<lambda>  s    &V r1   c                      dS )Nz/_pdist_backward requires pdist to be contiguousr5   r5   r1   r/   rT   z&meta__pdist_backward.<locals>.<lambda>  s    'X r1   r   )rE   rV   r6  r   r  )r  r   r  r  s       r/   meta__pdist_backwardr    ss     
LVV   
LXX   D0NOOOOr1   )r%  r$  c                   	
                      d          }                     d          }                     d          }                     |||f           t          j                                        dk    d            t          j                                        dk    d            t          j         j        j        cxk    o
j        k    nc  fd           j        }j        	|d         
|d         t          j        	d         
k    o	d         k    	
fd                                                                            S )	Nr   r   r   r   c                      dS Nzbatch1 must be a 3D tensorr5   r5   r1   r/   rT   zmeta_baddbmm.<locals>.<lambda>      ,H r1   c                      dS Nzbatch2 must be a 3D tensorr5   r5   r1   r/   rT   zmeta_baddbmm.<locals>.<lambda>  r!  r1   c                  8    dj          d j          dj          S )Nz+Input dtypes must be the same, got: input: z
, batch1: z
, batch2: r}   )batch1batch2r   s   r/   rT   zmeta_baddbmm.<locals>.<lambda>  s*    zdjzzTZT`zzlrlxzz r1   c            	      :    d d d d          d d          d	S Nz@Expected size for first two dimensions of batch2 tensor to be: [rf   z] but got: [r   r   ].r5   batch2_sizesbscontraction_sizes   r/   rT   zmeta_baddbmm.<locals>.<lambda>  sY    [[ [&[ [4@O[ [GSTU[ [ [ r1   )r   r\  rE   rV   rs   rM   r   r   )r   r%  r&  r%  r$  dim1dim2dim3batch1_sizesr+  r,  r-  s   ```      @@@r/   meta_baddbmmr2    sl    ;;q>>D;;q>>D;;q>>D;;dD)**D	L"$H$HIII	L"$H$HIII	L
fl2222fl2222zzzzzz   <L<L	aB#A	LQ2E,q/5E"E	
 	
 	
 	
 	
 	
   >>$))++&&&r1   c                N    t          j        |                                           S r,   rE   r   r   r   r   s     r/   meta_bernoullir6    s!     D!!,,...r1         ?c                     | S r,   r5   r   r  r   s      r/   meta_bernoulli_r:        Kr1   c                 N    t          j        |                                           S r,   r4  r9  s      r/   meta_bernoulli_pr=  $  s!     D!!,,...r1   c                 *    t          j        |           S r,   rE   r   r5  s     r/   meta_poissonr@  *  s     D!!!r1   c                     t          j        |
|                                 k     d            t          j        | t           j                  }t          j        |           |fS )Nc                      dS )NzJError in fused_moving_avg_obs_fake_quant_cpu: ch_axis must be < self.dim()r5   r5   r1   r/   rT   z6meta__fused_moving_avg_obs_fq_helper.<locals>.<lambda>B      \ r1   r}   )rE   rV   rs   r   bool)r   observer_onfake_quant_onrunning_minrunning_maxscale
zero_pointaveraging_const	quant_min	quant_maxch_axisper_row_fake_quantsymmetric_quantmasks                 r/   $meta__fused_moving_avg_obs_fq_helperrR  0  s]      
L$((**\\   D
333DT""D))r1   c                 L   t          j        |                                 dk    d            t          j        |                                dk    d            | j        \  |j        \  t          j        k    fd           |                               S )Nr   c                      dS )Nza must be 2Dr5   r5   r1   r/   rT   zmeta_mm.<locals>.<lambda>K      ~ r1   c                      dS )Nzb must be 2Dr5   r5   r1   r/   rT   zmeta_mm.<locals>.<lambda>L  rU  r1   c            	      "    d d  d d d	S )Nz/a and b must have same reduction dim, but got [rf   z] X [r)  r5   )M1M2Nr  s   r/   rT   zmeta_mm.<locals>.<lambda>Q  s-    [![[r[[PR[[VW[[[ r1   )rE   rV   rs   r   r   )abrX  rY  rZ  r  s     @@@@r/   meta_mmr]  H  s     
LA55666	LA55666GEArGEB	L
b[[[[[[[   ;;q!r1   c                      |r.t           fdt           j                  D                       S t          j         j                  S )Nc              3   >   K   | ]}|vrj         |         nd V  dS )r   Nr  )r;   r   dimsr   s     r/   rb   z+_compute_reduction_shape.<locals>.<genexpr>X  s5      UUqatmmTZ]]UUUUUUr1   )rU   r   r   r?   compute_reduction_output_shaper   )r   r`  rb  s   `` r/   r`  r`  V  sR     VUUUUUE$)DTDTUUUUUU/
DAAAr1   strc                     t          | t          j        j                  r| j        j        S t          | d          r1t          | j        d          r| j        j        dk    r| j        j        S dS )Nrp   rh   rn   rK  )r_   rE   _subclasses
FakeTensorfake_devicerh   hasattrrp   )r	  s    r/   rL  rL  a  sn    &%+677 	!&&!!FM6** M&((}!!vr1   input_tensorr   r  dilationis_transposedgroupsoutput_paddingc                    dt           dt           dt           dt           dt           dt           fd}dt           dt           dt           dt           dt           dt           dt           fd	}	|j        d
d          }
| j        d
d          }|r||j        d         z  }n;|j        d         }|j        d         |z  | j        d         k    rt          d          | j        d         |g}t          |t                    r|gt          |          z  }n,t          |          dk    r|d         gt          |          z  }t          |t                    r|gt          |          z  }n,t          |          dk    r|d         gt          |          z  }t          |t                    r|gt          |          z  }n,t          |          dk    r|d         gt          |          z  }d }|rXt          |t                    r|gt          |          z  }n/t          |          dk    r|d         gt          |          z  }n|}t          t          |                    D ]}|rH|                     |	||         ||         ||         |
|         ||         ||                              L|                     |||         ||         ||         |
|         ||                              |S )Nlnr  r   r9  rz  r@  c                 6    | d|z  z   ||dz
  z  z
  dz
  |z  dz   S )a  
        Formula to apply to calculate the length of some dimension of the output

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

        Args:
            ln: length of the dimension
            p: padding in that dim
            d: dilation in that dim
            k: kernel size in that dim
            s: stride in that dim
        Returns:
            The output length
        r   r   r5   )rn  r  r   r9  rz  s        r/   _formulaz+calc_conv_nd_return_shape.<locals>._formulax  s.     QU
Q!a%[(1,2Q66r1   r-   c                 <    | dz
  |z  d|z  z
  ||dz
  z  z   |z   dz   S )a  
        Formula to apply to calculate the length of some dimension of the output
        if transposed convolution is used.
        See: https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html

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

        Returns:
            The output length
        r   r   r5   )rn  r  r   r9  rz  r-   s         r/   _formula_transposedz6calc_conv_nd_return_shape.<locals>._formula_transposed  s2    " Q!|a!e#a1q5k1B6::r1   r   r   r   zInvalid channel dimensions)r  r   r   r_   r   r   r   r   )rh  r  r   r  ri  rj  rk  rl  rp  rr  kernel_sizer`  out_channels	ret_shapeoutput_padding_listr   s                   r/   calc_conv_nd_return_shaperw  n  s.   7S 7S 7S 7S 7S 7S 7 7 7 7"; ; ; ; ; ; ;QT ; ; ; ;& ,qrr"Kabb!D =Q/|A<?V#|'9!'<<<;<<<#A&5I&'"" )CII%	V		)s4yy('7## +)c$ii'	W		1:,T*(G$$ -:D		)	X!		QK=3t99,/3 1ng.. 	1#1"2SYY">  A%%#1!#4"5D		"A"03t99   	##GAJQKN1I'* 	 	 	 	 a'!*hqk;q>6RS9UU    r1   c                 \    t           j                            |           t           j        k    S r,   rE   _prims_commonsuggest_memory_formatchannels_lasttens    r/   is_channels_lastr    s!    44S99U=PPPr1   running_meanrunning_vartrainingexponential_average_factorepsilonc                 v     j         }||j         n|j         }	||j         n|j         }
 fd}                     |                               |                      }|r+                     |	          }                     |
          }n*                     d          }                     d          }|||fS )Nc                      t                     rt          j        S                      t          j                  rt          j        S t          j        S rB  )r  rE   r|  r6  r   )rh  s   r/   pick_memory_formatz2meta_miopen_batch_norm.<locals>.pick_memory_format  sJ    L)) 	'&&%%E4K%LL 	+**&&r1   r   r   )r   r   r   )rh  r  r  r  r  r  r  r  r   save_mean_shapesave_var_shaper  r   	save_meansave_vars   `              r/   meta_miopen_batch_normr    s     "I -9,Dl((&,O*5*A[&&v|N' ' ' ' ' 
 
 
+
+
.
.=O=O=Q=Q
.
R
RC 0 **?;;	)).99 **400	))$//	8##r1   c	           
            fd}	t           ||||||r|nd           }
d}d}                     |          dk    rd|
|<                        |
          }|                     |	                      }|S )Nc                  j   t                     dk    r+t                     st                    rt          j        S nt                     rt          j        S                      t          j                  rt          j        S                      t          j                  rt          j        S d S NrK  r   )rL  r  rE   r|  r6  r   preserve_format)rh  r  s   r/   r  z%meta_conv.<locals>.pick_memory_format		  s    |$$..-- +1A&1I1I +**+  -- +**%%E4K%LL 	)**''e6K'LL 	)((	) 	)r1   r   r   r   )rw  r   r   r   )rh  r  r  r   r  ri  rj  rl  rk  r  	shape_outinput_channels_dimoutput_channels_dimr   s   ``            r/   	meta_convr    s    
) 
) 
) 
) 
) 
) *'1T	 	I +,,11)*	%&

 
 
+
+C
&&1133&
4
4CJr1   mkldnnc
           
          t          | ||||d|g           }
|                     |
          }t          j        }|                                 dk    rt          j        }|                    |          }|S )NFr  r   )rw  r   rE   r|  rs   channels_last_3dr   )rh  r  r  r  r   ri  rk  attrscalars	algorithmr  r   out_memory_formats                r/   meta_mkldnn_convolution_defaultr  /	  s{     .&&'8UFB
 
	 $$Y//!/"" % 6ff#4f55
r1   c                 f    |                      g | j        d d         |j        d         R           S Nr   r   r   r   )rh  r  r  r  r  r  s         r/   meta_linear_pointwise_defaultr  F	  s8     %%&Q(:3B3(?&Qa&Q&QRRRr1   mklc                 f    |                      g | j        d d         |j        d         R           S r  r  )rh  packed_weightorig_weightr  r   s        r/   meta_mkl_linearr  Q	  s>    ))@,$SbS)@;+<Q+?@@  r1   onednnc           
          t          | ||||	d|
d           }|t          j        t          j        fv sJ |                     ||          }|                    t          j                  }|S )NFr}   r   )rw  rE   r2  r4  r   r   r|  )r<   x_scalex_zpww_scalew_zpr  r   r  ri  rk  output_scaleoutput_zero_pointoutput_dtyper  r  r  r  r   s                      r/   meta_qconv2d_pointwiser  [	  sw    ( .	
 	
	 u~>>>>>kk)<k88ff5#6f77
r1   c                     t          | j                  }|j        d         |d<   |	t          j        t          j        fv sJ |                     ||	          }|S )Nr   r   r}   )r   r   rE   r2  r4  r   )r<   r  r  r  r  r  r  r  r  r  post_op_namepost_op_argspost_op_algorithmr;  r   s                  r/   meta_qlinear_pointwiser  ~	  sU    " AG}}71:Ru~>>>>>kk,lk;;
r1   c                 x    t          | j                  }|j        d         |d<   |                     |          }|S )Nr   r   )r   r   r   )r<   r  r  r;  r   s        r/   meta_linear_dynamic_fp16r  	  s7     AG}}71:Rkk,''
r1   	quantizedr5   r   r   c                 4   t          | |||||          \  }}}|                                 dk    r|                     d          nd}	t          j        }
|                                 dk    r|||g}n|	|||g}t          j        || j        | j        |
          S Nr  r   r   rR  )#max_pool2d_checks_and_compute_shapers   r   rE   r|  rx   rM   rp   r   rs  r   r  ri  	ceil_modenInputPlaneoutputHeightoutputWidthr  r   r   s               r/   meta_quantized_max_pool2dr  	  s     0;9
 
		
 $)99;;!#3#3B+99;;!{;DDK{CD{+<'	
 
 
 	
r1   c                      t          j                                         k    o j                 k     fd           d S )Nc                  n    d  d d dd                                  d dj                  z   S )NzExpected a tensor of dimension z and tensor.size[z] == rf   zbut got : dimension z] = rs   r   )rs   dim_sizer   r	  s   r/   rT   z check_dim_size.<locals>.<lambda>	  sU    _#____W[___
f
f
f
f
ffl[cNd
f
fg r1   )rE   rV   rs   r   )r	  rs   r  r   s   ````r/   check_dim_sizer  	  so    	L

>X 6$ >	g 	g 	g 	g 	g 	g 	g    r1   c                 z   d } |d|          \  }}	t          j        t          |          dv d            t          |          dk    r||	}}
n3t          |          dk    r|d         |d         }}
n |d|          \  }
} |d|          \  }}t          j        |d u p|dk    d	            |                                 d
k    r|                     d          nd}|                     d          }|                     d          }|                     d          }t          ||||
d|          }t          ||	||d|          }t          j        |           }t          | ||	|
|||dd||||||           |                                 dk    r|||g}n||||g}t          j	        || j
        | j        |          S )Nc                      t          j        t          |          dv  fd           |d         }t          |          dk    r|n|d         }||fS )Nr   r   c                      d  dS )Nzavg_pool2d: 4 must either be a single int, or a tuple of two intsr5   r  s   r/   rT   z1meta_avg_pool2d.<locals>.unpack.<locals>.<lambda>	      ]4]]] r1   r   r   rE   rV   r   r  rs  HWs   `   r/   unpackzmeta_avg_pool2d.<locals>.unpack	  `    HH]]]]	
 	
 	
 FSQAACF!tr1   rs  r   r   r   c                      dS NzOavg_pool2d: stride must either be omitted, a single int, or a tuple of two intsr5   r5   r1   r/   rT   z!meta_avg_pool2d.<locals>.<lambda>	      a r1   r   r   r   r  c                      dS Nzdivisor must be not zeror5   r5   r1   r/   rT   z!meta_avg_pool2d.<locals>.<lambda>	      * r1   r  r  r  r   r   rR  )rE   rV   r   rs   r   pooling_output_shaper?   r{  pool2d_shape_checkrx   rM   rp   )r   rs  r   r  r  count_include_paddivisor_overrider  kHkWdHdWpadHpadWr  r  inputHeight
inputWidthr  r  r   r   s                         r/   meta_avg_pool2dr  	  s      VM;//FB	LFy aa   6{{aRB	V		F1IB&))B	7++JD$	LD 9$4$9**  
  %yy{{a//UZZ^^^QF**R..K**R..KBJ'Rr1iPPL&z2tRINNK/66M



		  $ yy{{a\;7\;?;k|#	   r1   c                     t          | ||||||dd|	|
||||           |                                 }|	}t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           d S )Nr   r   r   )r  rs   r  )r   
gradOutputr  r  r  r  r  r  r  r  r  r  r  r  
mem_formatr   nOutputPlanes                    r/   avg_pool2d_backward_shape_checkr  
  s    " 



		  $ 99;;DL:tTAX|<<<:tTAX|<<<:tTAX{;;;;;r1   c                 j   t          j        t          |          dk    pt          |          dk    d            |d         }t          |          dk    r|n|d         }	t          j        t          |          dk    p%t          |          dk    pt          |          dk    d            t          |          dk    r|n|d         }
t          |          dk    r|	nt          |          dk    r|
n|d         }t          j        t          |          dk    pt          |          dk    d            |d         }t          |          dk    r|n|d         }t          j        |d u p|dk    d            |j        }|                                dk    r|d	         nd}|d
         }|d         }|d         }t          ||||
d|          }t          ||	||d|          }t          j        |          }t          || |||	|
|||||||||           t          j	        ||j
        |j        |          S )Nr   r   c                      dS )NzKavg_pool2d: kernel_size must either be a single int, or a tuple of two intsr5   r5   r1   r/   rT   z*meta_avg_pool2d_backward.<locals>.<lambda>V
  s    ] r1   r   c                      dS r  r5   r5   r1   r/   rT   z*meta_avg_pool2d_backward.<locals>.<lambda>\
  r  r1   c                      dS )NzGavg_pool2d: padding must either be a single int, or a tuple of two intsr5   r5   r1   r/   rT   z*meta_avg_pool2d_backward.<locals>.<lambda>b
  s    Y r1   c                      dS r  r5   r5   r1   r/   rT   z*meta_avg_pool2d_backward.<locals>.<lambda>i
  r  r1   r  r  r  r  r   rR  )rE   rV   r   r   rs   r  r?   r{  r  rx   rM   rp   )gradOutput_r   rs  r   r  r  r  r  r  r  r  r  r  r  
input_sizer  r  r  r  r  r  r  s                         r/   meta_avg_pool2d_backwardr  H
  sz    
LKA6[!1!1Q!6]]   
QB;1$$+a.B	LFq@CKK1,@Fq0@aa   6{{aVAYB6{{a3v;;!+;+;RRB	LG.S\\Q.YY   1:Dw<<1$$44'!*D	LD 9$4$9**  
 J$yy{{a//Z^^QFR.KR.KBJ'Rr1iPPL&z2tRINNK,U33J#



  $ ;k| 	   r1   c                    t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }	t          j        | pt          |          dv d            |s|n|d         }
|s|nt          |          dk    r|
n|d         }|s|	nt          |          dk    r|
n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        | j        dv d	            t          j        | p|dk    d
            |                     d          }|                     d          }|                     d          }|                     d          }|                     d          }t          ||||
d|          }t          ||||d|          }t          ||	||d|          }t          | ||||	|
|||||ddd||||||dd           | j        dk    r|                     ||||f          S |                     |||||f          S )Nr   r   c                      dS NzFavg_pool3d: kernel_size must be a single int, or a tuple of three intsr5   r5   r1   r/   rT   z!meta_avg_pool3d.<locals>.<lambda>
      X r1   r   r   r   c                      dS NzJavg_pool3d: stride must be omitted, a single int, or a tuple of three intsr5   r5   r1   r/   rT   z!meta_avg_pool3d.<locals>.<lambda>
  rC  r1   c                      dS NzBavg_pool3d: padding must be a single int, or a tuple of three intsr5   r5   r1   r/   rT   z!meta_avg_pool3d.<locals>.<lambda>
      T r1   r  r  c                      dS Nz9non-empty 4D or 5D (batch mode) tensor expected for inputr5   r5   r1   r/   rT   z!meta_avg_pool3d.<locals>.<lambda>
      K r1   c                      dS r  r5   r5   r1   r/   rT   z!meta_avg_pool3d.<locals>.<lambda>
  r  r1   r  r  r  r   zavg_pool3d()T)check_input_sizer  )rE   rV   r   r   r   r  pool3d_shape_checkr   )r   rs  r   r  r  r  r  kTr  r  dTr  r  padTr  r  r  nslicesitimeiheightiwidthotimeoheightowidths                           r/   meta_avg_pool3dr  
  s    
LKF"XX   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+\\   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LGTT   1:Dw<<1$$44'!*Dw<<1$$44'!*D	L
fKK  
 
L5 0A 5**  
 ZZ]]FjjnnGJJrNNEjjnnGZZ^^F D"aCCE"7Bb!YGGG!&"dB9EEF





			-   2 zQ@AAAHIIIr1   c                    t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }	t          |          dk    r|n|d         }
t          j        | pt          |          dv d            |s|n|d         }|s|	nt          |          dk    r|n|d         }|s|
nt          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        |j        dv d	            t          j        | p|dk    d
            |                    d          }|                    d          }|                    d          }|                    d          }t          ||||d|          }t          ||	||d|          }t          ||
||d|          }t          || |||	|
||||||||||||d           |                    |j                  S )Nr  c                      dS r  r5   r5   r1   r/   rT   z*meta_avg_pool3d_backward.<locals>.<lambda>
  r  r1   r   r   r   c                      dS r  r5   r5   r1   r/   rT   z*meta_avg_pool3d_backward.<locals>.<lambda>
  rC  r1   c                      dS r   r5   r5   r1   r/   rT   z*meta_avg_pool3d_backward.<locals>.<lambda>  r  r1   r  c                      dS r  r5   r5   r1   r/   rT   z*meta_avg_pool3d_backward.<locals>.<lambda>  r  r1   c                      dS r  r5   r5   r1   r/   rT   z*meta_avg_pool3d_backward.<locals>.<lambda>  r  r1   r  r  r  r   zavg_pool3d_backward())	rE   rV   r   r   r   r  avg_pool3d_backward_shape_checkr   r   )r  r   rs  r   r  r  r  r  r	  r  r  r
  r  r  r  r  r  r  r  r  r  otime_for_shape_checkoheight_for_shape_checkowidth_for_shape_checks                           r/   meta_avg_pool3d_backwardr  
  s    
LKF"XX   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+\\   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LGTT   1:Dw<<1$$44'!*Dw<<1$$44'!*D	L
fKK  
 
L5 0A 5**  
 jjnnGJJrNNEjjnnGZZ^^F0D"aSS27Bb!YWW1&"dB9UU#





'  , ??5;'''r1   c                     t          j         j        dk    p
 j        dk     fd            j        d d         t	          |          z   }t          j                   }t          j        | j         j	        |          S )Nr   r  c                      d j          S )Nz"Expected 3D or 4D tensor, but got r  r   s   r/   rT   z*meta_adaptive_avg_pool2d.<locals>.<lambda>;      ATZAA r1   r  rR  )
rE   rV   r   r   rU   r?   r{  rx   rM   rp   )r   output_sizer;  r   s   `   r/   meta_adaptive_avg_pool2dr#  7  s    	L	Q($)q.AAAA   :crc?U;%7%77L/55M ;j{#	   r1   c                      t          j         j        dk    p
 j        dk     fd                                 j        d d         t          |          z             S )Nr  r  c                      d j          S )Nz"Expected 4D or 5D tensor, but got r  r   s   r/   rT   z*meta_adaptive_avg_pool3d.<locals>.<lambda>M  r!  r1   r  )rE   rV   r   r   r   rU   )r   r"  s   ` r/   meta_adaptive_avg_pool3dr&  I  sb    	L	Q($)q.AAAA   >>$*SbS/E+,>,>>???r1   c                      j         }t          d|          D ]2t          j                                       dk     fd           3t          j        |dk    p|dk    fd           t          j        j         j        k     fd           t          j        }t                    rt          j        }	                    j
                                      |          S )	Nr   r   c                       d j          d dS )Nz{adaptive_avg_pool2d_backward(): Expected grad_output to have non-zero                       size for non-batch dimensions,  with dimension  being emptyr  )grad_outr   s   r/   rT   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>X  s1     f6>nf fVWf f f r1   r   r  c                      d j          S )NzBadaptive_avg_pool2d_backward(): Expected 3D or 4D tensor, but got r  r   s   r/   rT   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>]  s    aUYU_aa r1   c                  (    dj          d j          S Nexpected dtype z! for `grad_output` but got dtype r}   )r+  r   s   r/   rT   z4meta__adaptive_avg_pool2d_backward.<locals>.<lambda>a  s    _$*__x~__ r1   r   )r   r   rE   rV   r   rM   r   r  r|  r   r   r   )r+  r   r   r   r   s   ``  @r/   "meta__adaptive_avg_pool2d_backwardr0  R  s   =D1d^^ 
 
MM!q f f f f f	
 	
 	
 	

 
L	TQYaaaa   
L
hn$_____   +M ,+>>$*%%((}(EEEr1   c                 b    t          | d           t          j        |t          j                  S )Nadaptive_avg_pool3d_backwardr   )!_adaptive_pool_empty_output_checkrE   r   r  r  r   s     r/   "meta__adaptive_avg_pool3d_backwardr5  i  s.     &k3QRRRD0NOOOOr1   r  c                       j         }t          d|          D ]3t          j                                       dk     fd           4d S )Nr   r   c                  $      dj          d dS )Nzc(): Expected grad_output to have non-zero size for non-batch dimensions, but grad_output has sizes r)  r*  r  )r  r  r   s   r/   rT   z3_adaptive_pool_empty_output_check.<locals>.<lambda>u  s>     ` `-8->` `PQ` ` ` r1   )r   r   rE   rV   r   )r  r  r   r   s   `` @r/   r3  r3  p  s{    D1d^^ 
 
Q!#     	
 	
 	
 	

 
r1   c                 8     j         }t          j        |dv  fd           t          d|          D ]2t          j                                       dk     fd           3t          j        t          |          dk    d            d}d}d} j         dk    r                     d          }|dz  }                     |dz
            }|\  }} j         d	k    r?|||f}                     |          }	                     |t          j        
          }
|	|
fS ||||f}t          j	                   }                     |          
                    |          }	                     |t          j        
          
                    |          }
|	|
fS )Nr   r  c                      d j          S )Nz:adaptive_max_pool2d(): Expected 3D or 4D tensor, but got: r  r   s   r/   rT   z*meta_adaptive_max_pool2d.<locals>.<lambda>      ZU[ZZ r1   r   r   c                       dj          d  dS )Nzjadaptive_max_pool2d(): Expected input to have non-zero size for non-batch dimensions, but input has sizes r)  r*  r  r   r   s   r/   rT   z*meta_adaptive_max_pool2d.<locals>.<lambda>  4    T',{T TDET T T r1   r   c                      dS )NzCadaptive_max_pool2d(): internal error: output_size.size() must be 2r5   r5   r1   r/   rT   z*meta_adaptive_max_pool2d.<locals>.<lambda>      U r1   r  r   r}   r   )r   rE   rV   r   r   r   r   r   r?   r{  r   )r   r"  r   dimHsizeBsizeDosizeHosizeWr   r   r   r   r   s   `           @r/   meta_adaptive_max_pool2drG  |  s    :D	LZZZZ   1d^^ 
 
JJqMMA    	
 	
 	
 	
 
LKAUU  
 DEEzQ

1	JJtax  E NFFzQFF+	ooi((//)5;/??G|E662	3E::ooi((++-+HH//)5;/??BB' C 
 
 G|r1   c                 :     j         }t          j        |dv  fd           t           d           t          j        j         j        k     fd           t          j                  }                    j                  	                    |          S )Nr9  c                      d j          S )NzKadaptive_max_pooling2d_backward(): Expected 3D or 4D grad_output, but got: r  r  s   r/   rT   z3meta_adaptive_max_pool2d_backward.<locals>.<lambda>  s    q^i^oqq r1   adaptive_max_pool2d_backwardc                  (    dj          d j          S r.  r}   )r  r   s   r/   rT   z3meta_adaptive_max_pool2d_backward.<locals>.<lambda>  s    c%+ccP[Pacc r1   r   )
r   rE   rV   r3  rM   r?   r{  r   r   r   )r  r   r   r   r   s   ``   r/   !meta_adaptive_max_pool2d_backwardrM    s     D	Lqqqq  
 &k3QRRR	L{((ccccc  
 /66M??5;''***GGGr1   c                 :     j         }t          j        |dv  fd           t          d|          D ]2t          j                                       dk     fd           3t          j        t          |          dk    d            d}d}d}|dk    r                     d          }|dz  }                     |          }|\  }}}|d	k    r||||f}	n|||||f}	                     |	          }
                     |	t          j        
          }|
|fS )Nr  c                      d j          S )Nz:adaptive_max_pool3d(): Expected 4D or 5D tensor, but got: r  r;  s   r/   rT   z*meta_adaptive_max_pool3d.<locals>.<lambda>  r<  r1   r   r   c                       dj          d  dS )Nzjadaptive_max_pool3d(): Expected input to have non-zero size for non-batch dimensions, but input has sizes r)  r*  r  r>  s   r/   rT   z*meta_adaptive_max_pool3d.<locals>.<lambda>  r?  r1   r   c                      dS )NzCadaptive_max_pool3d(): internal error: output_size.size() must be 3r5   r5   r1   r/   rT   z*meta_adaptive_max_pool3d.<locals>.<lambda>  rA  r1   r  r  r}   )r   rE   rV   r   r   r   r   r   )r   r"  r   dimDrC  rD  osizeTrE  rF  r   r   r   r   s   `           @r/   meta_adaptive_max_pool3drT    sg    :D	LZZZZ   1d^^ 
 
JJqMMA    	
 	
 	
 	
 
LKAUU  
 DEEqyy

1	JJtE(FFFqyyFFF3		E666:	
//)
$
$Cooiu{o;;G<r1   c                 V    t          | d           |                    |j                  S )Nadaptive_max_pool3d_backward)r3  r   r   )r  r   r   s      r/   !meta_adaptive_max_pool3d_backwardrW    s)     &k3QRRR??5;'''r1   c                 N    |t          d          |                     |          S )Nz:cannot repeat_interleave a meta tensor without output_size)r   r   )repeatsr"  s     r/   meta_repeat_interleave_TensorrZ    s+    WXXX[)))r1   c                     | j         j        sJ |j         j        sJ t          | j        |j                  }|                     |t          | j                             S r]  )rM   r   r#   r   r   r   )realimagr   s      r/   meta_complexr^    sZ     :'''':''''!$*dj99I>>)+Ftz+R+R>SSSr1   )
fill_valuer   r_  c                l    |                      ||                                 ft          j                  S r]  )r   rs   rE   r   )r   r   r_  s      r/   nonzero_staticra    s)     >>4,EJ>???r1   c           
          t          j        t                    d            g }t                    D ]d\  Ft          j        j        t           j        t           j        t           j        t           j        fv d            j        t           j        t           j        fv rǉ                                }t          |          t          j
        j        z    j        k     fd           t          j                  D ]dt          j
        j                  j        z            k     fd           |                    |                    d                     e8|                               O|                               f|t          j        t                     j        k     fd           dd lm} t%           |j                   t                     j        k     r-                    d            t                     j        k     -d}d}D ]|dk    rd}|dk    rd	} nd
}|sg }g }t                    D ]1\  *|                               |                               2t                    D ]1\  *|                               |                               2                     |           |g }	g }
g }t                    D ]_\  }D|r!|
                     j        |                    *|	                     j        |                    Kt%          j                  }`                     |	|z   |
z             S )Nc                      dS )Nz#at least one index must be providedr5   r5   r1   r/   rT   z#meta_index_Tensor.<locals>.<lambda>  s    (M r1   c                      dS )Nz?tensors used as indices must be long, int, byte or bool tensorsr5   r5   r1   r/   rT   z#meta_index_Tensor.<locals>.<lambda>  s    Y r1   c                      d j          S )N)too many indices for tensor of dimension r  r   s   r/   rT   z#meta_index_Tensor.<locals>.<lambda>  s    S	SS r1   c            	      :    dj          d  dj          dz    S )NzThe shape of the mask z
 at index z0 does not match the shape of the indexed tensor r  )r   r   jr9  r   s   r/   rT   z#meta_index_Tensor.<locals>.<lambda>  sP     !h !h !hPQ !h !hJN*!h !h`ade`e!h !h r1   r   c                  :    dj          dt                      dS )Nrf  z (got rg   )r   r   )r   r   s   r/   rT   z#meta_index_Tensor.<locals>.<lambda>*  s$    \DI\\SQX\\\\\ r1   r   Fr   T)rE   rV   rD  	enumeraterM   r   r  r  nonzeror   r   r   r   r   r   selecttorch._refs_refsr   r$   r   r   )r   r   r  rk  refsstatehas_contiguous_subspacer`  transposed_indicesbefore_shapeafter_shapereplacement_shapers   r   r   rh  r9  s   ``           @@@@r/   meta_index_Tensorrv  	  s   	Lg M MNNN &(Fg&& ! !5L
EIuz5:NNYY   {uz5:666--//KK"
Ndi/SSSS   uz** 8 8A&A$*QU*;;h h h h h h h h  
 MM'..A"6"677778 e$$$$MM%    G	LG	!\\\\\  
 (4('233G
g,,
"
"t g,,
"
" E# ' 'A:: aZZ}  ! #'
 # %!'** 	1 	1HAu A"))%000!'** 	1 	1HAu}A"))%000||D!!$ !LK#%(( 2 2
U=  5""4:c?3333##DJsO4444 $U[ 1 1>>,)::[HIIIr1   c                    d }d }d }|
d         r'|                      |                                          }|
d         r'|                      |                                          }|
d         r|                      |          }|||fS )Nr   r   r   r   r   )grad_output_input_weight_bias_sizes_optr   r  ri  
transposedrl  rk  output_maskbackend_grad_inputbackend_grad_weightbackend_grad_biass                 r/   meta_convolution_backwardr  m  s      1~ C)33FKKMMBB1~ E*44W\\^^DD1~ C(22>BB 35FGGr1   c                                        d          }                     d          }|                     ||f          } t          j                                        dk    d            t          j                                        dk    d            t          j                             d                               d          k    fd           t          j                             d                               d          k    fd           t          j        |                      d          |k    o|                      d          |k    d	            |                     |                                            S )
Nr   r   r   c                      dS r   r5   r5   r1   r/   rT   zmeta_addbmm.<locals>.<lambda>  r!  r1   c                      dS r#  r5   r5   r1   r/   rT   zmeta_addbmm.<locals>.<lambda>  r!  r1   r   c                  `    d                      d           d                     d           S )Nz8batch1 and batch2 must have same number of batches, got r   r   r   r%  r&  s   r/   rT   zmeta_addbmm.<locals>.<lambda>  s3    p6;;WX>>pp`f`k`klm`n`npp r1   c            
          d                      d           d                      d           d                     d           d                     d           d	S )Nz#Incompatible matrix sizes for bmm (r   r<   r   r   rg   r   r  s   r/   rT   zmeta_addbmm.<locals>.<lambda>  sl    6&++a.. 6 66;;q>> 6 6;;q>>6 6$*KKNN6 6 6 r1   c                      dS )Nz.self tensor does not match matmul output shaper5   r5   r1   r/   rT   zmeta_addbmm.<locals>.<lambda>  s    @ r1   )r   r\  rE   rV   rs   r   )r   r%  r&  r%  r$  r.  r/  s    ``    r/   meta_addbmmr    sg    ;;q>>D;;q>>D;;d|$$D	L"$H$HIII	L"$H$HIII	LA&++a..(ppppp   
LA&++a..(	
 	
 	
 	
 	
   
L		!51!5@@   >>$))++&&&r1   )
grad_scale	found_infc       	         t    | |||||fD ]-t          j        t          t                    fd           .d S )Nc                  (    dt                      S Nz'exponent must be a tensor list but got rh   ls   r/   rT   z#meta__fused_adam_.<locals>.<lambda>      Gd1ggGG r1   rE   rV   r_   r   )r   gradsexp_avgsexp_avg_sqsmax_exp_avg_sqsstate_stepslrbeta1beta2weight_decayepsamsgradmaximizer  r  r  s                  @r/   meta__fused_adam_r    s]    & E8[/;O 
 
q$GGGG	
 	
 	
 	

 
r1   c       	             | |||||fD ]-t          j        t          t                    fd           .d } ||            ||           ||           ||           ||          fS )Nc                  (    dt                      S r  r  r  s   r/   rT   z"meta__fused_adam.<locals>.<lambda>  r  r1   c                     d | D             S )Nc                 6    g | ]}t          j        |          S r5   r?  )r;   r  s     r/   r>   z=meta__fused_adam.<locals>.empty_like_list.<locals>.<listcomp>  s#    999 ##999r1   r5   )tensor_lists    r/   empty_like_listz)meta__fused_adam.<locals>.empty_like_list  s    99[9999r1   r  )r   r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  s                   @r/   meta__fused_adamr    s    & E8[/;O 
 
q$GGGG	
 	
 	
 	

: : : 	!!$$(( r1   c                 r    t          j                                         dk    d            t          j                                        dk    d            t          j         j        t           j        u  fd           t          j        j        t           j        u fd           t          j                             d                              d          k     fd                                                     d                              d          ft           j        	          S )
Nr   c                      dS )Nza must be a 2D tensorr5   r5   r1   r/   rT   zmeta__int_mm.<locals>.<lambda>      '> r1   c                      dS )Nzb must be a 2D tensorr5   r5   r1   r/   rT   zmeta__int_mm.<locals>.<lambda>  r  r1   c                      d j          S )Nzexpected self to be int8, got r}   )r[  s   r/   rT   zmeta__int_mm.<locals>.<lambda>      ::: r1   c                      d j          S )Nzexpected mat2 to be int8, got r}   )r\  s   r/   rT   zmeta__int_mm.<locals>.<lambda>  r  r1   r   r   c            
          d                      d           d                      d           d                     d           d                     d           d	S )Nz'Incompatible matrix sizes for _int_mm (r   r<   r   r   rg   r   r[  r\  s   r/   rT   zmeta__int_mm.<locals>.<lambda>  sl    ,affQii , ,!&&)) , ,66!99, , vvayy, , , r1   r}   )rE   rV   rs   rM   r  r   r   r  r  s   ``r/   meta__int_mmr    s"    
LA>>???	LA>>???	L	5:::::   
L	5:::::   
L	q		QVVAYY	
 	
 	
 	
 	
   ;;q		166!99-U[;AAAr1   c                 j    t          j                                         dk    d            t          j         j        t           j        u  fd                                d          }                     d          dz  }                     |dz  ||dz  z  d|dz  ft           j        	          S )
Nr   c                      dS Nzw must be a 2D tensorr5   r5   r1   r/   rT   z2meta__convert_weight_to_int4pack.<locals>.<lambda>  r  r1   c                      d j          S Nzexpected w to be uint8, got r}   r  s   r/   rT   z2meta__convert_weight_to_int4pack.<locals>.<lambda>       8qw88 r1   r   r      r1      r}   )rE   rV   rs   rM   uint8r   r   r  r  inner_k_tilesr   r9  s   `   r/    meta__convert_weight_to_int4packr    s    	LA>>???	L	5;8888   	
q		A	q		AA;;F-"$%Q		
 k    r1   c                 N    t          j                                         dk    d            t          j         j        t           j        u  fd                                d          }                     d          }                     ||dz  ft           j                  S )Nr   c                      dS r  r5   r5   r1   r/   rT   z:meta__convert_weight_to_int4pack_for_cpu.<locals>.<lambda>  r  r1   c                      d j          S Nzexpected w to be int32, got r}   r  s   r/   rT   z:meta__convert_weight_to_int4pack_for_cpu.<locals>.<lambda>  r  r1   r   r   r}   )rE   rV   rs   rM   r  r   r   r  r  s   `   r/   (meta__convert_weight_to_int4pack_for_cpur    s    	LA>>???	L	5;8888   	
q		A	q		A;;	
AFk    r1   c                     t          j                                         dk    d            t          j                                        dk    d            t          j         j        t           j        t           j        t           j        fv  fd           t          j        j        t           j        u fd                                 	                    d          	                    d          dz   j        	          S )
Nr   c                      dS Nzx must be a 2D tensorr5   r5   r1   r/   rT   z*meta__weight_int4pack_mm.<locals>.<lambda>   r  r1   r  c                      dS )Nzw must be a 4D tensorr5   r5   r1   r/   rT   z*meta__weight_int4pack_mm.<locals>.<lambda>!  r  r1   c                      d j          S Nz#expected x to be f32/f16/bf16, got r}   r<   s   r/   rT   z*meta__weight_int4pack_mm.<locals>.<lambda>$      ?ag?? r1   c                      d j          S r  r}   r  s   r/   rT   z*meta__weight_int4pack_mm.<locals>.<lambda>(  r  r1   r   r  r}   )
rE   rV   rs   rM   r2  r3  r4  r  r   r   r<   r  q_group_sizeq_scale_and_zeross   ``  r/   meta__weight_int4pack_mmr    s    	LA>>???	LA>>???	L	EM5=%.AA????   
L	5;8888   ;;qvvayy!&&))a-qw;???r1   c                     t          j                                         dk    d            t          j                                        dk    d            t          j         j        t           j        t           j        t           j        fv  fd           t          j        j        t           j        u fd                                 	                    d          	                    d           j                  S )Nr   c                      dS r  r5   r5   r1   r/   rT   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>/  r  r1   c                      dS r  r5   r5   r1   r/   rT   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>0  r  r1   c                      d j          S r  r}   r  s   r/   rT   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>3  r  r1   c                      d j          S r  r}   r  s   r/   rT   z2meta__weight_int4pack_mm_for_cpu.<locals>.<lambda>7  r  r1   r   r}   )
rE   rV   rs   rM   r2  r3  r4  r  r   r   r  s   ``  r/    meta__weight_int4pack_mm_for_cpur  -  s    	LA>>???	LA>>???	L	EM5=%.AA????   
L	5;8888   ;;qvvayy!&&))17;;;;r1   c                     t          j                                         dk    d            t          j         j        t           j        t           j        t           j        fv  fd           t          j                                        dk    d            t          j        j        t           j        u fd                                 	                    d          	                    d           j                  S )Nr   c                      dS r  r5   r5   r1   r/   rT   z*meta__weight_int8pack_mm.<locals>.<lambda>>  r  r1   c                      d j          S r  r}   r  s   r/   rT   z*meta__weight_int8pack_mm.<locals>.<lambda>A  r  r1   c                      dS r  r5   r5   r1   r/   rT   z*meta__weight_int8pack_mm.<locals>.<lambda>C  r  r1   c                      d j          S )Nzexpected w to be int8, got r}   r  s   r/   rT   z*meta__weight_int8pack_mm.<locals>.<lambda>F  s    7ag77 r1   r   r}   )
rE   rV   rs   rM   r2  r3  r4  r  r   r   )r<   r  q_scaless   `` r/   meta__weight_int8pack_mmr  <  s    	LA>>???	L	EM5=%.AA????   
LA>>???	L	5:7777   ;;qvvayy!&&))17;;;;r1   c                     t          j                                         dk     fd           t          j                                        dk    fd           t          j                             d                              d          k     fd           t          j        t	          j         j                  d            t          j        t	          j        j                  d            t          j        |dk    d	            t          j        d
v fd                                d          }                    d          } j        d d         }j        d d         }t          t          j	        ||                    }|
                    ||g                                |          S )Nr   c                  4    d                                   dS )Nz1cdist only supports at least 2D tensors, X1 got: Dr   )x1s   r/   rT   z$meta_cdist_forward.<locals>.<lambda>O      OBFFHHOOO r1   c                  4    d                                   dS )Nz1cdist only supports at least 2D tensors, X2 got: r  r   )x2s   r/   rT   z$meta_cdist_forward.<locals>.<lambda>S  r  r1   r   c                  `    d                      d           d                     d           S )Nz4X1 and X2 must have the same number of columns. X1: r   z X2: r   )r  r  s   r/   rT   z$meta_cdist_forward.<locals>.<lambda>W  s2    frwwr{{ffY[Y`Y`acYdYdff r1   c                      dS )Nz=cdist only supports floating-point dtypes, X1 got: {x1.dtype}r5   r5   r1   r/   rT   z$meta_cdist_forward.<locals>.<lambda>[  r   r1   c                      dS )Nz=cdist only supports floating-point dtypes, X2 got: {x2.dtype}r5   r5   r1   r/   rT   z$meta_cdist_forward.<locals>.<lambda>_  r   r1   r   c                      dS )Nz)cdist only supports non-negative p valuesr5   r5   r1   r/   rT   z$meta_cdist_forward.<locals>.<lambda>a  s    !L r1   Nr   r   c                      d  S )Nz%possible modes: None, 1, 2, but was: r5   )compute_modes   r/   rT   z$meta_cdist_forward.<locals>.<lambda>d  s    FFF r1   r  )rE   rV   rs   r   r?   is_float_dtyperM   r   r   broadcast_shapesextendr   )	r  r  r  r  r1r2batch_tensor1batch_tensor2r;  s	   `` `     r/   meta_cdist_forwardr  K  s   	L
AOOOO   
L
AOOOO   
L
rwwr{{"fffff   
LRX&&OO   
LRX&&OO   
LaLLMMM	L$FFFF   
B	BHSbSMMHSbSMM.}mLLMMLR!!!<<%%%r1   c                 @   |j         d         }|j         d         }|j         d         }|j         d d         }|j         d d         }	t          t          j        ||	                    }
|
                                }|                    ||g           t          j        |
          }|dk    s|dk    s|dk    s|dk    rt          j        |          S |t          |j                   k    r|	                    |          }t          j
        |t          j                  S )Nr   r  r   r   )r   r   rE   r  copyr  mathprod
zeros_liker\  r   r   )r  r  r  r  cdistc1r  r  r  r  rX  tensor1_expand_sizebatch_products                r/   meta_cdist_backwardr  o  s
    
"B	"B	"BHSbSMMHSbSMM 6}m T TUU.3355Bx(((I233M	Qww"''R1WW(:(:###d28nn,,YY*++Be.EFFFFr1   c	                 d    t          j        j        t           j        t           j        fv fd           t          j        j        t           j        t           j        fv fd           t          j        t          j         j                   fd                               d          }	|rt          j        |	dk    d            |	dz  }	                     |	                     d                    }
t          d          \  }}}|t          j        ||k    d            t          j        j
        dk    fd	           t          j                                                                        k    fd
           fdd fd}t                    dk    r                                        d                    }                                                              }||k    r*                    |	                     d                    }n                    d          }n | |
|          }|||fv s|s)                                        d                    }n                    d          }                    |	          }j        d         }||k    rC|rt          j        |dk    d            |dz  }                    | j        d                   }n'                    |                                          }|
|||fS )Nc                      d j          S )Nz(expected indices to be long or int, got r}   )r   s   r/   rT   z$meta_embedding_bag.<locals>.<lambda>      J7=JJ r1   c                      d j          S )Nz(expected offsets to be long or int, got r}   )rK  s   r/   rT   z$meta_embedding_bag.<locals>.<lambda>  r  r1   c                      d j          S )Nz/expected weight to be floating point type, got r}   )r  s   r/   rT   z$meta_embedding_bag.<locals>.<lambda>  s    P&,PP r1   r   r   c                      dS Nz1include_last_offset: numBags should be at least 1r5   r5   r1   r/   rT   z$meta_embedding_bag.<locals>.<lambda>  s    G r1   r   c                      dS )Nz@embedding_bag: per_sample_weights only supported with mode='sum'r5   r5   r1   r/   rT   z$meta_embedding_bag.<locals>.<lambda>  s    V r1   c                      d j          dS )Nz1expected per_sample_weights to be 1D tensor, got r  r  )per_sample_weightss   r/   rT   z$meta_embedding_bag.<locals>.<lambda>  s    bHZH_bbb r1   c                  ^    d                                  d                                   dS )Nz%expected per_sample_weights.numel() (z$ to be the same as indices.numel() (rg   r   )r   r  s   r/   rT   z$meta_embedding_bag.<locals>.<lambda>  sF    I8J8P8P8R8R I I6=mmooI I I r1   c                 P     | ||          o|                     d          dk    S Nr   r   r   )r  rI  r   padding_idxis_fast_path_index_selects       r/   is_fast_path_index_select_scalez;meta_embedding_bag.<locals>.is_fast_path_index_select_scale  s0    %%c6;??XELLQROOWXDX	
r1   c                     | j         t          j        k    s| j         t          j        k    o7|                     d          dk    o|                    d          dk    o|dk     S Nr   r   )rM   rE   rI   rG   r   )r  r   r  s      r/   r  z5meta_embedding_bag.<locals>.is_fast_path_index_select  s_    Y%+%@ej)@  

1" a  A%  a		
r1   c                 >    | | |||          S  | ||          S r,   r5   )r  rI  r   r  r  r  s       r/   is_fast_pathz(meta_embedding_bag.<locals>.is_fast_path  s5    223v{SSS,,S&+FFFr1   cpuc                      dS r  r5   r5   r1   r/   rT   z$meta_embedding_bag.<locals>.<lambda>  s    O r1   )rE   rV   rM   r   r  r?   r  r   r   r   r   r   rL  r   )r  r   rK  scale_grad_by_freqr3  sparser  include_last_offsetr  num_bagsr   MODE_SUM	MODE_MEANMODE_MAXr  
offset2bagbag_sizemax_indicesfast_path_sumnumBagsr  r  s   ```   `             @@r/   meta_embedding_bagr#    s    
L%*ei00JJJJ   
L%*ei00JJJJ   
LV\**PPPP  
 ||AH MGG	
 	
 	
 	AhA77F$)!HH!Hi%HVV	
 	
 	
 	#q(bbbb	
 	
 	
 	$$&&'--//9    	
 	
 	

 
 
 
 


 
 
G G G G G G 7u$$&&w||A77
$$W\\^^448!++Hfkk!nnEEKK!++A..KK$V-?UUIx(((( **7<<??;;JJ **1--J$$X..-"8" qLOO   1!++GV\!_EEKK!++HMMOO<<K:x44r1   c                     t          | ||g|R  \  }}}}t          |          dk    r'|                    |                                          }||||fS )Nr  )r#  rL  r   r   )r  r   rK  rA   r   r  r  r   s           r/   meta_embedding_bag_forward_onlyr%    sm    0B1#'1 1 1-FJ+ 7u$$$$W\\^^44:x44r1   c                 r    |r|S | j         j        s| j         j        r| j         S |rt          j        S | j         S r,   )rM   r   r   rE   r   )r   rM   promote_int_to_longs      r/   _get_reduction_dtyper(    sH     {$ (> {	 z;r1   r}   c                    t          | |d          }t          j        | j        |          }t	          | ||          }|                     ||          S )NT)r'  r}   )r(  r?   r_  r   r`  r   )r   r`  rb  rM   r  r;  s         r/   meta_nansumr*    sS     (u$OOOLT22D+E4AAL??<|?<<<r1   c           	          t          j        | j        t          t	          |                                                               }|                     |          S r,   )r?   ra  r   rU   r   rs   r   )r   r;  s     r/   meta_medianr,    sG    7U5--.. L ??<(((r1   c                    t          |           dk    rt          j        d           t          j        | j        |f          }t          | ||          }|                     |          |                     |t          j                  fS )NrK  zmedian CUDA with indices outputr}   )	rL  r?   alert_not_deterministicr_  r   r`  r   rE   r   )r   rs   rb  r;  s       r/   meta_median_mode_dimr/    s~     5V##%&GHHH

u{SF
3
3C+E3@@L%%EJ77 r1   c                     | S r,   r5   r   s    r/   meta_logical_not_r1  )  r;  r1   c                 p   t          j        t                    |                                 k    d            t                    |                                 z
  }d|z  t	          | j                  z   fdt          t                              D             }|                     |          S )Nc                      dS )NzZNumber of dimensions of repeat dims can not be smaller than number of dimensions of tensorr5   r5   r1   r/   rT   zmeta_repeat.<locals>.<lambda>2  s    l r1   r  c                 2    g | ]}|         |         z  S r5   r5   )r;   r   padded_sizerY  s     r/   r>   zmeta_repeat.<locals>.<listcomp>9  s&    LLL1;q>GAJ.LLLr1   )rE   rV   r   rs   rU   r   r   r   )r   rY  num_new_dimensionstarget_sizer5  s    `  @r/   meta_repeatr8  .  s    	LG

"ll   W

2++eDJ.?.??KLLLLLc'll8K8KLLLK>>+&&&r1   c                     | S r,   r5   r   s    r/   
meta_zero_r:  =  r;  r1   c                 n    t          |t          j                  rt          | j        |j                   | S r,   )r_   rE   r   rX   r   r   r   s     r/   meta_binop_inplacer=  B  s1     %&& 9
EK888Kr1   c                    d }d }d } ||           r ||          rt          d           ||           r ||          st          d          t          |t          j                  rt	          | j        |j                   | S )a*  
    Some checks for inplace ops.
    Checks for promotion rules for some dtypes.
    int.add/sub_(float) and bool.add/sub_(others) are rejected.
    Promoting in these in-place operations would require reallocating
    and copying over elements, hence not allowed.
    Checks for alpha param.
    c                     t          | t                    rt          j        | j                  S t          | t
                    S r,   )r_   r   r?   r  rM   r   ra   s    r/   is_integericz.meta_binop_inplace_alpha.<locals>.is_integerice  s7    c:&& 	,)#)444c7+++r1   c                     t          | t                    rt          j        | j                  S t          | t
                    S r,   )r_   r   r?   r  rM   r   r@  s    r/   
is_floaticz,meta_binop_inplace_alpha.<locals>.is_floatick  s7    c:&& 	.'	222c9---r1   c                     t          | t                    rt          j        | j                  S t          | t
                    S r,   )r_   r   r?   is_boolean_dtyperM   r   r@  s    r/   is_booleanicz.meta_binop_inplace_alpha.<locals>.is_booleanicq  s7    c:&& 	-)#)444c8,,,r1   z]Promotion of int.add/sub_(float) in in-place ops are not possible due to element size change.z_Promotion of book.add/sub_(others) in in-place ops are not possible due to element size change.)r   r_   rE   r   rX   r   )r   r   r$  rA  rC  rF  s         r/   meta_binop_inplace_alpharG  S  s    $, , ,. . .- - - |D 
jj// 
k
 
 	

 |D 
,,u"5"5 
m
 
 	
 %&& 9
EK888Kr1   c                 8    t          | t          j                  S Nr7   rC   r   r@   )r   kwargss     r/   
meta_roundrM    s"    <D   r1   c                 \    t          j        t          j        j                   fd           t          t           j                  r2t          j        t          j        j                   fd           d S t          j        t          t                     fd           d S )Nc                        dj          S )Nz7: Expected input tensor to have an integral dtype. Got r}   )r  r   s   r/   rT   z#shift_dtype_check.<locals>.<lambda>  s    7__SWS]__ r1   c                        dj          S )Nz6: Expected shift value to have an integral dtype. Got r}   r  rs  s   r/   rT   z#shift_dtype_check.<locals>.<lambda>  s    waaVYV_aa r1   c                        d S )Nz): Expected shift value to be an int. Got r5   rQ  s   r/   rT   z#shift_dtype_check.<locals>.<lambda>  s    wNNNN r1   )rE   rV   r?   r  rM   r_   r   r   )r  r   rs  s   ```r/   shift_dtype_checkrS    s    	Ltz**_____   #u|$$ 	
"39--aaaaa	
 	
 	
 	
 	

 	sG$$NNNNN	
 	
 	
 	
 	
r1   c                 \    t          d| |           t          | |t          j                  S )NrshiftrJ  rS  rC   r   r@   r<  s     r/   meta_rshiftsrW    6    he,,,e$C$K   r1   c                 \    t          d| |           t          | |t          j                  S )NlshiftrJ  rV  r<  s     r/   meta_lshiftsr[    rX  r1   c                 6    |                      | j                  S r,   r  r   s    r/   	meta_zeror]    s    >>$*%%%r1   c                     | S r,   r5   r   rs  s     r/   
meta_fill_r`    r;  r1   c                 *    t          j        |           S r,   r?  r_  s     r/   	meta_fillrb        D!!!r1   c                     | S r,   r5   r   s    r/   
meta_relu_re    r;  r1   c                 :    t          | |t          j                  S rI  rK  )r   r   r$  s      r/   meta__add_relurg    s&     e$C$K   r1         ?UUUUUU?c                 *    t          j        |           S r,   r?  r   noiselowerr  r  r   s         r/   meta_rrelu_with_noisern    s    
 D!!!r1   c                 R    t          j        |           t          j        |          fS r,   r?  rk  s         r/    meta_rrelu_with_noise_functionalrp    s%     D!!5#3E#:#:::r1   c                     | S r,   r5   )r   rm  r  r  r   s        r/   meta_rrelu_with_noise_rr    s	     Kr1   c                 *    t          j        |           S r,   r?  r   r   r   
accumulates       r/   meta_index_putrv    rc  r1   c                 :    t          | j        |j                   | S r,   rX   r   )r   rQ  values      r/   meta_masked_fill_rz    s    DJ
333Kr1   c                     |                      |                                                               t          j        |                     }|S rB  )r   r   r   r?   r{  )r   rQ  rI  masked_scales       r/   meta__masked_scaler}    sF    >>$))++..111$77 2  L r1   c                      t          j        |j        t           j        t           j        fv d            t          j         j        j        k     fd            S )Nc                      dS )NzMask must be bool or uint8r5   r5   r1   r/   rT   z&meta_masked_scatter_.<locals>.<lambda>  s    9U r1   c                  (    d j          dj          S )NzEmasked_scatter: expected self and source to have same dtypes but got r   r}   )r   r>  s   r/   rT   z&meta_masked_scatter_.<locals>.<lambda>  s(     :*: :+1<: : r1   )rE   rV   rM   rD  r  )r   rQ  r>  s   ` `r/   meta_masked_scatter_r    ss    	L
uz5;//1U1U   
L
fl"	: 	: 	: 	: 	:  
 Kr1   c                     t          | |          \  } }t          j        | t          j                  }t	          |||          S rB  )r$   rE   r   r   r  )r   rQ  r>  r   s       r/   meta_masked_scatterr    sA     "$--JD$d%2IJJJFf555r1   c                 ,    |                      |          S r,   rY  )r   rQ  r  s      r/   meta_masked_scatter_backwardr  
  s    >>%   r1   c                     | S r,   r5   rt  s       r/   meta_index_put_r    r;  r1   c                 6    |                      | j                  S r,   )viewr   r   s    r/   
meta_aliasr    s    99TZ   r1   c                   	
 t          j        |                                 dk    d            t          j        |                                dk    d            |                                 }|                                |d         	|d         
|d         }d         }	||ft          j        d         	k    od         
k    	
fd           |                              }|s][t          j                                        dk    d            t          j                                        k    fd	           |S )
Nr   c                      dS r   r5   r5   r1   r/   rT   z)common_meta_baddbmm_bmm.<locals>.<lambda>  r!  r1   c                      dS r#  r5   r5   r1   r/   rT   z)common_meta_baddbmm_bmm.<locals>.<lambda>  r!  r1   r   r   r   c            	      :    d d d d          d d          d	S r(  r5   r*  s   r/   rT   z)common_meta_baddbmm_bmm.<locals>.<lambda>(  sU     RSU R RR R+7?R R>J1oR R R r1   c                      dS )Nzself must be a 3D tensorr5   r5   r1   r/   rT   z)common_meta_baddbmm_bmm.<locals>.<lambda>1  s    6P r1   c                  8    d  d                                  S )Nz*Expected an input tensor shape with shape z but got shape: r   )r"  self_baddbmms   r/   rT   z)common_meta_baddbmm_bmm.<locals>.<lambda>4  s%    sss^j^o^o^q^qss r1   )rE   rV   rs   r   r   )r%  r&  is_bmmr  r1  res_rowsres_colsr   r+  r,  r-  r"  s      `    @@@@r/   common_meta_baddbmm_bmmr    sr   	L"$H$HIII	L"$H$HIII;;==L;;==L	aB#AAHAHx*K	LQ2E,q/5E"E	R 	R 	R 	R 	R 	R   k**F 
l.\%%''1,.P.PQQQ;.sssss	
 	
 	

 Mr1   c                 $    t          | |d          S )NT)r  )r   r   s     r/   meta_bmmr  :  s    "4t444r1   c                     | |z  }| |z  }|dk    r-t          |dk               t          |dk               k    r|dz  }|S r  )rD  )r<   yqr7  s       r/   div_rtnr  ?  sK    	QA	AA 	Avv4A;;$q1u++--	QHr1   c                     t          | |z   |z   ||dz
  z  z
  dz
  |r|dz
  ndz   |          dz   }|r|dz
  |z  | |z   k    r|dz  }|S r  )r  )	inputSize
kernelSizer  r  r   ri  r  
outputSizes           r/   pooling_output_shape_pad_lrr  I  s     	 *q.)* 	
 '-vzzA/ 	
 	
 		   Nf$	E(999!OJr1   c           	          t          j        |dk    d            t          j        dk    fd           t          j        dz
  z  dz   dz  k    fd           t          | ||          S )Nr   c                      dS )Nzstride should not be zeror5   r5   r1   r/   rT   z&pooling_output_shape.<locals>.<lambda>e  s    &A r1   c                      d  S )Nz'pad must be non-negative, but got pad: r5   )pads   r/   rT   z&pooling_output_shape.<locals>.<lambda>f  s    #RS#R#R r1   r   r   c                      d d d  S )NzApad should be at most half of effective kernel size, but got pad=z, kernel_size=z and dilation=r5   )ri  r  r  s   r/   rT   z&pooling_output_shape.<locals>.<lambda>i  s;    @PS @ @%@ @5=@ @ r1   )rE   rV   r  )r  r  r  r   ri  r  s    `` ` r/   r  r  d  s    	L1AABBB	LRRRRSSS	La8+a/A55	
 	
 	
 	
 	
 	
   ':sC9  r1   c           	      L   	
                                   }	t          j        dk    odk    d            t          j        |dk    o|dk    d            t          j        |dk    o|dk    d                                 d          dk    o                     d          dk    }|t          j        k    r6t          j        |dk    o|o                     d          dk    d	            nXt          j        |dk    r                     d          dk    r|p |dk    o|o                     d          dk     fd
           t          j        dz  k    odz  k    fd           t          j        dk    odk    
	fd           d S )Nr   c                      dS )NzCkernel size should be greater than zero, but got kH: {kH}, kW: {kW}r5   r5   r1   r/   rT   z$pool2d_shape_check.<locals>.<lambda>  rA  r1   c                      dS )Nz>stride should be greater than zero, but got dH: {dH}, dW: {dW}r5   r5   r1   r/   rT   z$pool2d_shape_check.<locals>.<lambda>  s    P r1   c                      dS )Nz\dilation should be greater than zero, but got dilationH: {dilationH}, dilationW: {dilationW}r5   r5   r1   r/   rT   z$pool2d_shape_check.<locals>.<lambda>  s    n r1   r   r   r  r   c                      dS )NzExpected 4D (batch mode) tensor expected for input with channels_last layout with optional 0 dim batch size for input, but got: {input.size()}r5   r5   r1   r/   rT   z$pool2d_shape_check.<locals>.<lambda>  s
     Q r1   c                  2    d                                   S )NzYExpected 3D or 4D (batch mode) tensor with optional 0 dim batch size for input, but got: r   r;  s   r/   rT   z$pool2d_shape_check.<locals>.<lambda>  s    ~pupzpzp|p|~~ r1   c                       d d d d  S )NzKpad should be smaller than or equal to half of kernel size, but got padW = z	, padH = z, kW = z, kH = r5   )r  r  r  r  s   r/   rT   z$pool2d_shape_check.<locals>.<lambda>  s>     >> >!%> >.0> >9;> > r1   c                  .    d d  d d d d dS NzGiven input size: (r<   z). Calculated output size: (z). Output size is too smallr5   )r  r  r  r  r  r  s   r/   rT   z$pool2d_shape_check.<locals>.<lambda>  sZ     #k # #K # #* # #$0# #3?# #BM# # # r1   )rs   rE   rV   r   r|  )r   r  r  r  r  r  r  	dilationH	dilationWr  r  r  r  r  r   r   
valid_dimsr  s   ```  ``  `````   @r/   r  r  s  s8   " 99;;DL	L
Q26UU   
L
Q26PP   
LA')a-nn  
 A!#:

1(:J+++AI;*;A!);Q Q	
 	
 	
 	
 	QY<5::a==A-<* A	?j?UZZ]]a-?~~~~	
 	
 	
 
L
a4+B!GtO	> 	> 	> 	> 	> 	> 	>   
Lq.\Q.	# 	# 	# 	# 	# 	# 	# 	# 	#    r1   r  r	  r  r  r
  r  r  pTpHpW	dilationTr  r  r  r  r  r  r  r  r  c           
      8   	
  j         }t          j        dk    odk    odk    fd           t          j        dk    odk    odk    fd           t          j        dk    odk    odk    fd           t          j        |dv  fd           t          |          D ]@|dk    rdk    rt          j                                       dk     fd           A|r-t          j        k    ok    ok    fd	           t          j        d
z  k    od
z  
k    od
z  	k    	
fd           t          j        dk    odk    odk    fd           d S )Nr   c                      d d  d S )Nz5kernel size should be greater than zero, but got kT: z, kH: z, kW: r5   )r  r	  r  s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  4    ,, ,, ,'), , r1   c                      d d  d S )Nz0stride should be greater than zero, but got dT: z, dH: z, dW: r5   )r  r
  r  s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  r  r1   c                      d d  d S )Nz9dilation should be greater than zero, but got dilationT: z, dilationH: z, dilationW: r5   )r  r  r  s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  s;    V#V V2;V VJSV V r1   r  c                        dj          S )Nz/: Expected 4D or 5D tensor for input, but got: r  )r  r   s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  s    7XX5;XX r1   r  c                  J      dj          d                               dS )NzZ: Expected input's non-batch dimensions to have positive length, but input has a shape of z and non-batch dimension z has length zero!)r   r   )r  r   r   s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  sG     M M-2[M M,1JJqMMM M M r1   c                  .    d d  d d d d dS )Nzinput image (T: r  r  z ) smaller than kernel size (kT:  kH:  kW: rg   r5   )r  r  r  r  r	  r  s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  s]    <5 < <g < <6 < <$&< <-/< <68< < < r1   r   c                  ,    d d d  d d d S )NzHpad should be smaller than or equal to half of kernel size, but got kT: r  r  z padT: z padW: z padH: r5   )r  r	  r  r  r  r  s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  sh    KK KK K%'K K02K K;=K KFHK K r1   r   c                  :    d d d  d d d d d dS r  r5   )r  r  r  r  r  r  r  s   r/   rT   z$pool3d_shape_check.<locals>.<lambda>  sv    (' ( (E ( (G ( (f ( ((/( (27( (:A( (DJ( ( ( r1   )r   rE   rV   r   r   )r   r  r	  r  r  r
  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r   r   s   `````````````````````  @r/   r  r    s   0 :D	L
Q$26$b1f	
 	
 	
 	
 	
 	
   
L
Q$26$b1f	
 	
 	
 	
 	
 	
   
LA9)a-9IM	
 	
 	
 	
 	
 	
   
LXXXXX  
 4[[ 
 
199aJJqMMA     	
 	
 	
 	
  
RK:GrM:fl        	
 	
 	
 
L
Q"6a26"q&B,	
 	
 	
 	
 	
 	
 	
 	
 	
   
L
3v{3w!|	
 	
 	
 	
 	
 	
 	
 	
 	
 	
    r1   c                    | j         }t          | |||||||	|
||||||||||||           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           d S )Nr  r   r   r   r   r  r  )r   r  r   r  r	  r  r  r
  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r   s                           r/   max_pool3d_backward_shape_checkr    s$   2 :D








+  0 ;dQh888;dQh666;dQh888;dQh7777D$(G4447D$(E2227D$(G4447D$(F33333r1   c                    | j         }t          | ||||||||	|
|ddd|||||||d           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           t          |||dz
  |           d S )Nr   Tr  r   r   r  )r   r  r  r	  r  r  r
  r  r  r  r  r  r  r  r  r  r  r  r  r   s                       r/   r  r  L  s    * :D








			-  2 ;dQh888;dQh666;dQh888;dQh77777r1   c                 P   d } |d|          \  }}t          j        t          |          dv d            t          |          dk    r||}
}	n |d|          \  }	}
 |d|          \  }} |d|          \  }}|                     d	          }|                     d
          }|                     d          }t	          j        |           }|t           j        k    r-t          j        |                                 dk    d            nQ|t           j        k    r+t          j        |                                 dv d            nt          j        dd            t          ||||	||          }t          ||||
||          }t          | |||	|
||||||||||           |||fS )Nc                      t          j        t          |          dv  fd           |d         }t          |          dk    r|n|d         }||fS )Nr  c                      d  dS )Nzmax_pool2d: r  r5   r  s   r/   rT   zEmax_pool2d_checks_and_compute_shape.<locals>.unpack.<locals>.<lambda>  r  r1   r   r   r  r  s   `   r/   r  z3max_pool2d_checks_and_compute_shape.<locals>.unpack  r  r1   rs  r  c                      dS )NzOmax_pool2d: stride must either be omitted, a single int, or a tuple of two intsr5   r5   r1   r/   rT   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  r  r1   r   r   r  ri  r  r  r   r  c                      dS )NzMnon-empty 4D (batch mode) tensor expected for input with channels_last layoutr5   r5   r1   r/   rT   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  s    c r1   r9  c                      dS )Nz9non-empty 3D or 4D (batch mode) tensor expected for inputr5   r5   r1   r/   rT   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>      O r1   Fc                      dS )Nz?Unsupport memory format. Supports only ChannelsLast, Contiguousr5   r5   r1   r/   rT   z5max_pool2d_checks_and_compute_shape.<locals>.<lambda>  s    U r1   )rE   rV   r   r   r?   r{  r|  rs   r   r  r  )r   rs  r   r  ri  r  r  r  r  r  r  r  r  r  r  r  r  r  r   r  r  s                        r/   r  r    s      VM;//FB	LFy aa   6{{aRB&))B	7++JD$!6*h77Iy**R..K**R..KBJ/66M+++IIKK1cc	
 	
 	
 	
 
%1	1	1IIKK6!OO	
 	
 	
 	

 	UU	
 	
 	

 (Rr9iXXL&z2tRIVVK



  $ k11r1   c                 P    t          |||||          \  }t          j        j         j        k     fd           |j        fd}	 |	             |	|           t          j                  }
t          j        j        j        j	        |
          S )Nc                  (    dj          d j          S )NzExpected dtype z  for `gradOutput` but got dtype r}   r4  s   r/   rT   z7meta_max_pool2d_with_indices_backward.<locals>.<lambda>  s    a$*aakN_aa r1   c                     t          | dz
             t          | dz
             t          | dz
             d S )Nr   r   r   )r  )r  r  r   r  r  s    r/   _check_dim_sizez>meta_max_pool2d_with_indices_backward.<locals>._check_dim_size  sQ    q$q,777q$q,777q$q+66666r1   rR  )
r  rE   rV   rM   r   r?   r{  rx   r   rp   )r  r   rs  r   r  ri  r  r   r  r  r   r  r   r  r  s   ``         @@@@r/   %meta_max_pool2d_with_indices_backwardr    s     	,k67Hi	 		
 
L
k''aaaaa  
 L9D7 7 7 7 7 7 7 7
 OK   OG/55M;
j{#	   r1   c                    t          | |||||          \  }}}|                                 dk    r|                     d          nd}	t          j        |           }
|                                 dk    r|||g}n|	|||g}t          j        || j        | j        |
          t          j        |t
          j	        | j        |
          fS r  )
r  rs   r   r?   r{  rE   rx   rM   rp   r   r  s               r/   meta_max_pool2d_with_indicesr    s     	,{FGXy	 		
  %yy{{a//UZZ^^^QF/66Myy{{a\;7\;?+<'		
 	
 	
 	+<'		
 	
 	
 r1   c           	         
 t          j         j        dv  fd            j        }t          |dz
  |          D ]G
t          j                             
          dk    d                                  d
 d           Ht          j        t                    dk    d	            t          j        t          |          dk    d
                                 d          }                     d                               d          |dk    r                     d          }nd}t          j         j        j        k    d            t          j        j        dk    fd                               d          }                    d          }                    d          
t          j        ||k    d           t          j        ||k    d            t          j        
dk    
fd           t          j        |d         d         z   dz
  k    fd           t          j        |d         d         z   dz
  k    fd                                            dk    r|||d         |d         g}	n||d         |d         g}	t          j        |	 j         j	                  t          j        |	t           j
         j	                  fS )Nr9  c                      d j          S )Nz:fractional_max_pool2d: Expected 3D or 4D tensor, but got: r  r   s   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>%  s    XTYXX r1   r   r   z^fractional_max_pool2d: Expected input to have non-zero  size for non-batch dimenions, but got r)  z emptyr   c                      dS )NzNfractional_max_pool2d: kernel_size musteither be a single int or tuple of Intsr5   r5   r1   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>3  	     2 r1   c                      dS )NzOfractional_max_pool2d: output_size must either be a single int or tuple of Intsr5   r5   r1   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>8  r  r1   r  r  r   r  r   c                      dS )Nz6Expect _random_samples to have the same dtype as inputr5   r5   r1   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>F  s    H r1   c                      d j          S )Nz1Expect _random samples to have 3 dimensions got, r  )random_sampless   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>J  s    YNDWYY r1   z=Expect _random_samples.size(0) no less then input batch size.c                      dS )Nz<Expect _random_samples.size(1) equals to input channel size.r5   r5   r1   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>V      N r1   c                      d  dS )Nz/Expect _random_samples.size(2) equals to 2 got .r5   )r   s   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>X  s    !WST!W!W!W r1   c                       dd          d  S )Nz%fractional_max_pool2d: kernel height r   z' is too large relative to input height r5   )input_heightrs  s   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>\  s    }A}}o{}} r1   c                       dd          d  S )Nz$fractional_max_pool2d: kernel width r   z& is too large relative to input width r5   )input_widthrs  s   r/   rT   z,meta_fractional_max_pool2d.<locals>.<lambda>`  s    z{1~zzmxzz r1   rM   rp   )rE   rV   r   r   r   r   rM   rs   rx   rp   r   )r   rs  r"  r  r   input_channelsinput_batchr   cr   r   r  r  s   `` `      @@@r/   meta_fractional_max_pool2dr  !  sb   	L	VXXXX   9D4!8T"" 
 
IIaLL1]6:iikk] ]ST] ] ]	
 	
 	
 	
 
LKA	2 	2  
 
LKA	2 	2   YYr]]N99R==L))B--Kqyyiill	L
n**HH   
Lq YYYY  
 	AAAAAA	L	[G   
L	^NN   
LaWWWWXXX	LAQ'!+|;}}}}}   
LAQ'!+{:zzzzz  
 xxzzQ^[^[^LAA? 	*;	
 	
 	

 	+;	
 	
 	
 r1   c                 T   t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        | pt          |          dv d            |s|n|d         }	|s|nt          |          dk    r|	n|d         }
|s|nt          |          dk    r|	n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        | j        d	v d
            | j        dk    r|                     d          nd}|                     d          }|                     d          }|                     d          }|                     d          }t          ||||	||          }t          ||||
||          }t          ||||||          }t          | |||||	|
|||||||||||||d           | j        dk    o!t          j        |           t           j	        k    }| j        dk    rQ| 
                    d          }|                                 o|                    t           j	                  }||||f}n|||||f}|                     |          }|                     |t           j                  }|r@|                    t           j	                  }|                    t           j	                  }||fS )Nr  c                      dS NzMmax_pool3d: kernel_size must either be a single int, or a tuple of three intsr5   r5   r1   r/   rT   z.meta_max_pool3d_with_indices.<locals>.<lambda>      _ r1   r   r   r   c                      dS NzQmax_pool3d: stride must either be omitted, a single int, or a tuple of three intsr5   r5   r1   r/   rT   z.meta_max_pool3d_with_indices.<locals>.<lambda>      c r1   c                      dS NzImax_pool3d: padding must either be a single int, or a tuple of three intsr5   r5   r1   r/   rT   z.meta_max_pool3d_with_indices.<locals>.<lambda>      [ r1   c                      dS NzJmax_pool3d: dilation must be either a single int, or a tuple of three intsr5   r5   r1   r/   rT   z.meta_max_pool3d_with_indices.<locals>.<lambda>  rC  r1   r  c                      dS r  r5   r5   r1   r/   rT   z.meta_max_pool3d_with_indices.<locals>.<lambda>  r  r1   r  r  r  r  r   zmax_pool3d_with_indices()r  r   r}   )rE   rV   r   r   r   r  r  r?   r{  r  rj  r6  r   r   r   )r   rs  r   r  ri  r  r	  r  r  r
  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r|  input_channels_last_checkr   r   r   s                                  r/   meta_max_pool3d_with_indicesr  v  s    
LKF"__   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+cc   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LG[[   
B7||q  gajB7||q  gajB	LH\\   I ]]a//		Xa[I ]]a//		Xa[I	L
fKK  
  %zQUZZ^^^AFjjnnGJJrNNEjjnnGZZ^^F BIyIIE"7BB	9MMG!&"b"iKKF








#+  2 	
aXE7>>%BXX  zQ$)OOA$6$6!)77999
'550 6 
 
 	
 eWf5		WeWf=	
//)
$
$Cooiu{o;;G Cff5#9f::**5+A*BB<r1   c                    t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }	t          |          dk    r|n|d         }
t          j        | pt          |          dv d            |s|n|d         }|s|	nt          |          dk    r|n|d         }|s|
nt          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        t          |          dv d            |d         }t          |          dk    r|n|d         }t          |          dk    r|n|d         }t          j        |j        d	v d
            |                    d          }|                    d          }|                    d          }|                    d          }|                     d          }|                     d          }|                     d          }t          || ||||	|
|||||||||||||||d           |j        dk    o!t          j        |          t           j        k    }|j        dk    rJ|	                    d          }|
                                 o|
                    t           j                  }|                    |j                  }|r |                    t           j                  }|S )Nr  c                      dS r  r5   r5   r1   r/   rT   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r  r1   r   r   r   c                      dS r  r5   r5   r1   r/   rT   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r  r1   c                      dS r  r5   r5   r1   r/   rT   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r  r1   c                      dS r  r5   r5   r1   r/   rT   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  rC  r1   r  c                      dS r  r5   r5   r1   r/   rT   z7meta_max_pool3d_with_indices_backward.<locals>.<lambda>  r  r1   r  r  r  r   z"max_pool3d_with_indices_backward()r  r  r   )rE   rV   r   r   r   r  r?   r{  r  rj  r6  r   r   r   )r  r   rs  r   r  ri  r  r   r	  r  r  r
  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r|  r  r  s                                 r/   %meta_max_pool3d_with_indices_backwardr     s    
LKF"__   
QB;1$$+a.B;1$$+a.B	L
+c&kkV+cc   	(vayB	Fc&kkQ&6&6F1IB	Fc&kkQ&6&6F1IB	LG[[   
B7||q  gajB7||q  gajB	LH\\   I ]]a//		Xa[I ]]a//		Xa[I	L
fKK  
 jjnnGJJrNNEjjnnGZZ^^FR  Er""Gb!!F#








,/  6 	
aXE7>>%BXX  zQ$)OOA$6$6!)77999
'550 6 
 
 	 --J I]]1G]HH
r1   gridc                 "    t          j         j        j        k     fd           t          j         j        t           j        k    oj        t           j        k     fd           t          j         j        d         j        d         k     fd           t          j        j        d          j        dz
  k     fd           t          d j                  D ]*t          j         j                 dk     fd           +d S )	Nc                  (    dj          d j          S )NzNgrid_sampler(): expected input and grid to be on same device, but input is on z and grid is on r~  r  r   s   r/   rT   z+check_grid_sampler_common.<locals>.<lambda>I  s0    A\A A37;A A r1   c                  (    dj          d j          S )NzTgrid_sampler(): expected input and grid to have torch.strided layout, but input has z and grid has )ro   r  s   r/   rT   z+check_grid_sampler_common.<locals>.<lambda>P  s0    CC C59[C C r1   r   c                  (    dj          d j          S )NzZgrid_sampler(): expected grid and input to have same batch size, but got input with sizes  and grid with sizes r  r  s   r/   rT   z+check_grid_sampler_common.<locals>.<lambda>W  s0    O %O OBF*O O r1   r   r   c                  .    dj         dz
   d j         S )Nz+grid_sampler(): expected grid to have size r   z, in last dimension, but got grid with sizes )r   r   r  s   r/   rT   z+check_grid_sampler_common.<locals>.<lambda>^  s.    ?%*q. ? ?26*? ? r1   c                       dj          d  dS )NzYgrid_sampler(): expected input to have non-empty spatial dimensions, but input has sizes r)  r*  r  r>  s   r/   rT   z+check_grid_sampler_common.<locals>.<lambda>g  r?  r1   )rE   rV   rp   ro   r  r   r   r   )r   r  r   s   ``@r/   check_grid_sampler_commonr
  F  st   	L#	
 	
 	
 	
 	
   
L%F$+*F	
 	
 	
 	
 	
   
LA$*Q-'	
 	
 	
 	
 	
   
L
2%*q.(	
 	
 	
 	
 	
   1ej!! 
 
KNQ    	
 	
 	
 	

 
r1   c                       e Zd ZdZdZdZdS )GridSamplerInterpolationr   r   r   N)ri   
__module____qualname__BILINEARNEARESTBICUBICr5   r1   r/   r  r  n  s        HGGGGr1   r  interpolation_modec                      t          j         j        dk    o j        j        k     fd           t          j         j        dk    o|t          j        j        k     d            d S )Nr  c                  (    dj          d j          S )Nzdgrid_sampler(): expected 5D input and grid with same number of dimensions, but got input with sizes r  r  r  s   r/   rT   z'check_grid_sampler_3d.<locals>.<lambda>w  s+    149K1 1$(J1 1 r1   c                      dS )Nz<grid_sampler(): bicubic interpolation only supports 4D inputr5   r5   r1   r/   rT   z'check_grid_sampler_3d.<locals>.<lambda>  r  r1   )rE   rV   r   r  r  ry  )r   r  r  s   `` r/   check_grid_sampler_3dr  t  s    	L
a3EJ$)3	
 	
 	
 	
 	
   
LJ!O M"&>&F&LL	
 	ON    r1   c                     |d         }|r!t          j        |t           j                  }nd }t          j        |t           j                  }	||	fS Nr   r   )rE   r  r   r   
r  r   r  r  padding_modealign_cornersr~  input_requires_gradr  	grad_grids
             r/   grid_sampler_2d_backward_metar    sX     &a. %e5;RSSS


 U5LMMMI	""r1   c                     t          | |           t          | ||           | j        d         }| j        d         }|j        d         }|j        d         }|j        d         }	|                     |||||	f          S )Nr   r   r   r   )r
  r  r   r   )
r   r  r  r  r  rZ  Cout_Dout_Hout_Ws
             r/   grid_sampler_3dr$    s{     eT***%'9:::AAAAJqMEJqMEJqME??Aq%6777r1   r  c                     t          ||           t          |||           |d         }|r!t          j        |t          j                  }nd }t          j        |t          j                  }	||	fS r  )r
  r  rE   r  r  r   r  s
             r/   grid_sampler_3d_backwardr&    s     eT***%'9:::%a. %!?
 
 


 
 U5STTTIy  r1   c                     |                     dd           }|st          j        |          }||d<   t          j        | g|R i |S )NrM   )rL   r?   	get_dtyperE   rx   )r   r_  rA   rL  rM   s        r/   fullr)    sU    JJw%%E ,
++F7O;t-d---f---r1   c                 n   |t           j        k    rt          j        |d u d            t          j        d|| j        n|||| j        n||          }| j        rN|                    |                                 | 	                                | 
                                           n;|                    |                                 |                                 d           |                    d           |S t          j                            | |||||          }|                    d           |S )Nc                      dS )Nz9memory format option is only supported by strided tensorsr5   r5   r1   r/   rT   zzeros_like.<locals>.<lambda>  r  r1   r   r   Tr|  )rE   
sparse_coorV   rx   rM   rp   	is_sparsesparse_resize_and_clear_r   
sparse_dim	dense_dimrs   _coalesced_r&   r   r  fill_)r   rM   ro   rp   rq   r   rt  s          r/   r  r    sA    !!!T!OO	
 	
 	

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

/
!
!# "  C IIaLLLJr1   c                 <    ddl m}                                  }t          j        |dk    d            dk    rn|z                                  }t          j         | |k              p ||k                fd           dk    rn|z   t                                                     }t                                                     }                                 |         z  z   }|= |=  	                    |||          S )Nr   guard_size_obliviousc                      dS )Nz-select() cannot be applied to a 0-dim tensor.r5   r5   r1   r/   rT   zmeta_select.<locals>.<lambda>  s    ? r1   c                  >    d d                                  d  S )Nzselect(): index z! out of range for tensor of size z at dimension r   rs   r   r   s   r/   rT   zmeta_select.<locals>.<lambda>  s8     ,5 , ,99;;, ,&), , r1   )
r   r5  rs   rE   r   r   r   r   r   r   )	r   rs   r   r5  r   r   new_sizer  new_storage_offsets	   ```      r/   meta_selectr;    sT   JJJJJJ88::D		??  
 ##sTzC99S>>D	  %$//V3G3GQU3V3V	
	, 	, 	, 	, 	, 	,	   aZZEEUT\EDIIKK  Hdkkmm$$J,,..C1HH3??8Z1CDDDr1   c                 *    t          j        |           S r,   r?   clone_preserve_strides)r   r  rs   r   s       r/   meta_select_scatterr?        '---r1   c                 *    t          j        |           S r,   r=  )r   r  rs   rk   rj   steps         r/   meta_slice_scatterrC  !  r@  r1   dim_post_exprwrap_scalarc                     |dk    r|sJ d}| }|dz
  }| |k     s| |k    rJ d|  d| d| d            | dk     r| |z  } | S )Nr   r   zdim z out of bounds (rf   rg   r5   )rs   rD  rE  r  r  s        r/   r   r   '  s~    {.C
!
Cc		S3YYY)R)R)RS)R)RC)R)R)RYY'
Qww}Jr1   c                 P    |                                  dk    rdn| j        |         S r  r  )r  rs   s     r/   ensure_nonempty_sizerH  3  s"    111!'#,.r1   c                 p    t                                           d          }t                                          d          }t          j        ||k    d            t	          |          D ]Ck    r;t          j        t                    t                     k     fd           Dd S )Nr   c                      dS )NzDIndex tensor must have the same number of dimensions as input tensorr5   r5   r1   r/   rT   z$gather_shape_check.<locals>.<lambda>=  s    V r1   c                  :    d dj          dj          d  z   S )Nz!Size does not match at dimension z expected index  to be no larger than self  apart from dimension r  )rs   r   r   r   s   r/   rT   z$gather_shape_check.<locals>.<lambda>C  s6    \A\\u{\\W
WWRUWWX r1   )r  rs   rE   rV   r   rH  )r   rs   r   	self_dims
index_dimsr   s   ```  @r/   gather_shape_checkrP  8  s    DHHJJ""IUYY[[!$$J	LZVV   9  88L$UA..2FtQ2O2OOX X X X X X X   r1   c                 F   ddl m} t          ||                                           } |                                dk              }|s<t          j        j        t
          j        k    fd           t          | |           | 
                    j                  S )Nr   r4  c                      d j          S )Nz2gather(): Expected dtype int64 for index, but got r}   r~   s   r/   rT   zmeta_gather.<locals>.<lambda>Q  s    VVV r1   )r   r5  r   rs   r   rE   rV   rM   r   rP  r   r   )r   rs   r   sparse_gradr5  wrapped_dimis_index_emptys     `    r/   meta_gatherrV  H  s    JJJJJJ dhhjj11K))%++--1*<==N 5K5:%VVVV	
 	
 	
 	4e444>>%+&&&r1   c                     |r@| dk    rdS | dk    rdS | dk    rdS | dk    rdS | d	k    rd
S t          j        dd            d S | dk    rdS | dk    rdS t          j        dd            d S )Nsum
REDUCE_ADDr  REDUCE_MULTIPLYmeanREDUCE_MEANamaxREDUCE_MAXIMUMaminREDUCE_MINIMUMFc                      dS )Nz=reduce argument must be either sum, prod, mean, amax or amin.r5   r5   r1   r/   rT   z#get_operator_enum.<locals>.<lambda>f  s    S r1   addmultiplyc                      dS )Nz/reduce argument must be either add or multiply.r5   r5   r1   r/   rT   z#get_operator_enum.<locals>.<lambda>n  s    $U r1   r9  )reduce_use_new_optionss     r/   get_operator_enumrg  X  s     e<$$ =####SS	
 	
 	
 	e<
""$$UUUVVVr1   c                      ddl m}  ||                                dk              r+t          j        |j        t          j        k     fd           |(t          j        |j        |j        k     fd           d S d S )Nr   r4  c                        dS )Nz"(): Expected dtype int64 for indexr5   method_names   r/   rT   z,scatter_gather_dtype_check.<locals>.<lambda>y  s    {FFF r1   c                        dS )Nz0(): Expected self.dtype to be equal to src.dtyper5   rj  s   r/   rT   z,scatter_gather_dtype_check.<locals>.<lambda>  s    {TTT r1   )r   r5  r   rE   rV   rM   r   )rk  r   r   src_optr5  s   `    r/   scatter_gather_dtype_checkrn  s  s    JJJJJJEKKMMQ.// 
K5:%FFFF	
 	
 	

 J'-'TTTT	
 	
 	
 	
 	
 r1   c                 "    t          | d          S r  )r  r   s    r/   ensure_nonempty_dimrp    s    sA;;r1   c                 z    ddl m}  |                                dk              rd S t          j        t                                                     t                                                    k    d            d}t                                                     }t          |          D ]1}t          |          }|k    r|t           |          k    rd} n2|s<:t          |          D ]*}t          |          }|t          |          k    rd} n+vt          j        t                                                     t                                                    k    d            t          j        |  fd           d S t          j        |  fd           d S )	Nr   r4  c                      dS NzCIndex tensor must have the same number of dimensions as self tensorr5   r5   r1   r/   rT   z%scatter_shape_check.<locals>.<lambda>  rA  r1   FTc                      dS rs  r5   r5   r1   r/   rT   z%scatter_shape_check.<locals>.<lambda>  s    Y r1   c                  D    dj          dj          d  dj          z   S )NExpected index rL  rM  z and to be no larger than src r  )rs   r   r   rm  s   r/   rT   z%scatter_shape_check.<locals>.<lambda>  s7    ZekZZdjZZYsYY'-YYZ r1   c                  4    dj          dj          d  z   S )Nrv  rL  rM  r  r8  s   r/   rT   z%scatter_shape_check.<locals>.<lambda>  s,    ZekZZdjZZ,s,,- r1   )	r   r5  r   rE   rV   rp  rs   r   rH  )	r   rs   r   rm  r5  is_wrong_shaperN  r   index_d_sizes	   ````     r/   scatter_shape_checkrz    s.   JJJJJJEKKMMQ.// 	LDHHJJ''+>uyy{{+K+KKUU  
 N#DHHJJ//I 9  +E15588.tQ7777!NE 8
  g1y!! 	 	A/q99L27A>>>>!% ? 

++/B599;;/O/OOYY	
 	
 	
 	Z Z Z Z Z Z Z	
 	
 	
 	
 	
 	- - - - - -	
 	
 	
 	
 	
r1   c                     t          ||                                           }t          d| ||           t          | |||           |t	          ||           d S d S )Nscatter)r   rs   rn  rz  rg  )r   rs   r   r  re  rf  rT  s          r/   scatter_meta_implr}    se     dhhjj11Ky$s;;;k5#666'?33333 r1   c                 \    t          | |||d           |                     | j                  S Nrb  r}  r   r   r   rs   r   r  s       r/   meta_scatter_addr    s,    dCU333>>$*%%%r1   c                 ,    t          | |||d           | S r  r}  r  s       r/   meta_scatter_add_r    s    dCU333Kr1   c                     t          |t          j                  r|nd }t          | ||||           |                     | j                  S r,   )r_   rE   r   r}  r   r   r   rs   r   src_or_valuer?  r  s         r/   meta_scatterr    sH     %\5<@@
J,,dCdCV444>>$*%%%r1   c                 h    t          |t          j                  r|nd }t          | ||||           | S r,   )r_   rE   r   r}  r  s         r/   meta_scatter_r    s9     %\5<@@
J,,dCdCV444Kr1           queryr   ry  	dropout_p	is_causalreturn_debug_maskrI  c                 *   |                      d          }|                      d          }|                      d          }	|                      d          }
|                     d          }|                     dd          }t          j        |                              dd          }t          j        |||	ft          j        | j                  }|rX|
dk    rdnd}t          j        |	|z            }|dk    rd}n|dk    rd}t          j        |||	|f| j	        | j                  }n!t          j        d| j	        | j                  }||d d |	|t          j        d	t          j
        d
          t          j        d	t          j
        d
          |f	S )Nr   r   r   r   r  @         r5   rn   )r   r  rE   r   rx   rI   rp   r  ceilrM   r   )r  r   ry  r  r  r  rI  r   	num_headsmax_seqlen_batch_qhead_dimmax_seqlen_batch_kquery_t	attention	logsumexpblocksize_cmax_seqlen_k
debug_masks                     r/   (meta__scaled_dot_product_flash_attentionr    s    AJ

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


 [%+elKKK
 	Bej888Bej888
 
r1   	attn_biascompute_log_sumexpc	           	      4   |                      d          }	|                      d          }
|                      d          }|                     d          }|                      d          }|                     d          }t          j        |	|
||f| j        | j                  }t          j        |	|
|ft          j        | j                  }t          j        dt          j        d          }t          j        dt          j        d          }||d d ||||d f	S )Nr   r   r   r   r  r5   rn   )r   rE   rx   rM   rp   rI   r   )r  r   ry  r  r  r  r  r  rI  r  r  S_QS_KVD_QKD_Vrt  
logsum_expseedoffsets                      r/   (meta__scaled_dot_product_cudnn_attentionr  $  s    	

1A

1A
**Q--C88A;;D::b>>D
**R..C
+q!S#&ek%,
O
O
OC	
Ask|  J ;rF;;;D[5:f===F 	
 
r1   r+  r  	cum_seq_q	cum_seq_kmax_qmax_kphilox_seedphilox_offsetc                 t   t          j        |                    dd                                        dd          }t          j        |                    dd                                        dd          }t          j        |                    dd                                        dd          }|||fS r  )rE   r   r  )r+  r  r   ry  r   r  r  r  r  r  r  r  r  r  rI  grad_qgrad_kgrad_vs                     r/   'meta__scaled_dot_product_flash_backwardr  O  s    , eooa3344>>q!DDFcmmAq1122<<QBBFeooa3344>>q!DDF66!!r1   	attn_maskc                 T   |                      d          }|                      d          }|                      d          }	|                      d          }
t          j        |           }t          j        ||	|ft          j        | j                                      dd          }||fS )Nr   r   r   r   r  )r   rE   r   rx   rI   rp   r  )r  r   ry  r  r  r  rI  r   r  r  r  r  r  s                r/   0meta__scaled_dot_product_flash_attention_for_cpur  k  s     AJ

1IAzz!}}H ''I	

 k|   i1oo  	 r1   c
                    |                     d          }
|                     d          }|                     d          }|                     d          }|                     d          }t          j        |
|||fd|j        |j                  }t          j        |
|||fd|j        |j                  }t          j        |
|||fd|j        |j                  }|||fS )Nr   r   r   r   r   r   r   r   r  )r   rE   empty_permutedrM   rp   )r+  r  r   ry  r   r  r  r  r  rI  r   r  r  len_qlen_kr  r  r  s                     r/   9meta__scaled_dot_product_flash_attention_for_cpu_backwardr    s    & AJ

1Izz!}}HJJqMMEHHQKKE!	Yx0k|	  F !	Yx0iz	  F !	Yx0k|	  F 66!!r1   c                    |                      dd          } |                     dd          }|                     dd          }|                     d          }|                     d          }	|                    d          }
|                     d          }|                     d          }|                    d          }t          j        ||	||| j        | j                  }|rt          j        |	dz            dz  nd}t          j        |||ft          j        | j                  }|                     dd          }t          j        dt          j	        d	          }t          j        dt          j	        d	          }||||fS )
Nr   r   r   r  r   r  r  r5   rn   )
r  r   rE   rx   rM   rp   r  r  rI   r   )r  r   ry  r  r  r  r  rI  r  r  rZ  r  KKvrt  logsumexp_dimr  r  r  s                      r/   ,meta__scaled_dot_product_efficient_attentionr    sY    OOAq!!E
--1

COOAq!!E

1A

1AA

2I

2A	BB
+aIrU\
R
R
RC.@GDIa"f%%**aM	
I}%k|  J --1

C ;rF;;;D[5:f===F
D&((r1   grad_input_maskc                    |                     d          }|                     d          }|                     d          }|                     d          }|                     d          }|                     d          }t          j        ||||fd|j        |j                  }t          j        ||||fd|j        |j                  }t          j        ||||fd|j        |j                  }d }||
d         r~|                     d          }|dz  dk    r|n
|dz   |dz  z
  }t          |                                           }||d<   t          j        ||j        |j                  }|d	d |f         }||||fS )
Nr   r   r   r   r  r  r   r1  .)r   rE   r  rM   rp   r   rx   )r+  r  r   ry  r  r   r  r  r  r  r  r  rI  r   r  r  r  
head_dim_vr  r  r  r  	grad_biaslastDimlastDimAligned	new_sizess                             r/   +meta__scaled_dot_product_efficient_backwardr    s   ( AJ

1IJJqMMEzz!}}HAJHHQKKE!	Yx0k|	  F !	Yx0iz	  F !	Yz2k|	  F I!3..$$$+bLA$5$57R<'TV,;V))**	&	"KY_Y5E
 
 
	 c8G8m,	669,,r1   c                     t          j        |          }t          j        |          }t          j        |          }|||fS r,   r?  )r+  r  r   ry  r   r  r  r  r  r  r  r  r  r  r  rI  r  r  r  s                      r/   'meta__scaled_dot_product_cudnn_backwardr  !  sA    . e$$Fc""Fe$$F66!!r1   window_size_leftwindow_size_right	seqused_kalibi_slopesc                    ||                      d          n|                                dz
  }||                      d          n|}||                     d          n|}|                      d          }|                      d          }t          j        |           }t          j        |||ft          j        | j                  }|	rX|dk    rdnd}t          j        ||z            }|dk    rd}n|dk    rd}t          j        ||||f| j	        | j                  }n!t          j        d| j	        | j                  }||t          j        d	t          j
        d
          t          j        d	t          j
        d
          |fS )Nr   r   r  r   r  r  r  r  r5   rn   )r   r   rE   r   rx   rI   rp   r  r  rM   r   )r  r   ry  r  r  r  r  r  r  r  rI  r  r  r  r  r   r  r  r  r  r  r  r  r  r  s                            r/   meta__flash_attention_forwardr  >  s   4 #,"3A9J9JQ9NJ*3*;A(1(9!u

2Izz"~~H  ''I	Y 23k|  I  L%]]ccy!3k!ABB$$LL3&&L[$6E+<
 
 


 [%+elKKK
 	Bej888Bej888 r1   c                     t          j        |          }t          j        |          }t          j        |          }|||fS r,   r?  )r+  r  r   ry  r   r  r  r  r  r  r  r  r  r  rI  r  r  
grad_querygrad_key
grad_values                       r/   meta__flash_attention_backwardr    sA    0 !%((J$$H!%((Jx++r1   cu_seqlens_qcu_seqlens_kmax_seqlen_qr  custom_mask_typecausal_diagonalseqlen_kwindow_sizec                    |                      d          }|                      d          }|                     d          }|                      d          }|                      d          }|                     d          }t          j        ||||| j        | j                  }||                     d          dz
  n|}|}||J |}||n|}|
rt          j        |dz            dz  nd}t          j        |||ft          j        | j                  }t          j        dt          j        d          }t          j        dt          j        d          }||||||fS )	Nr   r   r  r   r  r  r5   rn   )	r   rE   rx   rM   rp   r  r  rI   r   )r  r   ry  r  r  r  r  r  r  r  r  rI  r  r  r  r  r  rZ  r  r  r  rt  logsumexp_batch_dimactual_max_seqlen_qactual_max_seqlen_kr  r  r  r  s                                r/   !meta__efficient_attention_forwardr    si   , 	

1A

1AA

2I

2A	BB
+aIrU\
R
R
RC7C7O,++A..22VW'''**6*B,,4FM	%*++b00A  	i7k|  J ;rF;;;D[5:f===F
D&*=?RRRr1   bias_requires_gradnum_splits_keyshared_storage_dqdkdvc                    |rt          j        |j        d         |j        d         k    d            t          j        |j        d         |j        d         k    d            t          j        g |j        dd         d|j        d         |j        d         R |j        |j                  }|                    d	d          }|                    d	d          }|                    d	d
          }n<t          j        |          }t          j        |          }t          j        |          }||                    d          }|dz  dk    r|n
|dz   |dz  z
  }t          |                                          }||d<   t          j        ||j        |j                  }|dd |f         }nt          j        d|j                  }||||fS )Nr   c                      dS )Nz,seqlen must match for `shared_storage_dqdkdvr5   r5   r1   r/   rT   z4meta__efficient_attention_backward.<locals>.<lambda>  s    B r1   r   c                      dS )Nz3embedding dim must match for `shared_storage_dqdkdvr5   r5   r1   r/   rT   z4meta__efficient_attention_backward.<locals>.<lambda>  s    I r1   r   r  r   r  r  r   r1  .r5   r~  )
rE   rV   r   rx   rM   rp   rl  r   r   r   )r+  r  r   ry  r  r  r  r  r  r  r  r  r  r  r  rI  r  r  chunkr  r  r  r  r  r  r  s                             r/   "meta__efficient_attention_backwardr    s   2  -KNcil*BB	
 	
 	
 	KNcil*II	
 	
 	
 Eek!B$EEEKOEU[_EE+<
 
 

 \\"a((
<<A&&\\"a((

%e,,
#C((%e,,
))B--$+bLA$5$57R<'TV,;V%%	&	"K	DKPPP	c8G8m,		K5<888	xY66r1   scale_ascale_bscale_resultuse_fast_accumc                     d }t          j                                         dk    o                                dk     fd           t          j         | j                  o |j                   fd           t	                     dk    rd }	d }
d }t          j         |	                                           p
 |            fd	           t          j         |
                                          p
 |          fd
           t          j                             d          dz  dk     fd           t          j                            d          dz  dk    o                    d          dz  dk    fd           t          j        j        t           j        k    oj        t           j        k    d             j        \  }                    d          	                                dk    r	                                dk    rnt          j                                        dk    o                                dk    fd                               d          k    r                    d          dk    ro                    d          dk    rV                    d          k    r=t          j        
                                o
                                d            nt          j        dfd           ||n j        }t          j                             d                              d          | j                  S )Nc                 b    | t           j        t           j        t           j        t           j        fv S r,   )rE   r5  float8_e5m2float8_e4m3fnuzfloat8_e5m2fnuzr}   s    r/   is_fp8_typez#meta_scaled_mm.<locals>.is_fp8_type  s,    !!	
 
 	
r1   r   c                  \    d                                  d                                   S )Nz%Inputs must be 2D but got self.dim()=z and mat2.dim()=r   r   r   s   r/   rT   z meta_scaled_mm.<locals>.<lambda>$  s-    `

``TXT\T\T^T^`` r1   c                  (    dj          d j          S )Nz8Expected both inputs to be fp8 types but got self.dtype=z and mat2.dtype=r}   r  s   r/   rT   z meta_scaled_mm.<locals>.<lambda>(  s    s4:ssgkgqss r1   rK  c                 >    | d         | d         k    o| d         dk    S r  r5   r  s    r/   is_row_majorz$meta_scaled_mm.<locals>.is_row_major-  s"    !9vay(;VAY!^;r1   c                 2    | d         dk    o| d         dk    S r  r5   r  s    r/   is_col_majorz$meta_scaled_mm.<locals>.is_col_major0  s    !9>3fQi!m3r1   c                 f    |                      d          dk    p|                      d          dk    S r  r   )	tensor_2ds    r/   has_zero_dimz$meta_scaled_mm.<locals>.has_zero_dim3  s/    >>!$$)CY^^A->->!-CCr1   c                  2    d                                   S )Nz#self must be row_major, got stride r  r   s   r/   rT   z meta_scaled_mm.<locals>.<lambda>8      I$++--II r1   c                  2    d                                   S )Nz#mat2 must be col_major, got stride r  r   s   r/   rT   z meta_scaled_mm.<locals>.<lambda><  r  r1   r   r1  r   c                  4    d                      d           S )NzBExpected self.size(1) to be divisible by 16, but got self.size(1)=r   r   r   s   r/   rT   z meta_scaled_mm.<locals>.<lambda>@  s     gY]YbYbcdYeYegg r1   c                      d j          S )Nz>Expected both dimensions of mat2 to be divisble by 16 but got r  r  s   r/   rT   z meta_scaled_mm.<locals>.<lambda>D  s    aUYU_aa r1   c                      dS )Nz6Both scale_a and scale_b must be float (fp32) tensors.r5   r5   r1   r/   rT   z meta_scaled_mm.<locals>.<lambda>J  s    L r1   c                  \    d                                  d                                 S )NzLFor non-tensorwise scaling, scale tensors must be 2D, but got scale_a.dim()=z and scale_b.dim()=r   )r  r  s   r/   rT   z meta_scaled_mm.<locals>.<lambda>U  s2    ~Y`YdYdYfYf~~nunynyn{n{~~ r1   c                      dS )Nz@Both scale_a and scale_b must be contiguous for rowwise scaling.r5   r5   r1   r/   rT   z meta_scaled_mm.<locals>.<lambda>a  s    ^ r1   Fc                      d  d d                     d           d                     d           d                     d           d                     d           dS )	Nz}Invalid scaling configuration. For tensorwise scaling, both scales should be scalar. For rowwise scaling, scale_a should be (z, 1), scale_b should be (1, z). Got scale_a.size()=(r   rf   r   z) and scale_b.size()=(rg   r   )r:  r   r  r  s   r/   rT   z meta_scaled_mm.<locals>.<lambda>g  s    UCDU UbcU U 07||AU U CJ,,q//U U 07||A	U U CJ,,q//	U U U r1   r  )rE   rV   rs   rM   rL  r   r   r2  r   r   r6  rx   rp   )r   r   r  r  r  r  r  r  r  r  r  r  r9  
_out_dtyper:  r   s   ````          @@r/   meta_scaled_mmr    s   
 
 
 
L

a+DHHJJ!O`````   
LDJ;KK
$;$;sssss  
 4F""	< 	< 	<	4 	4 	4	D 	D 	D 	L''=<<+=+=IIII	
 	
 	
 	L''=<<+=+=IIII	
 	
 	
 	IIaLL2"gggg	
 	
 	
 	IIaLL2"=tyy||b'8A'=aaaa	
 	
 	
 	MU]*Mw}/MLL	
 	
 	
 z1IIaLL==??aGMMOOq$8$8 L"9w{{}}'9~~~~~   Q1$$LLOOq((LLOOq((LLOOq(( ))++G0E0E0G0G^^          	 	 	 (3J;tyy||TYYq\\DKXXXXr1   c                 `    t          | ||||d           |                     | j                  S NT)rf  r  r   rs   r   r  r?  r=  s         r/   meta_scatter_reduce_twor  t  s3     dCVTJJJJ>>$*%%%r1   c                 0    t          | ||||d           | S r  r  r  s         r/   meta_scatter_reduce__twor  {  s"    dCVTJJJJKr1   c                d    t          j        d                                 cxk     odk    nc  fd                                            dk    r&t          j        |t           j         j                  S t          j                             d          |t           j         j                  S )Nr   r   c                  2    d                                   S )Nz@The probabilty distributions dimensions must be 1 or 2, but got r   r;  s   r/   rT   z"meta_multinomial.<locals>.<lambda>  s    `SXS\S\S^S^`` r1   r   r  )rE   rV   rs   rx   r   rp   r   )r   num_samplesreplacementr   s   `   r/   meta_multinomialr    s     
L	EIIKK1````   yy{{a{;ejNNNN;

1{%*U\   r1   c                     d}| D ]}||z  }|S r  r5   )vsr7  vs      r/   multiply_integersr    s&    	A  	QHr1   c                     t          j        t                    k    fd           dz   t          j        t                     k     fd           t          j        t          d  dd          D                       ot          d D                        fd            d d         \  }}||gR S )Nc                  .    d  dt                     S )Nz%It is expected output_size equals to , but got size r  )num_spatial_dimsr"  s   r/   rT   z'upsample_common_check.<locals>.<lambda>  s#    k8HkkY\]hYiYikk r1   r   c                  .    d  dt                     S )Nz$It is expected input_size equals to r  r  )expected_input_dimsr  s   r/   rT   z'upsample_common_check.<locals>.<lambda>  s#    l7Jll[^_i[j[jll r1   c              3   "   K   | ]
}|d k    V  dS r   Nr5   r;   rz  s     r/   rb   z(upsample_common_check.<locals>.<genexpr>  s&      **aAE******r1   c              3   "   K   | ]
}|d k    V  dS r$  r5   r%  s     r/   rb   z(upsample_common_check.<locals>.<genexpr>  s&      2N2NQ1q52N2N2N2N2N2Nr1   c                      d  d S )NzDInput and output sizes should be greater than 0, but got input size z and output size r5   )r  r"  s   r/   rT   z'upsample_common_check.<locals>.<lambda>  s)     A A A3>A A r1   )rE   rV   r   rk  )r  r"  r   r  channelsr"  s   ```  @r/   upsample_common_checkr)    s   	LK,,kkkkk   +Q.	LJ..lllll  
 
L**:abb>*****Ns2N2N+2N2N2N/N/N	A 	A 	A 	A 	A   ""1"~FHH+{+++r1   c                 p    t          j                                         dk    p(t                                           dd                     fd           t                                           |d          }                     |                              t          j	                             S )Nr   r   c                  2    d                                   S )Nz>Non-empty 3D data tensor expected but got a tensor with sizes r   r;  s   r/   rT   z$upsample_nearest1d.<locals>.<lambda>      _QVQ[Q[Q]Q]__ r1   r   r   
rE   rV   r   r  r   r)  r   r   r?   r{  )r   r"  scalesfull_output_sizes   `   r/   upsample_nearest1dr1         
LA/

QRR0@AA____   -

kA   ??+,,//1%88 0   r1   c                     t          j                                         dk    p(t                                           dd                     fd           t                                           |d          }                     |          }t          j                   } j	        \  }}}} j
        j        dk    r|dk     rt           j        }|                    |          }|S )	Nr   r   c                  2    d                                   S Nz>Non-empty 4D data tensor expected but got a tensor with sizes r   r;  s   r/   rT   z$upsample_nearest2d.<locals>.<lambda>  r,  r1   r   r-  rK  r  r   )rE   rV   r   r  r   r)  r   r?   r{  r   rp   rh   r   r   )	r   r"  scales_hscales_wr0  r   r   rB   
n_channelss	   `        r/   upsample_nearest2dr9    s     
LA/

QRR0@AA____   -

kA   __-..F /66M  +Az1a|F""zA~~/];;FMr1   r"  r  r6  r7  c                 v    t          ||d          t          j         j        dk     fd           t	          d          D ]9t          j                                                k     fd           :                     |                              t          j	                             S )Nr   r-  r  c                      d j          S )NzFExpected grad_output to be a tensor of dimension 4 but got: dimension r  rJ  s   r/   rT   z-upsample_nearest2d_backward.<locals>.<lambda>  s    kYdYikk r1   c            
      R    d d           d d                                S )NzCExpected grad_output to have the same shape as output; output.size(z) = z but got grad_output.size(r   )r0  r  r   s   r/   rT   z-upsample_nearest2d_backward.<locals>.<lambda>  s]    J !J J'7':J J-.J J4?4D4DQ4G4GJ J r1   r   )
r)  rE   rV   r   r   r   r   r   r?   r{  )r  r"  r  r6  r7  r0  r   s   `    @@r/   upsample_nearest2d_backwardr=    s     -K!   
LAkkkk   1XX 
 
Q#3A#66     	
 	
 	
 	
   ,,//1+>> 0   r1   c                 p    t          j                                         dk    p(t                                           dd                     fd           t                                           |d          }                     |                              t          j	                             S )Nr   r   c                  2    d                                   S )Nz>Non-empty 5D data tensor expected but got a tensor with sizes r   r;  s   r/   rT   z$upsample_nearest3d.<locals>.<lambda>  r,  r1   r   r-  r   r.  )r   r"  scales_dr6  r7  r0  s   `     r/   upsample_nearest3drA    r2  r1   c                    t          j        |           t          j        | t           j                  }}||t          |t                    sJ t          |t                    sJ |j        }|                                }	t          ||          }t          ||          }|                    ||	           |                    ||	           t          ||           t          ||           ||fS ||fS )Nr}   )rg  rh  )
rE   r   r   r_   r   r   r   r   r  r!   )
r   stablers   
descendingr   r   r  r   r   
out_strides
             r/   	meta_sortrF    s     D!!5#3D#L#L#LqAg1&*-----':..... G	XXZZ
"6955#GY779j111Iz222F3333G4444wa4Kr1   c           	          t          j         j        dk     fd           t          j         j        j        k     fd                                d          wt          j        j        dk    fd           t          j                                        k    fd           t          j        j        j        k    fd           t          j        j        dk    fd                                d	          z  z  t          j                                        k     fd
           t          j        t           fdfD                       d            d S )Nr   c                       j          dS Nz != 2r  )input_gatess   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>+      ;3C0J0J0J r1   c                  &    j          d j          S N != r  )hidden_gatesrJ  s   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>.  s    ;$>>,*<>> r1   r   c                       j          dS )Nz != 1r  )
input_biass   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>2  s    jo3L3L3L r1   c                  6                                      d  S rM  r
  )
gates_sizerQ  s   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>5  s!    z''));;z;; r1   c                  &    j          d j          S rM  r  )hidden_biasrQ  s   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>9  s    z'@@[->@@ r1   c                       j          dS rI  r  )prev_hiddens   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>;  rK  r1   r   c            
      p                                      d                    d           d d d  d
S )NrN  r   z * z // z (aka rg   )r   r   )expected_prev_hidden_numelfactorrS  rJ  rW  s   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>?  sL    ;$$&&K,<,<Q,?,?JTZb| r1   c              3   8   K   | ]}|j         j         k    V  d S r,   r~  )r;   r<   rJ  s     r/   rb   z&rnn_cell_checkSizes.<locals>.<genexpr>B  sA       
 
 H**
 
 
 
 
 
r1   c                      dS )Nz%expected all inputs to be same devicer5   r5   r1   r/   rT   z%rnn_cell_checkSizes.<locals>.<lambda>F  s    7 r1   )rE   rV   r   r   r   r   rk  )rJ  rO  rQ  rU  rZ  rW  rY  rS  s   ``````@@r/   rnn_cell_checkSizesr]  #  s    
L!Q&(J(J(J(JKKK	L\//>>>>>   !!!$$JZ_)+L+L+L+LMMM*,;;;;;	
 	
 	
 	 11@@@@@	
 	
 	
 
L!Q&(J(J(J(JKKK!,!1!1!!4!4z!AV!K	L99   
L 
 
 
 
"J[I
 
 
 	
 	
 	87    r1   c                     t          | |||d|           t          j        | t          j                  }t          j        |t          j                  }t          j        |t          j                  }|||fS )Nr  r   )r]  rE   r   r   )rJ  rO  cxrQ  rU  	workspacehycys           r/   _thnn_fused_lstm_cell_metarc  J  sq     \:{ArRRR E<STTTI		"E,C	D	D	DB		"E,C	D	D	DBIr1   c                 t   t          |          dk    }|r%t          |          }|d         }| j        d         }n:|
r| j        d         n| j        d         }|
r| j        d         n| j        d         }d}|rdnd}|dk    r|n|}|r|||z  g}n|
r||||z  gn||||z  g}|                     |          }|	|z  ||g}|t          j        d| j                  }n|                    |          }|                    |	|z  ||g          }|rdnd}|                     |t          j                  }|||||fS )Nr   r   r   r   r~  r}   )r   r   r   rE   rx   rp   r  )r   r  weight_stride0
weight_bufhxr_  r3  hidden_size	proj_size
num_layersbatch_firstdropouttrainbidirectionalbatch_sizesdropout_stateis_input_packed
seq_length
mini_batchbatch_sizes_sumnum_directionsout_sizer   r   
cell_shaperb  ra  reserve_shapereserves                                r/   
_cudnn_rnnrz  Y  s   & +&&!+O %%
 ^
+a.'2FU[^^A
'2FU[^^A
'.QQQN%NNyyH 
$h&?@		 EZX%>??j(^*CD 	
 __Y''F~-z;GJ	z[5<000\\*%%	zN2JI	J	JB %AAAMoom5;o??G2r7J..r1   c                    |r| j         d         n| j         d         }|r| j         d         n| j         d         }|
}|r|||gn|||g}|                     |          }|t          j        d| j                  }n|                    |j                   }|t          j        d| j                  }n|                    |j                   }t          j        d| j        t          j                  }||||fS )Nr   r   r~  r   )r   r   rE   rx   rp   r  )r   w0w1w2w3hx_cx_r   ro  r3  rh  rj  
has_biasesrn  rk  rm  rr  rs  output_chanelsr   r   ra  rb  r`  s                           r/   mkldnn_rnn_layerr    s    & $/BQEKNJ#.BQEKNJ N 	6Z00*n5 
 __Y''F
{[5<000]]39%%
{[5<000]]39%%Ael%+FFFI2r9$$r1   c                     | j         dk    r%t          j        dk    pdk    fd           d S t          j        |                               dk    fd           d S )Nr   r   c                       d  S )Nz4: Expected reduction dim -1 or 0 for scalar but got r5   rs   r  s   r/   rT   z'zero_numel_check_dims.<locals>.<lambda>  s    wYYTWYY r1   c                       d  dS )Nz: Expected reduction dim z to have non-zero size.r5   r  s   r/   rT   z'zero_numel_check_dims.<locals>.<lambda>  s    wUUUUU r1   )r   rE   r   r   )r   rs   r  s    ``r/   zero_numel_check_dimsr    s    yA~~1H!r	YYYYY	
 	
 	
 	
 	

 	IIcNNaUUUUU	
 	
 	
 	
 	
r1   c                      |5t          ||                                          }t          ||            d S t          j        |                                dk     fd           d S )Nr   c                        dS )Nz@: Expected reduction dim to be specified for input.numel() == 0.r5   r  s   r/   rT   z%check_argmax_argmin.<locals>.<lambda>  s    t]]] r1   )r   rs   r  rE   rV   r   )r  r   rs   s   `  r/   check_argmax_argminr    sq    
S$((**--dC.....JJLLA]]]]	
 	
 	
 	
 	
r1   c                     t          d| |           t          j        | j        ||fnd           }t	          | ||          }|                     |t          j                  S )Nargmaxr}   )r  r?   r_  r   r`  r   rE   r   )r   rs   rb  r`  r   s        r/   argmax_argmin_metar    s[    $,,,
coSFF4PPD$T499E>>%u{>333r1   c                 l    |t           j        k    rt           j        }t          j        d||||          S )Nr5   r   )rE   jaggedr  rx   )rz  rM   ro   rp   rq   s        r/   scalar_tensorr    s<    
 ;
%v*   r1   c                    t          ||                                 d          }|                                 dk    rdn|                     |          }t          j        |dk    o||k    d            t          | j                  }t          |          dk    r|||<   |                     |          |                     |t          j	                  fS )NTrE  r   r   c                      dS )Nzk not in range for dimensionr5   r5   r1   r/   rT   ztopk_meta.<locals>.<lambda>  s    4R r1   r}   )
r   rs   r   rE   rV   r   r   r   r   r   )r   r9  rs   largestsorted	sliceSizetopKSizes          r/   	topk_metar    s     dhhjjd
;
;
;CXXZZ1__$))C..I	La*AN,R,RSSSDJH
8}}q>>(##T^^HEK^%P%PPPr1   c                     ||
J d            |                                 }|                                  }	t          j        ||	j        |	j        |	j                  S )Nz;segment_reduce(): Either lengths or offsets must be defined)rM   rp   ro   )r   rE   r   rM   rp   ro   )
r  r   rO  r?  rJ  rK  rL  rN  data_contiggrad_contigs
             r/   meta__segment_reduce_backwardr    sm     	w22D  322//##K//##K!!	   r1   c                    t          |                                 d          |                                 dk    r|                               nd}t          j        |dk    o||k    fd           t          | j        d          | j        dz   d          z             }|r.|                                 dk    r|                    d           |                     |          |                     |t          j	                  fS )NTr  r   r   c                      d  S )Nz9kthvalue(): selected number k out of range for dimension r5   r   s   r/   rT   zkthvalue_meta.<locals>.<lambda>  s    QCQQ r1   r}   )
r   rs   r   rE   rV   r   r   r  r   r   )r   r9  rs   rb  dimSizer   s     `   r/   kthvalue_metar    s     dhhjjd
;
;
;C $

QdiinnnAG	L	Q1<QQQQ  
 DSD!DJsQwyy$99::E 488::>>S!>>%  $..ek."J"JJJr1   c                    | | n|}t          j        |                                dk    d            |                                }| ,t          j        |                                 |k    d            |,t          j        |                                |k    d            t          j        |                                |k    d            t          j        |                                |k    d            t          j        |                                dk    d            t          j        |                                |d         |d	         z  d
z  k    d            d S )Nr   c                      dS N r5   r5   r1   r/   rT   z(checkLSTMBackwardSizes.<locals>.<lambda>  s    " r1   c                      dS r  r5   r5   r1   r/   rT   z(checkLSTMBackwardSizes.<locals>.<lambda>       r1   c                      dS r  r5   r5   r1   r/   rT   z(checkLSTMBackwardSizes.<locals>.<lambda>   r  r1   c                      dS r  r5   r5   r1   r/   rT   z(checkLSTMBackwardSizes.<locals>.<lambda>!       r1   c                      dS r  r5   r5   r1   r/   rT   z(checkLSTMBackwardSizes.<locals>.<lambda>"  r  r1   c                      dS r  r5   r5   r1   r/   rT   z(checkLSTMBackwardSizes.<locals>.<lambda>#  s    r r1   r   r   r  c                      dS r  r5   r5   r1   r/   rT   z(checkLSTMBackwardSizes.<locals>.<lambda>$  s    R r1   )rE   rV   rs   r   r   )grad_hygrad_cyr_  rb  r`  defined_gradexp_sizes          r/   checkLSTMBackwardSizesr    s2   %177wL	L!!##q(**555  ""HW\\^^x/<<<W\\^^x/<<<	Lh&

333	Lh&

333	LA%zz222	L""hqkHQK&?!&CCZZPPPPPr1   c                     | |dS t          | ||||           t          j        |t                    }t          j        |t                    }|r|                    dd          nd }|||fS )NNNNr   r   F)rb  )r  rE   r   legacy_contiguous_memory_formatrX  )	r  r  r_  rb  r`  has_bias
grad_gatesgrad_cxr  s	            r/   #_thnn_fused_lstm_cell_backward_implr  (  s    7?7GRY???!!@  J r1PQQQG4<F
q%000$Iw	))r1   c                 `   d }d }d }|d         r'|                     |                                           }|d         s|d         re|                     |                    d          |                     d          f          }|                     |                    d                    }|||fS )Nr   r   r   r   rx  )rz  ry  r{  r~  r  grad_weightr  s          r/   linear_backwardr  6  s    JKI1~ ;!++FKKMM::
1~ BQ B",,l.?.?.C.CV[[QS__-UVV **<+<+<R+@+@AA	Y//r1   c                     t           j                  dk    r j        d         ||z  z  dk    sJ d j         d|             d  fd} j        d         ||z  z  } j        d         |z  } j        d	         |z  }g  j        d d         |||R }                     |          }|                     |            
          }|S )Nr   r  r   z'Invalid input shape for pixel_shuffle: z with upscale_factor = c                 \    t           j                            |           t           j        k    S r,   ry  r}  s    r/   r  z,meta_pixel_shuffle.<locals>.is_channels_lastI  s!    "88==ATTTr1   c                  $               r+t                    dk    rt          j        S t          j        S                     t          j                  rt          j        S                     t          j                  rt          j        S d S r  )rL  rE   r   r|  r6  r  )r  r   s   r/   r  z.meta_pixel_shuffle.<locals>.pick_memory_formatL  s    D!! 	)4  F**..**e.EFF 	)**e.CDD 	)((	) 	)r1   r  r   r   )r   r   r   r   )	r   upscale_factorr  r   HrWrr   r   r  s	   `       @r/   meta_pixel_shuffler  C  s    	DJ!
2.>2Q RVW W W WdddTbdd !X WWU U U	) 	) 	) 	) 	) 	) 	
2>N:;A	B.	(B	B.	(B-$*SbS/-1-b-"--I
..
#
#C
&&1133&
4
4CJr1   c                 L   |                      | j                  }|                     |j                  }|                     |j                  }|                     |j                  }|                     |j                  }|                     |j                  }|||||||fS r,   r  )r   weight0weight1weight2weight3r  cx_tmpr   hy_cy_grad_output_r_optgrad_hy_r_optgrad_cy_r_optr   r3  rh  rj  r  rm  rn  ro  rk  r`  diff_xdiff_hxdiff_cxdiff_w1diff_w2diff_bs                                r/   mkldnn_rnn_layer_backwardr  a  s    4 __U[))FmmCI&&Gv|,,G..G..Gw}--F7GVVWgEEr1   )	out_int32r   c                    t          j        | |rt           j        nt           j                                                  S r]  )rE   r   r  r   r   )r   
boundariesr  r   s       r/   meta_bucketizer    s9     9=EKK%+  jllr1   d   c                 B    dt                     dk    r*t          j                                          fd           t          j        t	          t
                    fd           t          j        dk    fd           t          j        t	          t                    fd           t          j        t	          t                    fd           t          j        k    d	            t          j         j         j	        
          S )Nzhistc()r  c                      d j          dS )Nz%"histogram_cpu" not implemented for ''r}   r;  s   r/   rT   zmeta_histc.<locals>.<lambda>  s    LekLLL r1   c                  ,     dt                      S )Nz#: argument 'bins' must be int, not r  binsr  s   r/   rT   zmeta_histc.<locals>.<lambda>  s    7KKtDzzKK r1   r   c                       d  S )Nz: bins must be > 0, but got r5   r  s   r/   rT   zmeta_histc.<locals>.<lambda>  s    g#Q#Q4#Q#Q r1   c                  ,      dt                     S )Nz%: argument 'min' must be Number, not r  )r  r  s   r/   rT   zmeta_histc.<locals>.<lambda>      7LLcLL r1   c                  ,      dt                     S )Nz%: argument 'max' must be Number, not r  )r  r  s   r/   rT   zmeta_histc.<locals>.<lambda>  r  r1   c                      dS )Nz&{fn_name}: max must be larger than minr5   r5   r1   r/   rT   zmeta_histc.<locals>.<lambda>  s    %M r1   r   )
rL  rE   rV   r   r_   r   r   rx   rp   rM   )r   r  r  r  r  s   ````@r/   
meta_histcr    s@    G5U""##%%LLLL	
 	
 	
 
L4!!KKKKK   
LQQQQQRRR	L3LLLLL   
L3LLLLL   
LMMNNN;tELDDDDr1   c                     t                                           |d          }t          j                                         dk    p2t          d                                  dd          D                        fd                                |                              t          j	                             S )Nr   r-  r   c              3   "   K   | ]
}|d k    V  dS r$  r5   )r;   r   s     r/   rb   z,meta_upsample_bimode2d_aa.<locals>.<genexpr>  s&      !H!Ht$(!H!H!H!H!H!Hr1   r   c                  2    d                                   S r5  r   r;  s   r/   rT   z+meta_upsample_bimode2d_aa.<locals>.<lambda>  r,  r1   r   )
r)  r   rE   rV   r   rk  r   r   r?   r{  )r   r"  r  r6  r7  r0  s   `     r/   meta_upsample_bimode2d_aar    s     -

kA   
LHc!H!Huzz||ABB7G!H!H!HHH____   ??+,,//1%88 0   r1   c                 6   t          j        |                                dk    d            t          j        |                                dk    d            t          j        |j        j        d            t          j        |j        j        d            d S )Nr   c                      dS )Nz%found_inf must be a 1-element tensor.r5   r5   r1   r/   rT   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>      (O r1   c                      dS )Nz%inv_scale must be a 1-element tensor.r5   r5   r1   r/   rT   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r  r1   c                      dS )Nz!found_inf must be a float tensor.r5   r5   r1   r/   rT   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>      3 r1   c                      dS )Nz!inv_scale must be a float tensor.r5   r5   r1   r/   rT   z<_amp_foreach_non_finite_check_and_unscale_.<locals>.<lambda>  r  r1   )rE   rV   r   rM   r   )r   r  	inv_scales      r/   *_amp_foreach_non_finite_check_and_unscale_r    s    	LQ O O   
LQ O O   
L)33   
L)33    r1   c                 n    t          |                                           }|                     |          S r,   )r   r   r   )r   nanposinfneginfrH  s        r/   
nan_to_numr    s+     tyy{{##K>>+&&&r1   c                     | j         t          j        t          j        t          j        t          j        hvsJ d| j          d            | j        }t          ||          }t          ||          }||k    r| S t          | 	                                          }t          | 
                                          }||         ||         c||<   ||<   ||         ||         c||<   ||<   |                     ||           | S )Nz>torch.transpose_: in-place transposition is not supported for z layout)ro   rE   r  
sparse_cscr  
sparse_bscr   r   r   r   r   r  )r   dim0r.  ndimsr   r   s         r/   r{  r{    s	    		
	
 	
 	
 ]\\\	
 	
 	
 IE$&&D$&&Dt||		D$++--  F!'vd|F4L&,!$ZdDJT
T6"""Kr1   c                 (   | j         }| j        rF|                                 }|                                 }|dk    r|dk    sJ d| d| d            n&|                                 dk    sJ d| d            t          | d|dk     rdnd          S )	Nr   r   zEt_ expects a tensor with <= 2 sparse and 0 dense dimensions, but got z sparse and z dense dimensionsz6t_ expects a tensor with <= 2 dimensions, but self is r  r   )r   r-  r/  r0  rs   r{  )r   r  r/  r0  s       r/   t_r     s    IE~ 	M__&&
NN$$	!OO	Q HS]  H  Hkt  H  H  H !/. HHJJ!OOOLELLL OO dAEAIIqq1555r1   )r  r   sidesorterc                $    t          j        t           j                  dk    p j        d d         j        d d         k     fd           t          j        d u p j        j        k     fd           t          j        |dk    p| d           |rt           j        nt           j        }t          t           j                  r(t          j        |          	                                S t          j
        d| j        	          S )
Nr   r   c                  \    dt          j                   dt           j                   S )Nztorch.searchsorted(): boundaries tensor should be 1 dimension or the first N-1 dimensions of boundaries tensor and input value tensor must match, but we got boundaries tensor z and input value tensor r   r   )r   sorted_sequences   r/   rT   z#meta_searchsorted.<locals>.<lambda>  s>    5378M3N3N5 5 #'tz"2"25 5 r1   c                  d    dt           j                   dt          j                  ng  S )Nz[torch.searchsorted(): boundary and sorter must have the same size, but got boundary tensor z and got sorter tensor r  )r  r  s   r/   rT   z#meta_searchsorted.<locals>.<lambda>"  sK    B#'(=#>#>B B%+%7tFL!!!RB B r1   r   zetorch.searchsorted(): side and right can't be set to opposites, got side of left while right was Truer}   r5   r  )rE   rV   r   r   r  r   r_   r   r   r   rx   rp   )r  r   r  r   r  r  rM   s   ``   ` r/   meta_searchsortedr    s=    
LO!""a' 	9 "%CRC8	
 	
 	
 	
 	
	 	 	 
L$?//6<?	
 	
 	
 	
 	
   
L#e)	$   %5EKK%+E$%% KE222==???{2U?3IJJJJr1   c                 ~     t          j         t           j        t           j        t           j        fv fd           d S )Nc                      d  S )Nz/Unsupported input type encountered for isin(): r5   r}   s   r/   rT   z3_check_for_unsupported_isin_dtype.<locals>.<lambda>:  s    I%II r1   )rE   rV   rD  
complex128	complex64r}   s   `r/   !_check_for_unsupported_isin_dtyper  7  sC    	Lej%"2EODDIIII    r1   c                 |    |	r#t                               | ||||||||
|
  
        S t          | ||||||||
|
  
        S r,   )r&   _embedding_bag_sparse_backward!meta_embedding_bag_dense_backward)r  r   rK  r  r  maximum_indicesnum_weightsr  r3  r  r  r  s               r/   meta_embedding_bag_backwardr  >  st      
22
 
 	
 1
 
 	
r1   c
                 P    t          j         j        t           j        t           j        t           j        t           j        fv  fd           t          d          \  }
}}||k    rt          j        |d u                                | 	                    d          f          }|S )Nc                      d j          S )Nz$Unsupported input type encountered: r}   )r  s   r/   rT   z3meta_embedding_bag_dense_backward.<locals>.<lambda>x  s    CtzCC r1   r   r   )
rE   rV   rM   r3  r4  r2  float64r   r   r   )r  r   r  r  r  r  r  r3  r  r  r  r  r  index_grad_weights   `             r/   r  r  i  s     
L
u}enemU]SSCCCC   %*!HH!Hix_D0111TYYq\\'BCCr1   c                 0   t          d          \  }}}	|                     d          }
t          j        ||k    d           t          j        |                                 dk               t          j        |                                dk               |                    d          }t          j        |                                dk               t          j        |                    d          |
k               |                     |f          }|S )Nr   r   zHembedding_bag_backward: per_sample_weights only supported for mode='sum'r   r   )r   r   rE   rV   rs   r   )r  r  r   rK  r  r3  r  r  r  r  embedding_featuresr  r   s                r/   .meta_embedding_bag_per_sample_weights_backwardr    s     %*!HH!Hi1	LR   
Lq!!!	L!#$$$,,q//K	L"###	LQ#55666^^[N++FMr1   )assume_uniqueinvertc                   t          j        t          | t                    pt          |t                    d            t          | t                    st          j        | |j                  } t          |t                    st          j        || j                  }t          | j                   t          |j                   t          j        | t           j	                  S )Nc                      dS )Nz<At least one of elements and test_elements must be a Tensor.r5   r5   r1   r/   rT   zmeta_isin.<locals>.<lambda>  r  r1   r~  r}   )
rE   rV   r_   r   r	  rp   r  rM   r   rD  )elementstest_elementsr  r  s       r/   	meta_isinr!    s     
L8V$$I
=&(I(INN   h'' G<1EFFFmV,, L]8?KKK%hn555%m&9:::HEJ7777r1   r   c                     t          j        | dk    d            t          |t          j                  \  }}t          j        ||          S )Nr   c                      dS )Nz,polygamma(n, x) does not support negative n.r5   r5   r1   r/   rT   z meta_polygamma.<locals>.<lambda>  s    !O r1   rj  r}   )rE   rV   r   r   rk  r   )r   r   rB   r=   s       r/   meta_polygammar$    sW     
LaOOPPP(;H  OA| D5555r1   c                      t          d          )Nz.Tensor.item() cannot be called on meta tensors)r   r   s    r/   meta_local_scalar_denser&    s    
G
H
HHr1   max_lengthspadding_valuec                     t          |          dk    sJ t          |          dk    sJ |d         j        d         dz
  }|d         }||g| j        dd          R }|                     |          S r  )r   r   r   )r   rK  r'  r(  r  rQ  r;  s          r/   $meta__jagged_to_padded_dense_forwardr*    s     w<<1{q    
aAAAq,6<+,,LL)))r1   c                 ^    t          |           t                      d                         }|S )Nc                 8    t          | t          j                  S rI  rC   r   rk  r  s    r/   _fz)_create_unary_float_meta_func.<locals>._f  s$      =J
 
 
 	
r1   r6   r"   funcr.  s     r/   _create_unary_float_meta_funcr2    9    4]]
 
 ] 

 Ir1   c                 ^    t          |           t                      d                         }|S )Nc                 :    t          | |t          j                  S rI  r-  )r<   r  s     r/   r.  z*_create_binary_float_meta_func.<locals>._f  s&      q!@!M
 
 
 	
r1   r/  r0  s     r/   _create_binary_float_meta_funcr6    r3  r1   c                      t                      fd            } j         d}||_         t          t          t          |                    |          }|S )Nc                 T     | g|R i |}t          | j        |j                   | S r,   rx  )r   rA   rL  r   r.   s       r/   _fnz#_register_inplace_meta.<locals>._fn  s;    b''''''
CI666r1   rB   )r   ri   r6   getattrr&   )r.   r9  inplace_names   `  r/   _register_inplace_metar<    sk    
2YY    Y
 k$$$LCL
4-l33
4
4S
9
9CJr1   c                 ,    t          j         j        j        k     fd            g}t          t                    r<t          j         j        j        k     fd           |                               t          |dt          j        iS )Nc                  (    dj          d j          S )Nr/  z for `end`, but got dtype r}   )rj   rk   s   r/   rT   zlerp.<locals>.<lambda>  s    T%+TTTT r1   c                  (    d j          dj          S )Nr/  z for `weight`, but got dtype r}   )rk   r  s   r/   rT   zlerp.<locals>.<lambda>  s    ^ek^^PVP\^^ r1   r7   )	rE   rV   rM   r_   r   r   rC   r   r@   )rk   rj   r  rA   s   ``` r/   lerpr@  
  s     
Lsy TTTTT   3<D&*%% K6<'^^^^^	
 	
 	
 	F	=E  r1   )ry  c                <    t          | ||t          j                  S rI  rK  r   tensor1tensor2ry  s       r/   addcmulrE    s(     w0O0W   r1   c                    t          j        t          j        |j                  ot          j        |j                   d            t          | ||t          j                  S )Nc                      dS )N)zFInteger division with addcdiv is no longer supported, and in a future zErelease addcdiv will perform a true division of tensor1 and tensor2. z4The historic addcdiv behavior can be implemented as zA(input + value * torch.trunc(tensor1 / tensor2)).to(input.dtype) zfor integer inputs and as z6(input + value * tensor1 / tensor2) for float inputs. z?The future addcdiv behavior is just the latter implementation: z4(input + value * tensor1 / tensor2), for all dtypes.r5   r5   r1   r/   rT   zaddcdiv.<locals>.<lambda>-  s	     	
 r1   rJ  )rE   rV   r?   r  rM   rC   r   r@   rB  s       r/   addcdivrH  %  sr     
L"7=11 6&w}55	
		
 		
    w0O0W   r1   c                     i } dD ]#}t           |         }|D ]}|| vr||         | |<   $|                                 D ]\  }}t          |t          j        j                  r&t          |t                    sJ  |                    t          j        j	        j
                  |           t          j                            |                                d          r"|t           d         v rt          | d          |j        r|                                dv rd|                                v rt                              ||           d|                                v rt"                              ||           Hd|                                v rt$                              ||           {d	|                                v rt&                              ||           t(                              ||           d S )
N)rn   post_autogradpre_autogradCompositeImplicitAutogradrn   z is a CompositeImplicitAutograd op, we shouldn't register meta function for it. Instead, we should let the decomposition run and write meta kernels for the base operators.>   aten::cloneaten::copy_aten::rot90aten::_to_copyaten::empty_stridedaten::constant_pad_ndaten::as_strided_scatterzmkldnn::zmkl::zonednn::zquantized::)r   itemsr_   rE   _opsHigherOrderOperatorr   py_impl_CDispatchKeyr(   %_dispatch_has_kernel_for_dispatch_keyr  r   is_view2_meta_lib_dont_use_me_use_register_meta_for_mkldnnimpl/_meta_lib_dont_use_me_use_register_meta_for_mkl2_meta_lib_dont_use_me_use_register_meta_for_onednn5_meta_lib_dont_use_me_use_register_meta_for_quantized'_meta_lib_dont_use_me_use_register_meta)activate_meta_tablerh   registryopoop_overloadr.   s         r/   activate_metarf  I  sb    : 9 9-d3 	9 	9C---+3C=#C(	9 /4466 6N 6NR
 k5:#ABB 	+z222226EH0566r:::899 ;
 
 +	N 8@@@"" ; ; ;   A   	N 	 	 [--////BGGUWXXXXK,,....?DD[RTUUUU{//1111BGGUWXXXX+"2"2"4"444EJJ    8<<["MMMMm6N 6Nr1   r   r  r,   )NNNFr   r   F)NN)Tr  )r  )r  T)FF)TT)r5  )FTN)TFF)TF)r   )r7  N)r@  rb  )r5   r   r  F)r5   r   FTN)Fr   FNFr   )NF)r   F)rh  ri  FN)NNNNN)r   NNr   )NNF)r  FFN)r  FNN)r  FN)FN)FNNNN)NNNF)Nr   FNN)NNNN)r   TT)NNr   N)r  r   r   )r   )r  (]  r  enumr   	functoolsr   typingr   r   r   r   r	   rE   torch._prims_commonrz  r?   r
   r   r   torch._decompr   r   r   r   
torch._opsr   torch._primsr   r   r   r   r   r   r   r   r   r   r   r   torch._prims_common.wrappersr   r   r    r!   r"   rm  r#   r$   torch.utilsr%   r2   opsr&   libraryLibraryra  r6   rC   rO   rX   linspacelogspacer  rz   taker  r   r   r   r   cummaxcumminr   r   r   _fft_c2cr   _fft_r2cr   randpermgenerator_outr   r   r   randintr   r   low_outr   randr   _fft_c2rr   ro  r  r  
unsqueeze_r  _sparse_semi_structured_linearrb  rM   r  _sparse_semi_structured_mmr#  _sparse_semi_structured_addmmr'  _cslt_sparse_mmrD  r  r<  index_reducerD  index_reduce_rF  index_selectrI  segment_reducerV  r  	unary_outrZ  rs   rc  r  rf  rh  rp  rl  rq  _assert_asyncrt  msgrx  _printr{  _make_dep_tokenr  r  _functional_sym_constrain_ranger  r  (_functional_sym_constrain_range_for_sizer  _functional_assert_asyncr  r   r  r   r  r  r  r  _linalg_eighr  r  _linalg_eigvalslinalg_eigvalsr  
linalg_eigr  r  r  r  r  r  r  r  linalg_inv_exr  linalg_ldl_factor_exr
  linalg_ldl_solver  	linalg_lur  linalg_lu_factor_exr#  linalg_lu_solver,  	lu_unpackr2  r;  	linalg_qrrB  rF  rC  _linalg_svdrR  r  r  rb  rv  linalg_solve_triangularr~  r  r  _linalg_detr  r  r  r  reflection_pad1dr  replication_pad1dr  r  reflection_pad1d_backwardr  replication_pad1d_backwardr  r  reflection_pad2dr  replication_pad2dr  reflection_pad2d_backwardr  replication_pad2d_backwardr  r  reflection_pad3dr
  replication_pad3dr  reflection_pad3d_backwardreplication_pad3d_backwardr  _pdist_forwardrI   r  _pdist_backwardr  baddbmmr2  	bernoullir6  
bernoulli_r:  r  r=  poissonr@  _fused_moving_avg_obs_fq_helperrR  mmr]  r`  rL  rw  r  miopen_batch_normr  convolutionr  rX  _has_mkldnnr\  r  _convolution_pointwiser  _linear_pointwiser  has_mklr^  r  _mkl_linearr  r_  r  qconv2d_pointwiser  qlinear_pointwiser	  r  linear_dynamic_fp16linear_relu_dynamic_fp16r  r`  r  
max_pool2dr  r  
avg_pool2dr  r  avg_pool2d_backwardr  
avg_pool3dr  avg_pool3d_backwardr  _adaptive_avg_pool2dr#  _adaptive_avg_pool3dr&  _adaptive_avg_pool2d_backwardr0  _adaptive_avg_pool3d_backwardr5  r3  adaptive_max_pool2drG  rK  rM  adaptive_max_pool3drT  rV  rW  repeat_interleaverZ  r`   r^  ra  r   _unsafe_indexrv  convolution_backwardr  addbmmr  _fused_adam__fused_adamw_r  _fused_adamr  _int_mmr  _convert_weight_to_int4packr  #_convert_weight_to_int4pack_for_cpur  _weight_int4pack_mmr  _weight_int4pack_mm_for_cpur  _weight_int8pack_mmr  _cdist_forwardr  _cdist_backwardr  _embedding_bagr#  _embedding_bag_forward_onlyr%  r(  nansumr*  median	nanmedianr,  
dim_valuesr3  r   r/  logical_not_r1  repeatr8  zero_r:  mul_Scalardiv_logical_and_logical_or_logical_xor_r=  add_sub_rG  rounddecimalsrM  rS  
__rshift__rW  
__lshift__r[  zeror]  r2  r`  fillrb  relu_re  	_add_relurg  rrelu_with_noisern  rrelu_with_noise_functionalrp  rrelu_with_noise_rr  	index_put_unsafe_index_putrv  masked_fill_rz  _masked_scaler}  masked_scatter_r  masked_scatterr  masked_scatter_backwardr  
index_put_r  aliasr  r  bmmr  r  r  r  r  r  r  r  r   max_pool2d_with_indices_backwardr  max_pool2d_with_indicesr  fractional_max_pool2dr  max_pool3d_with_indicesr   max_pool3d_with_indices_backwardr   r
  r  r  grid_sampler_2d_backwardr  r$  r&  r)  r  rl  r;  select_scatterr?  slice_scatterrC  r   rH  rP  gatherrV  rg  rn  rp  rz  r}  scatter_addr  scatter_add_r  r|  r  ry  r?  value_reducer  scatter_r  #_scaled_dot_product_flash_attentionr  #_scaled_dot_product_cudnn_attentionr  ,_scaled_dot_product_flash_attention_backwardr  +_scaled_dot_product_flash_attention_for_cpur  4_scaled_dot_product_flash_attention_for_cpu_backwardr  '_scaled_dot_product_efficient_attentionr  0_scaled_dot_product_efficient_attention_backwardr  ,_scaled_dot_product_cudnn_attention_backwardr  _flash_attention_forwardr  _flash_attention_backwardr  _efficient_attention_forwardr  _efficient_attention_backwardSymIntr  
_scaled_mmr  scatter_reducetwotwo_outr  scatter_reduce_r  multinomialr  r  r)  r1  _upsample_nearest_exact1dr9  _upsample_nearest_exact2dr=  "_upsample_nearest_exact2d_backwardrA  _upsample_nearest_exact3dr   rC  values_stablerF  r]  _thnn_fused_lstm_cellrc  rz  r  r  r  r  argminr  r  topkr  _segment_reduce_backwardr  kthvaluer  r   r  r  r  r  pixel_shuffler  r  	bucketize
Tensor_outr  histcr  _upsample_bilinear2d_aa_upsample_bicubic2d_aar  r  r  r{  r   searchsortedr  r  _embedding_bag_backwardr  _embedding_bag_dense_backwardr  *_embedding_bag_per_sample_weights_backwardr  isinr!  	polygammar$  _local_scalar_denser&  _jagged_to_padded_dense_forwardr*  r2  r6  special_airy_aispecial_bessel_y0special_bessel_y1special_modified_bessel_i0special_modified_bessel_i1special_modified_bessel_k0special_modified_bessel_k1!special_scaled_modified_bessel_k0!special_scaled_modified_bessel_k1special_chebyshev_polynomial_tspecial_chebyshev_polynomial_uspecial_chebyshev_polynomial_vspecial_chebyshev_polynomial_w&special_shifted_chebyshev_polynomial_t&special_shifted_chebyshev_polynomial_u&special_shifted_chebyshev_polynomial_v&special_shifted_chebyshev_polynomial_wspecial_hermite_polynomial_hspecial_hermite_polynomial_hespecial_laguerre_polynomial_lspecial_legendre_polynomial_pr<  r@  rE  rH  lerp_addcmul_addcdiv_torch._refs.nn.functionaltorch._refs.specialrf  r5   r1   r/   <module>rQ	     sQ                9 9 9 9 9 9 9 9 9 9 9 9 9 9  # # # # # # + + + + + + + + + +            " ! ! ! ! ! U U U U U U U U                                     < ; ; ; ; ; ; ; ) ) ) ) ) ) y~*/-*?*?PV*W*W '
 
 
3   (* * *   t}-..
 

=5 5 5  /.5p 	!49=122' '  32' !)4+<+@ABB%' % % % %  CB%$ t%&&I I  '&I 	[$+/4;+>P  Xy!!  "!  !)4+<+@ABB/ /  CB/1H 1H 1Hl %t}'89::   ;:" %t}'89::   ;: t}*++"& 3 3 3 3 ,+3 t}$%% *
 
 
 
 &%
 $dl&6788
 *     98  $,"6788 *     98 	!49=122%)$tPT      32 %t}'89::K K  ;:K tz!""   #"0( ( ( t&''  (' t233
 "%)'+   6
	
 c] $   43B t.//
 (,	 
  $	   0/0 t122 	
'+  
  	 $   32D t#$$ ""'+"$/< /<,/<\/< 6
/< F	/<
 $/< /< /< /< /< /< /< %$/<d t ()) 	I 	I 	I
	I		I 	I L		I
 	I 	I 	I 	I 	I *)	I t!)** 	 	 	
			 	 L		
 	 	 	 	 	 +*	 t ())' ' *) ' t"*++
 !% $ $ W  W  W
 W W f	 W
 f W f W  W  W  W  W  W ,+ WF  $("4566   76 tx|     $("4566   76 tx|    tz!""6 6 #"6 tz~( ( (
 t!)**  +* t!%&&  '& t{"##  $# t#+,, ) ) ) ) -,) t'/00, , , 10, t3;<<   =<
 t08996 6 6 :96& t<DEE  FE
 t,011  21
sF sC s s s s   F  #        N (,
 


 !%
 
 
 
$ V S C    
 
F 
$ 
 
 
 
 
"  	   	   C     !)4+<+HIJJ]N++  s T    ,+ KJ" $,d.A.EFGGB B6 B B B  HGB  !!]N++	6 	 	 	 ,+ "!	Q QF Q Q Q Q t*++) )F )4 )F ) ) )  ,+) t"##J J JF J4 JF J J J  $#J t}) )6 )$ )6 ) ) )  ) t$%%) )6 )$ )6 ) ) )  &%) t&.// &  T    0/" 	$,d.M.QR  .f .6 .f . . .  .d t!)** &     +* )143L3PQRRT8V$$ 	  
  	
 666!"   %$ SR& %-t/D/HIJJ ' ' ''' '
 ' ' ' '  KJ'T &(:;<<S#s/3   f  fff>T8U     =<4 (0$2J2NOPPT8V$$ 	    	
 666!"   %$ QPD $,d.B.FGHH 4 4 444 4
 4 4 4 4 4  IH4n t~S#s 	$ $$$ $ 	$
 666!"$ $ $  $P tTz!2    * &(:;<<S# f C ffn8M     =<4 $,d.B.GHIIV[$11'v '%(F"G ' ' ' 21 JI'$ t'((   	" """ " SM	" " " )("J.
.
. 49d3i . . . .".
.
. 3-. 66>	. . . .(f V      t$%%
 ##!8 8 888 	8
 8 V8 	8 V8 6
8 6666)*8 8 8 &%8v ,4d6R6VWXX     	
   
&	    YX2 t$%%S#4((( +( +(
+(+( +( 	+(
 +( 66>+( +( +( )( &%+(^ t'((
 
 )(
 tz
 W WW	W W 	W
 W W W W  Wt  >#; #; #;L t$%%= =  &%= t%&&> >  '&>( ( (< t-..\S S  /.S t.//\T T  0/T2E 2E 2Ej t$%%= =  &%= t%&&> >  '&> &.&1'/'2	  \& &  &@<G <G <G~ t$%%= =  &%= t%&&> >  '&> &.&1'/'2	  \$( $(  $(N t"##

 

f 

 

v 

 

 

  $#

 t#$$Pv PV P Pf PQW P P P  %$P $dl&6788/0 ' ' ' '  98'2 &(:;<<&* / / / /  =</
 t$%%   &% t~  / / / ! /
 $dl&6788" " "  98" t3;<< * * * =<*. tw	 	  	B B B
 
 
 
* 7;b b,bLb $s)S.!b 49c>"	b
 DIsN#b b b U49c>23b b b bJQ Q Q t%-.."$,"$L"$ 5<
 "$ 5<(	"$
 %,'"$ "$ !&"$ "$ "$ "$ /."$J t'((),)L) ,) I	)
 #Y) 3i) ) I) ) ) ) )()X 	8 X
9>9N9N&&: :6 ]59#:BCC  DC, ]59#5=>>S S ?>S
 x 	:?-:O:O66;
 ;
7 
uy}0	1	1	 	 
2	1	
 :?9N9N&&: :6 ]59#5=>>    ?> D ]59#5=>>]59#5<==  >= ?>, ]59#7?@@]59#<DEE	 	 FE A@	 =BM<Q<QVV= =9 ]59&122 
 
 
 32
<   t&'' I I I ('IZ(< (< (<X t'/00E E 10EP t QJ QJ QJ   QJh t'((\K( K(  )(K(\ t(011  21" t(011@ @ 21@ t19::F F ;:F, t122\P P  32P
	
6 	
S 	
 	
 	
 	
 t'((UI+ +  )(+\ t011\H H  21H$ t'((UI' '  )('T t011\( (  21(
 t%,--* * * .-* $dl&6788T T  98T #+T-@-DEFF9; @ @ @# @3 @ @ @  GF@ 
!4#5#<=>>`J `J ?>`JF )1233H H 43H: #T[_566./q ' ' ' '  76'0 !)4+=+EFGG  !
 
 
 
 HG
2  ()**  !! ! ! ! +*!H ~B B  B* 0122  32& 89::  ;: ()**@ @ +*@ 0122< < 32< ()**< < +*< t"*++ &  & ,+ &F t#$$G G  %$G* t"*++
 	
a5 a5 a5 ,+a5H t/7885 5 985
 
 
 
 #T[_566=$ = = = =  76= #T^%;<==) ) >=) !			 	 Xy!!   "!	 	 t ())  *) t{"##' ' $#' tz!""  #" 				! !
 
 
 
 					 * * * *Z 
"DJ$7899  :9
 
 
" &(>?@@  A@ &(>?@@  A@ ty !!& & "!& 
!4:#4566  76 	 $)"2344" " 54" tz!""  #" t~$%% F     &% %&''RV" " "  ('" 0122RV; ; ; 32; &'((KO   )( &(>(FGHH" " " IH" t '((  )(
 t!)**  +* t#$$	 	 %$	 t"##6 6  $#6 t+,,! ! -,! t&''   (' tz!""! ! #"!   B tx  5 5 ! 5    6  ; ; ;h #-Z ZZZ 	Z 		Z
 	Z 	Z 	Z 	Z 	Z 	Z 	Z Z Z Z Z  !Z" #Z$ %Z& 'Z( )Z* +Z, -Z Z Z Zz;4 ;4 ;4|383838 38 		38
 	38 	38 	38 	38 	38 	38 	38 	38 38 38 38  !38" #38$ %38& '38 38 38 38lI2 I2 I2X t4<==( ( >=(V t+344 # # # 54#L t)122Q Q 32Qh t+,,UI d d d  -,dN t455\b b  65bJ%
V %
6 %
 %
 %
 %
P    t    v 3    $ t,455# # 65#$ t#$$8 8  %$8" t,--\;''! ! (' .-!, 	!"##. . $#. t&'' ) ) ) (')X t{E E  E@ t"*++. . ,+. t!)**. . . +*.
	 	 	C 	d 	 	 	 	/ / /
    t{"##' ' ' $#'   6
 
 
 
   
-
 -
 -
 -
b4 4 4 4 t'((& & )(&
 t !!  "!
 !	  & & &  & "	     89::
 #!5 55	5 5 	5
 5 5 E?5 5 5 ;:5p 89:: #!' ''	' ' 	'
 ' ' ' ' E?' ' ' ;:'T 9 ( "" """ 
" 	"
 
" " " " " " " " " " E?" " " 
". 8  "&! 	  	
   E?   
< A  #'!'" '"'"'" 
'" 	'"
 
'" '" '" '" '" E?'" '" '" 
'"T <=>> !$) $)$)	$) $) 	$)
 $) $) E?$) $) $) ?>$)N = " !4- 4-4-4- 
4- 	4-
 4- 
4- 4- 4- 4- 4- $Z4- 4- E?4- 4- 4- 
4-n 9 * "!" """ 
" 	"
 
" " " " " " " " " " "  E?!" " " 
"0 %   "&*'+"&%)9 99	9 9 	9
 9 9 9 9 9 9 E?9 sm9  }9 9 6"9 9 9 
9x & ( "&*'+#, ,,, 
, 	,
 
, , , , , , , , , , E?,  sm!,"  }#, , , 
,4 )    %!(,!%!%-S -S-S	-S -S 6
	-S
 6"-S 6"-S 3--S 3--S -S -S -S E?-S f%-S v-S #-S -S -S 
-S` * * "$("'%47 474747 
47 	47
 6
47 6"47 6"47 ,47 ,47 47 47 47 47 47 47  E?!47" SM#47$  %47 47 47 
47n '()) $(+/'+ aY aY
,aY
,aY \aY \	aY
 5<
 aY 5<(aY $aY aY aY aY *)aYH #')<)DEFF& & &  GF&
 t#'((   )(
  ($*:*>?@@	 	 	 	 	  A@	  , , ,* 	$d&D&LM 
 
 
 
 	$d&D&LM    . (0/7  !% $ %U\ 123 sEL012 uo	
 uo   : 	$d&D&LM 
 
 
 
 					    &$ $ $N t)122
    32 t&''4/ 4/ ('4/n t$,--$% $% .-$%N

 

 


 
 
 #T[%89::4 4 4 ;:4 t!)**   +* ty !!	Q 	Q 	Q "!	Q t,--LP    .-  %t}';<==Xy!!K K K "! >=K #("9 Q Q Q t7?@@	* 	* A@	* t#+,,	0 	0 -,	0 t!)**  +*: t-566F F 76FD %t~'@ABB27u      CB 
|E E E  E0 	!)4+F+NO     ( t>FGG  HG$ ')<=>>' ' '  ?>'
 uy~())  *): uy~ !!6 6 "!6" t !!
 
	+K +K +K +K  "!+K\   t+,, '
 '
 '
 -,'
T t122    32. t>??    @?0 ty8=e 8 8 8 8  8  t~6c 6 6F 6 6 6  6 t'((I& I I I )(I t3;<<
 	* **&\* c* 	* * * =<*       d2 3 3 3  d4 5 5 5  d4 5 5 5  d= > > >  d= > > >  d= > > >  d= > > >  dD E E E  dD E E E  tB C C C  tB C C C  tB C C C  tB C C C  tJ K K K  tJ K K K  tJ K K K  tJ K K K  t@ A A A  tA B B B  tA B B B  tA B B B   ty   " t|./       t|./      , 	ty))!!$,//!!$,//
                BN BN BNJ r1   