o
    ߗiI                 !   @   sJ3  U d dl Z d dlZd dlZd dlZd dlZd dlmZ d dl mZmZ d dlm	Z	m
Z
 d dlmZmZmZmZmZmZmZmZ d dlZd dlZd dlmZ d dlmZ d dlm  mZ d dlm Z m!Z!m"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,m-Z-m.Z.m/Z/m0Z0 d dl1m2Z3 d dl4m5Z5 ej6j7Z7g Z8ee9 e:d< ej;j<j=Z=G dd deZ>	d}dedej?de@fddZAeeAej?jBddZCeeAej?jBdZDeeAej?jEdZFde"deGde"fddZHe$e=jIe0deDde"d e"fd!d"ZIe$e=jJe0deDde"d e"fd#d$ZJe$e=jKe0deDde"de"d%eLd&eLfd'd(ZKe$e=jMe0deDd)e"d*eLd+eLd,eLd-e@d.e"fd/d0ZMe$e=jNjOgd1d2 ZPe$e=jNj"gd3e"fd4d5ZQe$e=jRe0 eDd6e"de"fd7d8ZRe$e=jSe0deDd)e"d6e"fd9d:ZSe$e=jTe0dd)e"d6e"d;eLd<eLfd=d>ZTe$e=jUe0 eDd6e"de"fd?d@ZUe$e=jVe0 eDd)e"d6e"de"fdAdBZVe$e=jWe0dd)e"d6e"d&eLfdCdDZWe$e=jXe0deDd)e"d6e"dEeLdFe@fdGdHZXe$e=jYe0deDd~dJe"d6e"dKe9fdLdMZYe$e=jZeDd)e"dNe"fdOdPZZe$e=j[e0 eDd6e"de"fdQdRZ[e$e=j\e0deDd)e"d6e"de"fdSdTZ\e$e=j]d6e"dUe"de"fdVdWZ]e$e=j^d)e"d6e"dUe"dee"e"f fdXdYZ^e$e=j_e0 eDd)e"d6e"dZe"d[eLd\eLd]e@dFe@de"fd^d_Z_e$e=j`e0deDd)e"d6e"d`e"de"fdadbZ`dce"ddeGfdedfZadgejbfdhdiZce$e=jde0 eDe>jejffd6e"dje"ddeGde"fdkdlZde$e=jge0deDd)e"dNe"dje"ddeGfdmdnZge$e=jhddodpZie$e=jje0 eDe>jejfdqfd6e"dje"ddeGd%eLfdrdsZje$e=jkjleDd)e"d6e"dje"ddeGd%eLf
dtduZke$e=jkjmeDd)e"d6e"dje"ddeGd%eLde"fdvdwZne$e=jojleDd)e"d6e"dje"ddeGdxeLf
dydzZoe$e=jojpeDd)e"d6e"dje"ddeGdxeLde"fd{d|Zqd)e"d6e"dje"dUee" ddeGd}eGd~e"de"fddZre$e=jse0deDd)e"d6e"deGde"fddZse$e=jte0dd)e"d6e"dje"dUee" ddeGd}eGd~e"de"fddZte$e=jue0dd)e"d6e"dje"dUee" ddeGd}eGd~e"de"fddZue$e=jve0 eDde>jejffd6e"dje"dUee" ddeGde"f
ddZve$e=jwe0deDde>jejffd)e"d6e"dje"dUee" ddeGde"fddZwe$e=jxe0 eDe>jejffdNe"dje"ddeGde"fddZxe$e=jye0deDe>jejffd)e"d6e"dje"ddeGde"f
ddZye$e=jze0 ddNe"de"deLfddZze$e=j{e0 de"de"de"fddZ{e$e=j|e0 d)e"deeG deGdeGdeGdeGfddZ|e$e=j}j"	 			dd6e"deGdeeG deeG deGf
ddZ~de"deGdeeG deeG deeGeGf f
ddZe$e=je0 	 			ddNe"de"deGdeeG deeG deGfddZe$e=je0 d)e"deeG deGdeGfddZe$e=je0 d)e"deeG deGdeGdeGf
ddZd)e"de"dejbfddZe$e=je0deCd)e"de"deGdejbfddZe$e=je0 eCd)e"de"deGdejbfddZdd Ze$e=je0 dNe"deeG deeG deeG deeG de"fddZe$e=je0 eDdNe"deeG deeG deeG deeG deeG de"fddZe$e=je0 d)e"de"d+eLfddÄZe$e=je0 dJe"deeG deGdeGdeGde"fddȄZe$e=jjleD	dd)e"d6e"deeL de"fdd˄Ze$e=je=jjle7je=jjle7jdNe"deLdee@ fdd΄Ze$e=je0ddЃdNe"deLdee@ fdd҄Ze$e=je0 de"deGde@fddՄZe$e=je0dd֍de"deGde@fdd؄Ze$e=je0 			ddUe"de"deGde@de@de"fdd߄Ze$e=je0 d)e"de"deGdeGde@f
ddZdeeG fddZdee" deGdeGdee" fddZdee" fddZdee" deGfddZdee" deGdeGfddZe$e=jjle=jjpg	ddee" deGdeGdee" de"f
ddZe$e=jjle=jjpg	 	dd6e"deeG deGdeee"  deee"  f
ddZe$e=jj"ddNe"deGdeGdee"df fddZe$e=jjl	 ddNe"deeG deGdee"df fddZe$e=jj"dd6e"deGdeGdee"df fddZe=jje7j	 dd6e"de"deGdee"df fddZe$e=je0dd֍eDdd6e"d e"de"d%eGd*eGf
ddZe$e=je0 eD			dd6e"d e"de"d%eGd*eGde@fddZe$e=je0 eDdd6e"d e"de"d%eGd*eGf
dd	Ze$e=jjleDd)e"dNe"d
e"de"dee" deGdeGdeGdeGdee@ deee" ee" ee" f fddZe$e=jjpd)e"dNe"d
e"de"dee" deGdeGdeGdeGdee@ dej"dej"dej"deee" ee" ee" f fddZdee" dee" fddZe$e=jjlde"dNe"deeG d
e"de"dUee" dee" dee@ deee" ee" ee" f fddZe$e=jjpde"dNe"deeG d
e"de"dUee" dee" dee@ dej"dej"dej"deee" ee" ee" f fddZdNe"dUee" dee" d ee" d!ee" d]e@d"eLdeLd#e@dee"e"e"ee" ee" f fd$d%Ze$e=je0dd&d'dNe"dUee" dee" d ee" d!ee" d]e@d"eLdeLdee"e"e"f fd(d)Ze=jjle7je=jjle7jdNe"dUee" dee" d ee" d!ee" d]e@d"eLdeLdee"e"e"f fd*d+Ze=jjle7jddee" fd,d-Ze$e=jjldNe"dUee" dee" d e"d!e"d"eLdeLdee"e"e"f fd.d/Ze$e=jjldNe"dUee" dee" d e"d!e"d]e@d"eLdeLdee"e"e"f fd0d1Ze$e=jjdNe"dUee" dee" d]e@d"eLdeLdee"e"e"f fd2d3Ze$e=jjldNe"dUee" dee" d e"d!e"d]e@d"eLdeLdee"e"e"e"e"f fd4d5ZdNe"dUee" dee" d e"d!e"deLd]e@de"fd6d7Ze$e=jjldNe"dUee" dee" d e"d!e"d"eLdeLdee"e"e"e"f fd8d9Ze$e=jjldNe"dUee" dee" d e"d!e"d"eLdeLdee"e"e"e"e"e"f fd:d;Ze$e=jjldNe"dUee" dee" d e"d!e"d"eLdeLdee"e"e"e"f fd<d=Ze$e=je0ddЃeDdd>d?Ze$e=je0 ddddddd@dee"e(f dgeejb dAeej dBe@dCe@dDeej fdEdFZe$e=je=je=jge0 dGdH Ze=jjle7je$e=je0ddАddIdNe"dUe"dee" d ee" d!ee" d]e@dJeLdKeLfdLdMZdNdO Ze$e=jjlde"dNe"dUee" d ee" d!ee" d&ee" d'ee" de@deLdee@ dPe"dee"ee" ee" f fdQdRZe$e=jjlde"dNe"dUee" d ee" d!ee" d&ee" d'ee" de@deLdee@ dee"ee" ee" f fdSdTZe$e=jjpde"dNe"dUee" d ee" d!ee" d&ee" d'ee" de@deLdee@ dej"dej"dej"dee"ee" ee" f fdUdVZe$e=jƃe0ddАddNe"d)e"dUe"d ee" d!ee" d&ee" dWee" dKeLfdXdYZe$e=jǃe0ddАddNe"d)e"dUe"d ee" d!ee" d&ee" dWee" dKeLdZe"fd[d\Ze$e=jȃe0 eDdNe"deeGeGf fd]d^Zd6e*de*deeG deGfd_d`Ze$e=j˃e0 d6e*de*deeG fdadbZe$e=j̃e0 dNe*de*deeG deeG deeG f
dcddZe$e=j̓ddede*deGde*dfe*d*e(f
dgdhZe$e=j΃e0 ddede*deGde*dfe*d*e(f
didjZddede*deGde*dfe*dke@d*e(fdldmZe$e=jjle=jjle7jddodpZe$e=jуde*deGde*dfe*fdqdrZe$e=j҃e0 de*deGde*dfe*fdsdtZde*deGde*dfe*dke@f
dudvZe$e=jԃe0dd`eDd6e"dee"e"f fdwdxZe$e=jՃe0 	n	q	dde"dyee@eGeLf dzee@eGeLf d{eej fd|d}Ze$e=j׃dd~dZאdd Zؐdd Ze$e=jjۃe$e=jjۃe$e=jjۃe=jj۠e7je=jj۠e7je=jj۠e7je=jj۠e7je=jj۠e7je=jj۠e7jdNe"deeeG  deeeL  de"fddZe$e=jjۃe$e=jjۃe$e=jjۃe=jj۠e7je=jj۠e7je=jj۠e7je=jj۠e7je=jj۠e7je=jj۠e7jdNe"deeeG  deeeL  de"fddZd}ddZe$e=jjle=jjpge=jjle7je=jjle7je0ddd	ddNe"deeG deeL de"fddZe$e=jjle=jjpge=jjle7je=jjle7je0ddd	ddNe"deeG deeL de"fddZe$e=jjle=jjpge=jjle7je=jjle7je0ddd		ddNe"deeG deeL deeL de"f
ddZe$e=jjle=jjpge=jjle7je=jjle7je0ddd		ddNe"deeG deeL deeL de"f
ddZe$e=jjle=jjpge=jjle7je=jjle7je0ddd			ddNe"deeG deeL deeL deeL de"fddZe$e=jjle=jjpge=jjle7je=jjle7je0ddd			ddNe"deeG deeL deeL deeL de"fddZeD	d}dNe"deeG deeeL  de@de"f
ddZdd Zdd Zdd Zdd Z	d}ddZdd Zdd Zd}ddZd}ddZdd Ze$e=jje=jje7je=jje7jdd Ze$e=jje=jje7je=jje7jdd Ze$e=jje=jje7je=jje7jdd Ze$e=jje=jje7je=jje7jdd Zdd Zd}ddZd}ddZdd Ze$e=jje=jje7je=jje7jdÐdĄ Ze$e=jje=jje7je=jje7jdŐdƄ ZdǐdȄ Zdɐdʄ Z e$e=jje=jje7je=jje7jdːd̄ Ze$e=jje=jje7je=jje7jd͐d΄ Ze$e=jjۃe=jj۠e7je=jj۠e7jdϐdЄ Ze$e=jjۃe=jj۠e7je=jj۠e7jdѐd҄ Ze$e=jjۃe$e=j	jۃe=j
j۠e7je=j
j۠e7je=jj۠e7je=jj۠e7je=j	j۠e7je=j	j۠e7jdӐdԄ Ze$e=j
jle=j
jpge0 	ddNe"deeG de@deeL de"f
d֐dׄZ
e$e=jjle=jjpge=jjle7je0 		ddNe"deeG de@deeL deeL de"fdؐdلZe$e=j	jle=j	jpge0 			ddNe"deeG de@deeL deeL deeL de"fdڐdۄZ	ddܐd݄Zdސd߄ Zdee" dee" de"de"fddZde+de"fddZeDdNe"deeG de@deeeL  de"f
ddZe$e=jjlde"de"de@fddZe$e=je=jge0 dd Ze$e=jgdd Ze$e=jgd}ddZe$e=jgdd Ze$e=jgdd Zd6e"dje"dUee" ddeGd}eGdee"e"f fddZe$e=je0dd~d6e"dje"dUee" ddeGd}eGdee"e"f fddZe$e=je0dd~d6e"dje"dUee" ddeGd}eGdee"e"f fddZde"deLde"fddZde"deLde"fdd Zde"de+fddZde+de"de"fddZdee" de"fdd	Zd
eGde@dgejbdAejfddZ de"deGdeGde@fddZ!de"deGdeGdeGde@f
ddZ"de"deeG de@fddZ#de"deeG de@fddZ$e$e=j%e0 eDde"deeG de@fddZ%	 	 		dde"de"deGdeGde@de@de"fdd Z&e$e=j'e0 eD	 	 	dde"de"deGdeGde@de"fd!d"Z'e$e=j(e0 eDd#d$ Z(e$e=j)e0 dde>jejffd%d&Z)d'ej"d(ej"d)e@de@fd*d+Z*e=j+jle7je=j+jpe7je0dd,dd-d.d/Z+e$e=j,jle=j,jpge=j,jle7je0 eD		ddNe"deeGeGf de@d0eeL d1eeL de"fd2d3Z-e$e=j,jۃe=j,j۠e7je=j,j۠e7je0 eD	dde"deeeGeGf  de@deeeLeLf  de"f
d4d5Z.e$e=j/e$e=j0e$e=j1eDe0 de"deeGdf de"fd6d7Z2e$e=j3e$e=j4e$e=j5eDe0 de"deeGdf de"fd8d9Z6de"deeGdf d:eeGeGeGge"f de"fd;d<Z7e$e=j8e$e=j9e$e=j:e0dd=d> Z;e$e=j<e0d?d@dddAdBdCZ<e$e=j=e0 dddDdEdFZ=e$e=j>jle=j>jpge0 dej?dddGde(dgeejb dHej@dAeej dBe@f
dIdJZAe$e=j>jBgdej?dddGde(de(dgeejb dHej@dAeej dBe@fdKdLZCe$e&dMdN ZDe$e=jEe=jEjle7je0 ddde>jejffdNe"dje"de(dOe(dUee" ddeGde"fdPdQZEe$e=jFe=jFjle7je0ddRdNe"dje"ddeGdee"e"f fdSdTZFe$e=jGjl	n	ddddUdVe"dWe"d3e"dXeLdYe@dZee" d+eeL dee"e"f fd[d\ZHd]d^ ZIe$e=jJge0 eDdd_d`ZJe$e=jKe0 dadb ZKe$e=jLdcdd ZLe$e=jMjle=jMjpgddded6e"dgeejb dee" de"fdfdgZNe$e=jOjle=jOjPgdd6e"deeG fdhdiZQe$ej<j=jRddjdkZRe$e=jSe0 dddldmdnZSe$e=jTjlddod6ej"d{eej dej"fdpdqZTe$e=jTjUddod{eej fdrdsZVddtdudvZWdddldwdxZXe$e=jYe0 dydz ZYe$e=jZdd{d|ZZeIe=j[e=j\ eIe=j]e=j eIe=j^e=j eIe=j_e=jJ eIe=j`e=jN eIe=jae=jb eIe=jce=jU eIe=jde=je eIe=jfe=jR eIe=jge=jh eIe=jie=jj eIe=jke=jl eIe=jme=jn eIe=joe=jp eIe=jqe=jr eIe=jse=jt eIe=jue=jv eIe=jwe=jx eIe=jye=jz eIe=j{e=j| eIe=j}e=j~ eIe=je=j eIe=je=j eIe=je=j eIe=je=j[ dS (      N)Enum)partialreduce)chainproduct)AnyCallablecastIterableListOptionalTupleUnion)	sym_floatsym_intTensorregister_decomposition)	out_dtype)IntLike
NumberTypesuggest_memory_format
TensorLikeTensorSequenceType)_maybe_convert_to_dtype_maybe_resize_out_safe_copy_outout_wrapper)_pytree)tree_map__all__c                   @   s   e Zd ZdZdZdZdS )	Reductionr         N)__name__
__module____qualname__NONEMEANSUM r*   r*   Z/home/ubuntu/transcripts/venv/lib/python3.10/site-packages/torch/_decomp/decompositions.pyr!   /   s    r!   Fftype_promotioncompute_dtype_onlyc                    s   t  fdd}|S )Nc                     sr   dd t j| i |D }tj|di\  fdd}fdd}t|| i t||}r4|S t||S )Nc                 S   s   g | ]	}t |tr|qS r*   )
isinstancer   .0xr*   r*   r+   
<listcomp>?   s
    
z-type_casts.<locals>.inner.<locals>.<listcomp>type_promotion_kindc                       t | tr
|  S | S Nr/   r   tor2   computation_dtyper*   r+   increase_precG      

z0type_casts.<locals>.inner.<locals>.increase_precc                    r5   r6   r7   r9   )result_dtyper*   r+   decrease_precM   r=   z0type_casts.<locals>.inner.<locals>.decrease_prec)pytreearg_tree_leavesutilselementwise_dtypesr   )argskwargs	flat_argsr<   r?   rr.   r,   r-   )r;   r>   r+   inner=   s   

ztype_casts.<locals>.inner)	functoolswraps)r,   r-   r.   rI   r*   rH   r+   
type_casts8   s   rL   T)r-   r.   )r-   r2   dimreturnc                 C   s$   t ||   D ]}| d} q| S N)rangerM   	unsqueeze)r2   rM   _r*   r*   r+   _unsqueeze_to_dimj   s   rT   
grad_inputout_gradyc                 C   s   | d||     S Nr"   conj_physicalrV   rW   r*   r*   r+   tanh_backwardp      r\   c                 C   s   | |d|     S rX   rY   r[   r*   r*   r+   sigmoid_backwardw   r]   r^   beta	thresholdc                 C   s.   ||   }t|| |k| | | |d  S N      ?)exptorchwhere)rV   r2   r_   r`   zr*   r*   r+   softplus_backward~   s   "rg   grad_outputalphascaleinput_scale	is_resultself_or_resultc           	      C   sb   || }|}|}|rt |dk| | ||  | | S t |dk| | | t ||  | | S Nr   )rd   re   rc   )	rh   ri   rj   rk   rl   rm   negcoefposcoef
negiptcoefr*   r*   r+   elu_backward   s   rr   c                 C      t | |S r6   )rd   	full_likeselfvaluer*   r*   r+   fill_scalar      rx   rw   c                    s(   t   dk fdd t|  S )Nr   c                      s   d    dS )Nz@fill only supports 0-dimension value tensor but got tensor with z dimensionsrM   r*   rw   r*   r+   <lambda>       zfill_tensor.<locals>.<lambda>)rd   _checkrM   atencopyru   r*   r{   r+   fill_tensor   s
   

r   rv   c                 C   s    t jt j| d ddddd S N   r   min   maxrd   clamprv   r*   r*   r+   hardsigmoid   s    r   c                 C   s   t |dk|dk @ | d dS )Ng      g      @gUUUUUU?        rd   re   rh   rv   r*   r*   r+   hardsigmoid_backward   s
   r   min_valmax_valc                 C   s   t ||k||kB d| S )Nr   r   )rh   rv   r   r   r*   r*   r+   hardtanh_backward   s   r   c                 C   s$   | t jt j| d dddd d S r   r   r   r*   r*   r+   	hardswish   s   $r   c              
   C   s,   t |dk dt |dk| |d d  | S )Nr   r         ?r   r   r*   r*   r+   hardswish_backward   s
   r   c                 C   s   t ||kd| S rn   r   )rh   rv   r`   r*   r*   r+   threshold_backward      r   negative_slopeself_is_resultc                 C   s   t |dk| | | S rn   r   )rh   rv   r   r   r*   r*   r+   leaky_relu_backward   s   r   nonegradapproximatec                 C   s   d}d}d}|dkrO|| d }d}|| }|| }	||||	   }
t |
}d| }d| }d| }d||  }|dd| |   }|| | }| ||  S |}|| d }ddt ||   }|t || d	  }| |||   S )
Ng;f?g;f?gmBP?tanhr   gHm?r"   r   g      )rd   r   erfrc   )r   rv   r   M_SQRT2	M_SQRT1_2
M_2_SQRTPIkBetakKappax_sqx_cuberI   
tanh_innerleftrightleft_derivativetanh_derivativeinner_derivativeright_derivativekAlphacdfpdfr*   r*   r+   gelu_backward   s,   
r   inputc                 C   s:   t t|}t |}|| d||   }| ||  S rX   )rd   r   Fsoftplussigmoid)rh   r   input_tanh_softplusinput_sigmoidoutr*   r*   r+   mish_backward  s   
r   c                 C   s   | t |  S r6   )rd   r   r   r*   r*   r+   silu  s   r   c                 C   s,   ddt |   }| | d|d|    S rX   )rd   rc   )rh   rv   r   r*   r*   r+   silu_backward  s   r   weightc                 C   s   t | dk| ||  S rn   r   )rv   r   r*   r*   r+   _prelu_kernel$  s   r   c                 C   s4   t |dk| ||  }t |dkd||  }||fS )Nr   r   r   )rh   rv   r   
input_gradweight_gradr*   r*   r+   _prelu_kernel_backward)  s   r   noiseloweruppertrainingc                 C   s6   |r|| dkr|  |S || d }t| |||S )Ngư>r#   )mulr   r   )rh   rv   r   r   r   r   r   r   r*   r*   r+   rrelu_with_noise_backward4  s   
r   bufferc                 C   sN   |dk }t |dd}t |dd}t t | }| |||d|     S )Nr   r"   rP   )rd   re   rc   abs)rh   rv   r   in_negative	max_derivsignrf   r*   r*   r+   log_sigmoid_backwardI  s
   r   loss	reductionc                 C   s0   |t jjkrt| S |t jjkrt| S | S r6   )r!   r(   rw   rd   meanr)   sum)r   r   r*   r*   r+   apply_loss_reductionV  s
   

r   dtypec                 C   s4   | t jkrt jS | t jkrt jS | t jkrt jS d S r6   )rd   	complex32float16	complex64float32
complex128float64r   r*   r*   r+   to_real_dtype_  s   


r   targetc                 C   s   | | d }t ||S )Nr#   )r   )rv   r   r   r   r*   r*   r+   mse_lossn  s   
r   c                 C   s,   |t jjkrd|  nd}|||  |  S )N       @)r!   r(   rw   numel)rh   r   r   r   normr*   r*   r+   mse_loss_backwardx  s   r   c                 C   sF   t j| ||d}| td}t j||dd}t |}t |||S )N)rM   r   z-infTrM   keepdim)rd   softmaxeqfloatall
zeros_likere   )rv   rM   r   r   maskedmasked_rowszerosr*   r*   r+   safe_softmax  s
   
r   rb   c                 C   s<   | |   }t||k d|d  | |d|  }t||S )Nr   r#   )r   rd   re   r   )rv   r   r   r_   r   r*   r*   r+   smooth_l1_loss  s   	&
r   c           	      C   sZ   |t jjkrd|  nd}|| }t|}||  }t||k || | |t| S ra   )r!   r(   rw   r   rd   r   re   r   )	rh   rv   r   r   r_   r   r2   abs_x	norm_gradr*   r*   r+   smooth_l1_loss_backward  s   

r   c                 C   *   t | ||||}t||j t||ddS NT	copy_fromcopy_toexact_dtype)r   r   shaper   )rh   rv   r   r   r_   rU   resultr*   r*   r+   smooth_l1_loss_backward_out     
r   deltac              
   C   s`   |t jjkrd|  nd}|| }t|| k | |  | t||k||  | || |  S ra   )r!   r(   rw   r   rd   re   )rh   rv   r   r   r   r   r2   r*   r*   r+   huber_loss_backward  s    r   c                 C   r   r   )r   r   r   r   )rh   rv   r   r   r   rU   r   r*   r*   r+   huber_loss_backward_out  r   r   ignore_indextotal_weightc                 C   s   |  dk rdnd}|tjjkr| | } ||}t||k|d}t|}	t|	||d}	|	  |     kr=dkrDn n| |} |d urcdd t	|  D }
|j
d |
|< ||
}| | } t||k| d} |	|  S )Nr#   r   r"   g      c                 S   s   g | ]}d qS r"   r*   r1   rS   r*   r*   r+   r3     r}   z&_nll_loss_backward.<locals>.<listcomp>)rM   r!   r(   rw   rR   rd   re   r   scatterrQ   r   reshape)rh   rv   r   r   r   r   r   channel_dimsafe_targetrU   	new_shaper*   r*   r+   _nll_loss_backward  s    	

 

r  c           
      C   s   |  dks
J dt|  |}||}|d dks'J d| d| |d }||d|}||||}t|}d| | | |  }	||  }tj||	g|dS )Nr   z*glu does not support 0-dimensional tensorsr#   z.Halving dimension must be even, but dimension z	 is size rb   rz   )rM   rB   canonicalize_dimsizenarrowrd   r   cat)
rh   rv   rM   wrap_dimnIn	inputSize	firstHalf
secondHalfgradInputFirstHalfgradInputSecondHalfr*   r*   r+   glu_backward  s   

r  c                 C   sr  d|    krdksJ d J d|  dksJ d|  dko)|  dk}|sC|jd |jd ksCJ d|j d|j d| dksXJ d	|j d
|  df|d u si| |jd ksiJ d|tjjkr|  dkr|   dkr| jd |jd ksJ d|jd  d|    d| jd  n|   dkr|  dksJ d| j t| ||||||S )Nr   r#   input tensor should be 1D or 2Dr"   ;0D or 1D target tensor expected, multi-target not supportedsize mismatch (got input: 
, target: ):expected total_weight to be a single element tensor, got: z (z
 elements)rP   z<weight tensor should be defined either for all or no classesz7Expected a tensor of dimension 1 and tensor.size[0] == z but got: dimension z and tensor.size[0] == z7Expected a single element grad_output tensor, but got: )rM   r   r   r!   r'   rw   r  )rh   rv   r   r   r   r   r   no_batch_dimr*   r*   r+   nll_loss_backward  s<   ("
r  c                 C   s   |  dksJ d|   |  dksJ d|   |jd |jd kr<|jd |jd kr<|jd |jd ksHJ d|j d	|j | dks\J d
|j d|  dt| ||||||S )N   zSonly batches of spatial inputs supported (4D tensors), but got input of dimension: r   zUonly batches of spatial targets supported (3D tensors) but got targets of dimension: r   r#   r"   r  r  r  z ( z, elements))rM   r   r   r  )rh   rv   r   r   r   r   r   r*   r*   r+   nll_loss2d_backward7  s*   r  c              	   C   s\   |d t t |  | dd |t t | | dd  }|d ur)|| }t||S )Nr"   r*   i)rd   maximumlog1pnew_fulllogr   )rv   r   r   r   r   r*   r*   r+   binary_cross_entropyZ  s   

r#  c                 C   sR   d}| ||  t j|d|  |d }|d ur|| }|tjjkr'||  }|S )Ng-q=r"   r   )rd   r   r!   r(   rw   r   )rh   rv   r   r   r   EPSILONr   r*   r*   r+   binary_cross_entropy_backwardp  s   
"r%  c                 C   s    t t |  | }t||S r6   )rd   r   rc   r   )r   r   r   r   r*   r*   r+   soft_margin_loss  s   
r&  c                 C   s6   ||  t || d  }|tjjkr||  }|S rX   )rd   r   r!   r(   rw   r   )rh   rv   r   r   rU   r*   r*   r+   soft_margin_loss_backward  s   	r'  r#   otherpc                 C   s   t j| | |dS )N)r)  )r   r   )r   r(  r)  r*   r*   r+   dist  r   r*  x1x2c           	      C   s   |  ddd}tj|tjd}| ddd}tj|tjd}t| d||gd}t|||gd}||j}|	d
 S )Nr#   rP   Tmemory_formatr   )powr   rd   	ones_likecontiguous_formatr  r   matmulmT	clamp_minsqrt)	r+  r,  x1_normx1_padx2_normx2_padx1_x2_r   r*   r*   r+   _euclidean_dist  s   r=  input_sizesstartendstepc                 C   s   |  |}t|| ||||S r6   )	new_zerosrd   slice_scatter)rh   r>  rM   r?  r@  rA  rU   r*   r*   r+   slice_backward  s   

rD  r"   c                 C   sp  ddl m}m} |  }|dkrtdt|  |}t|  }t| 	 }	|dkr0td|d ur6|nd}
|d ur>|nt
j}||
dk rM|
|| 7 }
||dk rY||| 7 }||
dk rbd}
n||
|| krn|| }
|||
k rw|
}n||t
jks|||| kr|| }|  |
|	|   }||
 }|| d | ||< |	|  |9  < | jrtd| ||	|S )Nr   )guard_size_obliviousstatically_known_truez,slice() cannot be applied to a 0-dim tensor.zslice step must be positiver"   z<Slice decomposition for quantized tensors aren't implemented)%torch.fx.experimental.symbolic_shapesrE  rF  rM   RuntimeErrorrB   r	  listr
  stridesysmaxsizestorage_offsetis_quantizedNotImplementedError
as_strided)rv   rM   r?  r@  rA  rE  rF  ndimsizesstrides	start_valend_valrM  lenr*   r*   r+   slice_forward  sD   	
rW  c                    s@   | j |  dtf fdd}||d d}|||  }||fS )zn
    Normalize start and end such that both are in the range
    [0, x.get_size()[dim]] and start <= end.
    rN   c                    s,   | d u r|S | dk r|   } t t| ||S rn   r   r   )valr   r   defaultdim_sizer*   r+   
clamp_wrap  s
   z(_normalize_start_end.<locals>.clamp_wrapr   )r   int)r2   rM   r?  r@  r]  r*   r[  r+   _normalize_start_end  s
   
r_  srcc              	   C   sB  t | j|}| j| }t| |||\}}t| j}|| |d  | ||< ||}|dkr;||kr;|dkr;| S d g|   }t	j
|| jd}	|	| | ||< t	j|| jt	jd}
|dkrht	|
|	|k}
||krtt	|
|	|k }
|dkrt	|
|	| | dk}
dg|   }d||< |
|}
t|
t||
|d| S )Nr"   r   devicerb  r   rP   )rB   r	  rQ  r   r_  rI  expandclonerM   rd   arangerb  onesboollogical_andviewr   re   _unsafe_masked_index)r   r`  rM   r?  r@  rA  r\  src_sizeindicesidxmask
mask_shaper*   r*   r+   rC    s,   




rC  indexc                 C   s   |  |}t|| ||S r6   )rB  rd   select_scatter)rh   r>  rM   rq  rU   r*   r*   r+   select_backward9  s   
rs  offsetdim1dim2c                 C   s   |  |}t|| |||S r6   )rB  rd   diagonal_scatter)rh   r>  rt  ru  rv  rU   r*   r*   r+   diagonal_backward@  s   
rx  input_dtypec                 C   s   | j |kr
||}|S r6   )r   r8   )rh   rU   ry  r*   r*   r+   _cast_grad_to_input_dtypeI  s   

rz  outputc                 C   s0   | | }||t j||dd  }t| || S NTr   )rd   r   rz  
contiguous)rh   r{  rM   ry  new_grad_outputrU   r*   r*   r+   _softmax_backward_dataQ  s
   
r  c                 C   s*   | t |t j| |dd  }t| ||S r|  )rd   rc   r   rz  )rh   r{  rM   ry  rU   r*   r*   r+   _log_softmax_backward_datac  s   
r  c           
      C   sZ   | |d  ||d   }t tjtj|d}|d||d}|d|| |d}	||	 S )z/Utility function to implement im2col and col2imr#   r"   r   rb  r   rP   )r   rd   rf  int64rR   )
input_dkernel_d
dilation_d	padding_dstride_drb  blocks_d	arange_kwblocks_d_indiceskernel_gridr*   r*   r+    _im2col_col2im_indices_along_dimo  s
   r  kernel_sizedilationpaddingrJ  c              	      s&  t tdkdd  t t dkdd  t tdkdd  t tdkdd  ddd	}|d
 | d | ddd |d | jt}t |dv odtdd dd  D fdd tdd tdd   D t tdd D  fdd |dk}|s| d} | j\}}	}
}\}}\}} \}}\}}t|
||||| j	}t|||||| j	}t
| ||||f}|dd}|d d d d ||f }|dddddd}|d}|d}|||	| | || }|s|d}|S ) Nr#   c                   S      dS )Nz"im2col(): only 2D kernel supportedr*   r*   r*   r*   r+   r|         zim2col.<locals>.<lambda>c                   S   r  )Nz$im2col(): only 2D dilation supportedr*   r*   r*   r*   r+   r|     r  c                   S   r  )Nz#im2col(): only 2D padding supportedr*   r*   r*   r*   r+   r|     r  c                   S   r  )Nz"im2col(): only 2D stride supportedr*   r*   r*   r*   r+   r|     r  Tc                 S   <   |rt dd | D nt dd | D }t|dd  d S )Nc                 s       | ]}|d kV  qdS r   Nr*   r1   r)  r*   r*   r+   	<genexpr>      z1im2col.<locals>.check_positive.<locals>.<genexpr>c                 s       | ]}|d kV  qdS r  r*   r  r*   r*   r+   r    r  c                   S   r  )Nz<{param_name} should be greater {'than' zero, but got {param}r*   r*   r*   r*   r+   r|     r  z0im2col.<locals>.check_positive.<locals>.<lambda>r   rd   r~   param
param_namestrictcondr*   r*   r+   check_positive     (zim2col.<locals>.check_positiver  r  r  Fr  rJ  r   r  c                 s       | ]}|d kV  qdS r  r*   r1   dr*   r*   r+   r    r  zim2col.<locals>.<genexpr>r   c                         dt   S )NzmExpected 3D or 4D (batch mode) tensor for input with possible 0 batch size and non-zero dimensions, but got: tupler*   r   r*   r+   r|         c                 s   s>    | ]\}}}}}d |d|  ||d    d  |  V  qdS )r"   r#   Nr*   r1   r   paddilkerstr*   r*   r+   r    s
    "
r/  c                 s   r  r  r*   )r1   cr*   r*   r+   r    r  c                      s6   dt dd   d d  d d d dS )	Nz!Given an input with spacial size r/  , kernel_size=, dilation=
, padding=	, stride=z9, the calculated shape of the array of sliding blocks is z*, but its components must be at least one.r  r*   r  r  output_sizer  r   rJ  r*   r+   r|     s    r  r   rP   r"   r      T)rd   r~   rV  r   r   r  ziprR   r  rb  r   r  permuter
  r  squeeze)r   r  r  r  rJ  r  rQ  batched_input	batch_dimr  input_hinput_wstride_hstride_w	padding_h	padding_w
dilation_h
dilation_wkernel_hkernel_wblocks_row_indicesblocks_col_indicespadded_inputr{  num_blocks_rownum_blocks_colr*   r  r+   im2col  sd   	



 




r  r  c              
      s  t tdkdd  t tdkdd  t tdkdd  t tdkdd  t tdkdd  d$d	d
}|d |d |ddd |d |d | jt}t |dv outdd dd  D fdd d d  }t d | dkfdd dd tD }	|	d |	d   t d  k fdd t  dk fdd |dk}
|
s| d} | j\}}\}}\}}\}}\}}| d d | g |	 } | dddd dd!} t	|||||| j
}t|d }t	|||||| j
}d"d tD }| d d t g| }d d ||f}tj||| dd#}t|| | | | f}|
sf|d}|S )%Nr#   c                   S   r  )Nzonly 2D output_size supportedr*   r*   r*   r*   r+   r|     r  zcol2im.<locals>.<lambda>c                   S   r  )Nzonly 2D kernel supportedr*   r*   r*   r*   r+   r|     r  c                   S   r  )Nzonly 2D dilation supportedr*   r*   r*   r*   r+   r|     r  c                   S   r  )Nzonly 2D padding supportedr*   r*   r*   r*   r+   r|     r  c                   S   r  )Nzonly 2D stride supportedr*   r*   r*   r*   r+   r|     r  Tc                 S   r  )Nc                 s   r  r  r*   r  r*   r*   r+   r    r  z1col2im.<locals>.check_positive.<locals>.<genexpr>c                 s   r  r  r*   r  r*   r*   r+   r    r  c                   S   r  )Nz9{param_name} should be greater than zero, but got {param}r*   r*   r*   r*   r+   r|     r  z0col2im.<locals>.check_positive.<locals>.<lambda>r  r  r*   r*   r+   r    r  zcol2im.<locals>.check_positiver  r  r  Fr  rJ  r  )r#   r   c                 s   r  r  r*   r  r*   r*   r+   r    r  zcol2im.<locals>.<genexpr>r/  c                      r  )NzmExpected 2D or 3D (batch mode) tensor for input with possible 0 batch size and non-zero dimensions, but got: r  r*   r  r*   r+   r|     r  r   r"   c                      s   dd  d  S )Nz|Expected size of input's first non-batch dimension to be divisible by the product of kernel_size, but got input.shape[-2] = r/  z and kernel_size=r*   r*   )r  r   r*   r+   r|     s
    c                 S   s:   g | ]\}}}}}d |d|  ||d    d  |  qS r"   r#   r*   r  r*   r*   r+   r3     s    "zcol2im.<locals>.<listcomp>rP   c                      4   d d d d d d  dd  d	S 
NzGiven output_size=r  r  r  r  z , expected input.size(-1) to be 	 but got rP   .r*   r*   Lr  r  r  r  r   rJ  r*   r+   r|         c                      r  r  r*   r*   r  r*   r+   r|     r  r   r  r  c                 S   s   g | ]
\}}|d |  qS r#   r*   )r1   or)  r*   r*   r+   r3   *      
accumulater  )rd   r~   rV  r   r   r  rR   r  r  r  rb  rT   rB  prodr   _unsafe_index_putr   r  r  )r   r  r  r  r  rJ  r  rQ  prod_kernel_sizecolr  out_hout_wr  r  r  r  r  r  r  r  indices_rowindices_coloutput_padded_sizer{  rn  r*   r  r+   col2im  s   




 



"

r  ro  c                 C   s$   | | | |  jt| d}|S Nr-  )type_asre  rB   r   )rh   ro  rj   rG   r*   r*   r+   native_dropout_backward7  s   	r  
input_size	dimensionr
  c           	      C   s   t |dkrt| dS tt ||}tj|| | jtjd}|d||	 }| 
d|d 	||d } | |}d| |f }tj||| dd S )Nr   rc  rP   r"   r6   Tr  )rV  rd   squeeze_copyrB   r	  rf  rb  int32unfoldflattenmovedimrB  r   r  r}  )	r   r  r  r
  rA  rM   rn  rU   rq  r*   r*   r+   unfold_backwardF  s   
r  epsc              	   C   st   |d ur|}d| }t t ||k||k| |d|   dS t t |dk|dk| |d|   |dtdS )Nrb   r   r*   nan)rd   re   ri  r!  r   )rh   rv   r  lohir*   r*   r+   logit_backwardY  s   r  trainc                 C   s&   |r|dkrt | ||d S |  S rn   )r   native_dropoutre  )r   r)  r  r*   r*   r+   dropoutn  s   r  out0out1c                 C   s   |r6|dkr6|dkrt | t j| t jdfS | jjstdt | |k}||  tdd|   }||fS | t j| t jdfS )Nr   r"   r   z?result type Float can't be cast to the desired output type Longrb   )	rd   r   rh  r   is_floating_pointrH  	rand_liker   r1  )r   r)  r  	bool_maskresr*   r*   r+   r  x  s   r  half_to_floatc                 C   s   |   } |r| jtjksJ tj| tjjd\}}| |} | 	 dkr*t
| }ntj| |dd}t
| | }|tj||dd }|sJ||}|S Nr4   r   T)r   )r}  r   rd   halfrB   rC   ELEMENTWISE_TYPE_PROMOTION_KINDDEFAULTr8   r   rc   amaxr   )r2   rM   r  r;   r>   unnormalizedx_maxr   r*   r*   r+   _softmax  s   


r  )r   c           	      C   s   |   } |r| jtjksJ tj| tjjd\}}| |} | 	 dkr'| }ntj
| |dd}| | }ttjt||dd}|| }|sL||}|S r  )r}  r   rd   r  rB   rC   r  r  r8   r   r  r"  r   rc   )	r2   rM   r  r;   r>   shiftedr  shifted_logsumexpr   r*   r*   r+   _log_softmax  s    


r
  rP   rm  padding_idxscale_grad_by_freqsparsec                 C   sJ   |   dks
J d|jdkr!| d|}|jdkr|d}|S | | S )Nr#   z'weight' must be 2-Dr"   r   )rM   rQ  index_selectr  )r   rm  r  r  r  r   r*   r*   r+   	embedding  s   	


r  num_weightsc                 C   s   t j| t jjd\}}| |} t|tj}|r8||f}t	|}t
j||g|dd}|| }	| |	d } t||k| j}
| |
d}| |f| j|jd   }t
j||g|dd|S )Nr   Tr  rP   r   )rB   rC   r  r  r8   r   rd   longrB  r1  r   r  rR   rT   rQ  masked_fillr   )rh   rm  r  r  r  r;   r>   countsrg  grad_weights_scalero  r   grad_weightr*   r*   r+   embedding_dense_backward  s&   	


r  c                 C   s   d}| D ]}||9 }q|S rX   r*   )r2   rG   ir*   r*   r+   r    s   
r  tensors
num_chunksc           	      C   s   g }| D ]H}|  }|| | d | | }||| kr7dgd |j| d  d|||  g }t||d}|d | t|dg }||| q|S )Nr"   r   r#   rP   )r
  rQ  r   constant_pad_ndrd   Sizeappendrj  )	r  rM   r  padded_tensorstensortensor_sizepad_along_dimr  	view_sizer*   r*   r+   
_pad_chunk  s   
r"  c                 C   s(   | d j }| D ]
}|j |kr dS qdS )Nr   FTrQ  )r  rQ  r  r*   r*   r+   have_same_ndims  s   

r$  c                 C   sB   | d   d | }| D ]}t|  d | |kdd  qd S )Nr   c                   S   r  )NzG_chunk_cat expects same sizes of 0,...,dim-1 dimensions for all tensorsr*   r*   r*   r*   r+   r|     r  z+leading_dimension_matches.<locals>.<lambda>)r
  rd   r~   )r  rM   leading_dim_sizesr  r*   r*   r+   leading_dimension_matches  s   r&  c                 C   s   t |dkdd  t t| dkdd  | d j}| d j}| D ]$}t | dkdd  t |j|kdd  t |j|kdd  q"t| rVt| d 	 |}nt |dkd	d  | D ]}t ||j
k d
d  qbt| | |S )Nr"   c                   S   r  )Nz&_chunk_cat expects positive num_chunksr*   r*   r*   r*   r+   r|     r  z._preprocess_chunk_cat_inputs.<locals>.<lambda>r   c                   S   r  )Nz0_chunk_cat expects a non-empty input tensor listr*   r*   r*   r*   r+   r|      r  c                   S   r  )Nz#_chunk_cat expects non-empty tensorr*   r*   r*   r*   r+   r|   %  r  c                   S   r  )Nz8_chunk_cat expects all input tensors with the same dtyper*   r*   r*   r*   r+   r|   (  r  c                   S   r  )Nz8_chunk_cat expects all inputs tensors on the same devicer*   r*   r*   r*   r+   r|   ,  r  c                   S   r  )NzK_chunk_cat expects non-negative dim when input tensors have different ndimsr*   r*   r*   r*   r+   r|   3  r  c                   S   r  )Nz3_chunk_cat expects dim < ndim for all input tensorsr*   r*   r*   r*   r+   r|   8  r  )rd   r~   rV  r   rb  r   r$  rB   r	  rM   rQ  r&  )r  rM   r  expected_dtypeexpected_devicer  r*   r*   r+   _preprocess_chunk_cat_inputs  s:   


r)  r   c                 C   sH   t | ||}t| ||}|d u rt||d S tj||d |d |S )Nr"   )r   )r)  r"  rd   r  )r  rM   r  r   r  r*   r*   r+   
_chunk_cat>  s   r*  split_sizesc                 C   sX   t j| ||d}|d u rdd |D S t||D ]\}}t||j t||dd qd S )Nrz   c                 S   s   g | ]	}|j tjd qS )r-  )re  rd   r2  )r1   sr*   r*   r+   r3   Z  s    z)split_with_sizes_copy.<locals>.<listcomp>Tr   )r   split_with_sizesr  r   r   r   )rv   r+  rM   r   splitsr{  splitr*   r*   r+   split_with_sizes_copyO  s   	r0  
split_size.c                 C      t j| ||S r6   )r   r/  r   )r   r1  rM   r*   r*   r+   unsafe_splitb     r3  c                 C   r2  r6   )r   r-  rZ  )r   r+  rM   r*   r*   r+   unsafe_split_with_sizesg  s   r5  c                    s   | j }|| } dkr|dksJ |  fS |  d   }ddlm} ||} fddt|D }  | |  |d< t| ||S )Nr   r"   )	guard_intc                       g | ]} qS r*   r*   r1   r  r1  r*   r+   r3   {  r}   zsplit.<locals>.<listcomp>rP   )r   detachrG  r6  rQ   rd   r/  )rv   r1  rM   r>  r\  chunksr6  r+  r*   r9  r+   r/  n  s   
r/  tensor_indices_or_sectionsc                    s   |j jdksJ |jtjksJ |  t dkp dk fdd  dkr9| }t|t	s3J | 
||S dd |D }| 
||S )Ncpur"   r   c                      s   d  dS )Nz{tensor_split expected tensor_indices_or_sections to be a zero-dimensional or one-dimensional tensor, but got a tensor with z dimsr*   r*   	split_dimr*   r+   r|     s    zAtensor_split_tensor_indices_or_sections_py_impl.<locals>.<lambda>c                 S   s   g | ]}|  qS r*   )itemr8  r*   r*   r+   r3         zCtensor_split_tensor_indices_or_sections_py_impl.<locals>.<listcomp>)rb  typer   rd   r  rM   r~   r@  r/   r   tensor_split)rv   r<  rM   sectionsrm  r*   r>  r+   /tensor_split_tensor_indices_or_sections_py_impl  s   

rE  mat1mat2c                 C   H   |   s|  st|}t|}|t|| }|dkr|S |||   S rn   )r  
is_complexr^  rd   mm)rv   rF  rG  r_   ri   r   r*   r*   r+   addmm  s   rK  use_geluc                 C   s<   t | ||||}|r| jrtj|ddS t|S t|S )Nr   )r   )rK  is_cudar   gelurelu)rv   rF  rG  r_   ri   rL  r   r*   r*   r+   _addmm_activation  s   

rP  vecc                 C   rH  rn   )r  rI  r^  rd   mv)rv   rF  rQ  r_   ri   r   r*   r*   r+   addmv  s   rS  r   rstdgammaNCHxWgroupoutput_maskc
              	      s  t j| ||dd t j|| dd t j|dd t|    k fdd tjfkfdd td u pJ  k fdd t \}
}t|dk fdd t| |	 j
d	gd
}| 	 j
d	gd
}d }d }d }|	d r:d|
  }d urt|d|

d	}t|d|

d	}t|dd|
}n&||

d	}||

d	}t|dtjd|
f|jd}| | | | | | }|  || |  }|d}t|d}t|d}t| |
|t||
| | }||j|j}|	d r_|	|
|	|
d  |d j
dgd
 }|	d	 rk|j
dgd
}|||fS )NF)allow_cpu_scalar_tensorsc                      s   d    dS )NzExpect input to have z	 elementsr*   r*   )rW  rX  rV  r*   r+   r|     rA  z,native_group_norm_backward.<locals>.<lambda>c                      s   d  d dj  S )NzExpect mean to have shape (, z
, but got r  r*   )rV  rY  r   r*   r+   r|         c                      s$   d  dd ur   S d S )NzExpect gamma to have z elements but got rP   )r   r*   )rW  rU  r*   r+   r|        $ r   c                      s   d  d S )NzExpect number of channels z, to be evenly-divisible by number of groups r*   r*   )rW  rY  r*   r+   r|     r}   r#   rz   rb   rP   r"   ra  r  )rB   check_same_devicecheck_same_shaperd   r~   r   r   divmodr   rj  r   rR   r  rg  rb  rT   r8   r   )rh   r   r   rT  rU  rV  rW  rX  rY  rZ  cpg_remdsdbd_inputd_gammad_biasr,  ds_valdb_valc1c2c3r*   )rW  rX  rV  rU  rY  r   r+   native_group_norm_backward  s   
 
""



$

rn  out2c
                C   d   t | |||||||||	
}|
||f}t|D ]\}}|d ur/t|| |j t||| dd q|S r   )rn  	enumerater   r   r   )rh   r   r   rT  rU  rV  rW  rX  rY  rZ  r  r  ro  r   rU   r  rG   r*   r*   r+   native_group_norm_backward_out1  s   
rr  c                 C   s   | d ur	|  |S | S r6   r8   )r2   r   r*   r*   r+   _maybe_castN  s   
rt  grad_outnormalized_shapebiasc           "         sf  |j }| }	t|j  fdd| |||fD \}
}}}|
d us$J |	t| }||d  }|d | }g }g }t|	D ]}||krJ|| q>|| q>t|}t|}ddl	m
} ||dksj||dkr|d rs||nd |d r|||d  nd |d r|||d  fS d fS t|| }t|| }|| | }|d ur|
| }n|
}|| }t||d}t||}t||d}t||}|| | }d }d } d }!|d r|| | }|d r|d urt|dkrt|
| |d} n|
| } |d r"|d ur"t|dkrt|
|d}!n|
 }!t||jt| |jt|!|jfS )	Nc                 3   s*    | ]}|d ur|   n|V  qd S r6   )r8   r}  r0   r:   r*   r+   r  c  
    
z-native_layer_norm_backward.<locals>.<genexpr>r   rE  r"   r#   TF)r   rM   rB   get_computation_dtyper   rV  rQ   r  r  rG  rE  rB  rT   rd   r   r   re  rt  )"ru  r   rv  r   rT  r   rw  rZ  input_shape
input_ndimgrad_out_cast
input_castweight_cast	bias_castaxis
inner_dims
outer_dimsinner_dim_indicesouter_dim_indicesr  rV  MrE  x_hat
grad_x_hatabrk  rl  rm  rI   rf  d_weightrh  r*   r:   r+   native_layer_norm_backwardU  sn   





r  c             	   C   s`   t | |||||||}||	|
f}t|D ]\}}|d ur-t|| |j t||| dd q|S r   )r  rq  r   r   r   )ru  r   rv  r   rT  r   rw  rZ  r  r  ro  r   rU   r  rG   r*   r*   r+   native_layer_norm_backward_out  s   
r  running_meanrunning_varmomentum
functionalc	                 C   sT  dgt td|   }	t| j}
|}|}|rt| j}
| j|
d}tj||	ddd\}}t	|| }| | | }t
||	}t
||	}|d ur]|| d| |  }|s]|| |d ur|  | jd  }t
||	}|||d   }|| d| |  }|s|| nT|d ur|d usJ |j|
dd}|}|j|
dd}|}|}dt||  }| jjdkr|}|}n
| d	}| d	}t||  d }t||  d }| | | }|d ur| }t||  d }|| }|d ur	| }t||  d }|| }| jjdkr|j| jd}|j| jd}|j| jd||||fS )
Nr   r#   r   T)rM   
correctionr   r"   )r   r   r=  r   )rI  rQ   rM   rB   rz  r   r8   rd   var_meanrsqrtr  copy_r   r   r6  rb  rB  rB  rT   r  )r   r   rw  r  r  r   r  r  r  reduction_dimsr;   new_running_meannew_running_var	input_acc
biased_varr   rT  r{  	save_mean	save_rstdnsqueezed_varunbiased_varinvstdr*   r*   r+   native_batch_norm_helper  st   





r  r  save_invstdc              
   C   ,   t | |||||||d	\}}	}
}}||	|
fS NFr  r   r   rw  r  r  r   r  r  r{  r  r  rS   r*   r*   r+   native_batch_norm  s   
r  c              
   C   sv   |d u r|d u rt | |||||S |d u rtd|d u r"td|r0t | |||||||S t | ||||||S )Nz`running_mean is None, but running_var is provided. They should both be None or both be provided.z`running_var is None, but running_mean is provided. They should both be None or both be provided.)r   _native_batch_norm_legitrH  $_native_batch_norm_legit_no_training)r   r   rw  r  r  r   r  r  r*   r*   r+   native_batch_norm_decomposition/  s&   r  c                    s|   |  |}|| d |   dkr4|dkr4 fdd|D }  | |  ||d < tjjj| ||S tjjj|  |S )Nr"   r   c                    r7  r*   r*   r  r9  r*   r+   r3   Z  r}   z(unsafe_chunk_py_impl.<locals>.<listcomp>)r
  rd   opsr   r5  rZ  r3  r   )r  r;  rM   r\  r+  r*   r9  r+   unsafe_chunk_py_implT  s   
r  c              
   C   s   t j| ||||d||S r  )r   r  rZ  )r   r   rw  r  r  r  r  r*   r*   r+   r  `  s   
r  c              
   C   r  r  r  r  r*   r*   r+   r  v  s   
r  c           
   
   C   s,   t | ||d d |||d	\}}}}	}	|||fS r  r  )
r   r   rw  r   r  r  r{  r  r  rS   r*   r*   r+   !_native_batch_norm_legit_no_stats  s   	
r  c              
   C   sP   t | |||||||d	\}}	}
}}|d usJ d|d us!J d||	|
||fS )NT#new_running_mean should not be None"new_running_var should not be Noner  )r   r   rw  r  r  r   r  r  r{  r  r  r  r  r*   r*   r+   #_native_batch_norm_legit_functional  s   r  c           	   	   C   sP   t j| ||||d|}d}|t jjjkrt j| |}t j|t j| j| j	dS )a  
    Return a reserve tensor for batch norm, used only by cudnn to pass forward state to the
    backward pass. This is needed for `_batch_norm_with_update` and `_batch_norm_no_update`,
    which support a variety of backends including cudnn. We create this tensor here to get
    the correct shape in the traced graph if we detect that will call the cudnn kernel,
    and rely on DCE to avoid materializing this tensor.
    Tr   )r   layoutrb  )
rd   _C_select_batch_norm_backend_BatchNormBackendCudnn(_get_cudnn_batch_norm_reserve_space_sizeemptyuint8r  rb  )	r   r   rw  r  r  r  r   backendreserve_sizer*   r*   r+   _get_batch_norm_reserve_tensor  s   r  c              
   C   sD   t | ||||d||d	\}}}	}
}
t| |||||dd}|||	|fS )NTFr   r  r  r   r   rw  r  r  r  r  r{  r  r  rS   reserver*   r*   r+   _batch_norm_with_update     
r  c              
   C   sh   t | ||||d||d	\}}}	}
}t| |||||dd}|
d us$J d|d us,J d|||	||
|fS )NTr  r  r  r  )r   r   rw  r  r  r  r  r{  r  r  new_rmnew_rvr  r*   r*   r+   "_batch_norm_with_update_functional  s   r  c              
   C   sD   t | ||||d||d	\}}}	}
}
t| |||||dd}|||	|fS )NFr  r  r  r*   r*   r+   _batch_norm_no_update   r  r  c                 C   sB   |d u sJ t | |k jt jd}|| |  d|  }||fS )Nr   rb   )rd   r  r8   r  r  )r   r)  	generatorro  r  r*   r*   r+   _fused_dropout_decomposition  s   r  )r   r  rb  
pin_memorynon_blockingr.  rb  r  r  r.  c          	      C   s
  |r|t jksJ d|rJ dt| t jttttfsJ |d u r6|d u r6|d u r6t| t jr4|  S | S d}t| t jrA| }nt 	| }|d uri||j
kri|d ura|jdkrat j||}d}t j|||}|d urx|sxt j||}d}|d urt j||dS |S )NTODOFr=  Tr-  )rd   stridedr/   r   r^  r   rh  complexre  scalar_tensorrb  rB  _primsconvert_element_type
device_put)	r2   r   r  rb  r  r  r.  dtype_convertedx_tensorr*   r*   r+   _to_copy%  s,   
r  c                 C   s
   t | S r6   )r   aliasr9   r*   r*   r+   nop_decompositionS  s   
r  out3exponential_average_factorepsilonc              
   C   s^   t | |||||||\}}	}
|r||	|
| jdtjdfS ||d|d| jdtjdfS )Nr  r   )r   r  rB  rd   r  )r   r   rw  r  r  r   r  r  r  r  r  r*   r*   r+   cudnn_batch_norm[  s"   
r  c                 C   s@   t |D ]\}}|dkr|| jk r| j| |ks| |} q| S rX   )rq  rQ  r   rR   )r2   broadcast_maskr  ro  r*   r*   r+   _broadcast_batch_norm_backward}  s
    
r  r  c                 C   s   t | |||||||||	
S r6   )native_batch_norm_backward)ru  r   r   r  r  r  r  r  r  rZ  r  r*   r*   r+   batch_norm_backward  s   r  c
           &         s  |j }
|d ur|j }n|
}t|j   fdd| ||||||fD \}}}}}}}|j}| }|dks9J dd}tt|||  }|}|}|rV|d urS|d usUJ n|d ur^|d us`J |}t|| }dg| }|| ||< g }t	|D ]}||kr|
| qzt||}d| }t||}t|||  |}t|| |}tt|| || |} |d u rt||d }!nt|| |}!|r|| |  }"||" | |! }#n||! }#|	d r|| }$nd }$|	d r|}%nd }%|#|
t|$|t|%|fS )Nc                 3   s&    | ]}|d ur|  n|V  qd S r6   rs  r0   r:   r*   r+   r    s
    
z-native_batch_norm_backward.<locals>.<genexpr>r#   z$rank of the input must be at least 2r"   rb   )r   rB   rz  r   rM   r  rI  rd   r  rQ   r  r  r   r   r8   rt  )&ru  r   r   r  r  r  r  r  r  rZ  ry  weight_dtyper}  r~  r  running_mean_castrunning_var_castsave_mean_castsave_invstd_castr{  
input_rankr  num_featuresr   r  r  reduction_axesr  r   grad_output_sumdot_p	grad_mean
proj_scale
grad_scaleprojrU   r  	grad_biasr*   r:   r+   r    s   
	



r  c
                C   rp  r   )r  rq  r   r   r   )ru  r   r   r  r  r  r  r  r  rZ  r  r  ro  r   rU   r  rG   r*   r*   r+   native_batch_norm_backward_out	  s&   
r  save_varc                 C       t || |||||d|g d
S NT)TTTr   r  )r   rh   r   r  r  r  r  r  r*   r*   r+   miopen_batch_norm_backward,	  s   r  reserveSpacec	           	      C   r  r  r  )	r   rh   r   r  r  r  r  r  r  r*   r*   r+   cudnn_batch_norm_backwardF	  s   r  c                    s  | j  | jttdv fdd | jdd  D ]}t|dkfdd qd |d  dkrjd |d  dkrjtdd	 tdd  |D }td
d	 tdd  ||D }tjj	| ||S dd dd  fdd}|d |d \}}}}	|d |d \}
}}}| dt
|d|
f }|	s|stj|ddS dd }|||||	dd\}}|||||dd\}}d }tt|jd t|jd D ]\}}|d u r|d|d d |f }q||d|d d |f  }q|||  S )Nr  c                      
   d  S )Nz9adaptive_avg_pool2d(): Expected 3D or 4D tensor, but got r*   r*   r#  r*   r+   r|   k	     
 z%adaptive_avg_pool2d.<locals>.<lambda>r/  r   c                         dt   dS )Nzjadaptive_avg_pool2d(): Expected input to have non-zero size for non-batch dimensions, but input has shape r  r  r*   r  r*   r+   r|   p	  s    rP   c                 s   s    | ]	\}}|| V  qd S r6   r*   )r1   r  r  r*   r*   r+   r  v	      z&adaptive_avg_pool2d.<locals>.<genexpr>c                 s   s&    | ]\}}}||d  |  V  qdS r"   Nr*   )r1   r  r  r,  r*   r*   r+   r  w	  s    
c                 S   s   t j| | |ddS )Ntruncrounding_moderd   divr  r  r  r*   r*   r+   start_index|	  s   z(adaptive_avg_pool2d.<locals>.start_indexc                 S   s    t j| d | | d |ddS )Nr"   r  r  r  r   r*   r*   r+   	end_index	      z&adaptive_avg_pool2d.<locals>.end_indexc                    s   t j| t jd}||| }| | d }| | }|dkp"|| dk }|r+|d7 }n|dkr3|d8 }t j| t jd}|d| }|rbt j| d |j|jd}	t ||	}||| }
|
| }n|}||||fS )Nrc  r"   r   rP   r  )rd   rf  r  rR   r  r   rb  minimum)in_sizeout_sizeorangei0	maxlengthin_size_modadaptive	range_maxrn  maxvali1length)rb  r  r  r*   r+   compute_idx	  s(   

z(adaptive_avg_pool2d.<locals>.compute_idx.r  )r   rP   rz   c                 S   s`   t |tr	| |fS |dk sJ ||dk}|dkrt|d}t| |d} t|| }| |fS )Nr   rP   r/  r  r   )r/   r   rR   rT   rd   r  )valsr  r  r  rM   ro  r*   r*   r+   
maybe_mask	  s   

z'adaptive_avg_pool2d.<locals>.maybe_mask)r  rM   r   )rb  r   rV  rd   r~   r  r  nnr  
avg_pool2drT   r   r   rQ   )r   r  r  rJ  kernelr  idxhlength_hrange_max_h
adaptive_hidxwlength_wrange_max_w
adaptive_wr  r  retr  jr*   )rb  r  rQ  r   r  r+   adaptive_avg_pool2da	  sN   

(  



&r   c           	      C   s   t d| d ttj| jd |  }ttj|}dg| j }| jd |  |d | < |tj|| j	d
||  d}| t| jd |  t| }tj|d|g| ddd
|jS )Nmax_unpoolingd_forward_outr"   ra  rP   Fr  )rB   alert_not_deterministicr   operatorr   r   rQ  r   rf  rb  rj  r  rB  rI  r  )	rv   rm  r  rM   nchwindices_nc_shapeindices_flatr{  r*   r*   r+   _max_unpoolnd	  s   	"r)  c                    s   t jt jkfdd t tdkfdd t jdv fdd t jjkfdd tdjD ] t  d	k fd
d q>t	dS )Nc                         d j  S )Nz2elements in indices should be type int64 but got: r   r*   )rm  r*   r+   r|   	      zmax_unpool2d.<locals>.<lambda>r#   c                      r  )NzMThere should be exactly two elements (height, width) in output_size, but got 
 elements.rV  r*   r  r*   r+   r|   	     r  c                         d j  dS )NzLInput to max_unpooling2d should be a 3d or 4d Tensor, but got a tensor with  dimensions.r#  r*   r   r*   r+   r|   	  s   c                         dj  d j  S NzBExpected shape of indices to be same as that of the input tensor (z%) but got indices tensor with shape: r  r*   )rm  rv   r*   r+   r|   	     
r"   r   c                         dj  d  dS )NzZmax_unpooling2d(): Expected input to have non-zero size for non-batch dimensions, but got  with dimension  being empty.r  r*   )r  rv   r*   r+   r|   
  
   )
rd   r~   r   r  rV  rQ  r   rQ   r
  r)  )rv   rm  r  r*   )r  rm  r  rv   r+   max_unpool2d	  s,   





	r9  c                    s  t jt jkdd  t jdv fdd t tdkfdd t tdkfdd t tdkfdd t jjkfd	d td
jD ] t  dk fdd qXt d dko~d
 dko~d dkfdd t	dS )Nc                   S   r  )Nz(elements in indices should be type int64r*   r*   r*   r*   r+   r|   
  r  zmax_unpool3d.<locals>.<lambda>r  r  c                      r0  )NzLInput to max_unpooling3d should be a 4d or 5d Tensor, but got a tensor with r1  r#  r*   r   r*   r+   r|   
      r   c                      r  )NzVThere should be exactly three elements (depth, height, width) in output_size, but got r,  r-  r*   r.  r*   r+   r|   #
  r/  c                      r  )NzRThere should be exactly three elements (depth, height, width) in stride, but got: r,  r-  r*   rJ  r*   r+   r|   *
  r}   c                      r  )NzSThere should be exactly three elements (depth, height, width) in padding, but got: r,  r-  r*   )r  r*   r+   r|   .
  r}   c                      r2  r3  r  r*   )rm  r   r*   r+   r|   2
  r4  r"   r   c                      r5  )NzZmax_unpooling3d(): Expected input to have non-zero size for non-batch dimensions, but got r6  r7  r  r*   )r  r   r*   r+   r|   ;
  r8  r#   c                      r  )Nz5strides should be greater than zero, but got stride: r*   r*   r=  r*   r+   r|   D
  r  )
rd   r~   r   r  rQ  rV  r   rQ   r
  r)  )r   rm  r  rJ  r  r*   )r  rm  r   r  r  rJ  r+   max_unpool3d
  sB   	







	"
r>  )ri   r  c                C      t | |||d|dS )NTinplaceri   
_index_addr2   rM   rq  r  ri   r*   r*   r+   
index_add_J
  s   	rE  c                C   r?  )NFr@  rB  rD  r*   r*   r+   	index_addV
  s   
rF  rA  c                   s"  t | jtjdkfdd jdkrdnd|jdkr*|ndtkfdd  dkr]t | jttkpQt 	t
  fdd |  }| jdk}|ri| dn| }d f }|rwtjntj}	|	|||dd	}
|r| S |r|
dS |
 S )
Nr"   c                      r0  Nz(Index should have dimension 1 or 0 (got r  r#  r*   rq  r*   r+   r|   o
  r<  z_index_add.<locals>.<lambda>r   c                      s   d d d S )NzNumber of indices (z') should be equal to tensor.size(dim) (z), for dim=r*   r*   )rM   
index_sizer  r*   r+   r|   u
      c                      s   dt   d dS )Nzalpha argument of type z cannot be safely cast to type !)rB  r*   )ri   python_typer*   r+   r|   |
  rJ  r6   Tr  )rB   canonicalize_dimsrQ  rd   r~   r
  dtype_to_typer   rh  is_weakly_lesser_typerB  rR   r   
index_put_	index_putr  r}  )r2   rM   rq  r  rA  ri   zero_dimr+  rn  rQ  r   r*   )ri   rM   rq  rI  rL  r  r+   rC  c
  s6   	

rC  r   c              
   C   s   t t| dkdd  t| }| d  }|dd  }tdd | D }|r,||f}n||f}|| }| d ||}dt| }	t|D ]+}
| |
 }t||	d||d f |}|rhtj	||d|
d}qFtj	||d|
d}qF|S )	Nr   c                   S   r  )Nz#received an empty list of sequencesr*   r*   r*   r*   r+   r|   
  r  zpad_sequence.<locals>.<lambda>r"   c                 s   s    | ]}| d V  qdS r  r
  r0   r*   r*   r+   r  
      zpad_sequence.<locals>.<genexpr>)r   r   rM   rq  )
rd   r~   rV  r
  r   r!  rQ   r   r  rr  )	sequencesbatch_firstpadding_valuesequences_sizemax_sizetrailing_dimsmax_lenout_dimsr   dim_paddingsr  currseqrowr*   r*   r+   pad_sequence
  s(   
ra  c                 C      t | |||ddS )NTrA  _index_copyr2   rM   rq  r  r*   r*   r+   index_copy_
     rg  c                 C   rb  )NFrc  rd  rf  r*   r*   r+   
index_copy
  r   ri  c          
         s   t | j|}t jdk fdd | jdk}|r | dn| } jdkr, dn  d|  f }|r:tjntj}||||}	|rG| S |rN|		dS |	
 S )Nr"   c                      r0  rG  r#  r*   rH  r*   r+   r|   
  r<  z_index_copy.<locals>.<lambda>r   r6   )rB   rM  rQ  rd   r~   rR   r   rP  rQ  r  r}  )
r2   rM   rq  r  rA  rR  r+  rn  rQ  r   r*   rH  r+   re  
  s   

re  c                 C   sR   t | d| }t t |  }| js| jr| d}n|}|t | |fS )Nr*   r  )rd   r  rB  rc   r   rM  is_xpur   )rv   r   rf   r   r*   r*   r+   log_sigmoid_forward
  s   rk  lowhighr  c                 C   s$   t j| jt|t|| j| j|dS )N)rl  rm  r   rb  r  )prims_uniform_helperr   r   r   rb  )r2   rl  rm  r  r*   r*   r+   uniform
  s   rp  c                 C   s   |  t| |||S r6   )r  rp  )rv   rl  rm  r  r*   r*   r+   uniform_
  s   rq  c                 C   s   t | d }|d ur"t|d u dd  tt ||kdd  |S |d urjt|d u dd  tt ||kdd  g }t|D ]%\}}t||krZ|| |d  t|  qB|t| |d  |  qB|S tddd  d S )	Nr#   c                   S   r  Nz9Must specify exactly one of output_size and scale_factorsr*   r*   r*   r*   r+   r|   
  r  z.upsample_compute_output_size.<locals>.<lambda>c                   S   r  N r*   r*   r*   r*   r+   r|   
  r  c                   S   r  rr  r*   r*   r*   r*   r+   r|   
  r  c                   S   r  rs  r*   r*   r*   r*   r+   r|   
  r  Fc                   S   r  rr  r*   r*   r*   r*   r+   r|     r  )rV  rd   r~   rq  r^  r  r   )r  r  scale_factorsspatial_dimensionsr  r,  r*   r*   r+   upsample_compute_output_size
  s.   rw  c                 C   s   | d u rd S | | S r6   r*   )scalesrn  r*   r*   r+   get_scale_value  s   ry  ru  c                 C   s2   t |  ||}|r|nd gt| }t| ||S r6   rw  r
  rV  _upsample_nearestr   r  ru  osizerx  r*   r*   r+   _upsample_nearest_vec  s   r~  c                 C   s6   t |  ||}|r|nd gt| }t| ||ddS NTexactrz  r|  r*   r*   r+   _upsample_nearest_exact_vec#  s   r  c                 C   s   g }t |}|r
dnd}t|D ]I}|| }| j| |  }	|| d ur,|	|	||   n|	| }
tj|tj| jd}|| |
 tj}t|d | D ]}|	d}qL|
| q|S )Nr   r   r  r"   rP   )rV  rQ   r   rd   rf  r   rb  r8   r  rR   r  )r   r  rx  r  rm  num_spatial_dimsrt  r  r}  isizerj   output_indicesinput_indicesrS   r*   r*   r+   !_compute_upsample_nearest_indices8  s   $r  )preserve_memory_formatr   rx  c                 C   s   t | ||gS r6   r{  r   r  rx  r*   r*   r+   upsample_nearest1dX  s   	r  c                 C   s   t | ||gddS r  r  r  r*   r*   r+   upsample_nearest_exact1dd     r  scales_hscales_wc                 C   s   t | |||gS r6   r  r   r  r  r  r*   r*   r+   upsample_nearest2dr  s   
r  c                 C   s   t | |||gddS r  r  r  r*   r*   r+   _upsample_nearest_exact2d  s   r  scales_dc                 C   s   t | ||||gS r6   r  r   r  r  r  r  r*   r*   r+   upsample_nearest3d  r  r  c                 C   s   t | ||||gddS r  r  r  r*   r*   r+   _upsample_nearest_exact3d  s   r  r  c           	      C   sp   t | |||d}d d g| }t| |}|jdkr6t| }| jd }| jjdkr0|dk r0t	j
}|j|d}|S )Nr  r  r"   cudar-  )r  r   _unsafe_indexrQ  rB   r   r   rb  rB  rd   r2  r}  )	r   r  rx  r  spatial_indicesrm  r   r.  
n_channelsr*   r*   r+   r{    s   


r{  c                    sb   |r|rd n|rd n|rd nd t   dks!J t  fddtdt  D S )Nr  r  r   r#   r   c                    s    g | ]}t ||   qS r*   r  r8  
group_sizeparamsr*   r+   r3     s    z!gather_params.<locals>.<listcomp>)rV  rQ   )r  
has_biaseshas_projectionsr*   r  r+   gather_params  s   r  c                 C   sh   |r!| d|  |d|  }}| d| d  |d| d  }}n| | || }}d\}}||||fS )Nr#   r"   NNr*   )r  hiddensr  bidirectional
cur_params
cur_hiddenbidir_paramsbidir_hiddenr*   r*   r+   params_hiddens  s   $r  c                 C   s2   ||ksJ | | d|||  | dd|S rn   )r  r  )r  last_batch_size
batch_sizer  r*   r*   r+   update_hidden_for_packed  s   r  c              	   C   s4   ||kr| S ||k sJ t | |d||| fS rn   )rd   concatr  )r  r  r  
inp_hiddenr*   r*   r+    update_hidden_for_packed_reverse  s   r  c                 C   s$  |d }|d }|r|d nd }	|r|d nd }
g }g }|r"|d n|d }| dd|}t| t|}|r>|d d d }|D ]-} | jd }||krLn|rVt||||}nt||||}|| |||	||
}|}|| q@|ru|  n	|| |  t	|d}|st	|dn|}||fS )Nr   r"   r#   r   rP   )
r  rd   r/  rI  r   r  r  r  reverser  )inphiddenr  r  	hidden_fnbatch_sizesr  	ih_weight	hh_weightih_biashh_biasstep_outputr  r  r  	split_inpr  r   
hidden_outr*   r*   r+   one_layer_rnn_data  s@   


r  c                        fdd}|S )Nc                    s    t ||||  S r6   r   linearr  r  r  r  r  r  nonlinearityr*   r+   rI   '  s   zrnn_cell.<locals>.innerr*   r  rI   r*   r  r+   rnn_cell&  s   r  c                    r  )Nc                    s$   t | ||}  t ||||  S r6   r  r  r  r*   r+   rI   .  s   zrnn_cell_data.<locals>.innerr*   r  r*   r  r+   rnn_cell_data-  s   r  c                 C   s   |d }|d }|r|d nd }|r|d nd }	t | ||}
|r&|
dn|
}
|d}g }|
D ]}|||||||	}|| q1|rH|  t|d}||dfS )Nr   r"   r#   r   )	r   r  fliprR   r  r  rd   r  r  )r  r  r  r  r  r  r  r  r  r  precomputed_inputr  r  r  r   r*   r*   r+   one_layer_rnn5  s   
r  c                 C   s   |d }|d }|r|d }|d }nt | }t | }|d d}	|d d}
g }d}|	d}d}d}d}d}|  } |	 }	|
 }
t jjj| |||||	|
|||||||||}|d |d |d }}}||	d|	dffS )Nr   r"   r#   r   F)
rd   r   r
  rR   r}  r  r   mkldnn_rnn_layerrZ  r  )r  r  r  r  r  w0w1w2w3hxcxr  modehidden_size
num_layersr  rW  r  outputsrW   hycyr*   r*   r+   mkldnn_one_layer_lstmK  sN   


r  c
                 C   s   |r|  ddn| } g }
t|D ]^}t||||\}}}}|r'||d k r'|nd}|	| |||\}}|
| |rI|	| |||dd\}}|
| |rXt||g| d } n|} |dkrn|rn||d k rntj| |dd} q|rw|  ddn| } | |
fS )Nr   r"   r   T)r  )r  )	transposerQ   r  r  rd   r  rM   r  )r   r  r  r  r  r  r  r  rW  layer_fnfinal_hiddensr  r  r  r  r  fwd_inp
fwd_hiddenbwd_inp
bwd_hiddenr*   r*   r+   _rnn_helper}  s,   



r  c	                 C   R   | d}	t||d}t| |	|||||||ttttjd
\}
}|
t|dfS Nr   Fr  )	unbindr  r  r   r  r  rd   r   stackr   r  r  r  r  r  r  r  rW  r  r   r  r*   r*   r+   rnn_tanh_input     
r  c	                 C   r  r  )	r  r  r  r   r  r  rd   rO  r  r  r*   r*   r+   rnn_relu_input  r  r  c	                 C   T   | d}	t||d}t| |	||||||dtt|ttjd
\}
}|
t|dfS Nr   Fr  r  )	r  r  r  r   r  r  rd   rO  r  datar  r  r  r  r  r  r  r  r  r   r  r*   r*   r+   rnn_relu_data  &   
r  c	                 C   r  r  )	r  r  r  r   r  r  rd   r   r  r  r*   r*   r+   rnn_tanh_data  r  r  c                 C   s   t ||||  }|d|}|d  }	|d  }
|d  }|d  }|
| |	|  }||  }|d u r;|nt ||d }||fS )Nr  r   r"   r#   r   r   r  chunkr   r   )r  r  r  r  r  	hr_weight	chunk_dimgateschunked_gatesin_gateforget_gate	cell_gateout_gater  r  r*   r*   r+   	lstm_cell*  s   r  c              
   C   s   |d }|d }|r|d nd }|r|d nd }t |dkr"|d nt |dkr,|d nd }	|d d}
|d d}t| ||}|rJ|dn|}g }|D ]} t| |
||||	dd\}
}||
 qP|rk|  t	|d}||

d|
dffS )Nr   r"   r#   r   r  r  r  )rV  rR   r   r  r  r  r  r  rd   r  r  )r  r  r  r  r  r  r  r  r  r  r  r  r  r  r   r*   r*   r+   one_layer_lstm8  s$   *r  c              
   C   s
  |d }|d }|r|d nd }|r|d nd }	t |dkr"|d nt |dkr,|d nd }
g }g }|r8|d n|d }t| t|}|rM|d d d }|d }|d }|dd||dd|}}|D ]l} | jd }t| ||} ||k r||d||| |d||| f |dd||dd|}}||krt	||d||| fd}t	||d||| fd}t
| ||||	|
dd\}}|}|| qf|r|  ||f}n|||f |  t| \}}t|dt|df}t|d}||fS )	Nr   r"   r#   r   r  r  rP   r  )rV  rd   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  r  orig_hxorig_cxr  r  r  r  hidden0hidden1r   r*   r*   r+   one_layer_lstm_dataS  s\   *

r   c                 C   s   dd }|| ||rt S tS )a*  Check whether we could use decompose lstm with mkldnn_rnn_layer.
    All the below conditions need to be met:
        * ``torch._C._get_mkldnn_enabled()`` returns ``True``.
        * All the input args are on CPU.
        * The dtypes of args are either torch.float or torch.bfloat16.
        * Inference.
        * ``has_projections`` returns ``False``.

    Args:
        * input: the input sequence to LSTM
        * hx: a tuple of the input hidden state and cell state ``(h_0, c_0)`` to LSTM
        * params: the weight and bias tensors of LSTM
    c           	      S   s   t j sdS | gt| tt| }dd |D }t|dkr$dS | }|t dkr1dS dd |D }|D ]}|t j	t j
fvrG dS q:| jrMdS |d d|d dk}|r_dS d	S )
NFc                 S      h | ]}|j qS r*   ra  r1   tr*   r*   r+   	<setcomp>      zEselect_one_layer_lstm_function.<locals>.use_mkldnn.<locals>.<setcomp>r"   r=  c                 S   r  r*   r   r  r*   r*   r+   r    r  r   r#   T)rd   r  _get_mkldnn_enabledrI  r   from_iterablerV  poprb  r   bfloat16requires_gradr
  )	r   r  r  r  devicesrb  dtypesr   r  r*   r*   r+   
use_mkldnn  s(   
z2select_one_layer_lstm_function.<locals>.use_mkldnn)r  r  )r   r  r  r  r*   r*   r+   select_one_layer_lstm_function  s   r  c	                 C   s   t |dks
J dt|||d d|d dk}tt|d |d }	t| ||}
t| |	||||||||

\}}tt| }|t|d dt|d dfS )Nr#   lstm expects two hidden statesr   r"   )	rV  r  r
  rI  r  r  r  rd   r  )r   r  r  r  r  r  r  r  rW  r  r  r   r  r*   r*   r+   	lstm_impl  s$   $"r  c	                 C   s   t |dks
J dt|||d d|d dk}tt|d |d }	t| |	||||||dtt|d
\}
}tt| }|
t	|d dt	|d dfS )Nr#   r  r   r"   F)r  )
rV  r  r
  rI  r  r  r   r   rd   r  r  r*   r*   r+   lstm_data_impl  s"   $
"r  c                 C   sr   |  dd}t||| dd}|d |d   }|d |d   }	|d |d |   }
||
 |	 |
 S )Nr   r"   r#   r   )r  r   r  r   r   r  r  r  r  r  r  chunked_igateschunked_hgates
reset_gate
input_gatenew_gater*   r*   r+   gru_cell  s   r  c                 C   s|   t | ||dd}t |||dd}|d |d   }|d |d   }	|d |d |   }
||
 |	 |
 S )Nr   r"   r   r#   r  r  r*   r*   r+   gru_cell_data  s   r  c	                 C   sJ   t ||d}t| |d||||||dtt|td
\}	}
|	t|
dfS )NFr   r  )r  r  r  r   r  r  rd   r  )r  r  r  r  r  r  r  r  r  r   r  r*   r*   r+   gru_impl_data  s   r  c	                 C   sH   t ||d}t| |d|||||||tttd
\}	}
|	t|
dfS )NFr   r  )r  r  r  r   r  r  rd   r  )r   r  r  r  r  r  r  r  rW  r   r  r*   r*   r+   gru_impl7  s   
r  c                 C   :   t |  ||}t|d}t|d}tjj| ||||S Nr   r"   )rw  r
  ry  rd   r  r   _upsample_bilinear2d_aar   r  align_cornersru  r}  scale_hscale_wr*   r*   r+   upsample_bilinear2d_aa_vecU     


r#  c                 C   r  r  )rw  r
  ry  rd   r  r   _upsample_bicubic2d_aar  r*   r*   r+   upsample_bicubic2d_aa_veca  r$  r&  c                 C   s4   t |  ||}|r|nd gt| }t| |||S r6   )rw  r
  rV  _upsample_linear)r   r  r   ru  r}  rx  r*   r*   r+   _upsample_linear_vecm  s   	r(  r   c                 C   s   t | |||gS r6   r'  )r   r  r   r  r*   r*   r+   upsample_linear1d{  s   r*  c                 C   s   t | ||||gS r6   r)  )r   r  r   r  r  r*   r*   r+   upsample_bilinear2d  s   r+  c                 C   s   t | |||||gS r6   r)  )r   r  r   r  r  r  r*   r*   r+   upsample_trilinear3d  s   r,  c                 C   s@   |r|dkr| d |d  S dS |d ur|dkrd| S | | S )Nr"   rb   r   r*   )r  r  r   rj   r*   r*   r+   _compute_scale  s    r-  c                 C   s   |r| | S | |d  d S Nr   r*   )rj   	dst_indexr   r*   r*   r+   _compute_source_index  s   r0  weightsweights_precisionc                 C   sB   t dd t| |D d|d >  }||? }t|ddtjS )Nc                 s   s,    | ]\}}| tj| tj V  qd S r6   )r8   rd   r  )r1   r,  r  r*   r*   r+   r    s    
z%_sum_tensors_uint8.<locals>.<genexpr>r"   r      )_sum_tensorsr  rd   r   r8   r  )r`  r1  r2  r{  r*   r*   r+   _sum_tensors_uint8  s   
r5  c                 C   sJ   t |  }d}t j||jd}d|d|d >   }|dk}||  S )N   ra  r   r"   i   )rd   r  r   rf  rb  r   )r1  
max_weightmax_weight_precision
precisionsvaluesro  r*   r*   r+   _compute_weight_precision  s   r;  c                    s  j d d \}}j dd  }t|tjtjjd\}fddfddtt|||D }tt| \}	g }
t	ddgg  D ]# d d g fd	dt
D  }t|}t|}|
| qKtt
D ]'}|	| |  d
dfddt|
d d d |
dd d D }
qut|
dksJ |
d }t}jjdkr|dk rtj}t|tjsJ |j|d} s| }|S )Nr#   r   c           	         s   t | | |}tj|jdjd}t|| jdd}|j|jd gdg| R  }|tj	}|d j| d d}|||fS )Nra  r   r   r   r   r"   r   )
r-  rd   rf  rb  r8   r0  r   r  r   r  )	inp_sizer  rx  nsqueezescale_factorr  x_f32r2   xp1)r   r   r   r*   r+   
get_values  s   
z$_upsample_linear.<locals>.get_valuesc                    s,   g | ]\}\}}} |||d  | qS r  r*   )r1   r  r<  r  rx  )rA  n_dimsr*   r+   r3     s    z$_upsample_linear.<locals>.<listcomp>r   r"   c                    s(   g | ]} | d kr| n| qS r  r*   )r1   k)r  xp1sxsr*   r+   r3        ( r   rb   c                    s$   g | ]\}}|t ||   qS r*   )rd   r   )r1   v1v2)xscaler*   r+   r3     s    r     r-  )r   rV  rB   rC   r  INT_TO_FLOATrq  r  rI  r   rQ   r   r  r   r  reversedr   r8   r   rb  rB  rd   r2  r/   r   r}  r  round)r   r  r   rx  n_batchr  	inp_sizesrS   r:  xs_f32vsrn  vr  r   r.  r*   )	r  r   r   rA  r   rB  rD  rE  rI  r+   r'    sF   

"


r'  r  r  c                 C   s   | j |j kS r6   r  )r  r  r*   r*   r+   is_same_size  ry   rS  c                 G   rs   r6   )r   rj  )r2   r   rD   r*   r*   r+   _reshape_alias  s   rT  c                 C   rs   r6   )r   rq  )r2   rm  r*   r*   r+   r    ry   r  c                 C   s   t | |||S r6   )r   rQ  )r2   rm  rw   r  r*   r*   r+   r  !  r4  r  c                 C   s   |D ]}|d urt |jt jt jfv dd  qt |jt jkdd  ddlm} ||  dkr@t j	
| |}| |j|S tt|D ]}|| }|d ur^|jd| |d d||< qFt| || |S )Nc                   S   r  Nz3tensors used as indices must be long or int tensorsr*   r*   r*   r*   r+   r|   ,  r  z&_unsafe_masked_index.<locals>.<lambda>c                   S   r  Nz*tensors used as masks must be bool tensorsr*   r*   r*   r*   r+   r|   1  r  r   ry  r"   rX  )rd   r~   r   r  r^  rh  rG  rE  r   _meta_registrationsmeta_index_Tensorr!  r   rQ   rV  r   r
  r   r  r  )r2   ro  rm  fillrq  rE  meta_resultr  r*   r*   r+   rk  &  s*   
rk  c                 C   s   |D ]}|d urt |jt jt jfv dd  qt |jt jkdd  |  dkr.|  S tt	|D ]}|| }|d urP|j
| | | |d d||< q4|| d}tj| ||ddS )	Nc                   S   r  rU  r*   r*   r*   r*   r+   r|   H  r  z5_unsafe_masked_index_put_accumulate.<locals>.<lambda>c                   S   r  rV  r*   r*   r*   r*   r+   r|   M  r  r   r"   rX  Tr  )rd   r~   r   r  r^  rh  r   re  rQ   rV  r   r
  r  r   r  )r2   ro  rm  r:  rq  r  masked_valuer*   r*   r+   #_unsafe_masked_index_put_accumulateB  s(   
$r\  c                 C   sV  |   }d}|dk rd}|d ur,|dkr&dg| }|jd ||< ||}n|}| | } t||k|d}	|	|}
t| ||
| }t||k|d}|tj	j
krb|dkrb| dd}||fS |d ur|| j}t|||
|}t||k|d}| }n	||k | }|tjj
kr| }||fS |tjj
kr| | }||fS )Nr"   r#   r   r*   r   )rM   r   rj  rd   re   rR   gatherr  r!   r'   rw   r!  rd  r   r8   r)   r(   )rv   r   r   r   r   rB  r  r   wr  safe_target_r   r   wsumr*   r*   r+   _nll_loss_forward\  sB   


ra  c                 C   s   |   dkr|   dksJ d|  dksJ d|   dko%|  dk}|s?| jd |jd ks?J d| j d|j d| jd	 }|d u s_|  dkrT| |ks_J d
| d|j t| ||||S )Nr   r#   r  r"   r  r  r  r  rP   z/weight tensor should be defined either for all z7 classes or no classes but got weight tensor of shape: )rM   r   r   ra  )rv   r   r   r   r   r  	n_classesr*   r*   r+   nll_loss_forward  s    	
rc  c                 C   s   t | ||||S r6   )ra  )rv   r   r   r   r   r*   r*   r+   nll_loss2d_forward  s   	rd  Ac                 C   s    |d |  |d  |  |  d S )Nr#   r   r"   r*   r2   re  r*   r*   r+   _upsample_cubic_convolution1  r  rg  c                 C   s(   ||  d|  |  d|  |  d|  S )Nr     r  r*   rf  r*   r*   r+   _upsample_cubic_convolution2  s   (ri  r  c           
      C   s   d}| j t dkrDtj| d|  gdd}tj| d d|  gdd}t||}t||}tj|dd\}}tj|dd\}}	|||	|fS t| d |t| |td|  |td|  |fS )Ng      r=  rb   r   rz   r   )rb  rd   r  ri  rg  r  )
r  re  tt1tt2w03w12r  r  r  r  r*   r*   r+    _upsample_get_cubic_coefficients  s   

rn  coeffstsc                 C   s    t |}tdd t| |D S )Nc                 s       | ]	\}}|| V  qd S r6   r*   r1   rk  rl  r*   r*   r+   r    r  z+_upsample_cubic_interp1d.<locals>.<genexpr>)rn  r4  r  )ro  rp  coeffs2r*   r*   r+   _upsample_cubic_interp1d  s   rt  c                 C   s   t tj| S r6   )r   rd   add)rp  r*   r*   r+   r4    s   r4  	num_stepsc                 C   sB   | dkrt jd||dS |s| d |  nd}t j| || ||dS )Nr"   r   rc  )stepsrb  r   )rd   r  linspace)rv  r   r   rb  r  r*   r*   r+   _linspace_from_neg_one  s   ry  thetahr^  c           	      C   s   | j }| j}t||||d|d}t|||||dd}tjd||d}tjjj|dddd}tjjj|dddd}tjjj|d	ddd}|| | S )
Nr"   )r"   r"   r"   r  )r   r#   constantr   r  r  rw   r"   r"   )r#   r   	r   rb  ry  rj  rd   rg  r  r  r  )	rz  r{  r^  r   r   rb  grid_xgrid_ygrid_oner*   r*   r+   _make_base_grid_4d  s   r  r  c                 C   s   | j }| j}t||||dd|d}t||||d|dd}t|||||ddd}	tjd||d}
tjjj|dddd}tjjj|dddd}tjjj|	d	ddd}	tjjj|
d
ddd}
|| |	 |
 S )Nr"   )r"   r"   r"   r"   r  )r   r   r|  r   r}  r  r#   r"   )r   r   r  )rz  r  r{  r^  r   r   rb  r  r  grid_zr  r*   r*   r+   _make_base_grid_5d  s   r  c           	      C   sL   |\}}}}t | |||d}|ddd| jd d}||||dS )Nr   rP   r   r"   r/  r#   )r  rj  r4  rR   r   )	rz  r
  r   r  rS   r{  r^  	base_gridgridr*   r*   r+   _affine_grid_generator_4d  s    r  c           
      C   sR   |\}}}}}t | ||||d}|ddd| jd d}	|	||||dS )Nr  rP   r  r"   r/  r   )r  rj  r4  rR   r   )
rz  r
  r   r  rS   r  r{  r^  r  r  r*   r*   r+   _affine_grid_generator_5d  s    r  c                 C   s@   t t|dv dd  t|dkrt| ||dS t| ||dS )Nr:  c                   S   r  )NzCaffine_grid_generator needs 4d (spatial) or 5d (volumetric) inputs.r*   r*   r*   r*   r+   r|   '  r  z'affine_grid_generator.<locals>.<lambda>r  r  )rd   r~   rV  r  r  )rz  r
  r   r*   r*   r+   affine_grid_generator!  s   
r  r  interpolation_modepadding_mode_expand_gridc                    sJ  t dv fdd t dv fdd dtdtdtffdd	dtd
tdtdtfdddtdtdtffdddtdtdtffdd}j\ |j\}}|dkscJ ru|d| d}dtdtdtffddt jjddddt j jdd dddtdtdtdt	f fdddtdtdtffdd
|d  }	|d! }
d"kr1||	}||
}|
 |
 d }}d }}||}}|| ||  }|| ||  }|| ||  }| |  }t
fd#d$|f|||f|||f|||ffD S dkrN||	}||
}| }| }
||dS |	}|
}|
 |
 | | }sud|d}dtdtdtf
fd%d&d'tdtffd(d)	t	fd*d$td+D }t||S ),N)r   r"   r#   c                      r  )NzInvalid interpolation mode r*   r*   )r  r*   r+   r|   @  r  z"_grid_sampler_2d.<locals>.<lambda>c                      r  )NzInvalid padding mode r*   r*   )r  r*   r+   r|   C  r  coordsr
  rN   c                    s0    r|d d n|d }|d d }| | | S r.  r*   )r  r
  r   ofsr  r*   r+   unnormalizeF  s   z%_grid_sampler_2d.<locals>.unnormalize	twice_low
twice_highc                 S   sv   ||kr	t | S |d }|| d }| |  }t ||}||  jt jd}t |d@ dk|| || | S )Nr#   r   r"   r   )rd   r   r   fmodfloorr8   int8re   )r  r  r  
coords_mincoords_spancoords2extraflipsr*   r*   r+   reflect_coordinatesQ  s   
z-_grid_sampler_2d.<locals>.reflect_coordinatesc                    sf   dkr| S dkrt | d|d S  r | dd|d  }n
| dd| d }t |d|d S )Nr   r"   r#   rP   r   )r  r
  coords_reflected)r   r  r  r*   r+   compute_coordinates]  s   z-_grid_sampler_2d.<locals>.compute_coordinatesc                    s   | |} ||S r6   r*   )r  r
  	coords_un)r  r  r*   r+   compute_source_indexi  s   

z._grid_sampler_2d.<locals>.compute_source_indexr#   r"   rE  ysc                    s,   t d| kt | k t d|k| k S rn   rd   ri  )rE  r  )iHiWr*   r+   in_bounds_condy  s   $z(_grid_sampler_2d.<locals>.in_bounds_condra  wsc                    sN   | |r	nd t  fdd| jtjd|jtjd|fD S )Nr"   c                 3   s*    | ]}t |d  V  qdS r  )rd   re   rj  r  )rV  r  r  oHoWr*   r+   r    rx  z1_grid_sampler_2d.<locals>.clip.<locals>.<genexpr>r   )r  r8   rd   r  )rE  r  r  )rW  rV  r  r  r  r  )r  r  r+   clip  s
   
z_grid_sampler_2d.<locals>.clipixiyc                    s&   | ||\}}} ||f | S r6   r*   )r  r  r^  idx_xidx_yw_)C_idxN_idxr  r  r*   r+   get_summand  s   z%_grid_sampler_2d.<locals>.get_summand).r   ).r"   r   c                 3   s"    | ]\}}} |||V  qd S r6   r*   )r1   r  r  r^  )r  r*   r+   r    s
    

z#_grid_sampler_2d.<locals>.<genexpr>c                    s     | } |}||dS rX   r*   )r  r  r2   rW   )r  r  r  r  r*   r+   get_value_bounded  s   

z+_grid_sampler_2d.<locals>.get_value_boundedr  c                    sF   | d  } d | | d | d |f}t |S )Nr"   r#   )rt  )r  iy_ofscs)r  ix_nwiy_nwtxr*   r+   	get_coeff  s   
z#_grid_sampler_2d.<locals>.get_coeffc                 3       | ]} |V  qd S r6   r*   )r1   r  )r  r*   r+   r    r  r  )rd   r~   r   r^  r   rj  rd  rf  rb  r   r  r4  rM  rR   r  rQ   rt  )r  r  r  r  r   r  r  rS   twor2   rW   r  r  ix_neiy_neix_swiy_swix_seiy_sew_nww_new_sww_se
ix_nearest
iy_nearesttyro  r*   )rW  r  rV  r  r  r  r   r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r+   _grid_sampler_2d/  sx   
 ( 




	





 

r  c                 C   s   t | ||||dS )N)r  r  r  r   )r  )r  r  r  r  r   r*   r*   r+   grid_sampler_2d  s   
r  c                    s`   t   dko dk fdd t  ddk fdd   jddS )Nr#   r"   c                      s   d    d   S )Nzmatrix @ vector expected, got r\  rz   r*   rv   rQ  r*   r+   r|     r]  zmv.<locals>.<lambda>r   c                      s*   d  d d  d d d dS )Nzsize mismatch, got input (r   r2   r"   z), vec (r  rS  r*   r  r*   r+   r|     s   * rz   )rd   r~   rM   r
  r   r  r*   r  r+   rR    s   rR  c                 C   sd   |d ur|d | d }d| |  |t |   }nd| |  t |  }|d ur-|| }t||S rX   )r   
logsigmoidr   )rv   r   r   
pos_weightr   
log_weightr   r*   r*   r+    binary_cross_entropy_with_logits  s   
r  tensor1tensor2is_outc           	         s   | j |j kr
| |fn|| f\}}ddlm  |j dkr |j dks"dS |jr)|s)dS | j dkr0dS  | dkr:dS |j}| }dg}t|dd  D ]}|||d   qLt	 fd	d
t
|tt||D S )Nr   ry  r   r#   FTr"   rP   c                 3   s*    | ]\}}} |d kp||kV  qdS r  r*   )r1   r   r   r
  ry  r*   r+   r  !  s
    
zshould_fold.<locals>.<genexpr>)rQ  rG  rE  r
  r   r   rJ  rL  r  r   r  rI  )	r  r  r  t1t2t1_shape	t1_strideexpected_strider
  r*   ry  r+   should_fold  s(    

r  )pass_is_out)r  c                C   sx  |   }|  }|dkr|dksJ |dkr |dkr t| |S |dkr.|dkr.t| |S |dkrD|dkrDttt| d|dS |dkrR|dkrRt| |S t| ||r||k}|ra|jn| }|sg|n	|dkro| 	 n| }|j
}t|d d }	ttj|	}
|  dk}|r|	|j
d  ||
|d }|rtjj|||	}|r|j S |S tjj|||	S |dkr|dkr|dkr| dnd}| d}| j
d d }|dkr|dn|d}|dkr|dnd}g }t|d D ]
}||| q|dkrA|dkrA|d |d krA|d dkr.| jr.t| d|S |d dkrA|jrAt| |dS tt||}|||g }t|}| ||||}|dk}|rv||g }||||d}n|||g }|||||}|}	|dkr|	| |dkr|	| |r||d|	S |||	S tddd	  d S )
Nr   r"   r#   rP   r/  r   Fc                   S   r  )Nz/both arguments to matmul need to be at least 1Dr*   r*   r*   r*   r+   r|     r  zmatmul.<locals>.<lambda>)rM   rd   dotrR  r  rJ  rR   r  r4  r  r   rI  r   r$  r   r  r  r  r   _unsafe_viewr}  r
  rQ   r
  r3  broadcast_shapesr  rd  bmmrj  r~   )r  r  r  dim_tensor1dim_tensor2r  r  r  sizes_1output_shapefolded_dim1t2_is_matrix	t1_foldedr{  r  m1batch_tensor1m2r)  batch_tensor2r  expand_batch_portiontensor1_expand_sizeexpand_batch_producttensor1_expanded
vector_rhstensor2_expand_sizetensor2_expandedr*   r*   r+   r3  )  s   	










r3  r!  r"  c                    s  j \}}t|d ||}t|d ||}tjtjjd\}}tj|d jdj	|d}	tj|d jdj	|d}
t
||
|}t
||	|}|d}| }| }|| dd}|| dd}|	tj}|	tj}|d ||d |d	 f}|d ||d |d	 ft|t|}d
\jtjkrtt|fddD fdd|D }fddfdd t fdd|D }jtjkrd usJ t||}ntdd t||D }t}|j|d}|S )Nr   r"   r   ra  r   rP   r   rb   r#   r  c                    .   g | ]}|d  >  t |d  t jqS r"   r   rd   r   r8   int16r1   r^  )weights_precision_xr*   r+   r3          z.upsample_bicubic2d_default.<locals>.<listcomp>c                    r  r  r  r  )weights_precision_yr*   r+   r3     r  c                    s<   t | d d }t |dd }td d ||g}|S r  )rd   r   r   r  )r  rE  y_idxx_idxrR  )in_hin_wr   r*   r+   load_bounded  s   z0upsample_bicubic2d_default.<locals>.load_boundedc                    sT   t  fddD }jtjkrd usJ t|S tdd t|D S )Nc                 3   s    | ]} |V  qd S r6   r*   )r1   x_ofs)r  rW   r*   r+   r    rT  zCupsample_bicubic2d_default.<locals>.get_x_interp.<locals>.<genexpr>c                 s   rq  r6   r*   rr  r*   r*   r+   r    r  )r  r   rd   r  r5  r4  r  )rW   src_x)r   ixs_ofsr  r  	weights_x)rW   r+   get_x_interp  s
   z0upsample_bicubic2d_default.<locals>.get_x_interpc                 3   r  r6   r*   )r1   y_ofs)r  r*   r+   r    r  z-upsample_bicubic2d_default.<locals>.<genexpr>c                 s   rq  r6   r*   rr  r*   r*   r+   r    r  r-  )r   r-  rB   rC   r  rK  rd   rf  rb  r8   r0  rR   r  r   r  rn  r   r  r;  r  r5  r4  r  r   r}  )r   r  r   r!  r"  rS   h_scale_factorw_scale_factorr   r  r  x_floaty_floatr2   rW   yscalerI  iys_ofs	weights_ysrc_yr   r.  r*   )	r  r  r  r   r  r  r  r  r   r+   upsample_bicubic2d_default  sR   




r  c                 C   s   t t|t| dkdd  |d u r2|d usJ ttttf tdd t| jdd  |D }|r6|nd\}}t	| ||||S )Nr"   c                   S   r  )Nz:Must specify exactly one of output_size and scale_factors.r*   r*   r*   r*   r+   r|     r  z(upsample_bicubic2d_vec.<locals>.<lambda>c                 s   s$    | ]\}}t t|| V  qd S r6   )r   r   )r1   r^  rj   r*   r*   r+   r    s
    
z)upsample_bicubic2d_vec.<locals>.<genexpr>r#   r  )
rd   r~   rh  r	   r   r^  r  r  r   r  )r  r  r   ru  r!  r"  r*   r*   r+   upsample_bicubic2d_vec  s   
r  c                        fdd}t  ||S )Nc                    s4   t j|  ||  jd}|d |d |    S )Nra  r"   )rd   rf  rb  r   r   middler   dim_idxr  r*   r+   rn    s   z_reflection_pad.<locals>.idx_reflection_or_replication_padr  r  rn  r*   r  r+   _reflection_pad     r  c                    r  )Nc                    s*   t j|  ||  jd}t |d|d S )Nra  r   r"   )rd   rf  rb  r   r  r  r*   r+   rn  -  s   z_replication_pad.<locals>.idxr  r  r*   r  r+   _replication_pad'  r  r  idx_fnc                    s   t d  t|   d  d fv  fdd | j  d  }|    } fddt D } fddt D }| }t D ]}d g|  }	||| || || |	|| < t||	}qFt	|}
|j
|
d}|S )	Nr#   r"   c                      s    d  d d  d d  dS )Nreflection_padzd requires r"   zD or r#   zD inputr*   r*   rz   r*   r+   r|   @       z0_reflection_or_replication_pad.<locals>.<lambda>c                        g | ]}d  d |   qS r  r*   r8  rM   r  r*   r+   r3   E  r  z2_reflection_or_replication_pad.<locals>.<listcomp>c                    $   g | ]}d  d |  d  qS r  r*   r8  r  r*   r+   r3   F  r^  r-  )rV  rd   r~   rM   r   rQ   r   r  rB   r   r}  )r  r  r  	inp_shapenc_dimpadding_leftpadding_rightr   r  rn  r.  r*   r  r+   r  8  s"   
 
r  c                    s\  t d dd |j d  D fddtD fddtD g }t|jD ]}dg|j }d||< |tj|j| |jd| q2|d    | d  
d	d
 
fddtD 
fddtD }
fddtD }fddtD 	t	
tj	fddtD }t|  d}	 fdd}
tjdd tD  D ]f}|tdg krqg }g }tD ]K}|| dkr| }	| }n0|| dkr|| }
| d| f}n|| dkr|| }
| | |  | d f}|| || q|
|	||}	q|	S )Nr#   c                 S   s   g | ]}|d  qS r  r*   )r1   r{  r*   r*   r+   r3   [  rA  z,_reflection_pad_backward.<locals>.<listcomp>c                    r  r  r*   r8  r  r*   r+   r3   ]  r  c                    r  r  r*   r8  r  r*   r+   r3   ^  r^  r"   rP   ra  c                 S   s   | \}}}t ||k||kS r6   r  )index_ranger  lbubr*   r*   r+   index_range_conditioni  s   
z7_reflection_pad_backward.<locals>.index_range_conditionc                    s   g | ]
}|  |  qS r*   r*   r8  r   xyzr*   r+   r3   w  r  c                    s   g | ]
} | |  qS r*   r*   r8  r&  r*   r+   r3   x  r  c                    s(   g | ]}d  |  |  |  qS r  r*   r8  )dhwr   r'  r*   r+   r3   y  rF  c                    s.   g | ]} | d | |  |  fqS r  r*   r8  )centerr(  r   r!  r*   r+   r3   }  s    "c                    s   g | ]} | qS r*   r*   r8  )r%  range_cr*   r+   r3     r]  r   c                    st   t D ]}|| d || d k }t|tr|r|   S qttjfdd|D }t| | d}| | S )Nr#   r"   c                    s   g | ]} |qS r*   r*   )r1   r"  )r%  r*   r+   r3     rA  z@_reflection_pad_backward.<locals>.accumulate.<locals>.<listcomp>r   )rQ   r/   rh  rJ   r   r   ri  rk  )r   r   index_rangesr  upper_less_than_lowerr  g)r  rM   rh   r%  r*   r+   r    s   z,_reflection_pad_backward.<locals>.accumulatec                 S   s   g | ]}g d qS ))rP   r   r"   r*   r  r*   r*   r+   r3     rA  r   )rV  r   rQ   rQ  r  rd   rf  rb  rj  rJ   r   r   ri  rk  	itertoolsr   r  )rh   r2   r  rm  r  
view_shapeleft_reflectright_reflectr  r   r  areaoutsr+  r   r"  r*   )r  r)  r(  rM   rh   r%  r  r   r!  r*  r'  r+   _reflection_pad_backwardT  sT   $
"
r4  r   r   r   c                C   s(   t j| ||d}t j| ||d}||fS )Nr   )rd   aminr  )rv   rM   r   r5  r  r*   r*   r+   aminmax  s   r6  r   c                C   s"   t jtt| d| |||dS )Nr   r   )r   r   rd   re   isnan)rv   rM   r   r   r*   r*   r+   nansum  s   "r8  r   r  rb  r  r  c             	   C   s   t jjd| d||||dS )Nr   r"   r9  r   rf  
start_step)r@  r   r  rb  r  r*   r*   r+   arange_default     
r<  c             	   C   s   t jj| |d||||dS )Nr"   r9  r:  )r?  r@  r   r  rb  r  r*   r*   r+   arange_start  r=  r>  c                  O   s   ddl m} || i |S )Nr   )out_dtype_dense)!torch._higher_order_ops.out_dtyper?  )rD   rE   r?  r*   r*   r+   out_dtype_decomp  s   rA  marginc           	         s  t t jd jd  t |dkp|dkdd  t jdko, dkfdd t jdko? kfdd d urdt t jdko\  k fdd dt jdd	}||  }|	d}|dkr|n|| }d ur|  }t j
 jd
}t |k|d}|tjjkr| S |tjjkr| |jd  S |jddS )Nr   r"   r#   c                   S   r  )Nz only p == 1 and p == 2 supportedr*   r*   r*   r*   r+   r|     r  z#multi_margin_loss.<locals>.<lambda>c                      r*  NzMExpected non-empty vector or matrix with optional 0-dim batch size, but got: r  r*   r;  r*   r+   r|     r+  c                         d  dj  S )Nz#inconsistent target size, expected r  r  r*   )nframer   r*   r+   r|     r  c                      rD  )Nz#inconsistent weight size, expected r  r  r*   )rM   r   r*   r+   r|     r  rU  ra  rz   )rd   
atleast_2d
atleast_1dr   r~   rQ  r   rR   r]  r5  rf  rb  re   r!   r(   rw   r   r)   r   )	r   r   r)  rB  r   r   urf   rn  r*   )rM   r   rE  r   r   r+   multi_margin_loss  sB   







rI  	is_targetc                    s  | j  |j t| } t|}| j d }tt dko |dk fdd ttdko2 k fdd tj||jd}|dk}tjt|||dd	d
}||k }t||d}tj	| d|d}	t||d}
tj
||
jddkdd}d|	jjdd |  }|d}|| }t|d|}|tjjkr|jdd }n|tjjkr| }n|jdd}|| j}||fS )Nr"   r#   r   c                      r  rC  r*   r*   )orig_input_shaper*   r+   r|      r  z0multilabel_margin_loss_forward.<locals>.<lambda>c                      s   d d  S )Nzinconsistent target size: z for input of size: r*   r*   rK  orig_target_shaper*   r+   r|   $  r}   ra  rP   Tr   rU  rz   rb   )r   rP   )r   rd   rF  r~   rV  rf  rb  r5  re   r]  anyrR   Tr5  r!   r(   rw   r   r   r)   r8   r   r  )r   r   r   rM   rn  is_endend_idxtarget_masktidx0rH  tidx1rJ  rf   r*   rL  r+   multilabel_margin_loss_forward  s@   





rU  )	attn_maskrj   querykey	dropout_p	is_causalrV  c          
   
      s   j }ttfdd t dko# dko# dkfdd t dk fdd tjd jd koMjd jd kdd  tjj| |d |d	\}}	|	d
dddj
tjd	dd
dd}||	fS )Nc                      r*  )Nz-query must be FP32, FP64, BF16, FP16 but got r   r*   )rW  r*   r+   r|   ^  r+  z<scaled_dot_product_flash_attention_for_cpu.<locals>.<lambda>r  c                      s"   d   d    d   S )Nz,q, k, v must be a 4 dimensional tensor, got r\  rz   r*   )rX  rW  rw   r*   r+   r|   b  s   " r   c                      r  )Nz&dropout probability must be zero, got r*   r*   )rY  r*   r+   r|   e  r  r   c                   S   r  )Nz&q, k, v should have the same head sizer*   r*   r*   r*   r+   r|   i  r  )rV  rY  rZ  dropout_maskrj   r#   r   r"   r-  )r   rd   r~   r  rM   r   r   "_scaled_dot_product_attention_mathrZ  r  r}  r2  )
rW  rX  rw   rY  rZ  rV  rj   r   r{  attnr*   )rY  rX  rW  rw   r+   *scaled_dot_product_flash_attention_for_cpuP  s@   
"&
"r^  c                    s   t |  fdd}|S )Nc                     s    | i |}| d  |S rn   )r  )rD   rE   r   outplace_opr*   r+   
inplace_op  s   z$register_inplace.<locals>.inplace_opr   )aten_opr`  ra  r*   r_  r+   register_inplace  s   rc  c                 C   sx   |   s|  st|}t|}t||}t|tjr |dkr$|| }|dkr*|S t|tjr4|dkr8| | } | | S )Nr"   r   )r  rI  r^  rd   r  r/   numbersNumber)rv   batch1batch2r_   ri   r   r*   r*   r+   baddbmm  s   rh  c                 C   s   t j| |ddS )Nr  r  r  )rv   r(  r*   r*   r+   floor_divide  s   ri  c                 C   s   t tj| jdS rX   )rJ   r   r$  r   r   )r  r*   r*   r+   	sym_numel  rh  rj  r   r   c                C   s.   |d u rt jj| g |dS t jj| g ||dS )Nr   rk  )r   r   dim_IntListIntList_out)rv   r   r   r*   r*   r+   sum_default  s   rn  c                 C   sB   t | tjs| S |d u rtj| tt|  S tj| |gS r6   )	r/   rd   r   r   r  dimsrI  rQ   rM   )rv   rM   r*   r*   r+   squeeze_default  s
   rp  c                    s`   t  fddtt| jD }|jtjkrtjnd }| jd|d|d}| ||	|j  |fS )Nc                 3   s    | ]	}| kr|V  qd S r6   r*   r8  rz   r*   r+   r    r  z)_weight_norm_interface.<locals>.<genexpr>r#   T)r   r   )
r  rQ   rV  r   r   rd   r	  r   r   r8   )rR  r-  rM   keep_dim
norm_dtyper   r*   rz   r+   _weight_norm_interface  s    rs  assume_uniqueinvertc                C   sp   t | tjstj| |jd} t |tjstj|| jd}| dt|  d k r0t| ||dS t| |||dS )Nra  g      $@g(\?rv  rt  )	r/   rd   r   r  rb  r   r0  isin_defaultisin_sorting)elementstest_elementsru  rv  r*   r*   r+   isin  s   r|  )r  c                C   sP   |d u rt j|  t j| jd}nt j|  |t j| jd}|| k | j}|S Nr  )r  r   rb  rd   randr
  r   rb  r8   r   )rv   r  raw_pr)  r*   r*   r+   	bernoulli  s   r  c                C   sP   |d u rt j|  t j| jd}nt j|  || j| jd}||k | j}|S r}  r~  )rv   r)  r  r  r*   r*   r+   bernoulli_p  s   r  rw  c                C   sr   |   dkrtj| tjdS | jg | jd|j R  }|s#||k}n||k}ttd|j d d}|j	|dS )Nr   r   r  rP   r"   rz   )
r   rd   
empty_likerh  rj  r   rQ  r  rQ   rN  )rz  r{  rv  r2   cmprM   r*   r*   r+   rx    s   
rx  c                C   s   |   }|  }|rIt||g}tj|dd\}}|dd  |d d k}	t|	ddgd}	|r5|	 }	t|	}
|
d||	}
|
d|   S t|\}}t	||}t
|| k |d}|| |k}|rm| n|}|| jS )NT)stabler"   rP   r   F)r  rd   r  sortr  logical_notr  ri  r   searchsortedre   r  r   )rz  r{  ru  rv  elements_flattest_elements_flatall_elementssorted_elementssorted_orderduplicate_maskro  sorted_test_elementsrS   rn  test_idxr  r*   r*   r+   ry    s$   
ry  c                 C   s   |  d}|| S rO   )r  )rv   rq  	flattenedr*   r*   r+   take:  s   
r  c                 C   s2   |d u rt j}|t jkrt|}tj| |j|dS r  )rd   r2  preserve_formatr   r   resizer   )rv   r(  r.  r*   r*   r+   	resize_asA  s
   
r  )F)r   r6   r  )r   NNr"   )rP   FFr  r  r~  )r"   r"   F)Fr   )r   rb   N)r   r"   Nr  )NNN)r   r   FT)r   r   Fr  )r   F(  rJ   r.  rd  r$  rK  enumr   r   r   r   r   typingr   r   r	   r
   r   r   r   r   rd   torch._meta_registrationstorch._primsr  rn  torch._prims_common_prims_commonrB   torch.nn.functionalr  r  r   r   r   r   torch._decompr   r@  r   r   r   r   r   r   torch._prims_common.wrappersr   r   r   r   torch.utilsr   r@   torch.utils._pytreer   r  DispatchKeyr    str__annotations___opsr  r   r!   r  rh  rL   r  compute_only_pw_cast_for_opmathpw_cast_for_opmathrK  pw_cast_for_int_to_realr^  rT   r\   r^   rg   r   rr   rY  Scalarrx   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r(   rw   r   _safe_softmaxr   r   r   rZ  rU   r   r   r   r   r  r  r  r  r#  r%  r&  r'  r*  r=  rD  slicerW  r_  rC  rs  rx  rz  r  r  r  r  r  r  r  r  r  py_implCompositeImplicitAutogradAutogradr  r  r
  r  r  r  r"  r$  r&  r)  r*  r0  r3  r5  r/  rC  r<  rE  rK  rP  rS  rn  rr  rt  r  r  r  r  r  unsafe_chunkr  r  r  no_statsr  r  r  r  r  r  _fused_dropoutr  r  rb  r.  r:  lift
lift_freshr  r  r  r  r  r  r  r  _adaptive_avg_pool2dr   r)  r9  r>  rE  rF  rC  ra  rg  ri  re  rk  rp  	Generatorrq  rw  ry  r  rQ  r  r  r~  _upsample_nearest_exact1dr  r  r  r  r  r{  r  r  r  r  r  r  r  r  r  r  rnn_tanhr   r  rnn_relur  r  r  r  r  r  r   r  lstmr  r  r  r  grur  r  r  r#  r%  r&  r+  r,  r*  r(  r-  r0  r5  r;  r'  rS  rT  r  r  r  rk  r\  ra  rc  rd  rg  ri  rn  rt  r4  ry  r  r  r  r  r  r  r  rR  r  r  r3  upsample_bicubic2dr  r  reflection_pad1dreflection_pad2dreflection_pad3dr  replication_pad1dreplication_pad2dreplication_pad3dr  r  reflection_pad1d_backwardreflection_pad2d_backwardreflection_pad3d_backwardr4  r6  r8  rf  r  r  r<  r?  r>  rA  rI  rU  +_scaled_dot_product_flash_attention_for_cpur^  rc  rh  ri  rj  r   rn  r  rM   rp  rs  r|  r  r)  r  rx  ry  r  r  addbmm_addbmmaddmm_addmv_baddbmm_fill_gelu_rN  
hardswish_	hardtanh_hardtanhhardsigmoid___iand____and____ilshift__
__lshift__rP  rQ  index_reduce_index_reduce__ior____or____irshift__
__rshift____ixor____xor__leaky_relu_
leaky_relulogit_logitrelu_rO  renorm_renormround_rM  scatter_r  scatter_add_scatter_addscatter_reduce_scatter_reducesilu_r*   r*   r*   r+   <module>   sz  
(

$ 
 

 
	




  *!	
9

'"
	P`
 
	
%


(


(
 00

	

W	

	
P
	
R		#

	

	



,
	

	

e	
%	$f-7
("$$





 




  		

.2
)


  ?
2
					

	
I"

5


 (
.$$


* 
'


*!
w
S


0


0




W


,

<

	D	

"
	&

