o
    hE                     @  s  d dl mZ d dlmZ d dlmZ d dlm  mZ d dl	m
Z
 d dlmZmZmZmZmZmZmZmZmZ d dlmZ d dlmZmZmZmZmZmZmZ d d	lm Z  ercd d
l!m"Z" d dlm#Z# eddG dd dZ$eddG dd dZ%d5ddZ&d6ddZ'eddG dd dZ(e(d dej)de(dd ej*dgZ+d7d(d)Z,ed8d*d+Z-eddG d,d- d-Z.ed8d.d/Z/d9d1d2Z0ed8d3d4Z1dS ):    )annotations)	dataclass)TYPE_CHECKINGN)	translate)		BaseCTypeBindingCTypeExpr
NamedCTypeopmath_tscalar_tStructuredImplSignatureVectorizedCType)with_native_function)ArgumentBaseTyBaseTypeDispatchKeyNativeFunctionsGroup
ScalarTypeUfuncKey)
OrderedSet)Sequence)UfunctorBindingsT)frozenc                   @  sb   e Zd ZU ded< ded< ded< dd	d
ZdddZdddZdddZdddZdddZ	dS )UfunctorSignaturer   gz
int | Nonescalar_tensor_idxstrnamereturnr   c                 C  s   t j| j| jtdS )N)r   r   )ufuncufunctor_argumentsr   r   r   self r%   g/var/www/html/construction_image-detection-poc/venv/lib/python3.10/site-packages/torchgen/dest/ufunc.py	argumentsI   s   
zUfunctorSignature.argumentslist[Binding]c                 C  s   dd |   jD S )Nc                 S  s   g | ]}| |j d qS )_)renamer   .0br%   r%   r&   
<listcomp>P   s    z,UfunctorSignature.fields.<locals>.<listcomp>)r'   ctorr#   r%   r%   r&   fieldsN   s   zUfunctorSignature.fieldsr   c                 C  s   t tS N)r   r   r#   r%   r%   r&   returns_typeR   s   zUfunctorSignature.returns_typec                 C  s   d dd |  D S )N
c                 s  s$    | ]}|j  d |j dV  qdS ) ;N)typer   )r,   fr%   r%   r&   	<genexpr>X      " z0UfunctorSignature.decl_fields.<locals>.<genexpr>)joinr0   r#   r%   r%   r&   decl_fieldsW   s   zUfunctorSignature.decl_fieldsc                 C  sL   d dd |  jD }d dd |  jD }| j d| d| dS )N, c                 s      | ]}|  V  qd S r1   declr,   ar%   r%   r&   r8   [       z5UfunctorSignature.inline_defn_ctor.<locals>.<genexpr>c                 s  s$    | ]}|j  d |j  dV  qdS )z_()Nr   r@   r%   r%   r&   r8   ^   r9   (z) : z {})r:   r'   r/   r   )r$   args_strinit_strr%   r%   r&   inline_defn_ctorZ   s   z"UfunctorSignature.inline_defn_ctorc                 C  s2   d dd |  jD }|    d| dS )Nr<   c                 s  r=   r1   r>   r@   r%   r%   r&   r8   b   rB   z/UfunctorSignature.decl_apply.<locals>.<genexpr>z operator()(z) const)r:   r'   applyr2   cpp_type)r$   rF   r%   r%   r&   
decl_applya   s   zUfunctorSignature.decl_applyN)r    r   r    r(   )r    r   r    r   )
__name__
__module____qualname____annotations__r'   r0   r2   r;   rH   rK   r%   r%   r%   r&   r   C   s   
 




r   c                   @  s:   e Zd ZU ded< ded< ded< dd	d
ZdddZdS )UfuncSignaturer   r   r   r   r   	compute_tr    r(   c                 C  s   t j| j| jdS )N)rS   )r!   ufunc_argumentsr   rS   r#   r%   r%   r&   r'   l   s   zUfuncSignature.argumentsctxSequence[Binding | Expr]c              	   C  ,   | j  dddd t||  D  dS )NrE   r<   c                 s      | ]}|j V  qd S r1   exprr@   r%   r%   r&   r8   p       z&UfuncSignature.call.<locals>.<genexpr>rC   r   r:   r   r'   r$   rU   r%   r%   r&   callo      ,zUfuncSignature.callNrL   )rU   rV   r    r   )rN   rO   rP   rQ   r'   r^   r%   r%   r%   r&   rR   f   s   
 
rR   r   r   r    boolc                 C  s"   t dd | jjjjD }|dkS )Nc                 s  s    | ]
}|j  rd V  qdS )   N)r6   is_tensor_liker@   r%   r%   r&   r8      s    

z<eligible_for_binary_scalar_specialization.<locals>.<genexpr>   )sum
functionalfuncr'   flat_non_out)r   num_tensorsr%   r%   r&   )eligible_for_binary_scalar_specialization   s   
ri   ?tuple[dict[ScalarType, dict[UfuncKey, UfunctorSignature]], str]c                 C  s  i }g }| j j}tjdtjdtjd i}t| r tjtjtjg}ntjg}tjtjfD ]}||vs8J d| dq*|D ]}||v r^t| || || jd}|| j	D ]
}||
|i |< qRq;d }	t }
tjtjfD ]#}||vrpqi|	d u rz|| j}	n|	|| jksJ d|
|| j	O }
qi|	d usJ | d|	 }t| || |d}|
D ]
}||
|i |< qt| d|	 ttd	}| | j }|d
|j d|  d|  d|  d|| d q;|d|fS )Nra   r   zcannot use z on non-binary function)r   r   z0ScalarOnly and Generic must have same ufunc namer)   ufunc::r   rS   z%
template <typename scalar_t>
struct z3 {
  using opmath_t = at::opmath_type<scalar_t>;
  z
  z
  __device__ z {
    return z	;
  }
};
r3   )outufunc_inner_loopr   CUDAFunctorOnSelfCUDAFunctorOnOtherCUDAFunctorri   r   r   supported_dtypes
setdefaultr   
ScalarOnlyGenericrR   r   r   r0   r'   rI   appendr;   rH   rK   r^   r:   )r   ufunctor_sigs	ufunctorsloopsscalar_tensor_idx_lookupkeyskufunctor_sigdtype
ufunc_namerr   lkr   	ufunc_sig	apply_ctxr%   r%   r&   compute_ufunc_cuda_functors   sv   	
r   c                   @  s&   e Zd ZU ded< ded< ded< dS ) BinaryScalarSpecializationConfigint
scalar_idxr   ctor_tensorr   	ufunc_keyN)rN   rO   rP   rQ   r%   r%   r%   r&   r      s   
 r   r$   )r   r   r   ra   otherr~   r   inner_loops!dict[UfuncKey, UfunctorSignature]
parent_ctxSequence[Binding]r   c           
      C  s   d}|d7 }t D ]K}|j|vrq||j }|jd }t|}|td| dt|jtt	d d
dd	 t|| jD }	|d
| d|j d|	 d| d	7 }q|tj }d
dd	 t|| jD }	|d|j d|	 d7 }|S )Nz+using opmath_t = at::opmath_type<scalar_t>;zif (false) {}
ra   ziter.scalar_value<opmath_t>(rC   )rZ   r6   r<   c                 s  rX   r1   rY   r@   r%   r%   r&   r8   
      
z0compute_ufunc_cuda_dtype_body.<locals>.<genexpr>zelse if (iter.is_cpu_scalar(z)) {
  z<scalar_t> ufunctor(z);
  iter.remove_operand(z");
  gpu_kernel(iter, ufunctor);
}c                 s  rX   r1   rY   r@   r%   r%   r&   r8     r   z
else {
  gpu_kernel(iter, z<scalar_t>(z
));
}
    )!BinaryScalarSpecializationConfigsr   r   listrv   r	   r
   r   r   r   r:   r   r'   r/   r   r   rq   )
r   r~   r   r   bodyconfigr}   r   rU   ufunctor_ctor_exprs_strr%   r%   r&   compute_ufunc_cuda_dtype_body   sH   







r   c           	      C  s   t | \}}t| t| tj}g }| D ]\}}|d| dt| |||	  d qd
|}t| }d| d|  d|  d|  d|j d| d	|j d
|j d|  d||	  dS )N"
AT_DISPATCH_CASE(at::ScalarType::,
  [&]() {
    
  }
)
r3   z

;
. {
  AT_DISPATCH_SWITCH(iter.common_dtype(), "",
    z
  );
}
REGISTER_DISPATCH(, &z)

 {
  ;
}
)r   r   r!   kernel_namer   CUDAitemsrv   r   r'   r:   StubSignature	type_defndispatch_declkernel_defnr   defndirect_call)	r   rw   rx   sigdtype_casesr~   inner_ufunc_sigsdtype_cases_strstub_sigr%   r%   r&   compute_ufunc_cuda#  sD   

r   c                   @  s   e Zd ZU ded< edddZedddZedd	d
Zd ddZdddZ	dddZ
dddZdddZdddZd!ddZd!ddZdS )"r   r   r   r    r   c                 C     t | jjjjj dS )N_stubr   r   re   rf   r   r#   r%   r%   r&   r   X     zStubSignature.namec                 C  r   )N_kernelr   r#   r%   r%   r&   r   \  r   zStubSignature.kernel_namec                 C  r   )N_fnr   r#   r%   r%   r&   	type_name`  r   zStubSignature.type_namer(   c                 C  s   t | jS r1   )r!   stub_argumentsr   r#   r%   r%   r&   r'   d  s   zStubSignature.argumentsc                 C  s$   |   }dddd |D  dS )Nzvoid(*)(TensorIteratorBase&, r<   c                 s  rX   r1   )r6   r@   r%   r%   r&   r8   i  r[   z%StubSignature.type.<locals>.<genexpr>rC   )r'   r:   )r$   cpp_argsr%   r%   r&   r6   g  s   zStubSignature.typec                 C  s   d| j  d| j dS )NzDECLARE_DISPATCH(r<   rC   )r   r   r#   r%   r%   r&   r   k     zStubSignature.dispatch_declc                 C  s   d| j  dS )NzDEFINE_DISPATCH(rC   rD   r#   r%   r%   r&   dispatch_defnn  s   zStubSignature.dispatch_defnc                 C  s(   d| j  dddd |  D  dS )Nzvoid z(TensorIteratorBase& iter, r<   c                 s  r=   r1   )r   r@   r%   r%   r&   r8   r  rB   z,StubSignature.kernel_defn.<locals>.<genexpr>rC   )r   r:   r'   r#   r%   r%   r&   r   q  s   (zStubSignature.kernel_defnc                 C  s   d| j  d|   S )Nzusing  = )r   r6   r#   r%   r%   r&   r   t  r   zStubSignature.type_defnrU   r   c              	   C  rW   )Nz(device_type(), *this, r<   c                 s  rX   r1   rY   r@   r%   r%   r&   r8   y  r[   z%StubSignature.call.<locals>.<genexpr>rC   r\   r]   r%   r%   r&   r^   x  r_   zStubSignature.callc              	   C  rW   )Nz(*this, r<   c                 s  rX   r1   rY   r@   r%   r%   r&   r8   }  r[   z,StubSignature.direct_call.<locals>.<genexpr>rC   )r   r:   r   r'   r]   r%   r%   r&   r   |  r_   zStubSignature.direct_callNrM   rL   )rU   r   r    r   )rN   rO   rP   rQ   propertyr   r   r   r'   r6   r   r   r   r   r^   r   r%   r%   r%   r&   r   T  s    
 






r   c                 C  sZ   t | }t| t| tj}d|  d|  d|  d|	  d|
|  dS )Nr3   r   z;

r   r   )r   r   r!   r   r   CPUr   r   r   r   r^   r'   )r   r   r   r%   r%   r&   compute_ufunc_cpu  s   r   dict[UfuncKey, UfuncSignature]c                   s|  t j|v sJ | d|  | t jt jhksJ |t j }d }t j|v r-|t j }g }g  |D ]3}t|jtrE|jjtt	j
krEq3|d|j d|j d  td|j t|jjtt q3|d ur|D ]5}t|jtr|jjtt	j
krqm|d|j d|j d  td	|j t|jjttt qmg }g }	| jjjjD ]:}
|
j sq|
jtt	jksJ |t|
jt|
jtt|
d
 |d ur|	t|
jt|
jttt|
d
 qd fdd}d|}|d ur%d| dddd |D  d||| dddd |	D  d|||	 dS d| dddd |D  d||| dS )Nr<   zauto _s_r   z.to<scalar_t>();_s_zauto _v_z$ = at::vec::Vectorized<scalar_t>(_s_z);_v_)r   nctypeargumentr-   r   r    list[Expr | Binding]c                   s   g }|   | |  |S r1   )extend)r-   rrU   r%   r&   with_ctx  s   

z.compute_ufunc_cpu_dtype_body.<locals>.with_ctxr3   z
cpu_kernel_vec(iter,
  [=](c                 s  r=   r1   r>   r+   r%   r%   r&   r8     rB   z/compute_ufunc_cpu_dtype_body.<locals>.<genexpr>z) { return z; },
  [=](c                 s  r=   r1   r>   r+   r%   r%   r&   r8     rB   z; }
);
z
cpu_kernel(iter,
  [=](c                 s  r=   r1   r>   r+   r%   r%   r&   r8     rB   )r-   r   r    r   )r   	CPUScalarr{   	CPUVector
isinstancer   r   r6   r   r   Scalarrv   r   r	   r
   r   r   r   r   re   rf   r'   rg   rb   Tensorr   r:   r^   )r   r~   r   r   scalar_loopvec_loopr   r-   scalar_bindingsvec_bindingsrA   r   body_strr%   r   r&   compute_ufunc_cpu_dtype_body  s    


	(
	


r   c                 C  sv  t | }| jj}i }tjtjfD ]f}g }||v r|| tj|v r-|tju r-|tj tj|v r8|tj |D ];}|| j	D ]3}|tju rMt
t}n|tju rYtt
t}nt||i }	||	vrtt| d|| j |d|	|< qAq:qg }
| D ]\}}	|
d| dt| ||	|  d q}d|
}d|  d|j d	| d
|  d|  d|j d|j dS )Nrk   rl   r   r   r   r3   z
namespace {

r   r   z#
  );
}

} // anonymous namespace

r   z
REGISTER_DISPATCH(r   z)
)r   rm   rn   r   r   r   rv   rt   ru   rr   r   r   r   AssertionErrorrs   rR   r   r   r   r'   r:   r   r   r   r   )r   r   ry   
ufunc_sigsr|   lksr   r~   rS   r   r   r   r%   r%   r&   compute_ufunc_cpu_kernel  sf   







r   )r   r   r    r`   )r   r   r    rj   )
r   r   r~   r   r   r   r   r   r    r   )r   r   r    r   )
r   r   r~   r   r   r   r   r   r    r   )2
__future__r   dataclassesr   typingr   torchgen.api.ufuncapir!   torchgen.api.translater   torchgen.api.typesr   r   r   r	   r
   r   r   r   r   torchgen.contextr   torchgen.modelr   r   r   r   r   r   r   torchgen.utilsr   collections.abcr   r   r   rR   ri   r   r   rp   ro   r   r   r   r   r   r   r   r%   r%   r%   r&   <module>   sR    ,$	 "

S
/0+
]