o
    8} i                    @  sp  U d dl mZ d dlZd dlZd dlZd dlZd dlZd dlmZm	Z	m
Z
 d dlmZmZ d dlmZ d dlmZmZmZmZmZ d dlZd dlm  mZ d dlm  mZ d dlm  mZ d dlm  mZ d dl m!Z! d dl"m#Z# d dl$m%Z% d d	l&m'Z' d d
l(m)Z)m*Z*m+Z+m,Z,m-Z-m.Z.m/Z/ d dl0m1Z1m2Z2m3Z3m4Z4 d dl5m6Z6m7Z7m8Z8m9Z9 d dl:m;Z;m<Z<m=Z=m>Z> d dl?m@Z@ d dlAmBZBmCZCmDZDmEZEmFZFmGZGmHZHmIZImJZJmKZKmLZLmMZMmNZNmOZOmPZPmQZQmRZRmSZSmTZTmUZUmVZVmWZWmXZXmYZYmZZZm[Z[ d dl\m]Z]m^Z^m_Z_m`Z` d dlambZb d dlcmdZdmeZemfZfmgZgmhZhmiZimjZjmkZk d dllmmZmmnZn er,d dlompZp edZqG dd denZre	dddgZsi atdeud< i avdeud< dd&d'Zw		(	)ddd4d5Zxddd6d7Zyejzdd8d9Z{	dd)dd:dd>d?Z|ddCdDZ}ddFdGZ~ddKdLZddRdSZddUdVZddXdYZdd^d_Zdd`daZddbdcZddddeZedfdgG dhdi diZedfdgG djdk dkZedfdgG dldm dmZedfdgG dndo doZedfdgG dpdq dqZe3ddrdsZddvdwZd dzd{ZedfdgG d|d} d}ZdddZdddZdddZdddZdddZdddZdddZe3dddZd	ddZe4d
ddZdddZdddZdddZe!jddddZdddZe!jddddZdddZdddÄZdddńZdddɄZdddӄZdddքZdddڄZdddZdddZdddZdddZedkre  dS dS (      )annotationsN)defaultdict
namedtupleOrderedDict)	dataclassfield)Path)AnyCallableLiteralTYPE_CHECKINGTypeVar)inductor_fallback_ops)cpp)	translate)BindingCppSignatureCppSignatureGroupDispatcherSignature
NamedCTypeNativeSignatureSpecialArgName)method_with_native_functionnative_function_managerwith_native_function with_native_function_and_indices)gen_aoti_c_shim*gen_static_dispatch_backend_call_signatureget_fallback_op_nameget_header_for_aoti) gen_functionalization_definition"gen_functionalization_registration.gen_functionalization_view_inverse_declarationGenCompositeViewCopyKernelgen_all_vmap_plumbing)ArgumentBackendIndexBackendMetadataBaseOperatorNameDEFAULT_KERNEL_NAMESPACEdispatch_device_mapDispatchKeyFRAGMENT_NAMESPACESFunctionSchemais_cuda_dispatch_keyis_generic_dispatch_keyis_ufunc_dispatch_keyis_xpu_dispatch_keyLocationNativeFunctionNativeFunctionsGroupNativeFunctionsViewGroupOperatorNameOptionalType
SchemaKindSelfArgumentSTRUCTURED_DISPATCH_KEYSTensorOptionsArgumentsTypeVariantViewSchemaKind)add_generated_native_functionsgen_composite_functional_kernelgen_composite_out_kernelpre_group_native_functions)SelectiveBuilder)assert_never	concatMapcontextFileManagermake_file_managermapMaybeNamespaceHelperTarget)
YamlDumper
YamlLoader)SequenceTc                      s   e Zd Zd fdd	Z  ZS )
LineLoaderFc                   s$   t  j||d}|jjd |d< |S )N)deep   __line__)superconstruct_mappingZ
start_markline)selfZnoderR   mapping	__class__ ./home/app/PyTorch/test/pytorch/torchgen/gen.pyrV      s   zLineLoader.construct_mapping)F)__name__
__module____qualname__rV   __classcell__r\   r\   rZ   r]   rQ      s    rQ   
ParsedYamlnative_functionsbackend_indiceszdict[str, ParsedYaml]_GLOBAL_PARSE_NATIVE_YAML_CACHEzdict[str, set[str]]_GLOBAL_PARSE_TAGS_YAML_CACHEdispatch_keyr,   
device_fmsdict[str, FileManager]
default_fmrH   returnc                   s(   | t fddt D d|}|S )Nc                 3  s     | ]\}}| r|V  qd S Nr\   ).0Zcheckdevicerg   r\   r]   	<genexpr>   s    
z1file_manager_from_dispatch_key.<locals>.<genexpr> )getnextr+   items)rg   rh   rj   fmr\   ro   r]   file_manager_from_dispatch_key   s   
rv   <stdin>Fesobject
valid_tagsset[str]ignore_keysset[DispatchKey] | Nonepathstrskip_native_fns_genboolc              	     s>  t | tsJ g }tt}| D ]_}t |tsJ d| t |dts)J |t||d |d  d us@J d| t fdd t	|||\}}	|
| t||	 W d    n1 siw   Y  qt| tdd }
|st|| | D ]\}}t|dd	t|pt||d
|
|< qt||
S )Nzexpected to be dict: rT   funczmissed 'func' in c                     s   d d  S Nzin z:
  r\   r\   funcslocr\   r]   <lambda>       z*parse_native_yaml_struct.<locals>.<lambda>c                   S  s   t tjdddi dS )NTFrg   Zuse_out_as_primaryZexternaldevice_guardindex)r'   r,   Z	Undefinedr\   r\   r\   r]   r      s    TFr   )
isinstancelistr   dictrr   intr3   rG   r4   Z	from_yamlappendr'   Z
grow_indexerror_check_native_functionsr@   rt   r/   r2   rb   )rx   rz   r|   r~   r   rsZbser   mindiceskvr\   r   r]   parse_native_yaml_struct   s<   



r   c              	     s   t | tsJ t }| D ]M}t |dtsJ |t||d  |dt fdd" | }|d}|dd}|dksEJ |	| W d    n1 sTw   Y  q|S )NrT   tagc                     s   d  d S r   r\   r\   r   tagsr\   r]   r      r   z(parse_tags_yaml_struct.<locals>.<lambda>descrq   )
r   r   setrr   r   r3   rG   copypopadd)rx   r~   r   r   Ze_inamer   r\   r   r]   parse_tags_yaml_struct   s   

r   c                 C  s\   | t vr*t| }tj|td}t|| dt | < W d    t |  S 1 s%w   Y  t |  S )NLoader)r~   )rf   openyamlloadrQ   r   )r~   frx   r\   r\   r]   parse_tags_yaml   s   

r   )r   loaded_yamltags_yaml_pathr   object | Nonec                C  sr   | t vr5t|}|d u r(t| }tj|td}W d    n1 s"w   Y  n|}t|||| |dt | < t |  S )Nr   )r~   r   )re   r   r   r   r   rQ   r   )r~   r   r|   r   r   rz   r   rx   r\   r\   r]   parse_native_yaml   s    	

r   r   Sequence[NativeFunction]Nonec           
      C  s  i }t t}| D ]}|||jj< ||jjj | q| D ]}|jd urS||j}|d us?J |jj d|j d|j d|jsSJ |jj d|j d|j dtt	j
}h d}|jjjD ] }|j|v rt|jj|jf|vrtd|j d|jj dqad	|jv rt|jjd
krt|jjdkrt|jjjdkr|jjj}|jsJ |jj dt|jd|j}	t||	 dksJ |jj d| dqd S )Nz0 is marked as a structured_delegate pointing to z, but z is missing.zY is not marked as structured. Consider adding 'structured=True' to the delegated operator>   )Z!_has_compatible_shallow_copy_typefrom)Zuniform_r   )zrandom_.fromr   zArgument name 'z' in function 'z' is a reserved Python keyword.Zinplace_viewZresize_Z
resize_as_Zset_z is marked with tag: inplace_view, but it doesn't follow the naming convention for inplace ops - the codegen expects the base name to have a trailing underscore. Fr   zw is marked with tag: inplace_view. The codegen expects there to be a corresponding out-of-place view op with the name 'z/' and matching schema, but it didn't find one. )r   r   r   r   r   structured_delegaterr   
structuredr   keywordkwlist	argumentsZflat_allr   AssertionErrorr   inplacer)   baseZdunder_methodlen)
r   Zfunc_mapZbase_func_mapr   Zdelegate_funcZPYTHON_RESERVED_KEYWORDSZEXCLUSION_LISTargZ	base_nameZout_of_place_base_namer\   r\   r]   r     sb   






	


r   sc                 C  sl   |  dd} |  dd} |  dd} |  dd} |  d	d
} |  dd} |  dd} |  dd} d|  dS )z1Convert a python string into a c++ string literal\z\\"z\"z\az\bz\f
z\nz\v	z\t)replacer   r\   r\   r]   
cpp_stringS  s   r   backendslist[BackendIndex]list[DispatchKey]c                 C  s2   t | dkrg S dd | D tjtjtjtjg S )Nr   c                 S     g | ]}|j qS r\   ro   )rm   Zbackendr\   r\   r]   
<listcomp>q      z(static_dispatch_keys.<locals>.<listcomp>)r   r,   CompositeImplicitAutograd%CompositeImplicitAutogradNestedTensorCompositeExplicitAutograd&CompositeExplicitAutogradNonFunctionalr   r\   r\   r]   static_dispatch_keysm  s   r   r   r4   backend_indexr'   DispatchKey | Nonec                 C  sN   | j d us
|| r|jS | jrtjS | jrtjS | jrtj	S | j
r%tjS d S rl   )r   
has_kernelrg   &has_composite_explicit_autograd_kernelr,   r   5has_composite_explicit_autograd_non_functional_kernelr   &has_composite_implicit_autograd_kernelr   4has_composite_implicit_autograd_nested_tensor_kernelr   )r   r   r\   r\   r]   get_static_dispatch_backendy  s   r   
str | Nonec                 C  sZ   |d u s| j r	d S g }|D ]}t| |}|d ur'|d| j d|  d qd|S )N#include <ATen/ops/__dispatch.h>r   )manual_kernel_registrationr   r   	root_namelowerjoin)r   r   outputr   rg   r\   r\   r]   static_dispatch_ops_header  s   

r   	list[str]c                 C  s   dd t | D S )Nc                 S     g | ]}d | dqS )#include <ATen/Functions.h>r\   )rm   rg   r\   r\   r]   r     s    
z1static_dispatch_extra_headers.<locals>.<listcomp>)r   r   r\   r\   r]   static_dispatch_extra_headers  s   r   sig"CppSignature | DispatcherSignaturecpp_sigr   c                 C  sd   d	dd}t |  }t | }|D ]}|jjtjkr"||} nqt||}ddd |D S )
Ninput_bindingslist[Binding]rk   c                 S  sT   g }| D ]#}|j dkr"tttj|jj|j |j|jd}|	| q|	| q|S )NZmemory_format)nctyper   defaultargument)
r   r   r   r    possibly_redundant_memory_formatr   typer   r   r   )r   Zoutput_bindingsZbindingZspl_mem_format_bindingr\   r\   r]   add_spl_memory_format_binding  s   
	z5translate_args.<locals>.add_spl_memory_format_binding, c                 s      | ]}|j V  qd S rl   Zexprrm   ar\   r\   r]   rp         z!translate_args.<locals>.<genexpr>)r   r   rk   r   )r   r   r   r   r   r   r   r   )r   r   r   Zsrc_bindingsZgoal_bindingsr   exprsr\   r\   r]   translate_args  s   

r   c           	   	   C  sj   t | |}| }t| |}||}|r|jr|jnt}|dd}d| d|j  d| d| d	S )N::nativerq   return ::();)	r   r   r   
get_kernelcpp_namespacer*   r   rg   r   )	r   r   r   r   r   r   backend_metadataZ	kernel_nsnsr\   r\   r]   %generate_static_dispatch_backend_call  s   


$r  c              	   C  s"  t j|ddd}| jr|j r|j}n|j}|d usJ | }t| |}t	
dd}|jrAd| dtj  d| d| d	S |jrVd| dtj  d| d| d	S |jrkd| dtj  d| d| d	S |jrd| dtj  d| d| d	S d	| d
ddd |D  dS )NFmethodZfallback_bindingr   rq   r   r   r   r  z5TORCH_CHECK(false, "Static dispatch does not support z forr   c                 S  s   g | ]}t |jqS r\   )r   rg   )rm   r   r\   r\   r]   r     s    z:generate_static_dispatch_fallback_call.<locals>.<listcomp>z ");)r   from_native_functionsymintr   
has_symintZsymint_signature	signaturer   r   r*   r   r   r,   r   r   r   r   r   r   r   r   r   )r   r   rd   Zcpp_sigsr   r   r   r  r\   r\   r]   &generate_static_dispatch_fallback_call  s*   
$$$$r  c                   sH  t |dks	 jrdS  fdd|D }t |dkr"t|  |d S t |dkr.t|  |S dd |  D }d|} jjj}g }g }|durN|d	 |dkr[|d
| d |dd| d |d g }	|D ]}
|	d|
j	 d |	dt|  |
 d qpt|  |}d}d|| d||	 d| dS )a  
    For a given `NativeFunction`, find out the corresponding backend and dispatch to it. If more than one
    backends exsit, fallback to static dispatch by determining dispatch key from inputs.
    Arguments:
        sig: A CppSignature or DispatcherSignature for this native function we want to use.
        f: NativeFunction to generate static dispatch.
        backend_indices: All available backends.
    Return:
        C++ code to call backend-specific functions, e.g., "return at::cpu::add(self, other, scale);"
    r   rq   c                   s.   g | ]}|  s jd ur|jtv r|qS rl   )r   r   rg   r;   )rm   br   r\   r]   r     s    

z#static_dispatch.<locals>.<listcomp>rS   c                 S  s6   g | ]}t |jtst |jtr|jj r|jqS r\   )r   r   r:   r&   r   is_tensor_liker   r   r\   r\   r]   r     s    


r   Nz>DispatchKeySet(c10::computeDispatchKey(dtype, layout, device))z$c10::detail::multi_dispatch_key_set()zDispatchKeySet _dk_set = z | ;z=DispatchKey _dk = c10::highestPriorityBackendTypeId(_dk_set);zcase DispatchKey:::r   z
		
    z
    switch (_dk) {
        z
        default:
            z
    }
    )
r   r   r  r  r   r   r   tensor_optionsr   rg   )r   r   rd   keysnative_tensor_argstensor_argsZtensor_optsZstmtsZsubexprsZdispatch_coder   ZfallbackZ	connectorr\   r  r]   static_dispatch  sN   
	


r  T)frozenc                   @  s6   e Zd ZU ded< eedZded< edd
dZdS )RegisterSchemarD   selector)default_factoryzdict[str, int]
known_tagsr   r4   rk   r   c                 C  s   | j |sd S dddd t|jD  d }|dkr(dtt|j dS d	}|| jvrBt	| j}|| j|< d
| d| d}| dtt|j d| j|  dS )N{r   c                 s  s    | ]}d | V  qdS )z	at::Tag::Nr\   )rm   r   r\   r\   r]   rp   N      z*RegisterSchema.__call__.<locals>.<genexpr>}z{}zm.def(z, {});
rq   z const std::vector<at::Tag> tags_z = ;
z, tags_z);
)
r  is_native_function_selectedr   sortedr   r   r   r   r  r   )rX   r   r   Z
maybe_tagsZidxr\   r\   r]   __call__J  s   "


&zRegisterSchema.__call__Nr   r4   rk   r   )	r^   r_   r`   __annotations__r   r   r  r   r%  r\   r\   r\   r]   r  E  s
   
 r  c                   @  ,   e Zd ZU ded< ded< edd	d
ZdS )ComputeOperators.Literal[Target.DECLARATION, Target.DEFINITION]targetr   static_dispatch_backend_indicesr   r4   rk   r   c                 C  s  t |j}|jj }| jtju rAd| d|  d|jjj d|jjj	 dt
t|j d|jddd	 d|jd
dd	 dS | jtju rd|j d| d| d| d| d| d}dD ]^}|ruddgdd | D  }d
}nddd | D }d}|}| d| }	d| d| d| d}
|st| jdkrt||| jd }
|d|j d!|j|	|d	 d"|
 d#7 }q_|S t| j d S )$Nz
struct TORCH_API z {
  using schema = z;
  using ptr_schema = schema*;
  // See Note [static constexpr char* members for windows NVCC]
  static constexpr const char* name = "aten::z3";
  static constexpr const char* overload_name = "z/";
  static constexpr const char* schema_str = z;
  static ZcallF)r   is_redispatching_fnZ
redispatchTz;
};

// aten::z.
static C10_NOINLINE c10::TypedOperatorHandle<z::schema> create_zQ_typed_handle() {
  return c10::Dispatcher::singleton()
      .findSchemaOrThrow(z::name, z::overload_name)
      .typed<z::schema>();
}
)FTr   dispatchKeySetc                 S  r   r\   r   r   r\   r\   r]   r     r   z-ComputeOperators.__call__.<locals>.<listcomp>c                 S  r   r\   r0  r   r\   r\   r]   r     r   r   z
    static auto op = create_z_typed_handle();
    return op.r   r  r   rd   r    {
    z
}
)r   from_schemar   r   unambiguous_namer+  rL   DECLARATIONr   overload_namer   r   defn
DEFINITIONr   r   r   r,  r  rE   )rX   r   r   r   Zdefnsr-  Zdispatcher_exprs_strZmethod_baseZdispatcher_callmethod_nameZfn_bodyr\   r\   r]   r%  c  s   	zComputeOperators.__call__Nr   r4   rk   r   r^   r_   r`   r'  r   r%  r\   r\   r\   r]   r)  ^  
   
 r)  c                   @     e Zd ZedddZdS )	ComputeFunctionr   r4   rk   r   c           
      C  s   t j|d|jd}|j }d}| D ]\}t|j}t|	 |	 }d
dd |D }|jr5d}	nd}	tj|jv rU|d	|j d
|  d|jj  d| d	7 }|rp|d|	 d|jdd d|jj  d| d	7 }q|S )NFr  rq   r   c                 S  r   r\   r   rm   r   r\   r\   r]   r     r   z,ComputeFunction.__call__.<locals>.<listcomp>zc10::SymIntZint64_tr.  
inline  {
    return at::_ops::::call(z);
}zZ
namespace symint {
  template <typename T, typename = std::enable_if_t<std::is_same_v<T, z>>>
  T)Zsuppress_symint_suffixz	);
  }
}
)r   r	  manual_cpp_bindingr   r  
signaturesr   r3  r   r   r   r
  r>   functionvariantsdeclr   r4  )
rX   r   	sig_groupr  resultr   
target_sigr   	exprs_strZ	intlike_tr\   r\   r]   r%    sD   





zComputeFunction.__call__Nr&  r^   r_   r`   r   r%  r\   r\   r\   r]   r>        r>  c                   @  r(  )ComputeTensorMethodr*  r+  r   r,  r   r4   rk   r   c                 C  s  t j|jvrd S |j rJ |jjjd usJ tj|d|j	d}| j
tju r;d}| D ]}||  d7 }q-|S | j
tjurFt| j
 d}| D ]7}t|j}t| | dd}ddd |D }|d	|j d
|jdd d|jj  d| d	7 }qL|S )NTr  rq   z const;
)r  r   c                 S  r   r\   r   r?  r\   r\   r]   r   
  r   z0ComputeTensorMethod.__call__.<locals>.<listcomp>r.  r@  zTensor::)prefixz const {
    return at::_ops::rB  );
}
)r>   r  rF  r   	is_out_fnr   Zself_argr   r	  rC  r+  rL   r5  rD  rG  r8  rE   r   r3  r   r   r7  r   r4  rX   r   rH  rI  r   rJ  r   rK  r\   r\   r]   r%    s:   


zComputeTensorMethod.__call__Nr&  r;  r\   r\   r\   r]   rN    r<  rN  c                   @  r=  )	ComputeRedispatchFunctionr   r4   rk   r   c                 C  s   t j|d|jd}d}| D ]8}t|j}t| | }d	dgdd |D  }|d|j d	|j
d
d d|jj  d| d	7 }q|S )NFr  rq   r   r/  c                 S  r   r\   r   r   r\   r\   r]   r   '  r   z6ComputeRedispatchFunction.__call__.<locals>.<listcomp>r.  r@  T)r-  rA  z::redispatch(rP  )r   r	  rC  rD  r   r3  r   r   r   r   rG  r   r4  rR  r\   r\   r]   r%    s$   

z"ComputeRedispatchFunction.__call__Nr&  rL  r\   r\   r\   r]   rS    rM  rS  c                 C  s   d| j jj d| j jj dS )Nz{"aten::z", "z"},)r   r   r6  r  r\   r\   r]   compute_aten_op8  s   rT  gr5   c                 C  s  | j sd S t| j> t| }t | }ddd |D }| jj}|d u r*d}d}| j r3| jjnd }|r)g |j	
 |j}dd |D }dd |D }	dd	d |	D }
d
|
 d}dd |D }ddd |D }g }t|D ]\}}d|	d | dg |	|d d   }d| d}|| jdd}| d|j d| d}d|j d}d|	|  d| d}g }|| d t|D ]!\}}||kr|d|j d q|d|j d|j d  q|d! d"|}|d#| d$| d%| d& qwd"|}ddgt|	 }d'| d(}d)}d*| d+| d#| d,}nd-}d-}d.| d/| d0| d1| d1| d2| d3W  d    S 1 sMw   Y  d S )4Nr   c                 s  s    | ]}|  V  qd S rl   )rG  r   r\   r\   r]   rp   D  s    z4compute_meta_function_declaration.<locals>.<genexpr>zat::impl::MetaBaseZvoidc                 S  s   g | ]	}|D ]}|qqS r\   r\   )rm   Zreplace_listelemr\   r\   r]   r   P  s
    z5compute_meta_function_declaration.<locals>.<listcomp>c                 S  s   g | ]}|j  qS r\   )r   upperrm   rV  r\   r\   r]   r   S      
c                 s  s    | ]	}d | dV  qdS )zbool z = falseNr\   )rm   Zparamr\   r\   r]   rp   V  s    
z
template <>c                 S  s   g | ]
}t j||jd qS ))binds)r   argument_typer   rX  r\   r\   r]   r   \      r"  c                 s  s(    | ]}|j d d d|j V  qdS )TZ	strip_ref N)cpp_typer   rX  r\   r\   r]   rp   a  s
    
truerS   zprecompute_out<Tr^  z set_r   z value)r   z already set"zstatic_assert(z == false, r  z ret;zret.z	 = value;z	 = this->r  zreturn ret;r   z
                    z {
                        z
                        z'
                    }
                z'using meta_return_ty = precompute_out <z>;Zmeta_return_tyz
                zG
                struct TORCH_API precompute_out {
                    z;
            };rq   zstruct TORCH_API structured_z
 : public r2  r  z meta(z);
};
)r   r   outmetar   Zmeta_argumentsr   Zstructured_inheritsprecomputedr   valuesr   	enumerater`  r   r   )rU  r   argsargs_strZparent_classZmeta_returnrd  Zprecomputed_valuesZprecomputed_elementsZprecomputed_template_parametersZprecomputed_template_params_strZprecompute_template_declZ#precomputed_elements_with_cpp_typesZprecomputed_elements_declZsetter_methodsirV  Zreturn_ty_templatesZ	return_tyZelem_cpp_tyr  Z
assert_msgZassert_stmtZconstruction_stmtsjZconstruction_blockZsetter_methods_declZmeta_return_template_paramsZmeta_return_typedefZprecomputed_declr\   r\   r]   !compute_meta_function_declaration>  s   








&rk  r  rD   c                 C  sB   t | jjj}|ds|drdS | jjjd u rdS || S )NZ_likeZnew_F)r   r   r   endswith
startswithr   r  r#  )r   r  r   r\   r\   r]   needs_backend_select  s   
rn  c                   @  r(  )ComputeBackendSelectz/Literal[Target.DEFINITION, Target.REGISTRATION]r+  rD   r  r   r4   rk   r   c                 C  s0  t || jsd S t|j}t|jdd}dd | D }t|j}|}|	 }d}| j
tju r|rP|jj s<J ddd |D }	d	| d
|	 d}
n|jj rXJ d| d}
d|j d|| d|
 d|jj  dddd |D  dS | j
tju rd|jj d| dS t| j
 d S )NTr
  c                 S  s(   g | ]}t |jtr|jj r|qS r\   )r   r   r&   r   r  r   r\   r\   r]   r     s    

z1ComputeBackendSelect.__call__.<locals>.<listcomp>z.c10::computeDispatchKey(dtype, layout, device)r   c                 s  r   rl   r0  r   r\   r\   r]   rp     r   z0ComputeBackendSelect.__call__.<locals>.<genexpr>z-DispatchKeySet _dk_set = c10::DispatchKeySet(z() | c10::detail::multi_dispatch_key_set(z);
DispatchKeySet _dk_mask = c10::DispatchKeySet(DispatchKeySet::FULL_AFTER, DispatchKey::BackendSelect);
DispatchKeySet _dk = c10::impl::computeDispatchKeySet(_dk_set, _dk_mask);z)DispatchKeySet _dk = c10::DispatchKeySet(r  z	// aten::z
C10_ALWAYS_INLINE
z {
  z
  return at::_ops::z::redispatch(
      _dk, c                 s  r   rl   r   r   r\   r\   r]   rp     r   rP  zm.impl("aten::z", TORCH_FN(z));)rn  r  nativer   r   r   r   r   r3  r   r+  rL   r8  Zhas_tensor_argr   r7  r4  REGISTRATIONrE   )rX   r   r   Z
native_sigr  Zdispatcher_sigr   Zdispatcher_exprsrg   r  Z
compute_dkr\   r\   r]   r%    sJ   


	zComputeBackendSelect.__call__Nr&  r;  r\   r\   r\   r]   ro    s
   
 ro  datac                 C  s2   dd t _ddd}t t| tj| d	t d
dS )Nc                 S  s   dS )NTr\   )rX   rs  r\   r\   r]   r         zformat_yaml.<locals>.<lambda>dumperr	   rs  rk   c                 S  s   |  | S rl   )Zrepresent_dictrt   )ru  rs  r\   r\   r]   dict_representer  s   z%format_yaml.<locals>.dict_representerFg    eA)Zdefault_flow_styleZDumperwidth)ru  r	   rs  r	   rk   r	   )rM   Zignore_aliasesZadd_representerr   r   dump)rs  rv  r\   r\   r]   format_yaml  s   

ry  c                 C  s\   | dkrdS | dkrdS zt | W S  ty-   zt| W  Y S  ty,   |  Y  Y S w w )Nra  TZfalseF)r   
ValueErrorfloatr   r\   r\   r]   pythonify_default  s   
r|  tr=   c                 C  s:   t | tr
t| jS t| dkrdS tj| dddd S )NTensorz
at::TensorF__placeholder__)mutabler[  r
  )r   r8   dynamic_typerV  r   r   Zargumenttype_typer`  )r}  r\   r\   r]   r  6  s   

r  rF  set[Variant]c                 C  s2   dg}t j| v r|d t j| v r|d |S )Nr=   r~  	namespace)r>   r  r   rE  )rF  	method_ofr\   r\   r]   compute_method_of_yamlC  s   



r  +tuple[list[dict[str, str]], dict[str, str]]c                 C  s   i }t | }g }tt| jj|D ]3\}\}}t|j|t j|dd	 d}|j
r@|j
|d< | j r@|j
|| jjj| j
< || q||fS )NFrp  )r  r   r   
field_name)r   Zreturn_namesrf  zipr   returnsr  r   Zreturn_typer`  r   rQ  r   rb  r   )r   name_to_field_namenamesr  ri  rr   Zretr\   r\   r]   compute_returns_yamlN  s   *


r  cpp_ar   schema_orderkwarg_only_setout_arg_setr  dict[str, str]c                C  sn   t | jtrd dd| j| jdd}| jd ur| j|d< |S t | jtr%tt | jtr5t	| j||||dS d S )Nzat::TensorOptionsFT)
annotationr  is_nullabler   r   
kwarg_onlyr   r  r  r  r  )
r   r   r<   r   r   r   r:   r   r&   compute_argument_yaml)r  r  r  r  r  r   r\   r\   r]   compute_cpp_argument_yaml  s,   

r  r   r&   c             	   C  s   | j rt| j nd t| j| j | jtj| ddd d}| j	d ur2t
tj| j	| jdd|d< | j|v r;d|d< | j|v rTd|d	< d|d
< | j|v rT|| j |d< | j }|d urn|jd urnt|jdkrn|j|d< |S )Nr  F)r[  r
  )r  r  r  r   r   rp  r   Tr  r   allocater  r   size)r  r   r  r   r  r   r   r\  r`  r   r|  Zdefault_exprZis_list_liker  rV  )r   r  r  r  r  r   lr\   r\   r]   r    s*   	




 
r  c                   s  t | \}dd | jjjD  dd | jjjD tj| ddd}|j } fdd|D }t| j	 } fdd|D }d	d |D }t
j| jjdd
 }| dd| d}	tdd |D ootj| jv}
tdt
| jfdt| jjjfdt| jjjfd| jfd| jd ur| jndfdd| j fd|fd|	fd|fdt| jfdd| jd u rdn| jfd|fd| jjjjfd |
fd!| jfd"| jfd#d$d%| jfgS )&Nc                 S     h | ]}|j qS r\   r0  r   r\   r\   r]   	<setcomp>  r   z+compute_declaration_yaml.<locals>.<setcomp>c                 S  r  r\   r0  r   r\   r\   r]   r    r   Fr  c              	        g | ]}t |d  dqS )Fr  )r  )rm   r  r  r  r  r\   r]   r         z,compute_declaration_yaml.<locals>.<listcomp>c              	     r  )Tr  )r  r   r  r\   r]   r     r  c              
   S  s0   g | ]}t j|d t d d d dD ]}|jqqS )F)r  Zcpp_no_default_argsZfaithfulr
  Zhas_tensor_options)r   r   r   r   )rm   r   r  r\   r\   r]   r     s    rp  z (r   r  c                 s  s    | ]	}t |jtV  qd S rl   )r   r   r<   r   r\   r\   r]   rp         z+compute_declaration_yaml.<locals>.<genexpr>r   Zoperator_namer6  r   category_overriderq   Zschema_stringaten::r   schema_order_cpp_signatureschema_order_argumentsr  )moderq  python_moduler  r   is_factory_methodZabstractr   )Zwith_gilF)Z
deprecatedFZhas_math_kernel)r  r   r   Zflat_kwarg_onlyrb  r   r	  r  r   r  r   returns_typer  r`  r   anyr>   r  rF  r   r   r   r6  r   r  r  r  r   Zis_abstractr   r   )r   r  rH  Zcpp_argsr   Zschema_order_jit_argumentsr  Zcpp_schema_order_typesZcpp_returnsr  r  r\   r  r]   compute_declaration_yaml  s^   

r  c                 C  s0   | j s| jd uo| j tjkp| j tjkS rl   )r   r   r   kindr9   
functionalr   r  r\   r\   r]   "has_autogenerated_composite_kernel/  s   r  dict[DispatchKey, BackendIndex]c              	     s   t  j}t  jj }t  j}ddd |D }d j t fdd|	 D t
jhkoD fdd|	 D t
jt
jhkt jpLt d}| d	| d
| dt| dS )Nr   c                 s  s    | ]	}|   V  qd S rl   )Z
no_defaultrG  r   r\   r\   r]   rp   <  r  z4compute_registration_declarations.<locals>.<genexpr>r  c                      h | ]\}}|  r|qS r\   r   rm   r   r   r  r\   r]   r  A      z4compute_registration_declarations.<locals>.<setcomp>c                   r  r\   r  r  r  r\   r]   r  C  r  )schemadispatchr   r_  r   z); // r   )
dispatcherr   r   r  r  r`  r   r   r   rt   r,   r   r   Zhas_composite_kernelr  jsondumps)r   rd   r   r  rg  rh  Zcomment_datar\   r  r]   !compute_registration_declarations5  s$   
	"r  "provided_op_registration_allowlistlist[str] | Noneop_selection_yaml_pathc                 C  sh   | d ur|d urJ dd }| d urt | }|d ur#t|dd}|S |d ur.t|}|S t }|S )NzhBoth provided_op_registration_allowlist and op_selection_yaml_path can NOT be provided at the same time.TF)r   rD   Z&from_legacy_op_registration_allow_listZfrom_yaml_pathget_nop_selector)r  r  Zop_registration_allowlistr  r\   r\   r]   get_custom_build_selectorV  s&   


r  3Sequence[NativeFunction | NativeFunctionsViewGroup]c                 C  s   ddd}t t}| D ]<}|j }|j}|tjkr.|j }||| vs'J ||| |< q||| vsAJ | d||   ||| |< qt	t
|| S )	Nd1dict[ViewSchemaKind | SchemaKind, NativeFunction]rk   /list[NativeFunction | NativeFunctionsViewGroup]c                 S  s\   g }t j| v r%| t j}| t jd }| tjd }|t|||d || 	  |S )N)view	view_copyview_inplace)
r?   Zaliasingr   Zaliasing_inplacer9   r  r   r6   extendre  )r  r   r  r  r  r\   r\   r]   maybe_create_view_groupx  s   
	zEget_grouped_by_view_native_functions.<locals>.maybe_create_view_groupz already in )r  r  rk   r  )r   r   r   Zview_signatureZview_schema_kindr?   Znon_aliasingr  r  r   rF   re  )rc   r  Zgrouped_by_viewsr   r  Z	view_kindr  r\   r\   r]   $get_grouped_by_view_native_functionsu  s   



r  /Sequence[NativeFunction | NativeFunctionsGroup]c                 C  s(   ddd}t | }tt|t| S )Nr   dict[SchemaKind, NativeFunction]rk   r  c                 S  s>   t | }|d u rtdd |  D rJ t|  S |gS )Nc                 s  s    | ]}d |j v V  qdS )	generatedN)r   rm   r   r\   r\   r]   rp     r   zJget_grouped_native_functions.<locals>.flatten_pre_group.<locals>.<genexpr>)r5   Z	from_dictr  re  r   )r  r  r\   r\   r]   flatten_pre_group  s
   
z7get_grouped_native_functions.<locals>.flatten_pre_group)r  r  rk   r  )rC   r   rF   re  )rc   r  Zpre_grouped_native_functionsr\   r\   r]   get_grouped_native_functions  s
   
r  )native_function_decl_gengrouped_native_functionsr  JCallable[[NativeFunctionsGroup | NativeFunction, BackendIndex], list[str]]dict[str, list[str]]c                 C  s   t t}| D ]B}t }t }| D ]5\}}||}	|	r+|	j}
|| ||
 nt}
t|dks=J d| d| ||
 	||| qq|S )NrS   z6Codegen only supports one namespace per operator, got z from )
r   r   r   rt   r  r  r   r*   r   r  )r  rd   r  ns_grouped_kernelsr   Znative_function_namespacesdispatch_keysrg   backend_idxr  r  r\   r\   r]   get_ns_grouped_kernels  s&   

r  r  c              
   C  sh   g }d}|   D ])\}}t|ddd}tt|}|d|j d|| d|j d	| q|S )Nr   rq      namespace_strZentity_nameZ	max_level	
        )
rt   rK   r   r   fromkeysr  prologuer   epiloguesplit)r  declarationsnewliner  kernels	ns_helperordered_kernelsr\   r\   r]   8get_native_function_declarations_from_ns_grouped_kernels  s(   r  c                 C  s   t | ||d}t|dS )a  
    Generate kernel declarations, in `NativeFunction(s).h`.
    :param grouped_native_functions: a sequence of `NativeFunction` or `NativeFunctionGroup`.
    :param backend_indices: kernel collections grouped by dispatch key.
    :param native_function_decl_gen: callable to generate kernel declaration for each `NativeFunction`.
    :return: a list of string, from the string with all declarations, grouped by namespaces, split by newline.
    r  rd   r  )r  )r  r  )r  rd   r  r  r\   r\   r]    get_native_function_declarations  s   r  %NativeFunction | NativeFunctionsGroupr  c                 C  s^   | | }|r(d|jv s(J dt| tr| jjn| jjj d|j d|j d|r-|jS tS )Nr   zThe kernel for function z with dispatch key z has a namespace z% and it's not ending with '::native'.)	r  r  r   r4   r   r   r  rg   r*   )r   r  r  r\   r\   r]   get_kernel_namespace  s   
 r  ru   rocmr
  skip_dispatcher_op_registrationgen_dispatch_helpersc        	           s  g }	t tt t t t}
d}tj|tj|||d d}tj|tj|||d d}tj|tj|||d d}|D ]A}t	||d
dd ||   || t|tr_|jn|jj}||
 vrot t|
< |
 | || q9D ]Kt dkrq}tdd|
 D ]}|
 | sqd| d	 d
||
 |  d7 q|	| d fdd| q}|	S )Nr   r  r
  class_method_namer  r   r  r   rq   r   )r  z
TORCH_LIBRARY_IMPL(r   z, m) {
    z
}zRegisterDispatchDefinitions.inic                     s,   j j  rdnd  dS )Nrq   )Zns_prologueZns_epilogueZdispatch_anonymous_definitionsZ"static_init_dispatch_registrationsZdeferred_dispatch_registrationsdispatch_namespaceZdispatch_namespaced_definitions)r  r  r   r\   Zanonymous_definitionsrg   Zkernel_namespaceZns_definitionsr  Zregistration_bodyr  r\   r]   r   j  s   z1get_native_function_definitions.<locals>.<lambda>)r   r   r   destRegisterDispatchKeyrL   ZNAMESPACED_DEFINITIONZANONYMOUS_DEFINITIONrr  r  r   r  r   r4   r  r  r   rK   r   Zsubstitute_with_templater  )ru   r  rg   r  r  r  r
  r  r  definitionsZregistrationsr  Zns_genZanonymous_genZreg_genr   r  r\   r  r]   get_native_function_definitions  s   			
r  c              
   C  s   g }t t}d}tj|tj||d d|d}	| D ]}
t|
|dd| }|| 	|	|
 q|
 D ]0\}}t|dkr>q3t|ddd	}tt|}|	d|j d|| d|j d
| q3|S )Nr   F)r  r  r  r
  r  rq  r   rq      r  r  )r   r   r  r  rL   NAMESPACED_DECLARATIONr  r   r   r  rt   r   rK   r   r  r  r   r  r  )r  rg   r  r  r  r
  r  r  r  r   r   r  r  r  r  r\   r\   r]   get_namespaced_declaration  sJ   		r  schema_selectortuple[list[str], str]c              
   C  s   t t}| D ]
}||j | qd}g }d }| D ]1\}}ttt||}	|dkr/|	}q|}d}
|tv r9dnd}|d| d| d|
|	 d	7 }q||fS )
Nrq   Zatenr   ZTORCH_LIBRARY_FRAGMENTZTORCH_LIBRARYr   r   z	, m) {
  z
};)	r   r   r  r   rt   rJ   r  r-   r   )rc   r  Zns_native_functionsZnative_functionschema_registrationsaten_schema_registrationsZcustom_namespacer  r   Zschema_registrations_bodyZtabZtorch_library_macror\   r\   r]   (get_native_function_schema_registrations  s4   r  structured_native_functionsSequence[NativeFunctionsGroup]static_dispatch_idxcpu_fmfunctions_keysset[DispatchKey]r  Sequence[DispatchKey]c              
     s  | dfdd dd D fddD | d
fdd | d	
fd
d | d
fdd t d| dfdd |	D ]4t||}|v rd d| ddfdd | dd 	fdd ~qMd S )NzNativeMetaFunctions.hc                     s   g t tt dS )N)ZNativeMetaFunctions_includesZ NativeMetaFunctions_declarationsr   rJ   rk  r\   )r  r\   r]   r     s
   z(gen_aggregated_headers.<locals>.<lambda>c                 S  s   g | ]
}t j|jv r|qS r\   r>   r  rF  rm   fnr\   r\   r]   r     s    z*gen_aggregated_headers.<locals>.<listcomp>c                   s   g | ]}| vr|qS r\   r\   r  )method_native_functionsr\   r]   r     s    MethodOperators.hc                     s   g t tttjd dS )Nr,  ZMethodOperators_includesZMethodOperators_declarationsr   rJ   r)  rL   r5  r\   )r  r  r\   r]   r     s   zOperators.hc                     s    dgt tttjd dS )Nz!#include <ATen/MethodOperators.h>r
  )ZOperators_includesZOperators_declarationsr  r\   )non_method_native_functionsr  r\   r]   r     s   Functions.hc                     s   t dgttt  dS )Nz#include <ATen/Operators.h>)r   ZFunctions_includesZFunctions_declarations)r   r   rJ   r>  r\   rc   r  r\   r]   r     s   )r  rd   zNativeFunctions.hc                     s   dg dS )Nz%#include <ATen/NativeMetaFunctions.h>)ZNativeFunctions_includesZNativeFunctions_declarationsr\   r\   )r  r\   r]   r   %  s   r   Functions_inl.h>DispatchKeyFunctions.hc                        t  dS N)rg   Zinline_headersr   r\   rg   inl_headersr\   r]   r   3     Functions_inl.hDispatchKeyFunctions_inl.hc                
     s$   g   t  dddS )NT)r  rg   r  r  r  r
  )!DispatchKeyFunctions_inl_includesr   dispatch_namespaced_declarations)r   r  r\   )rd   rg   r  r  r  r\   r]   r   ;  s   )writer  rv   write_with_template)rc   r  r  r  r  rd   r   rh   r  r  r  ru   r\   )rd   r  rg   r  r  r  rc   r  r  r  r  r  r]   gen_aggregated_headers  s\   
	

r  ops_fmc                   sJ  t t| D ]
}|j | qt t}|D ]}|j	|	 | q D ]^\	|	 dd
fdd |	 dd	
fdd |	g }dd	 |D td
krn|	 ddfdd t||t	j
d|	 dd	fdd q(dD ]\ |  d fdd q|	D ]y|vrq g  D ]<\	|	g }ttt	j| tj||
dd dd|td
krq	 |	 d ddfdd qt||}d d| ddfdd | d d!fd"d ~q|d#fd$d d S )%Nz_ops.hz
Operator.hc                     s   dt tttjd iS )Nr  r
  r  r\   )	functionsr  r\   r]   r   k  s   z*gen_per_operator_headers.<locals>.<lambda>.hz
Function.hc                     s2   t tfdd d dt tt  dS )Nc                   s   t |  dS )Nr   )r   r  r  r\   r]   r   ~  s    z<gen_per_operator_headers.<locals>.<lambda>.<locals>.<lambda>r   _ops.h>)Zstatic_dispatch_ops_headersZoperator_includesZfunction_definitions)r   rJ   r>  r\   )r   r   r  r\   r]   r   {  s   

c                 S  s    g | ]}t |tr|jr|qS r\   )r   r5   r   r  r\   r\   r]   r     s    z,gen_per_operator_headers.<locals>.<listcomp>r   z_meta.hzNativeMetaFunction.hc                        dt tt iS )NZmeta_function_declarationsr  r\   )structured_functionsr\   r]   r     s   r  z	_native.hzNativeFunction.hc                     s    rd d dS g  dS )Nr   z_meta.h>)Zextra_includesZnative_function_declarationsr\   r\   )r  is_structuredr   r\   r]   r     s   ))Z	Functionsrq   )Z	OperatorsZ_ops)ZNativeMetaFunctionsZ_meta)ZNativeFunctionsZ_nativec                     s.     dfddt  D   dg iS )NZ	_includesc                   s   g | ]
}d |   dqS )r   .h>r\   rm   r   )suffixr\   r]   r     r]  >gen_per_operator_headers.<locals>.<lambda>.<locals>.<listcomp>Z_declarations)r$  r  r\   )categoryfunctions_by_root_namer+  r\   r]   r     s
   

TFr  r   z_dispatch.hzDispatchKeyFunction.hc                     s
    dS )N)r  r  r\   r\   )r  r  r\   r]   r     s   r   r  r  r  c                     r  r  r  r\   r  r\   r]   r     r  r  r  c                     s   fddt  D g dS )Nc                   s   g | ]}d | d  dqS )r   r   r   r\   r*  )r  r\   r]   r     s    r,  )r  r  r  )r$  r\   )dispatch_namesr  r\   r]   r     s   
r	  c                     s   t dd   D g dS )Nc                 s  s2    | ]\}}t d d |D rd| dV  qdS )c                 s  s    | ]	}t j|jv V  qd S rl   r  r  r\   r\   r]   rp     r  zGgen_per_operator_headers.<locals>.<lambda>.<locals>.<genexpr>.<genexpr>r   r%  N)r  )rm   r   r   r\   r\   r]   rp     s    

z=gen_per_operator_headers.<locals>.<lambda>.<locals>.<genexpr>r  )r$  rt   r\   )r.  r\   r]   r     s
   )r   r   r   r   rt   r  rr   r   r  r  #compute_native_function_declarationr  r   rF   r  rL   r  rv   )rc   r  r  r  rd   r   rh   r  r  r  r  r  Zgrouped_functions_by_root_namegroupZgrouped_functionsru   r\   )r-  r  rg   r/  r  r   r.  r  r(  r   r  r'  r+  r]   gen_per_operator_headersL  s   
	
	
r2  core_fmper_operator_headersc                   s   |rt || ||	|
|||d nt||| ||	|||d |dfdd |dfdd |d fd	d |d
fdd dfdd}|d| dfdd}|d| d S )N)rc   r  r  r  rd   r   rh   r  r  r  r  )rc   r  r  r  r  rd   r   rh   r  r  r  zTensorBody.hc                     s2   t tttjd t tttjd dS )N)r+  r,  )Ztensor_method_declarationsZtensor_method_definitions)r   rJ   rN  rL   r5  r8  r\   r  r\   r]   r   ?  s&   	zgen_headers.<locals>.<lambda>zRedispatchFunctions.hc                     s   dt tt  iS )NZfunction_redispatch_definitions)r   rJ   rS  r\   rc   r\   r]   r   W  s   
zRegistrationDeclarations.hc                     s   d fddD iS )NZregistration_declarationsc                   s   g | ]}t | qS r\   )r  r  r1  r\   r]   r   a      z1gen_headers.<locals>.<lambda>.<locals>.<listcomp>r\   r\   )rd   rc   r\   r]   r   `  s   zVmapGeneratedPlumbing.hc                     s   t  S rl   r$   r\   r5  r\   r]   r   i  s    rk   r  c                    s   t  } t  } D ]"}|t|jjj ||jjjj | dd |j D  q|h d8 }ddd t	|D ddd t	| D dS )	Nc                 s  r   rl   r0  )rm   r   r\   r\   r]   rp   u  r   zAgen_headers.<locals>.gen_aten_interned_strings.<locals>.<genexpr>>   Znot_eqnotorZcomplZxor_eqZor_eqxorZbitorZbitandZand_eqandz \
c                 S  r   )z_(aten, r  r\   r*  r\   r\   r]   r         zBgen_headers.<locals>.gen_aten_interned_strings.<locals>.<listcomp>c                 S  r   )z_(attr, r  r\   r*  r\   r\   r]   r     r;  )Zaten_symbolsZattr_symbols)
r   r   r   r   r   r   updater  r   r$  )attrsr  r   r5  r\   r]   gen_aten_interned_stringsl  s   z.gen_headers.<locals>.gen_aten_interned_stringszaten_interned_strings.hc                     s   dd t iS )NZenum_of_valid_tagsz,
)r   r$  r\   )rz   r\   r]   gen_tags_enum  s   z"gen_headers.<locals>.gen_tags_enumz
enum_tag.h)rk   r  )r2  r  r  )rc   rz   r  r  r  r  rd   r3  r   rh   r  r  r  r  r4  r>  r?  r\   )rd   rc   r  rz   r]   gen_headers  s\   
	
$r@  view_groups"Sequence[NativeFunctionsViewGroup]aoti_fm
cpu_vec_fmforce_schema_registrationupdate_aoti_c_shimaoti_backendsextend_aoti_c_shimc           (        s  d}rd}|D ]ŉt ||
	|rdW
fdd}ndW
fdd} tt}D ]}t|tr;|jn|jj}|| | q1t	 t
jktrX|nddt|| ritng d	}dX	fdd}	jd dddd |t
jkrdnd|dhd D ]djjrtsqjjjjt
ju rՈ	|
u sJ 	d ddfdd |	d ddfdd qt
ju rdrd	d  d!d"fd#d qtd$ d%i D ]}| D ]}|jd ur||j<  nqq |v ri }D ]}t|}|tv r1|||< q"td&d' t|  D d(	  d)} t!d*dd+|ra|"| fd,d nBz*t#t$j%&|j'| }!|!( }"|"kszJ d-W d    n	1 sw   Y  W n t)y   t*t$j%&|j'|  d. Y nw dYfd0d1tr|nd|"d(	  dfd2d 	qdZfd3d4}#|
"d5|# }$|rt+, }$t-|$d6\ |
"d7 fd8d d[d;d<}%|
j.d=|%fd>dd?t/id@h dAdB |
"dCt0 |"dDt0 |"dEfdFd d\fdHdI}&tt }'dJdK t1dLd D dMdK t1dNd D |'2fdOd'D  |
j.dP|'|%|&dh dQdR |
"dSfdTd |
"dUfdVd d S )]Nz#include <c10/cuda/CUDAGuard.h>
#include <ATen/cuda/ATenCUDAGeneral.h>
#include <ATen/cuda/CUDADevice.h>
#include <ATen/cuda/CUDAContext.h>z#include <ATen/hip/impl/HIPGuardImplMasqueradingAsCUDA.h>
#include <ATen/hip/ATenHIPGeneral.h>
#include <ATen/hip/HIPDevice.h>
#include <ATen/hip/HIPContext.h>rk   r   c                    s   g } D ][}d}  |rd}n"t|tr%t fdd| D r%d}n|jr2tjtjfv r2d}|s5q| 	d|j
 d tjkrN| 	d|j
 d v r_| 	d|j
 d d	 qtt| S )
NFTc                 3  s    | ]}  |V  qd S rl   r  r  r"  r\   r]   rp     s    

z=gen_source_files.<locals>.operator_headers.<locals>.<genexpr>r   
_native.h>r)  r   r   )r   r   r5   r  r   r   r,   Metar   r   r   r$  r   )headersrU  Zis_registered)r   rg   r  r  r  r\   r]   operator_headers  s6   
z*gen_source_files.<locals>.operator_headersc                    s8   dg}  t jkr| d  v r| d d | S )Nz!#include <ATen/NativeFunctions.h>z#include <ATen/Functions.h>r   r   )r,   r   r   )rK  )rg   r  r\   r]   rL    s   

rq   )extra_cuda_headersZexternal_backend_headersZdispatch_headersops_headersZdispatch_helpersgnfr  r  c                   s    dt | g dd	iS )Ndispatch_definitionsT)	ru   r  rg   r  r  r  r
  r  r  )r  )rO  )r   rg   ru   r  r  r  r  r\   r]   "register_dispatch_key_env_callable	  s   z<gen_source_files.<locals>.register_dispatch_key_env_callableZRegisterz.cppzRegisterDispatchKey.cppc                 S     | j S rl   r   )xr\   r\   r]   r   *	  s    z"gen_source_files.<locals>.<lambda>r  rS   rP  )key_fnenv_callable
num_shardsbase_envsharded_keysZ	UfuncCPU_zUfuncCPU.cppc                     s"   t t  tdS )N)meta_declarationnative_declarationnative_definitions)rk  r  r0  Zcompute_ufunc_cpur\   )rd   rg   rU  r\   r]   r   :	  s   ZUfuncCPUKernel_zUfuncCPUKernel.cppc                     s   t  dS )N)r   r\  )r  Zcompute_ufunc_cpu_kernelr\   )rU  r   r\   r]   r   E	  s   z%#include <ATen/native/cuda/Loops.cuh>z$#include <ATen/native/hip/Loops.cuh>Z
UfuncCUDA_z.cuzUfuncCUDA.cuc                     s&   t t  tdS )N)r   cuda_headersrZ  r[  r\  )rk  r  r0  Zcompute_ufunc_cudar\   )rd   r]  rg   rU  r   r\   r]   r   Q	  s   zunrecognized z
 for ufuncc                 s  s    | ]\}}|V  qd S rl   r\   )rm   r   valuer\   r\   r]   rp   k	  s    
z#gen_source_files.<locals>.<genexpr>Zc_shim_r!  TheaderrH  Zincludesc                     s    S rl   r\   r\   )
new_headerr\   r]   r   }	  rt  a  

WARNING: The generated AOTInductor C shim header files have unexpectedly changed. This
indicates an AOTInductor fallback operator ABI backward compatibility breakage!!!
Only in a limited number of situations, this is allowed:

1. You added a fallback op to the inductor_fallback_ops list in torchgen/aoti/fallback_ops.py.
If that's the case, run `python torchgen/gen.py --update-aoti-c-shim` to update the existing
C shim header files.

2. You added a new default argument to an existing fallback op. This is clearly a BC breaking
change in the AOTInductor land. In this case, you need to keep a manual copy of that existing
fallback op in a file, e.g. torch/csrc/inductor/aoti_torch/c/shim.h, bump up the version
number of that fallback op in the newly generated C shim files, and update the cpp wrapper
codegen to generate the correct cpp call for this op. Contact AOTInductor team for assistance.

                        z
 not foundr   c                    sD   g } D ]}t | d}|d ur| | qdtt| S )N)rH  r   )r   r   r   r$  r   )rK  r   r`  )rd   rg   rH  fallback_native_functionsstructured_func_group_dictr\   r]   headers_for_aoti	  s   
z*gen_source_files.<locals>.headers_for_aotic                	     s    t  d d  dS )NFr   r_  )r   r\   )rd   rg   rH  extra_headersrb  rd  rc  r\   r]   r   	  s    c                    sL   fdd D } dd | D t tttj| t tttj| dS )Nc                   s   g | ]	}t | r|qS r\   )rn  r  r  r\   r]   r   	  
    
z@gen_source_files.<locals>.gen_backend_select.<locals>.<listcomp>c                 S  s   g | ]	}d |j  dqS )r   r%  rS  r  r\   r\   r]   r   	  s    )rN  Z!backend_select_method_definitionsZ%backend_select_function_registrations)r   rJ   ro  rL   r8  rr  )Zrelevant_fns)rc   r  r\   r]   gen_backend_select	  s"   
z,gen_source_files.<locals>.gen_backend_selectzRegisterBackendSelect.cpp)rc   r  zRegisterSchema.cppc                     s   rg n rg dS dS )N)r  r  r\   r\   )r  r  r  r\   r]   r   	  s   r  @NativeFunction | NativeFunctionsGroup | NativeFunctionsViewGroupc                 S  rR  rl   rS  r#  r\   r\   r]   key_func	  s   z"gen_source_files.<locals>.key_funczOperators.cppc                   s&   d| j  dgttj d| gdS )Nr   r)  r
  )rL  r  )r   r)  rL   r8  r#  r$  r\   r]   r   	  s   r      >   r   rL  r  )rU  rV  rX  rW  rY  zFunctions.cppzTensorMethods.cppzATenOpList.cppc                     r&  )NZaten_ops)r   rJ   rT  r\   r5  r\   r]   r   
  s   rU  c                   s.   ddd}|| t | t|  tj dS )	NrU  ri  rk   r   c                 S  s  t | tr.d| jj dd| jj dg}| jd ur,|d| jj dd| jj dg7 }|S t | trd| jj dd| jj dd| jj dd| jj dg}| jd urh|d| jj dd| jj dg7 }| j	d ur|d| j	j dd| j	j dg7 }|S d| j dd| j dgS )Nr   rI  r%  )
r   r6   r  r   r  r5   r  rb  r   r  )rU  rK  r\   r\   r]   gen_op_headers
  s<   




zPgen_source_files.<locals>.functionalization_env_callable.<locals>.gen_op_headers)rN  func_definitionsfunc_registrations)rU  ri  rk   r   )r    r!   r,   r   )rU  rl  )rd   r  r\   r]   functionalization_env_callable
  s   
(z8gen_source_files.<locals>.functionalization_env_callablec                 S     i | ]}|j j|qS r\   r   r   r  r\   r\   r]   
<dictcomp>U
  r6  z$gen_source_files.<locals>.<dictcomp>c                 S     t |  S rl   r   r   rU  r\   r\   r]   r   W
      c                 S  rp  r\   rq  r  r\   r\   r]   rr  Y
  rY  c                 S  rs  rl   rt  ru  r\   r\   r]   r   Z
  rv  c                 3  s,    | ]}|j j vr|j jvr|V  qd S rl   rq  r  )structured_mapview_mapr\   r]   rp   \
  s    zRegisterFunctionalization.cpp>   rN  rn  Zfunc_add_back_views_definitionsZ!func_add_back_views_registrationsrm  )rU  rV  rW  rY  zFunctionalInverses.hc                     s   dt t fddiS )NZview_inverse_declarationsc                   s
   t  | S rl   )r"   ru  rf  r\   r]   r   v
  s    z4gen_source_files.<locals>.<lambda>.<locals>.<lambda>)r   rJ   r\   )r  rA  r\   r]   r   s
  s   
zCompositeViewCopyKernels.cppc                     sN   dd D dd D  t tt tj t ttt ttdS )Nc                 S  s:   g | ]}d  dd |jdu r|jgn|j|jgD qS )r   c                 s  s&    | ]}d |j  d|j  dV  qdS )r   _ops.h>
#include <ATen/ops/rI  NrS  r  r\   r\   r]   rp   
  s    

@gen_source_files.<locals>.<lambda>.<locals>.<listcomp>.<genexpr>N)r   r  r  rm   rU  r\   r\   r]   r   
  s    
z6gen_source_files.<locals>.<lambda>.<locals>.<listcomp>c                 S  s,   g | ]}d  dd |j|j|jfD qS )r   c                 s  s8    | ]}|d urd|j vrd|j d|j dV  qd S )Nr  r   ry  rI  )r   r   r  r\   r\   r]   rp   
  s    
rz  )r   r   r  r  r{  r\   r\   r]   r   
  s    
)rN  Z#CompositeViewCopyKernel_DefinitionsZ(GeneratedCompositeFunctional_DefinitionsZ!GeneratedCompositeOut_Definitions)r   rJ   r#   r,   r   rA   rB   r\   )rd   r  rA  r\   r]   r   
  s<   
)rk   r   )rO  r  rk   r  )rk   r   )rk   r  )r  ri  rk   r   )rU  ri  rk   r  )3rv   r   r   r   r4   r  r  r   r   r   r,   r   r/   r  Zgen_registration_headersZgen_registration_helpersZwrite_sharded_with_templateCPUrb  Zufunc_inner_loopr1   r   r   r  CUDAr   r   r   r   r   tupler$  rt   r   r  r   osr~   r   install_dirreadFileNotFoundErrorprintrD   r  r  Zwrite_shardedr   r   rF   r  )(rc   r  r  rA  r  r  rd   rC  r3  rD  r   rh   r  r  r  rE  r4  r  rF  rG  rH  rM  rL  Zns_grouped_native_functionsZgrouped_native_functionr  Zregister_dispatch_key_base_envrQ  Z
func_groupr   Z	fallbacksZop_nameZheader_file_nameZold_fileZ
old_headerrh  r  rj  ro  Z
all_groupsr\   )r  r   rd   r]  rg   r  rH  re  rb  ru   r  rU  r  r  rd  r   rc   ra  r  r  r  r  r  rc  rw  r  rA  rx  r]   gen_source_files  sv  
*









	




9
r  c                   s   |  d fdd d S )NzDeclarations.yamlc                     s   t dd  D S )Nc                 S  s   g | ]}t |qS r\   )r  r  r\   r\   r]   r   
  s    z;gen_declarations_yaml.<locals>.<lambda>.<locals>.<listcomp>)ry  r\   r5  r\   r]   r   
  r   z'gen_declarations_yaml.<locals>.<lambda>)r  )r   rc   r\   r5  r]   gen_declarations_yaml
  s   
r  r   c                   C  s   t tj S )z
    If you're depending on torchgen out-of-tree, you can use the root to figure
    out the path to native_functions.yaml
    )r   __file__parentresolver\   r\   r\   r]   get_torchgen_root
  s   r  c            #        s*  t jdd} | jddddd | jdd	d
d | jdddd | jdddd | jdddddd | jddddd | jdddd | jdddd | jddd d | jd!d"d#d$d% | jd&d'd(d | jd)d*d#d+d% | jd,d-d#d.d% | jd/d0dd1d | jd2d3dd4d | jd5td#g d6g d6d7d8 | jd9dd:d | jd;dd<d |  tjj}tj	
jd=}tj	
jd>}d?d@lm} t }js|tj tj|v r||tj= js|tj tj|v r||tj= t|||}t| }|j|j} t|}	dAdB |	D }
t|}dCdB |D }j dD}t|jdEdEdF j dG}t|jdEdEdF j }t|jdEdEdF t |dH}t dI}t dI}t dI}t |dH}t |dH}dJ|i}jrvt dI|dK< tj!tj"tj#tj$tj%tj&tj'h}tj!tj"h}jr|tj jr|tj |tj j(rfdLdB|D }g }j)rه fdMdBj)D }j)D ]}t*|}||vr|| qdNj+v r)t,dmi dO|dP|	dQ|
dR|dS|dT|dU dV|dW|dX|dY|dZ|d[|d\|d]j-d^j.d_j/d`j0daj1db|dcj2 ddj+v rDt3|||	|
|| ||||||j-j/de dfj+v rPt4||dg j5rtj56 }|j7}|j8}|dhf|dif|djf|dkfgdldB |9 D  D ]\}} | | }!|j:| |  }"|;|!t|" qxd S d S )nNzGenerate ATen source files)Zdescriptionz-sz--source-pathz!path to source directory for ATenzaten/src/ATen)helpr   z-oz--output-dependenciesz:output a list of dependencies into the given file and exit)r  z	--dry-runZ
store_truez5run without writing any files (still updates outputs))actionr  z--per-operator-headersz2generate separate headers per operator in ATen/opsz-dz--install-dirz--install_dirzoutput directoryzbuild/aten/src/ATenz--aoti-install-dirz--aoti_install_dirz%output directory for AOTInductor shimz(torch/csrc/inductor/aoti_torch/generatedz--rocmz=reinterpret CUDA as ROCm/HIP and adjust filepaths accordinglyz--mpsz'Generate MPS registration code when setz--xpuz'Generate XPU registration code when setz--op-registration-whitelistz--op_registration_whitelist*zfilter op registrations by the whitelist (if set); each item is `namespace`::`operator name` without overload name; e.g.: aten::empty aten::conv2d ...)nargsr  z--op-selection-yaml-pathz--op_selection_yaml_pathaD  Provide a path to the operator selection (for custom build) YAML that contains the information about the set of selected operators and their categories (training, ...). Each operator is either a full operator name with overload or just a bare operator name. The operator names also contain the namespace prefix (e.g. aten::)z--backend-whitelistz--backend_whitelistzRfilter dispatch backend by the whitelist (if set), e.g.: CPU CUDA QuantizedCPU ...z--static-dispatch-backendz--static_dispatch_backendz?generate static dispatch code for the specific backend (if set)z!--skip-dispatcher-op-registrationz!--skip_dispatcher_op_registrationz0Avoid registering operators into the dispatcher.z--force-schema-registrationz--force_schema_registrationz}force it to generate schema-only registrations for all ops, includingthose that are not listed on --op-registration-whitelistz
--generate)rK  sourcesdeclarations_yamlzGenerate only a subset of files)r   r  Zchoicesr   r  z--update-aoti-c-shimzUpdate AOTInductor C shim after adding an entry to inductor_fallback_ops in torchgen/aoti/fallback_ops.py. WARNING: Do not use this unless you are sure what you are doing!!!z--extend-aoti-c-shima  This Flag indicates the generation of c shims for out-of-tree ATen ops,which is an extension to the In-tree ATen op c shims. This flag needs to be combined with---source-path=<out-of-tree native_functions.yaml>--aoti-install-dir=<in-tree aoti_install_dir>/extend   default is torch/csrc/inductor/aoti_torch/generated/extendWARNING: Do not use this unless you are sure what you are doing!!!znative/native_functions.yamlznative/tags.yamlr   )r  c                 S     g | ]	}t |tr|qS r\   )r   r5   r{  r\   r\   r]   r   r  rg  zmain.<locals>.<listcomp>c                 S  r  r\   )r   r6   r{  r\   r\   r]   r   x  s    z/coreT)parentsexist_okz/ops)optionsr  r  Zcudaxpuc                   s&   g | ]}t |st| jv r|qS r\   )r0   r   backend_whitelist)rm   r   r  r\   r]   r     s    c                   s   g | ]	} t | qS r\   )r,   parse)rm   keyr1  r\   r]   r     s    r  rc   r  r  rA  r  r  rd   rC  r3  rD  r   rh   r  r  r  rE  r4  r  rF  rG  rH  rK  )rc   rz   r  r  r  r  rd   r3  r   rh   r  r  r  r  r4  r  )rc   r   rq   Zcpu_vec_Zcore_Zops_c                 S  s   g | ]\}}|| d fqS )r   r\   )rm   rn   Z	device_fmr\   r\   r]   r     r  r\   )<argparseZArgumentParserZadd_argumentr   Z
parse_argsr  Zop_registration_whitelistr  r  r~   r   source_pathtorchgen.modelr  r   Zmpsr   r,   ZMPSr   r  ZXPUr   rf   rc   rd   r  r  r  r   mkdiraoti_install_dirrI   r|  r}  r   r   r   r   rJ  r  Zstatic_dispatch_backendr  Zgenerater  r  rE  r4  r  rF  rH  r@  r  Zoutput_dependenciesr  r   Zstemrt   r  Zwrite_outputs)#Zparserr  Znative_yaml_pathr   r  r|   Zparsed_yamlrz   rc   r  r  Z!native_functions_with_view_groupsrA  Zcore_install_dirZops_install_dirr  r3  r   rD  Zcuda_fmr  rC  rh   r  rG  r  r  Zdp_keyZdepfile_pathZdepfile_nameZdepfile_stemru   rO  Zvarnamer~   r\   )rd   r  r]   main
  s  		










	
r  __main__)rg   r,   rh   ri   rj   rH   rk   rH   )Nrw   F)rx   ry   rz   r{   r|   r}   r~   r   r   r   rk   rb   )rw   )rx   ry   r~   r   rk   r{   )r~   r   rk   r{   rl   )r~   r   r   r   r|   r}   r   r   r   r   rk   rb   )r   r   rk   r   )r   r   rk   r   )r   r   rk   r   )r   r4   r   r'   rk   r   )r   r4   r   r   rk   r   )r   r   rk   r   )r   r   r   r   rk   r   )r   r   r   r4   r   r'   rk   r   )r   r   r   r4   rd   r   rk   r   r:  )rU  r5   rk   r   )r   r4   r  rD   rk   r   )rs  ry   rk   r   )r   r   rk   ry   )r}  r=   rk   r   )rF  r  rk   r   )r   r4   rk   r  )r  r   r  r   r  r{   r  r{   r  r  rk   ry   )r   r&   r  r   r  r{   r  r{   r  r  rk   ry   )r   r4   rk   ry   )r   r4   rk   r   )r   r4   rd   r  rk   r   )r  r  r  r   rk   rD   )rc   r   rk   r  )rc   r   rk   r  )r  r  rd   r  r  r  rk   r  )r  r  rk   r   )r  r  rd   r  r  r  rk   r   )r   r  r  r'   rk   r   )ru   rH   r  r  rg   r,   r  r'   r  rD   r  r   r
  r   r  r   r  r   rk   r   )r  r  rg   r,   r  r'   r  rD   r  r   r
  r   rk   r   )rc   r   r  rD   rk   r  )rc   r   r  r  r  r  r  r   r  rD   rd   r  r   rH   rh   ri   r  r  r  r  r  r   rk   r   )rc   r   r  r  r  r   r  rD   rd   r  r   rH   rh   ri   r  rH   r  r  r  r  r  r   rk   r   ) rc   r   rz   r{   r  r  r  r  r  r   r  rD   rd   r  r3  rH   r   rH   rh   ri   r  rH   r  r  r  r  r  r   r4  r   rk   r   ),rc   r   r  r  r  r  rA  rB  r  rD   r  r   rd   r  rC  rH   r3  rH   rD  rH   r   rH   rh   ri   r  r  r  r  r  r   rE  r   r4  r   r  r   rF  r   rG  r  rH  r   rk   r   )r   rH   rc   r   rk   r   )rk   r   )rk   r   )Z
__future__r   r  	functoolsr  r   r  collectionsr   r   r   Zdataclassesr   r   Zpathlibr   typingr	   r
   r   r   r   r   Ztorchgen.api.dispatcherZapir  Ztorchgen.api.metarc  Ztorchgen.api.nativerq  Ztorchgen.api.structuredr   Ztorchgen.destr  Ztorchgen.aoti.fallback_opsr   Ztorchgen.apir   Ztorchgen.api.translater   Ztorchgen.api.typesr   r   r   r   r   r   r   Ztorchgen.contextr   r   r   r   Ztorchgen.gen_aoti_c_shimr   r   r   r   Z#torchgen.gen_functionalization_typer    r!   r"   r#   Ztorchgen.gen_vmap_plumbingr%   r  r&   r'   r(   r)   r*   r+   r,   r-   r.   r/   r0   r1   r2   r3   r4   r5   r6   r7   r8   r9   r:   r;   r<   r=   r>   r?   Z#torchgen.native_function_generationr@   rA   rB   rC   Z!torchgen.selective_build.selectorrD   Ztorchgen.utilsrE   rF   rG   rH   rI   rJ   rK   rL   Ztorchgen.yaml_utilsrM   rN   Zcollections.abcrO   rP   rQ   rb   re   r'  rf   rv   r   r   cacher   r   r   r   r   r   r   r   r   r  r  r  r  r)  r>  rN  rS  rT  rk  rn  ro  ry  r|  r  r  r  r  r  r  r  r  r  r  r  r0  r  r  r  r  r  r  r  r  r2  r@  r  r  r  r  r^   r\   r\   r\   r]   <module>   s    $	p(
!	
0
 
<






#

L\/-
|DC $Y / d0$y C     /	  
1
