o
    h                     @  sD  d dl m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	Z	d dl
Zd dlZd dlmZ d dl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 d
dlmZ d
dlmZ eG dd dZG dd dZ G dd dZ!eddG dd dZ"dd Z#dd Z$dd Z%dd Z&dd  Z'ej(e&ej)gd!Z*ej(e&ej+gd!Z,ej(e'ej-gd!Z.G d"d# d#Z/G d$d% d%Z0d&d' Z1d(d) Z2d*d+ Z3G d,d- d-Z4G d.d/ d/e4Z5G d0d1 d1e4Z6d2d3 Z7d4d5 Z8d6d7 Z9d8d9 Z:d:d; Z;e0 Z<ee<Z=d<d= Z>d>d? Z?G d@dA dAZ@G dBdC dCejAZBG dDdE dEZCG dFdG dGZDdS )H    )annotationsN)TupleListDict)	dataclass)TritonSemantic)TensorDescriptor   )InterpreterError)partial   )interpreter)irc                   @  sV   e Zd ZU dZded< ded< ejedZded< d	d
 Z	dd Z
dd Zdd ZdS )TensorHandlez
        data: numpy array
        dtype: triton type, either pointer_type or scalar_type.
        we don't store block_type here because the shape information is already available in the data field
        attr: a dictionary of attributes
    znp.arraydataztl.dtypedtype)default_factoryr   attrc                 C  s   t | j S N)boolr   allself r   `/home/www/facesmatcher.com/frenv_anti/lib/python3.10/site-packages/triton/runtime/interpreter.py__bool__#      zTensorHandle.__bool__c                 C  s$   | j }t|dr|j}t|ds|S )N
element_ty)r   hasattrr   )r   r   r   r   r   get_element_ty&   s
   

zTensorHandle.get_element_tyc                 C  s   t | j | jS r   )r   r   copyr   r   r   r   r   clone,      zTensorHandle.clonec                 C  s   || j |< d S r   )r   )r   keyvaluer   r   r   set_attr/   r   zTensorHandle.set_attrN)__name__
__module____qualname____doc____annotations__dataclassesfielddictr   r   r   r!   r%   r   r   r   r   r      s   
 r   c                   @  s   e Zd Zdd Zdd ZdS )BlockPointerHandlec                 C  s(   || _ || _|| _|| _|| _|| _d S r   )baseshapestridesoffsetsblock_shapeorder)r   r/   r0   r1   r2   r3   r4   r   r   r   __init__5   s   
zBlockPointerHandle.__init__c           	      C  s   | j  }|jd }t| j j| j}tj| jtd}t	t
| jD ]D}dgt
| j }| j| ||< | j| jt| j|  |}||| | j| j tj }||v rf||| j| jk @ |dk@ }q"t|| j jj}||fS )N   r   r	   r   )r/   r   primitive_bitwidthnpbroadcast_tor   r3   onesr   rangelenr2   arangereshaper1   astypeuint64r0   r   r   scalar)	r   boundary_checkdtype_ttZn_bytesptrsmasksdim
bcast_dimsoffr   r   r   materialize_pointers=   s   

" z'BlockPointerHandle.materialize_pointersN)r&   r'   r(   r5   rJ   r   r   r   r   r.   3   s    r.   c                   @  s(   e Zd Zddd	Zd
d ZdddZdS )TensorDescHandler/   r   r0   List[TensorHandle]r1   r3   	List[int]c                 C  s&   || _ t|| _|| _|| _|| _d S r   )r/   r=   ndimr0   r1   r3   )r   r/   r0   r1   r3   r   r   r   r5   O   s
   

zTensorDescHandle.__init__c                 C  s   | j j d dksJ dt| j| jksJ t| j| jks"J | jd d D ]}|j d dks8J dq)| jd j dksGJ dd S )N   r   zbase must be 16-byte alignedzstride must be 16-byte alignedr	   zlast dim must be contiguous)r/   r   itemr=   r1   rN   r3   )r   strider   r   r   validateW   s    zTensorDescHandle.validater2   c           	      C  s  t || jks	J | jjj}|jd }|d j| d dks"J dt| jj| j	}tj
| j	td}tt | j	D ]?}dgt | j	 }| j	| ||< || jt| j	|  |}||| | j| j tj }|d|k@ || j| jk @ }q:|jtjksJ t|| jjj}||fS )Nr6   rP   rO   r   z*block offset start must be 16-byte alignedr7   r	   )r=   rN   r/   r   r   r8   r   r9   r:   r3   r;   r   r<   r>   r?   r1   r@   rA   r0   r   rB   )	r   r2   Z	scalar_tyitemsizerE   rF   rG   rH   rI   r   r   r   rJ   `   s   

  z%TensorDescHandle.materialize_pointersN)r/   r   r0   rL   r1   rL   r3   rM   )r2   rL   )r&   r'   r(   r5   rS   rJ   r   r   r   r   rK   M   s    
	rK   T)frozenc                   @  s   e Zd ZU dZded< dZded< dZded< dZd	ed
< dZded< dZ	ded< dZ
d	ed< dZded< dZded< dZd	ed< dS )InterpreterOptionsNr-   extern_libsFr   debugTsanitize_overflowstrarch)fp8e5Zfp8e5b16fp8e4nvZfp8e4b8fp8e4b15z
Tuple[str]supported_fp8_dtypesr   !deprecated_fp8_dot_operand_dtypestf32default_dot_input_precision)ra   Ztf32x3Zieeeallowed_dot_input_precisionsr   intmax_num_imprecise_acc_defaultr   backend_name)r&   r'   r(   rW   r*   rX   rY   r[   r_   r`   rb   rc   re   rf   r   r   r   r   rV   s   s   
 rV   c                 C  sD   | t jkrt jS | t jkrt jS | t jkrt jS | t jkr t jS | S r   )	r9   uint8int8uint16int16uint32int32rA   int64r7   r   r   r   _get_signed_np_dtype   s   



rn   c                 C  st  t | tjrttjS i tjtttjttjtj	ttj	tj
ttj
tjttjtjttjtjttjtjttjtjttjtjttjtjttjtjttjtjttjtjttjtjttjtjttjtjttjtjttji}t | tjrt | jtjrttjS || j S ||  S r   )
isinstancetlpointer_typer9   r   rA   int1r   float16float32float64rh   rg   rj   ri   rl   rk   rm   bfloat16float8e5float8e5b16
float8e4nv
float8e4b8float8e4b15
block_typer   )Ztt_dtypeZnp_typesr   r   r   _get_np_dtype   sX   	

r}   c                 C  s  t td|j }t td|j }tj|  |d}||jd ? d@ }|j|j d }|j|j d }	|d|j> d @ }
|j}|j}||j? d|> d @ tj}|dk}t	|rtj
|tjd}t|jD ]}|
|? d@ }|j| ||dk< qh|
dk}d||  ||< || |||@ < |
| || > d|j> d @ |
|< tdt|| | d|	> d }||}||}|j|jkr|
|j|j ? d|j> d @ }|tjjkr|
d|j|j d > @ }||dk }||}n|
||j|j > d|j> d @ }|dk}t	|rH||j? d|> d @ tj}|dk}||@ }tj
|tjd}d| || |  ||< || || ? d|j||  > B ||< ||jd > ||j> B |B }|| jS )Nuintr7   r	   r   )getattrr9   r8   Z
frombuffertobytesZfp_mantissa_widthZexponent_biasr@   rl   any
zeros_liker<   maximumminimum_irZROUNDING_MODEZRTNEr?   r0   )inputZinput_dtypeZoutput_dtyperounding_modeZinput_uint_dtypeZoutput_unint_dtypeZ	input_binsignZinput_exponent_widthZoutput_exponent_widthZsignificandZ
bias_inputZbias_outputexponentZsubnormal_indexZbit_posiZ	bit_indexZzero_significand_indexZexponent_outputZsign_outputZsignificand_outputcut_offZnon_zero_exponent_indexshiftoutputr   r   r   _convert_float   sl   
$


r   c                 C  s
   t | S r   )matherfxr   r   r   _erf   s   
r   c                 C  s   t | t | d? S )N@   )rd   )abr   r   r   
_umulhi_64   s   r   )Zotypesc                   @  s   e Zd Zedd ZdS )ExtraFunctionsc                 C  s   t |j| j|||S r   )rp   tensorbuildercreate_fp_to_fphandle)r   Zdst_tyZfp_downcast_rounding	_semanticr   r   r   _convert_custom_types   s   z$ExtraFunctions._convert_custom_typesN)r&   r'   r(   staticmethodr   r   r   r   r   r      s    r   c                   @  s  e Zd Zejjejjejjejjejjejjejj	ejj	iZ
ejjejjejjejjejjejjejjejjejjejjejjejjejjejjejjejjejjejjejjejji
ZdddZdd Zdd Zd	d
 Zdd Zdd Zdd Zdd Zdd Z dd Z!dd Z"dd Z#dd Z$dd Z%dd  Z&d!d" Z'd#d$ Z(d%d& Z)d'd( Z*d)d* Z+d+d, Z,d-d. Z-d/d0 Z.d1d2 Z/d3d4 Z0d5d6 Z1d7d8 Z2d9d: Z3d;d< Z4d=d> Z5d?d@ Z6dAdB Z7dCdD Z8dEdF Z9dGdH Z:dIdJ Z;dKdL Z<dMdN Z=dOdP Z>dQdR Z?dSdT Z@dUdV ZAdWdX ZBdYdZ ZCd[dZ ZDd\dZ ZEd]dZ ZFd^dZ ZGd_dZ ZHd`dZ ZIdadb ZJdcdd ZKdedf ZLdgdZ ZMdhdZ ZNdidZ ZOdjdZ ZPdkdZ ZQdldZ ZRdmdZ ZSdndZ ZTdodZ ZUdpdZ ZVdqdZ ZWdrdZ ZXdsdZ ZYdtdZ ZZdudZ Z[dvdZ Z\dwdZ Z]dxdZ Z^dydZ Z_dzdZ Z`d{dZ Zad|dZ Zbd}dZ Zcd~dZ ZdddZ ZeddZ ZfddZ ZgddZ ZhddZ ZiddZ ZjddZ ZkddZ ZlddZ ZmddZ ZnddZ ZoddZ ZpddZ ZqddZ ZrddZ ZsddZ ZtddZ ZuddZ ZvddZ ZwddZ ZxddZ ZyddZ ZzddZ Z{ddZ Z|eKZ}eKZ~dd Zdd Zdd Zdd ZddZ ZddZ Zdd Zdd Zdd ZddZ ZddZ ZddZ ZddZ ZddZ ZddZ ZddZ ZddZ ZddZ ZddZ ZddZ Zdd Zdd ZddZ Zdd Zdd Zdd Zdd Zdd Zdd ZddĄ ZddƄ ZddȄ Zddʄ Zdd̄ Zdd΄ ZddЄ Zdd҄ ZddԄ Zddք Zdd؄ Zddڄ Zdd܄ Zddބ Zdd Zdd Zdd Zdd ZdddZdddZdddZdddZdddZd d ZdS (	  InterpreterBuilderreturnNonec                 C  s2   d | _ t | _i | _tj| jd< dd | jd< d S )NZconvert_custom_typesc                 S  s   dS )N)r	   r	   r	   r   )ZlhsTypeZrhsTyper   r   r   <lambda>  s    z-InterpreterBuilder.__init__.<locals>.<lambda>Zmin_dot_size)r[   rV   optionsZcodegen_fnsr   r   r   r   r   r   r5     s
   zInterpreterBuilder.__init__c                 C  sR   || j d k std|| j d k std|| j d k s!td|||f| _d S )Nr   zx >= grid_dim[0]r	   zy >= grid_dim[1]r   zz >= grid_dim[2])grid_dim
ValueErrorgrid_idxr   r   yzr   r   r   set_grid_idx   s   zInterpreterBuilder.set_grid_idxc                 C  s   |||f| _ d S r   )r   )r   nxnyZnzr   r   r   set_grid_dim)     zInterpreterBuilder.set_grid_dimc                 C     t jS r   )rp   rs   r   r   r   r   get_half_ty.     zInterpreterBuilder.get_half_tyc                 C  r   r   )rp   rv   r   r   r   r   get_bf16_ty1  r   zInterpreterBuilder.get_bf16_tyc                 C  r   r   )rp   rt   r   r   r   r   get_float_ty4  r   zInterpreterBuilder.get_float_tyc                 C  r   r   )rp   ru   r   r   r   r   get_double_ty7  r   z InterpreterBuilder.get_double_tyc                 C  r   r   )rp   rr   r   r   r   r   get_int1_ty:  r   zInterpreterBuilder.get_int1_tyc                 C  r   r   )rp   rh   r   r   r   r   get_int8_ty=  r   zInterpreterBuilder.get_int8_tyc                 C  r   r   )rp   rg   r   r   r   r   get_uint8_ty@  r   zInterpreterBuilder.get_uint8_tyc                 C  r   r   )rp   rj   r   r   r   r   get_int16_tyC  r   zInterpreterBuilder.get_int16_tyc                 C  r   r   )rp   ri   r   r   r   r   get_uint16_tyF  r   z InterpreterBuilder.get_uint16_tyc                 C  r   r   )rp   rl   r   r   r   r   get_int32_tyI  r   zInterpreterBuilder.get_int32_tyc                 C  r   r   )rp   rk   r   r   r   r   get_uint32_tyL  r   z InterpreterBuilder.get_uint32_tyc                 C  r   r   )rp   rm   r   r   r   r   get_int64_tyO  r   zInterpreterBuilder.get_int64_tyc                 C  r   r   )rp   rA   r   r   r   r   get_uint64_tyR  r   z InterpreterBuilder.get_uint64_tyc                 C  r   r   )rp   ry   r   r   r   r   get_fp8e4nv_tyU  r   z!InterpreterBuilder.get_fp8e4nv_tyc                 C  r   r   )rp   r{   r   r   r   r   get_fp8e4b15_tyX  r   z"InterpreterBuilder.get_fp8e4b15_tyc                 C  r   r   )rp   rz   r   r   r   r   get_fp8e4b8_ty[  r   z!InterpreterBuilder.get_fp8e4b8_tyc                 C  r   r   )rp   rw   r   r   r   r   get_fp8e5_ty^  r   zInterpreterBuilder.get_fp8e5_tyc                 C  r   r   )rp   rx   r   r   r   r   get_fp8e5b16_tya  r   z"InterpreterBuilder.get_fp8e5b16_tyc                 C     t ||S r   )rp   rq   )r   Zelt_tyZ
addr_spacer   r   r   
get_ptr_tyd     zInterpreterBuilder.get_ptr_tyc                 C  r   r   )rp   r|   )r   r   r0   r   r   r   get_block_tyg  r   zInterpreterBuilder.get_block_tyc                 C  s   t tj|gtjdtjS Nr7   )r   r9   arraybool_rp   rr   r   r$   r   r   r   get_int1j     zInterpreterBuilder.get_int1c                 C     t tj|gtjdtjS r   )r   r9   r   rg   rp   r   r   r   r   	get_uint8m  r   zInterpreterBuilder.get_uint8c                 C  r   r   )r   r9   r   rh   rp   r   r   r   r   get_int8p  r   zInterpreterBuilder.get_int8c                 C  r   r   )r   r9   r   ri   rp   r   r   r   r   
get_uint16s  r   zInterpreterBuilder.get_uint16c                 C  r   r   )r   r9   r   rj   rp   r   r   r   r   	get_int16v  r   zInterpreterBuilder.get_int16c                 C  r   r   )r   r9   r   rk   rp   r   r   r   r   
get_uint32y  r   zInterpreterBuilder.get_uint32c                 C  r   r   )r   r9   r   rl   rp   r   r   r   r   	get_int32|  r   zInterpreterBuilder.get_int32c                 C  r   r   )r   r9   r   rA   rp   r   r   r   r   
get_uint64  r   zInterpreterBuilder.get_uint64c                 C  r   r   )r   r9   r   rm   rp   r   r   r   r   	get_int64  r   zInterpreterBuilder.get_int64c                 C  r   r   )r   r9   r   rs   rp   r   r   r   r   get_fp16  r   zInterpreterBuilder.get_fp16c                 C  r   r   )r   r9   r   rt   rp   r   r   r   r   get_fp32  r   zInterpreterBuilder.get_fp32c                 C  r   r   )r   r9   r   ru   rp   r   r   r   r   get_fp64  r   zInterpreterBuilder.get_fp64c                 C  s   t tjdgt|d|S Nr   r7   )r   r9   r   r}   )r   typer   r   r   get_null_value  r   z!InterpreterBuilder.get_null_valuec                 C  s2   | j d u r	tdttj| j | gtjdtjS )Nzgrid_idx is Noner7   )r   r   r   r9   r   rl   rp   r   axisr   r   r   create_get_program_id  s   
 z(InterpreterBuilder.create_get_program_idc                 C  s    t tj| j| gtjdtjS r   )r   r9   r   r   rl   rp   r   r   r   r   create_get_num_programs  s    z*InterpreterBuilder.create_get_num_programsc                 C  s0   t tj|jtdtj}d }| ||||||S r   )r   r9   	ones_liker   r   rp   rr   create_masked_load)r   ptr_0_1is_volatilemaskotherr   r   r   create_load  s   zInterpreterBuilder.create_loadc                 C  s*   t tj|jtdtj}| |||d d S r   )r   r9   r   r   r   rp   rr   create_masked_store)r   r   valr   r   r   r   r   r   create_store  s   zInterpreterBuilder.create_storec           
      C  sN   |  }t|}|d u rttj|j|d|}t|j|j|j|}	t|	|S r   )r   r}   r   r9   r   r   _interpreterload)
r   rE   r   r   cache_modifiereviction_policyr   rD   dtype_npretr   r   r   r     s   
z%InterpreterBuilder.create_masked_loadc                 C  s   t |j|j|jS r   )r   storer   )r   rE   r$   r   r   r   r   r   r   r        z&InterpreterBuilder.create_masked_storec                 C  st   |j j}|j}|tjkr|tjks|tjkr.|tjkr.t|j||d t|}t	||jS t	|j
t||jS r   )r   rB   rp   rv   rt   r   r   viewr}   r   r@   )r   srcdst_typesrc_element_typedst_element_typer   r   r   r   	cast_impl  s   zInterpreterBuilder.cast_implc                 C     |  ||S r   r   r   r   r   r   r   r   r         zInterpreterBuilder.<lambda>c                 C  r   r   r   r   r   r   r   r     r   c                 C  r   r   r   r   r   r   r   r     r   c                 C  r   r   r   r   r   r   r   r     r   c                 C  r   r   r   r   r   r   r   r     r   c                 C  r   r   r   r   r   r   r   r     r   c                 C  r   r   r   )r   r   r   	is_signedr   r   r   r     r   c                 C  s4   |j j}|j}t|j|||t|}t||jS r   )r   rB   r   r   r   r}   r   )r   r   r   r   r   r   r   r   r   r   r     s   z"InterpreterBuilder.create_fp_to_fpc                 C  s   t |jt||jS r   )r   r   r   r}   rB   r   r   r   r   create_bitcast     z!InterpreterBuilder.create_bitcastc                 C  s   t ||j|j|jjS r   r   r   r   rB   )r   lhsrhsopr   r   r   	binary_op  r  zInterpreterBuilder.binary_opc                 C     |  ||tjS r   r  r9   addr   r  r  r   r   r   r         c                 C  r  r   r  r9   multiplyr
  r   r   r   r     r  c                 C  r  r   r  r9   divider
  r   r   r   r     r  c                 C  r  r   r  r9   fmodr
  r   r   r   r     r  c                 C  r  r   r  r9   subtractr
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r   r   create_idivr
  r   r   r   r     r   c                 C  r   r   r  r
  r   r   r   r     r   c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   )r  r9   Z
left_shiftr
  r   r   r   r     r  c                 C  r  r   )r  r9   right_shiftr
  r   r   r   r     r  c                 C  r  r   r  r9   r   r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r9   r   r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r9   Z
less_equalr
  r   r   r   r     r  c                 C  r  r   r  r9   lessr
  r   r   r   r     r  c                 C  r  r   r  r9   Zgreater_equalr
  r   r   r   r     r  c                 C  r  r   r  r9   Zgreaterr
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r9   equalr
  r   r   r   r     r  c                 C  r  r   r  r9   	not_equalr
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r   r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r  r
  r   r   r   r     r  c                 C  r  r   r   r
  r   r   r   r     r  c                 C  r  r   )r  r9   Zbitwise_andr
  r   r   r   r     r  c                 C  r  r   )r  r9   Zbitwise_xorr
  r   r   r   r     r  c                 C  r  r   )r  r9   Z
bitwise_orr
  r   r   r   r     r  c                 C  s&   t |jt|j|j |j |jjS r   )r   r   r9   r  r   rB   r
  r   r   r   r    s   &zInterpreterBuilder.create_idivc                 C  sD   t |jj}t |jj}|j||_|j||_| ||tjS r   )rn   r   r   r@   r  r9   r  )r   r  r  Z	lhs_dtypeZ	rhs_dtyper   r   r   create_ashr	  s
   zInterpreterBuilder.create_ashrc                 C  s   |j j}|tjks|tjkrtt|j |j |jjS ttd|j	d d  }|j 
|}|j 
|}t|||j	d ? }t|
||jjS )Nr~   r6   r   )r   r   r9   rm   rA   r   np_umulhi_u64rB   r   rT   r@   r  )r   r  r  r   Zcompute_dtypeZlhs_dataZrhs_dataZret_datar   r   r   create_umulhi  s   z InterpreterBuilder.create_umulhic                 C  s   t ||j|j|j|jjS r   r  )r   r  r  r   r  r   r   r   
ternary_op     zInterpreterBuilder.ternary_opc                 C     |  |||tjS r   )r%  r9   Zclip)r   arglohiZpropagate_nansr   r   r   r          c                 C  r'  r   )r%  r9   where)r   condr  r  r   r   r   r   !  r+  c                 C  s   t |j|j |j |jjS r   r  r   r   r   r   
create_fma#  r&  zInterpreterBuilder.create_fmac                 C  s   t ||j|jjS r   r  )r   r(  r  r   r   r   unary_op'  r   zInterpreterBuilder.unary_opc                 C  sZ   |j }|jd }ttd|j }|j|}d|> d }||@ t|}t||j jS )Nr	   r~   )	r   r8   r   r9   r   r   r}   r   rB   )r   r(  rD   Zmask_bitwidthZnp_uint_dtyper   r   r   r   r   r   create_fabs*  s   
zInterpreterBuilder.create_fabsc                 C     |  |tjS r   )r/  r9   cosr   r(  r   r   r   r   4      c                 C  r1  r   )r/  r9   expr3  r   r   r   r   5  r4  c                 C  r1  r   )r/  r9   Zexp2r3  r   r   r   r   6  r4  c                 C  r1  r   )r/  r9   absr3  r   r   r   r   7  r4  c                 C  r1  r   )r/  r9   floorr3  r   r   r   r   8  r4  c                 C  r1  r   )r/  r9   ceilr3  r   r   r   r   9  r4  c                 C  r1  r   )r/  r9   logr3  r   r   r   r   :  r4  c                 C  r1  r   )r/  r9   log2r3  r   r   r   r   ;  r4  c                 C  r1  r   r/  r9   sqrtr3  r   r   r   r   <  r4  c                 C  r1  r   r;  r3  r   r   r   r   =  r4  c                 C  r1  r   )r/  r9   sinr3  r   r   r   r   >  r4  c                 C  s0   |j jtjkrt|j nt|j }t||jjS r   )r   r   r9   rt   np_erf_fp32np_erf_fp64r   rB   )r   r(  r   r   r   r   
create_erf@  s   "zInterpreterBuilder.create_erfc                 C  s   t dt|j |jjS Nr	   )r   r9   r<  r   r   rB   r3  r   r   r   create_rsqrtD  r   zInterpreterBuilder.create_rsqrtc                 C  s   t |j||jjS r   )r   r   r?   r   rB   )r   r(  r0   Zallow_reorderr   r   r   r   H      c                 C     t t|j||jjS r   )r   r9   	transposer   r   rB   )r   r(  permr   r   r   create_transJ  r  zInterpreterBuilder.create_transc                 C  s   |j }|j }|jjdkr|j s|jjdkr6|j r6t||jtjd tj}t||jtjd tj}t	tj
|||j jd|j  |jjS )Nr6   r7   )r   r   r8   Zis_floatingr   rp   rs   r   r9   r   matmulrB   )r   r   r   dZinput_precisionZmax_num_imprecise_accZa_dataZb_datar   r   r   
create_dotM  s   
$zInterpreterBuilder.create_dotc                 C  s   t tj||tjdtjS r   )r   r9   r>   rl   rp   )r   ret_tystartstopr   r   r   create_make_rangeV  r   z$InterpreterBuilder.create_make_rangec                 C  sz   |d u rt tj|jtdtj}t|j|jt|j}tj	||d|fdd }|d  t
|j 8  < t |tjS )Nr7   r   )binsr<   )r   r9   r   r   r   rp   rr   r,  r   	histogramZlogical_notsumrl   )r   r   rO  r   rP  r   r   r   create_histogramY  s   z#InterpreterBuilder.create_histogramc                 C  s   t tj|j|j|d|jjS )Nr   )r   r9   Ztake_along_axisr   r   rB   )r   r   indicesr   r   r   r   create_gatherc  s   z InterpreterBuilder.create_gatherc                 C  s<   |  }|j}td|d }t|j||jtj  |jS )Nr	   r6   )	r   r8   maxr   r   r@   r9   rA   r   )r   r   offsetrD   Zelement_bitwidthZelement_bytewidthr   r   r   create_addptrh  s    z InterpreterBuilder.create_addptrc                 C  s   | |\}}| }	t|	}
|d u rd }n.|tjjkr(ttj|j	|
d|	}n|tjj
kr=ttj|j	td|
d|	}ntd| | ||||||S )Nr7   nanzunsupported padding option )rJ   r   r}   r   ZPADDING_OPTIONZPAD_ZEROr   r9   r   r   ZPAD_NANZ	full_likefloatr   r   )r   r   rC   Zpadding_optionr   r   r   rE   rF   rD   r   r   r   r   r   create_tensor_pointer_loado  s   z-InterpreterBuilder.create_tensor_pointer_loadc                 C  s    | |\}}| |||||S r   rJ   r   )r   r   r$   rC   r   r   rE   rF   r   r   r   create_tensor_pointer_store~     z.InterpreterBuilder.create_tensor_pointer_storec                 C  rD  r   )r   r9   expand_dimsr   r   rB   )r   r(  r   r   r   r   create_expand_dims  r  z%InterpreterBuilder.create_expand_dimsc                 C  rD  r   )r   r9   r:   r   r   rB   )r   r(  r0   r   r   r   create_broadcast  r  z#InterpreterBuilder.create_broadcastc                 C  s   t t|j|jg|jjS r   )r   r9   Zconcatenater   r   rB   r
  r   r   r   
create_cat  r&  zInterpreterBuilder.create_catc                 C  s    t tj|j|jgdd|jjS )NrP   rS  )r   r9   stackr   r   rB   r
  r   r   r   create_join  s    zInterpreterBuilder.create_joinc                 C  s(   t |jd |jjt |jd |jjfS )N).r   ).r	   r  )r   r   r   r   r   create_split  s   (zInterpreterBuilder.create_splitc                 C  s\   |j }t|jtjrttj||jd t	|jd|jj
S ttj||jt	|jd|jj
S r   )r0   ro   r   rp   r|   r   r9   fullr   r}   rB   )r   rK  r(  r0   r   r   r   create_splat  s   &"zInterpreterBuilder.create_splatc                 C  sB   || j vrtd| | j | }tt|j|j|j||jjS )Nunsupported semantic )ir_sem_to_interpreter_semr   r   r   Z
atomic_casr   r   rB   )r   r   cmpr   semscoper   r   r   create_atomic_cas  s   

 z$InterpreterBuilder.create_atomic_casc                 C  sf   || j vrtd| || jvrtd| | j | }| j| }tt||j|j|j||jjS )Nzunsupported rmwOp rh  )	ir_rmw_op_to_interpreter_rmw_opr   ri  r   r   Z
atomic_rmwr   r   rB   )r   ZrmwOpr   r   r   rk  rl  r   r   r   create_atomic_rmw  s   



"z$InterpreterBuilder.create_atomic_rmwc                 C     t d)Nz4extern_elementwise not supported in interpreter modeNotImplementedError)r   ZlibNameZlibPathsymbolZargListZretTypeisPurer   r   r   create_extern_elementwise     z,InterpreterBuilder.create_extern_elementwisec                 C  rp  )Nz,inline_asm not supported in interpreter moderq  )r   Z	inlineAsmconstraintsvaluesr   rt  packr   r   r   create_inline_asm  rv  z$InterpreterBuilder.create_inline_asmc                 C  s   d| j d  d| j d  d| j d  d}|r|d| 7 }|r*tjdd	d
 id |D ]}t|d|j   q,|rCtjd d d S d S )N(r   z, r	   r   ) r   c                 S  s   d| dS )N0x02xr   r   r   r   r   r     r   z1InterpreterBuilder.create_print.<locals>.<lambda>)	formatter)r   r9   Zset_printoptionsprintr   )r   prefixhexrx  ZisSignedmsgr$   r   r   r   create_print  s   *zInterpreterBuilder.create_printc                 C  s   |sJ | d S r   r   )r   	conditionmessager   r   r   create_assert  s   z InterpreterBuilder.create_assertc                 C  s   |sJ dd S )NzAssume failedr   )r   r  r   r   r   create_assume  r   z InterpreterBuilder.create_assumec                 C  s   d S r   r   r   r   r   r   create_barrier  s   z!InterpreterBuilder.create_barrierc                 C  s    dd |D }t ||||||S )Nc                 S     g | ]}|  qS r   r!   .0rW  r   r   r   
<listcomp>      z<InterpreterBuilder.create_make_block_ptr.<locals>.<listcomp>)r.   )r   r/   r0   r1   r2   r3   r4   new_offsetsr   r   r   create_make_block_ptr  s   z(InterpreterBuilder.create_make_block_ptrc                 C  sv   t |jt |krtddd |jD }t|j|j|j||j|j}t	t |D ]}|j|  j
|| j
7  _
q)|S )Nz len(ptr.offsets) != len(offsets)c                 S  r  r   r  r  r   r   r   r    r  z5InterpreterBuilder.create_advance.<locals>.<listcomp>)r=   r2   r   r.   r/   r0   r1   r3   r4   r<   r   )r   r   r2   r  r   r   r   r   r   create_advance  s   z!InterpreterBuilder.create_advancer/   r   r0   rL   r1   tensor_shaperM   r   r   c                 C  s   t ||||}|  |S r   )rK   rS   )r   r/   r0   r1   r  r   descr   r   r   create_make_tensor_descriptor  s   z0InterpreterBuilder.create_make_tensor_descriptorr  rK   rT  c                 C  s2   t |tsJ ||\}}| j||d ||ddS )NF)r   r   r   r   )ro   rK   rJ   r   )r   r  rT  r   r   rE   r   r   r   r   create_descriptor_load  s
   z)InterpreterBuilder.create_descriptor_loadr$   c                 C  s    | |\}}| |||d d S r   r\  )r   r  r$   rT  rE   r   r   r   r   create_descriptor_store  r^  z*InterpreterBuilder.create_descriptor_store	x_offsetsy_offsetc                 C  s   |j jj}t|}tj|jjd |jd g|d}d }d }	t	|jD ]\}
}t
|tj|g}| ||||	j||
d d f< q"t
||S )Nr   rP   r7   )r/   r   r   r}   r9   zerosr   r0   r3   	enumerater   rp   rl   r  )r   r  r  r  r   r   np_dtyperesultr   r   r   x_offsetrT  r   r   r   create_descriptor_gather  s   
  
z+InterpreterBuilder.create_descriptor_gatherc           	      C  sH   t |jD ]\}}t|j| |j}t|tj|g}| ||| qd S r   )r  r   r   r   rp   rl   r  )	r   r  r$   r  r  r   r  slicerT  r   r   r   create_descriptor_scatter  s
   z,InterpreterBuilder.create_descriptor_scatterc                 C  sZ   t |}d|jv rttjdd|d|jS |tjkr&ttjdd|d|jS td| )Nrd   r	   rP   r7   Tzunsupported type )r}   namer   r9   rf  rB   r   	TypeError)r   r   Znp_typer   r   r   get_all_ones_value  s   

z%InterpreterBuilder.get_all_ones_valueNr   r   )
r/   r   r0   rL   r1   rL   r  rM   r   r   )r  rK   rT  rL   )r  rK   r$   r   rT  rL   )r  rK   r  r   r  r   )r  rK   r$   r   r  r   r  r   )r&   r'   r(   r   ZMEM_SEMANTICZACQUIREr   ZRELEASEZRELAXEDZACQUIRE_RELEASEri  Z	ATOMIC_OPZADDZRMW_OPZFADDZMINZUMINMAXZUMAXANDORZXORZXCHGrn  r5   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   Zcreate_si_to_fpZcreate_ui_to_fpZcreate_fp_to_siZcreate_fp_to_uiZcreate_fp_extZcreate_fp_truncZcreate_int_castr   r   r  Zcreate_faddZcreate_fmulZcreate_fdivZcreate_fremZcreate_fsubZ
create_mulZcreate_precise_divfZcreate_sdivZcreate_udivZcreate_sremZcreate_uremZ
create_addZ
create_subZ
create_shlZcreate_lshrZcreate_minsiZcreate_minuiZcreate_minimumfZcreate_minnumfZcreate_maxsiZcreate_maxuiZcreate_maximumfZcreate_maxnumfZcreate_icmpSLEZcreate_icmpSLTZcreate_icmpSGEZcreate_icmpSGTZcreate_icmpULEZcreate_icmpULTZcreate_icmpUGEZcreate_icmpUGTZcreate_icmpEQZcreate_icmpNEZcreate_fcmpOLTZcreate_fcmpOGTZcreate_fcmpOLEZcreate_fcmpOGEZcreate_fcmpOEQZcreate_fcmpONEZcreate_fcmpULTZcreate_fcmpUGTZcreate_fcmpULEZcreate_fcmpUGEZcreate_fcmpUEQZcreate_fcmpUNEZ
create_andZ
create_xorZ	create_orZcreate_int_to_ptrZcreate_ptr_to_intr  r"  r$  r%  Zcreate_clampfZcreate_selectr.  r/  r0  Z
create_cosZ
create_expZcreate_exp2Zcreate_iabsZcreate_floorZcreate_ceilZ
create_logZcreate_log2Zcreate_precise_sqrtZcreate_sqrtZ
create_sinr@  rB  Zcreate_reshaperG  rJ  rN  rR  rU  rX  r[  r]  r`  ra  rb  rd  re  rg  rm  ro  ru  rz  r  r  r  r  r  r  r  r  r  r  r  r  r   r   r   r   r     sZ   	

	
	
r   c                   s*   t | |d fdd
}t| || d S )N)memberc                   s$   | |i dd |  D d iS )Nc                 S  s   i | ]\}}|d kr||qS )r   r   r  kvr   r   r   
<dictcomp>  s    z1_patch_attr.<locals>.<lambda>.<locals>.<dictcomp>r   items)r  argskwargssemanticr   r   r     s    z_patch_attr.<locals>.<lambda>)r   setattr)objr  r  r   
new_memberr   r  r   _patch_attr  s   r  c                 C  s2   t | D ]\}}tj|rt| ||| qd S r   )inspect
getmembersrp   core
is_builtinr  )pkgr   r  r  r   r   r   _patch_builtin  s
   r  c                   sJ   dd  dd }dd | _  fdd| _dd | _d	d | _t|| _d S )
Nc                 S  s   | j j}|jdkrt|S dS )Nr	   T)r   r   sizer   )r   r   r   r   r   	_get_bool  s   z%_patch_lang_tensor.<locals>._get_boolc                 S  sj   t t| jj| jj}| j sJ t| jj	}|d |d |d< |d< t
j| j|}t
j||S )NrP   )r   r9   rE  r   r   r   r   Zis_blocklistr0   rp   r  r|   r   )r   r   r3   Zres_tyr   r   r   _get_transpose"  s   z*_patch_lang_tensor.<locals>._get_transposec                 S     t | jjS r   )rd   r   r   r   r   r   r   r   *  r   z$_patch_lang_tensor.<locals>.<lambda>c                   s    | S r   r   r   r  r   r   r   +  s    c                 S  r  r   )reprr   r   r   r   r   r   r   ,  r   c                 S  r  r   )rZ   r   r   r   r   r   r   r   -  r   )	__index__r   __repr____str__propertyT)r   r  r   r  r   _patch_lang_tensor  s   


r  c                   @  s4   e Zd Zdd Zdd Zdd Zdd Zd	d
 ZdS )ReduceScanOpInterfacec                 C  s   || _ || _d S r   )r   
combine_fn)r   r   r  r   r   r   r5   3  s   
zReduceScanOpInterface.__init__c                 C  s0   |d ur|t |krtd| d| d S d S )Nzaxis z out of bounds for shape )r=   r   )r   r0   r   r   r   r   
check_axis7  s   z ReduceScanOpInterface.check_axisc                 C  s>   |D ]}t |tjjstdt| | |j| j qd S )Nzinput must be a tensor, got )	ro   rp   r  r   r   r   r  r0   r   )r   r   r(  r   r   r   check_tensor;  s
   z"ReduceScanOpInterface.check_tensorc                 C  s`   t |}t|dr|jr||}t|t|j}n
tj|g|d}|}tj	
t||j|S )Nr0   r7   )r}   r   r0   r@   rp   r|   r  r9   r   r  r   r   rB   )r   r   r   r  Zret_typer   r   r   	to_tensorA  s   
zReduceScanOpInterface.to_tensorc                 C  sJ   t |ts| |fd S | | | |}t |ttfr"t|S |fS Nr   )ro   tupleapplyr  
apply_implr  )r   r   r   r   r   r   r  K  s
   


zReduceScanOpInterface.applyN)r&   r'   r(   r5   r  r  r  r  r   r   r   r   r  1  s    
r  c                      sF   e Zd Z fddZdd Zdd Zddd	Zd
d Zdd Z  Z	S )	ReduceOpsc                      t  || || _d S r   )superr5   	keep_dims)r   r   r  r  	__class__r   r   r5   U     
zReduceOps.__init__c                 C  sN   g }|D ]}|d ur| | qd}| | |jj |j qt||fS r  )appendr  r   r   flattenr   r  )r   r   r   r   r   r   r   r   unravelY  s   zReduceOps.unravelc                   s  j } j \ }g }g } d jjj}|d| ||d d   } D ]}||jj |tj||jjjd q't	|d j
D ]}	t|	|d| |d d   t fddt|D }
| dkrt	t|D ]}|
| jj || < qvqEt fddt|D }jjg ||
R  }t|ts|fn|}t	t|D ]}t|| tjjr|| jj n|| || < qqEg }t|D ]6\}	}jr|d urt||}qt	t|D ]}t|d}qn|d u r| }|| |	 j q|S )Nr   r	   r7   c                 3  *    | ]\}} |  | jV  qd S r   r  r   r  iirI  )r   input_indexr   r   r   	<genexpr>r     ( z+ReduceOps.generic_reduce.<locals>.<genexpr>c                 3  r  r   r  r  Zoio)r   output_indexr   r   r   r  x  r  )r   r  r   r   r0   r  r9   r  r   r<   r  unravel_indexr  r  r=   rQ   r  fnro   rp   r  r   r  r_  r  )r   r   Zoriginal_axisr   
input_dataoutput_dataZinput_shapeZoutput_shaper(  r   Zinput_tuplej	acc_tuplecombine_fn_retr   r   _r   )r   r  r  r   r   generic_reducec  sN   zReduceOps.generic_reduceNc                 C  s   t |tr	|d n|}d }d }|r!| ||jj| j| jd|j}|r3| ||jj| j| jdtj	}|d ur?|d ur?||fS |d urE|S |d urK|S t
d)Nr   r   Zkeepdimsz-val_reduce_op and idx_reduce_op are both None)ro   r  r  r   r   r   r  r   rp   rl   r   )r   r   val_reduce_opidx_reduce_opr   idxr   r   r   min_max  s     zReduceOps.min_maxc                 C  s"   |  tj|jj| j| jd|jS )Nr  )r  r9   rQ  r   r   r   r  r   r   r   r   r   r   rQ       "zReduceOps.sumc                 C  s   | j tjjkr| j|d tjtjdS | j tjjkr&| j|d tj	tj
dS | j tjjkr8| j|d tj	d dS | j tjjkrJ| j|d tjd dS | j tjjkrX| |d S | |S )Nr   )r  r  )r  rp   standardZ_argmin_combine_tie_break_leftr  r9   minZargminZ_argmax_combine_tie_break_leftrV  ZargmaxZ_elementwise_maxZ_elementwise_min_sum_combinerQ  r  r  r   r   r   r    s   
zReduceOps.apply_implr   )
r&   r'   r(   r5   r  r  r  rQ  r  __classcell__r   r   r  r   r  S  s    

+r  c                      s<   e Zd Z fddZdd Zdd Zdd Zd	d
 Z  ZS )ScanOpsc                   r  r   )r  r5   reverse)r   r   r  r  r  r   r   r5     r  zScanOps.__init__c                 C  "   | j tj|jj| jd|jdgS NrS  r7   )r  r9   cumsumr   r   r   r   r  r   r   r   r    r  zScanOps.cumsumc                 C  r   r  )r  r9   cumprodr   r   r   r   r  r   r   r   r    r  zScanOps.cumprodc                   s  g }g }d j jj}D ]}||j j |tj||j jjd qt|d jD ]}t	|| t
 fddt|D } j dkr_tt|D ]}|| j j ||  < qOq+t
 fddtt D t
fddt|D }	jjg |	|R  }
t|
t
s|
fn|
}	tt|D ]}t|	| tjjr|	| j j n|	| ||  < qq+g }t|D ]\}}||| j q|S )Nr   r7   c                 3  s*    | ]\}} |  | jV  qd S r   r  r  )indexr   r   r   r   r    r  z'ScanOps.generic_scan.<locals>.<genexpr>c                 3  s.    | ]}|j kr | d  n | V  qdS )r	   NrS  )r  r   )r  r   r   r   r    s   , c                 3  r  r   r  r  )r   
prev_indexr   r   r   r    r  )r   r   r0   r  r9   r  r   r<   r  r  r  r  r   r=   rQ   r  r  ro   rp   r  r   r  )r   r   r  r  r0   r(  r   r   r  r  r  r   r   )r  r   r  r   r   generic_scan  s8    zScanOps.generic_scanc              	   C  s   g }| j r|D ]}|| tj|jj| jd|j qn|}| j	t
jjkr.| |d }n| j	t
jjkr=| |d }n| |}| j rV|D ]}tj|jj| jd|j_qG|S )NrS  r   )r  r  r  r9   flipr   r   r   r   r  rp   r  r  r  Z_prod_combiner  r  )r   r   Z	new_inputr(  r   r   r   r   r    s   &
zScanOps.apply_impl)	r&   r'   r(   r5   r  r  r  r  r  r   r   r  r   r    s    r  c                  C  s4   ddd} ddd}| t _|t _| t j_|t j_d S )NFc                 [     t |||| S r   )r  r  )r   r   r  r  r  r   r   r   _new_reduce  r"   z'_patch_reduce_scan.<locals>._new_reducec                 [  r  r   )r  r  )r   r   r  r  r  r   r   r   	_new_scan  r"   z%_patch_reduce_scan.<locals>._new_scan)F)rp   reduceZassociative_scanr  )r	  r
  r   r   r   _patch_reduce_scan  s   

r  c                 C  sx   dd }ddd}ddd}dd	 }|| _ || _|| _t| _|| j_t|d
d| _t|dd| _	t|dd| _
t  d S )Nc                 S  sB  | j dkr	| S | j dkr| S | j dkr| S | j dkr$| S | j dkr-| S | j dkr6| S | j dkr?| S | j dkrH| S | j d	krQ|	 S | j d
krZ|
 S | j dkrc| S | j dkrl| S | j dkru| S | j dkr~| S | j dkr| S | j dkr| S | j dkr| S td|  d)Nvoidrr   rh   rg   rj   ri   rl   rk   rm   rA   r\   r]   r^   Zfp16Zbf16Zfp32Zfp64zfail to convert z to ir type)r  Zget_void_tyr   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   )r   r   r   r   r   
_new_to_ir  sF   
















z$_patch_lang_core.<locals>._new_to_irc                 [  s6   |d u rd}|d u rd| }}n| |}}t |||S )Nr	   r   )r<   )Zarg1Zarg2stepr  rL  endr   r   r   
_new_range+  s   
z$_patch_lang_core.<locals>._new_range c                 S  s   | sJ |d S r   r   )r-  r  r   r   r   _new_static_assert4  r   z,_patch_lang_core.<locals>._new_static_assertc                 S  sn   t | tjs| S t |ttfs|gn|}dd |D }t|tdt| jkr.td| | j	
|| | S )Nc                 S  s"   g | ]}t |tjr|jn|qS r   )ro   rp   	constexprr$   r  r  r   r   r   r  =  s   " z7_patch_lang_core.<locals>._set_attr.<locals>.<listcomp>r	   z$len(values) != len(input.shape) for )ro   rp   r   r  r  r=   rV  r0   r   r   r%   )r   rx  r  r   r   r   	_set_attr7  s   z#_patch_lang_core.<locals>._set_attrztt.divisibilityr  ztt.contiguityztt.constancy)NN)r  )r<   Zstatic_rangeZstatic_assertr  Zstatic_printr   Zto_irr   Zmultiple_ofZmax_contiguousZmax_constancyr  )langr  r  r  r  r   r   r   _patch_lang_core  s   
(
	
r  c                 C  s   dd | j  D }t|dksJ d|D ] }t|t t|jt |tkr-t|jt t|j t	| qttj
jt d S )Nc                 S  s,   g | ]\}}t |r|ttjfv r|qS r   )r  ismodulerp   r  )r  r  r$   r   r   r   r  P  s   , z_patch_lang.<locals>.<listcomp>r	   z:triton.language must be visible from within jit'd function)__globals__r  r=   r  interpreter_builderr   rp   r   r  r  r  Ztensor_descriptor_base)r  Zlangsr  r   r   r   _patch_langO  s   


r  c                 C  s"   t | drt| | S t| |S )N_fields)r   r   )r(  contentsr   r   r   _tuple_create\  s   "r   c                 C  s  t | trjttjj| }tj	}d|   krdk r"n ntj	}n7d|   kr,dk r2n ntj
}n'd|   kr<dk rBn ntj}nd|   krLdk rRn ntj}ntd|  ttj| g|d|}t||S t| d	rttjj| }ttj|  gtjd|}t||S t | trt| tt| S t | trd
d | jD }| jd dksJ td|d< tt }|jt| jdd | jD |dd | jD dS | S )Ni   l        l        l         l            l            zUnsupported integer value r7   data_ptrc                 S     g | ]}t |qS r   _implicit_cvtr  sr   r   r   r  }  r  z!_implicit_cvt.<locals>.<listcomp>rP   r	   c                 S  r"  r   r#  r%  r   r   r   r    r  c                 S  s   g | ]}t |qS r   )rp   r  )r  r   r   r   r   r    rC  )r/   r0   r1   r3   ) ro   rd   rp   Z	str_to_tytritonruntimejitZmangle_typer9   rl   rk   rm   rA   r   r   r   r   r   r!  r  r   mapr$  r   r1   r  r   r   Zmake_tensor_descriptorr/   r0   r3   )r(  tyr   r   r1   r  r   r   r   r$  f  s@   




r$  c                 C  s   t | tjjjr| jS | S r   )ro   r'  r(  r)  TensorWrapperr/   )tr   r   r   _unwrap_tensor  s   r.  c                 C  s&   t |tjjjrtjj| |jS | S r   )ro   r'  r(  r)  r,  r   )r-  original_tensorr   r   r   _rewrap_tensor  s   r0  c                   @  s,   e Zd Zdd Zdd Zdd Zdd Zd	S )
GridExecutorc                   sN   ddl m || _|| _|| _fdd|j D   fdd|D | _d S )Nr	   _normalize_tyc                   s   i | ]	\}}| |qS r   r   )r  r  r+  r2  r   r   r    s    z)GridExecutor.__init__.<locals>.<dictcomp>c                   s   g | ]}  |d kr|qS )r  )get)r  r  )r*   r   r   r    s    z)GridExecutor.__init__.<locals>.<listcomp>)r)  r3  r  	arg_namesgridr*   r  
constexprs)r   r  r5  r6  r   )r*   r3  r   r5     s   zGridExecutor.__init__c                   sN   i  fdd  fdd|D }i }|  D ]
\}} |||< q||fS )Nc                   s   t | trt| t | S t | trt | j| j| j| jS t	| ds&| S t
| }|  vr>| }| | < |   }|jddd}||| | |  t|| d}|S )Nr!  r   cpu)Zdevice)r/  )ro   r  r   r*  r   r/   r0   r1   r3   r   r.  untyped_storager!  r8  Z	new_emptyZset_Zstorage_offsetr  rR   r0  )r(  Zunwrapped_argZstorageZcpu_arg_to_cpustoragesr   r   r;    s(   


z,GridExecutor._init_args_hst.<locals>._to_cpuc                   s   g | ]} |qS r   r   )r  r(  )r;  r   r   r    r  z/GridExecutor._init_args_hst.<locals>.<listcomp>r  )r   args_devr  args_hst
kwargs_hstr#   r$   r   r:  r   _init_args_hst  s   zGridExecutor._init_args_hstc           
        st   i  fdd t ||D ]	\}} || q| D ]\}}|| }	 ||	 q D ]	\}}|| q.d S )Nc                   s   t | drt| t|} }|  | f|   < d S t| tr4t| |D ]	\} } | | q(d S t| trB | j|j d S d S )Nr!  )	r   r.  r9  r!  ro   r  zipr   r/   )arg_devarg_hst	_from_cpur<  r   r   rE    s   
 

z1GridExecutor._restore_args_dev.<locals>._from_cpu)rA  r  rx  Zcopy_)
r   r=  r>  r  r?  rB  rC  r#   Z	kwarg_devZ	kwarg_hstr   rD  r   _restore_args_dev  s   zGridExecutor._restore_args_devc              
     sh  | ddrd S tj  fdd| D }||\}}tj tjjg|R i |}fdd| D }tj	rH	|nj	}t
|dksUJ d|ddt
|   }tj|  z,t|d	 D ]#}t|d
 D ]}t|d D ]}	t|||	 jdi | q{qsqkW n ty }
 ztjjjr tt|
|
d }
~
ww |||| d S )NZwarmupFc                   s    i | ]\}}| j v r||qS r   )r  r  )argspecr   r   r    s     z)GridExecutor.__call__.<locals>.<dictcomp>c                   s(   i | ]\}}|| j v r|nt|qS r   )r7  r$  )r  r  r(  r   r   r   r    s   (    z#grid must have at most 3 dimensions)r	   r   r	   r   r   )popr  getfullargspecr  r  r@  r  getcallargscallabler6  r=   r  r   r<   r   	Exceptionr'  ZknobsZcompilationZfront_end_debuggingr
   r  rF  )r   r=  r  r>  r?  r  r6  r   r   r   er   )rG  r   r   __call__  s8   


zGridExecutor.__call__N)r&   r'   r(   r5   r@  rF  rO  r   r   r   r   r1    s
    	#r1  c                   @  s   e Zd Zdd ZdS )ASTTransformerc                 C  sv   g }|j D ]
}|| |g7 }qt|dkrtdtjtjtjdt ddt d|j	tj
ddgg d	|_	|S )
Nr	   z&Multiple assignments are not supportedinterpreter_semantic)idctxr  )r$   r   rS  F)r$   )funcr  keywords)targetsvisitr=   r   astCall	AttributeNameLoadr$   Constant)r   nodenamestargetr   r   r   visit_Assign  s   
zASTTransformer.visit_AssignN)r&   r'   r(   ra  r   r   r   r   rP    s    rP  c                   @  sJ   e Zd Ze Zdd Zdd Zdd Zdd Zd	d
 Z	dd Z
dd ZdS )FunctionRewriterc                 K  s   || _ || _d| _d| _d S )Nr  r   )r  r  filenamedef_file_lineno)r   r  r  r   r   r   r5     s   
zFunctionRewriter.__init__c                 C  sh   z
t | j\}}W n ty   | j Y S w |  \| _| _| || _| 	|}| 
|}| |S r   )r  getsourcelinesr  rM  _get_jit_fn_file_linerc  rd  	_find_def
def_lineno_prepare_source_transform_ast_compile_and_exec)r   linesr  r   transformed_astr   r   r   rewrite_ast  s   
	


zFunctionRewriter.rewrite_astc                 C  s   ddl m}m} ||| jS )Nr	   )get_jit_fn_file_lineJITFunction)r)  ro  rp  r  )r   ro  rp  r   r   r   rf  2  s   z&FunctionRewriter._get_jit_fn_file_linec                 C  s0   d}t |D ]\}}| dr|d }q|S )Nr   zdef r	   )r  strip
startswith)r   rl  rh  r   liner   r   r   rg  6  s   zFunctionRewriter._find_defc                 C  s&   || j d d  }d|}t|S )Nr	   r  )rh  jointextwrapdedent)r   rl  r   r   r   r   ri  >  s   

z FunctionRewriter._prepare_sourcec                 C  s:   t |}| j|}t | | jd }t || |S rA  )rX  parseast_transformerrW  fix_missing_locationsrd  increment_lineno)r   r   Z
parsed_astrm  Z
inc_linenor   r   r   rj  C  s   


zFunctionRewriter._transform_astc                 C  s^   t || jdd}i | j}| jj}t  D ]\}}||vr"|||< qt||| || jj S )Nexec)rc  mode)	compilerc  r  r  r  globalsr  r{  r&   )r   rm  Zcompiled_codeZlocal_namespaceZ
fn_globalsr#   r$   r   r   r   rk  N  s   
z"FunctionRewriter._compile_and_execN)r&   r'   r(   rP  rx  r5   rn  rf  rg  ri  rj  rk  r   r   r   r   rb    s    rb  c                   @  s>   e Zd Zi ZdddZdd Zedd Z d	d
 Zdd ZdS )InterpretedFunctionr   r   c                   sN   | _ t|fi | _ fdd}| _t|}dd |j D  _d S )Nc                    s(   |d }   }t| j|| i |S )Nr6  rewriter1  r5  )r  r  r6  r  r   r   r   runa  s   z)InterpretedFunction.__init__.<locals>.runc                 S  s   g | ]}|j qS r   r  r  r   r   r   r  h  r+  z0InterpretedFunction.__init__.<locals>.<listcomp>)	r  rb  rewriterr  r  	signature
parametersrx  r5  )r   r  r  r  r  r   r   r   r5   ]  s   
zInterpretedFunction.__init__c                 C  s*   | j | jvr| j | j| j < | j| j  S r   )r  rewritten_fnr  rn  r   r   r   r   r  j  s   zInterpretedFunction.rewritec                 C  s   | j jS r   )r  r&   r   r   r   r   r&   o  s   zInterpretedFunction.__name__c                 C  s   |   }t|| j|S r   r  )r   r6  r  r   r   r   __getitem__s  s   zInterpretedFunction.__getitem__c              
   O  sJ   t | j |  }z||i |W S  ty$ } ztt||d }~ww r   )r  r  r  rM  r
   r  )r   r  r  r  rN  r   r   r   rO  w  s   
zInterpretedFunction.__call__Nr  )	r&   r'   r(   r  r5   r  r  r  rO  r   r   r   r   r  Y  s    

r  )E
__future__r   rX  ru  r  typingr   r   r   r   numpyr9   r'  Ztriton.languagelanguagerp   r+   r   Ztriton.language.semanticr   Ztriton.tools.tensor_descriptorr   errorsr
   	functoolsr   Z_C.libtritonr   r   r   r   r   r.   rK   rV   rn   r}   r   r   r   Z	vectorizert   r>  ru   r?  rA   r#  r   r   r  r  r  r  r  r  r  r  r  r   r$  r  rQ  r.  r0  r1  NodeTransformerrP  rb  r  r   r   r   r   <module>   sp    & @    	"`>N
$jE