
    kh                        d dl Z d dlZd dlZd dlmZmZ d dlZd dl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  G d
 d      Z G d d      Z G d d      Z ed       G d d             Zd Zd Zd Zd Zd Z  ejB                  eejD                  g      Z# ejB                  eejH                  g      Z% ejB                  e ejL                  g      Z' G d d      Z( G d d      Z)d Z*d Z+d  Z, G d! d"      Z- G d# d$e-      Z. G d% d&e-      Z/d' Z0d( Z1d) Z2d* Z3d+ Z4 e)       Z5d, Z6d- Z7 G d. d/      Z8 G d0 d1e jr                        Z: G d2 d3      Z; G d4 d5      Z<y)6    N)TupleList)	dataclass   )InterpreterError)partial   )interpreter)irc                   *    e Zd Zd Zd Zd Zd Zd Zy)TensorHandlec                 .    || _         || _        i | _        y)a  
            data: numpy array
            dtype: triton type, either pointer_type or scalar_type.
            we don't store block_type here because the shape information is already available in the data field
            attr: a dictionary of attributes
        N)datadtypeattr)selfr   r   s      V/var/www/teggl/fontify/venv/lib/python3.12/site-packages/triton/runtime/interpreter.py__init__zTensorHandle.__init__   s     	
	    c                 H    t        | j                  j                               S N)boolr   allr   s    r   __bool__zTensorHandle.__bool__   s    DIIMMO$$r   c                 h    | j                   }t        |d      r|j                  }t        |d      r|S )N
element_ty)r   hasattrr   )r   r   s     r   get_element_tyzTensorHandle.get_element_ty"   s1    

e\*$$E e\*r   c                 ^    t        | j                  j                         | j                        S r   )r   r   copyr   r   s    r   clonezTensorHandle.clone(   s    DIINN,djj99r   c                 "    || j                   |<   y r   )r   )r   keyvalues      r   set_attrzTensorHandle.set_attr+   s    		#r   N)__name__
__module____qualname__r   r   r   r"   r&    r   r   r   r      s    	%:r   r   c                       e Zd Zd Zd Zy)BlockPointerHandlec                 X    || _         || _        || _        || _        || _        || _        y r   )baseshapestridesoffsetsblock_shapeorder)r   r.   r/   r0   r1   r2   r3   s          r   r   zBlockPointerHandle.__init__1   s-    	
&
r   c                 f   | j                   j                         }|j                  dz  }t        j                  | j                   j
                  | j                        }t        j                  | j                  t              }t        t        | j                              D ]  }dgt        | j                        z  }| j                  |   ||<   | j                  |   j
                  t        j                  | j                  |         z   j                  |      }|||z  | j                  |   j
                  z  j                  t        j                         z   }||v s||| j"                  |   j
                  k  z  |dk\  z  } t%        || j                   j&                  j(                        }||fS )N   r   r   r   )r.   r   primitive_bitwidthnpbroadcast_tor   r2   onesr   rangelenr1   arangereshaper0   astypeuint64r/   r   r   scalar)	r   boundary_checkdtype_ttn_bytesptrsmasksdim
bcast_dimsoffs	            r   materialize_pointersz'BlockPointerHandle.materialize_pointers9   sb   99++---2tyy~~t/?/?@((5T--./ 	JCs4#3#344J"..s3JsO<<$))BIId6F6Fs6K,LLUUV`aC7S=4<<+<+A+AAII"))TTDn$tzz#';';!;<qI	J D$))//"8"89U{r   N)r'   r(   r)   r   rJ   r*   r   r   r,   r,   /   s    r   r,   c            	       N    e Zd Zdedee   dee   dee   fdZd Zdee   fdZy	)
TensorDescHandler.   r/   r0   r2   c                 \    || _         t        |      | _        || _        || _        || _        y r   )r.   r<   ndimr/   r0   r2   )r   r.   r/   r0   r2   s        r   r   zTensorDescHandle.__init__K   s*    	J	
&r   c                    | j                   j                  j                         dz  dk(  sJ d       t        | j                        | j
                  k(  sJ t        | j                        | j
                  k(  sJ | j                  d d D ])  }|j                  j                         dz  dk(  r$J d        | j                  d   j                  j                         dk(  sJ d       y )N   r   zbase must be 16-byte alignedzstride must be 16-byte alignedr   zlast dim must be contiguous)r.   r   itemr<   r0   rN   r2   )r   strides     r   validatezTensorDescHandle.validateS   s    yy~~""$r)Q.N0NN.4<< DII---4##$		111ll3B' 	RF;;##%*a/Q1QQ/	R||B$$))+q0O2OO0r   r1   c                    t        |      | j                  k(  sJ | j                  j                  j                  }|j
                  dz  }|d   j                  |z  dz  dk(  sJ d       t        j                  | j                  j                  | j                        }t        j                  | j                  t              }t        t        | j                              D ]  }dgt        | j                        z  }| j                  |   ||<   ||   j                  t        j                  | j                  |         z   j                  |      }|||z  | j                  |   j                  z  j!                  t        j"                        z   }|d|k  z  || j$                  |   j                  k  z  } t'        || j                  j                  j(                        }||fS )Nr5   rQ   rP   r   z*block offset start must be 16-byte alignedr6   r   )r<   rN   r.   r   r   r7   r   r8   r9   r2   r:   r   r;   r=   r>   r0   r?   r@   r/   r   rA   )	r   r1   	scalar_tyitemsizerE   rF   rG   rH   rI   s	            r   rJ   z%TensorDescHandle.materialize_pointers\   s   7|tyy(((IIOO..	//14  8+r1Q6d8dd6tyy~~t/?/?@((5T--./ 	FCs4#3#344J"..s3JsO3<$$ryy1A1A#1F'GGPPQ[\C8c>DLL,=,B,BBJJ299UUDQ#X&#

30D0D*DEE	F D$))//"8"89U{r   N)	r'   r(   r)   r   r   intr   rT   rJ   r*   r   r   rL   rL   I   sJ    '\ '$|2D 'tT`Oa '"3i'PD,> r   rL   T)frozenc                       e Zd ZU dZeed<   dZeed<   dZeed<   dZ	e
ed<   dZee
   ed	<   d
Zee
   ed<   dZe
ed<   dZee
   ed<   dZeed<   dZe
ed<   y)InterpreterOptionsNextern_libsFdebugTsanitize_overflowarch)fp8e5fp8e5b16fp8e4nvfp8e4b8fp8e4b15supported_fp8_dtypesr*   deprecated_fp8_dtypestf32default_dot_input_precision)rg   tf32x3ieeeallowed_dot_input_precisionsr   max_num_imprecise_acc_defaultr
   backend_name)r'   r(   r)   r\   dict__annotations__r]   r   r^   r_   strre   r   rf   rh   rk   rl   rX   rm   r*   r   r   r[   r[   n   sx    KE4"t"D#'^%*^(*5:*'--/I %*I)*!3*%L#%r   r[   c                    | 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   )	r8   uint8int8uint16int16uint32int32r@   int64r6   s    r   _get_signed_np_dtypery   |   s[    ww		xx		xx		xxLr   c                 `   t        | t        j                        r#t        j                  t        j
                        S i t        j                  t        j                  t              t        j                  t        j                  t        j                        t        j                  t        j                  t        j                        t        j                  t        j                  t        j                        t        j                  t        j                  t        j                        t        j                  t        j                  t        j                        t        j                  t        j                  t        j                        t        j                  t        j                  t        j                        t        j                  t        j                  t        j                        t        j                   t        j                  t        j                         t        j"                  t        j                  t        j"                        t        j
                  t        j                  t        j
                        t        j$                  t        j                  t        j                        t        j&                  t        j                  t        j                        t        j(                  t        j                  t        j                        t        j*                  t        j                  t        j                        t        j,                  t        j                  t        j                        t        j.                  t        j                  t        j                        i}t        | t        j0                        rVt        | j2                  t        j                        r#t        j                  t        j
                        S || j2                     S ||    S r   )
isinstancetlpointer_typer8   r   r@   int1r   float16float32float64rs   rr   ru   rt   rw   rv   rx   bfloat16float8e5float8e5b16
float8e4nv
float8e4b8float8e4b15
block_typer   )tt_dtypenp_typess     r   _get_np_dtyper      s~   (BOO,xx		""
$


BHHRZZ( 	

BHHRZZ( 	

BHHRZZ(	
 	"''" 	"((288$ 	"((288$ 			288BII& 	"((288$ 			288BII& 	"((288$ 			288BII& 	RXXbii(  	RXXbhh'!" 	*#$ 	rxx)%& 	rxx)'( 	*)H, (BMM*h))2??;88BII&&++,,Hr   c                    t        t        d|j                         }t        t        d|j                         }t        j                  | j	                         |      }||j                  dz
  z	  dz  }|j                  |j
                  z
  dz
  }|j                  |j
                  z
  dz
  }	|d|j
                  z  dz
  z  }
|j                  }|j                  }||j
                  z	  d|z  dz
  z  j                  t        j                        }|dk(  }t        j                  |      rt        j                  |t        j                        }t        |j
                        D ]  }|
|z	  dz  }|j
                  |z
  ||dk(  <   ! |
dk(  }d||   z
  ||<   ||z
  |||z  <   |
|   ||   z  d|j
                  z  dz
  z  |
|<   t        j                  dt        j                  ||z
  |z   d|	z  dz
              }|j                  |      }|j                  |      }|j                  |j                  kD  r|
|j
                  |j
                  z
  z	  d|j
                  z  dz
  z  }|t        j                  j                   k(  r*|
d|j
                  |j
                  z
  dz
  z  z  }||dkD  z   }|j                  |      }n>|
j                  |      |j
                  |j
                  z
  z  d|j
                  z  dz
  z  }|dk(  }t        j                  |      r||j
                  z	  d|z  dz
  z  j                  t        j                        }|dk7  }||z  }t        j                  |t        j                        }d|z
  ||   |z
  z
  ||<   ||   ||   z	  d|j
                  ||   z
  z  z  ||<   ||j                  dz
  z  ||j
                  z  z  |z  }|j#                  | j$                        S )Nuintr6   r   r   )getattrr8   r7   
frombuffertobytesfp_mantissa_widthexponent_biasr?   rw   any
zeros_liker;   maximumminimum_irROUNDING_MODERTNEr>   r/   )inputinput_dtypeoutput_dtyperounding_modeinput_uint_dtypeoutput_unint_dtype	input_binsigninput_exponent_widthoutput_exponent_widthsignificand
bias_inputbias_outputexponentsubnormal_indexbit_posi	bit_indexzero_significand_indexexponent_outputsign_outputsignificand_outputcut_offnon_zero_exponent_indexshiftoutputs                             r   _convert_floatr      s+   rT+*H*H)I#JK tL,K,K+L%MNemmo5EFI+881<=ED&99K<Y<YY\]](;;l>\>\\_``[%B%B BaGHK**J,,Kk;;;FZAZ^_@_`hhikiqiqrH!mO	vvo
 --	:{445 	HA%*d2I&1&C&Ca&GGIN#	H "-!1$%(@$@!=G+=U'/9:(3O(DP_H`(`+///14(6O$ jjBJJ:0E0SWX\qWquvVv$wxO%,,-?@O++01K%%(G(GG)k.K.KlNlNl.lm,000A57C--222!Q;+H+H<KiKi+ilm+m%noG!3w{!C/667IJ)001CD+==@]@]]_#$(F(F#F!"KM &*O	vvo
 +"?"??QJ^E^bcDcdllmomumuv"*a-),CCirxx8"#k/h6OR\6\!]o/A//RV[\kVl/l,0053IIJ/L?+l==AB<999;=OPF>>%++&&r   c                 ,    t        j                  |       S r   )matherfxs    r   _erfr      s    88A;r   c                 6    t        |       t        |      z  dz	  S )N@   )rX   )abs     r   
_umulhi_64r      s     FSVO""r   )otypesc                       e Zd Zed        Zy)ExtraFunctionsc                 d    t        j                  |j                  | j                  ||      |      S r   )r|   tensorcreate_fp_to_fphandle)r   dst_tyfp_downcast_rounding_builders       r   _convert_custom_typesz$ExtraFunctions._convert_custom_types   s(    yy11%,,H\]_effr   N)r'   r(   r)   staticmethodr   r*   r   r   r   r      s    g gr   r   c                      e Zd Zej                  j
                  ej                  j
                  ej                  j                  ej                  j                  ej                  j                  ej                  j                  ej                  j                  ej                  j                  iZ
ej                  j                  ej                  j                  ej                  j                  ej                  j                  ej                  j                  ej                  j                  ej                  j                   ej                  j                   ej                  j"                  ej                  j"                  ej                  j$                  ej                  j$                  ej                  j&                  ej                  j&                  ej                  j(                  ej                  j(                  ej                  j*                  ej                  j*                  ej                  j,                  ej                  j,                  i
ZddZd Zd Zd Zd Zd Zd Zd	 Zd
 Z d Z!d Z"d Z#d Z$d Z%d Z&d Z'd Z(d Z)d Z*d Z+d Z,d Z-d Z.d Z/d Z0d Z1d Z2d Z3d Z4d Z5d  Z6d! Z7d" Z8d# Z9d$ Z:d% Z;d& Z<d' Z=d( Z>d) Z?d* Z@d+ ZAd, ZBd- ZCd. ZDd/ ZEd0 ZFd1 ZGd2 ZHd3 ZId4 ZJd5 ZKd6 ZLd7 ZMd8 ZNd9 ZOd: ZPd; ZQd< ZRd= ZSd> ZTd? ZUd@ ZVdA ZWdB ZXdC ZYdD ZZdE Z[dF Z\dG Z]dH Z^dI Z_dJ Z`dK ZadL ZbdM ZcdN ZddO ZedP ZfdQ ZgdR ZhdS ZidT ZjdU ZkdV ZldW ZmdX ZndY ZodZ Zpd[ Zqd\ Zrd] Zsd^ Ztd_ Zud` Zvda Zwdb Zxdc Zydd Zzde Z{eJZ|eJZ}df Z~dg Zdh Zdi Zdj Zdk Zdl Zdm Zdn Zdo Zdp Zdq Zdr Zds Zdt Zdu Zdv Zdw Zdx Zdy Zdz Zd{ Zd| Zd} Zd~ Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zd Zdedee   dee   dee   fdZdedee   fdZdededee   fdZdededefdZdedededefdZd Zy)InterpreterBuilderNc                     d | _         t               | _        i | _        t        j
                  | j                  d<   d | j                  d<   y )Nconvert_custom_typesc                      y)N)r   r   r   r*   )lhsTyperhsTypes     r   <lambda>z-InterpreterBuilder.__init__.<locals>.<lambda>  s    r   min_dot_size)r_   r[   optionscodegen_fnsr   r   r   s    r   r   zInterpreterBuilder.__init__  sB    	)+3A3W3W/0+M(r   c                     || j                   d   k  st        d      || j                   d   k  st        d      || j                   d   k  st        d      |||f| _        y )Nr   zx >= grid_dim[0]r   zy >= grid_dim[1]r	   zz >= grid_dim[2])grid_dim
ValueErrorgrid_idxr   r   yzs       r   set_grid_idxzInterpreterBuilder.set_grid_idx  sf    4==##/004==##/004==##/00Aq	r   c                     |||f| _         y r   )r   )r   nxnynzs       r   set_grid_dimzInterpreterBuilder.set_grid_dim$  s    Rr   c                 "    t         j                  S r   )r|   r   r   s    r   get_half_tyzInterpreterBuilder.get_half_ty)      zzr   c                 "    t         j                  S r   )r|   r   r   s    r   get_bf16_tyzInterpreterBuilder.get_bf16_ty,      {{r   c                 "    t         j                  S r   )r|   r   r   s    r   get_float_tyzInterpreterBuilder.get_float_ty/  r   r   c                 "    t         j                  S r   )r|   r   r   s    r   get_double_tyz InterpreterBuilder.get_double_ty2  r   r   c                 "    t         j                  S r   )r|   rs   r   s    r   get_int8_tyzInterpreterBuilder.get_int8_ty5  s    wwr   c                 "    t         j                  S r   )r|   rr   r   s    r   get_uint8_tyzInterpreterBuilder.get_uint8_ty8      xxr   c                 "    t         j                  S r   )r|   ru   r   s    r   get_int16_tyzInterpreterBuilder.get_int16_ty;  r   r   c                 "    t         j                  S r   )r|   rt   r   s    r   get_uint16_tyz InterpreterBuilder.get_uint16_ty>      yyr   c                 "    t         j                  S r   )r|   rw   r   s    r   get_int32_tyzInterpreterBuilder.get_int32_tyA  r   r   c                 "    t         j                  S r   )r|   rv   r   s    r   get_uint32_tyz InterpreterBuilder.get_uint32_tyD  r   r   c                 "    t         j                  S r   )r|   rx   r   s    r   get_int64_tyzInterpreterBuilder.get_int64_tyG  r   r   c                 "    t         j                  S r   )r|   r@   r   s    r   get_uint64_tyz InterpreterBuilder.get_uint64_tyJ  r   r   c                 "    t         j                  S r   )r|   r   r   s    r   get_fp8e4nv_tyz!InterpreterBuilder.get_fp8e4nv_tyM      }}r   c                 "    t         j                  S r   )r|   r   r   s    r   get_fp8e4b15_tyz"InterpreterBuilder.get_fp8e4b15_tyP      ~~r   c                 "    t         j                  S r   )r|   r   r   s    r   get_fp8e4b8_tyz!InterpreterBuilder.get_fp8e4b8_tyS  r  r   c                 "    t         j                  S r   )r|   r   r   s    r   get_fp8e5_tyzInterpreterBuilder.get_fp8e5_tyV  r   r   c                 "    t         j                  S r   )r|   r   r   s    r   get_fp8e5b16_tyz"InterpreterBuilder.get_fp8e5b16_tyY  r  r   c                 .    t        j                  ||      S r   )r|   r}   )r   elt_ty
addr_spaces      r   
get_ptr_tyzInterpreterBuilder.get_ptr_ty\  s    vz22r   c                 .    t        j                  ||      S r   )r|   r   )r   r   r/   s      r   get_block_tyzInterpreterBuilder.get_block_ty_  s    }}UE**r   c                 ~    t        t        j                  |gt        j                        t        j
                        S Nr6   )r   r8   arraybool_r|   r~   r   r%   s     r   get_int1zInterpreterBuilder.get_int1b  s$    BHHeWBHH=rwwGGr   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  rr   r|   r  s     r   	get_uint8zInterpreterBuilder.get_uint8e  $    BHHeWBHH=rxxHHr   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  rs   r|   r  s     r   get_int8zInterpreterBuilder.get_int8h  s$    BHHeWBGG<bggFFr   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  rt   r|   r  s     r   
get_uint16zInterpreterBuilder.get_uint16k  $    BHHeWBII>		JJr   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  ru   r|   r  s     r   	get_int16zInterpreterBuilder.get_int16n  r  r   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  rv   r|   r  s     r   
get_uint32zInterpreterBuilder.get_uint32q  r  r   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  rw   r|   r  s     r   	get_int32zInterpreterBuilder.get_int32t  r  r   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  r@   r|   r  s     r   
get_uint64zInterpreterBuilder.get_uint64w  r  r   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  rx   r|   r  s     r   	get_int64zInterpreterBuilder.get_int64z  r  r   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  r   r|   r  s     r   get_fp16zInterpreterBuilder.get_fp16}  $    BHHeWBJJ?LLr   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  r   r|   r  s     r   get_fp32zInterpreterBuilder.get_fp32  r,  r   c                 ~    t        t        j                  |gt        j                        t        j                        S r  )r   r8   r  r   r|   r  s     r   get_fp64zInterpreterBuilder.get_fp64  r,  r   c                 X    t        t        j                  dgt        |            |      S Nr   r6   )r   r8   r  r   )r   types     r   get_null_valuez!InterpreterBuilder.get_null_value  s!    BHHaSd0CDdKKr   c                     | j                   t        d      t        t        j                  | j                   |   gt        j
                        t        j
                        S )Nzgrid_idx is Noner6   )r   r   r   r8   r  rw   r|   r   axiss     r   create_get_program_idz(InterpreterBuilder.create_get_program_id  sD    == /00BHHdmmD&9%:"((KRXXVVr   c                     t        t        j                  | j                  |   gt        j                        t
        j                        S r  )r   r8   r  r   rw   r|   r6  s     r   create_get_num_programsz*InterpreterBuilder.create_get_num_programs  s.    BHHdmmD&9%:"((KRXXVVr   c                     t        t        j                  |j                  t              t
        j                        }d }| j                  ||||||      S r  )r   r8   	ones_liker   r   r|   r~   create_masked_load)r   ptr_0_1is_volatilemaskothers          r   create_loadzInterpreterBuilder.create_load  sA    BLL>H&&sD%RMMr   c                     t        t        j                  |j                  t              t
        j                        }| j                  |||d d       S r  )r   r8   r<  r   r   r|   r~   create_masked_store)r   r>  valr?  r@  rB  s         r   create_storezInterpreterBuilder.create_store  s:    BLL>H''S$dCCr   c                    |j                         }t        |      }|+t        t        j                  |j
                  |      |      }t        j                  |j
                  |j
                  |j
                  |      }	t        |	|      S r  )r   r   r   r8   r   r   _interpreterload)
r   rE   rB  rC  cache_modifiereviction_policyrA  rC   dtype_nprets
             r   r=  z%InterpreterBuilder.create_masked_load  si    &&( *= tyy!I8TE		499ejj(KC**r   c                 l    t        j                  |j                  |j                  |j                        S r   )rJ  storer   )r   rE   r%   rB  rL  rM  s         r   rF  z&InterpreterBuilder.create_masked_store  s#    !!$))UZZCCr   c                    |j                   j                  }|j                  }|t        j                  k(  r|t        j                  k(  s&|t        j                  k(  rY|t        j                  k(  rFt        |j                  ||d       j                  t        |            }t        ||j                        S t        |j                  j                  t        |            |j                        S r   )r   rA   r|   r   r   r   r   viewr   r   r?   )r   srcdst_typesrc_element_typedst_element_typer   s         r   	cast_implzInterpreterBuilder.cast_impl  s    99++#??+0@BJJ0N

*/?2;;/N!#((,<>NPTUZZ[hiq[rsDhoo66h0G H(//ZZr   c                 &    | j                  ||      S r   rX  r   rT  rU  s      r   r   zInterpreterBuilder.<lambda>      $..h2O r   c                 &    | j                  ||      S r   rZ  r[  s      r   r   zInterpreterBuilder.<lambda>  r\  r   c                 &    | j                  ||      S r   rZ  r[  s      r   r   zInterpreterBuilder.<lambda>  r\  r   c                 &    | j                  ||      S r   rZ  r[  s      r   r   zInterpreterBuilder.<lambda>  r\  r   c                 &    | j                  ||      S r   rZ  r[  s      r   r   zInterpreterBuilder.<lambda>  s    sH0M r   c                 &    | j                  ||      S r   rZ  r[  s      r   r   zInterpreterBuilder.<lambda>  r\  r   c                 &    | j                  ||      S r   rZ  )r   rT  rU  	is_signeds       r   r   zInterpreterBuilder.<lambda>  s    T^^CQY=Z r   c                     |j                   j                  }|j                  }t        |j                  |||      j	                  t        |            }t        ||j                        S r   )r   rA   r   r   rS  r   r   )r   rT  rU  r   rV  rW  r   s          r   r   z"InterpreterBuilder.create_fp_to_fp  sU    99++#??chh(8:JMZ__`mnv`wxD(//22r   c                 r    t        |j                  j                  t        |            |j                        S r   )r   r   rS  r   rA   r[  s      r   create_bitcastz!InterpreterBuilder.create_bitcast  s%    CHHMM-*ABHOOTTr   c                 x    t         ||j                  |j                        |j                  j                        S r   r   r   r   rA   )r   lhsrhsops       r   	binary_opzInterpreterBuilder.binary_op  s(    Bsxx2CII4D4DEEr   c                 D    | j                  ||t        j                        S r   rl  r8   addr   ri  rj  s      r   r   zInterpreterBuilder.<lambda>  s    S"&&)I r   c                 D    | j                  ||t        j                        S r   rl  r8   multiplyrp  s      r   r   zInterpreterBuilder.<lambda>      S"++)N r   c                 D    | j                  ||t        j                        S r   rl  r8   dividerp  s      r   r   zInterpreterBuilder.<lambda>  s    S")))L r   c                 D    | j                  ||t        j                        S r   rl  r8   fmodrp  s      r   r   zInterpreterBuilder.<lambda>      S"'')J r   c                 D    | j                  ||t        j                        S r   rl  r8   subtractrp  s      r   r   zInterpreterBuilder.<lambda>  rt  r   c                 D    | j                  ||t        j                        S r   rr  rp  s      r   r   zInterpreterBuilder.<lambda>      sC(M r   c                 D    | j                  ||t        j                        S r   rv  rp  s      r   r   zInterpreterBuilder.<lambda>  s    S"))1T r   c                 &    | j                  ||      S r   create_idivrp  s      r   r   zInterpreterBuilder.<lambda>      )9)9#s)C r   c                 &    | j                  ||      S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   ry  rp  s      r   r   zInterpreterBuilder.<lambda>  r{  r   c                 D    | j                  ||t        j                        S r   ry  rp  s      r   r   zInterpreterBuilder.<lambda>  r{  r   c                 D    | j                  ||t        j                        S r   rn  rp  s      r   r   zInterpreterBuilder.<lambda>  s    sC(H r   c                 D    | j                  ||t        j                        S r   r}  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   )rl  r8   
left_shiftrp  s      r   r   zInterpreterBuilder.<lambda>  s    sC(O r   c                 D    | j                  ||t        j                        S r   )rl  r8   right_shiftrp  s      r   r   zInterpreterBuilder.<lambda>  s    S"..)Q r   c                 D    | j                  ||t        j                        S r   rl  r8   r   rp  s      r   r   zInterpreterBuilder.<lambda>      $..c2::*N r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>      T^^Cbjj-Q r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>      DNN3RZZ,P r   c                 D    | j                  ||t        j                        S r   rl  r8   r   rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   rl  r8   
less_equalrp  s      r   r   zInterpreterBuilder.<lambda>      DNN3R]],S r   c                 D    | j                  ||t        j                        S r   rl  r8   lessrp  s      r   r   zInterpreterBuilder.<lambda>      DNN3RWW,M r   c                 D    | j                  ||t        j                        S r   rl  r8   greater_equalrp  s      r   r   zInterpreterBuilder.<lambda>      DNN3REUEU,V r   c                 D    | j                  ||t        j                        S r   rl  r8   greaterrp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   rl  r8   equalrp  s      r   r   zInterpreterBuilder.<lambda>  s    4>>#sBHH+M r   c                 D    | j                  ||t        j                        S r   rl  r8   	not_equalrp  s      r   r   zInterpreterBuilder.<lambda>  s    4>>#sBLL+Q r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>      DNN3RXX,N r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>      DNN3R\\,R r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   r  rp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   )rl  r8   bitwise_andrp  s      r   r   zInterpreterBuilder.<lambda>      sC(P r   c                 D    | j                  ||t        j                        S r   )rl  r8   bitwise_xorrp  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                 D    | j                  ||t        j                        S r   )rl  r8   
bitwise_orrp  s      r   r   zInterpreterBuilder.<lambda>  s    t~~c3'N r   c                     t        |j                  t        j                  |j                  |j                        z
  |j                  z  |j                  j
                        S r   )r   r   r8   rz  r   rA   rp  s      r   r  zInterpreterBuilder.create_idiv  sC     SXX#(((CCPRUR[R[RbRbccr   c                 @   t        |j                  j                        }t        |j                  j                        }|j                  j                  |      |_        |j                  j                  |      |_        | j	                  ||t
        j                        S r   )ry   r   r   r?   rl  r8   r  )r   ri  rj  	lhs_dtype	rhs_dtypes        r   create_ashrzInterpreterBuilder.create_ashr  se    (8	(8	88??9-88??9-~~c377r   c                 R   |j                   j                  }|t        j                  k(  s|t        j                  k(  r>t        t        |j                   |j                         |j                  j                        S t        t        d|j                  dz  dz         }|j                   j                  |      }|j                   j                  |      }t        j                  ||      |j                  dz  z	  }t        |j                  |      |j                  j                        S )Nr   r5   r	   )r   r   r8   rx   r@   r   np_umulhi_u64rA   r   rW   r?   rs  )r   ri  rj  r   compute_dtypelhs_datarhs_dataret_datas           r   create_umulhiz InterpreterBuilder.create_umulhi	  s    BHH 2chh A399CSCSTT#B$u~~/AA/E.F(GHMxx}5Hxx}5H{{8X65>>A;MNH 6		8H8HIIr   c                     t         ||j                  |j                  |j                        |j                  j                        S r   rh  )r   ri  rj  rC  rk  s        r   
ternary_opzInterpreterBuilder.ternary_op  s.    Bsxx5::>@R@RSSr   c                 F    | j                  |||t        j                        S r   )r  r8   clip)r   arglohipropagate_nanss        r   r   zInterpreterBuilder.<lambda>  s    doocSUWY[][b[b>c r   c                 F    | j                  |||t        j                        S r   )r  r8   where)r   condri  rj  s       r   r   zInterpreterBuilder.<lambda>  s    sCQSQYQY1Z r   c                     t        |j                  |j                  z  |j                  z   |j                  j                        S r   rh  r   s       r   
create_fmazInterpreterBuilder.create_fma  s,    AFFQVVOaff4aggnnEEr   c                 b    t         ||j                        |j                  j                        S r   rh  )r   r  rk  s      r   unary_opzInterpreterBuilder.unary_op  s!    BsxxL#))*:*:;;r   c                 2   |j                   }|j                  dz
  }t        t        d|j                         }|j                  j                  |      }d|z  dz
  }||z  j                  t        |            }t        ||j                   j                        S )Nr   r   )	r   r7   r   r8   r   rS  r   r   rA   )r   r  rC   mask_bitwidthnp_uint_dtyper   rB  rO  s           r   create_fabszInterpreterBuilder.create_fabs"  s    99 33a7d8+F+F*G$HIxx}}]+]"a'd{  x!89C!1!122r   c                 B    | j                  |t        j                        S r   )r  r8   cosr   r  s     r   r   zInterpreterBuilder.<lambda>,      4==bff#= r   c                 B    | j                  |t        j                        S r   )r  r8   expr  s     r   r   zInterpreterBuilder.<lambda>-  r  r   c                 B    | j                  |t        j                        S r   )r  r8   exp2r  s     r   r   zInterpreterBuilder.<lambda>.      DMM#rww$? r   c                 B    | j                  |t        j                        S r   )r  r8   absr  s     r   r   zInterpreterBuilder.<lambda>/  s    DMM#rvv$> r   c                 B    | j                  |t        j                        S r   )r  r8   floorr  s     r   r   zInterpreterBuilder.<lambda>0  s    T]]3%A r   c                 B    | j                  |t        j                        S r   )r  r8   ceilr  s     r   r   zInterpreterBuilder.<lambda>1  r  r   c                 B    | j                  |t        j                        S r   )r  r8   logr  s     r   r   zInterpreterBuilder.<lambda>2  r  r   c                 B    | j                  |t        j                        S r   )r  r8   log2r  s     r   r   zInterpreterBuilder.<lambda>3  r  r   c                 B    | j                  |t        j                        S r   r  r8   sqrtr  s     r   r   zInterpreterBuilder.<lambda>4  s    DMM#rww,G r   c                 B    | j                  |t        j                        S r   r  r  s     r   r   zInterpreterBuilder.<lambda>5  r  r   c                 B    | j                  |t        j                        S r   )r  r8   sinr  s     r   r   zInterpreterBuilder.<lambda>6  r  r   c                     |j                   j                  t        j                  k(  rt	        |j                         nt        |j                         }t        ||j                  j                        S r   )r   r   r8   r   np_erf_fp32np_erf_fp64r   rA   )r   r  rO  s      r   
create_erfzInterpreterBuilder.create_erf8  sH    '*xx~~'Ck#((#UXU]U]I^C!1!122r   c                     t        dt        j                  |j                        z  |j                  j
                        S Nr   )r   r8   r  r   r   rA   r  s     r   create_rsqrtzInterpreterBuilder.create_rsqrt<  s+    A 113993C3CDDr   c                 t    t        |j                  j                  |      |j                  j                        S r   )r   r   r>   r   rA   )r   r  r/   allow_reorders       r   r   zInterpreterBuilder.<lambda>@  s*    \#((JZJZ[`Jacfclclcscs=t r   c                 ~    t        t        j                  |j                  |      |j                  j
                        S r   )r   r8   	transposer   r   rA   )r   r  perms      r   create_transzInterpreterBuilder.create_transB  s(    BLL48#)):J:JKKr   c                    |j                   }|j                   }|j                  j                  dk(  r|j                  j                         s3|j                  j                  dk(  r|j                  j                         rt	        ||j                  t
        j                  d       j                  t        j                        }t	        ||j                  t
        j                  d       j                  t        j                        }t        t        j                  |||j                   j                        |j                   z   |j                  j                        S )Nr5   r6   )r   r   r7   is_floatingr   r|   r   rS  r8   r   matmulrA   )r   r   r   dinput_precisionmax_num_imprecise_acca_datab_datas           r   
create_dotzInterpreterBuilder.create_dotE  s    GG&&!+0C0C0EGG&&!+0C0C0E#FAGGRZZFKKBJJWF#FAGGRZZFKKBJJWFBIIffAFFLLIAFFRTUT[T[TbTbccr   c                 ~    t        t        j                  ||t        j                        t        j                        S r  )r   r8   r=   rw   r|   )r   startstops      r   create_make_rangez$InterpreterBuilder.create_make_rangeN  s$    BIIeTBBHHMMr   c                     t        t        j                  |j                  |d|f      d   t        j
                        S )Nr   )binsr;   )r   r8   	histogramr   r|   rw   )r   r   r  s      r   create_histogramz#InterpreterBuilder.create_histogramQ  s1    BLLaYOPQRTVT\T\]]r   c                     t        t        j                  |j                  |j                  |      |j                  j
                        S )Nr7  )r   r8   take_along_axisr   r   rA   )r   rT  indicesr7  s       r   create_gatherz InterpreterBuilder.create_gatherT  s3    B..sxxDQSVS\S\ScScddr   c                     |j                         }|j                  }t        d|dz        }t        |j                  ||j                  j                  t        j                        z  z   |j                        S )Nr   r5   )	r   r7   maxr   r   r?   r8   r@   r   )r   r>  offsetrC   element_bitwidthelement_bytewidths         r   create_addptrz InterpreterBuilder.create_addptrY  se    %%'#66#3q#89CHH'86;;;M;Mbii;X'XXZ]ZcZcddr   c                    |j                  |      \  }}|j                         }	t        |	      }
|d }n|t        j                  j
                  k(  r,t        t        j                  |j                  |
      |	      }na|t        j                  j                  k(  r6t        t        j                  |j                  t        d      |
      |	      }nt        d|       | j                  ||||||      S )Nr6   nanzunsupported padding option )rJ   r   r   r   PADDING_OPTIONPAD_ZEROr   r8   r   r   PAD_NAN	full_likefloatr   r=  )r   r>  rB   padding_optionrL  rM  rA  rE   rF   rC   rN  rC  s               r   create_tensor_pointer_loadz-InterpreterBuilder.create_tensor_pointer_load`  s    ..~>e&&( *!Es11::: tyy!I8TEs11999 diiuX!VX`aE:>:JKLL&&tUE>?\ghhr   c                 T    |j                  |      \  }}| j                  |||||      S r   rJ   rF  )r   r>  r%   rB   rL  rM  rE   rF   s           r   create_tensor_pointer_storez.InterpreterBuilder.create_tensor_pointer_storeo  s/    ..~>e''eUNO\\r   c                 ~    t        t        j                  |j                  |      |j                  j
                        S r   )r   r8   expand_dimsr   r   rA   )r   r  r7  s      r   create_expand_dimsz%InterpreterBuilder.create_expand_dimss  s(    BNN388T:CII<L<LMMr   c                 ~    t        t        j                  |j                  |      |j                  j
                        S r   )r   r8   r9   r   r   rA   r   r  r/   s      r   create_broadcastz#InterpreterBuilder.create_broadcastv  s(    BOOCHHe<cii>N>NOOr   c                     t        t        j                  |j                  |j                  g      |j                  j
                        S r   )r   r8   concatenater   r   rA   rp  s      r   
create_catzInterpreterBuilder.create_caty  s/    BNNCHHchh+?@#))BRBRSSr   c                     t        t        j                  |j                  |j                  gd      |j                  j
                        S )NrQ   r!  )r   r8   stackr   r   rA   rp  s      r   create_joinzInterpreterBuilder.create_join|  s1    BHHchh%9CSYYEUEUVVr   c                     t        |j                  d   |j                  j                        t        |j                  d   |j                  j                        fS )N).r   ).r   rh  )r   rG  s     r   create_splitzInterpreterBuilder.create_split  sE    SXXf-syy/?/?@,sxxX^O_adajajaqaqBrssr   c           	         t        |j                  t        j                        rVt	        t        j                  ||j                  d   t        |j                              |j                  j                        S t	        t        j                  ||j                  t        |j                              |j                  j                        S r2  )
r{   r   r|   r   r   r8   fullr   r   rA   r;  s      r   create_splatzInterpreterBuilder.create_splat  s    cii/sxx{-PSPYPYBZ []`]f]f]m]mnnsxx}SYY?W XZ]ZcZcZjZjkkr   c                     || j                   vrt        d|       | j                   |   }t        t        j                  |j
                  |j
                  |j
                  |      |j                  j                        S )Nunsupported semantic )ir_sem_to_interpreter_semr   r   rJ  
atomic_casr   r   rA   )r   r>  cmprG  semscopes         r   create_atomic_casz$InterpreterBuilder.create_atomic_cas  sk    d4444SE:;;,,S1L33CHHchhRUVX[XaXaXhXhiir   c           	      X   || j                   vrt        d|       || j                  vrt        d|       | j                   |   }| j                  |   }t        t	        j
                  ||j                  |j                  |j                  |      |j                  j                        S )Nzunsupported rmwOp rI  )	ir_rmw_op_to_interpreter_rmw_opr   rJ  r   rJ  
atomic_rmwr   r   rA   )r   rmwOpr>  rG  rB  rM  rN  s          r   create_atomic_rmwz$InterpreterBuilder.create_atomic_rmw  s    <<<1%9::d4444SE:;;44U;,,S1L33E388SXXtyyZ]^`c`i`i`p`pqqr   c                     t        d      )Nz4extern_elementwise not supported in interpreter modeNotImplementedError)r   libNamelibPathsymbolargListretTypeisPures          r   create_extern_elementwisez,InterpreterBuilder.create_extern_elementwise  s    !"XYYr   c                     t        d      )Nz,inline_asm not supported in interpreter moderV  )r   	inlineAsmconstraintsvaluesr3  r]  packs          r   create_inline_asmz$InterpreterBuilder.create_inline_asm  s    !"PQQr   c                 *   d| j                   d    d| j                   d    d| j                   d    d}|r|d| z  }|rt        j                  dd	 i
       |D ]  }t        |d|j                   z           |rt        j                  d 
       y y )N(r   z, r   r	   ) r   c                     d| dS )N0x02xr*   r   s    r   r   z1InterpreterBuilder.create_print.<locals>.<lambda>  s    b3L r   )	formatter)r   r8   set_printoptionsprintr   )r   prefixhexrb  isSignedmsgr%   s          r   create_printzInterpreterBuilder.create_print  s    
 $--"#2dmmA&6%7r$--:J9K1MQvh<C52H*IJ 	*E#!EJJ<(()	*$/ r   c                     |sJ |        y r   r*   )r   	conditionmessages      r   create_assertz InterpreterBuilder.create_assert  s    &WI&yr   c                     |sJ d       y )NzAssume failedr*   )r   ru  s     r   create_assumez InterpreterBuilder.create_assume  s    )/)yr   c                      y r   r*   r   s    r   create_barrierz!InterpreterBuilder.create_barrier  s    r   c                 f    |D cg c]  }|j                          }}t        ||||||      S c c}w r   )r"   r,   )	r   r.   r/   r0   r1   r2   r3   r'  new_offsetss	            r   create_make_block_ptrz(InterpreterBuilder.create_make_block_ptr  s6    4;<&v||~<<!$w[RWXX =s   .c                    t        |j                        t        |      k7  rt        d      |j                  D cg c]  }|j                          }}t	        |j
                  |j                  |j                  ||j                  |j                        }t        t        |            D ]1  }|j                  |   xj                  ||   j                  z  c_        3 |S c c}w )Nz len(ptr.offsets) != len(offsets))r<   r1   r   r"   r,   r.   r/   r0   r2   r3   r;   r   )r   r>  r1   r'  r}  rO  r   s          r   create_advancez!InterpreterBuilder.create_advance  s    s{{s7|+?@@47KK@&v||~@@ 399ckk;PSP_P_adajajks7|$ 	3AKKN71:??2	3
	 As   C r.   r/   r0   tensor_shapec                 B    t        ||||      }|j                          |S r   )rL   rT   )r   r.   r/   r0   r  descs         r   create_make_tensor_descriptorz0InterpreterBuilder.create_make_tensor_descriptor  s"      eWlCr   r  r#  c                 |    t        |t              sJ |j                  |      \  }}| j                  ||d ||d      S )NF)rC  rL  rM  rA  )r{   rL   rJ   r=  )r   r  r#  rL  rM  rE   rB  s          r   create_descriptor_loadz)InterpreterBuilder.create_descriptor_load  sO    $ 0111..w7
d&&tTn7FTY ' [ 	[r   r%   c                 T    |j                  |      \  }}| j                  |||d d       S r   r5  )r   r  r%   r#  rE   rB  s         r   create_descriptor_storez*InterpreterBuilder.create_descriptor_store  s/    ..w7
d''eT4FFr   	x_offsetsy_offsetc                    |j                   j                  j                  }t        |      }t	        j
                  |j                  j                  d   |j                  d   g|      }d }d }	t        |j                        D ]F  \  }
}t        |t        j                        |g}| j                  ||||	      j                  ||
d d f<   H t        ||      S )Nr   rQ   r6   )r.   r   r   r   r8   zerosr   r/   r2   	enumerater   r|   rw   r  )r   r  r  r  r3  r   np_dtyperesultrL  rM  r   x_offsetr#  s                r   create_descriptor_gatherz+InterpreterBuilder.create_descriptor_gather  s    		** '9>>//2D4D4DR4HIQYZ$Y^^4 	lKAx#Hbhh7BG66tWnVefkkF1a4L	l FE**r   c                     t        |j                        D ]W  \  }}t        |j                  |   |j                        }t        |t        j
                        |g}| j                  |||       Y y r   )r  r   r   r   r|   rw   r  )	r   r  r%   r  r  r   r  slicer#  s	            r   create_descriptor_scatterz,InterpreterBuilder.create_descriptor_scatter  s]    $Y^^4 	?KAx A<E#Hbhh7BG((ug>	?r   c                     t        |      }d|j                  v r,t        t        j                  dd|      |j
                        S t        d|       )NrX   r   rQ   r6   zunsupported type )r   namer   r8   rF  rA   	TypeError)r   r3  np_types      r   get_all_ones_valuez%InterpreterBuilder.get_all_ones_value  sI    %GLL 2W =t{{KK/v677r   returnN)r'   r(   r)   r   MEM_SEMANTICACQUIRErJ  RELEASERELAXEDACQUIRE_RELEASErJ  	ATOMIC_OPADDRMW_OPFADDMINUMINMAXUMAXANDORXORXCHGrQ  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.  r0  r4  r8  r:  rD  rH  r=  rF  rX  create_si_to_fpcreate_ui_to_fpcreate_fp_to_sicreate_fp_to_uicreate_fp_extcreate_fp_trunccreate_int_castr   rf  rl  create_faddcreate_fmulcreate_fdivcreate_fremcreate_fsub
create_mulcreate_precise_divfcreate_sdivcreate_udivcreate_sremcreate_urem
create_add
create_sub
create_shlcreate_lshrcreate_minsicreate_minuicreate_minimumfcreate_minnumfcreate_maxsicreate_maxuicreate_maximumfcreate_maxnumfcreate_icmpSLEcreate_icmpSLTcreate_icmpSGEcreate_icmpSGTcreate_icmpULEcreate_icmpULTcreate_icmpUGEcreate_icmpUGTcreate_icmpEQcreate_icmpNEcreate_fcmpOLTcreate_fcmpOGTcreate_fcmpOLEcreate_fcmpOGEcreate_fcmpOEQcreate_fcmpONEcreate_fcmpULTcreate_fcmpUGTcreate_fcmpULEcreate_fcmpUGEcreate_fcmpUEQcreate_fcmpUNE
create_and
create_xor	create_orcreate_int_to_ptrcreate_ptr_to_intr  r  r  r  create_clampfcreate_selectr  r  r  
create_cos
create_expcreate_exp2create_iabscreate_floorcreate_ceil
create_logcreate_log2create_precise_sqrtcreate_sqrt
create_sinr  r  create_reshaper  r  r  r  r$  r*  r3  r6  r9  r<  r?  rB  rD  rG  rO  rT  r^  rd  rs  rw  ry  r{  r~  r  r   r   rX   r  rL   r  r  r  r  r  r*   r   r   r   r      s     ,";";"C"C  ,";";"C"C  ,";";"C"C((,*C*C*S*S	! 	<..22L//44<..22L//44<..22L//44<..22,--00<..22L//44'#N"%
3+HIGKIKIKIMMMLW
WN
D+D[ POOOOOOOMMOOZO3UF JKNKLKJKNKMJTCKCKJKJKHJMJOJQKNLNLQOPNNLNLQOPNSNMNVNPNSNMNVNPNMMQMMNPNSNVNNNRNMNPNSNVNNNRNPJPJNI&&d8	JT dMZMF<3 >J=J?K>KAL?K=J?KG?K=J3E uNLdN^e
ei]NPTWtljrZR0'*Y
		 L!	 l#		
 3i	[+; [d<FX [G,< G\ G\`am\n G	+-= 	+, 	+bn 	+?.> ?| ?`l ?,8?8r   r   c                 0    |dfd
}t        | ||       y )N)memberc           
      v     | |i |j                         D ci c]  \  }}|dk7  r|| c}}diS c c}}w )Nr   items)r  argskwargskvbuilders        r   r   z_patch_attr.<locals>.<lambda>  sZ     :hMS\\^AUEIQDEO BCA AU:h `g:h AUs   5)setattr)objr  r  r  
new_members      ` r   _patch_attrr    s    &, iJ Cz"r   c                     t        j                  |       D ]3  \  }}t        j                  j	                  |      s&t        | |||       5 y r   )inspect
getmembersr|   core
is_builtinr  )pkgr  r  r  s       r   _patch_builtinr     s@    **3/ 4f77f%T6734r   c                 v    d d }d | _         fd| _        d | _        d | _        t	        |      | _        y )Nc                 f    | j                   j                  }|j                  dk(  rt        |      S dS )Nr   T)r   r   sizer   )r   r   s     r   	_get_boolz%_patch_lang_tensor.<locals>._get_bool  s,    {{ "YY!^tDz55r   c                    t        t        j                  | j                  j                        | j                  j
                        }| j                  j                         sJ t        | j                  j                        }|d   |d   c|d<   |d<   t        j                  j                  | j
                  |      }t        j                  j                  ||      S )NrQ   )r   r8   r  r   r   r   r3  is_blocklistr/   r|   r  r   r   )r   r   r2   res_tys       r   _get_transposez*_patch_lang_tensor.<locals>._get_transpose  s    bll4;;+;+;<dkk>O>OPyy!!###499??++6r?KO(BR##DJJ<ww~~ff--r   c                 @    t        | j                  j                        S r   )rX   r   r   r   s    r   r   z$_patch_lang_tensor.<locals>.<lambda>  s    C(8(8$9 r   c                      |       S r   r*   )r   r  s    r   r   z$_patch_lang_tensor.<locals>.<lambda>  s    9T? r   c                 @    t        | j                  j                        S r   )reprr   r   r   s    r   r   z$_patch_lang_tensor.<locals>.<lambda>  s    4(8(8#9 r   c                 @    t        | j                  j                        S r   )rp   r   r   r   s    r   r   z$_patch_lang_tensor.<locals>.<lambda>  s    #dkk&6&6"7 r   )	__index__r   __repr____str__propertyT)r   r  r  s     @r   _patch_lang_tensorr    s8    6. :F2FO9FO7FN'FHr   c                   0    e Zd Zd Zd Zd Zd Zd Zd Zy)ReduceScanOpInterfacec                      || _         || _        y r   )r7  
combine_fn)r   r7  r!  s      r   r   zReduceScanOpInterface.__init__  s    	$r   c                 H    | |t        |      k\  rt        d| d|       y y )Nzaxis z out of bounds for shape )r<   r   )r   r/   r7  s      r   
check_axisz ReduceScanOpInterface.check_axis#  s4    E
 2uTF*CE7KLL !3r   c                     |D ]c  }t        |t        j                  j                        st	        dt        |             | j                  |j                  | j                         e y )Nzinput must be a tensor, got )	r{   r|   r  r   r   r3  r#  r/   r7  )r   r   r  s      r   check_tensorz"ReduceScanOpInterface.check_tensor'  sP     	2Cc277>>2 #?S	{!KLLOOCIItyy1	2r   c                 Z   t        |      }t        |d      rG|j                  r;|j                  |      }t	        j
                  |t        |j                              }nt        j                  |g|      }|}t        j                  j                  t        ||j                        |      S )Nr/   r6   )r   r   r/   r?   r|   r   r  r8   r  r  r   r   rA   )r   rO  r   r  ret_types        r   	to_tensorzReduceScanOpInterface.to_tensor-  sy     '3 SYY**X&C}}UDO<H((C51CHww~~l3=xHHr   c                 l    t        |t              s|f}| j                  |       | j                  |      S r   )r{   tupler%  
apply_implr   r   s     r   applyzReduceScanOpInterface.apply7  s0    %'IE% u%%r   c                     t        d      )Nzapply_impl not implementedrV  r,  s     r   r+  z ReduceScanOpInterface.apply_impl=  s    !">??r   N)	r'   r(   r)   r   r#  r%  r(  r-  r+  r*   r   r   r  r    s#    %M2I&@r   r  c                   >     e Zd Z fdZd Zd ZddZd Zd Z xZ	S )	ReduceOpsc                 4    t         |   ||       || _        y r   )superr   	keep_dims)r   r7  r!  r3  	__class__s       r   r   zReduceOps.__init__C  s    z*"r   c                     g }|D ]e  }||j                  |       d}|j                  | j                  |j                  j                  j	                         |j
                               g t        |      |fS )Nr   )appendr(  r   r   flattenr   r*  )r   r   r7  rO  r   s        r   unravelzReduceOps.unravelG  sn     	SD

4 

4>>$++*:*:*B*B*DdjjQR	S Sz4r   c                 $     j                   } j                   j                         \  }g }g }d   j                  j                  j                  }|d| ||dz   d  z   }D ]k  }|j                  |j                  j                         |j                  t        j                  ||j                  j                  j                               m t        |d   j                        D ]Y  }	t        j                  |	|      d| |dz   d  z   t         fdt        |      D              }
|   dk(  rGt        t        |            D ]/  }|
|   j                  j                  j                         ||   <   1 t         fdt        |      D              }  j                   j"                  g ||
 }t%        |t              s|fn|}t        t        |            D ][  }t%        ||   t&        j(                  j*                        r'||   j                  j                  j                         n||   ||   <   ] \ g }t        |      D ]  \  }	} j,                  rI|t        j.                  ||      }nBt        t        |            D ]  }t        j.                  |d      } n||j                         }|j                   j1                  ||	   j                                t        |      dk(  r|d   S t        |      S )Nr   r   r6   c              3   h   K   | ])  \  }}j                  |   |   j                         + y wr   r(  r   ).0iir  r   input_indexr   s      r   	<genexpr>z+ReduceOps.generic_reduce.<locals>.<genexpr>`  s/     sTYTVXYq~uRy Os   /2c              3   h   K   | ])  \  }}j                  |   |   j                         + y wr   r;  )r<  oior   output_indexr   s      r   r?  z+ReduceOps.generic_reduce.<locals>.<genexpr>f  s/     !wW\WY[\$..<%)//"R!wr@  )r7  r8  r   r   r/   r6  r8   r  r   r;   r  unravel_indexr*  r  r<   rR   r!  fnr{   r|   r  r   r3  r8  r(  )r   r   original_axisr7  
input_dataoutput_datainput_shapeoutput_shaper  r   input_tuplej	acc_tuplecombine_fn_retrO  r   _r>  rD  s   ``               @@r   generic_reducezReduceOps.generic_reduceQ  s   		ll5$))4t
Ahoo**00"1T*[-CC 	TCcjjoo.rxxCJJOO<Q<QRS	T z!}))* 	HA**1k:K&q.TAXY1GGLs]fgq]rssK4 A%s;/0 UA3>q>3H3H3M3M3R3R3TKN<0U "!w`iju`v!ww	!3!3!3!MY!M!M6@QV6W^.]k	s;/0 HAV`!!bggnnW69Q<3F3F3K3K3P3P3R;DQ<  N<0H	H"  - 	=GAt~~ ,>>$5D"3{#34 7!~~dA67 &yy{JJt~~dE!HNN;<	= SQs1v6E#J6r   c                    t        |t              r|d   n|}d }d }|rM| j                   ||j                  j                  | j
                  | j                        |j                        }|rQ| j                   ||j                  j                  | j
                  | j                        t        j                        }||||fS ||S ||S t        d      )Nr   r7  keepdimsz-val_reduce_op and idx_reduce_op are both None)r{   r*  r(  r   r   r7  r3  r   r|   rw   r   )r   r   val_reduce_opidx_reduce_oprG  idxs         r   min_maxzReduceOps.min_max|  s    &ue4a%..u||/@/@tyy[_[i[i!jlqlwlwxC..u||/@/@tyy[_[i[i!jlnltltuC?s8O_J_JLMMr   c                     | j                  t        j                  |j                  j                  | j
                  | j                        |j                        S )NrS  )r(  r8   sumr   r   r7  r3  r   r,  s     r   rZ  zReduceOps.sum  s<    ~~bffU\\%6%6TYYQUQ_Q_`bgbmbmnnr   c                 2   | j                   t        j                  j                  k(  r3| j	                  |d   t
        j                  t
        j                        S | j                   t        j                  j                  k(  r3| j	                  |d   t
        j                  t
        j                        S | j                   t        j                  j                  k(  r%| j	                  |d   t
        j                  d       S | j                   t        j                  j                  k(  r%| j	                  |d   t
        j                  d       S | j                   t        j                  j                  k(  r| j                  |d         S | j                  |      S )Nr   )rU  rV  )r!  r|   standard_argmin_combine_tie_break_leftrX  r8   minargmin_argmax_combine_tie_break_leftr&  argmax_elementwise_max_elementwise_min_sum_combinerZ  rQ  r,  s     r   r+  zReduceOps.apply_impl  s   ??bkkHHH<<abii<XX__ J JJ<<abii<XX__ < <<<<ad<SS__ < <<<<ad<SS__ 8 8888E!H%% &&u--r   r   )
r'   r(   r)   r   r8  rQ  rX  rZ  r+  __classcell__r4  s   @r   r0  r0  A  s$    # )7VN$o.r   r0  c                   6     e Zd Z fdZd Zd Zd Zd Z xZS )ScanOpsc                 4    t         |   ||       || _        y r   )r2  r   reverse)r   r7  r!  rj  r4  s       r   r   zScanOps.__init__  s    z*r   c                     | j                  t        j                  |j                  j                  | j
                        |j                        gS Nr!  r6   )r(  r8   cumsumr   r   r7  r   r,  s     r   rm  zScanOps.cumsum  s8    ryy):):KSXS^S^_``r   c                     | j                  t        j                  |j                  j                  | j
                        |j                        gS rl  )r(  r8   cumprodr   r   r7  r   r,  s     r   ro  zScanOps.cumprod  s8    rzz%,,*;*;$))LTYT_T_`aar   c           	          g }g }d   j                   j                  j                  }D ]k  }|j                  |j                   j                         |j                  t	        j
                  ||j                   j                  j                               m t        |d   j                        D ]|  }t	        j                  ||      t         fdt        |      D              } j                     dk(  rGt        t        |            D ]/  }||   j                   j                  j                         ||   <   1 t         fdt        t                    D              t         fdt        |      D              }	  j                  j                   g |	| }
t#        |
t              s|
fn|
}	t        t        |            D ][  }t#        |	|   t$        j&                  j(                        r'|	|   j                   j                  j                         n|	|   ||   <   ]  g }t        |      D ]3  \  }}|j                   j+                  ||   j                               5 |S )Nr   r6   c              3   h   K   | ])  \  }}j                  |   |   j                         + y wr   r;  )r<  r=  r  indexr   r   s      r   r?  z'ScanOps.generic_scan.<locals>.<genexpr>  s,     fur1%%)//Bfr@  c              3   V   K   | ]   }|j                   k(  r|   d z
  n|    " yw)r   Nr!  )r<  r   rr  r   s     r   r?  z'ScanOps.generic_scan.<locals>.<genexpr>  s-     "kTU1		>58a<uQx#O"ks   &)c              3   h   K   | ])  \  }}j                  |   |   j                         + y wr   r;  )r<  rB  rC  r   
prev_indexr   s      r   r?  z'ScanOps.generic_scan.<locals>.<genexpr>  s/     !uUZUWYZ$..:b	"P!ur@  )r   r   r/   r6  r8   r  r   r;   r  rE  r*  r  r7  r<   rR   r!  rF  r{   r|   r  r   r(  )r   r   rH  rI  r/   r  r   r   rM  rN  rO  rO  rr  ru  s   ``          @@r   generic_scanzScanOps.generic_scan  s!   
a$$** 	MCcjjoo.rxxSZZ__5J5JKL	M z!}))* 	HA$$Q.EfPYZdPeffDTYY1$s;/0 GA,0GNN,?,?,D,D,FKN5)G #"kY^_bch_iYj"kk
!!u^ghs^t!uu	!3!3!3!FY!F!F6@QV6W^.]k	s;/0 HAOY!!bggnnP6IaL,?,?,D,D,I,I,K;DQ<  N5)H	H"  - 	=GAtJJt~~dE!HNN;<	=
r   c           	         g }| j                   rf|D ]`  }|j                  | j                  t        j                  |j
                  j                  | j                        |j                               b n|}| j                  t        j                  j                  k(  r| j                  |d         }nM| j                  t        j                  j                  k(  r| j                  |d         }n| j!                  |      }| j                   rK|D ]F  }t        j                  |j
                  j                  | j                        |j
                  _        H t#        |      dk(  xr |d   xs t%        |      S )Nr!  r   r   )rj  r6  r(  r8   flipr   r   r7  r   r!  r|   r\  rd  rm  _prod_combinero  rv  r<   r*  )r   r   	new_inputr  rO  s        r   r+  zScanOps.apply_impl  s   	<< f  

dii0XZ]ZcZc!def I??bkk666++il+C__ 9 99,,y|,C ##I.C<< K"$''#**//		"J

K3x1}'Q55:5r   )	r'   r(   r)   r   rm  ro  rv  r+  re  rf  s   @r   rh  rh    s    ab<6r   rh  c                      dd} dd}| t         _        |t         _        | t         j                  _        |t         j                  _        y )Nc                 :    t        |||      j                  |       S r   )r0  r-  )r   r7  r!  r3  r  s        r   _new_reducez'_patch_reduce_scan.<locals>._new_reduce  s    z95;;EBBr   c                 :    t        |||      j                  |       S r   )rh  r-  )r   r7  r!  rj  r  s        r   	_new_scanz%_patch_reduce_scan.<locals>._new_scan  s    tZ177>>r   )F)r|   reduceassociative_scanr  )r}  r  s     r   _patch_reduce_scanr    s5    C? BI#B BGGN(BGGr   c                    d }d	d}d
d}d }|| _         || _        || _        t        | _        || j
                  _        t        |d      | _        t        |d      | _	        t        |d      | _
        t                y )Nc                 >   | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j	                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   d	k(  r|j                         S | j                   d
k(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j                         S | j                   dk(  r|j!                         S | j                   dk(  r|j#                         S t%        d|  d      )Nvoidr~   rs   rr   ru   rt   rw   rv   rx   r@   r`   rb   rd   fp16bf16fp32fp64zfail to convert z to ir type)r  get_void_tyget_int1_tyr   r   r   r   r   r   r   r   r	  r  r  r   r   r   r   r   )r   r  s     r   
_new_to_irz$_patch_lang_core.<locals>._new_to_ir  s   99&&((YY& &&((YY& &&((YY'!''))YY'!''))YY("((**YY'!''))YY("((**YY'!''))YY("((**YY'!''))YY)#))++YY*$**,,YY& &&((YY& &&((YY& ''))YY& ((**+D6=>>r   c                 :    |d}|d| }}n| |}}t        |||      S )Nr   r   )r;   )arg1arg2stepr  r  ends         r   
_new_rangez$_patch_lang_core.<locals>._new_range  s2    <D<D3Et3EUC&&r   c                     | sJ |       y r   r*   )r  rr  s     r   _new_static_assertz,_patch_lang_core.<locals>._new_static_assert"  s    Str   c                    t        | t        j                        s| S t        |t        t        f      s|gn|}|D cg c]*  }t        |t        j
                        r|j                  n|, }}t        |      t        dt        | j                              k7  rt        d|       | j                  j                  ||       | S c c}w )Nr   z$len(values) != len(input.shape) for )r{   r|   r   r  r*  	constexprr%   r<   r&  r/   r   r   r&   )r   rb  r  r  s       r   	_set_attrz#_patch_lang_core.<locals>._set_attr%  s    %+L!+FT5M!B&IOPAZ2<<8!''a?PPv;#aU[[!122CD6JKKdF+	 Qs   /Cztt.divisibility)r  ztt.contiguityztt.constancy)NN) )r;   static_rangestatic_assertrn  static_printr   to_irr   multiple_ofmax_contiguousmax_constancyr  )langr  r  r  r  s        r   _patch_lang_corer    sx    $?P'
 DJ"D+DD!DJJy/@AD!)/BD @Dr   c                 (   | j                   j                         D cg c]6  \  }}t        j                  |      s|t        t        j
                  fv s5|8 }}}t        |      dk\  sJ d       |D ]o  }t        |t               t        |j                  t               |t        k(  rt        |j                  t               t        |j                         t        |       q t        t        j
                  j                  t               y c c}}w )Nr   z:triton.language must be visible from within jit'd function)__globals__r  r  ismoduler|   r  r<   r  interpreter_builderr   r   r  r  $_experimental_tensor_descriptor_base)rF  rP  r%   langsr  s        r   _patch_langr  =  s    #%>>#7#7#9pxq%W=M=Me=TY^cegigngnboYoUpEpu:?XXX? t01t{{$782:499&9:4;;' 277??ATU qs   DDDc                 X    t        | d      r t        |       | S  t        |       |      S )N_fields)r   r3  )r  contentss     r   _tuple_creater  J  s.     $+3	#:949hS	S	(@SSr   c                    t        | t              rt        j                  t        j
                  j                  j                  |             }t        j                  }d| cxk  rdk  rn nt        j                  }nkd| cxk  rdk  rn nt        j                  }nLd| cxk  rdk  rn nt        j                  }n-d| cxk  rdk  rn nt        j                  }nt        d|        t        t        j                  | g|      |      }t        j                   ||      S t#        | d	      rt        j                  t        j
                  j                  j                  |             }t        t        j                  | j%                         gt        j                        |      }t        j                   ||      S t        | t&              rt)        | t+        t,        |             S | S )
Ni   l        l        l         l            l            zUnsupported integer value r6   data_ptr)r{   rX   r|   	str_to_tytritonruntimejitmangle_typer8   rw   rv   rx   r@   r   r   r  r   r   r  r*  r  map_implicit_cvt)r  tyr   r   s       r   r  r  T  sR   #s\\&..,,88=>S 5 HHEc!E!IIEs"U"HHEc!E!IIE9#?@@bhhuE:B?yy$$sJ\\&..,,88=>bhh'7ryyI2Nyy$$	C	S#mS"9::Jr   c                 z    t        | t        j                  j                  j                        r| j
                  S | S r   )r{   r  r  r  TensorWrapperr.   )ts    r   _unwrap_tensorr  p  s*    !V^^''556vvHr   c                     t        |t        j                  j                  j                        r4t        j                  j                  j	                  | |j
                        S | S r   )r{   r  r  r  r  r   )r  original_tensors     r   _rewrap_tensorr  v  sD    /6>>#5#5#C#CD~~!!//?3H3HIIHr   c                   $    e Zd Zd Zd Zd Zd Zy)GridExecutorc                 
   ddl m} || _        || _        || _        |j
                  j                         D ci c]  \  }}| ||       }}}|D cg c]  }|j                  |      dk(  s| c}| _        y c c}}w c c}w )Nr   )_normalize_tyr  )	r  r  rF  	arg_namesgridro   r  get
constexprs)r   rF  r  r  r  r  r  ro   s           r   r   zGridExecutor.__init__~  sx    &"	CECUCUC[C[C]^xtR4r!22^^,5bD9L9LT9RVa9a4b _bs   A:B -B c                    	 i 		fd|D cg c]
  } |       }}i }|j                         D ]  \  }} |      ||<    ||fS c c}w )Nc                 &   t        | t              rt        | t        |             S t	        | d      s| S t        |       }|j                         j                         vr1|j                         }|j                         |j                         <   |j                         j                            }|j                  dd      }|j                  ||j                         |j                         |j                                t        ||       }|S )Nr  r   cpu)device)r  )r{   r*  r  r  r   r  untyped_storager  r  	new_emptyset_storage_offsetr  rS   r  )r  unwrapped_argstoragecpu_arg_to_cpustoragess       r   r  z,GridExecutor._init_args_hst.<locals>._to_cpu  s    #u%$S#gs*;<<S*-
*3/M,,.779I'779/6{{}))+,}<<>GGIJG#--a->GLL-">">"@-BTBTBVXeXlXlXno$WcBGNr   r  )
r   args_devr  r  args_hst
kwargs_hstr$   r%   r  r  s
           @@r   _init_args_hstzGridExecutor._init_args_hst  sc    	" -55SGCL55 
 ,,. 	-JC%enJsO	-## 6s   A	c                    
 i 
fd
t        ||      D ]  \  }} 
||        |j                         D ]  \  }}||   }	 
||	        j                         D ]  \  }}|j                  |        y )Nc                 &   t        | d      rVt        |       t        |      }} | j                         |j                         f| j                         j                         <   y t	        | t
              rt        | |      D ]  \  } } | |        y y )Nr  )r   r  r  r  r{   r*  zip)arg_devarg_hst	_from_cpur  s     r   r  z1GridExecutor._restore_args_dev.<locals>._from_cpu  s    w
+#1'#:N7<SBIBYBYB[]d]t]t]vAw002;;=>GU+*-gw*? 0&Wggw/0 ,r   )r  r  rb  copy_)r   r  r  r  r  r  r  r$   	kwarg_dev	kwarg_hstr  r  s             @@r   _restore_args_devzGridExecutor._restore_args_dev  s    	0 !$Hh 7 	(GWgw'	( %lln 	,NC"3Ii+	, #+//"3 	#WgMM'"	#r   c                    |j                  dd      ry t        j                  | j                        }|j	                         D ci c]  \  }}||j
                  v s|| }}}| j                  ||      \  }}t        | j                         t        j                  | j                  g|i |}|j	                         D 	
ci c]!  \  }	}
|	|	| j                  v r|
n
t        |
      # }}	}
t        | j                        r| j                  |      n| j                  }t        |      dk  sJ d       |ddt        |      z
  z  z   }t        j                  |  	 t!        |d         D ]Q  }t!        |d         D ]>  }t!        |d         D ]+  }t        j#                  |||        | j                  d	i | - @ S 	 | j+                  ||||       y c c}}w c c}
}	w # t$        $ r}t'        t)        |            |d }~ww xY w)
NwarmupF   z#grid must have at most 3 dimensions)r   r   r   r	   r*   )popr  getfullargspecrF  r  r  r  r  getcallargsr  r  callabler  r<   r  r   r;   r   	Exceptionr   r  r  )r   r  r  argspecr  r  r  r  r  r  r  r  r   r   r   es                   r   __call__zGridExecutor.__call__  s   ::h& ((1#)<<>G41aQ',,5F!Q$GG#228VD*DGG ""477DXDD^b^h^h^jkQZQUWZTT__4c-:LLkk"*499"5tyy4994yA~DDD~eq3t9}--(($/	347^ (tAw (A"47^ (+88AqA$((( 	x6:F/ H l  	3"47+2	3s+   GG&GA"G   	H)G>>HN)r'   r(   r)   r   r  r  r  r*   r   r   r  r  |  s    c$8#.Gr   r  c                       e Zd Zd Zy)ASTTransformerc                    g }|j                   D ]  }|| j                  |      gz  } t        |      dkD  rt        d      t	        j
                  t	        j                  t	        j                  t	        j                  t	        j                  dt	        j                               dt	        j                               dt	        j                               dt	        j                               |j                  t	        j                  d	t	        j                               t	        j                  d
      gg       |_	        |S )Nr   z&Multiple assignments are not supportedr  )idctxlanguage)r%   r   r  semanticr(  r  F)r%   )funcr  keywords)targetsvisitr<   r   astCall	AttributeNameLoadr%   Constant)r   nodenamestargets       r   visit_AssignzASTTransformer.visit_Assign  s    ll 	*Fdjj())E	*u:>EFF XXmm--chh(
.SZdjmjrjrjtu#5;FCHHJX **chh*?SXXZP,,U+-79;
 r   N)r'   r(   r)   r   r*   r   r   r  r    s    r   r  c                   D    e Zd Z e       Zd Zd Zd Zd Zd Z	d Z
d Zy)	FunctionRewriterc                 <    || _         || _        d| _        d| _        y )Nr  r   )rF  r  filenamedef_file_lineno)r   rF  r  s      r   r   zFunctionRewriter.__init__  s    $%r   c                 J   	 t        j                  | j                        \  }}| j	                         \  | _        | _        | j                  |      | _        | j                  |      }| j                  |      }| j                  |      S # t        $ r | j                  cY S w xY wr   )r  getsourcelinesrF  r  _get_jit_fn_file_liner  r  	_find_def
def_lineno_prepare_source_transform_ast_compile_and_exec)r   linesrP  rT  transformed_asts        r   rewrite_astzFunctionRewriter.rewrite_ast  s    	--dgg6HE1 /3.H.H.J+t+../""5)--c2%%o66  	77N	s   "B
 
B"!B"c                 B    ddl m}m}  | || j                              S )Nr   )get_jit_fn_file_lineJITFunction)r  r  r  rF  )r   r  r  s      r   r  z&FunctionRewriter._get_jit_fn_file_line  s    :#K$899r   c                 z    d}t        |      D ]*  \  }}|j                         j                  d      s&|dz   }, |S )Nr   zdef r   )r  strip
startswith)r   r  r
  r   lines        r   r	  zFunctionRewriter._find_def  sD    
 ' 	#GAtzz|&&v.U
	# r   c                 r    || j                   dz
  d  }dj                  |      }t        j                  |      S )Nr   r  )r
  jointextwrapdedent)r   r  rT  s      r   r  z FunctionRewriter._prepare_source  s4    doo)*+ggens##r   c                     t        j                  |      }| j                  j                  |      }t        j                  |       | j
                  dz
  }t        j                  ||       |S r  )r  parseast_transformerr  fix_missing_locationsr  increment_lineno)r   rT  
parsed_astr  
inc_linenos        r   r  zFunctionRewriter._transform_ast  s[     YYs^
..44Z@!!/2))A-
_j9r   c                    t        || j                  d      }i | j                  }| j                  j                  }t               j                         D ]  \  }}||vs|||<    t        |||       || j                  j                     S )Nexec)r  mode)	compiler  r  rF  r  globalsr  r$  r'   )r   r  compiled_codelocal_namespace
fn_globalsr$   r%   s          r   r  z"FunctionRewriter._compile_and_exec)  s    $--fU)T[[/WW((
!)//+ 	(JC*$"'
3	( 	]J8tww//00r   N)r'   r(   r)   r  r  r   r  r  r	  r  r  r  r*   r   r   r  r    s-    $&O&7(:$
	1r   r  c                   :    e Zd Zi ZddZd Zed        Z d Zd Zy)InterpretedFunctionNc                      | _         t        |fi | _         fd}| _        t	        j
                  |      }|j                  j                         D cg c]  }|j                   c} _	        y c c}w )Nc                  h    |d   }j                         } t        |j                  |      | i |S )Nr  rewriter  r  )r  r  r  rF  r   s       r   runz)InterpretedFunction.__init__.<locals>.run<  s6    &>DB9<DNND94J6JJr   )
rF  r  rewriterr1  r  	signature
parametersrb  r  r  )r   rF  r  r1  r3  r  s   `     r   r   zInterpretedFunction.__init__8  sb    (6v6	K
 %%b)	*3*>*>*E*E*GHQ!&&HHs   A3c                     | j                   | j                  vr1| j                  j                         | j                  | j                   <   | j                  | j                      S r   )rF  rewritten_fnr2  r  r   s    r   r0  zInterpretedFunction.rewriteE  sJ    77$+++)-)B)B)DDdgg&  ))r   c                 .    | j                   j                  S r   )rF  r'   r   s    r   r'   zInterpretedFunction.__name__J  s    wwr   c                 P    | j                         }t        || j                  |      S r   r/  )r   r  rF  s      r   __getitem__zInterpretedFunction.__getitem__N  s     \\^B55r   c                     t        | j                         | j                         }	  ||i |S # t        $ r}t	        t        |            |d }~ww xY wr   )r  rF  r0  r  r   r  )r   r  r  rF  r  s        r   r  zInterpretedFunction.__call__R  sO    DGG\\^	3t&v&& 	3"47+2	3s   / 	AAAr  )	r'   r(   r)   r6  r   r0  r  r9  r  r*   r   r   r,  r,  4  s0    LI*
    63r   r,  )=r  r  r  typingr   r   r   numpyr8   r  triton.languager  r|   dataclassesr   errorsr   	functoolsr   _C.libtritonr
   rJ  r   r   r   r,   rL   r[   ry   r   r   r   r   	vectorizer   r  r   r  r@   r  r   r   r  r  r  r  r0  rh  r  r  r  r  r  r  r  r  r  NodeTransformerr  r  r,  r*   r   r   <module>rD     s   
        ! $  6 $ : 4" "J $
& 
& 
&	@='@
# bll45bll45Z<g gv8 v8r#4(.!@ !@H].% ].@;6# ;6|) K\
VT2 )* \G \G~S(( (B1 B1J%3 %3r   