
    Xh                    h   d dl mZ d dlZd dlZd dlZd dlmZmZmZ d dl	Z	d dl
Zd dlZd dlmZ d dlZd dlmZ d dlmZ d dlmZ ddlmZ d d	lmZ d
dlmZ d
dlmZ e G d d                      Z G d d          Z  G d d          Z! ed           G d d                      Z"d Z#d Z$d Z%d Z&d Z' ej(        e&ej)        g          Z* ej(        e&ej+        g          Z, ej(        e'ej-        g          Z. G d d          Z/ G d d           Z0d! Z1d" Z2d# Z3 G d$ d%          Z4 G d& d'e4          Z5 G d( d)e4          Z6d* Z7d+ Z8d, Z9d- Z:d. Z; e0            Z< ee<          Z=d/ Z>d0 Z? G d1 d2          Z@ G d3 d4ejA                  ZB G d5 d6          ZC G d7 d8          ZDdS )9    )annotationsN)TupleListDict)	dataclass)TritonSemantic)TensorDescriptor   )InterpreterError)partial   )interpreter)irc                  l    e Zd ZU dZded<   ded<    ej        e          Zded<   d	 Z	d
 Z
d Zd ZdS )TensorHandlez
        data: numpy array
        dtype: triton type, either pointer_type or scalar_type.
        we don't store block_type here because the shape information is already available in the data field
        attr: a dictionary of attributes
    znp.arraydataztl.dtypedtype)default_factoryr   attrc                N    t          | j                                                  S N)boolr   allselfs    l/var/www/tools.fuzzalab.pt/emblema-extractor/venv/lib/python3.11/site-packages/triton/runtime/interpreter.py__bool__zTensorHandle.__bool__#   s    DIMMOO$$$    c                b    | j         }t          |d          r|j        }t          |d          |S )N
element_ty)r   hasattrr    )r   r   s     r   get_element_tyzTensorHandle.get_element_ty&   s<    
e\** 	%$E e\** 	%r   c                Z    t          | j                                        | j                  S r   )r   r   copyr   r   s    r   clonezTensorHandle.clone,   s     DINN,,dj999r   c                    || j         |<   d S r   )r   )r   keyvalues      r   set_attrzTensorHandle.set_attr/   s    	#r   N)__name__
__module____qualname____doc____annotations__dataclassesfielddictr   r   r"   r%   r)    r   r   r   r      s           NNNOOO""4888D8888% % %  : : :    r   r   c                      e Zd Zd Zd ZdS )BlockPointerHandlec                Z    || _         || _        || _        || _        || _        || _        d S r   )baseshapestridesoffsetsblock_shapeorder)r   r6   r7   r8   r9   r:   r;   s          r   __init__zBlockPointerHandle.__init__5   s1    	
&


r   c                   | 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                       |          }|||z  | j        |         j        z                      t          j                  z   }||v r ||| j        |         j        k     z  |dk    z  }t%          || j         j        j                  }||fS )N   r   r
   r   )r6   r"   primitive_bitwidthnpbroadcast_tor   r:   onesr   rangelenr9   arangereshaper8   astypeuint64r7   r   r   scalar)	r   boundary_checkdtype_ttn_bytesptrsmasksdim
bcast_dimsoffs	            r   materialize_pointersz'BlockPointerHandle.materialize_pointers=   sL   9++---2ty~t/?@@(555T-..// 	J 	JCs4#3444J".s3JsO<$)BId6Fs6K,L,LLUUV`aaC7S=4<+<+AAII")TTTDn$$tz#';!;<qID$)/"899U{r   N)r*   r+   r,   r<   rS   r2   r   r   r4   r4   3   s2              r   r4   c                  $    e Zd ZddZd	 ZddZdS )TensorDescHandler6   r   r7   List[TensorHandle]r8   r:   	List[int]c                f    || _         t          |          | _        || _        || _        || _        d S r   )r6   rE   ndimr7   r8   r:   )r   r6   r7   r8   r:   s        r   r<   zTensorDescHandle.__init__O   s2    	JJ	
&r   c                   | j         j                                        dz  dk    s
J d            t          | j                  | j        k    sJ t          | j                  | j        k    sJ | j        d d         D ],}|j                                        dz  dk    s
J d            -| j        d         j                                        dk    s
J d            d S )N   r   zbase must be 16-byte alignedzstride must be 16-byte alignedr
   zlast dim must be contiguous)r6   r   itemrE   r8   rY   r:   )r   strides     r   validatezTensorDescHandle.validateW   s    y~""$$r)Q...0N...4<  DI----4#$$	1111l3B3' 	R 	RF;##%%*a///1Q////|B$))++q0002O00000r   r9   c                r   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                       |          }|||z  | j        |         j        z                      t          j                  z   }|d|k    z  || j        |         j        k     z  }|j        t          j        k    sJ t'          || j        j        j                  }||fS )Nr>   r\   r[   r   z*block offset start must be 16-byte alignedr?   r
   )rE   rY   r6   r   r    r@   r   rA   rB   r:   rC   r   rD   rF   rG   r8   rH   rI   r7   r   rJ   )	r   r9   	scalar_tyitemsizerN   rO   rP   rQ   rR   s	            r   rS   z%TensorDescHandle.materialize_pointers`   s   7||ty((((IO.	/14 8+r1Q6668d666ty~t/?@@(555T-..// 	F 	FCs4#3444J".s3JsO3<$ry1A#1F'G'GGPPQ[\\C8c>DL,=,BBJJ29UUUDQ#X&#
30D*DEEEzRY&&&&D$)/"899U{r   N)r6   r   r7   rV   r8   rV   r:   rW   )r9   rV   )r*   r+   r,   r<   r_   rS   r2   r   r   rU   rU   M   sN        ' ' ' 'P P P     r   rU   T)frozenc                      e Zd ZU dZded<   dZded<   dZded<   dZd	ed
<   dZded<   dZ	ded<   dZ
d	ed<   dZded<   dZded<   dZd	ed<   dS )InterpreterOptionsNr1   extern_libsFr   debugTsanitize_overflowstrarch)fp8e5fp8e5b16fp8e4nvfp8e4b8fp8e4b15z
Tuple[str]supported_fp8_dtypesr2   !deprecated_fp8_dot_operand_dtypestf32default_dot_input_precision)rr   tf32x3ieeeallowed_dot_input_precisionsr   intmax_num_imprecise_acc_defaultr   backend_name)r*   r+   r,   rf   r.   rg   rh   rj   rp   rq   rs   rv   rx   ry   r2   r   r   re   re   s   s         KE"""""D'^^^^^46%6666'-----/I IIII)*!****%L%%%%%%r   re   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   )	rA   uint8int8uint16int16uint32int32rI   int64r?   s    r   _get_signed_np_dtyper      sW    w	x	x	x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          j                  rJt          | j        t          j                  rt          j        t          j                  S || j                 S ||          S r   )
isinstancetlpointer_typerA   r   rI   int1r   float16float32float64r|   r{   r~   r}   r   r   r   bfloat16float8e5float8e5b16
float8e4nv
float8e4b8float8e4b15
block_typer    )tt_dtypenp_typess     r   _get_np_dtyper      sX   (BO,, #x	"""
$

BHRZ(( 	
BHRZ(( 	
BHRZ((	
 	"'"" 	"(28$$ 	"(28$$ 		28BI&& 	"(28$$ 		28BI&& 	"(28$$ 		28BI&& 	RXbi((  	RXbh''!" 	**#$ 	rx))%& 	rx))'( 	**) H, (BM** -h)2?;; 	'8BI&&&+,,Hr   c                   t          t          d|j                   }t          t          d|j                   }t          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                      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        k    rl|
|j        |j        z
  z	  d|j        z  dz
  z  }|t          j        j        k    r!|
d|j        |j        z
  dz
  z  z  }||dk    z   }|                    |          }n3|
                    |          |j        |j        z
  z  d|j        z  dz
  z  }|dk    }t          j	        |          r||j        z	  d|z  dz
  z                      t          j                  }|dk    }||z  }t          j
        |t          j                  }d|z
  ||         |z
  z
  ||<   ||         ||         z	  d|j        ||         z
  z  z  ||<   ||j        dz
  z  ||j        z  z  |z  }|                    | j                  S )Nuintr?   r
   r   )getattrrA   r@   
frombuffertobytesfp_mantissa_widthexponent_biasrH   r   any
zeros_likerD   maximumminimum_irROUNDING_MODERTNErG   r7   )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   r#J+*H#J#JKK %ML,K%M%MNNemmoo5EFFFI+81<=ED&9K<YY\]](;l>\\_``[%B BaGHK*J,Kk;;FZAZ^_@_`hhikiqrrH!mO	vo 6
 -	:::{455 	H 	HA%*d2I&1&Ca&GGIN##!,!1$%(@$@!=G+=U'/9:(3O(DP_H`(`+//14(6O$ jBJ:0E0SWX\qWquvVv$w$wxxO%,,-?@@O++011K%(GGG)k.KlNl.lm,00A57C-222!Q;+H<Ki+ilm+m%noG!3w{!C/667IJJ)001CDD+=@]]_#$(F#F!"KM &*O	vo L
 +"??QJ^E^bcDcdllmomuvv"*a-),CCirx888"#k/h6OR\6\!]o/A//RV[\kVl/l,053IIJ/L?+l=AB<99;=OPF>>%+&&&r   c                *    t          j        |           S r   )matherfxs    r   _erfr      s    8A;;r   c                F    t          |           t          |          z  dz	  S )N@   )rw   )abs     r   
_umulhi_64r      s     FFSVVO""r   )otypesc                  $    e Zd Zed             ZdS )ExtraFunctionsc                j    t          j        |j                            | j        ||          |          S r   )r   tensorbuildercreate_fp_to_fphandle)r   dst_tyfp_downcast_rounding	_semantics       r   _convert_custom_typesz$ExtraFunctions._convert_custom_types   s-    y*::5<Qeffhnooor   N)r*   r+   r,   staticmethodr   r2   r   r   r   r      s2        p p \p p p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/ ZDd0 ZEd1 ZFd2 ZGd3 ZHd4 ZId5 ZJd6 ZKd7 ZLd8 ZMd9 ZNd: ZOd; ZPd< ZQd= ZRd> ZSd? ZTd@ ZUdA ZVdB ZWdC ZXdD ZYdE ZZdF Z[dG Z\dH Z]dI Z^dJ Z_dK Z`dL ZadM ZbdN ZcdO ZddP ZedQ ZfdR ZgdS ZhdT ZidU ZjdV ZkdW ZldX ZmdY ZndZ Zod[ Zpd\ Zqd] Zrd^ Zsd_ Ztd` Zuda Zvdb Zwdc Zxdd Zyde Zzdf Z{dg Z|eKZ}eKZ~dh Zdi 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 Zd Zd ZddZddZddZddZddZd ZdS )InterpreterBuilderreturnNonec                    d | _         t                      | _        i | _        t          j        | j        d<   d | j        d<   d S )Nconvert_custom_typesc                    dS )N)r
   r
   r
   r2   )lhsTyperhsTypes     r   <lambda>z-InterpreterBuilder.__init__.<locals>.<lambda>  s    I r   min_dot_size)rj   re   optionscodegen_fnsr   r   r   s    r   r<   zInterpreterBuilder.__init__  sE    	)++3A3W/0+M+M(((r   c                    || j         d         k     st          d          || j         d         k     st          d          || j         d         k     st          d          |||f| _        d S )Nr   zx >= grid_dim[0]r
   zy >= grid_dim[1]r   zz >= grid_dim[2])grid_dim
ValueErrorgrid_idxr   r   yzs       r   set_grid_idxzInterpreterBuilder.set_grid_idx   sv    4=###/0004=###/0004=###/000Aq	r   c                    |||f| _         d S 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.  
    zr   c                    t           j        S r   )r   r   r   s    r   get_bf16_tyzInterpreterBuilder.get_bf16_ty1  
    {r   c                    t           j        S r   )r   r   r   s    r   get_float_tyzInterpreterBuilder.get_float_ty4  r   r   c                    t           j        S r   )r   r   r   s    r   get_double_tyz InterpreterBuilder.get_double_ty7  r   r   c                    t           j        S r   )r   r   r   s    r   get_int1_tyzInterpreterBuilder.get_int1_ty:  	    wr   c                    t           j        S r   )r   r|   r   s    r   get_int8_tyzInterpreterBuilder.get_int8_ty=  r   r   c                    t           j        S r   )r   r{   r   s    r   get_uint8_tyzInterpreterBuilder.get_uint8_ty@  	    xr   c                    t           j        S r   )r   r~   r   s    r   get_int16_tyzInterpreterBuilder.get_int16_tyC  r  r   c                    t           j        S r   )r   r}   r   s    r   get_uint16_tyz InterpreterBuilder.get_uint16_tyF  
    yr   c                    t           j        S r   )r   r   r   s    r   get_int32_tyzInterpreterBuilder.get_int32_tyI  r  r   c                    t           j        S r   )r   r   r   s    r   get_uint32_tyz InterpreterBuilder.get_uint32_tyL  r  r   c                    t           j        S r   )r   r   r   s    r   get_int64_tyzInterpreterBuilder.get_int64_tyO  r  r   c                    t           j        S r   )r   rI   r   s    r   get_uint64_tyz InterpreterBuilder.get_uint64_tyR  r  r   c                    t           j        S r   )r   r   r   s    r   get_fp8e4nv_tyz!InterpreterBuilder.get_fp8e4nv_tyU  
    }r   c                    t           j        S r   )r   r   r   s    r   get_fp8e4b15_tyz"InterpreterBuilder.get_fp8e4b15_tyX  
    ~r   c                    t           j        S r   )r   r   r   s    r   get_fp8e4b8_tyz!InterpreterBuilder.get_fp8e4b8_ty[  r  r   c                    t           j        S r   )r   r   r   s    r   get_fp8e5_tyzInterpreterBuilder.get_fp8e5_ty^  r   r   c                    t           j        S r   )r   r   r   s    r   get_fp8e5b16_tyz"InterpreterBuilder.get_fp8e5b16_tya  r  r   c                ,    t          j        ||          S r   )r   r   )r   elt_ty
addr_spaces      r   
get_ptr_tyzInterpreterBuilder.get_ptr_tyd  s    vz222r   c                ,    t          j        ||          S r   )r   r   )r   r   r7   s      r   get_block_tyzInterpreterBuilder.get_block_tyg  s    }UE***r   c                t    t          t          j        |gt          j                  t          j                  S Nr?   )r   rA   arraybool_r   r   r   r(   s     r   get_int1zInterpreterBuilder.get_int1j  s'    BHeWBH===rwGGGr   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r{   r   r%  s     r   	get_uint8zInterpreterBuilder.get_uint8m  '    BHeWBH===rxHHHr   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r|   r   r%  s     r   get_int8zInterpreterBuilder.get_int8p  s'    BHeWBG<<<bgFFFr   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r}   r   r%  s     r   
get_uint16zInterpreterBuilder.get_uint16s  '    BHeWBI>>>	JJJr   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r~   r   r%  s     r   	get_int16zInterpreterBuilder.get_int16v  r)  r   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r   r   r%  s     r   
get_uint32zInterpreterBuilder.get_uint32y  r.  r   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r   r   r%  s     r   	get_int32zInterpreterBuilder.get_int32|  r)  r   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  rI   r   r%  s     r   
get_uint64zInterpreterBuilder.get_uint64  r.  r   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r   r   r%  s     r   	get_int64zInterpreterBuilder.get_int64  r)  r   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r   r   r%  s     r   get_fp16zInterpreterBuilder.get_fp16  '    BHeWBJ???LLLr   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r   r   r%  s     r   get_fp32zInterpreterBuilder.get_fp32  r;  r   c                t    t          t          j        |gt          j                  t          j                  S r"  )r   rA   r#  r   r   r%  s     r   get_fp64zInterpreterBuilder.get_fp64  r;  r   c                f    t          t          j        dgt          |                    |          S Nr   r?   )r   rA   r#  r   )r   types     r   get_null_valuez!InterpreterBuilder.get_null_value  s+    BHaSd0C0CDDDdKKKr   c                    | j         t          d          t          t          j        | j         |         gt          j                  t          j                  S )Nzgrid_idx is Noner?   )r   r   r   rA   r#  r   r   r   axiss     r   create_get_program_idz(InterpreterBuilder.create_get_program_id  sF    = /000BHdmD&9%:"(KKKRXVVVr   c                    t          t          j        | j        |         gt          j                  t
          j                  S r"  )r   rA   r#  r   r   r   rE  s     r   create_get_num_programsz*InterpreterBuilder.create_get_num_programs  s/    BHdmD&9%:"(KKKRXVVVr   c                    t          t          j        |j        t                    t
          j                  }d }|                     ||||||          S r"  )r   rA   	ones_liker   r   r   r   create_masked_load)r   ptr_0_1is_volatilemaskothers          r   create_loadzInterpreterBuilder.create_load  sG    BL>>>HH&&sD%RMMMr   c                    t          t          j        |j        t                    t
          j                  }|                     |||d d           S r"  )r   rA   rK  r   r   r   r   create_masked_store)r   rM  valrN  rO  rQ  s         r   create_storezInterpreterBuilder.create_store  s@    BL>>>HH''S$dCCCr   c                
   |                                 }t          |          }|)t          t          j        |j        |          |          }t          j        |j        |j        |j        |          }	t          |	|          S r"  )r"   r   r   rA   r   r   _interpreterload)
r   rN   rQ  rR  cache_modifiereviction_policyrP  rL   dtype_nprets
             r   rL  z%InterpreterBuilder.create_masked_load  sq    &&(( **= ty!I!I!I8TTE	49ej(KKC***r   c                L    t          j        |j        |j        |j                  S r   )rY  storer   )r   rN   r(   rQ  r[  r\  s         r   rU  z&InterpreterBuilder.create_masked_store  s    !$)UZCCCr   c                   |j         j        }|j        }|t          j        k    r|t          j        k    s |t          j        k    r\|t          j        k    rLt          |j        ||d                               t          |                    }t          ||j                  S t          |j        
                    t          |                    |j                  S r   )r   rJ   r   r   r   r   r   viewr   r   rH   )r   srcdst_typesrc_element_typedst_element_typer   s         r   	cast_implzInterpreterBuilder.cast_impl  s    9+#?++0@BJ0N0N
**/?2;/N/N!#(,<>NPTUUZZ[hiq[r[rssDho666h0G0G H H(/ZZZr   c                .    |                      ||          S r   rg  r   rc  rd  s      r   r   zInterpreterBuilder.<lambda>      $..h2O2O r   c                .    |                      ||          S r   ri  rj  s      r   r   zInterpreterBuilder.<lambda>  rk  r   c                .    |                      ||          S r   ri  rj  s      r   r   zInterpreterBuilder.<lambda>  rk  r   c                .    |                      ||          S r   ri  rj  s      r   r   zInterpreterBuilder.<lambda>  rk  r   c                .    |                      ||          S r   ri  rj  s      r   r   zInterpreterBuilder.<lambda>  s    sH0M0M r   c                .    |                      ||          S r   ri  rj  s      r   r   zInterpreterBuilder.<lambda>  rk  r   c                .    |                      ||          S r   ri  )r   rc  rd  	is_signeds       r   r   zInterpreterBuilder.<lambda>  s    T^^CQY=Z=Z r   c                    |j         j        }|j        }t          |j        |||                              t          |                    }t          ||j                  S r   )r   rJ   r   r   rb  r   r   )r   rc  rd  r   re  rf  r   s          r   r   z"InterpreterBuilder.create_fp_to_fp  sV    9+#?ch(8:JMZZ__`mnv`w`wxxD(/222r   c                v    t          |j                            t          |                    |j                  S r   )r   r   rb  r   rJ   rj  s      r   create_bitcastz!InterpreterBuilder.create_bitcast  s*    CHMM-*A*ABBHOTTTr   c                ^    t           ||j        |j                  |j        j                  S r   r   r   r   rJ   )r   lhsrhsops       r   	binary_opzInterpreterBuilder.binary_op  s'    BBsx22CI4DEEEr   c                D    |                      ||t          j                  S r   r{  rA   addr   rx  ry  s      r   r   zInterpreterBuilder.<lambda>  s    S"&)I)I r   c                D    |                      ||t          j                  S r   r{  rA   multiplyr  s      r   r   zInterpreterBuilder.<lambda>      S"+)N)N r   c                D    |                      ||t          j                  S r   r{  rA   divider  s      r   r   zInterpreterBuilder.<lambda>  s    S"))L)L r   c                D    |                      ||t          j                  S r   r{  rA   fmodr  s      r   r   zInterpreterBuilder.<lambda>      S"')J)J r   c                D    |                      ||t          j                  S r   r{  rA   subtractr  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>      sC(M(M r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  s    S")1T1T r   c                .    |                      ||          S r   create_idivr  s      r   r   zInterpreterBuilder.<lambda>      )9)9#s)C)C r   c                .    |                      ||          S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r}  r  s      r   r   zInterpreterBuilder.<lambda>  s    sC(H(H r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   )r{  rA   
left_shiftr  s      r   r   zInterpreterBuilder.<lambda>  s    sC(O(O r   c                D    |                      ||t          j                  S r   )r{  rA   right_shiftr  s      r   r   zInterpreterBuilder.<lambda>  s    S".)Q)Q r   c                D    |                      ||t          j                  S r   r{  rA   r   r  s      r   r   zInterpreterBuilder.<lambda>      $..c2:*N*N r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>      T^^Cbj-Q-Q r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>      DNN3RZ,P,P r   c                D    |                      ||t          j                  S r   r{  rA   r   r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r{  rA   
less_equalr  s      r   r   zInterpreterBuilder.<lambda>      DNN3R],S,S r   c                D    |                      ||t          j                  S r   r{  rA   lessr  s      r   r   zInterpreterBuilder.<lambda>      DNN3RW,M,M r   c                D    |                      ||t          j                  S r   r{  rA   greater_equalr  s      r   r   zInterpreterBuilder.<lambda>      DNN3REU,V,V r   c                D    |                      ||t          j                  S r   r{  rA   greaterr  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r{  rA   equalr  s      r   r   zInterpreterBuilder.<lambda>  s    4>>#sBH+M+M r   c                D    |                      ||t          j                  S r   r{  rA   	not_equalr  s      r   r   zInterpreterBuilder.<lambda>  s    4>>#sBL+Q+Q r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>      DNN3RX,N,N r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>      DNN3R\,R,R r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   r  r  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   )r{  rA   bitwise_andr  s      r   r   zInterpreterBuilder.<lambda>      sC(P(P r   c                D    |                      ||t          j                  S r   )r{  rA   bitwise_xorr  s      r   r   zInterpreterBuilder.<lambda>  r  r   c                D    |                      ||t          j                  S r   )r{  rA   
bitwise_orr  s      r   r   zInterpreterBuilder.<lambda>  s    t~~c3'N'N r   c                    t          |j        t          j        |j        |j                  z
  |j        z  |j        j                  S r   )r   r   rA   r  r   rJ   r  s      r   r  zInterpreterBuilder.create_idiv  s9     SX#((C(CCPRUR[Rbcccr   c                $   t          |j        j                  }t          |j        j                  }|j                            |          |_        |j                            |          |_        |                     ||t
          j                  S r   )r   r   r   rH   r{  rA   r  )r   rx  ry  	lhs_dtype	rhs_dtypes        r   create_ashrzInterpreterBuilder.create_ashr	  sf    (88	(88	8??9--8??9--~~c3777r   c                   |j         j        }|t          j        k    s|t          j        k    r2t          t          |j         |j                   |j        j                  S t          t          d|j	        dz  dz             }|j         
                    |          }|j         
                    |          }t          j        ||          |j	        dz  z	  }t          |
                    |          |j        j                  S )Nr   r>   r   )r   r   rA   r   rI   r   np_umulhi_u64rJ   r   rb   rH   r  )r   rx  ry  r   compute_dtypelhs_datarhs_dataret_datas           r   create_umulhiz InterpreterBuilder.create_umulhi  s    BH 2 2ch A A39CSTTT#B(Gu~/AA/E(G(GHHMx}55Hx}55H{8X665>A;MNH 6 6	8HIIIr   c                j    t           ||j        |j        |j                  |j        j                  S r   rw  )r   rx  ry  rR  rz  s        r   
ternary_opzInterpreterBuilder.ternary_op  s+    BBsx5:>>@RSSSr   c                F    |                      |||t          j                  S r   )r  rA   clip)r   arglohipropagate_nanss        r   r   zInterpreterBuilder.<lambda>   s    doocSUWY[][b>c>c r   c                F    |                      |||t          j                  S r   )r  rA   where)r   condrx  ry  s       r   r   zInterpreterBuilder.<lambda>!  s    sCQSQY1Z1Z r   c                `    t          |j        |j        z  |j        z   |j        j                  S r   rw  r   s       r   
create_fmazInterpreterBuilder.create_fma#  s%    AFQVOaf4agnEEEr   c                R    t           ||j                  |j        j                  S r   rw  )r   r  rz  s      r   unary_opzInterpreterBuilder.unary_op'  s!    BBsxLL#)*:;;;r   c                    |j         }|j        dz
  }t          t          d|j                   }|j                            |          }d|z  dz
  }||z                      t          |                    }t          ||j         j                  S )Nr
   r   )	r   r@   r   rA   r   rb  r   r   rJ   )r   r  rL   mask_bitwidthnp_uint_dtyper   rQ  r^  s           r   create_fabszInterpreterBuilder.create_fabs*  s    9 3a7$H8+F$H$HIIx}}]++]"a'd{  x!8!899C!1222r   c                B    |                      |t          j                  S r   )r  rA   cosr   r  s     r   r   zInterpreterBuilder.<lambda>4      4==bf#=#= r   c                B    |                      |t          j                  S r   )r  rA   expr  s     r   r   zInterpreterBuilder.<lambda>5  r  r   c                B    |                      |t          j                  S r   )r  rA   exp2r  s     r   r   zInterpreterBuilder.<lambda>6      DMM#rw$?$? r   c                B    |                      |t          j                  S r   )r  rA   absr  s     r   r   zInterpreterBuilder.<lambda>7  s    DMM#rv$>$> r   c                B    |                      |t          j                  S r   )r  rA   floorr  s     r   r   zInterpreterBuilder.<lambda>8  s    T]]3%A%A r   c                B    |                      |t          j                  S r   )r  rA   ceilr  s     r   r   zInterpreterBuilder.<lambda>9  r   r   c                B    |                      |t          j                  S r   )r  rA   logr  s     r   r   zInterpreterBuilder.<lambda>:  r  r   c                B    |                      |t          j                  S r   )r  rA   log2r  s     r   r   zInterpreterBuilder.<lambda>;  r   r   c                B    |                      |t          j                  S r   r  rA   sqrtr  s     r   r   zInterpreterBuilder.<lambda><  s    DMM#rw,G,G r   c                B    |                      |t          j                  S r   r  r  s     r   r   zInterpreterBuilder.<lambda>=  r   r   c                B    |                      |t          j                  S r   )r  rA   sinr  s     r   r   zInterpreterBuilder.<lambda>>  r  r   c                    |j         j        t          j        k    rt	          |j                   nt          |j                   }t          ||j        j                  S r   )r   r   rA   r   np_erf_fp32np_erf_fp64r   rJ   )r   r  r^  s      r   
create_erfzInterpreterBuilder.create_erf@  sG    '*x~'C'Ck#(###UXU]I^I^C!1222r   c                j    t          dt          j        |j                  z  |j        j                  S Nr
   )r   rA   r  r   r   rJ   r  s     r   create_rsqrtzInterpreterBuilder.create_rsqrtD  s(    A 1 11393CDDDr   c                f    t          |j                            |          |j        j                  S r   )r   r   rG   r   rJ   )r   r  r7   allow_reorders       r   r   zInterpreterBuilder.<lambda>H  s(    \#(JZJZ[`JaJacfclcs=t=t r   c                f    t          t          j        |j        |          |j        j                  S r   )r   rA   	transposer   r   rJ   )r   r  perms      r   create_transzInterpreterBuilder.create_transJ  s%    BL488#):JKKKr   c                <   |j         }|j         }|j        j        dk    r|j                                        s)|j        j        dk    r|j                                        r|t	          ||j        t
          j        d                               t          j                  }t	          ||j        t
          j        d                               t          j                  }t          t          j
        |||j         j                  |j         z   |j        j                  S )Nr>   r?   )r   r   r@   is_floatingr   r   r   rb  rA   r   matmulrJ   )r   r   r   dinput_precisionmax_num_imprecise_acca_datab_datas           r   
create_dotzInterpreterBuilder.create_dotM  s    G&!++0C0C0E0E+G&!++0C0C0E0E+#FAGRZFFKKBJWWF#FAGRZFFKKBJWWFBIffAFLIIIAFRTUT[Tbcccr   c                t    t          t          j        ||t          j                  t          j                  S r"  )r   rA   rF   r   r   )r   ret_tystartstops       r   create_make_rangez$InterpreterBuilder.create_make_rangeV  s'    BIeTBBBBHMMMr   c                   |8t          t          j        |j        t                    t
          j                  }t          j        |j        |j        t          j        |j                            }t          j	        ||d|f          d         }|dxx         t          j
        |j                                                  z  cc<   t          |t
          j                  S )Nr?   r   )binsrD   )r   rA   rK  r   r   r   r   r  r   	histogramlogical_notsumr   )r   r   r-  rQ  r.  s        r   create_histogramz#InterpreterBuilder.create_histogramY  s    <TYd C C CRWMMDx	49bmDI.F.FGGLDD	BBB1E	!ty1155777Irx000r   c                t    t          t          j        |j        |j        |          |j        j                  S )NrF  )r   rA   take_along_axisr   r   rJ   )r   rc  indicesrF  s       r   create_gatherz InterpreterBuilder.create_gatherc  s/    B.sxDQQQSVS\Scdddr   c                    |                                 }|j        }t          d|dz            }t          |j        ||j                            t          j                  z  z   |j                  S )Nr
   r>   )	r"   r@   maxr   r   rH   rA   rI   r   )r   rM  offsetrL   element_bitwidthelement_bytewidths         r   create_addptrz InterpreterBuilder.create_addptrh  se    %%''#6#3q#899CH'86;;M;Mbi;X;X'XXZ]Zcdddr   c                   |                     |          \  }}|                                }	t          |	          }
|d }n|t          j        j        k    r*t          t          j        |j	        |
          |	          }n_|t          j        j
        k    r8t          t          j        |j	        t          d          |
          |	          }nt          d|           |                     ||||||          S )Nr?   nanzunsupported padding option )rS   r"   r   r   PADDING_OPTIONPAD_ZEROr   rA   r   r   PAD_NAN	full_likefloatr   rL  )r   rM  rK   padding_optionr[  r\  rP  rN   rO   rL   r]  rR  s               r   create_tensor_pointer_loadz-InterpreterBuilder.create_tensor_pointer_loado  s    ..~>>e&&(( **!EEs1::: ty!I!I!I8TTEEs1999 diuX!V!V!VX`aaEEK>KKLLL&&tUE>?\ghhhr   c                d    |                     |          \  }}|                     |||||          S r   rS   rU  )r   rM  r(   rK   r[  r\  rN   rO   s           r   create_tensor_pointer_storez.InterpreterBuilder.create_tensor_pointer_store~  s5    ..~>>e''eUNO\\\r   c                f    t          t          j        |j        |          |j        j                  S r   )r   rA   expand_dimsr   r   rJ   )r   r  rF  s      r   create_expand_dimsz%InterpreterBuilder.create_expand_dims  s%    BN38T::CI<LMMMr   c                f    t          t          j        |j        |          |j        j                  S r   )r   rA   rB   r   r   rJ   )r   r  r7   s      r   create_broadcastz#InterpreterBuilder.create_broadcast  s%    BOCHe<<ci>NOOOr   c                r    t          t          j        |j        |j        g          |j        j                  S r   )r   rA   concatenater   r   rJ   r  s      r   
create_catzInterpreterBuilder.create_cat  s*    BNCHch+?@@#)BRSSSr   c                v    t          t          j        |j        |j        gd          |j        j                  S )Nr\   r3  )r   rA   stackr   r   rJ   r  s      r   create_joinzInterpreterBuilder.create_join  s/    BHch%9CCCSYEUVVVr   c                    t          |j        d         |j        j                  t          |j        d         |j        j                  fS )N).r   ).r
   rw  )r   rV  s     r   create_splitzInterpreterBuilder.create_split  s<    SXf-sy/?@@,sxX^O_adajaqBrBrssr   c           	     r   |j         }t          |j        t          j                  rLt          t          j        ||j        d         t          |j                            |j        j
                  S t          t          j        ||j        t          |j                            |j        j
                  S rA  )r7   r   r   r   r   r   rA   fullr   r   rJ   )r   r(  r  r7   s       r   create_splatzInterpreterBuilder.create_splat  s    ci// 	lsx{-PSPYBZBZ [ [ []`]f]mnnnsx}SY?W?W X X XZ]ZcZjk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   rY  
atomic_casr   r   rJ   )r   rM  cmprV  semscopes         r   create_atomic_casz$InterpreterBuilder.create_atomic_cas  sc    d444:S::;;;,S1L3CHchRUVVX[XaXhiiir   c           	         || j         vrt          d|           || j        vrt          d|           | j         |         }| j        |         }t          t	          j        ||j        |j        |j        |          |j        j                  S )Nzunsupported rmwOp rZ  )	ir_rmw_op_to_interpreter_rmw_opr   r[  r   rY  
atomic_rmwr   r   rJ   )r   rmwOprM  rV  rQ  r^  r_  s          r   create_atomic_rmwz$InterpreterBuilder.create_atomic_rmw  s    <<<9%99:::d444:S::;;;4U;,S1L3E38SXtyZ]^^`c`i`pqqqr   c                     t          d          )Nz4extern_elementwise not supported in interpreter modeNotImplementedError)r   libNamelibPathsymbolargListretTypeisPures          r   create_extern_elementwisez,InterpreterBuilder.create_extern_elementwise  s    !"XYYYr   c                     t          d          )Nz,inline_asm not supported in interpreter moderg  )r   	inlineAsmconstraintsvaluesrB  rn  packs          r   create_inline_asmz$InterpreterBuilder.create_inline_asm  s    !"PQ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 
           d S d S )N(r   z, r
   r   ) r   c                    d| dS )N0x02xr2   r   s    r   r   z1InterpreterBuilder.create_print.<locals>.<lambda>  s    LLLL r   )	formatter)r   rA   set_printoptionsprintr   )r   prefixhexrs  isSignedmsgr(   s          r   create_printzInterpreterBuilder.create_print  s    
 N$-"MMdmA&6MM$-:JMMM 	 <v<<C 	K52H2H*IJJJJ 	* 	*E#(EJ((()))) 	0$//////	0 	0r   c                     |sJ |             d S r   r2   )r   	conditionmessages      r   create_assertz InterpreterBuilder.create_assert  s    &&W,&&&&&r   c                    |s
J d            d S )NzAssume failedr2   )r   r  s     r   create_assumez InterpreterBuilder.create_assume  s    ))/)))))r   c                    d S r   r2   r   s    r   create_barrierz!InterpreterBuilder.create_barrier  s    r   c                B    d |D             }t          ||||||          S )Nc                6    g | ]}|                                 S r2   r%   .0r9  s     r   
<listcomp>z<InterpreterBuilder.create_make_block_ptr.<locals>.<listcomp>  s     <<<&v||~~<<<r   )r4   )r   r6   r7   r8   r9   r:   r;   new_offsetss           r   create_make_block_ptrz(InterpreterBuilder.create_make_block_ptr  s.    <<G<<<!$w[RWXXXr   c                t   t          |j                  t          |          k    rt          d          d |j        D             }t          |j        |j        |j        ||j        |j                  }t          t          |                    D ](}|j        |         xj
        ||         j
        z  c_
        )|S )Nz len(ptr.offsets) != len(offsets)c                6    g | ]}|                                 S r2   r  r  s     r   r  z5InterpreterBuilder.create_advance.<locals>.<listcomp>  s     @@@&v||~~@@@r   )rE   r9   r   r4   r6   r7   r8   r:   r;   rD   r   )r   rM  r9   r  r^  r   s         r   create_advancez!InterpreterBuilder.create_advance  s    s{s7||++?@@@@@CK@@@ 39ck;PSP_adajkks7||$$ 	3 	3AKN71:?2
r   r6   r   r7   rV   r8   tensor_shaperW   rr  r   c                R    t          ||||          }|                                 |S r   )rU   r_   )r   r6   r7   r8   r  rr  descs          r   create_make_tensor_descriptorz0InterpreterBuilder.create_make_tensor_descriptor  s)      eWlCCr   r  rU   r5  c                    t          |t                    sJ |                    |          \  }}|                     ||d ||d          S )NF)rR  r[  r\  rP  )r   rU   rS   rL  )r   r  r5  r[  r\  rN   rQ  s          r   create_descriptor_loadz)InterpreterBuilder.create_descriptor_load  s^    $ 011111..w77
d&&tTn7FTY ' [ [ 	[r   r(   c                d    |                     |          \  }}|                     |||d d           S r   rG  )r   r  r(   r5  rN   rQ  s         r   create_descriptor_storez*InterpreterBuilder.create_descriptor_store  s5    ..w77
d''eT4FFFr   	x_offsetsy_offsetc                   |j         j        j        }t          |          }t	          j        |j        j        d         |j        d         g|          }d }d }	t          |j                  D ]E\  }
}t          |t          j                  |g}|                     ||||	          j        ||
d d f<   Ft          ||          S )Nr   r\   r?   )r6   r   r    r   rA   zerosr   r7   r:   	enumerater   r   r   r  )r   r  r  r  rB  r   np_dtyperesultr[  r\  r   x_offsetr5  s                r   create_descriptor_gatherz+InterpreterBuilder.create_descriptor_gather  s    	* ''9>/2D4DR4HIQYZZZ$Y^44 	l 	lKAx#Hbh77BG66tWnVeffkF1aaa4LLFE***r   c                    t          |j                  D ]X\  }}t          |j        |         |j                  }t          |t          j                  |g}|                     |||           Yd S r   )r  r   r   r   r   r   r  )	r   r  r(   r  r  r   r  slicer5  s	            r   create_descriptor_scatterz,InterpreterBuilder.create_descriptor_scatter  sq    $Y^44 	? 	?KAx A<<E#Hbh77BG((ug>>>>	? 	?r   c                   t          |          }d|j        v r*t          t          j        dd|          |j                  S |t          j        k    r*t          t          j        dd|          |j                  S t          d|           )Nrw   r
   r\   r?   Tzunsupported type )r   namer   rA   rW  rJ   r$  	TypeError)r   rB  np_types      r   get_all_ones_valuez%InterpreterBuilder.get_all_ones_value  s    %%GL  2W = = =t{KKK  4w ? ? ?MMM666777r   Nr   r   )
r6   r   r7   rV   r8   rV   r  rW   rr  r   )r  rU   r5  rV   )r  rU   r(   r   r5  rV   )r  rU   r  r   r  r   )r  rU   r(   r   r  r   r  r   )r*   r+   r,   r   MEM_SEMANTICACQUIRErY  RELEASERELAXEDACQUIRE_RELEASEr[  	ATOMIC_OPADDRMW_OPFADDMINUMINMAXUMAXANDORXORXCHGrb  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  r2  r4  r6  r8  r:  r=  r?  rC  rG  rI  rS  rW  rL  rU  rg  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   ru  r{  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+  r1  r6  r<  rE  rH  rK  rM  rP  rS  rU  rX  r`  re  ro  ru  r  r  r  r  r  r  r  r  r  r  r  r  r2   r   r   r   r     s        ,";"C ,";"C ,";"C(,*C*S	! 	<.2L/4<.2L/4<.2L/4<.2,-0<.2L/4'#N N N N" " "% % %
                                    3 3 3+ + +H H HI I IG G GK K KI I IK K KI I IK K KI I IM M MM M MM M ML L LW W W
W W WN N N
D D D+ + +D D D[ [ [ POOOOOOOOOOOMMMOOOZZO3 3 3U U UF F F JIKNNKLLKJJKNNKMMJTTCCKCCKJJKJJKHHJMMJOOJQQKNNLNNLQQOPPNNNLNNLQQOPPNSSNMMNVVNPPNSSNMMNVVNPPNMMMQQMMMNPPNSSNVVNNNNRRNMMNPPNSSNVVNNNNRRNPPJPPJNNI&&d d d8 8 8	J 	J 	JT T T dcMZZMF F F< < <3 3 3 >=J==J??K>>KAAL??K==J??KGG??K==J3 3 3E E E utNL L Ld d dN N N1 1 1e e e
e e ei i i] ] ]N N NP P PT T TW W Wt t tl l lj j jr r rZ Z ZR R R0 0 0' ' '* * *  Y Y Y
  
 
 
 
[ [ [ [G G G G	+ 	+ 	+ 	+? ? ? ?8 8 8 8 8r   r   c                X    t          |          |dfd
}t          | ||           d S )N)memberc                T     | |i d |                                 D             diS )Nc                &    i | ]\  }}|d k    ||S )r   r2   )r  kvs      r   
<dictcomp>z1_patch_attr.<locals>.<lambda>.<locals>.<dictcomp>  s;     AV AV AVEIQDEDTDT BCADTDTDTr   r   items)r  argskwargssemantics      r   r   z_patch_attr.<locals>.<lambda>  sd     :kAV AVMS\\^^AV AV AV:k :k bj:k :k :k r   )r   setattr)objr  r  r   
new_memberr  s        @r   _patch_attrr    sS    g&&H&, l l l l l lJ Cz"""""r   c                    t          j        |           D ]6\  }}t          j                            |          rt          | |||           7d S r   )inspect
getmembersr   core
is_builtinr  )pkgr   r  r  s       r   _patch_builtinr     sW    *3// 4 4f7f%% 	4T673334 4r   c                    d d }d | _         fd| _        d | _        d | _        t	          |          | _        d S )Nc                R    | j         j        }|j        dk    rt          |          ndS )Nr
   T)r   r   sizer   )r   r   s     r   	_get_boolz%_patch_lang_tensor.<locals>._get_bool  s(    { "Y!^^tDzzz5r   c                   t          t          j        | j        j                  | j        j                  }| j                                        sJ t          | j        j	                  }|d         |d         c|d<   |d<   t          j                            | j        |          }t          j                            ||          S )Nr\   )r   rA   r  r   r   r   rB  is_blocklistr7   r   r  r   r   )r   r   r:   res_tys       r   _get_transposez*_patch_lang_tensor.<locals>._get_transpose"  s    bl4;+;<<dk>OPPy!!#####49?+++6r?KO(BR##DJ<<w~~ff---r   c                4    t          | j        j                  S r   )rw   r   r   r   s    r   r   z$_patch_lang_tensor.<locals>.<lambda>*  s    C(8$9$9 r   c                     |           S r   r2   )r   r$  s    r   r   z$_patch_lang_tensor.<locals>.<lambda>+  s    99T?? r   c                4    t          | j        j                  S r   )reprr   r   r   s    r   r   z$_patch_lang_tensor.<locals>.<lambda>,  s    4(8#9#9 r   c                4    t          | j        j                  S r   )ri   r   r   r   s    r   r   z$_patch_lang_tensor.<locals>.<lambda>-  s    #dk&6"7"7 r   )	__index__r   __repr____str__propertyT)r   r*  r$  s     @r   _patch_lang_tensorr5    se    6 6 6. . . :9F2222FO99FO77FN''FHHHr   c                  ,    e Zd Zd Zd Zd Zd Zd ZdS )ReduceScanOpInterfacec                "    || _         || _        d S r   )rF  
combine_fn)r   rF  r9  s      r   r<   zReduceScanOpInterface.__init__3  s    	$r   c                ^    |(|t          |          k    rt          d| d|           d S d S )Nzaxis z out of bounds for shape )rE   r   )r   r7   rF  s      r   
check_axisz ReduceScanOpInterface.check_axis7  sE    E

 2 2KTKKEKKLLL  2 2r   c                    |D ]`}t          |t          j        j                  st	          dt          |                     |                     |j        | j                   ad S )Nzinput must be a tensor, got )	r   r   r  r   r   rB  r;  r7   rF  )r   r   r  s      r   check_tensorz"ReduceScanOpInterface.check_tensor;  sj     	2 	2Cc27>22 M !KS		!K!KLLLOOCIty1111	2 	2r   c                `   t          |          }t          |d          rD|j        r=|                    |          }t	          j        |t          |j                            }nt          j        |g|          }|}t          j	        
                    t          ||j                  |          S )Nr7   r?   )r   r!   r7   rH   r   r   r(  rA   r#  r  r   r   rJ   )r   r^  r   r  ret_types        r   	to_tensorzReduceScanOpInterface.to_tensorA  s     ''3   	SY 	**X&&C}UDOO<<HH(C5111CHw~~l3==xHHHr   c                   t          |t                    s|                     |f          d         S |                     |           |                     |          }t          |t
          t          f          rt          |          n|fS Nr   )r   tupleapplyr=  
apply_implr(  )r   r   r^  s      r   rD  zReduceScanOpInterface.applyK  sw    %'' 	,::ui((++%   ooe$$'dE];;HuSzzz#Hr   N)r*   r+   r,   r<   r;  r=  r@  rD  r2   r   r   r7  r7  1  sj        % % %M M M2 2 2I I II I I I Ir   r7  c                  >     e Zd Z fdZd Zd ZddZd Zd Z xZ	S )		ReduceOpsc                Z    t                                          ||           || _        d S r   )superr<   	keep_dims)r   rF  r9  rJ  	__class__s       r   r<   zReduceOps.__init__U  s(    z***"r   c                    g }|D ]f}||                     |           d}|                     |                     |j        j                                        |j                             gt          |          |fS rB  )appendr@  r   r   flattenr   rC  )r   r   rF  r^  r   s        r   unravelzReduceOps.unravelY  s     	S 	SD

4    

4>>$+*:*B*B*D*DdjQQRRRRSzz4r   c                      j         }                      j                   \  }g }g }d         j        j        j        }|d|         ||dz   d          z   }D ]Y}|                    |j        j                   |                    t          j        ||j        j        j                             Zt          |d         j
                  D ]}	t          j        |	|          d|         |dz   d          z   t           fdt          |          D                       }
|         dk    rMt          t          |                    D ]/}|
|         j        j                                        ||         <   0t           fdt          |          D                       }  j        j        g ||
R  }t%          |t                    s|fn|}t          t          |                    D ]\}t%          ||         t&          j        j                  r$||         j        j                                        n||         ||         <   ]g }t          |          D ]\  }	} j        rM|t          j        ||          }nKt          t          |                    D ]}t          j        |d          }n||                                }|                                         ||	         j                             |S )Nr   r
   r?   c              3  l   K   | ].\  }}                     |         |         j                  V  /d S r   r@  r   )r  iir!  r   input_indexr   s      r   	<genexpr>z+ReduceOps.generic_reduce.<locals>.<genexpr>r  sB      ssTYTVXYq~uRy O Ossssssr   c              3  l   K   | ].\  }}                     |         |         j                  V  /d S r   rR  )r  oior   output_indexr   s      r   rU  z+ReduceOps.generic_reduce.<locals>.<genexpr>x  sB      !w!wW\WY[\$..<%)/"R"R!w!w!w!w!w!wr   )rF  rO  r   r   r7   rM  rA   r  r   rD   r#  unravel_indexrC  r  rE   r]   r9  fnr   r   r  r   rJ  rJ  r@  )r   r   original_axisrF  
input_dataoutput_datainput_shapeoutput_shaper  r   input_tuplej	acc_tuplecombine_fn_retr^  r   _rT  rY  s   ``               @@r   generic_reducezReduceOps.generic_reducec  sE   	ll5$)44t
Aho*0"1T6*[-CC 	T 	TCcjo...rxCJO<QRRRSSSSz!})** 	H 	HA*1k::K&qv.TAXYY1GGLssssss]fgq]r]rsssssK4 A%%s;//00 U UA3>q>3H3M3R3R3T3TKN<00U "!w!w!w!w!w!w`iju`v`v!w!w!www	!3!3!MY!M!M!M!M6@QV6W6Wk^..]k	s;//00 H HAV`!!bgnW6 W6 4H9Q<3F3K3P3P3R3R3R;DQ<  N<00H  -- 	= 	=GAt~ 	# ,>$55DD"3{#3#344 7 7!~dA667 &yy{{JJt~~dE!HN;;<<<<
r   Nc                   t          |t                    r|d         n|}d }d }|r;|                      ||j        j        | j        | j                  |j                  }|r@|                      ||j        j        | j        | j                  t          j	                  }||||fS ||S ||S t          d          )Nr   rF  keepdimsz-val_reduce_op and idx_reduce_op are both None)r   rC  r@  r   r   rF  rJ  r   r   r   r   )r   r   val_reduce_opidx_reduce_oprV  idxs         r   min_maxzReduceOps.min_max  s    &ue44?a% 	y..u|/@ty[_[i!j!j!jlqlwxxC 	v..u|/@ty[_[i!j!j!jlnltuuC?s8O_J_JLMMMr   c                    |                      t          j        |j        j        | j        | j                  |j                  S )Nrh  )r@  rA   r0  r   r   rF  rJ  r   r   r   s     r   r0  zReduceOps.sum  s6    ~~bfU\%6TYQUQ_```bgbmnnnr   c                   | j         t          j        j        k    r2|                     |d         t
          j        t
          j                  S | j         t          j        j        k    r2|                     |d         t
          j	        t
          j
                  S | j         t          j        j        k    r(|                     |d         t
          j	        d           S | j         t          j        j        k    r(|                     |d         t
          j        d           S | j         t          j        j        k    r|                     |d                   S |                     |          S )Nr   )rj  rk  )r9  r   standard_argmin_combine_tie_break_leftrm  rA   minargmin_argmax_combine_tie_break_leftr8  argmax_elementwise_max_elementwise_min_sum_combiner0  rf  ro  s     r   rE  zReduceOps.apply_impl  s   ?bkHHH<<abi<XXX_ JJJ<<abi<XXX_ <<<<<ad<SSS_ <<<<<ad<SSS_ 88888E!H%%% &&u---r   r   )
r*   r+   r,   r<   rO  rf  rm  r0  rE  __classcell__rK  s   @r   rG  rG  S  s        # # # # #     ) ) )VN N N N$o o o. . . . . . .r   rG  c                  6     e Zd Z fdZd Zd Zd Zd Z xZS )ScanOpsc                Z    t                                          ||           || _        d S r   )rI  r<   reverse)r   rF  r9  r  rK  s       r   r<   zScanOps.__init__  s(    z***r   c                    |                      t          j        |j        j        | j                  |j                  gS Nr3  r?   )r@  rA   cumsumr   r   rF  r   ro  s     r   r  zScanOps.cumsum  s5    ry):KKKSXS^__``r   c                    |                      t          j        |j        j        | j                  |j                  gS r  )r@  rA   cumprodr   r   rF  r   ro  s     r   r  zScanOps.cumprod  s5    rz%,*;$)LLLTYT_``aar   c           	         g }g }d         j         j        j        }D ]Y}|                    |j         j                   |                    t	          j        ||j         j        j                             Zt          |d         j                  D ]}t	          j	        ||          t           fdt          |          D                       } j                 dk    rMt          t          |                    D ]/}||         j         j                                        ||         <   0t           fdt          t                              D                       t           fdt          |          D                       }	  j        j        g |	|R  }
t#          |
t                    s|
fn|
}	t          t          |                    D ]\}t#          |	|         t$          j        j                  r$|	|         j         j                                        n|	|         ||         <   ]g }t          |          D ]9\  }}|                                         ||         j                             :|S )Nr   r?   c              3  l   K   | ].\  }}                     |         |         j                  V  /d S r   rR  )r  rS  r!  indexr   r   s      r   rU  z'ScanOps.generic_scan.<locals>.<genexpr>  s?      ffur1%%)/BBffffffr   c              3  T   K   | ]"}|j         k    r|         d z
  n|         V  #dS )r
   Nr3  )r  r   r  r   s     r   rU  z'ScanOps.generic_scan.<locals>.<genexpr>  s>      "k"kTU1	>>58a<<uQx"k"k"k"k"k"kr   c              3  l   K   | ].\  }}                     |         |         j                  V  /d S r   rR  )r  rW  rX  r   
prev_indexr   s      r   rU  z'ScanOps.generic_scan.<locals>.<genexpr>  sB      !u!uUZUWYZ$..:b	"P"P!u!u!u!u!u!ur   )r   r   r7   rM  rA   r  r   rD   r#  rZ  rC  r  rF  rE   r]   r9  r[  r   r   r  r   r@  )r   r   r]  r^  r7   r  r   r   rb  rc  rd  r^  r  r  s   ``          @@r   generic_scanzScanOps.generic_scan  s   
a$* 	M 	MCcjo...rxSZ_5JKKKLLLLz!})** 	H 	HA$Q..EffffffPYZdPePefffffDTY1$$s;//00 G GA,0GN,?,D,D,F,FKN5))G #"k"k"k"k"kY^_bch_i_iYjYj"k"k"kkk
!!u!u!u!u!u!u^ghs^t^t!u!u!uuu	!3!3!FY!F!F!F!F6@QV6W6Wk^..]k	s;//00 H HAOY!!bgnP6 P6 -HIaL,?,D,I,I,K,K,K;DQ<  N5))H  -- 	= 	=GAtJJt~~dE!HN;;<<<<
r   c           	     B   g }| j         rW|D ]S}|                    |                     t          j        |j        j        | j                  |j                             Tn|}| j	        t          j        j        k    r|                     |d                   }nK| j	        t          j        j        k    r|                     |d                   }n|                     |          }| j         r4|D ]1}t          j        |j        j        | j                  |j        _        2|S )Nr3  r   )r  rM  r@  rA   flipr   r   rF  r   r9  r   rq  ry  r  _prod_combiner  r  )r   r   	new_inputr  r^  s        r   rE  zScanOps.apply_impl  s   	< 	 f f  
di0X0X0XZ]Zc!d!deeeef I?bk666++il++CC_ 999,,y|,,CC ##I..C< 	K K K"$'#*/	"J"J"J

r   )	r*   r+   r,   r<   r  r  r  rE  rz  r{  s   @r   r}  r}    sz            a a ab b b  <      r   r}  c                     dd} dd}| t           _        |t           _        | t           j        _        |t           j        _        d S )NFc                J    t          |||                              |           S r   )rG  rD  )r   rF  r9  rJ  r  s        r   _new_reducez'_patch_reduce_scan.<locals>._new_reduce  s"    z955;;EBBBr   c                J    t          |||                              |           S r   )r}  rD  )r   rF  r9  r  r  s        r   	_new_scanz%_patch_reduce_scan.<locals>._new_scan  s"    tZ1177>>>r   )F)r   reduceassociative_scanr  )r  r  s     r   _patch_reduce_scanr    sX    C C C C? ? ? ? BI#B BGN(BGr   c                   d }d
d}dd}d }|| _         || _        || _        t          | _        || j        _        t          |d          | _        t          |d          | _	        t          |d	          | _
        t                       d S )Nc                F   | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         d	k    r|	                                S | j         d
k    r|
                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S | j         dk    r|                                S t%          d|  d          )Nvoidr   r|   r{   r~   r}   r   r   r   rI   rk   rm   ro   fp16bf16fp32fp64zfail to convert z to ir type)r  get_void_tyr   r   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+   9&&(((Y&  &&(((Y&  &&(((Y'!!'')))Y'!!'')))Y(""((***Y'!!'')))Y(""((***Y'!!'')))Y(""((***Y'!!'')))Y)##))+++Y*$$**,,,Y&  &&(((Y&  &&(((Y&  '')))Y&  ((***=D===>>>r   c                B    |d}|d| }}n| |}}t          |||          S )Nr
   r   )rD   )arg1arg2stepr  r)  ends         r   
_new_rangez$_patch_lang_core.<locals>._new_range+  s7    <D<D3EEt3EUC&&&r    c                    | s
J |            d S r   r2   )r  r  s     r   _new_static_assertz,_patch_lang_core.<locals>._new_static_assert4  s    Sr   c                X   t          | t          j                  s| S t          |t          t          f          s|gn|}d |D             }t          |          t          dt          | j                            k    rt          d|           | j	        
                    ||           | S )Nc                T    g | ]%}t          |t          j                  r|j        n|&S r2   )r   r   	constexprr(   r  r  s     r   r  z7_patch_lang_core.<locals>._set_attr.<locals>.<listcomp>=  s/    PPPAZ2<88?!''aPPPr   r
   z$len(values) != len(input.shape) for )r   r   r   r(  rC  rE   r8  r7   r   r   r)   )r   rs  r  s      r   	_set_attrz#_patch_lang_core.<locals>._set_attr7  s    %++ 	L!+FT5M!B!BN&PPPPPv;;#aU[!1!12222JDJJKKKdF+++r   ztt.divisibilityr  ztt.contiguityztt.constancy)NN)r  )rD   static_rangestatic_assertr  static_printr   to_irr   multiple_ofmax_contiguousmax_constancyr  )langr  r  r  r  s        r   _patch_lang_corer    s    $? $? $?P' ' ' '   
 
 
 DJ"D+DD!DJy/@AAAD!)/BBBD @@@Dr   c                   d | j                                         D             }t          |          dk    s
J d            |D ]y}t          |t                     t          |j        t                     |t          k    rt          |j        t                     t          |j                   t          |           zt          t          j
        j        t                     d S )Nc                j    g | ]0\  }}t          j        |          |t          t          j        fv .|1S r2   )r  ismoduler   r  )r  re  r(   s      r   r  z_patch_lang.<locals>.<listcomp>P  sA    pppxq%W=Me=T=TpY^cegignboYoYoUYoYoYor   r
   z:triton.language must be visible from within jit'd function)__globals__r  rE   r   interpreter_builderr   r   r   r5  r  r  tensor_descriptor_base)r[  langsr  s      r   _patch_langr  O  s    pp2>#7#7#9#9pppEu::???X???  t0111t{$78882::49&9:::4;'''2713FGGGGGr   c                v    t          | d          r t          |           | n t          |           |          S )N_fields)r!   rB  )r  contentss     r   _tuple_creater  \  s;     $+3	#:#:S9499h	S		(@S@SSr   c                F   t          | t                    rt          j        t          j        j                            |                     }t          j	        }d| cxk    rdk     rn nt          j	        }nid| 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                            |                     }t          t          j        |                                 gt          j                  |          }t          j        ||          S t          | t&                    r#t)          | t+          t,          |                     S t          | t.                    rd
 | j        D             }| j        d         dk    sJ t          j        d          |d<   t5          t7                                }|                    t-          | j                  d | j        D             |d | j        D                       S | S )Ni   l        l        l         l            l            zUnsupported integer value r?   data_ptrc                ,    g | ]}t          |          S r2   _implicit_cvtr  ss     r   r  z!_implicit_cvt.<locals>.<listcomp>}  s     999=##999r   r\   r
   c                ,    g | ]}t          |          S r2   r  r  s     r   r  z!_implicit_cvt.<locals>.<listcomp>  s     777=##777r   c                6    g | ]}t          j        |          S r2   )r   r  )r  r   s     r   r  z!_implicit_cvt.<locals>.<listcomp>  s     BBBQaBBBr   )r6   r7   r8   r:   ) r   rw   r   	str_to_tytritonruntimejitmangle_typerA   r   r   r   rI   r   r   r#  r   r!   r  rC  r  mapr  r	   r8   r  r   r   make_tensor_descriptorr6   r7   r:   )r  tyr   r   r8   r  s         r   r  r  f  s   #s %\&.,88==>>S    5     HEEc!!!!E!!!!!IEEs""""U"""""HEEc!!!!E!!!!!IEE?#??@@@bhuE:::B??y$$$sJ 
\&.,88==>>bh'7ryIII2NNy$$$	C		 
S#mS"9"9:::	C)	*	* 

99S[999{2!####l1oo!"4"6"677..sx((77SY777BB#/BBB	 / 
 
 	
 Jr   c                \    t          | t          j        j        j                  r| j        S | S r   )r   r  r  r  TensorWrapperr6   )ts    r   _unwrap_tensorr    s(    !V^'566 vHr   c                    t          |t          j        j        j                  r*t          j        j                            | |j                  S | S r   )r   r  r  r  r  r   )r  original_tensors     r   _rewrap_tensorr    s@    /6>#5#CDD J~!//?3HIIIHr   c                  &    e Zd Zd Zd Zd Zd ZdS )GridExecutorc                    ddl m || _        || _        || _        fd|j                                        D             fd|D             | _        d S )Nr
   )_normalize_tyc                .    i | ]\  }}| |          S r2   r2   )r  r  r  r  s      r   r  z)GridExecutor.__init__.<locals>.<dictcomp>  s)    ^^^xtR4r!2!2^^^r   c                F    g | ]}                     |          d k    |S )r  )get)r  r  r.   s     r   r  z)GridExecutor.__init__.<locals>.<listcomp>  s2    bbbD9L9LT9R9RVa9a9a49a9a9ar   )r  r  r[  	arg_namesgridr.   r  
constexprs)r   r[  r  r  r.   r  s       @@r   r<   zGridExecutor.__init__  sr    &&&&&&"	^^^^2CUC[C[C]C]^^^bbbbIbbbr   c                    i fdfd|D             }i }|                                 D ]\  }} |          ||<   ||fS )Nc                F   t          | t                    rt          | t          |                     S t          | t                    r/t	           | j                  | j        | j        | j                  S t          | d          s| S t          |           }|                                                                vr=|                                }|                                |                                <   |                                                                         }|                    dd          }|                    ||                                |                                |                                           t'          ||           }|S )Nr  r   cpu)device)r  )r   rC  r  r  r	   r6   r7   r8   r:   r!   r  untyped_storager  r  	new_emptyset_storage_offsetr#  r^   r  )r  unwrapped_argstoragecpu_arg_to_cpustoragess       r   r  z,GridExecutor._init_args_hst.<locals>._to_cpu  sx   #u%% 
$S#gs*;*;<<<C!122 'GCH%%IKO	   S*-- 
*3//M,,..7799II'7799/6{{}}))++,}<<>>GGIIJG#--a->>GLL-">">"@"@-BTBTBVBVXeXlXlXnXnooo$WcBBBGNr   c                &    g | ]} |          S r2   r2   )r  r  r  s     r   r  z/GridExecutor._init_args_hst.<locals>.<listcomp>  s!    555SGGCLL555r   r  )	r   args_devr  args_hst
kwargs_hstr'   r(   r  r  s	          @@r   _init_args_hstzGridExecutor._init_args_hst  s    	 	 	 	 	 	0 6555H555 
 ,,.. 	- 	-JC%gennJsOO##r   c                  
 i 
fd
t          ||          D ]\  }} 
||           |                                D ]\  }}||         }	 
||	                                           D ]\  }}|                    |           d S )Nc                   t          | d          rot          |           t          |          }} |                                 |                                f|                                                                 <   d S t	          | t
                    r$t          | |          D ]\  } } | |           d S t	          | t                    r | j        |j                   d S d S )Nr  )	r!   r  r  r  r   rC  zipr	   r6   )arg_devarg_hst	_from_cpur  s     r   r  z1GridExecutor._restore_args_dev.<locals>._from_cpu  s    w
++ 6#1'#:#:N7<S<SBIBYBYB[B[]d]t]t]v]vAw0022;;==>>>GU++ 6*-gw*?*? 0 0&WgIgw////0 0G%566 6	',555556 6r   )r  r  rs  copy_)r   r  r  r  r  r  r  r'   	kwarg_dev	kwarg_hstr  r  s             @@r   _restore_args_devzGridExecutor._restore_args_dev  s    		6 		6 		6 		6 		6 		6 !$Hh 7 7 	( 	(GWIgw'''' %llnn 	, 	,NC"3IIi++++"*//"3"3 	# 	#WgMM'""""	# 	#r   c                    |                     dd          rd S t          j         j                  fd|                                D             }                     ||          \  }}t           j                   t          j         j        g|R i |} fd|                                D             }t           j	                  r 	                    |          n j	        }t          |          dk    s
J d            |ddt          |          z
  z  z   }t          j        |  	 t          |d                   D ][}t          |d	                   D ]C}t          |d
                   D ]+}	t                              |||	             j        di | ,D\nF# t          $ r9}
t           j        j        j        r t)          t+          |
                    |
d }
~
ww xY w                     ||||           d S )NwarmupFc                .    i | ]\  }}|j         v ||S r2   )r  )r  r  r  argspecs      r   r  z)GridExecutor.__call__.<locals>.<dictcomp>  s+    GGG41aQ',5F5F!Q5F5F5Fr   c                L    i | ] \  }}||j         v r|nt          |          !S r2   )r  r  )r  r  r  r   s      r   r  z)GridExecutor.__call__.<locals>.<dictcomp>  s:    kkkQZQUWZTT_44cc-:L:Lkkkr      z#grid must have at most 3 dimensions)r
   r   r
   r   r2   )popr  getfullargspecr[  r  r  r  getcallargscallabler  rE   r  r   rD   r   	Exceptionr  knobscompilationfront_end_debuggingr   r.  r  )r   r  r  r  r  r  r  r   r   r   er  s   `          @r   __call__zGridExecutor.__call__  s%   ::h&& 	F (11GGGG6<<>>GGG#228VDD*DG "47DXDDDDDkkkk^b^h^h^j^jkkk"*49"5"5Dtyy494yyA~~~D~~~eq3t99}--($//		347^^ ( (tAw ( (A"47^^ ( (+88AqAAA$(((
  	3 	3 	3|'; "477++2	3
 	x6:FFFFFs   /A1F! !
G$+4GG$N)r*   r+   r,   r<   r  r  r  r2   r   r   r  r    sY        c c c!$ !$ !$F# # #2 G  G  G  G  Gr   r  c                      e Zd Zd ZdS )ASTTransformerc           	        g }|j         D ]}||                     |          gz  }t          |          dk    rt          d          t	          j        t	          j        t	          j        dt	          j                              dt	          j                              |j	        t	          j
        d          gg 	          |_	        |S )
Nr
   z&Multiple assignments are not supportedinterpreter_semantic)idctxr@  )r(   r   r  F)r(   )funcr  keywords)targetsvisitrE   r   astCall	AttributeNameLoadr(   Constant)r   nodenamestargets       r   visit_AssignzASTTransformer.visit_Assign  s    l 	* 	*Fdjj(())EEu::>>EFFF XSX1GSXZZ%X%X%X_j#&8::/ / /6:j#,UZB[B[B[5\gik k k
 r   N)r*   r+   r,   r&  r2   r   r   r  r    s#            r   r  c                  L    e Zd Z e            Zd Zd Zd Zd Zd Z	d Z
d ZdS )	FunctionRewriterc                >    || _         || _        d| _        d| _        d S )Nr  r   )r[  r  filenamedef_file_lineno)r   r[  r  s      r   r<   zFunctionRewriter.__init__  s%    $%r   c                `   	 t          j        | j                  \  }}n# t          $ r
 | j        cY S w xY w|                                 \  | _        | _        |                     |          | _        | 	                    |          }| 
                    |          }|                     |          S r   )r  getsourcelinesr[  r  _get_jit_fn_file_liner*  r+  	_find_def
def_lineno_prepare_source_transform_ast_compile_and_exec)r   linesre  rc  transformed_asts        r   rewrite_astzFunctionRewriter.rewrite_ast  s    	-dg66HE11 	 	 	7NNN	 /3.H.H.J.J+t+..//""5))--c22%%o666s    33c                D    ddl m}m}  | || j                            S )Nr
   )get_jit_fn_file_lineJITFunction)r  r8  r9  r[  )r   r8  r9  s      r   r.  z&FunctionRewriter._get_jit_fn_file_line2  s7    ::::::::##KK$8$8999r   c                    d}t          |          D ]1\  }}|                                                    d          r|dz   }2|S )Nr   zdef r
   )r  strip
startswith)r   r4  r0  r   lines        r   r/  zFunctionRewriter._find_def6  sP    
 '' 	# 	#GAtzz||&&v.. #U
r   c                x    || j         dz
  d          }d                    |          }t          j        |          S )Nr
   r  )r0  jointextwrapdedent)r   r4  rc  s      r   r1  z FunctionRewriter._prepare_source>  s8    do)**+ggenns###r   c                    t          j        |          }| j                            |          }t          j        |           | j        dz
  }t          j        ||           |S r  )r  parseast_transformerr  fix_missing_locationsr+  increment_lineno)r   rc  
parsed_astr5  
inc_linenos        r   r2  zFunctionRewriter._transform_astC  s^     Ys^^
.44Z@@!/222)A-
_j999r   c                    t          || j        d          }i | j        }| j        j        }t                                                      D ]\  }}||vr|||<   t          |||           || j        j                 S )Nexec)r*  mode)	compiler*  r  r[  r  globalsr  rJ  r*   )r   r5  compiled_codelocal_namespace
fn_globalsr'   r(   s          r   r3  z"FunctionRewriter._compile_and_execN  s    $-fUUU)T[/W(
!))//++ 	( 	(JC*$$"'
3]J888tw/00r   N)r*   r+   r,   r  rD  r<   r6  r.  r/  r1  r2  r3  r2   r   r   r(  r(    s        $n&&O& & &7 7 7(: : :  $ $ $
	 	 	1 1 1 1 1r   r(  c                  B    e Zd Zi Zd	dZd Zed             Z d Zd ZdS )
InterpretedFunctionr   r   c                     | _         t          |fi | _         fd}| _        t	          j        |          }d |j                                        D              _        d S )Nc                 t    |d         }                                 } t          |j        |          | i |S )Nr  rewriter  r  )r  r  r  r[  r   s       r   runz)InterpretedFunction.__init__.<locals>.runa  s<    &>DB9<DND994J6JJJr   c                    g | ]	}|j         
S r2   r  r  s     r   r  z0InterpretedFunction.__init__.<locals>.<listcomp>h  s    HHHQ!&HHHr   )	r[  r(  rewriterrW  r  	signature
parametersrs  r  )r   r[  r  rW  rZ  s   `    r   r<   zInterpretedFunction.__init__]  s    (66v66	K 	K 	K 	K 	K
 %b))	HH)*>*E*E*G*GHHHr   c                    | j         | j        vr&| j                                        | j        | j         <   | j        | j                  S r   )r[  rewritten_fnrY  r6  r   s    r   rV  zInterpretedFunction.rewritej  s?    7$+++)-)B)B)D)DDdg& ))r   c                    | j         j        S r   )r[  r*   r   s    r   r*   zInterpretedFunction.__name__o  s    wr   c                V    |                                  }t          || j        |          S r   rU  )r   r  r[  s      r   __getitem__zInterpretedFunction.__getitem__s  s#    \\^^B555r   c                    t          | j                   |                                 }	  ||i |S # t          $ r"}t	          t          |                    |d }~ww xY wr   )r  r[  rV  r  r   r.  )r   r  r  r[  r  s        r   r  zInterpretedFunction.__call__w  sn    DG\\^^	32t&v&&& 	3 	3 	3"477++2	3s   2 
AAANr  )	r*   r+   r,   r]  r<   rV  r3  r`  r  r2   r   r   rR  rR  Y  sy        LI I I I* * *
     X 6 6 63 3 3 3 3r   rR  )E
__future__r   r  r@  r  typingr   r   r   r   numpyrA   r  triton.languagelanguager   r/   r   triton.language.semanticr   triton.tools.tensor_descriptorr	   errorsr   	functoolsr   _C.libtritonr   rY  r   r   r   r4   rU   re   r   r   r   r   r   	vectorizer   r  r   r  rI   r  r   r   r  r   r5  r7  rG  r}  r  r  r  r  r  r  r  r  r  r  NodeTransformerr  r(  rR  r2   r   r   <module>rn     s   " " " " " " 



   $ $ $ $ $ $ $ $ $ $                 ! ! ! ! ! ! 3 3 3 3 3 3 ; ; ; ; ; ; $ $ $ $ $ $       6 6 6 6 6 6 $ $ $ $ $ $        6       4# # # # # # # #L $
& 
& 
& 
& 
& 
& 
& 
&	 	 	  @=' =' ='@  
# # # bl4555bl4555Z<<<p p p p p p p pD8 D8 D8 D8 D8 D8 D8 D8N# # #4 4 4( ( (.I I I I I I I ID]. ]. ]. ]. ].% ]. ]. ].@; ; ; ; ;# ; ; ;|) ) ) K K K\
H 
H 
HT T T! ! !H )(** %~&9::     gG gG gG gG gG gG gG gGT    S(    B1 B1 B1 B1 B1 B1 B1 B1J%3 %3 %3 %3 %3 %3 %3 %3 %3 %3r   