o
    灛iD                     @   sB  d dl Z d dlZd dlZd dlZd dlmZ d dlmZ d dl	m
Z
 d dlZd dlmZ d dlmZ dZedkreed	Zejd
dd ejddedddd ejddeddd ejddeddd ejddeddd ejdd edd!d ejd"d#ed$dd% ejd&d'ed(dd% e ZejrejnejZejrejneeZeejZejd eej ej ej!eZ"ej#e"Z$e"j%&e$ e'e$ejZ(ej)*d)Z)e+e)dksJ e,e-d*d+ ej.*d)Z.d,e
e fd-d.Z/d/ej0 d0ej1 Z2e/e.e2g Z3d1d2 Z4d3d4 e5e.D Z6d5d4 e67 D Z6d6d4 e5e.D Z8d7d4 e87 D Z8d8d4 e5e.D Z.d99d:d; e8: D Z;d<d; e8< D Z=e=d=ej0 d>ej1 g7 Z=e6: D ]Z>e>d?v sgJ d@e> qYdAd; e67 D Z?dBd; e67 D Z@ejAjBe?e@dCZCe@D ]
ZDe8EeDdi qejAjFe(e8e.eCdDZGej0ej1dEZHejIeGeHdFZJg ZKg ZLe.< D ]ZDeDe@vreKe(jKeD g7 ZKeLe.eD g7 ZLqee.: eCZMdG9ee3eMgZNee OeJjPdH dIdJ ZQi dKeNdLejdMe+eQdNdO9dPd; eReQdddI eQdddI D d,dO9dQd; eReKeLD dRdO9dSd; e.< D dTdO9dUd; eKD dVe+eKdWe=dXeJjSjTdYej0dZdG9e;e2gd[e)d  d\e)d d]e)dI d^dZUd_D ]@ZVeeWjd`eV  ZXeYdae3 dGeM daeV ZdbZ[e[\eeX] j^dci eU W d   n	1 sw   Y  q^dS dS )d    N)ArgumentParser)Path)List)kernel_suffix	ty_to_cppa  
Triton ahead-of-time compiler:

This program compiles the kernel with name `kernel-name` in the file at the
provided `path` into self-contained C source-code that embeds the `cubin`
data along with utilities to load, unload and launch the kernel.

signature is provided as a list of (optionally divisibility-hinted) types
or constexpr values, e.g.

`compile.py --kernel-name kernel --signature "*fp32:16, i32:16, 1024, i32" --out-name kernel /path/to/kernel.py`

will compile triton.JITFunction of name `kernel` inside the file `/path/to/kernel.py`.
Said kernel will be specialized such that argument 0, 1 are assumed to be multiple of 16,
and argument 2 is assumed to be a compile-time constant of value 1024, i.e. it won't be part of the generated prototype.

The resulting entry point will have signature

CUresult kernel_{specialization_suffix}(CUstream stream, unsigned gX, unsigned gY, unsigned gZ, float* arg0, int32_t arg1, int32_t arg2)

Different such specialized entry points can be combined using the `linker.py` script.

NOTE: when resolving the scope of /path/to/kernel.py, the file will be executed from within its parent directory with the python interpreter
used to run this `compile.py` script
__main__)descriptionpathzTPath to Python source containing desired kernel in its scope. File will be executed.)helpz--kernel-namez-n zName of the kernel to compileT)typedefaultr   requiredz--num-warpsz-w   z$Number of warps to launch the kernel)r   r   r   z--num-stagesz-ns   z/Number of stages (meta-parameter of the kernel)z
--out-namez-onz Out name for the compiled kernelz
--out-pathz-ozOut filenamez--signaturez-szSignature of the kernel)r   r   r   z--gridz-gzLaunch grid of the kernel,c                 C   s
   |  dS )N )strip)s r   V/sda-disk/www/egybert/egybert_env/lib/python3.10/site-packages/triton/tools/compile.py<lambda>F   s   
 r   	signaturec                 C   s,   t  }|d|   | d d S )Nr      )hashlibsha256updatejoinencode	hexdigest)r   mr   r   r   hash_signatureH   s   r"   warpsxstagesc                 C   sF   zt | }|W S  ty   Y nw zt| }|W S  ty"   Y d S w N)int
ValueErrorfloat)r   retr   r   r   	constexprP   s   r*   c                 C   s,   i | ]\}}d |v r|t |d d qS ):r   )r*   split.0ir   r   r   r   
<dictcomp>]   s   , r0   c                 C      i | ]\}}|d ur||qS r%   r   r.   kvr   r   r   r0   ^       c                 C   s   i | ]	\}}|t |qS r   )r*   r-   r   r   r   r0   _   s    c                 C   r1   r%   r   r2   r   r   r   r0   `   r5   c                 C   s(   i | ]\}}|t vr||d d qS )r+   r   )	constantsr,   r-   r   r   r   r0   a      ( xc                 C   s   g | ]}t |qS r   )str)r.   r4   r   r   r   
<listcomp>b   s    r:   c                 C   s$   g | ]}t j|  d t|  qS )=)kernel	arg_namesr6   r.   r/   r   r   r   r:   c   s   $ z
num_warps=znum_stages=)r      z#Only 1 and 16 are valid hints, got c                 C      g | ]
\}}|d kr|qS )r?   r   r.   r/   hr   r   r   r:   i       c                 C   r@   )r   r   rA   r   r   r   r:   j   rC   )divisible_by_16
equal_to_1)fnr6   r   attrs)	num_warps
num_stages)options_cubin   kernel_nametriton_kernel_namebin_sizebin_dataz, c                 C   s   g | ]\}}d | | qS )0xr   )r.   r8   yr   r   r   r:      r5   c                 C   s"   g | ]\}}t | d | qS r   r   )r.   nametyr   r   r   r:      s   " full_signaturec                 C   s(   g | ]}t t|  d tj|  qS rU   )r   r   r<   r=   r>   r   r   r   r:      r7   arg_pointersc                 C   s   g | ]}d | qS )&r   )r.   argr   r   r   r:      s    num_argskernel_docstringsharedrH   	algo_infogridXgridYgridZ_placeholder)rB   czcompile..wr   )_binasciir   importlib.util	importlibsysargparser   pathlibr   typingr   tritontriton.compiler.code_generatorr   triton.backends.nvidia.driverr   desc__name__parseradd_argumentr9   r&   
parse_argsargsout_namerO   out_pathr
   arg_pathinsertparentutilspec_from_file_locationstemspecmodule_from_specmodloaderexec_modulegetattrr<   gridr,   lenlistmapr   r"   rH   rI   meta_sigsig_hashr*   	enumeratehintsitemsr6   r   values	const_sigkeys
doc_stringrB   rD   rE   compilerAttrsDescriptorrG   r/   r   	ASTSourcesrcoptscompileccinfor=   	arg_typessuffix	func_namehexlifyasmhex_zipmetadatar^   paramsext__file__template_pathwith_suffixopenfpwrite	read_textformatr   r   r   r   <module>   s    



.	
$ g