a
    ¶DfRG                     @   s   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mZm	Z	 ddl
mZ ddlmZ dd Zd'd
dZd(ddZd)ddZG dd dZG dd dZdd Zd*ddZd+ddZdd  Zed,d#d$Zd-d%d&ZdS ).    N)contextmanager)AnyDictList   )language)runtimec                 C   sL   d | } dddd|  dg}t|}|tjjd}dd |D }|S )	N,
nvidia-smi-i0z--query-gpu=z--format=csv,noheader,nounitsc                 S   s   g | ]}t |qS  )int.0xr   r   [/nfs/NAS7/SABIOD/METHODE/ermites/ermites_venv/lib/python3.9/site-packages/triton/testing.py
<listcomp>       znvsmi.<locals>.<listcomp>)join
subprocesscheck_outputdecodesysstdoutencodingsplit)attrscmdoutretr   r   r   nvsmi   s    

r!      c                 C   s  dd l }|j |j kr$td|   |d urT|D ]}|  |d d |_q6|j }|j	| |   W d    n1 s0    Y  |j
  |jjdd}|jjdd}|  |  |  |j
  ||}tdt|| }	|j }|j	|B t|	D ](}
|d ur4|D ]}d |_q&|   qW d    n1 sT0    Y  |j
  g }d}t|D ]X}
|jjdd}|jjdd}|  |  |  |j
  ||||	 g7 }qx||| S )Nr   zQCannot capture graph in default stream. Please use side stream in benchmark code.TZenable_timingr   
   )torchcudaZcurrent_streamZdefault_streamRuntimeErrorZdetach_Zrequires_grad_gradZ	CUDAGraphgraphsynchronizeEventrecordZreplayelapsed_timemaxr   rangemeantensoritem)fnrepgrad_to_noner%   r   gstart_event	end_eventestimate_msn_repeatir    Z	n_retriesr   r   r   do_bench_cudagraph   sP    

$





*

r<      d   Tr0   c                    s  |dv sJ dd l  |    j  |r@ jtd jdd}n jtd jdd} jjdd} jjdd}	|  td	D ]}
|	  |   q|	   j  |
|	d	 }td
t|| }td
t|| } fddt|D } fddt|D }	t|D ]}
|   qt|D ]H}|d urD|D ]}d |_q6|	  ||   |   |	|   q$ j   jdd t||	D  jd}|d ur؈ | j| jd }t|d
kr|d }|S t || S )N)minr.   r0   Zmedianr   g    Ar&   )dtypedeviceg    ATr#      r   c                    s   g | ]} j jd dqS Tr#   r&   r+   r   r;   r%   r   r   r      r   zdo_bench.<locals>.<listcomp>c                    s   g | ]} j jd dqS rC   rD   rE   rF   r   r   r      r   c                 S   s   g | ]\}}| |qS r   )r-   )r   ser   r   r   r      r   )r@   )r%   r&   r*   emptyr   int8r+   r,   r/   Zzero_r-   r.   r(   r1   zipfloatZquantiletolistlengetattrr2   )r3   Zwarmupr4   r5   Z	quantilesZ
fast_flushZreturn_modecacher7   r8   _r9   Zn_warmupr:   r;   r   timesr    r   rF   r   do_benchQ   sL    





 
rS    c                 C   sN  dd l }dd l}t| |js&|| } t||js<||}|d u rHd}t|rZ|| jn|}|d u rjd}t|r||| jn|}t| |jr| j|jkr|  } | 	 
   } t||jr|j|jkr| }|	 
   }| jdks|jdkr|jj| |||dd d S |j| |||dsJt| d|  d	| d
| d| d
d S )Nr   g{Gz?g        r   T)atolrtolZ	equal_nan)rU   rV    z is not close to z (atol=z, rtol=))numpyr%   
isinstanceZTensorr1   callabler@   bfloat16rL   cpudetachsizeZtestingZassert_allcloseZallcloseAssertionError)r   yrU   rV   err_msgnpr%   r   r   r   assert_close   s2    

rd   c                   @   sL   e Zd ZdZdee ee eee ee eeeef eeeedddZ	dS )		Benchmarkzk
    This class is used by the :code:`perf_report` function to generate line plots with a concise API.
    rT   FN)x_namesx_valsline_arg	line_vals
line_names	plot_nameargsxlabelylabelx_logy_logc                 C   sL   || _ || _|
| _|| _|| _|| _|| _|| _|| _|	| _	|| _
|| _dS )a  
        Constructor.
        x_vals can be a list of scalars or a list of tuples/lists. If x_vals is a list
        of scalars and there are multiple x_names, all arguments will have the same value.
        If x_vals is a list of tuples/lists, each element should have the same length as
        x_names.

        :param x_names: Name of the arguments that should appear on the x axis of the plot.
        :type x_names: List[str]
        :param x_vals: List of values to use for the arguments in :code:`x_names`.
        :type x_vals: List[Any]
        :param line_arg: Argument name for which different values correspond to different lines in the plot.
        :type line_arg: str
        :param line_vals: List of values to use for the arguments in :code:`line_arg`.
        :type line_vals: List[Any]
        :param line_names: Label names for the different lines.
        :type line_names: List[str]
        :param plot_name: Name of the plot.
        :type plot_name: str
        :param args: Dictionary of keyword arguments to remain fixed throughout the benchmark.
        :type args: Dict[str, Any]
        :param xlabel: Label for the x axis of the plot.
        :type xlabel: str, optional
        :param ylabel: Label for the y axis of the plot.
        :type ylabel: str, optional
        :param x_log: Whether the x axis should be log scale.
        :type x_log: bool, optional
        :param y_log: Whether the y axis should be log scale.
        :type y_log: bool, optional
        N)rf   rg   ro   rh   ri   rj   rp   stylesrm   rn   rk   rl   )selfrf   rg   rh   ri   rj   rk   rl   rm   rn   ro   rp   colorrq   r   r   r   __init__   s    .zBenchmark.__init__)rT   rT   FFNN)
__name__
__module____qualname____doc__r   strr   r   boolrt   r   r   r   r   re      s(         
re   c                   @   s4   e Zd Zdd ZdeeeedddZddd	Zd
S )Markc                 C   s   || _ || _d S N)r3   
benchmarks)rr   r3   r}   r   r   r   rt     s    zMark.__init__F)bench	save_path
show_plots
print_datac              
      sp  dd l }dd lm} dd l}	|j}
dd |jD }dd |jD }t|j}|	j||
 | | d}|jD ]
 t	 tt
fs fdd|D  t t|krtdt| d  tt| }g g g   }}}|jD ]z}| jf i ||j|i|j|}z|\}
}}W n$ ty6   |d d   }
}}Y n0 ||
g7 }||g7 }||g7 }qt | | | |jt|< qj|jr|  | }|d }t|jD ]\}}||d	  ||d
   }}|jr|j| d nd }|jr|j| d nd }|j|| || |||d |  s|  s|t}|t}|j|| ||d|d q|   |!|j"pv| |#|j$ |%|j&rdnd |'|j(rdnd |r|)  |r|*|j+,||j d |||j  }|r(|j-d dkr(|j./ \}}|| ||  |d< |rDt0|jd  t0| |rl|j1|j+,||j dddd |S )Nr   c                 S   s   g | ]}| d qS )-minr   r   r   r   r   r     r   zMark._run.<locals>.<listcomp>c                 S   s   g | ]}| d qS )-maxr   r   r   r   r   r     r   )columnsc                    s   g | ]} qS r   r   )r   rQ   r   r   r   r     r   z	Expected z values, got r   r   r   )labelrs   Zlsg333333?)alphars   logZlinearz.png   ZDiff:z.csvz%.1fF)Zfloat_formatindex)2osZmatplotlib.pyplotZpyplotZpandasrj   listrf   Z	DataFramerg   rZ   tuplerN   
ValueErrordictrK   ri   r3   rh   rl   	TypeErrorlocrk   figureZsubplot	enumeraterq   ZplotZisnullallZastyperL   Zfill_betweenZlegendZ
set_xlabelrm   Z
set_ylabelrn   Z
set_xscalero   Z
set_yscalerp   showZsavefigpathr   shaper   rM   printZto_csv)rr   r~   r   r   r   Zdiff_colZkwragsr   ZpltpdZy_meanZy_minZy_maxrf   ZdfZx_argsZrow_meanZrow_minZrow_maxra   r    axZfirst_xr;   colZstyZcol0Zcol1r   r   r   _run
  sr    

$

 

"z	Mark._runrT   c                 K   s   t | jt}|r| jgn| j}g }|rDttj|dd}	|	d |D ]:}
|| j	|
|||fi | |rH|	d|
j
 d qH|r|	d |r|r|d S |S d S )Nzresults.htmlwz<html><body>
z<image src="z.png"/>
z</body></html>
r   )rZ   r}   re   openr   r   r   writeappendr   rk   )rr   r   r   r   Z	return_dfkwargsZhas_single_benchr}   Z
result_dfshtmlr~   r   r   r   runM  s"    

zMark.runN)F)FFrT   F)	ru   rv   rw   rt   re   ry   rz   r   r   r   r   r   r   r{     s   Cr{   c                    s    fdd}|S )z
    Mark a function for benchmarking. The benchmark can then be executed by using the :code:`.run` method on the return value.

    :param benchmarks: Benchmarking configurations.
    :type benchmarks: List of :class:`Benchmark`
    c                    s
   t |  S r|   )r{   )r3   r}   r   r   <lambda>i  r   zperf_report.<locals>.<lambda>r   )r}   wrapperr   r   r   perf_reportb  s    r   c                 C   sf   ddl }ddlm} | s tjj} |s.|j }|j|d }|j|d }|| d d d	 }|S )
z return DRAM bandwidth in GB/s r   Nr   driverZmem_clock_rateZmem_bus_widthr   g    .A   )	r%   r   r   backendCUDAr&   current_deviceutilsget_device_properties)r   rA   r%   r   Zmem_clock_khzZ	bus_widthZbw_gbpsr   r   r   get_dram_gbpsm  s    
r   c           
      C   s   dd l }ddlm} |s tjj}|s.|j }|j|d d }|j	|}|d dk rn| |j
kshJ d}nV| |j|jfv rd}n@| |j
|j|jfv rd}n&| |jtjtjtjfv rd	}ntd
|| | d }	|	S )Nr   r   r   multiprocessor_count   r      i   i   dtype not supported&.>)r%   r   r   r   r   r&   r   r   r   get_device_capabilityfloat16float32int32r\   int16rJ   tlZ
float8e4nvZfloat8e4b15Zfloat8e5r'   
r@   Z
clock_rater   rA   r%   r   Znum_subcoresZ
capabilityZops_per_sub_coretflopsr   r   r   get_max_tensorcore_tflops|  s(    
r   c                     s    fdd}|S )Nc                    s   t   fdd}|S )Nc            
         s   dd l }|t  }  | k}|r|dkrtjjd }tj	d dd}d|v shJ d|d j
jj}| d	j d
| d}tjddd|gd|d}	|	jdksJ ddt|	jv sJ n| i | d S )Nr   zcuda-memcheck__file__PATH1)r   ZPYTORCH_NO_CUDA_MEMORY_CACHINGrequestz@memcheck'ed test must have a (possibly unused) `request` fixturez::[]Zpytestz-vsT)capture_outputenvz7cuda-memcheck returned an error: bounds checking failedzERROR SUMMARY: 0 errors)psutilProcessr   getppidnameitemsr   realpath__globals__environnodeZcallspecidru   r   r   
returncodery   r   )
rl   r   r   Z	ppid_nameZrun_cuda_memcheckr   r   Ztest_idr   r   )target_kwargstest_fnr   r   r     s    z1cuda_memcheck.<locals>.decorator.<locals>.wrapper)	functoolswraps)r   r   r   )r   r   	decorator  s    z cuda_memcheck.<locals>.decoratorr   )r   r   r   r   r   cuda_memcheck  s    r   F    c              
   c   s$  zt g d t dddd|  d|  g t dddd| d| g tdgd	 }td
gd	 }t||  dk sJ d|  dt|| dk sJ d| dd|  }d| d }||fV  W t g d t g d t g d n,t g d t g d t g d 0 d S )N)r
   r   r   -pmr   r
   r   r   z--lock-gpu-clocks=r	   z--lock-memory-clocks=zclocks.current.smr   zclocks.current.memoryr$   zGPU SMs must run at z MHzg 3O?i   gMbP?)r
   r   r   r   r   )r
   r   r   z-rgc)r
   r   r   z-rmc)r   r   r!   abs)Zref_sm_clockZref_mem_clockZcur_sm_clockZcur_mem_clockr   Zgbpsr   r   r   set_gpu_clock  s6      r   c           
      C   s   dd l }ddlm} |s tjj}|s.|j }|j|d d }|j	 }|d dk r| |j
krhd}q| |jkrxd}qtd	n.| |j
krd}n| |j|jfv rd}ntd	|| | d
 }	|	S )Nr   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   get_max_simd_tflops  s*    





r   )r"   N)r=   r>   NNTr0   )NNrT   )NN)NN)r   r   )NN)r   r   r   r   
contextlibr   typingr   r   r   rT   r   r   Z_C.libtriton.tritonr   r!   r<   rS   rd   re   r{   r   r   r   r   r   r   r   r   r   r   <module>   s(   	
<
L
%B^

