Vh d dl mZ d dlmZmZ d dlZd dlZd dlZd dlm Z de
de e
fdZ dej j fd Zd
e
deee e f fd Zd Zd
ZddddefdZed ZdefdZdefdZdefdZy) )contextmanager)AnycastN)Timerfilenamereturnc d}d}d }g }t | 5 }|j j | }t | D ] \ }}|dk( r |j | } | dk( r#|d | }
||dz
j d }|
j d D cg c] } | t
| d } } |j dj | d d d |S c c} w # 1 sw Y |S xY w) Nzzr T)keepends ) openreadsplit enumeratefind
splitlineslenappendjoin)
r BEGINENDpfxgraphsf
split_strsi split_strend_locsxliness
K/home/dcms/DCMS/lib/python3.12/site-packages/torch/utils/jit/log_extract.py
extract_irr$ s E
C
C
F
h *1VVX^^E*
%j 1 *LAyAv nnS)G"}(7#A QU#..04C+,<<<+FGaQs3xy\GEG MM"''%. ) * * M H * Ms B C C)$C C C'inp_typec | j }| j }| j }| j }|J |J |J |J t j
|||| S )N)sizestridedevicedtype)sizesstridesr) r* torch
empty_strided)r% r' r( r) r* s r# make_tensor_from_typer/ s{ >> D
F
__
F NN E De TT irc | t j j | d }|j g }|j D ] }t |j
t j j r&|j t j dd \t |j
t j j r&|j t j dd t |j
t j j rNt t j j |j
}|j t | 4t |j
t j j r*|j t j dd dk( t# d|j
t j j% d| }t j j' |j( ||fS ) NT)parse_tensor_constantsg?d r r z,A default value is not implemented for type forward)r- _Cparse_irmakeMultiOutputIntoTupleinputs
isinstancetype FloatTyper randomuniformIntTyperandint
TensorTyper r/ BoolTypeNotImplementedError_create_function_from_graph!_jit_pass_erase_shape_informationgraph)r1 rF r9 inp
tensorTypefuncs r# load_graph_and_inputsrJ ) sy HH b >E ""$
F||~ c chhj%(("4"4 5 MM&..S1 2
EHH$4$4
5 MM&..C0 1
EHH$7$7
8 ehh 1 1388:>J MM/
; <
EHH$5$5
6 MM&..A.!3 4%(TUXU]U]U_T`&ab b c 88 / / 5 AD HH..tzz: &>r0 c ` t d| |d }|j }|j dz S )Nz fn(*inputs))fnr9 )stmtglobals )r blocked_autorangemedian)rL r9 test_runsttimess r# time_cudarU > s0
=*GHA
!E <<$ r0 c t j }t | D ] } | | t j }||z
|z dz S )NrO )time perf_counterrange)rL r9 rR r _es r# time_cpur\ C sM A
9
F A
EY %%r0
) warmup_runsrR c t | \ }}t | D ] } || d }|D ]7 }t |t j s |j
j dk( } n |J |rt ||| }|S t ||| }|S )Ncpu) rJ rY r: r- Tensorr) r; r\ rU ) r1 r9 r_ rR rF rZ is_cpuinputouts r# run_testrf J s $R(HE1
;
v F eU\\ *\\&&%/F 06(5&)
,C J =FeVU^<_C Jr0 c / K t j j d } d t j j | y # t j j | w xY ww)NF)r- r6 _get_graph_executor_optimize)argskwargs old_optimizes r# no_fuserrl Y sF 8888?L<
--l;--l;s A+A A+!A((A+c Z t 5 t | | cd d d S # 1 sw Y y xY w)N)rl rf r1 r9 s r# run_baseline_no_fusionro a s& $F#$ $ $s !*c |rdgndg}t j j | }t j j d 5 t | | cd d d t j j | S # 1 sw Y nxY w t j j | y # t j j w xY w)N)DYNAMICr] )STATICr] fuser1)r- jitset_fusion_strategyfuserrf )r1 r9 dynamicstrat old_strats r# run_nncrz f s 1%, !>2BII11%8
YY__X
& (B' ( ( %%i0 ( ( ( %%i0 %%i0s$ AB* A= B* =BB* *!C c t j j d 5 t | | cd d d S # 1 sw Y y xY w)Nfuser2)r- rt rv rf rn s r# run_nvfuserr} o s1 " $F#$ $ $s 6?)
contextlibr typingr r r= r- rW torch.utils.benchmarkr strlistr$ r6 rA r/ tuplerJ rU r\ floatrf rl ro rz r} r0 r# r s %
' c * UEHH$7$7 Uc eCcN&; *
& )+b
U
< <$% $
1E 1$u $r0