§
    l¬¸ji   ã                   óü   — d dl Z d dlZd dlmZ d dlmZ d dlmZmZmZm	Z	m
Z
mZ dZd„ Zd„ Zded	ej        fd
„Zded	ej        fd„Zdej        j        j        fd„Zd„ Zd„ Zd„ Zd„ Zd„ Zd„ Zd„ Zdej        fd„ZdS )é    N)Únn)Ú_ConvNd)Úcalculate_conv2d_flopsÚcalculate_linearÚcalculate_normÚcalculate_softmaxÚcalculate_zero_opsÚl_prodé   c                 ó>   — | xj         t          ¦   «         z  c_         dS )zHIncrementally add zero operations to the model's total operations count.N)Ú	total_opsr	   ©ÚmÚxÚys      úg/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/thop/vision/basic_hooks.pyÚzero_opsr      s   € à€K„KÕ%Ñ'Ô'Ñ'€K„K€K€Kó    c                 óp   — | j         �.t          j        t          ¦  «                             d¦  «         dS dS )zTCharge nothing for an embedding gather, warning when max_norm makes that incomplete.Nz'max_norm renormalization is not counted)Úmax_normÚloggingÚ	getLoggerÚ__name__Úwarningr   s      r   Úcount_embeddingr      s8   € à„zÐÝÔ�(Ñ#Ô#×+Ò+Ð,UÑVÔVÐVÐVÐVð Ðr   r   r   c           
      óæ   — |d         }| xj         t          t          |j        ¦  «        t          |j        ¦  «        t          | j        j        ¦  «        | j        | j        d¬¦  «        z  c_         dS )úmCalculate and add the number of convolutional operations (FLOPs) for a ConvNd layer to the model's total ops.r   F©Ú
input_sizeÚoutput_sizeÚkernel_sizeÚgroupsÚbiasÚ	transposeN©r   r   ÚlistÚshapeÚweightr"   r#   r   s      r   Úcount_convNdr)       sc   € à	ˆ!Œ€Aà€K„KÕ)Ý˜œ‘=”=Ý˜œ‘M”MÝ˜œœÑ(Ô(ØŒxØŒVØðñ ô ñ €K„K€K€Kr   c           
      óæ   — |d         }| xj         t          t          |j        ¦  «        t          |j        ¦  «        t          | j        j        ¦  «        | j        | j        d¬¦  «        z  c_         dS )r   r   Tr   Nr%   r   s      r   Úcount_convtNdr+   6   sc   € à	ˆ!Œ€Aà€K„KÕ)Ý˜œ‘=”=Ý˜œ‘M”MÝ˜œœÑ(Ô(ØŒxØŒVØðñ ô ñ €K„K€K€Kr   c                 óÆ   — |d         }t          |                     ¦   «         ¦  «        }t          | dd¦  «        st          | dd¦  «        r|dz  }| xj        |z  c_        dS )ziCalculate and add the FLOPs for a batch normalization layer, including elementwise and affine operations.r   ÚaffineFÚelementwise_affineé   N)r   ÚnumelÚgetattrr   )r   r   r   Úflopss       r   Úcount_normalizationr3   D   sg   € ð 	
ˆ!Œ€Aå˜1Ÿ7š7™9œ9Ñ%Ô%€EÝˆq�(˜EÑ"Ô"ð ¥g¨aÐ1EÀuÑ&MÔ&Mð Ø�‰
ˆØ€K„K�5Ñ€K„K€K€Kr   c                 ól   — |d         }| j         s$| xj        |                     ¦   «         z  c_        dS dS )z]Calculate and update the total operation counts for a PReLU layer using input element number.r   N)Útrainingr   r0   r   s      r   Úcount_prelur6   Z   s;   € à	ˆ!Œ€AàŒ:ð !Ø	ˆŒ�q—w’w‘y”yÑ ˆŒˆˆð!ð !r   c                 ó:  — |d         }t          | t          j        ¦  «        rdn| j        }|€|                     ¦   «         dv rdnd}|                     ¦   «         r|                     ¦   «         |         nd}|r|                     ¦   «         |z  nd}| xj        t          ||¦  «        z  c_        t          | t          j        ¦  «        r$| xj        |                     ¦   «         z  c_        dS t          | t          j	        ¦  «        r| xj        |z  c_        dS dS )zWCalculate and update the total operation counts for a Softmax layer in a PyTorch model.r   éýÿÿÿN>   r   r   é   r   )
Ú
isinstancer   Ú	Softmax2dÚdimÚsizer0   r   r   ÚSoftminÚ
LogSoftmax)r   r   r   r<   Ú	nfeaturesÚ
batch_sizes         r   Úcount_softmaxrB   b   s  € à	ˆ!Œ€Aå˜1�bœlÑ+Ô+Ð
6ˆ"ˆ"°´€CØ
€{à—5’5‘7”7˜iÐ'Ð'ˆaˆa¨Qˆð "#§¢¡¤Ð/�—’‘”˜”�¨a€IØ+4Ð;�—’‘”˜iÑ'Ð'¸!€Jà€K„KÕ$ Z°Ñ;Ô;Ñ;€K„KÝ�!•R”ZÑ Ô ð "Ø	ˆŒ�q—w’w‘y”yÑ ˆŒˆˆÝ	�A•r”}Ñ	%Ô	%ð "Ø	ˆŒ�zÑ!ˆŒˆˆð"ð "r   c           	      óö  ‡
‡‡‡‡— | j         }t          |t          ¦  «        r"|ft          | t          j        ¦  «        rdndz  }t          |¦  «        }t          | j        t          ¦  «        r| j        f|z  n| j        }t          | j        t          ¦  «        r| j        f|z  n| j        }d}t          |d         j	        | d…         |j	        | d…         |||¦  «        D ]M\  }}	Š
ŠŠ| j
        r‰ |‰z   fnd|f\  ŠŠ|t          ˆ
ˆˆˆˆfd„t          |	¦  «        D ¦   «         ¦  «        z  }ŒN| xj        t          |d         j	        d| …         ¦  «        |z  |                     ¦   «         z   z  c_        dS )zDCalculate and update the total operation count for an AvgPool layer.r9   r/   r   r   Nc              3   ó”   •K  — | ]B}t          t          |‰z  ‰z
  ‰z   ‰¦  «        t          |‰z  ‰z
  ‰¦  «        z
  d ¦  «        V — ŒCdS )r   N)ÚmaxÚmin)Ú.0ÚiÚkÚlowerÚpÚsÚuppers     €€€€€r   ú	<genexpr>z count_avgpool.<locals>.<genexpr>�   s[   øè è € ÐhÐhÐUV•s�3˜q 1™u q™y¨1™}¨eÑ4Ô4µs¸1¸q¹5À1¹9ÀeÑ7LÔ7LÑLÈaÑPÔPÐhÐhÐhÐhÐhÐhr   )r!   r:   Úintr   Ú	AvgPool3dÚlenÚstrideÚpaddingÚzipr'   Úcount_include_padÚsumÚranger   r
   r0   )r   r   r   ÚkernelÚdimsrR   rS   Úwindowsr=   ÚoutputrI   rJ   rK   rL   rM   s             @@@@@r   Úcount_avgpoolr\   v   sŠ  øøøøø€ àŒ]€FÝ�&�#ÑÔð GØ�¥:¨aµ´Ñ#>Ô#>ÐE˜a˜aÀAÑFˆÝˆv‰;Œ;€DÝ#-¨a¬h½Ñ#<Ô#<ÐJˆaŒhˆ[˜4ÑÐÀ!Ä(€FÝ%/°´	½3Ñ%?Ô%?ÐNˆqŒyˆl˜TÑ!Ð!ÀQÄY€GØ€GÝ!$ Q q¤T¤Z°°°°Ô%7¸¼À$ÀÀÀ¼È&ÐRXÐZaÑ!bÔ!bð ið iÑˆˆf�a˜˜AØ)*Ô)<ÐK˜˜˜D 1™H�~�~À1ÀdÀ)‰ˆˆuØ•3ÐhÐhÐhÐhÐhÐhÐhÐhÕZ_Ð`fÑZgÔZgÐhÑhÔhÑhÔhÑhˆˆØ€K„K•6˜!˜Aœ$œ* V t e VÔ,Ñ-Ô-°Ñ7¸!¿'º'¹)¼)ÑCÑC€K„K€K€Kr   c           	      óº   — | xj         t          d„ t          |d         j        |j        ¦  «        D ¦   «         ¦  «        |                     ¦   «         z   z  c_         dS )zgCalculate and update the total operation counts for an AdaptiveAvgPool layer from the windows it pools.c              3   ó‚   ‡‡K  — | ]8\  ŠŠ‰‰k    r‰n(t          ˆˆfd „t          ‰¦  «        D ¦   «         ¦  «        V — Œ9dS )c              3   óD   •K  — | ]}|d z    ‰z  ‰z   |‰z  ‰z  z
  V — ŒdS )r   N© )rG   ÚjrH   Úos     €€r   rN   z/count_adap_avgpool.<locals>.<genexpr>.<genexpr>�   sB   øè è € Ð UÐ UÀq Q¨¡U 8¨a¡<°1Ñ#4Ð!5¸¸A¹À¹
Ñ!BÐ UÐ UÐ UÐ UÐ UÐ Ur   N)rV   rW   )rG   rH   rb   s    @@r   rN   z%count_adap_avgpool.<locals>.<genexpr>�   so   øøè è € ð 
ð 
á��1ð �a’�ˆAˆA�SÐ UÐ UÐ UÐ UÐ UÍEÐRSÉHÌHÐ UÑ UÔ UÑUÔUð
ð 
ð 
ð 
ð 
ð 
r   r   N)r   r
   rT   r'   r0   r   s      r   Úcount_adap_avgpoolrc   …   se   € ð €K„KÝð 
ð 
å˜A˜aœDœJ¨¬Ñ0Ô0ð
ñ 
ô 
ñ 	
ô 	
ð �'Š'‰)Œ)ñ		ñ€K„K€K€Kr   c                 ó  — dddddddœ                      | j        ¦  «        }|€7t          j        t          ¦  «                             d| j        › d	�¦  «         d}| xj        ||                     ¦   «         z  z  c_        dS )
zMUpdate total operations counter for upsampling layers based on the mode used.r   é   é   i  é   )Únearestznearest-exactÚlinearÚbilinearÚbicubicÚ	trilinearNzmode z* is not implemented yet, take it a zero op)ÚgetÚmoder   r   r   r   r   Únelement)r   r   r   Úops_per_elements       r   Úcount_upsamplerq   ˜   s�   € ð ØØØØØðð ÷ 
‚cˆ!Œ&�k„kð ð ÐÝÔ�(Ñ#Ô#×+Ò+Ð,f°A´FÐ,fÐ,fÐ,fÑgÔgÐgØˆØ€K„K�? Q§Z¢Z¡\¤\Ñ1Ñ1€K„K€K€Kr   c                 óx   — | j         }|                     ¦   «         }| xj        t          ||¦  «        z  c_        dS )zWCounts total operations for nn.Linear layers using input and output element dimensions.N)Úin_featuresr0   r   r   )r   r   r   Ú	total_mulÚnum_elementss        r   Úcount_linearrv   ©   s7   € à”€Ið —7’7‘9”9€Là€K„KÕ# I¨|Ñ<Ô<Ñ<€K„K€K€Kr   c                 ó†   — | xj         t          | j        dz   | j        z  |                     ¦   «         ¦  «        z  c_         dS )z0Count both contractions of each bilinear output.r   N)r   r   Úin1_featuresÚin2_featuresr0   r   s      r   Úcount_bilinearrz   ³   s7   € à€K„KÕ# Q¤^°aÑ%7¸1¼>Ñ$IÈ1Ï7Ê7É9Ì9ÑUÔUÑU€K„K€K€Kr   c                 óÀ  — t          |t          ¦  «        r|n|df\  }}t          | dd¦  «        }|s|                     ¦   «         dk    r|j        d         n|j        d         }|                     ¦   «         dk    rdn|j        |rdnd         }| j        du| j        z   }|�|j        d         }	n?|d         }
|s|
                     ¦   «         dk    r|
j        d         n|
j        d         |z   }	| j        d|z  | j        z  |	|z
  | j        | j	        z   z  z   z  }d|z  |	z  | j        z  }t          || j        z  |z  |	¦  «        }| xj        |||z   z  |z   z  c_        dS )	a  Count projections, attention matrix products, and softmax.

    MultiheadAttention calls its projection weights functionally, so none of this work reaches a child-module hook.
    Returned weights reveal appended key positions; calls that suppress them fall back to the key input.
    NÚbatch_firstFr/   éþÿÿÿr   r   éÿÿÿÿ)r:   Útupler1   r<   r'   Úbias_kÚadd_zero_attnÚ	embed_dimÚkdimÚvdimr   Ú	num_headsr   )r   r   r   ÚoutÚweightsr|   Útgt_lenrA   ÚappendedÚattendedÚkeyÚprojectionsÚ	attentionÚsoftmaxs                 r   Úcount_multihead_attentionr�   ¸   sq  € õ # 1¥eÑ,Ô,Ð;�1�1°1°d°)�L€CˆÝ˜!˜]¨EÑ2Ô2€KØ*ÐN¨c¯gªg©i¬i¸1ªn¨nˆcŒi˜ŒmˆmÀ#Ä)ÈAÄ,€GØ—g’g‘i”i 1’n�n��¨#¬)¸Ð4K°A°AÈ!Ô*L€JØ” Ð$¨¬Ñ7€HØÐØ”= Ô$ˆˆà�ŒdˆØ%0ÐT°C·G²G±I´IÀ²N°N�C”I˜b”M�MÈÌ	ÐRSÌÐX`Ñ`ˆð ”+  W¡¨q¬{Ñ!:¸hÈÑ>QÐVWÔV\Ð_`Ô_eÑVeÑ=fÑ!fÑg€KØ�G‘˜hÑ&¨¬Ñ4€IÝ 
¨Q¬[Ñ 8¸7Ñ BÀHÑMÔM€GØ€K„K�: ¨yÑ!8Ñ9¸GÑCÑC€K„K€K€Kr   )r   Útorchr   Útorch.nn.modules.convr   Úthop.vision.calc_funcr   r   r   r   r	   r
   Úmultiply_addsr   r   ÚTensorr)   r+   ÚmodulesÚ	batchnormÚ
_BatchNormr3   r6   rB   r\   rc   rq   rv   rz   ÚMultiheadAttentionr�   r`   r   r   ú<module>r™      s¾  ðð €€€à €€€Ø Ð Ð Ð Ð Ð Ø )Ð )Ð )Ð )Ð )Ð )ðð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð €ð(ð (ð (ð
Wð Wð Wð�Gð  5¤<ð ð ð ð ð,�Wð  E¤Lð ð ð ð ð	˜2œ:Ô/Ô:ð 	ð 	ð 	ð 	ð,!ð !ð !ð"ð "ð "ð(Dð Dð Dðð ð ð&2ð 2ð 2ð"=ð =ð =ðVð Vð Vð
D Ô!6ð Dð Dð Dð Dð Dð Dr   