afe.tvm_converter.custom_convert_maps
=====================================

.. py:module:: afe.tvm_converter.custom_convert_maps

.. autoapi-nested-parse::

   TODO(Joey.Chou):
       Use registration method to register
           * custom_convert_map
           * MergeComposite Pattern _PATTERN_TABLE
           * tvm_func_to_awesome_attributes



Attributes
----------

.. autoapisummary::

   afe.tvm_converter.custom_convert_maps.AVAILABLE_ANNOTATIONS
   afe.tvm_converter.custom_convert_maps.CUSTOM_CONVERT_MAP_DICT
   afe.tvm_converter.custom_convert_maps.TVM_FUNC_TO_AWESOME_ATTRIBUTES_DICT


Classes
-------

.. autoapisummary::

   afe.tvm_converter.custom_convert_maps.GroupConv2DTranspose
   afe.tvm_converter.custom_convert_maps.ConstantMultiplyAdd
   afe.tvm_converter.custom_convert_maps.RenamerWithAttrsChecker


Module Contents
---------------

.. py:class:: GroupConv2DTranspose

   Models: CenterNet
   Author: Alicja Kwasniewska, Joey Chou

   Decompose a nn.conv2d_transpose with groups > 1 into splits, multiple nn.conv2d_transpose,
   and concatenate as below:

                     |
                   split
           --------------------------- ....
           |               | | | | ...
      tuple_get_item       .....
           |
    nn.conv2d_transpose
           |
           |               | | | | ...
           --------------------------- ....
                     |
                   tuple
                     |
                concatenate
                     |


   .. py:attribute:: ONNX_op_name
      :value: 'ConvTranspose'



   .. py:method:: map_impl()
      :staticmethod:



   .. py:method:: tvm_func_to_awesome_attributes(params: afe.tvm_converter.parameters.TVMConverterParams, func: afe._tvm._defines.TVMFunction) -> afe.ir.attributes.AwesomeAttributes
      :staticmethod:


      Functions to extract decomposed group conv2d_transpose Relay attributes and
      translate it into SiMa IR Conv2DTransposeAttrs



.. py:class:: ConstantMultiplyAdd

   Merge multiply with constant to add op.
   Do it for both inputs if possible.


   .. py:method:: tvm_func_to_awesome_attributes(params: afe.tvm_converter.parameters.TVMConverterParams, func: afe._tvm._defines.TVMFunction) -> afe.ir.attributes.AwesomeAttributes
      :staticmethod:


      Functions to extract decomposed constant, mul, add relay attributes from composite
      constant_multiply_add and translate it into SiMa IR ConstantMultiplyAddAttrs.



.. py:class:: RenamerWithAttrsChecker(new_name, checker)



   .. py:attribute:: checker


.. py:data:: AVAILABLE_ANNOTATIONS
   :value: ['int8', 'int16']


.. py:data:: CUSTOM_CONVERT_MAP_DICT

.. py:data:: TVM_FUNC_TO_AWESOME_ATTRIBUTES_DICT