srkreddy1238 commented on issue #4102: URL: https://github.com/apache/tvm/issues/4102#issuecomment-783872081
@wx3000 understand the complexity around control flow ops. But, I am worried about two copies of independent ops across parsers. Understanding the complexity of existing parser and to avoid redundant ops across parsers here are my thoughts. - Version independent ops implementation can move to a common file (tf_op.py) - Common entry point for the parser from_tensorflow with graphdef or concrete_function (tensorflow.py). - Quick inspection of graph can decide the parser version to use (TF1.x / TF2.x). - Parser and corresponding exclusive ops can go to individual parsers (tf_parser_1_x.py & tf_parser_2_x.py) Or the tf_parser_x_x. py can hold only the parser and tf_op.py can hold all operator implementations. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
