While importing a model into TVM, we noticed this linear transform not being simplified. Trying various passes didn’t help either, even on a mock graph reproducing this conv2d-add-add pattern.
The expected simplification is to have the adds collapsed into 1 add.
Any suggestions?