TensorFlow's Conv2D dot product implementation?

I would like to run a Python code of NN which is using Conv2D, but i'd like to change the dot product inside the conv2D operation in C++. To be more specific about the background, conv operator should contain product between the filter kernel and a window to get a filtered window (for each window within image). I've already searched in Tensorflow's conv_ops.cc documentation, but didn't find any dot product (or any product). I'd like to be referenced to the place where the actual dot product happens. Thanks!