Create an Ingress Module #29
Description
As detailed in #16, our goal is to have a common infrastructure for all examples, pipelines and external usage. The first stage is to help bring other formats into MLIR.
As a first step, we can common up the torch-mlir usage into a module that gets used by the existing examples. This module should live in lighthouse/ingress (as per #23) and used by any example that currently calls it directly. The idea is to hide the complexity behind a simpler API.
The torch-mlir project can convert graphs from PyTorch, Tensorflow and ONNX formats into MLIR of various forms. This module needs to be flexible enough to use that API for different inputs and outputs. The current usage is simple enough for this module to be a simple wrapper.
However, #3 shows that the intent is to make the torch-mlir more robust and powerful, but currently that usage is frail (needs to build and copy files across), so we can start with the simple cases we have today and when torch-mlir improves its usability by third party projects, we only need to change the ingress module.
Key deliverable: Abstract away the repeated usage in examples into a single module and change those examples to use this module.
Further refinement of the module, its API and torch-mlir usage should be done in future issues.