I am a new beginner at MLIR and I have read the toy language tutorial for the MLIR.
I want to build a DNN compilation flow using MLIR. I mean from the DNN model(tf model) to generate the target code (CPU or GPU backend).
Any study suggestions or some learning road map for the compiler beginner?
Thanks a lot.
You could look at IREE and also this thread on the MLIR/TF forum: https://groups.google.com/a/tensorflow.org/g/mlir/c/quu_s4KMzuM/m/Yn1ztQqqBAAJ
For the path below the mid-level dialects, you’ll find enough documentation and information (esp. on dialect conversions) in MLIR proper. For the higher level dialects, you’ll need to build and work with TF and MLIR in TF. Looking at the test cases for various dialect conversion passes will also be helpful.