[XLA related] Sort out my thought and notes about XLA related

This post could be a little bit unstructured because it's for my reference in notes.
I recently found that there are several slides in SlideShare which contain very good information and source code analysis/study about XLA related as follows:



TensorFlow local Python XLA client
https://www.slideshare.net/ssuser479fa3/tensorflow-local-python-xla-client

TensorFlow XLA RPC
https://www.slideshare.net/ssuser479fa3/tensorflow-xla-rpc

Bridge TensorFlow to run on Intel nGraph
https://www.slideshare.net/ssuser479fa3/bridge-tensorflow-to-run-on-intel-ngraph-backends-v05

Tensorflow dynamically loadable SXLA plugin
https://www.slideshare.net/ssuser479fa3/tensorflow-dynamically-loadable-sxla-plugin-98370414

Here is ONNX-XLA project in Github that is very interesting:
Using the RPC of TensorFlow XLA, the ONNX model is converted to TensorFlow XLA computation, and the converted computation is sent to the server by the RPC, and it is processed.
https://github.com/varunjain99/onnx-xla.git

About Onnx Compiler & Optimization Tools for inferencing:
1. Glow : Facebook Glow is the default interpreter CPU, GPU, and optional. Glow doesn't must use Onnx. It can be through Pytorch => Caffe2 => Glow => inference engine after optimization (AOT or JIT)
2. NNVM + TVM
3. ONNC
4. nGraph

Which is best/suitable?