I authored FxpMathOps very early in MLIR’s history and it is now fairly superseded (especially now with the quant.region op in place). I am planning to remove it so we can focus on what we see as the longer term path forward. Is there any reliance on this or objections?
While doing this, I also decided to remove the corresponding Quantizer, as it had some dependencies and was equally experimental (and a new proposal is being worked on).
Please comment on the review if you wish any of this to not be deleted.
We do have a review that has been outstanding for quite some time which adds a dependency on one header in Quantizer from QuantOps (and it was commented that we should move that one file to QuantOps vs adding the dependency). The review has been outstanding for a month. If it gets picked back up, it is easy enough to pull the Statistics header from history and revive it (and that would be useful). https://reviews.llvm.org/D74703 @Vooblin