[Mlir-commits] [mlir] Fix tosa::TransposeOp::inferReturnTypeComponents() (PR #88656)

[Eric Kunze]

eric-k256 wrote:

I'm okay with the quantized tests going away. If it makes sense to convert them to i8 tests, that would be preferable instead of deletion, but we may have sufficient coverage without i8. 
I'd like to start removing quantized tensor support from the dialect. The spec thinks of tensor data as integers and holds zero points where necessary. Quantized scales get turned into RESCALE ops, leaving the quant parameters unused.

https://github.com/llvm/llvm-project/pull/88656