No --convert-tensor-to-* in mlir-opt? Is there any flag for mlir-opt to convert tensors to std/affine/arith/torch dialects?
Going from the tensor domain to lower levels is more involved than a simple “conversion” because it usually involved going from a “pure value based” computation to something with side-effect and explicit buffers. This is why it has a dedicated process and doc here: Bufferization - MLIR
Oh right, will have a read through. Thanks!