Lowering dynamic shape of tensor.pack to vector dialect.
Disable expand_shape and collapse_shape lowering to vector dialect. Because those two operation is actually a tensor view operation, memref dialect has a pass to optimize it as a view operation.
Tracking issue300.
memref
dialect has a pass to optimize it as a view operation.Depends on PR100.
We will optimize mask operation as
vector.maskload
andvector.maskstore
in for loop in next PR.