We should allow the user to use memref.collapse_shape, memref.expand_shape, tensor.collapse_shape, tensor.expand_shape.
We should also make more specific operations squeeze and unsqueeze which work like they do in numpy and can be used to add/remove dimensions of size 1.
The syntax and logic for collapse_shape and expand_shape might be non-trivial, since you have to deal with static vs dynamic dimensions.
We should allow the user to use
memref.collapse_shape,memref.expand_shape,tensor.collapse_shape,tensor.expand_shape.We should also make more specific operations
squeezeandunsqueezewhich work like they do in numpy and can be used to add/remove dimensions of size 1.The syntax and logic for
collapse_shapeandexpand_shapemight be non-trivial, since you have to deal with static vs dynamic dimensions.