Many functions are missing from caffe.
For example

1. elementwise abs
2. sum of matrix, col sum row sum and total sum. (though can be done by gemv, it might be more convenient to provide a wrapper, multipliers can be allocated at run time and store in a global map).
3. non-in-place scaling, currently it is in place.
4. elementwise sign

A large proportion of the commonly needed function are elementwise operations.
Will we consider bringing in additional dependencies? e.g. thrust