since it turns out it doesn't help with speed to do it that way. (Cython code shouldn't need much modification to support having it in different parts of memory, though the loss of cache locality might hurt. Ideally it'd support either stacked or unstacked features, if that's possible without too much code complexity.)
since it turns out it doesn't help with speed to do it that way. (Cython code shouldn't need much modification to support having it in different parts of memory, though the loss of cache locality might hurt. Ideally it'd support either stacked or unstacked features, if that's possible without too much code complexity.)