pytaco.tensor.pack
- tensor.pack()
Packs a tensor.
When performing inserts, taco will place new values in a buffer to avoid recreating sparse structures on every insert. As a result, data is not immediately updated to the tensor. Taco will automatically do this when needed (for example before printing, computing, writing to a file, iterating etc..) but we expose this mechanism to allow the user to explicitly pack a tensor themselves.
It is advised to explicitly pack tensors before taking timing measurements since this is not considered to be part of the kernel computation.
Examples
>>> import pytaco as pt >>> a = pt.tensor([2, 2]) >>> a.insert([0, 0], 10) # 10 is in a buffer >>> a.pack() # 10 is now in matrix structure.