make block-sparse tensors faster

google / TensorNetwork

A library for easy and efficient manipulation of tensor networks.

Apache License 2.0

1.81k stars 356 forks source link

make block-sparse tensors faster #536

Open mganahl opened 4 years ago

mganahl commented 4 years ago

for small rank tensors with small bond dimensions, block-sparse tensor contractions are currently not very fast. We should change this. Possible options:

port code to cython
use dense encoding for small rank, small-D tensors
add block-wise encoding for small-rank, small-D tensors

chaserileyroberts commented 4 years ago

So based on our internal conversation, we're going with adding a blockwise encoding scheme in addition to our current elementwise scheme. This should be better for few legged tensors like MPS systems.