Tensor reuse if there was a full tensor and a partial tensor

[Richard1688Sun]

Extending off #65

What would happen within a subgraph, if 1 operation loads a full T0 but another operation only requires a strip of T0? Would this be intra-subgraph reuse or will we need to double count the T0 strip?

Does the situation change if T0 was fully retained from the previous subgraph? What decides the tensor renaming when fetching from slow memory?

Thanks for all the clarification!

[papp-pal-andras]

Good question, I'm also interested in this one!