According to the design of sequence decoder,
a LodExtend Operator is needed to link the previous state to the current state of RNN/sequence decoder.

It functions like

• inputs
  • source : LoDTensor
  • target, the lod to fit, LoDTensor
• outputs
  • the expanded source that has the same LoD with target

Some real examples:

• source: a 1-level LoDTensor with 2 sequences
  • LoD: [0, 3, 5]
  • data: [0, 1, 2, 3, 4, 5]
• target: a 1-level LoDTensor with 5 sequences
  • LoD: [0, 2, 3, 5, 6, 6, 8]
• output:
  • LoD same with target
  • data: [0, 0, 1, 2, 2, 3, 5, 5]

The function LodExpand in another PR can be used.