Skip to content

WIP: compute_layer resource-reuse-API for parallel inputs

Javier Duarte requested to merge ejk/resource-reuse-api into master

Created by: ejk43

First draft of the compute_layer "resource reuse API" for parallel inputs only.

Still to do:

  1. Add option for "serial" inputs
  2. Code generation from keras with new struct-based API

Thought I'd open the PR so we can look at the potential edits.

Works pretty well for the "simple-demo" example project (just change the unroll_factor to achieve a lower DSP usage)

Feel free to suggest any edits here! Especially API related.

Merge request reports