v3.16.0
·
1352 commits
to develop
since this release
Release version 3.16.0
2019.09.06
New Feature
- Codegen:
- clusters can now be printed by using CodegenClusterTask
- Apollo optimizations are now supported in Preesm and SPiDER
- To activate it, Apollo flag must be set to true in the CodeGen workflow task
- CMakeLists.txt of the application must be modified
- Scheduling:
- Mapping/scheduling task (pisdf-mapper.list) can now perform energy-aware mapping/scheduling (It will increase scheduling/mapping task time, as it tests several mappings)
- The information is taken from energy tab in scenario and the optimization will consider the performance objective to search for the best PEs configuration
- EnergyAwareness parameter will enable this optimization
- EnergyAwarenessFirstConfig selects the starting point: First (as less PEs as possible), Max (all PEs), Middle (half amount of PEs of each type) and Random (random starting point)
- EnergyAwarenessSearchType selects the type of searching done: Thorough (changes PEs number one by one) and halves (Divides in halves the remaining available PEs and goes up/down depending if the FPS reached are below/above the objective)
Changes
- PiSDF:
- PiGraph can be declared as a cluster. Hierarchy of cluster may be ignore by SR and DAG transformation;
- Add operation to get the unique data port of a SrdagActor (init/end);
- Change PiSDF task categories in task reference;
- Clustering:
- cluster are now mappable into component by retrieving common possible mapping of included actors
- make code clearer for ClusteringBuilder
- refactored the way of getting clusterizable couples
- clusters now have timings
- Schedule:
- Readapt the way of exhibiting data parallelism and print expression
- Add communication nodes
- Refactor:
- Fix major and critical sonar issues;