This property is used during the code generation to ensure the the actors mapped to a specific component instance are properly executed on the corresponding one. There is no constraint in the model forbidding to set the same ID to several components. Indeed the architecture model could represent 8 x86 cores with 2 CUDA GPUs and 1 shared memory, having the x86 IDs ranging from 0 to 7, the GPU IDs set to 0 and 1, and the memory ID set to 0. However the current code generation will throw errors if several Operator (i.e. x86 or GPU, but not the memory) share the same ID.

This new property causes warnings when unset. We have updated most of the tutorials and our application repository (preesm-apps) to reflect this update. If your workflow ‘Error on Warning’ settings is set to true, this warning will cause a failure.

Allong with this new property, we have fixed the code generation to properly assign the affinity of the main thread (the thread running on the main operator). We have also disabled the affinity settings on Windows and added a new C preprocessor test on PREESM_NO_AFFINITY to disable affinity setting if defined.

In previous versions, the code generation was making use of 2 macros, VERBOSE and LOOP_SIZE. To avoid possible conflicts with application constants, we have prefixed them with PREESM_. This requires manual change in the preesm.h header.

Also, the task reference for the code generation now display properly the effect of these macros.

@PreesmTask(
	id = "my.unique.workflow.task.identifier",
	name = "Example Task",
	inputs = {@Port(name = "PiMM", type = PiGraph.class)},
	outputs = {@Port(name = "PiMM", type = PiGraph.class)}
)

The main objective of this update is to centralize the task configuration, that was split in 3 files previously:

• The Java file was holding the behavior;
• The plugin.xml file was registering the task in the Preesm framework, declaring inputs/outputs;
• The doc/workflow_tasks.tex file was holding the task documentation;

With this new annotation mechanism, the tex file becomes obsolete as we can generate the documentation from the annotations. Also, the plugin.xml file will only need to register the plugin as a Preesm plugin only once for all the tasks it contains.

We have updated our documentation on new task creation. Also, complete task reference is now automatically generated. More complete examples can be found in the code base.

Migration from previous mechanism is straightforward:

1. Locate the plugin.xml file declaring the tasks (the file will exhibit errors after update);
2. For each task declaration:
   • Copy the content of the ‘type’ field (that is the class name), press CTRL+SHIFT+T, paste the name, press Enter;
   • Add the annotation with the content that was in the plugin.xml (id, inputs, outputs);
   • Remove the task declaration from plugin.xml
3. Remove the “org.preesm.workflow.tasks” extension and add “org.preesm.commons.plugin” instead (without any child)

The workflow parameters control the workflow output and behavior as follows:

• ‘Error on Warning’: If set to ‘true’ (default), the workflow will stop with failure when the log prints a warning or severe entry;
• ‘Verbose Level’: Controls the output of the log. Default value is INFO. Possible values range in {ALL, FINEST, FINER, FINE, CONFIG, INFO, WARNING, SEVERE, NONE}, as defined in java.util.logging.Level.

Note that, as the error on warning parameter is set to ‘true’ by default, this may change the behavior of existing workflows throwing warnings.

The procedure is similar to the one presented in our documentation.

Also, following the release 3.0.0, the DFTools repository has been archived. That means the repository is now read-only and we stop its support. All the releases are still available on our archived repository at http://preesm.insa-rennes.fr/repo/update-site-backups/.

Among the changes, we would like to emphasize the integration of the DFTools code base in Preesm, as there has been no reported use of DFTools apart from Preesm source code. This will greatly ease the maintenance of the Workflow, Loggers, Slam and IBSDF/MapperDAG code.

Also, a significative refactoring has been done on the merged code base. All the plugins and packages have been renamed into org.preesm.*; Many classes have been moved to respect the new architecture, as depicted here: https://hackmd.io/cYswTMtXTUew_NmjaRxUeQ.

The release process has also been tuned to minimize the footprint of the generated update site (from 600MB to ~6MB). This enables the deployment of snapshot releases, as soon as the proper platform will be identified, in order to improve the integration testing with Spider development cycle.

This update will mostly impact the developers as they will have to clean their workspace and reimport the Preesm projects, as explained here: https://preesm.github.io/docs/buildpreesm/#import-preesm-projects.

Transitional update site and features have been deployed to the old URL in order to have transparent updates.

Please also note that the Command Line Interface wrapper (https://github.com/preesm/preesm-cli) has been updated to use the updated application identifier.

Releases are available through the usual distribution mechanisms: https://preesm.github.io/get/.