Workflow Tasks Reference

This page has been generated. Last update : 2020.10.01; for Preesm version 3.21.0

This page references the available workflow tasks.

Graph Transformation

Data-parallel Transformation

Identifier: fi.abo.preesm.dataparallel.DataParallel
Implementing Class: fi.abo.preesm.dataparallel.DataParallel
Short description: Detect whether an SDF graph is data-parallel and provide its data-parallel equivalent Single-Rate SDF and its re-timing information.

Inputs

SDF (of SDFGraph)

Outputs

CySDF (of SDFGraph)
Info (of RetimingInfo)

Description

An SDF graph is data-parallel when for each actor of the SDF graph, all of its instances can be executed at the same time. For instance, all strictly acyclic SDF graphs are data-parallel. This task increases the scope of this analysis to generic SDF graphs.

The task analyses an input SDF graph and reports if it is data-parallel by analysing each strongly connected component of the graph. If a strongly connected component requires a re-timing transformation for it to be data-parallel, then the transformation is carried out such that the corresponding strongly connected component in the output single-rate SDF is data-parallel. The re-timing transformation modifies the delays in the original SDF graph such that the final single-rate SDF output is data-parallel.

However, if a strongly connected component of the SDF is not data-parallel, then the plugin reports the actors of this strongly connected component. In this case, the strongly connected component at the output single-rate SDF graph is same as that of the single-rate transformation on the original SDF.

The data-structure INFO describes mapping of delays in original FIFO to delays in the transformed SDF. This non-trivial initialization of delays in FIFOs is represented using SDF-like graphs where FIFO initialization function is represented as an actor. The data-structure INFO is provided for sake of completion and is not being used by other plugins. The data-structure INFO can change in future based on the design of the initialization of the FIFOs.

Parameters

None.

Documented Errors

Message	Explanation
DAGComputationBug	There is a bug in implementation due to incorrect assumption. Report the bug by opening an issue and attaching the graph that caused it.

Clustering

Identifier: org.ietr.preesm.Clustering
Implementing Class: org.preesm.algorithm.clustering.OldClustering
Short description: Undocumented

Inputs

SDF (of SDFGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

SDF (of SDFGraph)

Description

Workflow task responsible for clustering hierarchical actors.

Parameters

None.

Cluster Partitioner

Identifier: cluster-partitioner
Implementing Class: org.preesm.algorithm.clustering.partitioner.ClusterPartitionerTask
Short description: Undocumented

Inputs

PiMM (of PiGraph) : Input PiSDF graph
scenario (of Scenario) : Scenario

Outputs

PiMM (of PiGraph) : Output PiSDF graph

Description

Undocumented

Parameters

Number of PEs in clusters

The number of PEs in compute clusters. This information is used to balance actor firings between coarse and fine-grained levels.

Value	Effect
Fixed:=n	Where $n\in \mathbb{N}^*$.

SCAPE transformation

Identifier: scape.task.identifier
Implementing Class: org.preesm.algorithm.clustering.scape.ClusteringScapeTask
Short description: This task cluster actors in order to match parallelism the target architecture.

Inputs

scenario (of Scenario)

Outputs

PiMM (of PiGraph)
scenario (of Scenario)
cMem (of ClusterMemory)

Description

SCAPE stand for Scaling up of Cluster of Actor on Processing Element. The task happen upstream the standard resource allocation process and aim to simplify the dataflow graph structure while preserving the parallelism that correspond the one of the target architecture. The process is based on agglomerative clustering method which revolves around identifying actors with comparable attributes and behaviors within a dataflow graph, ultimately grouping them to form subgraphs with shared features. Subsequently, a sub-code is generated for each subgraph, encapsulating the behavior of the grouped actors. This sub-code replaces the hierarchical actor, transforming it into a standard actor defined by the newly created code.

Parameters

Level number

This parameter works with SCAPE mode 0 and 2. It corresponds to the hierarchical level on which particular clusters (URC, SRV, LOOP, SEQ) are identified. Higher levels are left as such. Lower levels are coarsely clustered.

Value	Effect
Integer	Hierarchical level are coarsely clustered untill this hierarchical level value.

SCAPE mode

The SCAPE method admit three method, each one is the extension of the former one.

The original SCAPE method only takes into account two patterns for clustering, which are Actor Unique RepetitionCount (URC) and Single Repetition Vector (SRV). This method is parameterised, meaning it accepts a number as a parameter that corresponds to the number of hierarchical levels to be coarsely clustered. The clustering occurs at this specified level, which implies that there can be as many clustering configurations as there are hierarchical levels in the input graph. The goal is two reduce the data parallelism to the target.
The first extension introduces two additional patterns to the original SCAPE method: LOOP for cycles and SEQ for sequential parts. This extended method retains its parameterised nature, with the aim of reducing data parallelism and enhancing pipeline parallelism to align with the intended target.
The second extension of the SCAPE method takes into account the hierarchical context when choosing the clustering approach. This allows for the adaptation of parallelism or the coarsening of identified hierarchical levels based on the context.

Value	Effect
0	Reduce excessive data parallelism to match the parallelism of the target on the specified level.
1	Reduce excessive data parallelism and add parallelism to sequential part to match the parallelism of the target on the specified level.
2	Consider hierarchical context and reduce excessive data parallelism and add parallelism to sequential part to match the parallelism of the target on each admissible level.

Stack size

This parameter defines the stack size, which is set to a default value of 1048576, equivalent to 1MB. The algorithm generates a C file for each cluster and allocates internal buffers statically using arrays. Once the stack size limit is reached, buffers are then allocated dynamically using malloc and freed using free.

Value	Effect
Integer	Cluster-internal buffers are allocated statically up to this value, then dynamically.

Static PiMM to IBSDF - Deprecated

Identifier: org.ietr.preesm.experiment.pimm2sdf.StaticPiMM2SDFTask
Implementing Class: org.preesm.algorithm.pisdf.pimm2sdf.StaticPiMM2SDFTask
Short description: Transforms a static PiSDF Graph into an equivalent IBSDF graph.

Inputs

PiMM (of PiGraph)
scenario (of Scenario)

Outputs

SDF (of SDFGraph)

Description

In Preesm, since version 2.0.0, the Parameterized and Interfaced SDF (PiSDF) model of computation is used as the frontend model in the graphical editor of dataflow graphs. This model makes it possible to design dynamically reconfigurable dataflow graphs where the value of parameters, and production/consumption rates depending on them, might change during the execution of the application. In former versions, the Interface Based SDF (IBSDF) model of computation was used as the front end model for application design. Contrary to the PiSDF, the IBSDF is a static model of computation where production and consumption rates of actors is fixed at compile-time.

The purpose of this workflow task is to transform a static PiSDF graph into an equivalent IBSDF graph. A static PiSDF graph is a PiSDF graph where dynamic reconfiguration features of the PiSDF model of computation are not used.

Parameters

None.

Single-Rate Transformation - Deprecated

Identifier: org.ietr.preesm.experiment.pimm2srdag.StaticPiMM2SrDAGTask
Implementing Class: org.preesm.algorithm.pisdf.pimm2srdag.StaticPiMM2SrDAGTask
Short description: Transforms an SDF graph into an equivalent single-rate SDF graph.

Inputs

PiMM (of PiGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

DAG (of DirectedAcyclicGraph)
PiMM (of PiGraph)

Description

Parameters

ExplodeImploreSuppr

(Deprecated: use at your own risks)

This parameter makes it possible to remove most of the explode and implode actors that are inserted in the graph during the single-rate transformation. The resulting SDF graph is an ill-constructed graph where a single data input/output port of an actor may be connected to several First-In, First-Out queues (Fifos).

Value	Effect
false	(default) The suppression of explode/implode special actors is not activated.
true	The suppression of explode/implode special actors is activated.

Documented Errors

Message	Explanation
Graph not valid, not schedulable	Single-rate transformation of the SDF graph was aborted because the top level was not consistent, or it was consistent but did not contained enough delays — i.e. initial data tokens — to make it schedulable.

Hierarchy Flattening - Deprecated

Identifier: org.ietr.preesm.plugin.transforms.flathierarchy
Implementing Class: org.preesm.algorithm.transforms.HierarchyFlattening
Short description: Transforms a hierarchical IBSDF graph into an equivalent SDF graph.

Inputs

SDF (of SDFGraph)

Outputs

SDF (of SDFGraph)

Description

The purpose of this workflow task is to flatten several levels of the hierarchy of an IBSDF graph and produce an equivalent SDF graph. A hierarchical IBSDF graph is a graph where the internal behavior of some actors is described using another IBSDF subgraph instead of a C header file. When applying this transformation, hierarchical IBSDF actors of the first n levels of hierarchy are replaced with the actors of the IBSDF subgraph with which these hierarchical actors are associated.

Parameters

None.

Documented Errors

Message	Explanation
Inconsistent Hierarchy, graph can’t be flattened	Flattening of the IBSDF graph was aborted because one of the graph composing the application, at the top level or deeper in the hierarchy, was not consistent.

PiSDF Flattener

Identifier: org.ietr.preesm.pimm.algorithm.pimm2flat.StaticPiMM2FlatPiMMTask
Implementing Class: org.preesm.model.pisdf.statictools.PiSDFFlattenerTask
Short description: Undocumented

Inputs

PiMM (of PiGraph)

Outputs

PiMM (of PiGraph)

Description

Undocumented

Parameters

Perform optimizations

Undocumented

Value	Effect
true / false	If true, tries to remove redundant special actors and self loops on delays.

PiSDF Single-Rate Transformation

Identifier: pisdf-srdag
Implementing Class: org.preesm.model.pisdf.statictools.PiSDFToSingleRateTask
Short description: Undocumented

Inputs

PiMM (of PiGraph)

Outputs

PiMM (of PiGraph)

Description

Undocumented

Parameters

Consistency_Method

Undocumented

Value	Effect
LCM
Topology

PiSDF Transform Perfect Fit Delay To End Init

Identifier: org.ietr.preesm.pisdf.transformperfectfitdelay
Implementing Class: org.preesm.model.pisdf.statictools.PiSDFTransformPerfectFitDelayToEndInitTask
Short description: Undocumented

Inputs

PiMM (of PiGraph)

Outputs

PiMM (of PiGraph)

Description

Undocumented

Parameters

None.

Node Partitioner

Identifier: node.partitioner.task.identifier
Implementing Class: org.preesm.algorithm.node.partitioner.NodePartitionerTask
Short description: Undocumented

Inputs

scenario (of Scenario)

Outputs

None.

Description

This is part of the SimSDP multinode multicore HPC resource allocation. The purpose of this task partition graph into subgraphs, each associated to a multcicore node architecture. This generate subgraphs, topgraph, node architectures and scenarios.

Parameters

archi path

Path of the CSV multinode architecture from the Archi/ folder.

Value	Effect
path/in/proj	Path within the Preesm project containing the workflow where the task is instantiated. E.g: SimSDP_node.csv.

Partitioning mode

Partitioning method to build subgraphs.

Value	Effect
equivalentTimed	The original SimSDP partitioning method. The method estimate the workload per subgraph to reach balanced-worload partitioning.
random	The method associate an random workload per subgraph to reach random-worload partitioning.
manual	The method consider the original subgraphs from the given graph.

Graph Exporters

SDF Exporter - Deprecated

Identifier: org.ietr.preesm.plugin.exportXml.sdf4jgml
Implementing Class: org.preesm.algorithm.io.xml.SDFExporter
Short description: Create a new *.graphml file containing the exported SDF graph.

Inputs

SDF (of SDFGraph)

Outputs

None.

Description

The purpose of this task is to create a new *.graphml file containing where the exported IBSDF graph will be written. The exported graph can then be visualized and exported using the former Preesm graph editor for IBSDF graph, which was replaced with the PiSDF graph editor since version 2.0.0. This task is generally used to export intermediary graphs generated at different step of a workflow execution. For example, to visualize the SDF graph resulting from the flattening of an IBSDF graph, or to understand the parallelism that was exposed by the single-rate transformation.

Parameters

path

Path of the directory within which the exported *.graphml file will be created. If the specified directory does not exist, it will be created.

Value	Effect
path/in/proj	Path within the Preesm project containing the workflow where the ”SDF Exporter” task is instantiated. Even if the workflow of a Preesm project A is executed with a scenario from a different project B, the *.graphml file will be generated within the specified directory of project A.

Exported SDF graphs will be named automatically, usually using the same name as the original SDF graph processed by the workflow. If a graph with this name already exists in the given path, it will be overwritten.

Example: Algo/generated/singlerate
	path/in/proj/name.graphml	Path within the Preesm project containing the workflow where the ”SDF Exporter” task is instantiated. Even if the workflow of a Preesm project A is executed with a scenario from a different project B, the *.graphml file will be generated within the specified directory of project A.

Exported SDF graph will be named using the string with the graphml extension at the end of the given path. If a graph with this name already exists in the given path, it will be overwritten.

Example: Algo/generated/singlerate/myexport.graphml

Documented Errors

Message	Explanation
Path is not a valid path for export.	The value set for parameter path is not a valid path in the project.

DAG Exporter - Deprecated

Identifier: org.ietr.preesm.mapper.exporter.DAGExportTransform
Implementing Class: org.preesm.algorithm.mapper.exporter.DAGExportTransform
Short description: Create a new *.graphml file containing the exported Directed Acyclic Graph (DAG).

Inputs

DAG (of DirectedAcyclicGraph)

Outputs

None.

Description

The purpose of this task is to create a new *.graphml file containing where the exported DAG will be written. The exported graph can then be visualized and exported using the former Preesm graph editor for IBSDF graph, which was replaced with the PiSDF graph editor since version 2.0.0. This task is generally used to export intermediary graphs generated by the mapping and scheduling tasks of the workflow.

Parameters

path

Path of the directory within which the exported *.graphml file will be created. If the specified directory does not exist, it will be created.

Value	Effect
path/in/proj	Path within the Preesm project containing the workflow where the ”SDF Exporter” task is instantiated. Even if the workflow of a Preesm project A is executed with a scenario from a different project B, the .graphml file will be generated within the specified directory of project A. Exported SDF graphs will be named automatically, usually using the same name as the original SDF graph processed by the workflow. If a graph with this name already exists in the given path, it will be overwritten. Example: Algo/generated/singlerate*
path/in/proj/name.graphml	Path within the Preesm project containing the workflow where the ”SDF Exporter” task is instantiated. Even if the workflow of a Preesm project A is executed with a scenario from a different project B, the .graphml file will be generated within the specified directory of project A. Exported SDF graph will be named using the string with the graphml extension at the end of the given path. If a graph with this name already exists in the given path, it will be overwritten. Example: Algo/generated/singlerate/myexport.graphml*

Value

Effect

path/in/proj

Path within the Preesm project containing the workflow where the ”SDF Exporter” task is instantiated. Even if the workflow of a Preesm project A is executed with a scenario from a different project B, the *.graphml file will be generated within the specified directory of project A. Exported SDF graphs will be named automatically, usually using the same name as the original SDF graph processed by the workflow. If a graph with this name already exists in the given path, it will be overwritten. Example: Algo/generated/singlerate

path/in/proj/name.graphml

Path within the Preesm project containing the workflow where the ”SDF Exporter” task is instantiated. Even if the workflow of a Preesm project A is executed with a scenario from a different project B, the *.graphml file will be generated within the specified directory of project A. Exported SDF graph will be named using the string with the graphml extension at the end of the given path. If a graph with this name already exists in the given path, it will be overwritten. Example: Algo/generated/singlerate/myexport.graphml

Documented Errors

Message	Explanation
Path <given path> is not a valid path for export. <reason>	The value set for parameter path is not a valid path in the project.

Parameters exporter

Identifier: pisdf-export.parameters
Implementing Class: org.preesm.model.pisdf.serialize.ParametersExporterTask
Short description: Undocumented

Inputs

PiMM (of PiGraph)

Outputs

None.

Description

Export parameters of the graph as C header with define. Exports only static parameters. Name of file is: _preesm_params.h

Parameters

path

Undocumented

Value	Effect
/Code	default path, relative to the project

PiSDF Exporter

Identifier: pisdf-export
Implementing Class: org.preesm.model.pisdf.serialize.PiSDFExporterTask
Short description: Undocumented

Inputs

PiMM (of PiGraph)

Outputs

None.

Description

Undocumented

Parameters

path

Undocumented

Value	Effect
/Algo/generated/pisdf/	default path

hierarchical

Undocumented

Value	Effect
true/false	Export the whole hierarchy (default: true). When set to true, will export all the hierarchy in the folder given by ‘path’, replacing refinement paths. Note: exporting hierarchical graph with this option set to false can cause the the consistency check fail if the children graphs do not exist.

Schedulers

External Scheduling from DAG

Identifier: org.ietr.preesm.plugin.mapper.external
Implementing Class: org.preesm.algorithm.mapper.ExternalMappingFromDAG
Short description: Undocumented

Inputs

DAG (of DirectedAcyclicGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

This class imports schedule expressed in dedicated json format. It is experimental and limited to flat PiMM and a few architectures (regular x86 and Odroid). See package org.preesm.algorithm.mapper.schedule for the json format.

Parameters

SCHEDULE_FILE

Undocumented

Value	Effect
/schedule.json	default value

External Scheduling from PiSDF

Identifier: pisdf-mapper.external
Implementing Class: org.preesm.algorithm.mapper.ExternalMappingFromPiMM
Short description: Undocumented

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

SCHEDULE_FILE

Undocumented

Value	Effect
/schedule.json	default value

List Scheduling from SDF - Deprecated

Identifier: org.ietr.preesm.plugin.mapper.fast
Implementing Class: org.preesm.algorithm.mapper.FASTMapping
Short description: Undocumented

Inputs

SDF (of SDFGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

displaySolutions

Undocumented

Value	Effect
false	Undocumented

fastTime

Undocumented

Value	Effect
100	Undocumented

fastLocalSearchTime

Undocumented

Value	Effect
10	Undocumented

Fast Scheduling from DAG - Deprecated

Identifier: org.ietr.preesm.plugin.mapper.fastdag
Implementing Class: org.preesm.algorithm.mapper.FASTMappingFromDAG
Short description: Undocumented

Inputs

DAG (of DirectedAcyclicGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

displaySolutions

Undocumented

Value	Effect
false	Undocumented

fastTime

Undocumented

Value	Effect
100	Undocumented

fastLocalSearchTime

Undocumented

Value	Effect
10	Undocumented

Fast Scheduling from PiSDF

Identifier: pisdf-mapper.fast
Implementing Class: org.preesm.algorithm.mapper.FASTMappingFromPiMM
Short description: Undocumented

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

displaySolutions

Undocumented

Value	Effect
false	Undocumented

fastTime

Undocumented

Value	Effect
100	Undocumented

fastLocalSearchTime

Undocumented

Value	Effect
10	Undocumented

List Scheduling from SDF - Deprecated

Identifier: org.ietr.preesm.plugin.mapper.listscheduling
Implementing Class: org.preesm.algorithm.mapper.ListSchedulingMapping
Short description: Undocumented

Inputs

SDF (of SDFGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

List Scheduling from DAG - Deprecated

Identifier: org.ietr.preesm.plugin.mapper.listschedulingfromdag
Implementing Class: org.preesm.algorithm.mapper.ListSchedulingMappingFromDAG
Short description: Undocumented

Inputs

DAG (of DirectedAcyclicGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple / Switcher	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

EnergyAwareness

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

List Scheduling from PiSDF

Identifier: pisdf-mapper.list
Implementing Class: org.preesm.algorithm.mapper.ListSchedulingMappingFromPiMM
Short description: Undocumented

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

EnergyAwareness

Undocumented

Value	Effect
True	Turns on energy aware mapping/scheduling

EnergyAwarenessFirstConfig

Undocumented

Value	Effect
First	Takes as starting point the first valid combination of PEs
Middle	Takes as starting point half of the available PEs
Max	Takes as starting point all the available PEs
Random	Takes as starting point a random number of PEs

EnergyAwarenessSearchType

Undocumented

Value	Effect
Thorough	Analyzes PE combinations one by one until the performance objective is reached
Halves	Divides in halves the remaining available PEs and goes up/down depending if the FPS reached are below/above the objective

Simple Scheduling from DAG - Deprecated

Identifier: org.ietr.preesm.plugin.mapper.simple
Implementing Class: org.preesm.algorithm.mapper.MainCoreMappingFromDAG
Short description: Undocumented

Inputs

DAG (of DirectedAcyclicGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

None.

Simple Scheduling from PiSDF

Identifier: pisdf-mapper.simple
Implementing Class: org.preesm.algorithm.mapper.MainCoreMappingFromPiMM
Short description: Undocumented

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

None.

PFast Scheduling from SDF - Deprecated

Identifier: org.ietr.preesm.plugin.mapper.pfast
Implementing Class: org.preesm.algorithm.mapper.PFASTMapping
Short description: Undocumented

Inputs

SDF (of SDFGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

displaySolutions

Undocumented

Value	Effect
false	Undocumented

fastTime

Undocumented

Value	Effect
100	Undocumented

fastLocalSearchTime

Undocumented

Value	Effect
10	Undocumented

nodesMin

Undocumented

Value	Effect
5	Undocumented

procNumber

Undocumented

Value	Effect
1	Undocumented

fastNumber

Undocumented

Value	Effect
100	Undocumented

PFast Scheduling from DAG - Deprecated

Identifier: org.ietr.preesm.plugin.mapper.pfastdag
Implementing Class: org.preesm.algorithm.mapper.PFASTMappingFromDAG
Short description: Undocumented

Inputs

DAG (of DirectedAcyclicGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

displaySolutions

Undocumented

Value	Effect
false	Undocumented

fastTime

Undocumented

Value	Effect
100	Undocumented

fastLocalSearchTime

Undocumented

Value	Effect
10	Undocumented

nodesMin

Undocumented

Value	Effect
5	Undocumented

procNumber

Undocumented

Value	Effect
1	Undocumented

fastNumber

Undocumented

Value	Effect
100	Undocumented

PFast Scheduling from PiSDF

Identifier: pisdf-mapper.pfast
Implementing Class: org.preesm.algorithm.mapper.PFASTMappingFromPiMM
Short description: Undocumented

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Undocumented

Parameters

edgeSchedType

Undocumented

Value	Effect
Simple	Undocumented

simulatorType

Undocumented

Value	Effect
LooselyTimed	Undocumented

Check

Undocumented

Value	Effect
True	Undocumented

Optimize synchronization

Undocumented

Value	Effect
False	Undocumented

balanceLoads

Undocumented

Value	Effect
false	Undocumented

displaySolutions

Undocumented

Value	Effect
false	Undocumented

fastTime

Undocumented

Value	Effect
100	Undocumented

fastLocalSearchTime

Undocumented

Value	Effect
10	Undocumented

nodesMin

Undocumented

Value	Effect
5	Undocumented

procNumber

Undocumented

Value	Effect
1	Undocumented

fastNumber

Undocumented

Value	Effect
100	Undocumented

Periodic Scheduling from PiSDF to old DAG

Identifier: pisdf-mapper.periodic.DAG
Implementing Class: org.preesm.algorithm.mapper.PeriodicMappingFromPiMMTask
Short description: Schedule and maps actors according to their periods.

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

DAG (of DirectedAcyclicGraph)
ABC (of LatencyAbc)

Description

Schedule and map actors according to their periods thanks to a list scheduler. Only works for homogeneous architectures, does not take into account communication times. Works also if there are no periods in the graph. Result is exported in the same format as pisdf-mapper.list standard scheduler.

Parameters

None.

Stats exporters

Internode stats exporter

Identifier: IntranodeExporterTask.identifier
Implementing Class: /org.preesm.algorithm.node.simulator.IntranodeExporterTask

Inputs

ABC (of LatencyAbc)
cMem (of ClusterMemory)

Outputs

None.

Description

This task exports the CSV file in order to generate the bar chart (occupation, speedup) for SIMSDP intra-node analysis (Thread partitioning workflow).

Description

Internode stats exporter

Identifier: InternodeExporterTask.identifier
Implementing Class: /org.preesm.algorithm.node.simulator.InternodeExporterTask

Inputs

ABC (of LatencyAbc)
void (from Gantt Exporter)

Outputs

void (for Multicriteria exporter)

Description

This task exports the CSV file in order to generate the trend chart (latency, standard deviation workload) for SIMSDP inter-node analysis (Simulation workflow). Depending on the recognized operating system, nodes are simulated on SimGrid if the system is linux, Preesm otherwise.

Multicriteria exporter

Identifier: RadarExporterTask.identifier
Implementing Class: /org.preesm.algorithm.node.simulator.RadarExporterTask

Inputs

ABC (of LatencyAbc)
void (from Internode Stats exporter)

Outputs

None.

Description

This task exports CSV files to generate a radar chart (final latency, memory, energy, cost per network) for SIMSDP multinet analysis (Node simulation workflow). This task takes input metrics, processes them, and exports the results to CSV files. The exported CSV files are used analyzed with a given notbook.

ABC Gantt exporter

Identifier: org.ietr.preesm.stats.exporter.StatsExporterTask
Implementing Class: org.preesm.algorithm.mapper.stats.exporter.StatsExporterTask
Short description: This task exports scheduling results as a *.pgantt file that can be viewed using the ganttDisplay viewer [1].

Inputs

ABC (of LatencyAbc)
scenario (of Scenario)

Outputs

None.

Description

This task exports scheduling results as a *.pgantt file that can be viewed using the ganttDisplay viewer [1]. The exported *.pgantt file uses the XML syntax.

Parameters

path

Path of the exported *.pgantt file. If the specified directory does not exist, it will not be created.

Value	Effect
/stats/xml/	Path within the Preesm project containing the workflow where the ”Gantt Exporter” task is instantiated. Exported Gantt will be named as follows: /path/in/proj/ stats.pgantt. If a graph with this name already exists in the given path, it will be overwritten.

ABC Gantt displayer

Identifier: org.ietr.preesm.plugin.mapper.plot
Implementing Class: org.preesm.algorithm.mapper.ui.stats.StatEditorAbcTask
Short description: Displays the result of a mapping/scheduling algorithm as a Gantt diagram.

Inputs

ABC (of LatencyAbc)
scenario (of Scenario)

Outputs

None.

Description

Displays the result of a mapping/scheduling algorithm as a Gantt diagram.

Parameters

None.

Synthesis Gantt displayer and exporter

Identifier: gantt-output
Implementing Class: org.preesm.algorithm.mapper.ui.stats.StatsEditorSynthesisTask
Short description: Undocumented

Inputs

PiMM (of PiGraph)
scenario (of Scenario)
architecture (of Design)
Schedule (of Schedule)
Mapping (of Mapping)
Allocation (of Allocation)

Outputs

None.

Description

Undocumented

Parameters

display

Specify if statistics, including Gantt diagram, must be displayed or not.

Value	Effect
true/false	Undocumented

file path

Folder to store Gantt diagram as xml file. Path is relative to the project, put “/” if at root, may be empty.

Value	Effect
/path/to	Undocumented

Memory Optimization

MEG Updater

Identifier: org.ietr.preesm.memory.exclusiongraph.MemExUpdater
Implementing Class: org.preesm.algorithm.memory.allocation.tasks.MemExUpdater
Short description: Relax memory allocation constraints of the MEG using scheduling information.

Inputs

DAG (of DirectedAcyclicGraph)
MemEx (of MemoryExclusionGraph)

Outputs

MemEx (of MemoryExclusionGraph)

Description

The MEG used in Preesm can be updated with scheduling information to remove exclusions between memory objects and make better allocations possible.

Parameters

Verbose

How verbose will this task be during its execution. In verbose mode, the task will log the start and completion time of the update, as well as characteristics (number of memory objects, density of exclusions) of the MEGs both before and after the update.

Value	Effect
false	(Default) The task will not log information.
true	The task will log build and MEG information.

Memory Allocation

Identifier: org.ietr.preesm.memory.allocation.MemoryAllocatorTask
Implementing Class: org.preesm.algorithm.memory.allocation.tasks.MemoryAllocatorTask
Short description: Perform the memory allocation for the given MEG.

Inputs

MemEx (of MemoryExclusionGraph) : Input Memory Exclusion Graph

Outputs

MEGs (of Map) : Map associating, for each memory element in the architecture, according to the chosen Distribution parameter value, a Memory Exclusion Graph annotated with allocation information (i.e. buffer addresses, etc.).

Description

Workflow task responsible for allocating the memory objects of the given MEG.

Parameters

Verbose

Verbosity of the task.

Value	Effect
True	Detailed statistics of the allocation process are logged
False	Logged information is kept to a minimum

Allocator(s)

Specify which memory allocation algorithm(s) should be used. If the string value of the parameters contains several algorithm names, all will be executed one by one.

Value	Effect
Basic	Each memory object is allocated in a dedicated memory space. Memory allocated for a given object is not reused for other.
BestFit	Memory objects are allocated one by one; allocating each object to the available space in memory whose size is the closest to the size of the allocated object. If MEG exclusions permit it, memory allocated for a memory object may be reused for others.
FirstFit	Memory objects are allocated one by one; allocating each object to the first available space in memory whose size is the large enough to allocate the object. If MEG exclusions permit it, memory allocated for a memory object may be reused for others.
DeGreef	Algorithm adapted from DeGreef (1997)}. If MEG exclusions permit it, memory allocated for a memory object may be reused for others.

Distribution

Specify which memory architecture should be used to allocate the memory.

Value	Effect
SharedOnly	(Default) All memory objects are allocated in a single memory bank accessible to all PE.
DistributedOnly	Each PE is associated to a private memory bank that no other PE can access. (Currently supported only in the MPPA code generation.)
Mixed	Both private memory banks and a shared memory can be used for allocating memory.
MixedMerged	Same as mixed, but the memory allocation algorithm favors buffer merging over memory distribution.

Best/First Fit order

When using FirstFit or BestFit memory allocators, this parameter specifies in which order the memory objects will be fed to the allocation algorithm. If the string value associated to the parameters contains several order names, all will be executed one by one.

Value	Effect
ApproxStableSet	Memory objects are sorted into disjoint stable sets. Stable sets are formed one after the other, each with the largest possible number of object. Memory objects are fed to the allocator set by set and in the largest first order within each stable set.
ExactStableSet	Similar to ‘ApproxStableSet’. Stable set are built using an exact algorithm instead of a heuristic.
LargestFirst	Memory objects are allocated in decreasing order of their size.
Shuffle	Memory objects are allocated in a random order. Using the ‘Nb of Shuffling Tested’ parameter, it is possible to test several random orders and only keep the best memory allocation.
Scheduling	Memory objects are allocated in scheduling order of their ‘birth’. The ‘birth’ of a memory object is the instant when its memory would be allocated by a dynamic allocator. This option can be used to mimic the behavior of a dynamic allocator. (Only available for MEG updated with scheduling information).

Data alignment

Option used to force the allocation of buffers (i.e. Memory objects) with aligned addresses. The data alignment property should always have the same value as the one set in the properties of the Memory Scripts task.

Value	Effect
None	No special care is taken to align the buffers in memory.
Data	All buffers are aligned on addresses that are multiples of their size. For example, a 4 bytes integer is aligned on 4 bytes address.
Fixed:=n	Where $n\in \mathbb{N}^*$. This forces the allocation algorithm to align all buffers on addresses that are multiples of n bytes.

Nb of Shuffling Tested

Number of random order tested when using the Shuffle value for the Best/First Fit order parameter.

Value	Effect
$n\in \mathbb{N}^$*	Number of random order.

Documented Errors

Message	Explanation
The obtained allocation was not valid because mutually exclusive memory objects have overlapping address ranges. The allocator is not working.	When checking the result of a memory allocation, two memory objects linked with an exclusion in the MEG were allocated in overlapping memory spaces. The error is caused by an invalid memory allocation algorithm and should be corrected in the source code.
The obtained allocation was not valid because there were unaligned memory objects. The allocator is not working.	When checking the result of a memory allocation, some memory objects were found not to respect the Dala alignment parameter. The error is caused by an invalid memory allocation algorithm and should be corrected in the source code.

Memory Bounds Estimator

Identifier: org.ietr.preesm.memory.bounds.MemoryBoundsEstimator
Implementing Class: org.preesm.algorithm.memory.allocation.tasks.MemoryBoundsEstimator
Short description: Compute bounds of the amount of memory needed to allocate the MEG

Inputs

MemEx (of MemoryExclusionGraph)

Outputs

MemEx (of MemoryExclusionGraph)
BoundMin (of Long)
BoundMax (of Long)

Description

The analysis technique presented in [1] can be used in Preesm to derive bounds for the amount of memory that can be allocated for an application. The upper bound corresponds to the worst memory allocation possible for an application. The lower bound is a theoretical value that limits the minimum amount of memory that can be allocated. By definition, the lower bound is not always reachable, which means that it might be impossible to find an allocation with this optimal amount of memory. The minimum bound is found by solving the Maximum Weight Clique problem on the MEG. This task provides a convenient way to evaluate the quality of a memory allocation.

Parameters

Verbose

How verbose will this task be during its execution. In verbose mode, the task will log the name of the used solver, the start and completion time of the bound estimation algorithm. Computed bounds are always logged, even if the verbose parameter is set to false.

Value	Effect
false	(Default) The task will not log information.
true	The task will log build and MEG information.

Solver

Specify which algorithm is used to compute the lower bound.

Value	Effect
Heuristic	(Default) Heuristic algorithm described in [1] is used. This technique find an approximate solution.
Ostergard	Östergård’s algorithm [2] is used. This technique finds an optimal solution, but has a potentially exponential complexity.
Yamaguchi	Yamaguchi et al.’s algorithm [3] is used. This technique finds an optimal solution, but has a potentially exponential complexity.

MEG Builder

Identifier: org.ietr.preesm.memory.exclusiongraph.MemoryExclusionGraphBuilder
Implementing Class: org.preesm.algorithm.memory.allocation.tasks.MemoryExclusionGraphBuilder
Short description: Builds the Memory Exclusion Graph (MEG) modeling the memory allocation constraints.

Inputs

DAG (of DirectedAcyclicGraph)
scenario (of Scenario)

Outputs

MemEx (of MemoryExclusionGraph)

Description

The memory allocation technique used in Preesm is based on a Memory Exclusion Graph (MEG). A MEG is a graph whose vertices model the memory objects that must be allocated in memory in order to run the generated code. In the current version of Preesm, each of these memory objects corresponds either to an edge of the Directed Acyclic Graph (DAG) or to a buffer corresponding to ”delays” of the graph that store data between executions of a schedule. In the MEG, two memory objects are linked by an edge (called an exclusion) if they can not be allocated in overlapping memory spaces.

Parameters

Verbose

How verbose will this task be during its execution. In verbose mode, the task will log the start and completion time of the build, as well as characteristics (number of memory objects, density of exclusions) of the produced MEG.

Value	Effect
false	(Default) The task will not log information.
true	The task will log build and MEG information.

Memory Scripts

Identifier: org.ietr.preesm.memory.script.MemoryScriptTask
Implementing Class: org.preesm.algorithm.memory.allocation.tasks.MemoryScriptTask
Short description: Executes the memory scripts associated to actors and merge buffers.

Inputs

DAG (of DirectedAcyclicGraph)
MemEx (of MemoryExclusionGraph)
scenario (of Scenario)

Outputs

MemEx (of MemoryExclusionGraph)

Description

Executes the memory scripts associated to actors and merge buffers. The purpose of the memory scripts is to allow Preesm to allocate input and output buffers of certain actors in overlapping memory range.

Parameters

Check

Verification policy used when checking the applicability of the memory scripts written by the developer and associated to the actor.

Value	Effect
Thorough	Will generate error messages with a detailed description of the source of the error. This policy should be used when writting memory scripts for the first time.
Fast	All errors in memory script are still detected, but error messages are less verbose. This verification policy is faster than the Thorough policy.
None	No verification is performed. Use this policy to speed up workflow execution once all memory scripts have been validated..

Data alignment

Option used to force the allocation of buffers with aligned addresses. The data alignment property should always have the same value as the one set in the properties of the Memory Allocation task.

Value	Effect
None	No special care is taken to align the buffers in memory.
Data	All buffers are aligned on addresses that are multiples of their size. For example,a 4 bytes integer is aligned on 4 bytes address.
Fixed:=$n$	Where $n\in \mathbb{N}^*$. This forces the allocation algorithm to align all buffers on addresses that are multiples of n bytes.

Log Path

Specify whether, and where, a log of the buffer matching optimization should be generated. Generated log are in the markdown format, and provide information on all matches created by scripts as well as which match could be applied by the optimization process.

Value	Effect
path/file.txt	The path given in this property is relative to the ”Code generation directory” defined in the executed scenario.
empty	No log will be generated.

Verbose

Verbosity of the workflow task.

Value	Effect
True	The workflow task will be verbose in the console.
False	The workflow task will be more quiet in the console.

Serial Memory Bounds

Identifier: org.ietr.preesm.memory.bounds.SerialMemoryBoundsEstimator
Implementing Class: org.preesm.algorithm.memory.allocation.tasks.SerialMemoryBoundsEstimator
Short description: Compute bounds of the amount of memory needed to allocate the MEGs.

Inputs

MEGs (of Map)

Outputs

None.

Description

This task computes the memory bounds (see Memory Bound Estimator Task) for several MEGs, like the one produced by the Memory Allocation task.

Parameters

Verbose

Value	Effect
false	(Default) The task will not log information.
true	The task will log build and MEG information.

Solver

Specify which algorithm is used to compute the lower bound.

Value	Effect
Heuristic	(Default) Heuristic algorithm described in [1] is used. This technique find an approximate solution.
Ostergard	Östergård’s algorithm [2] is used. This technique finds an optimal solution, but has a potentially exponential complexity.
Yamaguchi	Yamaguchi et al.’s algorithm [3] is used. This technique finds an optimal solution, but has a potentially exponential complexity.

Synthesis

Simple Synhtesis

Identifier: pisdf-synthesis.simple
Implementing Class: org.preesm.algorithm.synthesis.PreesmSynthesisTask
Short description: Schedule and map actors, and allocate their memory.

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

Schedule (of Schedule)
Mapping (of Mapping)
Allocation (of Allocation)

Description

Schedule and map actors and their communications, and allocate the buffer memory.Multiple available schedulers. Output is working only for the new code generation workflow tasks codegen2.

Parameters

scheduler

Scheduler used to schedule and map the tasks.

Value	Effect
simple	Naive greedy list scheduler.
legacy	See workflow task pisdf-mapper.list.
periodic	List scheduler (without communication times) respecting actor or graph periods, if any.
choco	Optimal scheduler (without communication times) respecting actor or graph periods, if any.

allocation

Allocate the memory for buffers.

Value	Effect
simple	Undocumented
legacy	Undocumented

Code Generation

Spider Codegen

Identifier: org.ietr.preesm.pimm.algorithm.spider.codegen.SpiderCodegenTask
Implementing Class: org.preesm.codegen.xtend.spider.SpiderCodegenTask
Short description: Generate Spider code for dynamic PiSDF.

Inputs

PiMM (of PiGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

None.

Description

Generate Spider code for dynamic PiSDF.

Parameters

scheduler

Runtime scheduler to use.

Value	Effect
list_on_the_go	Undocumented
round_robin	Undocumented
round_robin_scattered	Undocumented
list	(Default)

memory-alloc

Runtime memory allocation to use.

Value	Effect
special-actors	Undocumented
dummy	(Default)

shared-memory-size

Size of the shared memory allocated by Spider.

Value	Effect
$n$	$n > 0$ bytes. (Default = 67108864)

papify

Use of PAPIFY. Select type of feedback given too

Value	Effect
off	PAPIFY is off
dump	PAPIFY is on. Print csv files
feedback	PAPIFY is on. Give feedback to the GRT
both	PAPIFY is on. Print csv files and give feedback to the GRT

apollo

Whether to use Apollo.

Value	Effect
true / false

verbose

Whether to log.

Value	Effect
true / false

trace

Whether to trace what is happening at runtime.

Value	Effect
true / false

stack-type

Type of stack to use

Value	Effect
static	Use static stack
dynamic	Use dynamic stack

graph-optims

Whether to optimize the graph at runtime or not

Value	Effect
true / false

energy-awareness

Whether to activate or not energy-aware mapping/scheduling.

Value	Effect
true/false	Enable/disable energy-aware mapping/scheduling

Spider2 Codegen

Identifier: org.preesm.codegen.xtend.Spider2CodegenTask
Implementing Class: org.preesm.codegen.xtend.spider2.Spider2CodegenTask
Short description: Generate code for spider2 library for dynamic PiSDF.

Inputs

scenario (of Scenario)

Outputs

None.

Description

Generate code for spider2 library for dynamic PiSDF.

Parameters

generate archi

Whether to generate archi file from slam description or not

Value	Effect
true / false	(Default = true)

generate cmakelist

Whether to generate CMakeList.txt or not

Value	Effect
true / false	(Default = true)

move includes

Whether to move the includes of the project into the generated include folder

Value	Effect
true / false	(Default = false)

scheduler

Runtime scheduler to use.

Value	Effect
bestfit_list_scheduling	(Default)
round_robin_list_scheduling	Undocumented
greedy_scheduling	Undocumented

use-verbose

Whether to enable verbose log.

Value	Effect
true / false

enable-srdag-optims

Whether to optimize the srdag-graphs at runtime or not

Value	Effect
true / false

Code Generation

Identifier: org.ietr.preesm.codegen.xtend.task.CodegenTask
Implementing Class: org.preesm.codegen.xtend.task.CodegenTask
Short description: Generate code for the application deployment resulting from the workflow execution.

Inputs

MEGs (of Map)
DAG (of DirectedAcyclicGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

None.

Description

This workflow task is responsible for generating code for the application deployment resulting from the workflow execution.

The generated code makes use of 2 macros that can be overridden in the preesm.h user header file:

PREESM_VERBOSE : if defined, the code will print extra info about actor firing;
PREESM_LOOP_SIZE : when set to an integer value $n > 0$, the application will terminate after $n$ executions of the graph.
PREESM_NO_AFFINITY : if defined, the part of the code that sets the affinity to specific cores will be skipped;

When the loop size macro is omitted, the execution can be stopped by setting the global variable preesmStopThreads to 1. This variable is defined in the main.c generated file, and should be accessed using extern keyword.

Parameters

Printer

Specify which printer should be used to generate code. Printers are defined in Preesm source code using an extension mechanism that make it possible to define a single printer name for several targeted architecture. Hence, depending on the type of PEs declared in the architecture model, Preesm will automatically select the associated printer class, if it exists.

Value	Effect
C	Print C code and shared-memory based communications. Currently compatible with x86, c6678, and arm architectures.
InstrumentedC	Print C code instrumented with profiling code, and shared-memory based communications. Currently compatible with x86, c6678 architectures..
XML	Print XML code with all informations used by other printers to print code. Compatible with x86, c6678.

Papify

Enable the PAPI-based code instrumentation provided by PAPIFY

Value	Effect
true/false	Print C code instrumented with PAPIFY function calls based on the user-defined configuration of PAPIFY tab in the scenario. Currently compatibe with x86 and MPPA-256

Apollo

Enable the use of Apollo for intra-actor optimization

Value	Effect
true/false	Print C code with Apollo function calls. Currently compatibe with x86

Code Generation with cluster

Identifier: org.ietr.preesm.codegen.xtend.task.CodegenClusterTask
Implementing Class: org.preesm.codegen.xtend.task.CodegenWithClusterTask
Short description: Generate code for the application deployment resulting from the workflow execution.

Inputs

MEGs (of Map)
DAG (of DirectedAcyclicGraph)
scenario (of Scenario)
architecture (of Design)
CS (of Map)

Outputs

None.

Description

This workflow task is responsible for generating code for the application deployment resulting from the workflow execution.

The generated code makes use of 2 macros that can be overridden in the preesm.h user header file:

PREESM_VERBOSE : if defined, the code will print extra info about actor firing;
PREESM_LOOP_SIZE : when set to an integer value $n > 0$, the application will terminate after $n$ executions of the graph.
PREESM_NO_AFFINITY : if defined, the part of the code that sets the affinity to specific cores will be skipped;

Parameters

Printer

Value	Effect
C	Print C code and shared-memory based communications. Currently compatible with x86, c6678, and arm architectures.
InstrumentedC	Print C code instrumented with profiling code, and shared-memory based communications. Currently compatible with x86, c6678 architectures..
XML	Print XML code with all informations used by other printers to print code. Compatible with x86, c6678.

Papify

Enable the PAPI-based code instrumentation provided by PAPIFY

Value	Effect
true/false	Print C code instrumented with PAPIFY function calls based on the user-defined configuration of PAPIFY tab in the scenario. Currently compatibe with x86 and MPPA-256

Code Generation 2

Identifier: codegen2
Implementing Class: org.preesm.codegen.xtend.task.CodegenTask2
Short description: Undocumented

Inputs

PiMM (of PiGraph)
scenario (of Scenario)
architecture (of Design)
Schedule (of Schedule)
Mapping (of Mapping)
Allocation (of Allocation)

Outputs

None.

Description

Undocumented

Parameters

Printer

Value	Effect
C	Print C code and shared-memory based communications. Currently compatible with x86, c6678, and arm architectures.
InstrumentedC	Print C code instrumented with profiling code, and shared-memory based communications. Currently compatible with x86, c6678 architectures..
XML	Print XML code with all informations used by other printers to print code. Compatible with x86, c6678.

Papify

Enable the PAPI-based code instrumentation provided by PAPIFY

Value	Effect
true/false	Print C code instrumented with PAPIFY function calls based on the user-defined configuration of PAPIFY tab in the scenario. Currently compatibe with x86 and MPPA-256

Code Generation (SimSDP)

Identifier: org.ietr.preesm.codegen.xtend.task.CodegenSimSDPTask
Implementing Class: /org.ietr.preesm.codegen.xtend.task.CodegenSimSDPTask
Short description: Undocumented

Inputs

MEGs (of Map)
DAG (of DirectedAcyclicGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

None.

Description

This workflow task is responsible for generating code for the multinode HPC application deployment resulting from the workflow execution. This generated code operate with the SimSDPmain.c file connecting thread-based file with MPI functions.

The generated code makes use of 2 macros that can be overridden in the preesm.h user header file:

PREESM_VERBOSE : if defined, the code will print extra info about actor firing;
PREESM_LOOP_SIZE : when set to an integer value $n > 0$, the application will terminate after $n$ executions of the graph.
PREESM_NO_AFFINITY : if defined, the part of the code that sets the affinity to specific cores will be skipped;

When the loop size macro is omitted, the execution can be stopped by setting the global variable preesmStopThreads to 1. This variable is defined in the sub.c generated file, and should be accessed using extern keyword.

Parameters

Printer

Value	Effect
C	Print C code and shared-memory based communications. Currently compatible with x86, c6678, and arm architectures.
InstrumentedC	Print C code instrumented with profiling code, and shared-memory based communications. Currently compatible with x86, c6678 architectures..
XML	Print XML code with all informations used by other printers to print code. Compatible with x86, c6678.

Papify

Enable the PAPI-based code instrumentation provided by PAPIFY

Value	Effect
true/false	Print C code instrumented with PAPIFY function calls based on the user-defined configuration of PAPIFY tab in the scenario. Currently compatibe with x86 and MPPA-256

Apollo

Enable the use of Apollo for intra-actor optimization

Value	Effect
true/false	Print C code with Apollo function calls. Currently compatibe with x86

Other

Hypervisor

Identifier: hypervisor.task.identifier
Implementing Class: /org.preesm.algorithm.hypervisor.HypervisorTask.java
Short description: Undocumented

#### Inputs

void : Connector from scenario

#### Parameters

Iteration

Choose the number of SimSDP iteration to balance workload partitioning between architecture nodes.

Value	Effect
Integer	The resource allocation process iterate until the number of iteration complete.

Cluster Scheduler

Identifier: cluster-scheduler
Implementing Class: org.preesm.algorithm.clustering.scheduler.ClusterSchedulerTask
Short description: Undocumented

Inputs

PiMM (of PiGraph) : Input PiSDF graph
scenario (of Scenario) : Scenario

Outputs

PiMM (of PiGraph) : Output PiSDF graph
CS (of Map) : Map of Cluster Schedule

Description

Undocumented

Parameters

Target

Choose if the whole input graph will be scheduled rather than just clusters.

Value	Effect
Cluster	Clusters are scheduled.
Input graph	Input graph is scheduled.

Optimization criteria

Specify the criteria to optimize. If memory is choosen, some parallelizable actors will be sequentialized to minimize memory space. On the other hand, if performance is choosen, the algorithm will exploit every parallelism possibility.

Value	Effect
Memory	Minimize memory space of resulting clusters
Performance	Maximize performance of resulting clusters

Parallelism

Specify if resulting Cluster Schedules have to contain parallelism information.

Value	Effect
True	Cluster Schedules contain data parallelism information.
False	Cluster Schedules are purely sequential.

SDF4J Exporter - Deprecated

Identifier: sdf4j-export
Implementing Class: org.preesm.algorithm.io.xml.SDF4JGMLExporter
Short description: Undocumented

Inputs

SDF (of SDFGraph)

Outputs

None.

Description

Undocumented

Parameters

path

Undocumented

Value	Effect
__	Undocumented

Latency Evaluation

Identifier: org.ietr.preesm.latency.LatencyEvaluationPlugin
Implementing Class: org.preesm.algorithm.latency.LatencyEvaluationTask
Short description: Undocumented

Inputs

SDF (of SDFGraph)
scenario (of Scenario)

Outputs

SDF (of SDFGraph)
scenario (of Scenario)
latency (of Double)

Description

Undocumented

Parameters

multicore

Undocumented

Value	Effect
true/false

method

Undocumented

Value	Effect
FAST	(default) Hierarchical method
FLAT_LP	Based on Flattening the hierarchy
FLAT_SE	Based on Flattening the hierarchy

GetPiMM

Identifier: org.ietr.preesm.mapper.getpimm
Implementing Class: org.preesm.algorithm.mapper.GetPiMMFromDAGTask
Short description: Undocumented

Inputs

DAG (of DirectedAcyclicGraph)

Outputs

PiMM (of PiGraph)

Description

Undocumented

Parameters

None.

Hierarchical Scheduler - Deprecated

Identifier: hsceduler
Implementing Class: org.preesm.algorithm.mapper.algo.HScheduleTask
Short description: Undocumented

Inputs

None.

Outputs

None.

Description

Undocumented

Parameters

None.

Implementation Exporter

Identifier: org.ietr.preesm.plugin.mapper.exporter.ImplExportTransform
Implementing Class: org.preesm.algorithm.mapper.exporter.ImplExportTransform
Short description: Undocumented

Inputs

DAG (of DirectedAcyclicGraph)

Outputs

xml (of String)

Description

Undocumented

Parameters

path

Undocumented

Value	Effect
__	Undocumented

Single rate SDF to DAG Transformation

Identifier: org.ietr.preesm.mapper.SDF2DAGTransformation
Implementing Class: org.preesm.algorithm.mapper.graphtransfo.SDF2DAGTransformation
Short description: Undocumented

Inputs

SDF (of SDFGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

DAG (of DirectedAcyclicGraph)

Description

Undocumented

Parameters

None.

Memory Exclusion Graph Mapper

Identifier: org.ietr.preesm.memory.distributed.MapperTask
Implementing Class: org.preesm.algorithm.memory.allocation.tasks.MapperTask
Short description: Undocumented

Inputs

MemEx (of MemoryExclusionGraph)

Outputs

MemExes (of Map)

Description

Undocumented

Parameters

Verbose

Undocumented

Value	Effect
? C {True, False}

Distribution

Specify which memory architecture should be used to allocate the memory.

Value	Effect
SharedOnly	(Default) All memory objects are allocated in a single memory bank accessible to all PE.
DistributedOnly	Each PE is associated to a private memory bank that no other PE can access. (Currently not supported by code generation.)
Mixed	Both private memory banks and a shared memory can be used for allocating memory.
MixedMerged	Same as mixed, but the memory allocation algorithm favors buffer merging over memory distribution.

Activity Exporter of Tokens and Quanta

Identifier: org.ietr.preesm.algorithm.moa.activity.ActivityExporter
Implementing Class: org.preesm.algorithm.moa.activity.ActivityExporter
Short description: Undocumented

Inputs

ABC (of LatencyAbc)

Outputs

None.

Description

Undocumented

Parameters

path

Undocumented

Value	Effect
stats/mat/activity	Undocumented

human_readable

Undocumented

Value	Effect
Yes	Undocumented

Activity Exporter of Custom Quanta

Identifier: org.ietr.preesm.algorithm.moa.activity.CustomQuantaExporter
Implementing Class: org.preesm.algorithm.moa.activity.CustomQuantaExporter
Short description: Undocumented

Inputs

ABC (of LatencyAbc)

Outputs

None.

Description

Undocumented

Parameters

xls_file

Undocumented

Value	Effect
stats/mat/custom_quanta_in/quanta_in$SCENARIO$.xls_

path

Undocumented

Value	Effect
stats/mat/activity

human_readable

Undocumented

Value	Effect
Yes

Activity Exporter of Tokens and Quanta for a single ABC

Identifier: org.ietr.preesm.algorithm.moa.activity.MonoActivityExporter
Implementing Class: org.preesm.algorithm.moa.activity.MonoActivityExporter
Short description: Undocumented

Inputs

ABC (of LatencyAbc)

Outputs

None.

Description

Undocumented

Parameters

path

Undocumented

Value	Effect
stats/mat/activity

human_readable

Undocumented

Value	Effect
Yes

Malleable Parameters setter

Identifier: pisdf-mparams.setter
Implementing Class: org.preesm.algorithm.mparameters.SetMalleableParametersTask
Short description: Set the malleable parameters default value according to the best schedule found.

Inputs

PiMM (of PiGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

PiMM (of PiGraph)

Description

Set the malleable parameters default value in the scenario according to the best schedule found.Works only on homogeneous architectures. Different strategies are possible, exhaustive search or heuristics.

Parameters

1. Comparisons

Order of comparisons of the metrics (T for throughput or P for power or E for energy or L for latency or M for makespan, separated by >). Latency is indexed from 1 to the maximum number of pipeline stages allowed.

Value	Effect
T>P>L	Metrics are compare from left to right.

2. Thresholds

Objectives of the metrics. Threshold if it is any integer higher than 0, minimize it otherwise.

Value	Effect
0>0>0	In the same order as the metrics.

3. Params objectives

Tells to minimize (-) or maximize (+) a parameter (after main objectives). May be empty.

Value	Effect
>	Syntax: >+parentGraphName/parameterName>-…

4. Number heuristic

Use a DSE heuristic on all malleable parameter expressions which are integer numbers. Only a subset of their expressions are explored.

Value	Effect
false	False disables the heuristic.

5. Retry with delays

Use a DSE heuristic to try to add delays if it improves the throughput. See workflow task pisdf-delays.setter. Number of pipelines is inferred automatically.

Value	Effect
false	False disables the heuristic.

6. Log path

Export all explored points with associated metrics in a csv file.

Value	Effect
/Code/generated/	Path relative to the project root.

Automatic Placement of Delays

Identifier: pisdf-delays.setter
Implementing Class: org.preesm.algorithm.pisdf.autodelays.AutoDelaysTask
Short description: Puts delays in a flat PiMM, in order to speed up the execution.

Inputs

PiMM (of PiGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

PiMM (of PiGraph)

Description

Puts delays in a flat PiMM, in order to speed up the execution. Works only on homogeneous architectures. The heuristic will perform a search of all simple cycles, so the task may take a long time to run if many cycles are present.

Parameters

Selection cuts

Number of graph cuts to consider, higher or equal to the maximum number of cuts.

Value	Effect
4	Split the graph in zones of equivalent work load.

Maximum cuts

Maximum number of graph cuts induced by the added delays. Each graph cut adds one pipeline stage. If delays are already present, the values are summed.

Value	Effect
1

Test best choco ?

Computes all topological graph cuts with a CP solver. All topological cuts are evaluated with a list scheduler to select the best.

Value	Effect
false	False disables this comparison.

Test scheduling ?

Whether or not a schedule must be generated at the end.

Value	Effect
false	False disables this feature.

Fill cycles ?

Whether or not the cycles must be broken with extra delays.

Value	Effect
false	False disables this feature.

Periods Prescheduling Checker

Identifier: org.ietr.preesm.pimm.algorithm.checker.periods.PeriodsPreschedulingChecker
Implementing Class: org.preesm.algorithm.pisdf.periods.PeriodsPreschedulingCheckTask
Short description: Check necessary condition to schedule graphs with periods (at top level or in actors).

Inputs

PiMM (of PiGraph)
scenario (of Scenario)
architecture (of Design)

Outputs

PiMM (of PiGraph)

Description

Check necessary condition to schedule graphs with periods (at top level or in actors). Works only on flat graphs.

Parameters

Selection rate (%)

Percentage of periodic actors to consider.

Value	Effect
100	All periodic actors are checked.

Periodic scheduling (without output)

Identifier: pisdf-synthesis.void-periodic-schedule
Implementing Class: org.preesm.algorithm.synthesis.schedule.VoidPeriodicScheduleTask
Short description: Schedule and maps actors according to their periods.

Inputs

PiMM (of PiGraph)
architecture (of Design)
scenario (of Scenario)

Outputs

None.

Description

Parameters

solver

Algorithm used to schedule and map.

Value	Effect
list	List scheduler.
choco	Optimal CP scheduler.

Throughput Evaluation

Identifier: org.ietr.preesm.throughput.ThroughputPlugin
Implementing Class: org.preesm.algorithm.throughput.ThroughputEvaluationTask
Short description: Undocumented

Inputs

SDF (of SDFGraph)
scenario (of Scenario)

Outputs

throughput (of Double)
SDF (of SDFGraph)
scenario (of Scenario)

Description

Undocumented

Parameters

method

Undocumented

Value	Effect
SR	Schedule-Replace technique
ESR	Evaluate-Schedule-Replace method
HPeriodic	Hierarchical Periodic Schedule method
Classical	Based on Flattening the hierarchy

SDF2HSDF

Identifier: org.ietr.preesm.plugin.transforms.sdf2hsdf
Implementing Class: org.preesm.algorithm.transforms.HSDFTransformation
Short description: Undocumented

Inputs

SDF (of SDFGraph)

Outputs

SDF (of SDFGraph)

Description

transform a SDF graph into a HSDF graph, that is into a single rate graph

Parameters

None.

Algorithm Iterator

Identifier: org.ietr.preesm.algorithm.transforms.IterateAlgorithm
Implementing Class: org.preesm.algorithm.transforms.IterateAlgorithm
Short description: Undocumented

Inputs

SDF (of SDFGraph)

Outputs

SDF (of SDFGraph)

Description

Undocumented

Parameters

nbIt

Undocumented

Value	Effect
1

setStates

Undocumented

Value	Effect
true

Papify Engine - Deprecated

Identifier: org.ietr.preesm.codegen.xtend.task.CodegenPapifyEngineTask
Implementing Class: org.preesm.codegen.xtend.task.CodegenPapifyEngineTask
Short description: Deprecated - does nothing (as of v3.9.1). See parameter ‘Papify’ in CodegenTask’.

Old doc: Generate the required instrumentation code for the application based on the PAPIFY tab information.

Inputs

scenario (of Scenario)
DAG (of DirectedAcyclicGraph)

Outputs

DAG (of DirectedAcyclicGraph)

Description

Deprecated - does nothing (as of v3.9.1). See parameter ‘Papify’ in CodegenTask’.

Old doc: This workflow task is responsible for generating the instrumentation of the code for the application based on the PAPIFY tab information.

The generated code makes use of 1 macro that enables/disables the monitoring in the preesm.h user header file:

_PREESM_PAPIFY_MONITOR : if defined, the code monitoring will take place;

Parameters

None.

PiSDF BRV Exporter

Identifier: pisdf-brv-export
Implementing Class: org.preesm.model.pisdf.brv.BRVExporter
Short description: Undocumented

Inputs

PiMM (of PiGraph)

Outputs

PiMM (of PiGraph)

Description

Undocumented

Parameters

path

Undocumented

Value	Effect
/stats/xml/	default value

PiSDF Checker

Identifier: pisdf-checker
Implementing Class: org.preesm.model.pisdf.check.PiMMAlgorithmCheckerTask
Short description: Undocumented

Inputs

PiMM (of PiGraph)

Outputs

PiMM (of PiGraph)

Description

Undocumented

Parameters

None.

scenario

Identifier: org.ietr.preesm.scenario.task
Implementing Class: org.preesm.model.scenario.workflow.AlgorithmAndArchitectureScenarioNode
Short description: Undocumented

Inputs

None.

Outputs

scenario (of Scenario)
architecture (of Design)
PiMM (of PiGraph)

Description

Undocumented

Parameters

None.

Slam Flattener

Identifier: org.ietr.preesm.archi.slam.flattening
Implementing Class: org.preesm.model.slam.utils.SlamHierarchyFlattening
Short description: Undocumented

Inputs

architecture (of Design)

Outputs

architecture (of Design)

Description

Undocumented

Parameters

depth

Undocumented

Value	Effect
n > 0	default = 1

Graph Transformation

Data-parallel Transformation

Inputs

Outputs

Description

Parameters

Documented Errors

See Also

Clustering

Inputs

Outputs

Description

Parameters

Cluster Partitioner

Inputs

Outputs

Description

Parameters

Number of PEs in clusters

SCAPE transformation

Inputs

Outputs

Description

Parameters

Level number

SCAPE mode

Stack size

See Also

Static PiMM to IBSDF - Deprecated

Inputs

Outputs

Description

Parameters

See Also

Single-Rate Transformation - Deprecated

Inputs

Outputs

Description

Parameters

ExplodeImploreSuppr

Documented Errors

See Also

Hierarchy Flattening - Deprecated

Inputs

Outputs

Description

Parameters

Documented Errors

See Also

PiSDF Flattener

Inputs

Outputs

Description

Parameters

Perform optimizations

PiSDF Single-Rate Transformation

Inputs

Outputs

Description

Parameters

Consistency_Method

PiSDF Transform Perfect Fit Delay To End Init

Inputs

Outputs

Description

Parameters

Node Partitioner

Inputs

Outputs

Description

Parameters

archi path

Partitioning mode

Graph Exporters

SDF Exporter - Deprecated

Inputs

Outputs

Description

Parameters

path