ADFG User’s guide

ADFG (Affine DataFlow Graph) tool checks the scheduling feasibility of an Ultimately Cyclo-Static DataFlow (UCSDF) graph which represents a system of actors (or processes) and their dependencies (data production and consumption rates), aka channels between them. If the graph is schedulable, regarding to a specified policy, runtime parameters are computed to help the system implementation or build (if it is an embedded system)

ADFG accepts two kind of input files: SDF3 standard (modified for csdf) and .adfg. You also can export the provided SDF3 examples into .adfg with the results (adfg model enables to store the computed periods, at the opposite of SDF3). ADFG comes with a command line interface tool and some Eclipse plugins.

##Using ADFG tool as a command line

Syntax:

• adfg_sched FILENAME ANALYSIS [OPTIONS]
• adfg_sched -h for help

This command takes a SDF3 file FILENAME (describing the system) and a scheduling policy ANALYSIS as mandatory arguments. The number of processors available in the system is also mandatory if a multi-processor policy is chosen, it is set with the option -p.

FILENAME: ADFG or SDF3 file to analyse (.adfg or .xml).

ANALYSIS: specifies the scheduling policy (several variants of EDF or SP policies, UNI or MULTI processors). See below for details. The accepted values are:

• EDF, Mono processor: EDF_UNI, EDF_UNI_DA, EDF_UNI_HDA, EDF_MULT_MBS_HDA, EDF_UNI_PQPA
• EDF, Multi processors: EDF_MULT_MBS_ID, EDF_MULT_BF_UF_ID, EDF_MULT_BF_SQPA_CD, EDF_MULT_BF_PQPA, GEDF_MULT
  
• SP, Mono processor: SP_UNI, SP_UNI_LOP, SP_UNI_UDH
• SP, Multi processors: SP_MULT_MBS, SP_MULT_BF_FBBFFD, SP_MULT_BF_SRTA

OPTIONS:

• -b (–keepMapping) : Use the given actor partition (i. e. computed only if not provided). (default: false)
• -c (–cheddar) FILENAME : Export actors properties in the XML Cheddar format.
• -d (–dot) FILENAME : Export graph properties in the DOT format.
• -g (–WCETgcd) : Reduce all WCET by their greatest common divisor (for simulation). (default: false)
• -h (–help) : Print usage. (default: true)
• -i (–keepInitToken) : Use the given initial buffer tokens (i. e. computed only if not provided). (default: false)
• -m (–keepMaxToken) : Use the given max buffer sizes (i. e. computed only if not provided). (default: false)
• -n (–newFile) FILENAME : Create a results file (ADFG .adfg format, or SDF3 .xml if constructed by a SDF3 file).
• -p (–nbProcs) POS INT : Number of processors. (default: 1)
• -q (–quiet) : Print only errors (disable verbose mode). (default: false)
• -r (–compareResult) : Compare result with the given input. (default: false)
• -s (–strategy) [BUFFER_MIN | THROUGHPUT_MAX | BALANCED] : Solving objective.
• -t (–keepPeriod) : Use the given actor periods (i. e. computed only if not provided). (default: false)
• -u (–updateXML) : Update intial file with computed results. (default: false)
• -v (–verbose) : Print more detailed information. (default: false)
• -y (–yartiss) FILENAME : Export actors properties in the XML Yartiss format.
• -z (–zeroDelay) : All channel initial number of tokens (delays) must be zero. (default: false)

###Example

It is assumed that the setup of ADFG has been executed. The command

adfg_sched Beamformer.xml EDF_UNI

provides this result.

##Using ADFG tool under Eclipse

It is assumed that you have installed ADFG according to the installation procedure, and the setup of ADFG has been executed. Using adfg under Eclipse is quite similar to the adfg_sched command line.

Run Eclipse,

• select a sdf3 file and right click,
• select in the menu the line “ADFG for SDF” and select a scheduling policy (example),
• set the options and run ADFG (example)
• the result is written in the console windows (example).

##About the scheduling policies:

• EDF (Early Deadline First)
  • Mono processor (not restricted to weakly connected DFG):
    • EDF_UNI
    • EDF_UNI_DA Deadline Adjustment to avoid preemptions
    • EDF_UNI_HDA Hybbrid Deadline Adjustment
    • EDF_UNI_PQPA Parametric Quick Processor demand Analysis
  • Multi processors (restricted to weakly connected DFG except EDF_MULT_PQPA):
    • EDF_MULT_MBS_HDA Scotch, Hybrid Deadline Adjustment
    • EDF_MULT_MBS_ID Scotch, Implicit Deadline
    • EDF_MULT_BF_UF_ID Best Fit, Implicit Deadline
    • EDF_MULT_BF_SQPA_CD Best Fit, Constrained Deadline
    • EDF_MULT_BF_PQPA Best Fit, Parametric Quick Processor demand Analysis
  • Global multi processors (restricted to weakly connected DFG):
    • GEDF_MULT
• SP (Static Priority)
  • Mono processor (restricted to weakly connected DFG):
    • SP_UNI Deadline Monotonic
    • SP_UNI_LOP Linear Odering Problem to minimize total buffer size
    • SP_UNI_UDH Utilization Distance Heuristic to minimize total buffer size
  • Multi processors (restricted to weakly connected DFG):
    • SP_MULT_MBS Scotch
    • SP_MULT_BF_FBBFFD Best Fit, Fisher Baruah Baker First Fit Decreasing
    • SP_MULT_BF_SRTA Best Fit, Symbolic Response Time Analysis