run_vtr_flow

This script runs the VTR flow for a single benchmark circuit and architecture file.

The script is located at:

$VTR_ROOT/vtr_flow/scripts/run_vtr_flow.py

Basic Usage

At a minimum run_vtr_flow.py requires two command-line arguments:

run_vtr_flow.py <circuit_file> <architecture_file>

where:

  • <circuit_file> is the circuit to be processed

  • <architecture_file> is the target FPGA architecture

Note

The script will create a ./temp directory, unless otherwise specified with the -temp_dir option. The circuit file and architecture file will be copied to the temporary directory. All stages of the flow will be run within this directory. Several intermediate files will be generated and deleted upon completion. Users should ensure that no important files are kept in this directory as they may be deleted.

Output

The standard out of the script will produce a single line with the format:

<architecture>/<circuit_name>...<status>

If execution completed successfully the status will be ‘OK’. Otherwise, the status will indicate which stage of execution failed.

The script will also produce an output files (*.out) for each stage, containing the standout output of the executable(s).

Advanced Usage

Additional optional command arguments can also be passed to run_vtr_flow.py:

run_vtr_flow.py <circuit_file> <architecture_file> [<options>] [<vpr_options>]

where:

  • <options> are additional arguments passed to run_vtr_flow.py (described below),

  • <vpr_options> are any arguments not recognized by run_vtr_flow.py. These will be forwarded to VPR.

For example:

run_vtr_flow.py my_circuit.v my_arch.xml -track_memory_usage --pack --place

will run the VTR flow to map the circuit my_circuit.v onto the architecture my_arch.xml; the arguments --pack and --place will be passed to VPR (since they are unrecognized arguments to run_vtr_flow.py). They will cause VPR to perform only packing and placement.

Detailed Command-line Options

Note

Any options not recognized by this script is forwarded to VPR.

-starting_stage <stage>

Start the VTR flow at the specified stage.

Accepted values:

  • odin

  • abc

  • scripts

  • vpr

Default: odin

-ending_stage <stage>

End the VTR flow at the specified stage.

Accepted values:

  • odin

  • abc

  • scripts

  • vpr

Default: vpr

-power

Enables power estimation.

See Power Estimation

-cmos_tech <file>

CMOS technology XML file.

See Technology Properties

-delete_intermediate_files

Delete intermediate files (i.e. .dot, .xml, .rc, etc)

-delete_result_files

Delete result files (i.e. VPR’s .net, .place, .route outputs)

-track_memory_usage

Record peak memory usage and additional statistics for each stage.

Note

Requires /usr/bin/time -v command. Some operating systems do not report peak memory.

Default: off

-limit_memory_usage

Kill benchmark if it is taking up too much memory to avoid slow disk swaps.

Note

Requires ulimit -Sv command.

Default: off

-timeout <float>

Maximum amount of time to spend on a single stage of a task in seconds.

Default: 14 days

-temp_dir <path>

Temporary directory used for execution and intermediate files. The script will automatically create this directory if necessary.

Default: ./temp

-valgrind

Run the flow with valgrind while using the following valgrind options:

  • –leak-check=full

  • –errors-for-leak-kinds=none

  • –error-exitcode=1

  • –track-origins=yes

-min_hard_mult_size <int>

Tells ODIN II the minimum multiplier size that should be implemented using hard multiplier (if available). Smaller multipliers will be implemented using soft logic.

Default: 3

-min_hard_adder_size <int>

Tells ODIN II the minimum adder size that should be implemented using hard adders (if available). Smaller adders will be implemented using soft logic.

Default: 1

-adder_cin_global

Tells ODIN II to connect the first cin in an adder/subtractor chain to a global gnd/vdd net. Instead of creating a dummy adder to generate the input signal of the first cin port of the chain.