Sub-Modular Node Design

The mflowgen nodes we provide for DC synthesis and Innovus place-and-route nodes are designed to be submodular. This means that each node (which is already modular on its own) is further split into individual scripts for various purposes (e.g., reporting, setting up variables and margins, tweaking constraints). An mflowgen parameter called order then allows the user to parameterize how the node is run from a Python interface. For example, the user can remove a script that is not needed for their design, add an additional custom script for their design, or simply define a reordering of the existing scripts.

Take a look at the design initialization node (i.e., cadence-innovus-init), which is responsible for reading in the post-synthesis design files from Synopsys DC and executing floorplanning. This node configuration file lists the following inputs and outputs:

input	innovus-foundation-flow	Scripts generated by the Innovus Foundation Flow.
input	adk	ASIC design kit interface to the process technology and library files.
input	design.v	The post-synthesis gate-level netlist.
input	design.sdc	Constraints dumped from synthesis.
output	design.checkpoint	The working Innovus database after the node finishes.

The configuration file also defines the order parameter, which lists scripts in the following order (note that the contents of these scripts can be found in the scripts subdirectory):

main.tcl	Main script from the Innovus Foundation Flow for design initialization.
quality-of-life.tcl	Useful variables and settings when working with Innovus.
floorplan.tcl	Creates the floorplan with the core area and all margins.
pin-assignments.tcl	Places IO pins along the perimeter of the design.
make-path-groups.tcl	Creates timing path groups to help the timing engine prioritize certain paths.
reporting.tcl	Dumps a variety of reports.

The order parameter determines the order in which the scripts are run. For example, floorplanning (floorplan.tcl) will be executed after we read in the design (main.tcl) and set up quality-of-life variables (quality-of-life.tcl), with reporting done at the end of the node (reporting.tcl). The order parameter can be accessed as a normal python list in your mflowgen graph defined in construct.py. For example, the parameter can be printed like this:

>>> init = Node( 'cadence-innovus-init', default=True )
>>> order = init.get_param( 'order' )
>>> print( order )
['main.tcl', 'quality-of-life.tcl', 'floorplan.tcl',
'pin-assignments.tcl', 'make-path-groups.tcl', 'reporting.tcl']

The commands run Innovus with a special script that simply loops through the order parameter and runs each item one by one. It searches for each script in both the node’s local scripts directory (for the default versions of the scripts) as well as the inputs directory (where the user can supply custom versions of the scripts) with priority given to the inputs. This allows users to extend the node with new scripts.

Note

If the same script is found in both the scripts and inputs directories, priority is given to the copy in the inputs directory.

For example, suppose that “scripts/foo.tcl” exists and the flow designer then supplies “inputs/foo.tcl”. In this case, “inputs/foo.tcl” is run and “scripts/foo.tcl” is ignored.

For example, if we had a node called custom-init that provided new-last-step.tcl as an output, we could parameterize the init node to append the new script at the end:

• main.tcl

• quality-of-life.tcl

• floorplan.tcl

• pin-assignments.tcl

• make-path-groups.tcl

• reporting.tcl

• new-last-step.tcl (a new custom script)

The mflowgen construct script would use the mflowgen API for extending the list of inputs in a node (i.e., extend_inputs()) and extend the init node inputs with the list of all outputs of our custom node (i.e., all_outputs()). Running make graph would then produce a graph with a new input edge into the init node like this:

The mflowgen construct script that does the above would look like this:

init        = Node( 'cadence-innovus-init', default=True )
custom_init = Node( ... ) # Comes from somewhere and
                          # has an output 'new-last-step.tcl'

init.extend_inputs( custom_init.all_outputs() )

g.add_node( init )
g.add_node( custom_init )
g.connect_by_name( custom_init, init )

Finally, we would parameterize the order of the init node to include the new input script:

order = init.get_param( 'order' )
order.append( 'new-last-step.tcl' ) # Append the script to run last
init.update_params( { 'order': order } )

print( order ) # -> [ 'main.tcl', ..., 'new-last-step.tcl']

Note that we could simply have replaced the entire init node with our own custom version. However, it can be very useful to reuse the majority of an existing node while tweaking just a small part of it.