metis_unicorn

Implementation of hMETIS

PHONEY [Pronounce 'poney'], Partitioning of Hypergraph Obviously Not EasY

Note: - It is very important at the moment that the clusters have the same ID as their
        position in the Graph.clusters list.
      - All .out files should be in the same <dir> folder. However, the script will 
        look the parent folder if they can't be found.

Usage:
    phoney.py   [-d <dir>] [-w <weight>] [--seed=seed] [--algo=algo] [--path=path] [--fix-pins]
                [--simple-graph] [--custom-fixfile] [--netsegments] [--ub=UBfactor] [--onetoone]
    phoney.py --help

Options:
    -d <dir>            Design directory containing the cluster information (.out files and such)
    -w <weight>         Amount of edge weights to consider, [1, 10]
    --seed=seed         RNG seed
    --algo=algo         0: hMETIS [default: 0]
                        1: PaToH
                        2: Circut
                        3: Custom partitioner
                        4: Random partitioner
    --path=path         Full path to the partitioning tool
    --simple-graph      If set, translate the hypergraph into a simple graph
    --fix-pins          If set, force all standard cells connected to a pin to be placed
                        on the bottom die.
    --custom-fixfile    Use existing fixfile for hmetis, located in <dir> and named 'fixfile.hgr'
    --netsegments       Use alternative net segments for the graph generation. Needs a file named 'WLnets_segments.out'.
    --ub=UBfactor       Hmetis' UBfactor. eg ub=1 means partitions allowed to a disbalance of 49/51
    --onetoone          Flag set if there is not clustering applied to the design.
    -h --help           Print this help

Weight types

• 0: 01_NoWires_Area
• 1: 02_1_NoWires_Area
• 2: 03_TotLength_Area
• 3: 04_1_TotLength_Area
• 4: 05_AvgLength_Area
• 5: 06_1_AvgLength_Area
• 6: 07_NoWiresXTotLength_Area
• 7: 08_1_NoWiresXTotLength_Area
• 8: 09_NoWires+TotLength_Area
• 9: 10_1_NoWires+TotLength_Area

Input files

ClustersArea.out

Header: Name Type InstCount Boundary Area

• Column 1: <cluster name>, same as in Clusters.out
• Column 2: Type of cluster, exclusive by default.
• Column 3: Amount of instances in the cluster [integer]
• Column 4: Coordinates of the cluster's bottom left corner, (<x coordinate [float]>,<y coordinate [float]>)
• Column 5: Coordinates of the cluster's top right corner, (<x coordinate [float]>,<y coordinate [float]>) Columns 4 and 5 only make sense if the cluster is rectangular shaped.
• Column 6: area of the cluster in µm2 [float]

ClustersInstances.out

Each line: <cluster name> <instance name 1> <...> <instance name n>.

InstancesPerNet.out

Each line is <net name> <instance name 1> <...> <instance name n>.

WLnets.out

Header: NET NUM_PINS LENGTH

Then each line is <net name> <number of pins [integer]> <length in µm [float]>.

bb.out

Memory block file specific to some designs. The relevant lines should be as follows: <instance name> <instance type> <amount of such instance> <total area of the gates in the instance> <total physical area of the instance> <porosity [default: unspecified]> <total area of the cumulated instance> <width> <height> <orientation> <type of standard cells inside>

Output files

.part.2

.cut

.tcl

.part

.hgr

connectivity_partition.txt

cutLength_partition.txt

raw_hyperedges.out