TritonPart : A Constraints Driven General Hypergraph Partitioner

TritonPart is a constraints-driven general hypergraph partitioner. Currently our repo is under active development.

Table of Content

• OpenROAD includes the source code for TritonPart. The codes are available in OpenROAD/src/par.
• titan23_benchmark provides the titan23 benchmark suites.
• regression provides the scripts for benchmarking TritonPart.

Features

• Start of the art multiple-constraints driven partitioning “multi-tool”
• Optimizes cost function based on user requirement
• Permissive open-source license
• Solves multi-way partitioning with following features:
  • Multidimensional weights on vertices and hyperedges
  • Multilevel coarsening and refinement framework
  • Fixed vertices
  • Timing-driven partitioning framework
  • Community constraints: Groups of vertices need to be in same partition

How to build TritonPart

The first step, independent of the build method, is to download the repository:

• We implement our TritonPart based on OpenROAD app. Please check the dependencies of OpenROAD app first.
• We use Google OR-Tools and CPLEX as our ILP solver. Please install Google OR-Tools following the instructions and CPLEX.
• Please unload Anaconda from the environment.
• Run following scripts

mkdir build
cd build
cmake ../OpenROAD/
make

How to partition a hypergraph in the way you would using hMETIS

The example tcl command is as follows.

triton_part_hypergraph -hypergraph_file des90.hgr -num_parts 3 -balance_constraint 2 -seed 2

You can run this example with following comands.

cd test
source run.sh

How to partition a hypergraph with fixed vertices constraints

The example tcl command is as follows.

triton_part_hypergraph -hypergraph_file sample.hgr -num_parts 2 -balance_constraint 20 -seed 2 -fixed_file sample.fix

You can run this example with following comands.

cd test
source run_fix.sh

How to partition a hypergraph with embedding constraints

The example tcl command is as follows.

triton_part_hypergraph -hypergraph_file sample.hgr -num_parts 2 -balance_constraint 20 -seed 2 \
-placement_file sample.emb -placement_wt_factors { 1.0 1.0 } -placement_dimension 2

You can run this example with following comands.

cd test
source run_emb.sh

How to run regression test

We recommend users to specify their parameters by modifying the regression.py from Line 131 to Line 148. You can run the regression scripts as following:

python regression.py | tee summary.log