DATE'25 "Timing-Driven Global Placement by Efficient Critical Path Extraction"

We provide the implementation of the method proposed in the paper. It is built upon the popular open-source infrastructure DREAMPlace.

Highlights of the update

• Optional HeteroSTA backend for faster timing analysis through the timer_engine configuration.
• Improved determinism in the pin-to-pin attraction flow, which addresses the nondeterministic behavior fixed in the merged updates.
• Ready-to-run benchmark JSONs for both the original OpenTimer-based flow and the HeteroSTA-based flow.
• Thanks to Zizheng Guo (gzz@pku.edu.cn) and Shenglu Hua for providing these updates.

Build with Docker

We highly recommend the use of Docker to enable a smooth environment configuration.

The following steps are borrowed from DREAMPlace repository. We make minor revisions to make it more clear.

1. Get the code and put it in folder DATE25-TDP.

2. Get the container:

• Option 1: pull from the cloud limbo018/dreamplace.

  docker pull limbo018/dreamplace:cuda
  

• Option 2: build the container.

  docker build . --file Dockerfile --tag your_name/dreamplace:cuda
  

3. Enter bash environment of the container. Replace limbo018 with your name if option 2 is chosen in the previous step.

• Option 1: Run with GPU on Linux.

  docker run --gpus 1 -it -v $(pwd):/DATE25-TDP limbo018/dreamplace:cuda bash
  

• Option 2: Run with CPU on Linux.

  docker run -it -v $(pwd):/DATE25-TDP limbo018/dreamplace:cuda bash
  

4. cd /DATE25-TDP.

5. Build.

   mkdir build
   cd build
   cmake .. 
   make
   make install
   

6. Get benchmarks: download the cases here: https://drive.google.com/file/d/1HsAW_qcRje_-Ex1anWqAEQOKpGeCxpZa/view?usp=sharing. Unzip the package and put it in the following directory:

   install/benchmarks/iccad2015.hs
   

Test

Run our method on case superblue1 of ICCAD2015 timing-driven placement contest:

case=superblue1
python dreamplace/Placer.py test/iccad2015.pin2pin/$case.json

Or you can run all 8 cases by:

cd install
./run.sh

If you have already built and installed the project, you can skip the rebuild step in the helper script:

cd install
SKIP_BUILD=1 ./run.sh

The iccad2015 contest's official evaluation kit can be found at Google Drive link.

Caution

The default configuration for Critical Path Extraction uses 8 threads to accommodate various CPU cores and RAM capacities, impacting only the execution speed without affecting timing performance. For reproducing the speeds reported in the paper, adjust the thread count to 52 as specified in DATE25-TDP/thirdparty/OpenTimer/ot/timer/path.cpp at line 426.

Switch the timer to HeteroSTA

You can get a 5.7x end-to-end speedup by switching the timer to HeteroSTA.

1. Obtain a free license by visiting the website HeteroSTA, then set it as an environment variable "HeteroSTA_Lic".

2. Modify the JSON file to enable HeteroSTA by changing the following parameters in /DATE25-TDP/test/iccad2015.pin2pin/$case.json, or directly use the preconfigured files under test/iccad2015.hs/.

   JSON parameter	VALUE
   timer_engine	heterosta
   sdc_input	benchmarks/iccad2015.hs/$case/$case.hs.sdc

3. Run our method integrated with HeteroSTA on case superblue1 of ICCAD2015 timing-driven placement contest:

   case=superblue1
   cd install
   python dreamplace/Placer.py test/iccad2015.hs/$case.json
   

4. To keep the legacy behavior, configurations without an explicit timer_engine continue to use OpenTimer by default.