search

limbo018 / DREAMPlace

Deep learning toolkit-enabled VLSI placement
BSD 3-Clause "New" or "Revised" License
705 stars 204 forks source link

RUDY implementation #75

Closed HAOHANZHOU closed 5 months ago

HAOHANZHOU commented 2 years ago

Hi, does dreamplace/ops/rudy implements RUDY metric mentioned in paper: Fast and Accurate Routing Demand Estimation for Efficient Routability-driven Placement https://ieeexplore.ieee.org/document/4211973 It seems there is no horizontal or vertical congestion concept mentioned in article, but I found them in implementation. rudy

limbo018 commented 2 years ago

Hi, the RUDY implementation is more based on the following paper. Cheng, Chih-Liang Eric. "RISA: Accurate and efficient placement routability modeling." Proceedings of the 1994 IEEE/ACM international conference on Computer-aided design. 1994.

You are very welcome to contribute new OPs and extend the routability estimation.

YihangQiu commented 2 years ago

Hi! Prof. Lin.

I don't know how to eliminate the randomness caused by the atomicAdd() function in _rudy_cudakernel.cu. I would like the program to run twice with the same result when I use the route utilization map generated by rudy.

Thanks for your reply! image

limbo018 commented 2 years ago

Hi, there are commits to make rudy deterministic in another branch, which is an ongoing task. I can pick the commits to master branch in today or tomorrow.

limbo018 commented 2 years ago

Hi, please checkout develop branch for this commit (https://github.com/limbo018/DREAMPlace/commit/5199a626314e7e390c593ffa248efe633b37827f). We have added a deterministic_flag to RUDY as well. Remember to turn deterministic_flag to 1 in json.

YihangQiu commented 2 years ago

Okay, thank you Prof. Lin.

ybiao-he commented 1 year ago

Hi Dr. Lin,

I just went through the paper (Cheng, Chih-Liang Eric. "RISA: Accurate and efficient placement routability modeling." Proceedings of the 1994 IEEE/ACM international conference on Computer-aided design. 1994.). It seems that they consider the existing wire and blockage while computing the congestion value, they should reduce the routing resource supply, but in the implementation, I did not find any information of existing wires and blockage in the input of RUDY function/class. Are the existing wires and blockages considered in this implementation?

limbo018 commented 1 year ago

Fixed macros are handled by initial_horizontal_utilization_map and initial_vertical_utilization_map in our implementation. If you trace these variables, they are initialized in place_io.

ybiao-he commented 1 year ago

i.c. Thanks very much!