I had two problems while processing global placement of RePlAce. When I ran the global placement(GP) with the benchmark of ISPD 2015 contest, there were two benchmarks that could not process the GP stage smoothly. ( mgc_superblue16_a and mgc_superblue19 ) *benchmark of ISPD 2015 : https://www.ispd.cc/contests/15/web/downloads.html (Hidden benchmark designs ) I processed the following commands in my tcl file: read_lef {tech.lef} read_lef ${cells.lef} read_def ${floorplan.def} global_placement -density 1.0 After I processed benchmark of mgc_superblue19, the last 2 rows of error messages are: [NesterovSolve] Iter: 680 overflow: 0.18357 HPWL: 14598922301 [ERROR GPL-0307] RePlAce divergence detected. Re-run with a smaller max_phi_cof value. When I try smaller max_phi_cof value ( 1.1,1.3) with following command: global_placement -density 1.0 -max_phi_coef 1.03 There still had error message: [NesterovSolve] Iter: 680 overflow: 0.18357 HPWL: 14598922301 [ERROR GPL-0307] RePlAce divergence detected. Re-run with a smaller max_phi_cof value. And I further try smaller max_phi_cof global_placement -density 1.0 -max_phi_coef 1.01 The error will be solved, but the overflow will not decrease less than 0.1 after 5000 iterations of NesterovSolve. The last 2 rows of messages are: [NesterovSolve] Iter: 4990 overflow: 0.362241 HPWL: 12902711426 [NesterovSolve] Iter: 5000 overflow: 0.362278 HPWL: 12902395788 Another problem is that after I processing mgc_superblue16_a, the error message showed that : [ERROR GPL-0305] RePlAce diverged at newStepLength. But I have no idea how to solve this problem. Please help me, thanks!

The example verilog project for tinytapeout 2 has the same issue. If you were using OpenLane, the suggested fix for small diverging designs is setting PL_RANDOM_INITIAL_PLACEMENT, which generates a random initial placement in a python script. Maybe you could do something similar?

Because I'm using openroad's RePlAce stage, I think it can't work in my case. But thanks for your advise!