Incremental GPL can't properly handle the cells inserted by `replace_arith_modules`

[AcKoucher]

Description

When running Secure-CI for #9465, ngt45/swerv_wrapper failed during GRT due unsolved congestion:

[ERROR GRT-0116] Global routing finished with congestion. Check the congestion regions in the DRC Viewer

The cause of the failure seems to be a dense blob of cells in the middle of the design (in the image, Routing Congestion heat map for 5_1_grt-failed.odb):

The real problem looks to be that during the global_placement -incremental call that happens right after replace_arith_modules at 3_4_place_resized sub-stage, GPL puts nearly 800 of the new logic cells in the same place:

These end up being spread apart subsequently by the legalizer, forming the blob we see in the first image:

.odb after replace_arith_modules	.odb after global_placement -incremental	3_5_place_dp.odb

Suggested Solution

GPL should handle the cells inserted by replace_arith_modules in a better way.

Additional Context

Here's the link for the design's artifact.

[AcKoucher]

@precisionmoon @povik FYI.

[eder-matheus]

It's worth mentioning that the ~800 new cells are not placed before the global_placement -incremental call.