[ERROR GPL-0305] RePlAce diverged during gradient descent calculation - after a few seconds

[oharboe]

Describe the bug

untar and run global_place_skip_io_tag_array_64x184_asap7_base_2026-03-18_14-47.tar.gz

v2.0-29792-g38e979fbd4

after a few seconds:

     2600 |   0.1039 |  8.991379e+03 |   +0.43% |  2.12e+33 |      
[INFO GPL-1014] Final placement area: 51.09 (+0.00%)
[ERROR GPL-0305] RePlAce diverged during gradient descent calculation, resulting in an invalid step length (Inf or NaN). This is often caused by numerical instability or high placement density. Consider reducing placement density to potentially resolve the issue.
Error: global_place_skip_io.tcl, 15 GPL-0305
openroad> 

Expected Behavior

Complete

Environment

v2.0-29792-g38e979fbd4

To Reproduce

See above

Relevant log output

Screenshots

No response

Additional Context

Bump PLACE_DENSITY to 0.20 to work around

[gudeh]

We have 152 TIE cells starting at origin and moving together all the time.

[gudeh]

I am working on a branch for rsz to not set the TIEs incorrectly as placed.

On the gpl side, we already have a random offset to avoid issues similar to this, the offset is having no effect for these TIEs because they are being initiated out of the core area.

[gudeh]

Hi @oharboe, I believe recently merged PR #10074 should have fixed this issue.

I cannot reproduce with the fix because we need to run the previous stages. Would you like to check on your end with the recent update?

[oharboe]

Hi @oharboe, I believe recently merged PR #10074 should have fixed this issue.

Great!

I cannot reproduce with the fix because we need to run the previous stages. Would you like to check on your end with the recent update?

I haven't run into this recently and I don't have a record of the preceeding stages. I will report a new issue if I run into this again.