Underflow error after antenna repair

[gadfort]

Describe the bug

I'm getting a [ERROR GRT-0183] Net 22574: heap underflow during 3D maze routing.
After antenna repair

Expected Behavior

No error.

Environment

26Q2-555-gd71446d6e9
OpenROAD 26Q2-258-g45a537e4a3

To Reproduce

sc_issue_tiny_rocket_ihp130_sg13g2_stdcell_1p2_route.antenna_repair_0_20260427-095744.tar.gz

tar xvf sc_issue_tiny_rocket_ihp130_sg13g2_stdcell_1p2_route.antenna_repair_0_20260427-095744.tar.gz
cd sc_issue_tiny_rocket_ihp130_sg13g2_stdcell_1p2_route.antenna_repair_0_20260427-095744
./run.sh

Relevant log output

[INFO ANT-0002] Found 103 net violations.
[INFO ANT-0001] Found 153 pin violations.
Starting antenna repair with sg13g2_antennanp cell
[INFO GRT-0012] Found 103 antenna violations.
[INFO GRT-0302] Inserted 56 jumpers for 53 nets.
[INFO GRT-0012] Found 51 antenna violations.
[INFO DPL-0006] Core area: 1292215.68 um^2, Instances area: 580369.96 um^2, Utilization: 44.9%
[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
[INFO DPL-1101] Legalizing using diamond search.
Movements Summary
---------------------------------------
Total cells:                   33333
Diamond Move Success:          33333 (100.00%)
Diamond Move Failure:              0
Rip-up and replace Success:        0 (  0.00% of diamond failures)
Rip-up and replace Failure:        0
Total Placement Failures:          0
---------------------------------------
[INFO GRT-0015] Inserted 154 diodes.
[ERROR GRT-0183] Net _22574_: heap underflow during 3D maze routing.
Error: sc_antenna_repair.tcl, 44 GRT-0183

Screenshots

No response

Additional Context

No response

[jfgava]

It looks fine now. The issue was related to incorrect resource initialization when running repair_antennas.
I'm finishing the PR

[jfgava]

PR merged.
@gadfort let me know if you find any issues.

[mole99]

Could this also be related to this issue here, or was it design specific? #10156

[jfgava]

Could this also be related to this issue here, or was it design specific? #10156

Yes, it looks the same. The problem happened when calling repair_antennas with Fast Route uninitialized.

[gadfort]

@jfgava with this fix I'm getting overflow errors now on different designs:

caliptra_sha512-skywater130_sky130hd-issue.zip

[jfgava]

@jfgava with this fix I'm getting overflow errors now on different designs:

caliptra_sha512-skywater130_sky130hd-issue.zip

Interesting. In this new case, we already start with 2 overflows. I'll investigate.
I fixed another issue related to wrong routing resources a few days ago.
Is the input ODB updated?
Are you getting overflow in any GRT test? Or just related to Antenna Repair?

[gadfort]

@jfgava this is a different design that started to fail with this change, the CI I run, stops on first failure, so this is the only one I saw.

[jfgava]

@jfgava this is a different design that started to fail with this change, the CI I run, stops on first failure, so this is the only one I saw.

Got it. I suspect the problem is occuring before the repair antennas stage.
Can you provide a reproducible from an earlier stage?
I'm not sure what is your flow. In ORFS, we run global_route -> repair_design -> repair_timing -> repair_antennas
I think the problem might be in incremental routing during repair_timing

[gadfort]

@jfgava I don't have access to the ones before the failure on github. Is the issue that the information in ODB is wrong and therefore you need the database earlier? I think it's just global route to antenna repair, so no timing repair after global route