The-OpenROAD-Project
CoralNPU: global place hits "FixedDelay conversion out of range: INF"
Describe the bug
When running a version of CoralNPU with fakeRAM, global place hits:
[INFO RSZ-0038] Inserted 37621 buffers in 3760 nets.
Iter | Area | Removed | Inserted | Pins
| | Buffers | Buffers | Remaining
-------------------------------------------------------
0 | +0.0% | 0 | 0 | 1377210
137700 | +0.5% | 315 | 10225 | 1239510
275400 | +0.5% | 315 | 10278 | 1101810
413100 | +0.9% | 326 | 17527 | 964110
550800 | +1.3% | 335 | 24054 | 826410
[ERROR RSZ-1008] FixedDelay conversion out of range: INF
Error: global_place.tcl, 51 RSZ-1008
Command exited with non-zero status 1
Elapsed time: 40:15.02[h:]min:sec. CPU time: user 4827.41 sys 1574.13 (265%). Peak memory: 9851352KB.
Expected Behavior
successful run or user-actionable error message
Environment
8c985e9b2ac113bf6f1dc8befd1e623a3c9194d7 - orfs git commit tag
To Reproduce
mkissue tarball: https://drive.google.com/file/d/14LQM10E0G1CAmxYXI3DHRRz_lHRfAHfI/view?usp=drive_link
Relevant log output
[INFO RSZ-0038] Inserted 37621 buffers in 3760 nets.
Iter | Area | Removed | Inserted | Pins
| | Buffers | Buffers | Remaining
-------------------------------------------------------
0 | +0.0% | 0 | 0 | 1377210
137700 | +0.5% | 315 | 10225 | 1239510
275400 | +0.5% | 315 | 10278 | 1101810
413100 | +0.9% | 326 | 17527 | 964110
550800 | +1.3% | 335 | 24054 | 826410
[ERROR RSZ-1008] FixedDelay conversion out of range: INF
Error: global_place.tcl, 51 RSZ-1008
Command exited with non-zero status 1
Screenshots
No response
Additional Context
No response
Are you using -hier?
I'm using hierarchical synthesis, but not really setting other design-specific flow variables.
@jhkim-pii is this likely to be the same as you are looking at with remove buffers and -hier or something different?
I think it is different issue. Looks like -hier option is not used in this case.
FYI. I'm also seeing failures when trying to use the new private PDK and I'm not using the -hier option. One run has the INF error and one has message with a really big number. LMK when resolved so that I can re-run.
I opened a new PR https://github.com/The-OpenROAD-Project/OpenROAD/pull/8839 that fixes multiple remove_buffer issues. The new PR changed remove_buffer behavior.
I am not sure if it can resolve this issue or not.
If you see any function or QoR issues after the new PR is merged, please think of this as one of suspects.