Unexpected failures of OpenROAD flow

[ishwars1618]

Recently, I have been trying to synthesize circuits through OpenROAD (with asap7 library), but some of those runs either failed or hung depending on the core area and clock period constraints (utilization was kept constant at 80%). My example runs for a netlist with 10^4 instances are shown in the table below. I have also attached the input Verilog netlist, and an example of the config.mk and constraint.sdc files I used.

Can you please explain the following observations, as shown in the table below?
(a) For area 2048, reducing clock period from 3200 to 400 results in completion.
(b) Flow completes with area 52 and clock period 400, but hangs when area reduced to 51. However, completes again with area 51 only at a lower clock period 100.

Can you please help me understand what is going on, and suggest guidelines for reliable configuration for OpenROAD flow to complete?

Thanks,
Ishwar S.

verilog netlist: tc3_1e4.v.zip
config.mk: config.mk.zip
constraint.sdc: constraint.sdc.zip

The table below says "HANG @ <some step>" to indicate that OpenROAD hangs indefinitely after printing the line "<some step> report_checks -path_delay max" for the given core area and clock period.

[maliberty]

What is ASYNC_DFFHx1_ASAP7_75t_R? it is in your .v but not in the asap7 platorm.

[ishwars1618]

The ASYNC_DFFHx1_ASAP7_75t_R cell is in the ASAP7 distribution (asap7sc7p5t_SEQ_RVT_FF_nldm_201020.lib). Looks like it could be a DFF with asynchronous reset. This Verilog was generated by the Artificial Netlist Generator program (https://github.com/daeyeon22/artificial_netlist_generator).

[maliberty]

I see this is in the older _27 rev of asap7 and was removed in _28. I'll pull it from there.

There isn't any simple answer to make everything work. We'll have to look at the issues one by one and see what they are.

[maliberty]

I started with the hang reported in floorplan at the lower-left. I can't reproduce it as the design goes through without delay for me.

Please package a test case using 'make floorplan_issue' so that it is complete.

[ishwars1618]

I tried attaching the test case as you suggested as a *.tar.gz file. But the upload failed as the file size was more than 25MB allowed. Any suggestions?

BTW, the hang usually happens after the following messages (if that gives any hint as to the issue):
floorplan final report_checks -path_delay max
global place report_checks -path_delay max

Thanks for the help.

[maliberty]

Put the file in google drive (or similar) and share a link here.

[ishwars1618]

Test case at this link: https://drive.google.com/file/d/1lkZcY8JI0vXlmPhLf5ZmRSeScx3n4RyP/view?usp=sharing

Thanks!

[maliberty]

Would you give the ORFS version you are using? (git rev-parse HEAD).

I see your OR version is from the end of Sept.

[ishwars1618]

git rev-parse HEAD gives this:

72d7253

[ishwars1618]

72d7253

[ishwars1618]

I tried giving the long string that was printed by the command you gave, but after pasting it here it became the link that is showing up.

[ishwars1618]

Hi Matt
Were you able to reproduce the hang at your end with my testcase?
Thanks,
Ishwar S.

[maliberty]

Yes and I've sent the problem along to our sta developer (where it is hanging).

[maliberty]

While waiting on 3200,2048 could you package a test for 400,51 ?

[ishwars1618]

This test case hangs in the placement step and not in the floorplanning step. Would another command be needed to package it, instead of floorplan_issue?

[maliberty]

make global_place_issue. In general it uses the name of the script from flow/scripts as <script>_issue in the target.

[ishwars1618]

Thanks for letting me know.
In that case, here's the package after using global_place_issue. https://drive.google.com/file/d/1KeTF_1oBhbdYhqbmhQMCtyt1DyZaQJDo/view?usp=sharing

[maliberty]

I unpacked this global place issue but there is no results/ directory. Had you run this case before packaging it?

[ishwars1618]

Apologies for the delay.

Here is the full package, after running the test case beforehand: https://drive.google.com/file/d/1JO0UJO-jhaprlMgKLGwnZn4Ux9njgEoY/view?usp=drive_link.

[maliberty]

The hang at 3200,2048 is resolved by The-OpenROAD-Project/OpenROAD#4358. I haven't run the full flow but floorplan finishes.

[maliberty]

I just tried both of these with OR commit 59acbf4 and neither hung.

[ishwars1618]

Thank you very much for pointing me to that. Should I use this commit or a more recent commit to rebuild OpenROAD?

Also, I am encountering a signal 11 error with some circuits. I'm wondering if this issue is related to the above or if this was resolved with another later commit.

Here is a package with an example of such an error: https://drive.google.com/file/d/1J5IR9FvRRfwJh_ctukBQx4-K0utr5yvb/view?usp=sharing.

[ishwars1618]

Hello Matt,

I tried to build OpenROAD with the commit that you pointed me to (59acbf4). However, the build failed with the following error: /home/isuriyap/OR/OpenROAD-flow-scripts/tools/OpenROAD/src/gui/src/staGuiInterface.cpp:234:38: error: no matching function for call to ‘sta::GraphDelayCalc::loadCap(sta::Pin* const&, sta::Parasitic*&, const sta::RiseFall*, sta::DcalcAnalysisPt*&)’.

I am attaching below a full log with cmake output and make output, including errors and warnings.
https://drive.google.com/file/d/1uRC-Jyb9ndl75JyQTeQ1RujXRg-hDinP/view?usp=sharing

Thanks,
Ishwar S.

[maliberty]

Most likely you need to update to update the sta submodule with git submodule update --init --recursive

[ishwars1618]

I rebuilt OpenROAD with a more recent commit, and the problem I had before is resolved. Thank you for the fix and the suggestion.