Closed gudeh closed 1 week ago
I can't reproduce the issue. The valgrind report runs an older version of the OpenROAD, where repair_antennas was dependant on orderWires. With the new antenna checker, we don't have orderWires calls, and Valgrind doesn't report memory issues.
Describe the bug
The issue is happening currently on master. I noticed this on a previous PR of mine, when I updated some metrics from a run I did locally on my PC, and updated the results on /tools/OpenROAD/test. After doing so, the metrics were still failing on jenkins (but successful locally).
The memory bug generates a non-determinism to some of the tests. The first divergence in the logs happens on GRT-0054 with the number of diodes,as can be seen on the printscreen, left is jenkins:![image](https://github.com/The-OpenROAD-Project/OpenROAD/assets/37420320/806b7ced-5266-4f85-8d5d-f835e9b3c6eb)
I was able to reproduce this bug by comparing my local run of
/tools/OpenROAD$ ./test/regression jpeg_sky130hd.tcl
versus the log file I downloaded from jenkins:OpenROAD-Private <3464>
.Here is a print showing the commit hash locally and from jenkins:![image](https://github.com/The-OpenROAD-Project/OpenROAD/assets/37420320/f6129ec3-84b6-4a32-ad9a-1e3c6c811249)
I also ran the test with Valgrind on a previous PR of mine, search for "invalid" keyword to find the memory bug with Valgrind. Available here: https://drive.google.com/file/d/1-D9rB7HQGah5yzKJSX6hVrYazpc5-3k_/view?usp=sharing
Expected Behavior
No difference from test ran locally versus available in jenkins.
Environment
To Reproduce
cd tools/OpenROAD ./test/regression jpeg_sky130hd
Relevant log output
No response
Screenshots
No response
Additional Context
No response