The-OpenROAD-Project
Gui gets stuck after link_design
Describe the bug
After running the test case provided in #5340, after link_design finishes running and sta tries to report, the gui doesn't show nothing in scripting and keeps "running".
Expected Behavior
The report should be visible in Scripting and the GUI should be unpaused.
Environment
HEAD of Master
To Reproduce
openroad -gui test.tcl with the test case from the issue mentioned in the description.
Relevant log output
No response
Screenshots
No response
Additional Context
No response
I tried this in text mode and it is the same. It appears that link_design just takes a long time and it has nothing to do with the GUI. How large is this design?
There is something odd about this design. There are huge numbers of high fanout nets. I suspect you have some sort of connectivity problem in this design. Some examples which repeat for many different values of the integers:
bp_processor/coh_resp_hor_link_li_2__0__data__43_ has 8586 pins Net bp_processor/coh_req_ver_link_li[507] has 5716 pins Net bp_processor/cc/lce_cmd_link_li[1054] has 14280 pins
I run the following script:
# Setting lef files
set tech_dir "/scripts/MacroPlacement/Enablements/NanGate45"
set tech_lef "${tech_dir}/lef/NangateOpenCellLibrary.tech.lef"
set std_lef "${tech_dir}/lef/NangateOpenCellLibrary.macro.mod.lef"
set lefs "
./fakeram45_256x16.lef
"
# Setting lib files
set libs "
${tech_dir}/lib/NangateOpenCellLibrary_typical.lib \
./fakeram45_256x16.lib \
"
read_lef $tech_lef
read_lef $std_lef
foreach lef_file ${lefs} {
read_lef $lef_file
}
foreach lib_file ${libs} {
read_liberty $lib_file
}
puts "read ariane_fp.def"
read_verilog ./ariane.v
link_design ariane
read_sdc ./ariane.sdc
read_def -floorplan_initialize ./ariane_fp.def
# Liberty units are fF,kOhm
set_layer_rc -layer metal1 -resistance 5.4286e-03 -capacitance 7.41819E-02
set_layer_rc -layer metal2 -resistance 3.5714e-03 -capacitance 6.74606E-02
set_layer_rc -layer metal3 -resistance 3.5714e-03 -capacitance 8.88758E-02
set_layer_rc -layer metal4 -resistance 1.5000e-03 -capacitance 1.07121E-01
set_layer_rc -layer metal5 -resistance 1.5000e-03 -capacitance 1.08964E-01
set_layer_rc -layer metal6 -resistance 1.5000e-03 -capacitance 1.02044E-01
set_layer_rc -layer metal7 -resistance 1.8750e-04 -capacitance 1.10436E-01
set_layer_rc -layer metal8 -resistance 1.8750e-04 -capacitance 9.69714E-02
# No calibration data available for metal9 and metal10
#set_layer_rc -layer metal9 -resistance 3.7500e-05 -capacitance 3.6864e-02
#set_layer_rc -layer metal10 -resistance 3.7500e-05 -capacitance 2.8042e-02
set_wire_rc -signal -layer metal3
set_wire_rc -clock -layer metal5
The process gets stuck at the link_design ariane step when I start OpenROAD with the -gui option. However, if I start OpenROAD from the shell without the -gui option, it does not hang at that step.
As a temporary workaround, I execute the link_design command in shell and then invoke gui::show to bypass the issue. Nonetheless, I believe this is a bug that needs to be resolved.
For debugging purposes, I've uploaded my run script and related code here: https://github.com/shifengzhicheng/hierarchical-mpl/tree/5e4a74494c7337b4ad79356a0baa481ec926d190/log/Issue_20250326
What version of OR are you testing with? "HEAD of Master" is not descriptive at it changes constantly.
I'm using the master branch after merging commit f7f634f886bb446dc12c5d254105ba65a6b90340.
I retested with be6d1c688c219e29569c2fffa8cf60b95d9fbd12 from March 18th and the issue is still here. Without GUI (all print iterations and durations printed in terminal console) - 100 print iterations take 1 ms - link_design takes 3 mn - 100 print iterations take 1 ms With GUI: - 100 print iterations take 6 ms (all printed in terminal console with final duration) - Only "Iteration 0" printed in GUI console - Then a popup appears about "OpenROAD not responding", clicking "wait" makes it disappear but the popu reappears 2 s later - link_desing actually finish with a duration of 5 mn (printed in the terminal console) - "Iteration 1" appears in GUI console - Then nothing more happens
I found it takes 19 minutes for me w/o the GUI. I don't think machine speed would explain it taking 3min for you. Are we running the same data?
Yes indeed and my setup is not particularly powerful (VM using 8 cores on a HP EliteBook G9). I used test case https://github.com/user-attachments/files/16179674/slow.zip from https://github.com/The-OpenROAD-Project/OpenROAD/issues/5340.