PDNGEN crashes - Githubissues

macd commented 1 year ago

Describe the bug

When running the following script in the OpenROAD/src/pdn/test directory, it segfaults (OOM)

Expected Behavior

No crashes.

Environment

macd@macd-NUC9:~/road/OpenROAD$ ./etc/Env.sh
Git commit: cf9e3b282573a32a7ae3678cd97074a5697276ae
kernel: Linux 5.15.0-56-generic
os: Ubuntu 22.04.1 LTS (Jammy Jellyfish)
cmake version 3.22.1
gcc (Ubuntu 11.3.0-1ubuntu1~22.04) 11.3.0
Ubuntu clang version 14.0.0-1ubuntu1

To Reproduce

run the following script in the OpenROAD/src/pdn/test directory

source "helpers.tcl"
read_lef Nangate45/Nangate45_tech.lef
read_lef Nangate45/Nangate45_stdcell.lef
read_def nangate_gcd/floorplan.def
add_global_connection -net VDD -pin_pattern VDD -power
add_global_connection -net VSS -pin_pattern VSS -ground
set pwr [[ord::get_db_block] findNet "VDD"]
set gnd [[ord::get_db_block] findNet "VSS"]
set region "NULL"
set name "NULL"
set secondary {}
set switched_power "NULL"
pdn::make_region_domain $name $pwr $switched_power $gnd $secondary $region
define_pdn_grid -name "Core"
add_pdn_stripe -followpins -layer metal1

Relevant log output

openroad> source "helpers.tcl"
read_lef Nangate45/Nangate45_tech.lef
read_lef Nangate45/Nangate45_stdcell.lef
read_def nangate_gcd/floorplan.def
add_global_connection -net VDD -pin_pattern VDD -power
add_global_connection -net VSS -pin_pattern VSS -ground
set pwr [[ord::get_db_block] findNet "VDD"]
set gnd [[ord::get_db_block] findNet "VSS"]
set region "NULL"
set name "NULL"
set secondary {}
set switched_power "NULL"
pdn::make_region_domain $name $pwr $switched_power $gnd $secondary $region
define_pdn_grid -name "Core"
add_pdn_stripe -followpins -layer metal1
[INFO ODB-0222] Reading LEF file: Nangate45/Nangate45_tech.lef
[INFO ODB-0223]     Created 22 technology layers
[INFO ODB-0224]     Created 27 technology vias
[INFO ODB-0226] Finished LEF file:  Nangate45/Nangate45_tech.lef
[INFO ODB-0222] Reading LEF file: Nangate45/Nangate45_stdcell.lef
[INFO ODB-0225]     Created 135 library cells
[INFO ODB-0226] Finished LEF file:  Nangate45/Nangate45_stdcell.lef
[INFO ODB-0128] Design: gcd
[INFO ODB-0130]     Created 54 pins.
[INFO ODB-0131]     Created 482 components and 2074 component-terminals.
[INFO ODB-0133]     Created 385 nets and 1110 connections.
[INFO PDN-0101] Using VDD as power net for Core domain.
[INFO PDN-0102] Using VSS as ground net for Core domain.
Segmentation fault (core dumped)

Screenshots

No response

Additional Context

Exposure of this to end users is maybe minimal, but roughly what happens is that a voltage domain is defined before a core domain. Then when one of the commands does a findDomain("Core") (which is in PdnGen.cc) the very first thing it does is to call "ensureCoreDomain()" which winds up adding an new superfluous domain (because coredomain is not set) Then when getDomainRows, from domain.cpp, is called it hits an finite recursion between getRows and getDomainRows. I guess the fix would be either to 1) disallow creating a voltage domain before the core domain, or 2) put in a call to ensureCoreDomain() as the very first thing in makeRegionVoltageDomain, (in PdnGen.cc), but honestly, I don't know this code well enough to do much more.

gadfort commented 1 year ago

@macd why are you directly calling the make_region_domain? I can add a check that the region is NULL, but you should not be calling them directly in TCL.