timing_driven global placement is getting killed always

[NAGENDRA-ARASAN-K]

After the macro placement I am running global_placement -timing_driven but it is running for long time and after it is getting killed (even after trying many switching options) and exitting from the openroad. I need to stack trace it. Is there any debugger for the same? As for my knowledge there is something called openroad workbench as the debugger "https://docs.actian.com/openroad/11.1/index.html#page/Programming/Debugging_Your_Application.htm". But nothing is mentioned in the openroad documentation about the debugger neither there is no documentation available for the debugger too.
Or is there any other ways to perform the timing_driven so that it won't get killed??

[maliberty]

My first guess when a process is being killed is that you are running out of memory. Please check if you have sufficient resources to run the job.

[NAGENDRA-ARASAN-K]

Yeah you are right, it is been -OOM killed. (Out of Memory).
But I am allocating 120G mem to run it. This is my command to allocate resources salloc --x11 --gpus=1 --cpus-per-task=16 --mem=120G --nodes=1 --time=12:00:00 --hint=compute_bound --cpu-freq=high /bin/bash. Does 120G not sufficient? And also I tried increasing the no of gpus to 3 and mem to 180G too How to prevent it?

[maliberty]

It sounds sufficient but I also don't know what size design you are running. How many placeable instances?

[NAGENDRA-ARASAN-K]

I am using around 2400000 std cells and 60 locked macros

[maliberty]

180G seems excessive even for that sized design. How much memory does it take without timing_driven?

[NAGENDRA-ARASAN-K]

I tried for routability driven I didn't check the exact memory but i gave only 60G.
And sometimes it is killed even during rtl_macro_placer and throws this error
Signal 11 received
Stack trace:
0# 0x000000000107724D in openroad
1# 0x00007FC7B123E6F0 in /lib64/libc.so.6
2# odb::dbIntProperty::getValue() in openroad
3# mpl::ClusteringEngine::computeHypergraph(int) in openroad
4# mpl::ClusteringEngine::createDataFlow() in openroad
5# mpl::ClusteringEngine::run() in openroad
6# mpl::HierRTLMP::runMultilevelAutoclustering() in openroad
7# mpl::HierRTLMP::run() in openroad
8# mpl::MacroPlacer::place(int, int, int, int, int, float, int, float, int, int, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, char const*) in openroad
9# mpl::rtl_macro_placer_cmd(int, int, int, int, float, int, float, int, int, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, float, char const*) in openroad
10# 0x0000000001A95860 in openroad
11# TclNRRunCallbacks in /lib64/libtcl8.6.so
12# 0x00007FC7B552B0AA in /lib64/libtcl8.6.so
13# Tcl_EvalEx in /lib64/libtcl8.6.so
14# Tcl_Eval in /lib64/libtcl8.6.so
15# 0x00000000010766EA in openroad
16# Tcl_MainEx in /lib64/libtcl8.6.so
17# main in openroad
18# 0x00007FC7B1229590 in /lib64/libc.so.6
19# __libc_start_main in /lib64/libc.so.6
20# _start in openroad
Segmentation fault (core dumped)
timed out waiting for input: auto-logout

[maliberty]

That stack isn't in global placement, its in macro placement. Can you provide a test case?

[NAGENDRA-ARASAN-K]

Test case?
I didn't get you. And I wanted you to ask why that is erroring out in macro placement?
For global placement
I tried for routability driven I didn't check the exact memory but i gave only 60G.

[maliberty]

Can you provide the data you are running that demonstrates the issue?

[NAGENDRA-ARASAN-K]

Sorry sir, my company doesn't allows to share the data

[maliberty]

There isn't enough info to see what the issue is.

[maliberty]

Professional support under NDA is available from https://precisioninno.com/

[NAGENDRA-ARASAN-K]

ok thank you just now noticed.

[NAGENDRA-ARASAN-K]

Hi and there is also another issue during placement process some of the instances are out of the die area and it is detected during the global_route stage only. Is there any way to rectify it?

[maliberty]

The image is broken