@User After GDS write completed (over 11 hours of total flow time) and during LEF generation, magic exited with a

kill signal

. This occurred in the docker and the magic version is 8.3.87. I've provided the context of the log below. If there's anything else I can provide (or you want me to create a GitHub issue), let me know.

Writing cell sky130_fd_sc_hd__buf_12

[INFO]: GDS Write Complete

[INFO]: Writing abstract LEF

Generating LEF output /openLANE_flow/designs/decred_nonblock/runs/8mFP50DENS50STRAT1_20ns/results/magic/decred_nonblock.lef for cell decred_nonblock:

Diagnostic:  Write LEF header for cell decred_nonblock

Diagnostic:  Writing LEF output for cell decred_nonblock

[ERROR]: during executing: "magic -noconsole -dnull -rcfile /home/dracex/PDK_ROOT/sky130A/libs.tech/magic/sky130A.magicrc /openLANE_flow/scripts/magic.tcl </dev/null |& tee >&@stdout /openLANE_flow/designs/decred_nonblock/runs/8mFP50DENS50STRAT1_20ns/logs/magic/magic.log"

[ERROR]: Exit code: 1

[ERROR]: Last 10 lines:

child killed: kill signal

[ERROR]: Please check magic  log file

[ERROR]: Dumping to /openLANE_flow/designs/decred_nonblock/runs/8mFP50DENS50STRAT1_20ns/error.log

while executing

"try_catch magic  -noconsole  -dnull  -rcfile $::env(MAGIC_MAGICRC)  $::env(SCRIPTS_DIR)/magic.tcl  </dev/null  |& tee $::env(TERMINAL_OUTPUT) $::env(m..."

(procedure "run_magic" line 14)

invoked from within

"run_magic"

(procedure "run_non_interactive_mode" line 27)

invoked from within

"run_non_interactive_mode {*}$argv"

invoked from within

"if { [info exists flags_map(-interactive)] || [info exists flags_map(-it)] } {

puts_info "Running interactively"

if { [info exists arg_values(-file)..."

(file "./flow.tcl" line 168)

How large is your design? It seems like it ran out of memory and swap since it received a kill signal?

yosys_1.stat.rpt shows number of cells at 377k. Since then I have tried a smaller version of the design at 283k cells and it still fails with the same error. The VM I'm running in has 16G of RAM.

@Ahmed Ghazy: Openlane is using "cif *hier write disable" for all digital standard cell layouts? Because assuming so, the GDS output should be very fast and very low overhead.

@Tim Edwards: Yes, that's the default for std cell layouts. @Matt Aamold: Are you using the latest on the develop branch?

@Ahmed Ghazy I'm using develop branch of openlane as of Nov 30. Have changes been made since then that you think would resolve this?

Not really; I only added

cif *array write disable

besides the existing https://github.com/efabless/openlane/blob/e63096fec826f1bad03700a26cb6c6abb6a7c5c7/scripts/magic.tcl#L75.

@Ahmed Ghazy: "cif *array" will do nothing because openlane doesn't generate arrays. Otherwise, the script looks okay so we'll have to try to duplicate the problem and see what's causing it.

@Ahmed Ghazy Sure, could you point me to where these files would be in the "runs" directory (e.g., results/magic/*)?

results/routing/design_name.def

and

tmp/merged.lef

. You can open it on openlane's GH repo for now.

What's the best way to share a large file (DEF) that is about 462MB? The LEF is only about 2MB.

Yea, that sounds good.

@Ahmed Ghazy https://github.com/efabless/openlane/issues/119

@Matt Aamold: It took around 30 mins to generate both LEF and GDS using magic

8.3.91

, which will be the default in a bit in openlane. It used around 9 GB of memory most of the time. You don't have to re-run the whole flow if you want to retry this; you could use the interactive mode

flow.tcl -interactive -design decred_nonblock -tag 8mFP50DENS50STRAT1_20ns

and then use

run_magic

.

I'll try that. Thanks for looking into this.