Hi everyone, a quick question, I am getting these ...
# openlane
a
Hi everyone, a quick question, I am getting these errors during
run_magic_spice_export,
Copy code
box 55873 307343 55907 307362
feedback add "Illegal overlap (types do not connect)" medium
box 56977 307343 57011 307362
feedback add "Illegal overlap (types do not connect)" medium
box 58081 307343 58115 307362
feedback add "Illegal overlap (types do not connect)" medium
box 60013 307343 60047 307362
feedback add "Illegal overlap (types do not connect)" medium
box 61117 307343 61151 307362
feedback add "Illegal overlap (types do not connect)" medium
box 62221 307343 62255 307362
feedback add "Illegal overlap (types do not connect)" medium
box 62588 307349 62612 307362
feedback add "Illegal overlap (types do not connect)" medium
box 211358 306763 211362 306797
There are no DRC violation during routing, but the process keeps bombing at this step. Any leads as to what could be the possible reason? I am using magic version 8.3.176.
Copy code
Extracting azadi_soc_top into azadi_soc_top.ext:
Created database crash recovery file /tmp//MAG2140.mexcXY
azadi_soc_top: 34 fatal errors
Total of 34 errors (check feedback entries).
exttospice finished.
Using technology "sky130A", version 1.0.156-0-g7e29496
[ERROR]: There are illegal overlaps (e.g., routes over obstructions) in your design.
[ERROR]: See /project/openlane/azadi_soc_top/runs/azadi_soc_top_13th_june_10pm_run/logs/magic/31-magic_ext2spice.feedback.txt for more.
p
I have exactly the same problem, did you found a solution yet?
r
I'm seeing this now as well with two macros in
user_project_wrapper
I did a better job of manual placement of the macros using the
macro.cfg
file and it finished successfully for me. In the original placement, macros were definitely needing to be routed around to get to IO at the wrapper level. Not sure if that would cause this or not ­čĄĚÔÇŹÔÖé´ŞĆ
a
@Philipp G├╝hring, the problem solved as soon as I removed the obstructions. This is bizarre though, as there were no overlaps during routing.
m
How to remove obstructions. Is it allowed to do for the fabrication?
a
I had set a
GLB_RT_OBS
parameter for metal5 layer in my config.tcl file, and commenting the variable solved the problem for me. I used
GLB_RT_MAXLAYER
instead to avoid using metal5.
m
I will also try..thank you so much
­čĹŹ 1
r
I'm unfortunately seeing this trying to generate "user_project_wrapper" Any way to debug?
m
@Russell Friesenhahn I've just debugged this with someone else
they had metal 5 in their macro
must be turned off before hardening: set ::env(GLB_RT_MAXLAYER) 5
not sure why this isn't in the example project config
probably the design is so small it doesn't use metal5
m
Yes might be not using..how to turn it off..
m
set ::env(GLB_RT_MAXLAYER) 5
r
I have very little met5 in my macro
Copy code
grep met5 routing/stc0_core.def 
TRACKS X 1700 DO 320 STEP 3400 LAYER met5 ;
TRACKS Y 1700 DO 323 STEP 3400 LAYER met5 ;
+ RECT met5 ( -710 -710 ) ( 710 710 ) 
    NEW met5 ( 403420 620500 ) ( 476100 620500 )
    NEW met5 ( 288420 596700 ) ( 552460 596700 )
    NEW met5 ( 206540 583100 ) ( 710700 583100 )
m
you need to have none I think
1st thing I would try
r
so in the macro config.tcl do
set ::env(GLB_RT_MAXLAYER) 4
?
to prevent it from using met5 at all?
m
no 5
the layers are off by 1 because of LI
r
okay, and I see now the default is 6, thanks, that could be it, must've just luckily found that trip wire on a rebuild last night that barely pushed me over the edge
m
yeah, I've just opened an issue on caravel_user_project. I think it only works there because there's so much space the router by chance never uses metal5
(my working theory)
r
no that makes sense, it won't jog up unless the cost function reaches a threshold (my coarse understanding of those algorithms)
do you know how to translate the box coordinates so that I could've seen this in the layout?
p
@Russell Friesenhahn I developed a script to convert these, but I can't find it at the moment. Basically you have to divide the coordinates by 100 I think and make rect's out of it
r
@Philipp G├╝hring I will look into that next week. I could've sworn I tried that. But now that I have a very clear idea what I'm looking for it might make more sense. Thanks, and would love to see the script if you find it.
p
I have converted them to metal5 and appended it to the .mag file and then opened the .mag file in magic
r
Thanks, I'll try that
p
Usage: perl feedback2mag.pl <feedback.txt >>layout.mag
Try this, I couldn't find the original script anymore, but that should work
r
Thanks!