https://open-source-silicon.dev logo
Channels
Powered by
m

mehdi

07/14/2020, 10:59 PM
@User please use this: https://github.com/msaligane/OpenROAD-flow/tree/openroad/flow/platforms/sw130 I setup the hs lib for you and just realized I made "custom" changes to get the flow running. But this should work for the new libs as is (I did not include finishing steps).
u

20Mhz

07/15/2020, 3:44 AM
do you know if repair_clock_inverters a native command? my binary doens’t seem to have it…
m

mehdi

07/15/2020, 3:45 AM
which branch of the tools submodules are you using?
u

20Mhz

07/15/2020, 3:47 AM
i cloned the original repo, the master branch for building/installing
m

mehdi

07/15/2020, 3:48 AM
use the openroad branch. And when you are building use: ./build_openroad.sh -d -p -l to build the latest version of the tools
u

20Mhz

07/15/2020, 3:54 AM
ok, so this one is new? is there one stop shop for UI doc?… I’m still familiarizing with the whole suite…
m

mehdi

07/15/2020, 3:54 AM
When you run the build_openroad use --latest. It pulls the head of all submodules.
No worries, yes we need to update the doc (this was added recently).
👍 1
u

20Mhz

07/15/2020, 5:02 AM
getting some odd errors during global_place.tcl 
[PROC] Begin InitialPlace ...
[InitialPlace]  Iter: 1 CG Error: 1.08983e-07 HPWL: 18309304
[InitialPlace]  Iter: 2 CG Error: 4.8921e-08 HPWL: 14469696
[InitialPlace]  Iter: 3 CG Error: 9.21233e-08 HPWL: 14452772
[InitialPlace]  Iter: 4 CG Error: 9.02186e-08 HPWL: 14453353
[InitialPlace]  Iter: 5 CG Error: 9.43531e-08 HPWL: 14464670
bails out at the end: 
[INFO]   NewStepLength = 30.916
[INFO]   NumBackTrak = 11
[ERROR] RePlAce divergence detected.
Please decrease init_density_penalty value (REPL-3)
any clues?
I’m simply trying to build the gcd block
m

mehdi

07/15/2020, 11:40 AM
Did you change the size of gcd?
u

20Mhz

07/15/2020, 12:20 PM
only to try to fix this, do you have any insight on what it means?
changing init_density_penalty doens’t seem to help
m

mehdi

07/15/2020, 12:25 PM
My experience with the tiny designs is that the placer diverges when the core area is too small. Can you try increasing it?
u

20Mhz

07/15/2020, 12:31 PM
I’m setting it by utilization, currently at 60%
Running Global Placement
[INFO] DBU = 1000
[INFO] SiteSize = (480, 3330)
[INFO] CoreAreaLxLy = (19680, 19980)
[INFO] CoreAreaUxUy = (90240, 89910)
[INFO] NumInstances = 292
[INFO] NumPlaceInstances = 227
[INFO] NumFixedInstances = 65
[INFO] NumDummyInstances = 0
[INFO] NumNets = 263
[INFO] NumPins = 895
[INFO] DieAreaLxLy = (0, 0)
[INFO] DieAreaUxUy = (110450, 110450)
[INFO] CoreAreaLxLy = (19680, 19980)
[INFO] CoreAreaUxUy = (90240, 89910)
[INFO] CoreArea = 4934260800
[INFO] NonPlaceInstsArea = 305294400
[INFO] PlaceInstsArea = 2977819200
[INFO] Util(%) = 64.330109
[INFO] StdInstsArea = 2977819200
[INFO] MacroInstsArea = 0
do you know if this is off by default? how to enable?
Copy code
#ifdef ENABLE_CIMG_LIB
    pe.SaveCellPlotAsJPEG(
        string("InitPlace ") + to_string(i), false,
        string("./plot/cell/ip_") + to_string(i));
#endif
mage to build, results unexpected xD
Copy code
InitialPlace]  Iter: 1 CG Error: 6.41183e-08 HPWL: 8071794

[CImg] *** CImgIOException *** [instance(1060,1060,1,3,0x7fe8fb672000,non-shared)] CImg<unsigned char>::save_other(): Failed to save file './plot/cell/ip_1.jpg'. Format is not natively supported, and no external commands succeeded.
libc++abi.dylib: terminating with uncaught exception of type cimg_library::CImgIOException: [instance(1060,1060,1,3,0x7fe8fb672000,non-shared)] CImg<unsigned char>::save_other(): Failed to save file './plot/cell/ip_1.jpg'. Format is not natively supported, and no external commands succeeded.
/bin/bash: line 1: 69880 Abort trap: 6           openroad -no_init -exit ./scripts/global_place.tcl
m

mehdi

07/15/2020, 1:02 PM
@20Mhz sorry I don't see all these errors. I can take a look after my morning meetings. I think the previous cmd is for visuals.
u

20Mhz

07/15/2020, 1:04 PM
sure, my though was to enable the plots to get more insight on the convergence problem, I manage to rebuild, but CIMG code bailed out, I’m guessing there is some sort of version issue. need to shift to work myself now, to be continued, thanks
I was missing -D cimg_use_jpeg flag. now works, will see
m

mehdi

07/15/2020, 1:23 PM
It generates really cool visuals on larger designs usually 👌
u

20Mhz

07/15/2020, 1:25 PM
looking odd?
thats just after the first failure
Copy code
[InitialPlace]  Iter: 1 CG Error: 6.41183e-08 HPWL: 8071794
INFO: ./plot/cell/ip_1.jpg image has been saved
m

mehdi

07/15/2020, 1:31 PM
As I mentioned the placer diverges on tiny area butnI think this issue was fixed. I don't see this divergence when I increase the core size. I will check later and get back to you. Can you try a larger design? Aes or jpeg?
u

20Mhz

07/15/2020, 1:31 PM
ok
found it, I had accidentally removed the density option, at least now I have pretty screenshots :)
m

mehdi

07/15/2020, 2:12 PM
😅 great! What are you using them for?
u

20Mhz

07/15/2020, 2:20 PM
I’m just testing, what I meant is that I created the screenshot for debugging, but at the end It was just a typo xD
m

mehdi

07/15/2020, 2:23 PM
BTW, what is the change you made to get the visuals?
u

20Mhz

07/15/2020, 2:27 PM
OpenROAD-flow/tools/build/OpenROAD/src/replace/CMakeFiles/replace.dir/flags.make
Copy code
CXX_DEFINES = -DOPENROAD_BUILD -D ENABLE_CIMG_LIB -D cimg_use_jpeg
CXX_INCLUDES = -I /usr/local//opt/libjpeg/include -I /opt//X11/include/ -I/Users/ronaldv/Projects/repositories/CImg ...
✅ 1
@mkk
Powered by