citypw has quit [Ping timeout: 250 seconds]
citypw has joined #symbiflow
citypw has quit [Ping timeout: 250 seconds]
<digshadow> nats`: have some more time now. Let me know if you still have any qs
<digshadow> litghost is a better contact for 07x series, but I can try to answer qs
citypw has joined #symbiflow
citypw has quit [Ping timeout: 258 seconds]
<nats`> I'll write the splitted version this evening to test it I think it could work correctly and I'll also try with 2018 to see if the leak is still here
<digshadow> nats`: FYI we are locked to 2017.2 unless something really critical comes pu
<digshadow> up
storm_ has quit [Quit: Page closed]
<nats`> yep remember that :)
<nats`> but I wanted to see if I can run it on 2018
<nats`> just you know to know :D
sxpert has quit [Ping timeout: 250 seconds]
sxpert has joined #symbiflow
igorangst has joined #symbiflow
<digshadow> litghost: I'm fixing / reviewing IOB_SIGN issue I saw a while back
<digshadow> tentatively, looks like this will result in a *negative tilegrid offset*
<digshadow> do you have any fundamental objections to this?
<digshadow> the issue is that only half of the normal tile is present, both as displayed in the GUi and in the bitstream
<digshadow> as the normal tile is 4 words high, but we only have the Y1 case
<digshadow> I fully expect this to at least break some of our tools on prints
igorangst has quit [Ping timeout: 258 seconds]
igorangst has joined #symbiflow
efg has joined #symbiflow
igorangst has quit [Ping timeout: 246 seconds]
<litghost> > tentatively, looks like this will result in a *negative tilegrid offset* -> Why not base the tile at the lower address, and shift everybody else up?
<litghost> Ultimately I don't actually care about a negative offset, as long as fasm2frames.py does the right thing, and xc7patch accepts the input
<litghost> If you can round trip (bit -> fasm -> bit -> fasm) with no errors, than I'm good
<digshadow> litghost: its the word offset, not the tile address
<digshadow> we're out of things to go back to
<litghost> A negative word offset just causes the frame address to drop below the base address, so shifting the base address down would resolve that.
<digshadow> litghost: hmm not sure I agree with that
<digshadow> it would at the very least introduce a lot more special cases than it would solve
<digshadow> in any case, I just tested with segprint and the iob fuzzer
<digshadow> and both seemed happy with it
<digshadow> I'll put the details in the PR for you to review
<digshadow> push back if you think it will be problematic
<tpb> Title: Iob sign by mcmasterg · Pull Request #488 · SymbiFlow/prjxray · GitHub (at github.com)
swetland has joined #symbiflow
Rahix has joined #symbiflow
<nats`> litghost digshadow I'm starting testing my parallel 072 fuzzer
<nats`> it seems I have something working
<nats`> I'll run it long time to test if someone is motivated I can make a zip of that dirty code and you can try
igorangst has joined #symbiflow
<nats`> https://i.imgur.com/cS6s7YI.png <= screencap of fuzzer running :)
<nats`> https://i.imgur.com/nxduqxW.png <= change of block seems to help !
<nats`> maybe it's just a feeling but it seems to go faster too
<nats`> seems 1/3rd in 10 minutes oO
<nats`> there is a problem I think
Rahix has quit [Quit: Rahix]
<nats`> my fuzzer ended work in 30 minutes....
<nats`> I'm not sure it's good :p
<nats`> I get 2.7GB downhill file and a 3.6GB uphill file
<nats`> is it ok ?
<nats`> anyway if it's good it means we can generate fuzzer 072 in 30 minutes with not eating more than 8GB
<nats`> and it can be even more optimized with more parallel block if more ram is available
<litghost> nats: Sounds great
<litghost> That size looks about right
<nats`> litghost, how could we check diff between results file ?
<nats`> do you have them ?
<nats`> I could download them pretty fast I have FFTH
<nats`> FTTH
<litghost> Assuming the files were emitted in order, lets start with sha1sum
<nats`> ahhhhhh you know you're smart ? :D
<litghost> If we are lucky, they'll just match
* nats` dumb
<nats`> anyway where can I get original one because I can't run 072 fuzzer on my computer
<litghost> I was planning to just run it on my machine
<nats`> someone must have them no ?
<nats`> 2ade66e56811d9753e48d51aa0fdcc1692e5d1ef uphill_wires.txt
<nats`> I concated all file together
<litghost> 072 gets consume in 074 and only tileconn.json is output
<litghost> 072 generates input data
<nats`> ce520ef646faf050651189290e1128818c19d426 downhill_wires.txt
<litghost> It doesn't directly output to the database
<nats`> yep I saw that
<nats`> I was thinking someone would have kept them
<nats`> should I make a clean patch and put a PR or zip it for you or someone else to test it ?
<nats`> uhhmmm wait
<nats`> maybe I can feed them to 074
<nats`> we could compare output
<nats`> http://www.rfporn.org/~nats/symbi/ <= you can have the 3 needed fuzzer file here
<tpb> Title: Index of /~nats/symbi/ (at www.rfporn.org)
<nats`> if you want to run it put it in a folder because it'll generate a huge mess
<nats`> and you need to be in the prjxray env
Ultrasauce has quit [Quit: Ultrasauce]
Ultrasauce has joined #symbiflow
<nats`> oky I adapted it in fuzzer 072 with good build dir etc
<nats`> I just started it one last time to see if it "works"
<nats`> I'll make the PR after that and someone will have to compare output file
<litghost> It's worth mentioning that if your version of 072 completes in 30 minutes, that is actually faster than the old version on my machine
<litghost> it still hasn't completed, and has been running for more than 1 hr
<nats`> sure
<nats`> can you get my fuzzer and run it too ?
<nats`> by the way I have a server with a core i3 on it if needed to deploy a build system
<nats`> maybe as the leak becomes huge the vivado process slows down
<nats`> it would be really possibl
<nats`> possible
<litghost> sure
<nats`> already at 20% !
<nats`> in about 8 minutes
<nats`> I launched it with time command like you recommended
<nats`> don't forget I run it on 4 processor too :D
<nats`> it may explains why it goes faster
<nats`> 30%
igorangst has left #symbiflow [#symbiflow]
<nats`> 50%
<nats`> about 4 minutes per 10% :)
vup2 has quit [Remote host closed the connection]
vup2 has joined #symbiflow
<nats`> litghost
<nats`> Work done !
<nats`> true
<nats`> real 31m49,181s
<nats`> touch build/specimen_001/OK
<nats`> user 104m14,978s
<nats`> sys 17m32,685s
<nats`> not bad
<nats`> and it generates the two file in the end
<nats`> I push on my fork and PR but it'll be a PR to be checked by someone
vup2 has quit [Quit: No Ping reply in 180 seconds.]
vup2 has joined #symbiflow
<tpb> Title: Fixing Bug #171 and #465 by natsfr · Pull Request #489 · SymbiFlow/prjxray · GitHub (at github.com)
<nats`> pull request made
<nats`> I'll write the bug explanation in bed I'm sleepy :)
<nats`> good night folks
vup2 has quit [Remote host closed the connection]
vup2 has joined #symbiflow
tpb has quit [Remote host closed the connection]
tpb has joined #symbiflow