08:40 <kbeckmann> _florent_: thanks for the reply! yeah that makes a lot of sense, migrating is probably a pretty high risk in its own. being able to use nMigen modules together with litex sounds like a good start.

14:40 <somlo> daveshah: this is going to start out as a high level, dumb-user question... Back about a month ago, I used to be able to sometimes reach 60MHz for Litex+Rocket on the versa (with -abc9 -nowidelut and the xc7dsp patch, still a PR at the time). FWIW, commits were yosys #c7f1368, trellis #f692269, nextpnr #927077e.

14:40 <somlo> With recent versions (right now it's yosys #2daa568, trellis #e2e10bf, nextpnr #c365dd1 but I've been about it for the last week or so) I'm only reaching about 10MHz less on average, and couldn't hit 60MHz after running the build in a loop over the last 24+ hours.

14:40 <somlo> Is there something else I should try, flag-wise, and/or do you know of a possible explanation as to why the average Fmax would have dropped by 10MHz?

14:41 <somlo> Sorry, I know it's not a lot to go on :)

14:46 <daveshah> I can only think it's related to the 36-bit BRAM change

14:47 <daveshah> I can't think why that would reduce Fmax though

15:17 <acathla> somlo, any debug wire left somewhere?

15:34 <somlo> daveshah: maybe I'm just unlucky, 60MHz was never a straightforward, reliable outcome; it's just that I've been looping harder and longer than I remember before without managing to land on one...

15:34 <somlo> acathla: same design as before, no additional debug wires added in recent memory :)

15:35 <somlo> daveshah: maybe I should just focus on the trellis board and forget about versa -- then I won't have to '-nowidelut' to barely squeeze everything within 45k

15:36 <somlo> btw, does the 36-bit BRAM thing have an 'off' switch I could try?

15:42 <daveshah> There isn't one as I wasn't expecting regressions, but you could try removing the "PDPW" sections from https://github.com/YosysHQ/yosys/blob/master/techlibs/ecp5/bram.txt

15:42 <tpb> Title: yosys/bram.txt at master · YosysHQ/yosys · GitHub (at github.com)

15:42 <daveshah> Or, reverting https://github.com/YosysHQ/yosys/pull/1425 in its entirety

15:42 <tpb> Title: ecp5: Add support for mapping 36-bit wide PDP BRAMs by daveshah1 · Pull Request #1425 · YosysHQ/yosys · GitHub (at github.com)

15:58 <somlo> daveshah: thanks, I'll give that a shot, and report back if I notice a difference...

15:59 <lolsborn> People running diamond, are you doing so using LiteX on Windows or Linux? I've had nothing but issues running it on Ubuntu, but I'm not sure what the cause is.

16:00 <daveshah> I've used LiteX with Diamond on Linux without any problems

16:01 <daveshah> What issues are you seeing? Does Diamond run alright on its own?

16:01 <lolsborn> I can find the error, but basically it was running out of stack when running synth

16:02 <daveshah> Might be a memory inference issue

16:02 <lolsborn> I have only 8gb of ram on this machine, I don't know if that is enough, but the system wasn't showing it was memory constrained when running

16:05 <lolsborn> This is what the error looks like https://gist.github.com/lolsborn/1f54a68c23e3719d2e5d64d778f24c2b

16:05 <tpb> Title: diamond error · GitHub (at gist.github.com)

16:12 <_florent_> lolsborn: strange, are you building an example target from litex-boards? I'm also using Diamond on Linux without issues

16:13 <daveshah> Seems like that error is coming from Tcl

16:14 <daveshah> Which is even odder

16:14 <daveshah> The only thing I can think of is Diamond version. I've only ever used 3.10

16:15 <lolsborn> I haven't tried synthing other designs on this machine in diamond. I'm just building the ulx3s target with picorv32. Works perfectly with Trellis.

16:16 <lolsborn> Using Diamond 3.11.0

16:16 <lolsborn> I might just try reinstalling

16:17 <lolsborn> daveshah: are you running Ubuntu?

16:17 <daveshah> No, Arch Linux

16:17 <lolsborn> I had to unpack the rpm and run the install scripts manually

16:17 <daveshah> I've previously run Diamond on Ubuntu 18.04 without issues too

16:18 <daveshah> Yep, I did the same

16:19 <lolsborn> I'll try reinstalling

16:20 <daveshah> I'd try 3.10 at the same time

16:29 <lolsborn> will do

18:20 <lolsborn> This is where I'm at with diamond on ubuntu https://gist.github.com/lolsborn/53d956b6d22e13c590050fab89c2289c this seems to happen with 3.10 or 3.11 the command I am running is at the bottom

18:20 <tpb> Title: Issue with LiteX using diamond · GitHub (at gist.github.com)

18:47 <somlo> daveshah: after reverting yosys PR 1425 (but keeping trellis and nextpnr at yesterday's latest), I'm still hovering in the 45-50MHz Fmax

18:48 <somlo> maybe it's just that I've exhausted my quota of good luck earlier on, and now all the random seeds I end up with are balancing out the karma :)

18:49 <daveshah> I can't see anything obvious on the nextpnr side that would be causing issues

18:49 <daveshah> Could have been another Yosys change or just bad luck as you say

18:51 <somlo> ok, I'll write it off to karma, keep spinning, while I work on migrating to the trellis board :)

19:05 <lolsborn> looks like this might be the issue I'm having https://electronics.stackexchange.com/questions/327527/lattice-icecube2-error-synplify-pro-321

19:05 <tpb> Title: fpga - Lattice iCEcube2, error synplify_pro 321 - Electrical Engineering Stack Exchange (at electronics.stackexchange.com)

19:37 <_florent_> somlo: can you try to revert https://github.com/enjoy-digital/litex/commit/1425a68d9e68cd6af4508d4369512684d1b97138 and rebuild? that's the only change i can think of regarding litex during last month that could impact timings

19:37 <tpb> Title: wishbone2csr: refactor using FSM, reduce latency (make it asynchronou… · enjoy-digital/litex@1425a68 · GitHub (at github.com)

19:54 <somlo> _florent_: I'll try that one next, thanks!

20:45 <lolsborn> I finally got Diamond working, man that was painful. The only thing I think could improve in LiteX is discovery of Diamond if --gateware-tool-path is not specified. /opt/Diamond seems arbitrary since the rpm installs to /usr/local/diamond and if you create a symlink with a trailing / it breaks. The real issue on Ubuntu though is /bin/sh being a symlink to dash and breaking many of the scripts in diamond

20:50 <lolsborn> fwiw 3.11 seems to work fine

20:57 <_florent_> lolsborn: ok good for Diamond, i'll try to add automatic discovery soon

21:01 <lolsborn> I may make a script that installs converts diamond rpm to deb, installs it and then changes all the scripts from sh to bash just because everyone in my office uses ubuntu/debian and I think it would be useful for other folks