sb0 changed the topic of #m-labs to: ARTIQ, Migen, MiSoC, Mixxeo & other M-Labs projects :: fka #milkymist :: Logs http://irclog.whitequark.org/m-labs
jkeller has quit [Quit: Page closed]
FelixVi has joined #m-labs
<davidc__> sb0: that and "oh, that bank of buttns you get to by pressing shift-3"
<davidc__> sb0: I mean, its obvious. Shift-1 goes back to the main screen, shift-2 crashes the application unless you are using this specific hardware that nobody has, so don't do that one, then shift-3 gets to the screen you want
<davidc__> (actual behaviour of a radio system I used in the past)
<sb0> cr1901_modern, link where?
<cr1901_modern> sb0: On the misoc repo, under the "how to install gcc" section
<sb0> FelixVi, the artix7 transceivers are 6.6gbps max. and 10G ethernet is relatively uncommon and expensive.
<FelixVi> sb0: it's a 4-lane xaui intf
<FelixVi> something like 3 gbps each lane
<sb0> depends which
<FelixVi> GMII
<sb0> with 10g sfp it's not that
<FelixVi> so you're right if you go directly to sfp things are different
<FelixVi> yeah, I keep forgetting that
<FelixVi> sb0: did you write the dfi bus by any chance?
<sb0> yes
<FelixVi> the address logic is what gives me a headache right now
<FelixVi> in pipistrello it's super fast, but on the saturn platform that becomes the bottleneck
<FelixVi> I don't understand why
<FelixVi> same optimization setting... I am playing with those now to see if I can get it to go away
<sb0> incorrect timing constraints?
<FelixVi> there's just period constraints
<FelixVi> 1 input clock and 5 derived ones
<FelixVi> so that's unlikely - and the ucf doesn't have anything else
<sb0> do you have async domains that ise thinks are related when they are not?
<FelixVi> that might be the case - i am running without resource sharing, without equiv reg removal and with reg balancing now
<davidc__> FelixVi: there are 10GBASE gearboxes that go from 2x5GBPS (or lower) to 10GBASE modulation suitable for driving a SFP+, but they aren't cheap/easy parts
<davidc__> probably cheaper to move to an FPGA that has the native transceiver
<FelixVi> aha, so now it runs at 62.5 MHz rather than just 25 MHz
<FelixVi> davidc__: there's some xilinx eval boards, but those were also ~2.5k ea (iirc)
<FelixVi> sb0: but even with that, optimization of the address logic domain is still screwy - per required 8.000 ns -> actual per 8.000 ns
<sb0> X&A FPGAs with fast transceivers are expensive
<FelixVi> this one would be a good place to start
<sb0> try the polarfire stuff
<sb0> X&A transceivers are trash
<sb0> polarfire may or may not be trash
<FelixVi> do you think they give me an eval board if I walk into their office? :)
<sb0> maybe. try it
<FelixVi> yeah, their board is only $1500
<FelixVi> mpf300 if that tells you anything
<FelixVi> does misoc support microsemi's toolchain? :)
<FelixVi> hmm, fmax for saturn is currently 67 MHz
<FelixVi> and that's limited by the address clock domain
<FelixVi> What I don't get is why pipistrello has no problem
<sb0> the address clock domain?
<FelixVi> pll[2] which clocks the dfi bus
<FelixVi> dq and clk are through serdes and a is from logic
<sb0> isn't that the system domain?
<FelixVi> pll[5] is sys
<FelixVi> pll[3] is clk
<sb0> maybe you have a wrong relationship between those two PLL outputs?
<FelixVi> and pll[0] is data (4x)
<sb0> phase and/or frequency
<FelixVi> yeah, just trying the same phase settings as pipistrello
<FelixVi> freq is right
<FelixVi> sb0: there isn't anything on the dfi bus that can be pipelined?
<FelixVi> I should probably look what exactly is going on there
<FelixVi> and how that interact with l2cache
<FelixVi> alright, it does 75 MHz now, I don't think it can be pushed to 83.333 MHz
<FelixVi> but if it passes memtest, I'll be pretty happy
<sb0> what was the problem?
<FelixVi> the phase offset hurt the pll[2] domain
<FelixVi> now the cpu is limiting - which is great
<FelixVi> but it fails memtest
<FelixVi> the address in the random address test is always off by 1
<FelixVi> so there is a timing hazard with that
<FelixVi> I'll try shifting the address phase and hope it goes away
rohitksingh_work has joined #m-labs
<GitHub97> [sinara] marmeladapk tagged Kasli/v1.0 at 7637ca1: https://git.io/vFAZF
<sb0> rjo, considering that _florent_'s RTM has one non-functional DAC, how do we get jesd204 sync developed?
<sb0> _florent_, can you use the hkg boards remotely?
<FelixVi> sb0: I've had no luck with phase adjustment
<FelixVi> will now let it run over night, generating 128 SoCs with different phase and bitslip and check em in the morning
<FelixVi> I'd be curious if I'm the only person seeing this on LX45-csg324-2sg FPGAs
<FelixVi> This is for 75 MHz fcpu
<FelixVi> either way, I'll make a table in the morning and will report back about what I see
FelixVi has quit [Ping timeout: 255 seconds]
<GitHub84> [smoltcp] whitequark closed pull request #66: Allow receiving broadcast packets in UDP (master...master) https://github.com/m-labs/smoltcp/pull/66
<GitHub113> smoltcp/master d8b25cd Jeremy Soller: Allow receiving broadcast packets via UDP sockets.
<GitHub113> [smoltcp] whitequark pushed 1 new commit to master: https://github.com/m-labs/smoltcp/commit/d8b25cdce85eabe72e72aefad1cc20298a27a766
<travis-ci> m-labs/smoltcp#441 (master - d8b25cd : Jeremy Soller): The build passed.
<GitHub111> [migen] jordens pushed 1 new commit to master: https://git.io/vFAVY
<GitHub111> migen/master 13ee25b Robert Jordens: kasli: add eem connectors from v1.0 release
<bb-m-labs> build #205 of migen is complete: Success [build successful] Build details are at http://buildbot.m-labs.hk/builders/migen/builds/205
<rjo> sb0: sync the dac to rtio
<sb0> rjo, shouldn't there be another sync mechanism on a single board, i.e. standard JESD204 sync?
<GitHub195> [smoltcp] phil-opp commented on issue #75: @whitequark I think the most important options are implemented now (message type, requested ip, server identifier). Are there any other options that we need? Other things that I should add/change? https://github.com/m-labs/smoltcp/pull/75#issuecomment-346603933
rohitksingh_work has quit [Read error: Connection reset by peer]
<sb0> rjo, FYI: I put the RTMs on Sayma 1 and 2, and they're now clocked at 100MHz with the SynthNV
<sb0> the signal level and parameters other than the frequency are the same as they were with the FMC DAC cards
<rjo> sb0: i don't get what you have in mind. the "other" sync mechanism consists of (a) aligning SYSREF to RTIO plus (b) implementing subclass 1 in jesd204b
<sb0> the "other" is (b)
<sb0> that would sync the two DACs on a board, right?
<rjo> that would only give you alignment between the two dacs on a rtm.
<sb0> yes
<sb0> but that's the first step
<rjo> (a) and (b) are independent
<sb0> yes
<sb0> my understanding is that _florent_ should do (b). but only one DAC works on his board.
<rjo> afaik (b) is with _florent_ and (a) is with us. (b) can be verified with a single DAC using (a) and RTIO
<rjo> what's the triage/diagnosis of the other DAC?
<_florent_> rjo: the SERDESPLL is not locking
<sb0> whitequark, have you tried artiq-2 and the switch?
<rjo> _florent_: and that is after applying the changes that greg implements on the power supplies and the clock termination?
FabM has quit [Ping timeout: 248 seconds]
FabM has joined #m-labs
FabM has quit [Ping timeout: 240 seconds]
mumptai has joined #m-labs
<felix_> rjo: hmm, any reason that the xilinx-xc7.cfg tcl file in openocd is in the cpld folder and not in the fpga folder?
<rjo> felix_: when i started this, there was only the virtex2 stuff in the cpld folder already. and there was no fpga folder. i followed that lead. someone else then did a fpga folder. if you transition all the configs and sources that's fine by me.
<felix_> ok
<rjo> but imho those categories are a bit weird anyway. i'd do it either by manufacturer or merge cpld/fpga.
mumptai has quit [Quit: Verlassend]