#milkymist IRC log for Monday, 2012-03-05

GitHub171[scripts] xiangfu pushed 1 new commit to master: http://git.io/-YzlmA07:15
GitHub171[scripts/master] reflash_m1.sh update the --data parameter when use --qi - Xiangfu Liu07:15
cladamwwolfspraul, i'll wait maybe 2 ~ 3 days for list reviews, meanwhile to prepare layout notes.08:26
wolfspraulyes, good08:29
Fallenoucladamw: got your plastic feet for the M1 :) thx !08:49
cladamwFallenou, hi great ! :-)08:50
wolfspraulfor the ADV7181C and D sourcing query, I just looked up the suitable p/n for the D, and I think it's ADV7181DBCPZ12:31
wolfspraulso the planned one is ADV7181CBSTZ, and if we go to D, it would probably be ADV7181DBCPZ12:32
wpwrakany word from AD ?12:32
wolfspraulnot much variety there, the only other version is a 'W' one for automotive12:32
wolfspraulin the que12:32
cladamwyup, should be BCPZ12:35
cladamwno else p/n choice we could pick.12:35
cladamwi've no news on D version now.12:36
wolfspraulforget D, we focus on C12:38
wolfspraulshould have info from shanghai soon12:38
wpwrakah, one bit of information for the layout house: they can move around LEDs in the matrix as they wish.12:38
wolfspraulwho wants a 12 USD analog video decoder :-)12:38
wolfspraulwe have to hurry up and improve m1...12:38
wpwrak(12 USD video decoder) ? is the D priced differently from the C ?12:39
wolfspraulall roughly the same12:39
wolfspraulat a time when whole routers or wifi phones cost less than 20 USD after assembly, such chips look like an anacronism to me anyway12:40
wpwrakah ;-)12:40
wolfspraulbut we need the C now, and then move forward. and we'll find it.12:40
cladamwwpwrak, what did you mean about "one bit of ...." ? you meant those reserved leds ?12:41
wpwrakcladamw: no, all of them. if the layout people find the arrangement inconvenient, they can put the LEDs at different places in the matrix.12:42
wolfspraulwpwrak: lekernel where do you prefer the m1 kicad schematics to live - github or qi-hardware ?12:43
wolfspraulI'm fine with that, though I will continue to maintain a meaningful set of services for people to collaborate on hardware at qi-hw, of course12:46
wolfspraulit's just starting actually :-)12:46
wolfspraulbut for the schematics histories, we can point to kicad easily12:46
lekernelthey have a much nicer interface, and let's happily leave the sysadmin and web development work to other people (who are also better at this...)12:47
wolfspraulsorry I mean point to github12:47
wolfspraulI will maintain the qi servers, but of course github is another scale etc.12:47
wolfspraulin the long run they will end like sourceforge :-012:47
lekernelwhat do you mean?12:47
cladamwwpwrak, your meaning is that we don't need to follow 'text' liked I marked 'MIDI RX' or 'RF' to under related connector, but relies on house's convenience and later we make each led's reference. did i realize correctly ?12:47
wolfspraulbut in the short run we should use whatever accelerates us now, and makes people happy12:47
wolfspraulso I am totally fine with that12:47
wpwrakcladamw: yup12:48
wolfspraullekernel: don't know, I think the most important for such sites is to not be too fragmented12:48
lekernelI have nothing against SF, except their interface is lousy and the approval process tedious12:48
wolfsprauland I don't know how github can manage this, but it's their problem now mine12:48
lekernelgithub is much better on both: good and fast interface, no approva12:48
cladamwwpwrak, okay12:48
wolfspraulgee 'not mine'12:48
wolfspraulI am ok with github, as I said12:48
wolfsprauleven the histories can point there12:49
wolfspraulyes github has been doing outstanding work12:49
wpwrak(SF) "download" links you can't wget ? i lack the words to adequately express my sentiments about that12:49
lekernelwolfspraul: what do you mean, "fragmented"?12:50
wpwrak(github) i guess as long as we use if just as repository, there's little risk. i like all the "value added" things better under our direct control12:51
wolfspraullekernel: over time github will have to manage a database they understand less and less12:52
wolfspraulthat lack of understanding is difficult to deal with, imho12:52
wolfspraulbut I really don't care, as i said I think github is doing a great job12:52
wolfspraulI was just asking where we should put those files12:52
lekerneldatabase? it's just a collection of repositories12:52
lekernelthere isn't much to understand12:52
wolfspraulok :-)12:53
lekernelunless I missed something...12:53
wolfspraulno it's good, seems wpwrak agrees as well so that's settled12:54
lekernelif we switch to kicad, let's add a second video input and a series-7 FPGA13:00
lekerneland ddr2 or ddr313:01
wolfspraultotally with you13:03
wolfspraulthe second video input is done how?13:03
lekernelanother adv7181 or similar circuit13:03
wolfspraulhow about digital video-in ?13:03
lekernelhmm... why not13:04
lekernelif we get the ddr3 to work correctly, we should have a lot of bandwidth13:04
wolfspraulyes, if the 2nd one would be digital-in that would be perfect as it provides a smooth transition from something that works (adv7181) to something that first needs to be developed and optimized13:06
wolfspraulI haven't read enough about hdmi, displayport, sdi or whatever to know what makes sense though digitally13:06
lekernelwhat about 1x adv7181 + 2x digital ?13:07
wolfspraullekernel: did you see the question yesterday - how do you feel about removing the din-midi connectors and including a usb-midi cable instead? (not now, but assuming usb-midi works flawlessly one day)13:07
lekernelthe digital ones are cheap as long as we have the memory bandwidth13:07
lekernelit's just a connector13:07
lekerneland some routes13:07
wolfspraulwhat are the differences between the standards - what exist even? hdmi/dp/sdi?13:08
lekernelbtw - can HDMI ports be used both ways?13:08
lekernelinput vs. output switchable by software?13:08
lekernelthat would be awesome13:08
wolfsprauldon't know13:08
lekernelSDI is very specialized13:09
wolfspraul7-series and ddr3 I agree, we should go there asap (after all the things in motion now have settled and we are still alive, of course)13:09
lekernelit's for very expensive equipment on big stages13:09
lekernelwe can have it, of course13:09
wolfspraulI know too little about it, need to see whether that's a sales opportunity or not13:09
lekernelbut it'd be a different product, on which HDMI makes no sense13:09
lekernelwe'd put the board in a rack and sell it for 10000¬13:09
wolfspraulI also think we should switch to spi flash instead of nor, finally the various nor problems grow on me13:10
wpwrak(kicad) step 1: make an exact replica of the existing M1r4. step 2: clean up :) step 3: do new fancy things.13:10
wolfspraulhow about displayport for video-in ?13:10
wpwrak(kicad) step 1 is important to get people aboard. it's much easier to copy from an existing design than to do something new13:10
wolfspraulso let's first investigate hdmi and dp, I guess (for digital video-in)13:10
wolfspraulwpwrak: no worries13:10
lekerneldisplayport looks USB-style messy (just had a quick look at it) and not very common13:11
lekernelhdmi is the most straightforward - even simpler than DVI because it doesn't have the analog pins13:11
wolfspraulone question is also what allows us to create the highest performance13:11
wolfspraulthat is something I can judge very badly, but you can13:11
wolfspraul'us' means with our design, fpga, milkymist soc, etc.13:12
wolfspraulmakes no sense that we run after some buzzwords only to then deliver mediocre performance13:12
lekernelif we have a DVI input, we might have problems with analog devices (not pun intended)13:12
wolfsprauldon't understand13:12
lekernelDVI can work both with analog signals (= VGA) or digital signals (= HDMI)13:13
wpwrak(nor) /me does a happy little dance :)13:13
ThihiDVI-I has both, there is of course DVI-D which only has the digital.13:13
ThihiDisplayPort is not common yet, but it is steadily gaining momentum.13:14
lekernel...and when you have a DVI input, you should be prepared to accept both digital and analog signals13:14
wolfspraulwpwrak: [nor] the 'grows on me' is mainly how uncommon nor is for that task, and that I don't really buy the 'boots faster' argument13:15
wolfspraulplus it's expensive, plus lots of varieties / sourcing issues13:15
wolfspraullekernel: how strongly do you feel in favor of the nor flash over a spi flash?13:16
lekernelif we have DP, seems we can still happily ignore this micro-packet mess for a while13:16
lekernelslide "DisplayPort uses a layered protocol for Isochronous AV Stream Transport" makes me wish for the death of DP already13:19
lekernelsounds like there are revolving doors between the USB and VESA design committees13:20
lekernelah, also there are electrical level problems too: see the "link training" slide13:22
lekernelI'm not even sure the FPGA IOBs will support it => need for additional transceiver chip etc.13:22
lekernelwolfspraul: (nor/spi) NOR was supposedly less design effort (though with the "corruption" series of bugs this idea has pretty much been proven false)13:25
wolfsprauloh I'm sure the spi switch will be painful as well, but as I keep thinking about these things over time I start to feel when we make a big change like -7 series and ddr3, that one may just go in as well13:26
wolfspraulat least that's the right opportunity without disrupting the milkymist continuity unnecessarily, imho13:26
wolfsprauleither then or never :-)13:26
lekernelits main advantage remains that it's easy to do execute-in-place13:27
lekernelwhich is used to boot the system when the SDRAM is not working yet13:27
lekernelwe have the CPU executing code directly from the NOR13:27
lekerneland no SDRAM (which that code is bringing up)13:27
wolfspraulgood point, wasn't really on my mind13:27
lekernelit can be done with SPI flash too, but we need to implement and debug full gateware-based SPI access13:28
lekernelit'll be slower than NOR, but we don't really need speed for booting, and the CPU cache should keep it quite fast anyway13:29
lekernelanother good side of SPI is it frees many FPGA I/Os13:30
lekernelso... well, I guess we should give it a try13:31
wolfspraulhow much slower do you think it might be?13:37
wolfspraulthe 'fast' was always a nor argument as well, but I never remember any specific number13:38
wolfspraulmy feeling is we are talking about a difference of less than 500ms or so13:38
wolfspraulin other words - other factors of a real product boot procedure are far more significant anyway13:38
lekernelah: http://hackipedia.org/Hardware/video/connectors/DisplayPort/VESA%20DisplayPort%20Standard%20v1.1a.pdf13:42
wolfspraulwhich of the -7 series fpga and ddr3 chips would you like to start with?13:48
wolfspraulor do you think make the most sense to start with?13:49
lekernellet's say any ddr3-1600 chip from micron13:54
lekernelmaybe two of them, for more bandwidth13:54
lekerneland a FPGA slightly larger than what we have now - maybe from the mid-range ones (kintex), if they're not too expensive13:55
lekernelbtw, if we let the user input a HDCP master key themselves, does the product still violate DMCA? :-)14:00
wolfsprauldon't even start with that14:00
wolfspraulI think if we do anything hdmi, we should not mention 'hdmi' *anywhere*14:00
wolfspraulwe need to use a new term that just describes the electrical/protocol/video parameters14:00
Hodapplekernel: by telling me that the HDCP master key can violate the DMCA, you are violating the DMCA.14:01
Hodapplekernel: fucker, why do you hate capitalism?14:01
wolfspraulI think kintex chips may well cost hundreds of usd14:02
wolfspraulbut we need to see14:02
lekernelwolfspraul: I don't know, usually they had the virtex series for that14:02
lekernelit was spartan/virtex14:02
wolfspraulmaybe they are thousands now :-)14:02
lekernelnow it's artix/kintex/virtex14:02
larsclekernel: it does not (violate the DMCA). I think14:10
Hodapplarsc: I wouldn't worry so much about actual DMCA violation; I'd worry about anyone with enough lawyers thinking they could make a case that it does.14:11
larscI wouldn't ;)14:14
Action: Hodapp shrugs14:14
GitHub68[migen] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/migen/commit/c4c22c9ca071816f7754d7eb2da46de04e5e8f3314:46
GitHub68[migen/master] sim: signal writes working - Sebastien Bourdeauducq14:46
wpwrak(gateware-level spi access) somehow, that doesn't sound very scary :)14:50
lekernelshouldn't be, but those things aren't very exciting and yet can take a lot of time if pesky bugs show up15:01
wpwrakwell, not everything we do can be new and sexy :)15:02
GitHub199[migen] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/migen/commit/aac97525585b1e3591969ba3771a8d34b12b7d6119:38
GitHub199[migen/master] sim: basic functionality working - Sebastien Bourdeauducq19:38
lekernelhttp://pastebin.com/uEF1Yvmb (the first zeros are due to the assertion of reset)19:43
kristianpaullekernel: http://sylvestre.ledru.info/blog/sylvestre/2012/02/29/rebuild_of_the_debian_archive_with_clang20:06
kristianpaulgithub is okay as you keep the files over other places :-)20:07
kristianpaulalso forks are nice too follow, and the social thing, dont see more added values besides that20:09
lekernelguess the social thing doesn't really matter here as we're (almost) the only hardware project ...20:11
Fallenouand we don't like being social *joke*20:12
HodappI found milkymist via github.20:12
lekernelwith only a few dozen followers M1 SoC used to be the #1 verilog project (and then the more fashionable bitcoin miner took over)20:12
kristianpauloh sure, cooll will be automated stuff you can get from the repos20:12
kristianpauli love that kicad diff from qi-hw for example20:12
kristianpaulminer fever :)20:13
kristianpaulreally? http://www.corelabs.cn/site/milkymist/20:23
kristianpaul" CORE Labs will be hosting some researchers in late August 2010, as well as collaborating with the development of its design."20:24
kristianpaulwell, perhaps, i wasnt aware of milkymist at that date..20:24
lekernelkristianpaul: no, it did not happen22:32
wpwraklekernel: hmm, not sure if such a flagrant violation of copyright is a good idea23:16
wpwraklekernel: i have a less aggressive solution: http://projects.qi-hardware.com/index.php/p/eda-tools/source/tree/master/dsv/dsv23:19
wpwraklekernel: this reads a file describing the location of data sheets, then downloads them. after that, they're available locally, under the name(s) we assign. e.g., we could use "dsv adc" instead of "dsv adv7181c". or even things like "dsv q1"23:21
wpwraksome may still consider it illegal, but it's at least a lot less blatant23:21
--- Tue Mar 6 201200:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!