#milkymist IRC log for Friday, 2012-06-01

rejonhi guys01:22
wolfspraulhey01:22
wpwrakoh, the ghost of rejon speaks to us ! :)01:24
rejonoh man, that sounds horrible01:34
wpwrakfor the last few months, you've been seen a lot but never heard :)01:35
rejonYeah, its unlike me I know ;)01:40
wpwrakbtw, will you go to FISL this year ? i missed the deadline again but i'll try to sneak in ...01:46
wolfspraulthe other day when talking about -ng, -7/ddr3, mirteo etc. the topic of how to simplify the m1 came up01:52
wolfspraulI had some thoughts on this as well, but it mostly involves USB :-)01:52
wolfspraulbut quickly, for the record. the big value I see in the m1 board/product is that it is still one unified whole set, a whole product that works today01:52
wolfspraulin full awareness of the thousands of bugs lurking everywhere01:53
wolfspraulbugs and missing features01:53
wolfspraulbut I don't want to compromise the *set* we got to work together here, and fall into separate hacking projects where everybody can happily claim that this or that feature works, but unfortunately they all don't work together, anywhere01:53
wolfspraulputting that aside, I can think of removing the following parts on m1, over time:01:54
wolfspraul1) native 5-pin midi01:54
wolfspraulthat one is my #1. 5-pin midi is deader than dead. I don't think a single of the m1 users has used it, or will ever use it01:54
wolfspraulit's next to impossible to even buy these cables still01:54
wolfspraulabsolutely no new devices with this standard will come out anymore01:54
wolfspraulwe have started to support usb-midi, and if we continue with that (it's receive only now), then if anyone really still has some dust-collecting old midi devices they have to use with m1, we can refer them to those 5 USD dongles01:55
wolfspraulthe need for that is so low that I wouldn't even think of including them with m101:55
wolfspraulthere are a few things we would be dropping still, such as the hardware-level passthrough01:56
wolfspraulor maybe electrical features in our native midi support that a usb-dongle could not do?01:56
wolfspraulbut I hesitate to claim that until we specifically know what such feature could be (long cables?), especially since I think nobody will ever try anyway, as I said 5-pin midi is dead01:56
wpwrak(midi) i have !! even bought an expensive MIDI cable, which took about an hour of searching ;-))01:56
wolfspraulI think you agree with my points though, I try to make it as clear as possible01:57
wolfspraulwe can remove that native subsystem01:57
wolfspraulnobody uses it, or will ever use it01:57
wolfspraulwe can refer people to usb-midi dongles, especially after we more fully support them (in & out)01:57
wolfspraulbut there is no rush for that removal, at the current volume it doesn't really matter whether it stays or goes01:57
wolfspraulnext01:58
wolfspraul2. rca analog video in01:58
wolfspraula bit harder case to make, but the reality is that our support and excitement for analog video in is muted01:58
wpwraki don't feel an urgency for dropping legacy midi but yes, it's usefulness seems rather limited01:58
wolfspraulzero :-)01:58
wpwrak(no rush) exactly01:58
wolfspraula zombie01:58
wolfsprauloh sure01:58
wolfspraullemme continue with analog-in01:58
wolfspraulso I think the #1 competition to the quality of that video-in we have today is any 15 USD webcam01:59
wolfspraulour best argument against that is 'latency will be bad'01:59
wolfspraulbut that sounds like a lame excuse really, because nobody tried hard01:59
wolfspraulso what I can foresee is that over the next year or so, with Linux/MMU hopefully happening, and the USB driver issue at some point being addressed, we get a chance to give some usb webcam a try01:59
wolfspraul*after* such usb webcam works, we can compare the performance to what we have over analog video-in and the ADI decoder02:00
wolfsprauland I somehow think who will win02:00
wolfspraulwe pay more for the old-style analog camera than for a usb cam02:00
wolfspraulwe need to source a separate 12V power supply for the camera02:00
wolfspraulwe only scan half of the lines, and nobody seems to be motivated to improve that old stuff02:00
wolfspraulwe pay 12 USD for the ADI decoder alone02:01
wpwrakno idea about webcam latency. could be intrinsic in the way they compress, but i don't really know. also the cost of internal memory limits the amount of buffering they could possibly do02:01
wolfsprauleven if the latency of the webcam would be 10ms more, I would drop faster than anyone could make another argument :-)02:01
wolfspraulthat old analog video-in is a temporary replacement, just the right thing (!) to do to get the m1 package off the ground02:01
wolfspraulbut nothing for the long run02:01
wolfspraulfor sure not02:01
wpwrakimproving that old stuff may not be all that trivial. you'd have to de-interlace the signal, which is messy work.02:01
wolfspraulsure02:02
wolfspraulI know02:02
wolfsprauland those codec chips are a loosing proposition as well02:02
wolfsprauleverybody with a sane mind will migrate away from them02:02
wolfspraulbut the order here is: get usb webcam support working first. that will take 1 year, best case.02:02
wolfspraulonce a new (even other) digital solution works, drop the whole analog video-in thing02:02
wolfspraulafter such new solution works02:02
wolfsprauli.e.: not now02:03
wolfspraul3. audio02:03
wolfspraulsame argument as with video02:03
wolfspraulwe have a quite difficult subsystem there02:03
wolfspraulyet we cannot really excel at any particularly powerful audio feature because that subsystem is hardwired for some use cases with the wolfson codec02:03
wolfspraulone day maybe with linux/mmu/usb in better shape, we can plug in a 1-2 USD usb audio card, and may have the same performance for the user as with that entire audio subsystem02:04
wolfspraulthe wolfson codec alone will cost us more than a USB audio dongle02:04
wolfspraulthere is no way anyone could resist the business case of removing that internal subsystem02:04
wolfspraulI see that even later than the analog video-in though, but it's on the same line of thought02:04
wolfspraulmy strategy is: protect the *completeness* of the m1 experience through to the user02:05
wolfsprauldon't rip out some stuff and leave users with even less of what they had before02:05
wolfspraulfirst get the newer/better/cheaper solution to work02:05
wolfspraulwith midi we may almost be there already02:05
wolfspraulwith other subsystems such as analog video-in, audio, vga, we need some more time and get the replacement working first02:06
wolfspraulwhich can take years02:06
wolfspraulbut then we can remove all that old & expensive stuff02:06
wolfspraulthat sums it up :-)02:06
wolfspraulthat's my contribution to the "how can we simplify m1, make it cheaper, highlight the uniqueness/power of the milkymist soc" debate02:08
wpwrak(audio) that one has a reasonably good use case. digital audio isn't used all that much. and if you add too many dongles, the whole device becomes a mess02:08
wolfspraulit's #3 in my list for a reason02:09
wolfspraulI just explain my thinking, not any actions02:09
wolfspraulnone of #2 and #3 will even work anytime soon (the replacement)02:09
wolfspraulthe fact is that I can buy a usb-audio dongle for 1.50 USD02:10
wolfsprauland it won't get more expensive :-)02:10
wolfspraulthe wolfson codec alone costs what? 4 USD?02:10
wolfspraulI don't even want to know, because the business case is already a lost one to start with02:10
wolfspraulif m1 could get some really unique things out of audio, that would be different02:10
wolfspraullike a 20 channel synthesizer/mixer/whatever02:10
wpwrakthe point is that the product looks untidy if it's one box with a gazillion dongles hanging out02:10
wolfspraulbut that's not the case, we use the subsystem only to feed a simple line-in to m102:10
wolfspraulsure we could put the dongle inside the case02:11
wolfspraulbut that's all premature02:11
wolfspraulI explain my 'simplification' thinking, because I also want to move towards products that allow milkymist soc to shine02:11
wpwrakthen you have to worry about mechanical integration. are any of these cheap dongles designed for panel mounting ?02:11
wolfspraulas opposed to us lugging around tons of legacy02:11
wolfspraulI won't discuss it, it's too hypothetical02:11
wpwrak;-)02:12
wolfspraulof course the audio subsystem stays for years02:12
wolfspraullet's survive those years first02:12
wolfspraulmy point before even explaining my thinking was that we should protect the *completeness* that we have achieved02:12
wolfspraulI will02:12
wolfspraul640x480 vga may sound crappy02:12
wolfspraulbut it works02:12
wpwrakmaybe we can find a cheaper codec. could even be usb, used internally. that's a pretty common choice for an internal low-performance peripheral bus02:13
wolfspraulsure02:13
wolfspraulthat discussion is premature though02:13
wolfspraullet's focus our energy on where milkymist soc is unique/powerful02:13
wolfsprauland we have an audio subsystem now02:13
wolfspraulit's not that we have to decide whether/how to do one - we have one already02:13
wolfspraulof course I won't touch it02:14
wolfsprauladmittedly my simplifaction strategy involves a lot of usb02:14
wolfspraulwhich causes understandable headaches02:14
wolfspraulfor any of this to work, we probably need linux/mmu, we need a different usb core, etc.02:14
wpwrakthe unique thing, for the VJ, are the smooth iterative video effects. and, if we put a bit more effort into it, also the ability to tweak them (midi or maybe something fancier in the future)02:15
wolfspraulso - much later :-)02:15
wolfspraulsure02:15
wolfsprauldid I make a point for removing/breaking anything that works?02:15
wolfspraulnope02:15
wolfsprauleven removal of 5-pin midi is premature now02:16
wolfspraulbut I could see us doing that in the next major update maybe02:16
wolfspraulthere are no users of that and never will be...02:16
wpwraki think anyone who wants to use midi on m1rc3 may find it useful because the usb ports may already be taken up by keyboard and mouse02:18
wpwrakbut m1r4 won't have that problem either :)02:18
wolfspraulyes02:18
lekernel_wolfspra1l: continue on that line, then add some stuff that make the product really unique and not just some slow computer, and you'll end up with something like the mirteo :)10:05
lekernel_good to see the ideas are sinking in :p10:06
lekernelbtw, if we have HDMI hooked directly to the fpga, we can have bidirectional video ports.10:06
lekerneland the -7 series have an ADC for audio-in, and can PWM audio-out... so will be easy to have audio off the expansion header10:08
wolfspra1lgood [audio]11:12
lekernelin defense of the midi-port, it's great for communicating with external AVRs (arduino etc.) without opening the box. as always, it's just that no one cares.11:13
lekernelwhich of course doesn't mean it should be kept11:14
kristianpaulPWM audio-out, really and get decent performance?14:20
mumptaihi14:20
mumptaimulti-bit deltasigma is kind-of okay14:21
lekernelwolfspra1l: how did you submit your talk to campus party?20:11
GitHub50[milkymist-mmu-simulation] fallen pushed 4 new commits to master: http://git.io/0jKfBA20:37
GitHub50[milkymist-mmu-simulation/master] Update ISim wave visualization window config file - Yann Sionneau20:37
GitHub50[milkymist-mmu-simulation/master] Indent fix - Yann Sionneau20:37
GitHub50[milkymist-mmu-simulation/master] Cleaning code - Yann Sionneau20:37
GitHub78[milkymist-mmu] fallen pushed 5 new commits to mmu-bios: http://git.io/gQWX-Q20:58
GitHub78[milkymist-mmu/mmu-bios] Add tests for DTLB exception handling - Yann Sionneau20:58
GitHub78[milkymist-mmu/mmu-bios] Increase stack size - Yann Sionneau20:58
GitHub78[milkymist-mmu/mmu-bios] Add MMU handling functionality - Yann Sionneau20:58
Fallenoua nice dtlb-miss-exception handling test running successfully in ISim :)21:06
FallenouWill try soon on FPGA21:06
wpwrakwohoo ! i hear trumpets ! horses ! the sounds of a million marching feet ! it's the armies of the mighty penguin !21:49
Fallenouhehe :) Hope the army will arrive soon enough !22:08
Fallenouif more tests tend to prove dtlb lookups and exceptions are reliable, I think I might soon start working on ITLB22:08
Fallenouwhich should not be so different to implement22:09
Fallenoubut really more hard to debug22:09
Fallenous/more hard/harder/g22:09
Fallenouwill shoot an email soon with a sum up about dtlb exception handling and a few logs of tests on FPGA and ISim22:10
GitHub68[misp] sbourdeauducq pushed 1 new commit to master: http://git.io/ohQ62Q22:17
GitHub68[misp/master] Memory allocator - Sebastien Bourdeauducq22:17
GitHub151[milkymist-ng] sbourdeauducq pushed 1 new commit to master: http://git.io/4in2RQ22:18
GitHub151[milkymist-ng/master] software/libbase: uintptr_t - Sebastien Bourdeauducq22:18
wolfspra1llekernel: I was invited by Nelson and he helped me through the paperwork23:24
wpwrakFallenou: with dtlb working, itlb should be a breeze :)23:27
--- Sat Jun 2 201200:00

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