#qi-hardware IRC log for Monday, 2011-04-11

wpwrakwhitequark: could be either, depending on the board's complexity. maybe we can just call it "heroic" ? ;-)00:07
whitequarkaround of 15 smd components there, with .016" width wires00:14
wpwrak16 mil ? so ... power electronics. are you designing a board to bridge the 50 Hz and the 60 Hz net in japan ? :)00:15
wpwrakwell, i have a board with 30 mil traces and i think i also managed to break one ... that was with a bit too much haste on the acid, though00:19
wpwrakthe flip side is that i had it etched in something like ten seconds ;-)00:19
qi-bot[commit] Werner Almesberger: f10/Makefile (brd): funny, the "brd" target never worked http://qi-hw.com/p/wernermisc/3b04bf000:23
whitequarkwpwrak: not power at all. and they look a bit too thin, anyway00:29
whitequarkand when I place the photonegative above the board, there is no visible gap between the image and the actual wires on board00:31
whitequarkI'll post a photo tomorrow, I think00:32
whitequark(maybe my design is crappy, and someone will tell me that. that would be very good, actually)00:32
wpwrakwhitequark: (no gap) does that mean that the traces are the same width as the design or that they're no wider than the design ?00:34
whitequarkwpwrak: they're the same width, through they do not look so (probably due to an optical illusion. white on black looks wider than copper on PCB)00:36
whitequarkthe .016 trace looks a bit too narrow near 0805 components, through I usually lay out digital lines with .01200:37
wpwrakwhitequark: (15 components) that's simple than atusb, 150% of the complexity of atben. both are 8 mil designs. atben has trace failure ratio of about 25%, or less. did you check for problems before etching ?00:38
wpwrak(illusions) yeah, you can never tell ;-)00:39
whitequark8 mil!!!00:39
whitequarknow I feel myself like an idiot.00:40
wpwrakwelcome to the wonderful world of toner transfer ;-)00:40
whitequarkdo we both use same mils?00:41
wpwraktruth be told, i don't think my traces are really 8 mil. most likely, they're ~10 mil. otherwise, i wouldn't have so many shorts in zones with (nominally) 8 mil clearance.00:41
whitequarkhmm.00:42
whitequarkI didn't really tried 8mil traces at all00:42
wpwrakof course, shorts are easy to remove. all they take is one swift cut :)00:42
wpwrakyou're using a photochemical process, right ?00:43
whitequarkyeah00:43
whitequarkdry film photoresist00:43
whitequarkit works WAY better than toner transfer (for me)00:43
wpwrakthat should give you even better accuracy than toner transfer, yes00:43
whitequarkat least it is repetitive, i.e. I get exactly the same crap each time00:43
whitequarkand the failure rate is maybe 10 or 50 times less than with toner transfer. I was not able to do a single board without errors with the latter00:44
wpwrakfor toner transfer, all the magic is in the paper. plus, you have to be either very steady at ironing, or automate the proces00:44
wpwrak(errors) easily fixed errors or real problems ?00:45
whitequarkfor the toner transfer, they often were serious00:45
whitequarkfor the photoresist, they are often fixed by a cut or a small piece of thin wire00:45
whitequarkI use laminator to attach the photoresistive film to the board, anyway00:46
wpwrakthin wire is level 3 ;-) a scratch / a bit of acid resist pen before etching would be level 1. a cut or a solder bridge after etching would be level 200:46
whitequarkah yes, acid resist pen. cd marker works somehow for that00:47
wpwraki almost always have level 1 problems. some of them may not be very serious, though.00:48
whitequarkI think that my worst problem is with etching itself00:48
wpwrakwhat acid ?00:48
whitequarkI use FeCl3 for that00:48
wpwrakshould be "friendly", although a mess00:49
wpwrakdo you reuse it a lot ?00:49
whitequarkyes00:49
wpwrakthat may be the problem00:49
whitequarkI think no. if you leave it exposed to air, it eventually oxidizes all the copper00:49
whitequarkand, furtherly, FeCl2 converts itself to FeCl300:50
wpwrakerm, how long to you leave your boards in the acid ?00:50
whitequarkI check them regularily, and remove when all copper not covered by the resist is etched00:50
whitequark~20-30mins at room temp.00:50
wpwrakokay. 30 min is still okay.00:50
whitequark(oxidizing trick) for that to work, you need to add 35% HCl to it sometimes00:51
wpwrakyou're moving the board around while etching ?00:51
whitequarkyes, several times00:51
wpwraklike, all the time ? :)00:51
whitequarkotherwise it'll stall at some point00:51
whitequarkno, 2 or 3 times maybe00:51
wpwrakoh no, it won't stall. it will just surprise you ;-)00:52
whitequarkhmmm00:52
whitequarkwhat kind of surprise?00:52
wpwrakuneven etching. stops in some areas (because it's blocked by byproducts) yet happily etches on in others.00:52
wpwrakit's basically the same as with an excessively aggressive acid. also there, you can't control the distribution.00:53
whitequarkwell, my problem, as I understand it, is: it should etch the unneeded copper from top to bottom, and it should not etch the traces from sides. but it still does00:54
wpwrakyou have to move it around a bit more00:54
whitequarksounds reasonably, I'll try that with next board00:54
whitequarkthanks00:54
wpwrakalso, FeCl3 is supposed to be warmed up. it's a mess, though.00:55
whitequarkgrhm, warming it up isn't easy at all00:55
wpwrakif you have acceptable ventilation, i would recommend giving HCl+H2O2 a try. that acid doesn't last very long (hours, not days), but it's so incredibly cheap that you usually don't care.00:56
whitequarkI should either get some equipment, or convince my parents that the spots of FeCl3 all over the kitchen is an acceptable byproduct of etching...00:57
wpwrakit works at room temperature and is transparent. its downside is that the Cl evaporates. so if you leave it inside, it will attack its surroundings.00:57
wpwrakFeCl3 is a pig ;-)00:58
whitequarkwhen it is cold, I manage to not leave traces anywhere00:58
whitequark(HCl+H2O2) there is a terribly small problem. you can't freely buy acids (and more than ~3% H2O2) in russia00:58
wpwrakcold FeCl3 is also the one that doesn't work very well ;-)00:58
whitequarkthey're regulated by some idiotic act written to prevent illegal drug manufacturing or something like that00:59
whitequarkHCl, H2SO4, H2O2, benzene, phenol, even KMnO400:59
wpwrakone nice thing about HCl+H2O2 is that, if you let it evaporate, it produces nice crystals. you can sweep them up and throw them away with solid waste. (that may not be entirely according to the rules, but there shouldn't really be any danger in this)01:00
wpwrakHCl ? that's used all over the place.01:01
wpwrakH2O2 is used to desinfect wounds.01:01
wpwrak(at low concentrations. at higher concentrations, it's more like causing them ...)01:01
whitequark(H2O2) yeah, low concentrated one can be bought freely01:02
whitequarkbut anything more ~10% is not available01:03
wpwrak10% should be plenty ;-)01:03
whitequarkHCl too01:03
wpwrakfor etching PCBs, the usual concentration is 5%. it does mean that you can't easily reoxidize your acid, though. but hey, it's cheap. just mix some new.01:04
wpwrakMCl .. hmm, maybe you have to look at places for professional/industrial supplies. in general, even ~35% should be in very common use.01:05
whitequarkwell, the singke place where I've found some HCl was my university. technically, teachers aren't allowed to give students more than needed for labs, but they can sometimes look the other way (especially if you ask him)...01:07
whitequark*single01:07
wpwrakthe HCl only gets really nasty if you give it a lot of oxygen. so mix 30+% HCl and 30+% peroxide, toss some copper into it, and watch the fun ;-) (preferably from a safe distance)01:07
whitequarkindustrial suppliers require a company to work with, not an invividual01:07
whitequarksame with professional ones AFAIK01:08
wpwrakmaybe HCl is know under some other name. in english and spanish, it's "muriatic" acid. it's a very old chemical, so it may have many "strange" names.01:08
wpwrakdon't you have hardware stores in russia ?01:08
wpwrak(construction/plumbing/etc. supplies)01:09
whitequark(about name) well, maybe I'm not a chemistry genius, but I cannot know it THAT bad :/01:10
wpwrakHCl is also used for swimming pools. of course, that may be something that's not so popular in russia :)01:10
wpwrak(name) you may be surprised ;-) maybe ask someone who works in the field.01:11
whitequark(pools) I think they use not the HCl itself, but Cl2 gas. it decomposes to HCl+HOCl in water, with the latter giving all the effect01:12
whitequark(name) I sure I know how it is called, really. that's so basic knowledge01:12
whitequark(stores) hmm, I never thought (and never heard) that I may find HCl there. may be a place to check01:13
wpwrakwhitequark: (pools) it's used to control the pH level. a quick google find (alas, with imperial units): http://www.deh.enr.state.nc.us/ehs/quality/wph.htm01:14
whitequarkwpwrak: hmm... maybe it's not controlled here? at least I never heard of that01:15
wpwrakwhitequark: (uncontrolled pH level) well, have fun with your pool then ;-))01:16
wpwrak"Metal fittings, pump impeller, heater core may corrode", "Plaster walls are etched", "Eye irritation may occur", ...01:17
whitequarkuh-oh01:17
wpwrakand there's also the very subtle "Chlorine activity is slowed and inefficient". in other words, the chemicals won't get you, but the bugs will ;-)01:18
whitequarkwell, a quick googling (in russian) reveals a lot of pool tools sellers, so probably I'm wrong01:18
whitequarkdidn't knew that managing a pool is so hard, hm01:18
wpwrakseems to be pretty sophisticated, yes. i wonder how many pool owners know all this ;-)01:20
wpwrakbut yes, pool supplies shops may be a good place to ask for HCl. they may sell it by the barrel :)01:21
whitequarkI don't know anyone with a personal pool anyway01:21
whitequarkheh01:21
whitequarkthe construction market is another good place01:22
wpwrakother places would be shops with plumbing supplies. i'm not quite sure that the plumbers use it for, but at the local hardware supermarket, they keep HCl in bottles of 5 l (or maybe more) in the plumbing section.01:22
wpwrakyeah01:22
wpwrakit's also used for cleaning, concrete and other things01:22
wpwrakjust don't use it on those dainty marble surfaces :)01:23
whitequarkahh, carbonates...01:24
wpwrakyes, could be for lime/scale removal01:25
wpwrakwhitequark: anyway, for HCl+H2O2, you want to use some sort of balcony. not necessarily for the etching itself, but for placing the acid after use, to let it quietly evaporate01:29
whitequarkI think that warming FeCl3 is less difficult, currently01:29
whitequarkbut I'll record your method, thanks01:29
whitequarkbtw, it's 5:30AM here and I'll go to bed :)01:30
wpwrakwarming FeCl3 also has its traps ;-)01:30
wpwraka good time for some rest. just before the evil day star rises ;-)01:30
whitequarkyeah. night is the best time for work01:31
wpwrakwhitequark: couldn't agree more ;-)01:34
kristianpaulhttp://michelebavaro.blogspot.com/2011/04/benefit-of-glonass.html03:34
whitequarkwpwrak: hm. so, what are the traps of warm FeCl3?05:22
qi-bot[commit] Xiangfu Liu: add variable for FLICKERNOISE http://qi-hw.com/p/m1s/158f7bc06:22
qi-bot[commit] Xiangfu Liu:  merge compile-rtems to compile-flickernoise http://qi-hw.com/p/m1s/50c608106:22
qi-bot[commit] Xiangfu Liu: update git address http://qi-hw.com/p/m1s/5b6b26106:22
qi-bot[commit] Xiangfu Liu: remove autotest-m1.git, fix the wallpaper name http://qi-hw.com/p/m1s/81466a906:50
kyakwolfspraul: i don't like to get excited too early, but today i will visit my post office and pick up Ben :)09:11
kyakthey sent me a notification that it has arrived09:12
wolfspraulhe09:12
wolfspraullet's wait until you open the box, it boots, etc.09:12
kyaktoday is exactly one month09:12
kyakyeah09:12
wolfspraulfirst let's wait until we have the real bottom line, then it's time for a conclusion...09:12
kyakright09:13
qi-bot[commit] Xiangfu Liu: using cat create the flash.batch file http://qi-hw.com/p/m1s/f085a9209:22
qi-bot[commit] Xiangfu Liu: using mktemp for create tmp file http://qi-hw.com/p/m1s/bc7f29b09:30
wpwrakwhitequark: (traps) risk of the outer container breaking due to heat stress. risk of overheating. limited choice of locations (e.g., the kitchen oven may be less convenient and less spill-tolerant than, say, the balcony)09:38
whitequarkwpwrak: it's quite cold at the balcony now (not to say it is... eh... cluttered up)09:41
whitequarkand no wall sockets there, so I'd leave the door open etc.09:42
whitequarkhm.09:42
wpwrakwhitequark: yeah, an open door will do, too. just have some air flow. the issue is more finding a place where a) nobody will bump into your setup, and b) where spill can be tolerated. of course, you can use temporal separation instead of or complementing spacial separation :)09:46
wpwrakkristianpaul: those pictures make a lot more sense once you realize they're from the UK ;-)09:47
whitequarkwpwrak: what about ammonia (or ammonium?) persulphate?10:06
whitequarkit is cheap, easily obtained and does not have color10:06
wpwrakwhitequark: i tried it once, briefly, and got uneven etching. but yes, it could be an option.10:09
Action: wpwrak wonders what happened with the panelized gerbers. it's been a while since that order went out ...10:24
kyakwpwrak: can two atben's communicate between each other? or two atusb's?10:31
wpwrakkyak: sure10:31
kyakso they are absolutely the same, basically it's just the form factor that is differenet?10:32
wpwrakkyak: atusb = atben+usb_to_spi (more or less)10:32
kyakah, ok10:32
kyaki wonder if they are going to be sold in pairs10:33
kyaki.e. atben+atusb10:33
wpwrakkyak: thet're not absolutely identical: atben doesn't use the transceiver's clock output while atusb used it to clock the cpu. furthermore, atben doesn't have a reset input (so you have to power-cycle it) while atusb does.10:33
wpwrakkyak: dunno what tuxbrain has decided on10:34
kyakwpwrak: so the atben cpu is running on internal clock?10:37
wpwrakkyak: the atben transceiver has its own crystal, yes. there's no "CPU" on atben, just the RF chip10:38
kyakbut these are two same RF chips on atben/atusb?10:38
wpwrakatben and atusb use the same RF chip. atusb also has an ATmega32U2 to interface the transceiver (SPI) with USB10:39
wpwrak"transceiver" = "RF chip"10:39
kyakdid you decide not to use the external clock for atben to save space?10:40
wpwrakkyak: i tried to use the ben's system clock to clock the transceiver, which would have eliminated the crystal. unfortunately, the ben's clock is too noisy for this.10:42
kyakjust trying to understand, it seems that atben is "worse" than atusb?10:44
wpwrakkyak: atusb saves one crystal (the avr would typically also have one) by re-using the transceiver clock.10:44
wpwrakkyak: atben is simpler and cheaper than atusb.10:44
kyakdo you have any idea what the price could be for both of them?10:45
wpwrakkyak: here you can see what's on them: http://downloads.qi-hardware.com/people/werner/wpan/tmp/2boards-20110305.jpg10:45
wpwrakmagnified: http://downloads.qi-hardware.com/people/werner/wpan/tmp/2boards-20110305-full.jpg10:45
wpwrakkyak: tuxbrain once mentioned an estimate of EUR 25 for atben and EUR 35 for atusb, but i don't know if these are still current.10:47
kyakwpwrak: is there going to be a casing for them?10:48
wpwrakkyak: that's still under investigation :)10:49
wpwrakkyak: one idea is to ship them with a bit of silicone, so can you make your own casing.10:49
kyakwpwrak: seems there are no LEDs, too? My eye is not trained, i couldn't see it even on magnified picture10:50
wpwrakkyak: but there are still a few unknown parameters, such as whether the material is easy enough to handle, what quantity is needed, and how it affects RF performance. wouldn't be so good if it absorbed or reflected 90% of the signal ;-)10:51
wpwrakkyak: atusb has one LED10:51
kyakwpwrak: silicone hasn't done any harm to anyone, ever ;)10:51
kyakwpwrak: i think it should be posible to attached one atben to my linksys router (maybe i'll have to revert the SDIO card mod) and turn my linksys router into a WPAN-capable device :)10:54
kyakthat would be cool10:54
wpwraksounds like a fun project :)10:54
wpwraknote that atben requires the ability to cycle the voltage supply (for reset)10:55
wpwrakwell, in theory, you don't need that, but it's safer if you have that ability10:56
kyaki can cycle the router itself, or come up with an idea how to detach atben from the router10:58
kyakwpwrak: do you think there could be some problems with 2.4 kernel?10:59
wpwrakkyak: well yes, 2.4 doesn't have any in-kernel support for IEEE 802.15.411:00
kyakluckily, i should be able to use 2.6 with wrt54gl..11:01
wpwrakgood :)11:01
whitequarkkyak: do you have wrt54gl? :)11:21
kyakwhitequark: yeah :)11:22
whitequarkkyak: one of worst routers I've seen. slow CPU, 100M switch and costs too much11:32
kyakwhitequark: it's a router, what kind of CPU do you expect? :) It' sdoing it's job fine, and i don't even have 1Gb network card to complain about 100 Mb switch11:37
kyakcosts too much? i doubt it is still sold11:37
kyakmoreover, it was a present and back in those days (around 5 years ago) it was maybe the first home router running Linux11:38
kyakwpwrak: could you correct me if i'm wrong: i need to find at least 4 GPIOs to be able to communicate with atben. The 4 GPIOs are already used for SDIO mode, and i would need to find other spare GPIOs (there is no option to share GPIOs?)11:41
wpwrakkyak: you need at least one more, for the interrupt (that is, unless you want to poll)11:43
wpwrakkyak: there's also another signal that can be useful to have (to enter sleep mode and to precisely trigger a transmission). but it's less important than the others.11:44
kyakall right, so it counts to 5-6 GPIOs.. I'll check if wrt54gl has those11:45
kyakthere must be other spare GPIOs, at least those controlling LEDs: http://www.hendlsofen.de/WRT54GL/eng/WRT54GL_SDMod.html11:46
kyak6 LEDs in a row, conveniently :)11:47
whitequarkkyak: you'll probably need to cut traces to leds then. I doubt they can work simultaneously with atben11:51
whitequarkkyak: oh yes, 5 years ago it was a very good router. it is still sold, yeah, but it is nothing compared to e.g. last tplink model. 1gbit, fast cpu and usb for the same cost11:53
kyakwhitequark: hmm, having a look at GPIO7, which they took from LED, seems that it was not necessary to cut the traces?11:55
kyakwhitequark: sure, it's not a modern router :)11:55
kyaki wish i had USB, too11:55
whitequarkkyak: 3k RUR for gigabit wireless-n router with excellent linux support isn't a lot at all :)12:00
kyakwhitequark: let me finish beating the crap out of my old good linksys before you persuade me to go and buy a new router :)12:02
whitequarkkyak: if you'd follow my advice, at the time you'll brick and/or fry your old router you'd already have a replacement ;)12:05
Action: kyak tries to extrapolate whitequark's advice to womenfolks12:10
whitequarkkyak: in the cpu world, that's called pipelining. I like to apply this technology to various human activites12:15
whitequarkconsider, for example, eating icecream. by the time the first one has melted completely, you'd eat a half of second and buy third12:17
wpwrakwhitequark: and if you had concentrated on just the first one, you'd have eaten it before it melted in your fingers ;-)12:21
kyakplus you would spend three times less money :)12:22
kyakdamn is it hard to find schematics for that router..12:22
whitequarkwpwrak: yeah, that's the essence of pipelining12:44
whitequarkkyak: openwrt folks have good docs for it12:44
whitequarkkyak: here: http://wiki.openwrt.org/oldwiki/port.gpio12:46
kyakwhitequark: oh yeah, thanks12:49
kyakit definitely lacks GPIOs to manage both SDIO and atben at the same time12:54
kristianpaulwpwrak: labsurlab related pics?13:05
wpwrakkristianpaul: hmm ?13:05
kristianpaul04:44 < wpwrak> kristianpaul: those pictures make a lot more sense once you realize they're from the UK ;-)13:05
kristianpaulbrb (breakfast)13:06
wpwrakkristianpaul: i meant this one: http://michelebavaro.blogspot.com/2011/04/benefit-of-glonass.html13:06
wpwrakkristianpaul: the last one looks weird until you realize that it's from the UK where they drive on the wrong side of the street (it's fun to ride in someone's car in front there. at each intersection, you think they're about to commit suicide :)13:08
kristianpaulwpwrak: ah, yes :-)13:27
whitequarkkyak: the sd card is bitbanged too13:27
kyakwhitequark: right, so?..13:28
whitequarkkyak: well, it's not SDIO then. just plain SPI13:30
Action: kyak shrugs13:33
wpwrakwhitequark: could be any of the three modes :) 4-bit SDIO, 1-bit SDIO, SPI.13:37
larscis there 1-bit SDIO?13:39
larscapparently there is13:41
kristianpaulsure13:41
wpwraklarsc: i actually didn't check ... but there's certainly 1-bit SD13:42
kristianpaulSIE have 1-bit SDIO i remenber13:42
larscalthough it uses an additional pin for IRQs13:42
wpwrakah, "1 bit" is the bus width. still has a lot of signals.13:43
kristianpaul(lot of signals) thats the fun part ;-)13:44
whitequarkwpwrak: is there any code for proprietary sd protocol in linux core for bitbanging?13:46
qi-bot[commit] Werner Almesberger: tools/lib/: atnet now works (tried atrf-txrx send and receive, atrf-rssi) http://qi-hw.com/p/ben-wpan/424653513:46
whitequarkI think no13:46
whitequarkanyway, it would not be useful at all, as the SDIO has one main advantage over SPI, speed, and that's low anyway with bitbanging13:47
larscsdio won't work with bitbanging13:49
larscat least not with the current linux drivers13:49
larscmmc bitbanging is done in mmc spi mode13:49
wpwrakwhitequark: i don't see any bit-banging driver. it wouldn't be useless, because you still have the wider bus (in case of 4-bit SD(IO))13:50
wpwraklarsc: why wouldn't it work ?13:50
whitequarkwpwrak: and, given the gpio framework, you'd call slow bitbanging functions a lot more. that's a C function call per bit13:50
larscwpwrak: because there is no support13:50
wpwrakwhitequark: i didn't say it would be blindingly fast ;-)13:51
wpwraklarsc: okay, but one could write a driver :)13:51
larscmaybe13:51
whitequarkwpwrak: then, what are advantages of wider bus? more gpio pins doing useless work?13:52
wpwrakthe good thing about linux is that "that's not possible" hardly exists, it's usually "nobody has gotten around to doing it yet" ;-)13:52
wpwrakwhitequark: they would make it a bit faster than just spi13:52
whitequarkwpwrak: (linux) more "only you in the whole universe need that, and every other man who tried to do thatis dead"13:53
wpwrakwhitequark: spi: 2.5 toggles per bit. 4-bit sdio: 4 toggles per 4 bits13:53
wpwrak(on average)13:54
whitequarkwpwrak: 2.5 toggles?13:54
whitequarksdio does have clock, too13:54
wpwrakwhitequark: set data if it changed = 0.5, raise clock = 1, drop clock = 113:55
whitequarkand a read, too13:55
whitequarkit has the same cost as write13:55
wpwrakfor tx ?13:55
whitequarkah yes.13:55
wpwrakread may actually be even more expensive13:55
whitequarkon controllers where gpio bitbanging is worth implenting, it is a cpu-accessible register read (or write for toggles)13:56
larscand you need to change direction13:56
wpwrakbtw, on the ben, i can poll a gpio at about 10 MHz. (that was a loop that counts cycles until a pin changes)13:57
larscwpwrak: with gpio_get_value() or direct register access?13:57
wpwraklarsc: if you alternate tx and rx, which fortunately doesn't happen so much in block transfers :)13:58
whitequarkwpwrak: that's 33 cycles per poll, right?13:58
wpwraklarsc: direct access13:58
wpwrakwhitequark: in that order, yes. maybe the rate was actually a bit higher, could have been 12 MHz. (i didn't measure the true pulse length)13:59
larscreading from the register alone should probably kill a few cycles14:00
whitequarkwpwrak: what was your statement? while(SOMEGPIOREG & (1 << bit)) i++; ?14:00
wpwrakwhitequark: (on controllers where gpio bitbanging is worth implenting) do yo uhave cpus where the gpio registers are inaccessible ? ;-)14:01
whitequarkwpwrak: sometimes i2c expanders are used, when there are not a lot of gpios on cpu itself14:01
whitequarkwpwrak: this is especially the case on PDAs14:01
wpwrakdo i--; while (!(*pdpin & 0x1000));14:02
wpwrakwhitequark: okay, via i2c, it would be indeed evil ;-) bitbang i2c, then bitbang the gpios on i2c ;-)14:02
whitequarkwpwrak: that translates to ~15 mips instructions14:07
whitequarkhttp://dpaste.com/530902/14:08
whitequark12, to be precise14:09
wpwrakwhitequark: did you set -O9 ? :)14:09
whitequarkaha14:09
whitequarknow it's better14:09
wpwrak;-)14:09
whitequarkhttp://dpaste.com/530903/14:10
whitequarkthe percentage of nops became much higher ;)14:10
whitequark8 instructions14:10
whitequarkwith one pipeline flush, it'd be 16 cycles per poll14:11
larscwhitequark: did you compile for mips32?14:11
wpwrakwhitequark: you should use "i" somewhere. it was optimized out.14:12
whitequarklarsc: no. with -march=mips32 it removes all the load delay slots14:12
larscwhitequark: thought so14:12
whitequarkhttp://dpaste.com/530907/14:13
larscwpwrak: the --i goes into the branch delay slot14:13
wpwraklarsc: yup ;-)14:14
whitequarkthe return with delay slot is just _weird_14:14
wpwraklarsc: is the jz4720/40 "mips32" ? from what i saw, it's not "mips2".14:15
larscwpwrak: it is14:15
whitequarkwpwrak: it's R4k-compatible, and that is AFAIK mips3214:16
larscbtw. the peripheral bus is clocked at 82 Mhz14:16
larsciirc14:16
Action: wpwrak remembers the day where he was proud to have an SGI with a 12 MHz R2000 on his desk ;-)14:17
whitequarklarsc: it's divided by 3 in 4750, but I don't know what's in 472014:17
larscwhitequark: iirc 414:18
larscbut could be 3 as well14:18
whitequarklarsc: also, I have had a strange feeling that registers are accessed through AHB after reading manuals, however that may be grossly wrong14:19
whitequarksomething in the clock generation unit made me think so, through I don't quite remember what it was14:21
larsci think it is clocked at the same rate14:23
wpwrakyeah, no performance improvement between default architecture and mips32. i.e. the one delay slot has no cost.14:24
whitequarklarsc: the AHB? yes. then, peripheral register accesses won't impose any additional delay14:25
larsci doubt that you can access mmio access without any delay14:30
whitequarkwhy not?14:33
larscbecause that would mean they would be as fast as normal cpu registers14:38
whitequarklarsc: I'd say 'as fast as normal memory14:38
whitequarkand that sounds perfectly fine for me14:38
larscmy 'feelings' tell me it's not the case.14:44
larscbut i can't give you hard numbers14:44
whitequarkhm14:45
larscbut Arned Bergmann told me once that mmio could easily take 100 cycles or more14:47
larschttps://lkml.org/lkml/2011/1/19/8914:47
Jay7whitequark: ping15:04
Jay7whitequark: about tplink15:04
Jay7whitequark: what model you suggests to be able to reflash with something self-build with openwrt e.g.? :)15:04
Action: Jay7 have old enough WR543g here15:05
whitequarkJay7: the latest model, TL-WR1043ND, is very good and cheap enough ($100 here)15:05
whitequarkand supports openwrt very good15:05
whitequark(or vice-versa)15:06
Jay7nice15:06
Jay7I'll consider :)15:06
whitequarkit is also somewhat hackable. the case is easily disassembled (well... after some practice:) and you have at least uart with uboot15:07
whitequarkusb2.0 is good, too15:07
C-Keenthe usb chip on the nanonote can only act as a gadget right?15:08
whitequarkyes15:08
C-Keenpity :)15:09
whitequarki has been told on #mips-linux that mips32 does not require delay slots at all15:52
whitequark*load delay slots15:52
whitequarkuh-oh18:34
whitequarkmy flux is a bit weird18:34
whitequarkis it normal for a neutral flux-gel to eat all the oxide layer in a ~minute after distributing it all over the board?18:35
mththere are branch delay slots, but I never heard about load delay slots18:35
whitequarkmth: check See MIPS Run, they have all that stuff described nicely18:36
jesperjwpwrak: The M36 chip in the phone, MT6223CA, is based on ARM7EJ-S.21:30
jesperjIt would be sweet if linux could run on it, but it is very likely not possible heh21:31
jesperjfound this project though http://opensrc.sec.samsung.com/profile.html which "intends to support almost full 2.6 kernel features on MMU-less ARM architectures"21:31
jesperjThe M36L0T7050T3ZAQ chip turned out to be NOR flash memory: "NOR Density: 128Mb RAM Density: 32Mb"21:33
jesperjI found out all sort of stuff about the phone though21:34
jesperjIt most likely runs "MTK OS"21:35
jesperjI found it interesting that the M36 chip can run 3 LCD modules at the same time21:36
wpwrakjesperj: (3 LCD) wow ;-)21:54
wpwrakjesperj: 32 Mb is only 4 MB. if that's all the RAM in the system, linux won't be very happy21:54
jesperj<wpwrak> jesperj: (3 LCD) wow ;-) < sarcasm?21:57
jesperjuuuugh... damn MegaBIT21:58
jesperj:(21:58
jesperjI read that as MB21:58
jesperjgrr21:58
jesperjThanks for pointing it out though22:01
wpwraki'm sure marketing would *love* an even smaller unit than a bit ;-)22:01
wpwrakmaybe a "hunch", 1/100 bit ? then you'd have 3.2 Gigahunches !22:02
jesperjwpwrak: They'd love it even better if they could make it look like MB/GB/TB... "It has a stunning 2* TerraBite hdd! (* Don't read this but that equals about 20 MB)"22:05
jesperj"Bite" seems like a good candidate22:06
wpwrakah yes, Bite would be good. or l6 TB, with l6 = 2^-2222:12
GNUtoouhmmm22:15
GNUtoonot sure they could do import that in france, since the word is vulgar there22:15
GNUtooif they do it would be very problematic....lol22:16
wpwrakGNUtoo: reminds me of that spanish girl who shouted encouragements ("vite ! vite !") to someone, while others looked in disbelief :)22:21
GNUtoohttp://fr.wiktionary.org/wiki/bite22:23
GNUtoolook in the vulgar section22:23
GNUtoobut maybe you don't speak french22:23
wpwrakGNUtoo: oh, i know it. et oui, j'ai vecu pour huit ans a Lausanne ;-)22:25
GNUtoook22:25
wpwrakGNUtoo: (which is where the aforementioned incident occurred - in spanish, the pronunciation of "v" and "b" is considered interchangeable ;-)22:26
GNUtooahh22:29
GNUtoonow I understand22:29
GNUtoolol22:29
whitequarkjesperj: you can still very well use some RTOS on that chip, it'd be enough for telephony stack to run (if you manage to assemble one)22:47
jesperjwhitequark: cool22:58
--- Tue Apr 12 201100:00

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