Previous day | Jump to hour: 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 | Next day

Seconds: Show Hide | Joins: Show Hide | View raw
Font: Serif Sans-Serif Monospace | Size: Small Medium Large

Click in the nick column to highlight everything a person has said.
The Logo icon identifies that the person is a core developer (has commit access).

#rockbox log for 2010-02-20

00:00:08pixelmanot at the moment, have to think about it
00:00:40 Part anethema
00:00:53pixelma() is not nice as you will need %( and %) then and using braces is quite common
00:01:16JdGordon_not necessarily
00:02:35AlexPJdGordon_: Now I get 716 if (id < SYSTEMFONTCOUNT || id >= MAXFONTS-1) 717 return WPS_ERROR_INVALID_PARAM; 725 }
00:03:05JdGordon_the numbers are?
00:03:30AlexPsorry, one mo
00:03:40 Join Rob2222 [0] (
00:04:05JdGordon_oh, gdb output? that means the the font is being loaded into the id for the remote user font
00:04:31JdGordon_oh bloody hell
00:04:34*JdGordon_ appologises
00:04:44JdGordon_yeah, but that line back.. bloody stupid of me
00:05:00JdGordon_FFS :p
00:05:02AlexPheh, OK :)
00:05:15JdGordon_id -= SYSTEMFONTCOUNT; should be id -= FONT_UI
00:05:24JdGordon_right area, wrong fix :p
00:05:26pixelmaand I still want preset name on it's own line or only with the preset number max as it is in SVN (line too long)
00:05:26AlexPOK, one mo
00:05:33pixelmaand scrolling lines
00:06:39 Quit Bagder (Ping timeout: 260 seconds)
00:07:02pixelmaoh, I said I'll take care of that... but only if own FMS don't need a hard reset for me
00:07:11 Quit Rob2223 (Ping timeout: 256 seconds)
00:07:31pixelmaremoving %pm from my FMS didn't help
00:07:38*JdGordon_ doesnt understand the but part?
00:08:10AlexPJdGordon_: p id $1 = 1 p font_ids[id] $2 = 3
00:08:32AlexPShould I see if it works now?
00:08:39JdGordon_yep, just hit c
00:09:10pixelmaa hard reset is the only way to exit the radio screen with an own FMS is a hard reset currently (on my Ondio) so it's not nice to test FMSs
00:09:54 Quit perfectdrug (Quit: perfectdrug)
00:10:48*JdGordon_ heading home
00:11:00JdGordon_please nmentino that in the tracker incase I forget
00:11:05 Quit JdGordon_ (Quit: Miranda IM! Smaller, Faster, Easier.
00:11:57 Quit ender` (Quit: Connection Reset by Gypsies with Wire Cutters)
00:13:44 Join kugel [0] (~kugel@rockbox/developer/kugel)
00:13:50AlexPJdGordon: It now gives an error on parsing if I misspell the font in the %Fl, but the wps still doesn't use it
00:14:04*pixelma wonders about something
00:21:20 Quit mt (Quit: ChatZilla 0.9.86 [Firefox 3.5.7/20091221164558])
00:25:45 Quit MaadMan (Quit: Verlassend)
00:26:14 Join ecio_ [0] (~ecio@
00:26:31 Nick knine is now known as kaniini (
00:28:07 Quit freddyb (Ping timeout: 260 seconds)
00:29:59 Quit ecio (Ping timeout: 260 seconds)
00:30:00 Nick ecio_ is now known as ecio (~ecio@
00:30:40pixelmaJdGordon: any own FMS with a FMS tag makes it break. I just tried "..." as the only things in the FMS and it works - whereas "%Tn" as the only thing doesn't
00:31:00 Join Farthen [0] (
00:31:14 Join Omlet05 [0] (
00:31:32pixelmaalthough it is correctly parsed and shows me the preset name
00:31:37 Quit Omlet05 (Client Quit)
00:33:00 Quit Tomis (Read error: Operation timed out)
00:33:27 Quit huelk (Read error: Operation timed out)
00:33:50 Join Sajber^ [0] (~Sajber^
00:34:18 Quit robin0800 (Remote host closed the connection)
00:34:25pixelmaand not sure if it is relatad (guess not) but I still get unused variable warnings in radio.c (line 592, variables in question are "minutes" and "hours"
00:34:26 Join Tomis [0] (~Tomis@
00:34:28 Quit Omlet (Ping timeout: 276 seconds)
00:39:27 Quit piotrekm (Quit: piotrekm)
00:42:33TucknDargevaerts, I've tried to get the card in my mini to be as thick as the original hard drive, but I can't get the click working... do you by any chance have any other suggestions?
00:43:28pixelmaI think people successfully used pieces of cardboard to fill the gap
00:44:39TucknDaryeah, I'm trying but no... weird. guess I'll have to keep experimenting.
00:45:11JdGordonAlexP: bloody hell :p
00:45:31CIA-8New commit by funman (r24778): rbutil: link to the Sansa forums when the user is requested to download a Sansa AMS firmware
00:46:16 Quit petur (Quit: Zzzzzz)
00:49:19CIA-8New commit by jdgordon (r24779): fix possible out-of-bounds error on remote lcd targets if they try loading a font to id==2
00:49:52 Quit Tomis (Read error: Connection reset by peer)
00:50:57JdGordonTucknDar: I didnt use anything to fill the gap.. my clickwheel still works fine
00:51:07 Join Tomis [0] (~Tomis@
00:51:33TucknDarmy doesn't... another test now though. Thanks for your replies, btw! Appreciate it!
00:51:53JdGordonminig1 or g2?
00:52:15JdGordonsame, yeah, some thick card should do the trick
00:52:33pixelmahopefully you didn't damage the cable?
00:52:37 Join Strife1989 [0] (
00:52:58TucknDardon't think so, as it works with apple firmware, which I obviously don't want to use ;)
00:54:37CIA-8New commit by jdgordon (r24780): and actually fix multifont on remote lcd targets also
00:54:53JdGordonAlexP: there you go!
00:55:10AlexPthank you :)
00:55:13pixelmaJdGordon: I just tested the pm on its own line theory with a test.wps containing %?mh<%pm|AAA> (and on a second line with %?mh<HHH|AAA> to see if the hold tag works and it does) - and the first line always shows the peakmeters, whether on hold or not...
00:55:15 Join karashata [0] (
00:56:08JdGordonok, I guess thats not really surprising. did we ever expect %pm to work in conditionals?
00:56:44pixelmayou seemed to :P
00:56:52JdGordonapart from me :)
00:56:53***Saving seen data "./dancer.seen"
00:57:00JdGordon(who knows nothing about actually using the skins)
00:57:17 Nick Strife1989 is now known as Strife89|Desktop (
00:58:06 Quit jgarvey (Quit: Leaving)
00:58:30AlexPJdGordon: ta, multifont works nicely now :)
00:58:37JdGordonweeeee :)
00:59:01AlexPI'll try out fms shortly
01:02:04 Join crwl [0] (
01:05:10 Join CaptainKewl [0] (
01:05:23 Join CaptainKwel [0] (
01:07:22 Quit Strife89|Desktop (Quit: Log out, log in.)
01:08:31 Join Strife1989 [0] (
01:09:39saratogaha the new mdct library flips the order of the input and output arguments
01:09:56 Join tomers [0] (
01:10:00 Quit CaptainKewl (Ping timeout: 264 seconds)
01:12:07 Nick Strife1989 is now known as Strife89|Desktop (
01:14:38 Quit tomers (Client Quit)
01:16:22 Quit TucknDar (Quit: Miranda IM! Smaller, Faster, Easier.
01:24:03AlexPJdGordon: limits exceeded! :
01:24:10AlexPWhere do I increase them?
01:24:18JdGordonwhich limit?
01:24:34JdGordonprobably apps/gui/skin_engine/skin_buffer.c at the top
01:24:54AlexPdunno, the sim just says ERR: Failed parsing on line 31 : ERR: Limits exceeded
01:25:03AlexPAlthough neither my wps or fms have a line 31
01:27:32JdGordonyeah, probably that file ten
01:27:41AlexPhang on
01:27:43 Quit bertrik (Ping timeout: 246 seconds)
01:29:32CIA-8New commit by saratoga (r24781): Use new MDCT library for libfaad. Speeds up AAC-LC by 2.5MHz.
01:30:20 Join BHSPitMonkey [0] (~stephen@unaffiliated/bhspitmonkey)
01:32:06saratogaI guess we can remove the old MDCT now
01:32:12saratogaits not used for anything
01:32:28JdGordonAlexP: are you using different backdrops for the 3 possible skins? (sbs/setting, wps, fms)?
01:32:47AlexPJdGordon: no sbs and at the moment the fms and wps are using the same one
01:32:52AlexP(or trying to)
01:33:05JdGordonlots of images? or fonts?
01:33:26AlexP1 additional font in the wps, two in the sbs
01:33:40AlexPand not many images, no - just vol, battery etc from cabbiev2
01:33:43JdGordon30K, shouldnt be such a big deal
01:34:10AlexPA slight ofddity is the error at line 31 - the wps has 24 lines, the fms 23
01:34:17JdGordonwe really need to get better debug info from skins :/
01:34:28AlexPsorry, two additional fonts in the fms not sbs
01:34:38AlexPThere is no sbs
01:34:41JdGordonremote target though, so the remote maybe?
01:34:57AlexPah, possibly - I haven't touched that yet
01:35:55AlexPyes, the cabbie rwps has 34 lines :)
01:36:19AlexPI'll cut that down into one of my own with a sngle line or something and see if it goes away
01:37:30AlexPyes, that was it
01:37:51AlexPbut without remote wps or fms I'm already at 38620/44032
01:38:04AlexPwithout any images to speak of, and three additional fonts
01:38:21AlexPwell, with a few images
01:38:26AlexPbut not going mental :)
01:39:16 Join phanboy4 [0] (
01:39:58JdGordonthat seems too high
01:40:09Unhelpfulamiconn: pong?
01:40:17AlexPI'll finish up and then upload somewhere
01:40:21 Join kimi-sharamin [0] (
01:40:28JdGordonwhich target?
01:41:12amiconnUnhelpful: I don't understand your ARMv6 division optimisation (udiv32_arm.S lines 241ff)
01:41:56saratogaoddly enough on the nano2g rbutil said i already had a bootloader installed, and i'm pretty sure i didn't
01:42:15amiconnYour're using smmla (should be equivalent to the smlal with low word cleared used on ARM <v6, which is understandable and saves two cycles
01:42:34AlexPJdGordon: I'm going to bed in a bit - I'll have a play tomorrow then bug you on Sunday :)
01:42:44JdGordonAlexP: easiest fix is apps/gui/skin_engine/skin_buffer.c line 58, change that 2* to 4* or something
01:42:45amiconnBut then you're using smmul, which needs correction if the sign bit is set
01:42:46 Quit karashata (Ping timeout: 265 seconds)
01:42:51JdGordonsounds good
01:43:31amiconnHowever, tst + not-taken branch + smmul is even one cycle more than just using umull
01:44:58Unhelpfulamiconn: the reciprocal won't have the sign bit set, and the test ensures that the numerator does not either. this benches faster on beast, but whether it's faster in practice probably depends on the frequency with which the branch is taken.
01:45:02 Quit Tomis (Read error: Connection reset by peer)
01:45:26saratogaoh shit plugged my ipod into a sansa USB cable
01:45:52amiconnI understand what the test does, however, I would expect umull to be faster here, as it doesn't need the test + branch
01:46:38saratogastrange that they fit so well
01:46:43Unhelpfulalso both test and branch should fit into result delays for multiplies - the smmla uses the result of the mul, which means it can't execute for 3c (iirc) after the mul completes.
01:48:05 Quit Farthen (Read error: Connection reset by peer)
01:48:14Unhelpfuland the smmla result is used in the following smmul, so the smmul will block for 3c if reached as well
01:48:49Unhelpfulbut unless i've misunderstood the arm1136 manual and ASDG the tst and bmi can execute without penalty during these delays?
01:49:31amiconnHmm, that is true
01:50:02Unhelpfuli should probably remove the note about this being an APE-specific optimization as well... from what i see it should be successful as long as the branch is not taken *too* often and is predicted successfully most of the time.
01:50:17amiconnThis whole streak of multiplications depending on each other's result is rather bad for the arm11 pipeline
01:50:19 Join Tomis [0] (~Tomis@
01:51:18amiconnWhat I don't understand as well is why this case would only account for 1 in 10^7 cases
01:51:22Unhelpfulit is, but it give me lots of places to put tests and branches for special cases.
01:51:47Unhelpfulah, well, it does on APE (i counted) which is why i said i should remove the note about this being APE-specific.
01:52:39amiconnI would expect this to occur much more often with random test data
01:54:31Unhelpfulthe test set my benchmark uses isn't random, either.... but are neither are most actual divisions, and i suspect that in typical uses sign bit is set <50% of the time. values in actual use are probably more likely to be evenly distributed among lengths than across all possible values
01:57:07 Quit BHSPitMonkey (Remote host closed the connection)
01:58:19Unhelpfulbut even if the branch is taken exactly half of the time, it shouldn't be a *huge* expense... ASDG gives branch as taking 1c on correct prediction, 4c on incorrect prediction, but it's in a 3c delay slot already. also the failed-prediction case is given as needing flags 2c early, but since the smmla between tst and bmi is a 2c instruction, that should be satisfied as well.
02:00:36Unhelpfulthe test set i used for benchmarks consists of 1-bit, 2-bit, 3-bit, etc up to 32-bit values, one of each, and each repeated at each possible left-shift position... so there is exactly 1 1-bit value in the set, 2 2-bit values, 3 3-bit values, etc. the benchmark divides every possible combination of values in the set. i could easily run with a pre-generated array of random numbers, instead, though.
02:08:32 Quit moos (Ping timeout: 260 seconds)
02:12:35Unhelpfulthere's also another way to handle the "main" division path (the one followed if none of the special-case branches occur). a goldschmidt divider takes two multiplies per iteration, but they can be done independently. this should reduce stalling on arm11 and should be almost free of stalls on arm9e, and can use the same initial-estimate table as the N-R version (and therefore the same trick of using the estimate-table value directly for
02:12:35Unhelpful large enough divisors).
02:13:50Unhelpful*however* goldschmidt does not converge as quickly as N-R - two iterations are required to get as good an approximation, and the second one would need to use umull on arm9e.
02:17:07Unhelpfulif it's placed carefully perhaps a test/branch could cheaply avoid the second iteration when the divisor is the right size. i never got to tuning the goldschmidt divider that finely, and then the implementation got lost (i hadn't committed what i *thought* i had comitted)
02:18:09 Quit komputes (Remote host closed the connection)
02:29:53 Quit GeekShadow (Read error: Connection reset by peer)
02:34:29 Quit kimi-sharamin (Read error: Connection reset by peer)
02:35:30 Join Kitr88 [0] (
02:36:30 Join karashata [0] (
02:37:19 Quit kugel (Remote host closed the connection)
02:37:58Unhelpfulfunman: i modified your test to print amount written every 256B while writing to iram. it never prints anything. now trying with print every iteration...
02:38:18Unhelpfulhangs after writting 0x20 bytes
02:38:35 Quit Kitar|st (Ping timeout: 252 seconds)
02:40:28 Quit Kitr88 (Ping timeout: 272 seconds)
02:45:36 Join Kitar|st [0] (
02:46:02Unhelpfulthe last address it writes to successfully seems to be 0x81000018
02:49:09 Quit archivator (Quit: Leaving)
02:51:22Unhelpfulleavittx: *real* 3d accel support? or only some 2D blend/blit/scale stuff like the beast's IPU can do? if it can draw textured tris or quads, even with only 2D coordinates, that might at least be something to start with.
02:56:56***Saving seen data "./dancer.seen"
03:04:59CIA-8New commit by uchida (r24782): commit FS #10424 and FS #10425 ...
03:07:56 Join ucchan [0] (
03:11:12 Part fleebailey33
03:15:56ucchanThereis a question. My task(FS #10424, FS #10425) cannot be close because the authority doesn't exist.
03:16:00ucchanHow should I do to close my task?
03:19:51 Quit nimak (Ping timeout: 256 seconds)
03:22:58 Join robin0800 [0] (
03:23:08 Join nima [0] (
03:24:32 Quit ps-auxw (Ping timeout: 248 seconds)
03:25:04 Join ps-auxw [0] (~arneb@2001:470:c807:0:1532:4e5f:2ad3:4123)
03:26:28 Quit robin0800 (Client Quit)
03:26:55 Join robin0800 [0] (
03:27:45Unhelpfulucchan: are they committed, or is there some reason they ought to be rejected without further consideration?
03:42:08 Quit GodEater (Ping timeout: 248 seconds)
03:43:12 Quit robin0800 (Remote host closed the connection)
03:44:11ucchanI committed the patch that exists in these tasks, then I want to close them.
03:44:51 Quit Adnyxo (Quit: Leaving)
03:48:00 Quit ps-auxw (Ping timeout: 248 seconds)
03:50:53 Join ps-auxw [0] (~arneb@2001:470:c807:0:1532:4e5f:2ad3:4123)
03:52:16 Quit ucchan (Quit: Leaving...)
03:57:39 Join Schmogel [0] (
04:05:21Unhelpfulah... you probably need developer rights on the tracker
04:09:38 Join S_a_i_n_t [0] (S_a_i_n_t@
04:11:40 Quit TheSeven (Disconnected by services)
04:11:53 Join The_Seven [0] (~theseven@rockbox/developer/TheSeven)
04:12:03 Nick The_Seven is now known as TheSeven (~theseven@rockbox/developer/TheSeven)
04:16:18 Quit phanboy4 (Read error: Connection reset by peer)
04:20:57 Quit mikroflops (Remote host closed the connection)
04:21:04 Join mikroflops_ [0] (
04:26:37 Quit Schmogel (Read error: Connection reset by peer)
04:33:53 Join DerPapst1 [0] (
04:35:39tmztsaratoga: do you know what pins are different? I'm considering getting a radio with dtwo docks but want sansa and a touch to work. it would also be interesting to see AAP support on the sansas
04:36:01 Quit DerPapst (Ping timeout: 248 seconds)
04:41:49 Quit efyx_ (Remote host closed the connection)
04:43:20 Quit S_a_i_n_t (Quit: There are 10 types of people, those who understand binary, and those who don't.)
04:45:48 Quit MethoS- (Remote host closed the connection)
04:46:31 Join Barahir [0] (
04:49:55 Quit Barahir_ (Ping timeout: 252 seconds)
04:51:39 Join S_a_i_n_t [0] (S_a_i_n_t@
04:53:39 Nick jfc^2 is now known as jfc (
04:54:56 Quit DerPapst1 (Quit: Leaving.)
04:57:00***Saving seen data "./dancer.seen"
05:05:01 Quit ecio (Quit: Colloquy for iPhone -
05:05:29 Join ecio1 [0] (
05:07:44leavittxUnhelpful: I don't know details. Even google didn't helped me. But it seems to me that real. :)
05:09:30 Part froggyman
05:50:09 Quit tmzt (Ping timeout: 248 seconds)
05:51:05 Join tmzt [0] (
05:51:45 Quit shai_ (Ping timeout: 248 seconds)
06:00:54 Nick ecio1 is now known as ecio (
06:31:43 Quit karashata (Quit: The fluffy dragon has left completely!)
06:35:29Unhelpfulamiconn: i tried replacing the 528 selected values with 528 random ones. handling large numerators in a separate branch is still faster, though only marginally. the random values were generated in python, which uses MT, so they should be quite random enough.
06:56:01 Quit panni_ (Quit: ( :: NoNameScript 3.81 :: ))
06:57:04***Saving seen data "./dancer.seen"
07:04:08 Join CPICH_ [0] (
07:05:28 Quit ecio (Ping timeout: 246 seconds)
07:17:58 Join ecio [0] (~ecio@
07:18:55 Quit CPICH_ (Ping timeout: 245 seconds)
07:21:33 Quit nima (Read error: Connection reset by peer)
07:21:34 Join nimak [0] (
07:24:35 Join moos [0] (moos@rockbox/staff/moos)
07:27:19 Quit martian67 (Ping timeout: 268 seconds)
07:29:27CIA-8New commit by unhelpful (r24783): Clarify comments in ARMv6 divider regarding special-case handling of large (high bit set) numerators.
07:35:14 Join Horschti [0] (~Horscht2@xbmc/user/horscht)
07:39:02 Quit Horscht (Ping timeout: 268 seconds)
07:39:13 Join Guest88353 [0] (
07:42:48CIA-8New commit by tomers (r24784): Comment out lcd_drawline() DEBUGF messages which show in various simulators
07:46:00 Join CaptainKewl [0] (
07:49:35 Quit CaptainKwel (Ping timeout: 252 seconds)
07:54:45 Join tomers [0] (
08:03:00 Join stoffel [0] (
08:05:35 Quit liar (Ping timeout: 245 seconds)
08:09:59amiconntomers: Those debug messages were there for a reason.
08:10:23tomersamiconn: what what the reason?
08:10:56tomersamiconn: s/what/was/
08:11:03amiconnSeveral plugin developers just used lcd_drawline() in places where they should have used lcd_hline() or lcd_vline() for performance reasons
08:11:39tomersamiconn: Should the 'user' of the simulator have to see all these debug messages?
08:11:46amiconnThose plugins showed *massive* output from these messages, so I was able to find and fix them
08:12:20amiconnI left those messages in place in case it will happen again
08:12:40tomersbut it does happen often today
08:12:58tomersdoes those prints indicate something is wrong?
08:13:12tomersor need to be addressed to?
08:13:46tomersyou use the simulator for your own reasons and then you get all that debug output which seems irrelevant...
08:13:58amiconnOf course there are some calls to lcd_drawline() which do *occasionally* show this message (because a random line is exactly horizontal or vertical by coincidence)
08:14:29tomersso do you want me to revert?
08:14:56tomersdon't we need to keep the simulator output clean?
08:15:15amiconnThe latter is unfortunately unavoidable for this optimisation hint, the former is something that should be fixed in the plugin
08:16:17amiconnWhy should we need to keep the simulator out put clean? The simulator is meant for developers, and debug output should point out problems
08:17:03tomersyeah but most developers want the output to show whatever is relevant for them.
08:17:22amiconnThe question is whether this optimisation hint should be kept enabled all the time, but if you really found plugins which output lots of those messages, you should rather check those plugins whether they can be optimized...
08:17:22tomersthey don't want to get usb messages, lcd messaged, storage messages, etc.
08:19:16tomersi think that most developers (like me) are not aware of that, and even if they see it they don't really know it is a hint for them to use lcd_[vh]line for optimisation. I guess that from time to time someone should enable these messages and review all plugins
08:19:34tomersbut that's not going to happen, i guess
08:20:16tomersmaybe i'll wrap these messages with #ifndef HIDE_LCD_LINE_WARNING or something?
08:21:19amiconnNah. There are already more than enough ifdefs
08:21:24tomersso i guess i should revert this commit so that these messages will show by default?
08:22:28amiconnHmm. I would tend to say yes, but your other concern is also valid.
08:23:02amiconnI wonder if there are other similar debugging aids which are forgotten by now because they're disabled
08:23:43amiconn(this one is an optimisation hint though, it doesn't point out functional bugs)
08:25:01amiconnDo you remember which plugins printed this message a lot?
08:25:21tomersamiconn: maybe we should re-enable these prints, and change the message to something like '<filename>:<line>: consider using lcd-[hv]line instead...', and use a single macro for that
08:25:33tomersamiconn: I can't remember. sorry.
08:26:12amiconnWhat would 'filename' and 'line' be?
08:26:24tomers__FILENAME__ and __LINE__
08:27:02tomerse.g. firmware/drivers/lcd-16bit-vert.c:361
08:27:21amiconnThose would always be the same for a certain sim, e.g. lcd-16bit.c and 354 or 361
08:28:34amiconnBut that would be misleading imo. The caller (e.g. a plugin) should use the specialised versions if possible
08:28:55tomersCurrent message, "lcd_drawline() called for vertical line - optimisation.\n" is rather confusing. 'optimisation' sounds like it notifies you of an optimisation that took place, not that you should consider using other function for performance reasons.
08:29:31amiconnIt does both
08:30:14tomersI don't understands ^^^, can you elaborate?
08:30:51amiconnlcd_drawline() takes this optimisation, and notifies you that you can save the decision overhead if you call lcd_hline() and lcd_vline() directly
08:31:47tomersi got that part, but I didn't get what you said about the caller (e.h. a plugin), do you mean that it should print "bubbles:100" instead of lcd-16bit.c:200 ?
08:31:58Unhelpfulit looks like goldschmidt can produce an inverse in 4 multiplies, the same number as newton-raphson. pros: multiplies in goldschmidt are on two values and can be carried out (somewhat) independently, for large-enough divisors we could potentially do one iteration instead of two. cons: there are some shifts require between iterations if we don't want to see performance destroyed.
08:32:25amiconnYes that's what I meant, but afaik this is not possible
08:33:28Unhelpfulamiconn: it's *somewhat* possible - wrap lcd_drawline in a macro of the same name that tests and prints the message. since the macro exists at the callsite it should output the correct filename/line.
08:34:06amiconnBut the idea of printing the location where a debug message comes from is good, imo.
08:34:23amiconnIt makes it easier to identify them if you know where they come from
08:35:51amiconnUnhelpful: (1) That won't work for plugins though, as they are loaded dynamically
08:36:13*Unhelpful will try implementing goldschmidt in arm asm again and see where it gets him
08:36:46Unhelpfulamiconn: i wonder what happens if you try to #define rb->lcd_drawline(...) ... nothing nice probably.
08:37:15amiconnI would expect that the preprocessor complains
08:37:32amiconn- and > are not valid in macro names afaik
08:37:51Unhelpfulindeed they're not.
08:39:01amiconnHmm. Enabling the debug messages now is more difficult than disabling them if they're enabled by default
08:39:10Unhelpfulyou could also have lcd_drawline take the filename and line number as args on simulator builds, and then the same define would work for plugins.
08:40:01Unhelpfulie #define lcd_drawline(args) lcd_drawline(args, __FILENAME__, __LINE__)
08:41:51amiconnNot always
08:42:38amiconnThere are plugins which call lcd_drawline() via a function pointer
08:42:50 Join petur [0] (~petur@rockbox/developer/petur)
08:43:46amiconnAt least these plugins are calling several lcd_* functions via pointer - I'm not sure whether lcd_drawline() is among those.
08:47:14 Nick kaniini is now known as knine (
08:56:45 Join linuxstb [0] (~linuxstb@rockbox/developer/linuxstb)
08:57:07***Saving seen data "./dancer.seen"
09:02:08Unhelpfulthe cutoff point for stopping at 1 goldschmidt iteration appears to be divisors >= 0x9e000 (that's the smallest such cuttoff that is a valid arm immediate, anyway).
09:05:05 Join flydutch [0] (
09:09:21 Quit CaptainKewl (Ping timeout: 272 seconds)
09:19:22leavittxDevs: why didn't you finally swiched to git?
09:19:45leavittxThe git mirror of svn is nice, but...
09:21:53Unhelpfulleavittx: serial numbering of revisions for one thing - you can always tell if one build is newer/older than another. and it's already svn. and devs who want to use git already can - it's not like being able to git push would be so very much easier than git svn dcommit.
09:22:05Unhelpfulalso i believe the majority of devs prefer to use svn
09:23:06 Join bmbl [0] (~Miranda@unaffiliated/bmbl)
09:23:59 Join TopyMobile_ [0] (
09:27:47 Quit TopyMobile__ (Ping timeout: 256 seconds)
09:44:11 Quit tomers (Ping timeout: 272 seconds)
09:48:26*pixelma wonders if USB HID should really be enabled by default
09:50:31pixelmaafter experiencing myself that some OSs can't cope with it (which is why the option to turn it off was implemented) and reading some reports from other people, I start thinking that the default should be off - at least for the release
09:51:24 Quit Zarggg (Ping timeout: 264 seconds)
09:52:06pixelmaHID is a plus and I never heard of any other MP3 players that offers it, whereas USB data connection is something everyone expects to be working
10:06:58 Join bertrik [0] (
10:14:27 Join Bagder [0] (~daniel@rockbox/developer/bagder)
10:16:18ThomasAHpixelma: I don't know about the effects of having the player as HID, but the hard disc accelerometer being presented as a joystick device caused some troubles for me on LInux
10:17:16 Join stripwax [0] (
10:27:05 Join ender` [0] (
10:37:20leavittxWhy doesn't rb->lcd_clear_display(); clear display after rb->splash?
10:39:24leavittxok, how to clear display after rb->splash()?
10:39:41amiconnYou need to update afterwards
10:40:27amiconnlcd_clear_display() only changes the framebuffer, like all lcd_* drawing functions
10:43:29leavittxamiconn: what do you mean by "update afterwards"?
10:43:47 Join DerPapst [0] (
10:48:49*stripwax doh's - new MDCT doesn't have an icode_attr - that'll give an extra MHz-or-so bonus..
10:49:36leavittxstripwax: you mean rb->lcd_update, right? Not helps. Look smb on my code plz:
10:50:42stripwaxleavittx - I don't see you calling rb->lcd_clear_display - you need to call both lcd_clear_display AND lcd_update, right?
10:51:06stripwaxI don't know what "look smb on my code plz" means. Please use real words in this channel, helps.
10:51:57leavittxSorry, will not repeat anymore :-\
10:53:55 Join m3dlg [0] (~m3dlg@
10:57:10***Saving seen data "./dancer.seen"
10:58:17stripwaxlooks like wma about 5.5Mhz faster on coldfire (h120) with icode_attr in new mdct. will double check vs unchanged, but can't believe we missed that!
11:00:45stripwaxin fact it can *almost* run unboosted at 40Mhz at 96kbps ..
11:00:55leavittxQuestion: after calling lcd_clear_display() display becomes black. How do I restore normal rockbox background with logo?
11:10:45 Quit m3dlg (Ping timeout: 240 seconds)
11:11:35stripwaxleavittx - if you can think of another (existing) rockbox plugin that has the behaviour that you want to copy, then you can look at the sourcecode for that other plugin and see how it is done..
11:12:55pixelmaI guess you have to keep an eye on the drawmodes (and foreground/background definitions)
11:15:38CIA-8New commit by uchida (r24785): libpcm: linear pcm decode logic separates according to each bitspersample, endian, and signess.
11:24:36 Quit stoffel (Remote host closed the connection)
11:50:46stripwaxcorrection - it runs almost unboosted at 40Mhz at *128* kbps. h120 wma at 96kbps only 37MHz
11:52:10 Join Buschel [0] (
11:52:21Buschelstripwax: \o/
11:52:42Buschelstripwax: what clock was needed before mdctexp?
11:54:46stripwaxBuschel - 43Mhz or more I think. I actually don't think I still have those timings to hand though :(
11:55:02stripwaxThat's 43Mhz for 96kbps.
11:55:42stripwaxI'm tempted to do a rebuild of the old code and put a definitive comparison on the wiki page ..
11:58:24Unhelpfulhrm, so the estimated division result may be low by as much as 2, or high by as much as 1. wonder how best to correct that...
11:59:28stripwaxExcellent, ICODE gives us another 0.5Mhz boost on arm also. So 96kbps on ipod video now 24.99MHz :)
12:00:19CIA-8New commit by stripwax (r24786): Adding ICODE for imdct (and its constituent ifft bits) gives 0.5MHz boost on arm (ipod video) and about 5MHz boost on coldfire (H120)
12:01:42stripwaxthat svn comment really makes no sense at all, sorry. 0.5MHz boost on ipod vorbis, and 5MHz on h120 wma.
12:02:10stripwax(measuring ipod wma improvement now, suspect minimal though)
12:02:18Buschelstripwax: the results are really impressive! :o)
12:02:20Buschelgood work!
12:03:21*Buschel needs to optimize mpc again to keep the performance headroom to "other" codecs :)
12:13:02stripwaxBuschel - according to this, we're now a clear 2MHz faster on coldfire vorbis, and 7MHz faster on coldfire wma, versus 3 months ago.
12:13:33stripwaxa bit surprised that coldfire vorbis hasn't seen as big an improvement, but I have a suspicion about that (revtab being in ram not iram..)
12:15:50 Quit bertrik (Ping timeout: 276 seconds)
12:19:06 Join m3dlg [0] (~m3dlg@
12:19:29 Join kugel [0] (~kugel@rockbox/developer/kugel)
12:19:46kugelstripwax: I think most of the time you should s/arm/PP/
12:21:06Unhelpfulhave you tested on arm9tdmi, arm9e, and arm11? arm9tdmi has 1c loads (with delayed result availability) and arm9e/arm11 have the 2c multiplier w/ respectively 1c and 2c delay for result after the multiply completes execution
12:21:54stripwaxkugel - probably right, good point
12:22:11Unhelpfulon all three you can benefit from loading data well before it's needed, and on the last two putting in non-dependent ops around multiplies will help
12:22:21stripwaxUnhelpful - nope. I have tested only on the targets I have access to (h120 and ipod video). Can you, please?
12:22:54stripwaxluckily I didn't hand-optimise *everything*, so gcc should be able to do some suitable scheduling
12:23:30 Quit m3dlg (Ping timeout: 245 seconds)
12:24:00Unhelpfulstripwax: i'll try to find time this weekend, then. there aren't any very-working arm9e targets, and i don't have any arm9tdmi, but i can benchmark on the beast... just checkout the mdct_exp branch and build?
12:24:46kugelit's in trunk now I believe
12:24:56stripwaxyep, it's all in trunk...
12:25:21stripwaxso you'd probably be best to test_codec before you upgrade (or build an old version, or something)
12:25:39Unhelpfuloh... hrm, i assume we made sure it was at least not slower on non-PP before moving it to trunk? ;)
12:25:57Unhelpfulif you pester somebody w/ nano2g i believe that's arm9tdmi
12:26:14stripwaxYes, we made sure it was faster on both PP and coldfire (I tested h120 and I believe someone else tested H340)
12:26:38Unhelpfulyes, but not all ARM targets are PP ;)
12:26:38stripwaxUnhelpful - oh, cool!
12:27:02 Join MethoS- [0] (~clemens@
12:27:09stripwaxTheSeven - would you be able to check test_codec performance on nano2g versus a few revisions ago? (i.e. before mdctexp committed to trunk)?
12:27:13kugelfx and samsa are 9tdmi also
12:27:25stripwaxUnhelpful - right. saratoga tested some kind of e2x0
12:27:31stripwaxI don't know if that's pp
12:27:55TheSevenstripwax: sure, if you can name me an exact revision number that i should test
12:28:16Unhelpfulif you look in tools/configure targets that are arm9xx*bunchofjunk*e are arm9e, other arm9 targets are arm9tdmi
12:28:26 Join Omlet [0] (
12:29:17TheSevenand can anybody point me to a link to those codec test files? they are being referred to on the codec performance page, but there is no link (or i didn't find it)
12:29:39amiconnUnhelpful, stripwax: ARM9e is e.g. Cowon D2, and the D2 is usable for this kind of tests
12:29:42stripwaxTheSeven - fab. ok so r24786 versus r24711 (although I see some nano2g fixes put in after 24711 so I don't know if 24711 works for you)
12:30:07TheSevenah, i did miss the underscore when guessing :-)
12:30:12amiconnAfaik the D2 can even be used to play music from SD, just the internal flash isn't usable (ftl missing)
12:30:22kugelthegeek: sorry :(
12:30:45stripwaxanyone other than shotofadds have a D2?
12:30:47Unhelpfulamiconn: that might need some fixing to test_codec? doesn't it always write logs in /?
12:30:47TheSevenstripwax: should all be minor stuff (power management etc.)
12:30:53stripwaxor, shotofadds, if you're listening..
12:31:01stripwaxTheSeven - ah ok
12:31:08amiconnUnhelpful: Well, gevaerts did several tests on D2 for me already
12:31:20kugelstripwax: tomers, gevaerts and Llorean have one
12:31:22amiconnAnd ARM9 v4 can be tested on Gigabeat F/X
12:31:39 Join stoffel [0] (
12:31:44Unhelpfulamiconn: remind me to bother him about these dividers, he needs more excuses to make up division puns, anyway :)
12:32:12Unhelpfulalthough i guess i can benchmark them well enough on clipv2, just not via the plugin i wrote for it.
12:32:32TheSevenstripwax: should i test with/without eabi?
12:32:41amiconnKeep in mind that the D2 has no iram (actually it does, but afaik nobody found out how to use it properly)
12:32:57stripwaxTheSeven - whatever works best for you. I didn't use eabi
12:33:03amiconnSo no small-dividers special table...
12:33:06 Join piotrekm [0] (~piotrek@unaffiliated/piotrekm)
12:33:43TheSevenstripwax: they both work fine - so i'll probably do all 4 combinations, so that we also have a result whether eabi makes a difference
12:33:46 Join TheSphinX^ [0] (
12:33:50Unhelpfulamiconn: we don't use that on arm9e anyway
12:33:57amiconnGigabeat F/X also uses no iram - iirc the reasoning was that it's too small to be of use on swcodec
12:34:30kugel4k only
12:34:44amiconnIirc the S3C2440 has 4KB of IRAM. On SH1, which has the same amount, we do use it though - and it's very beneficial
12:34:57amiconnBut then SH1 has no cache
12:35:20kugelmaybe it's used for target specific stuff, I can imagine rolo runs from it
12:35:31TheSeventhat s3c2440 actually is only present because they need it as as steppingstone buffer :-)
12:35:48Unhelpfulthe only divider specialization for APE on arm9e/arm11 is that if there's iram a copy of the divider goes there and is used by the UDIV macro
12:35:54TheSeventhat iram on*
12:39:22amiconnIt's very interesting that the benefit of using ldm varies wildly between arm versions and variants
12:39:53amiconnOn arm7tdmi it is *very* useful in that a single load takes 3 cycles, whereas ldm takes n+2 cycles
12:39:55Unhelpfulinitially i wrote the branch-for-large-numerator optimization only for APE, because i knew it would be so rarely taken there... but it turned out to bench faster in general as well.
12:40:10 Quit ecio (Ping timeout: 245 seconds)
12:40:49amiconnOn arm9 (both v4 and v5) this advantage (almost) vanishes as a single load is 1 cycle (as long as you don't use it in the next cycle), and ldm is n cycles
12:40:56amiconnOf course it still reduces code size
12:41:50amiconnOn armv6 it becomes very useful again, because a single load is 1 cycle, and ldm is *also just one cycle, as the fetches happen in the background
12:42:27amiconnOf course this needs proper design wrt latencies
12:43:07stripwaxthe bulk of the arm-specific code is optimising register (re)usage and ldm/stm - but since it was tested for arm7tdmi I don't know what impact will be on arm9/arm6 - sounds like there could be some scheduling problems if I ldm followed by using it immediately
12:43:55TheSevendamn, my internet connection is faster than copying the test files to the ipod
12:44:17amiconnIirc I squeezed quite a bit of performance out of the ape predictor on armv6 by just reordering instructions, without negative impact on the earlier arm versions
12:44:50Unhelpfulreordering doesn't really matter to arm7... and only barely to arm9tdmi
12:45:46Unhelpfuland i don't think reordering that will help arm9e/arm11 will ever hurt arm9tdmi, will it?
12:45:54amiconnAh yes - r19210; 20% speedup
12:46:01kugelstripwax: what files would I test?
12:46:03stripwaxwhat targets use armv6?
12:46:04amiconnNo it shouldn't
12:46:05stripwaxkugel - ?
12:46:12kugelfor the mdct comparison
12:46:12amiconnstripwax: Just Gigabeat S
12:47:22kugelstripwax: I mean which type of files are relevant for this?
12:47:39kugelwma,ogg. what else?
12:47:45stripwaxkugel - Start with vorbis and wma for now, just for ease of comparison
12:48:30stripwaxreally all that use the mdct are relevant. so atrac3, cook, a52. But I don't know how well those are represented in the existing test_files downloadable stuff
12:53:55 Join anewuser [0] (anewuser@unaffiliated/anewuser)
12:57:13***Saving seen data "./dancer.seen"
12:58:28 Quit kugel (Ping timeout: 265 seconds)
13:00:08TheSevenerm, does make reconf lose the information that a build dir is supposed to use eabi?
13:00:43 Join Farthen [0] (
13:04:25TheSevenok, after some build problems, test_codec is finally running. on r24711-noeabi for now
13:05:43TheSeven64kaahce roughly realtime, does that make sense, or are we having boosting problems again?
13:06:41TheSeven190.13 MHz
13:08:29 Join teru [0] (
13:09:37TheSevengrr, doom fails to build again
13:14:03TheSevenape c2000: 66.4 MHz, ape c4000: 301.16 MHz
13:16:27TheSevenc3000: 102.7 MHz
13:16:52*amiconn wonders how large the filter coefficients in demac can become
13:16:55stripwaxTheSeven - can you try vorbis and wma first please - ape doesn't use mdct so wouldn't have any delta to pre-mdctexp changes
13:17:23amiconnIf 15 bits were enough, the swar stuff could be sped up
13:17:27*stripwax tries to fix red - atrac3 iram full on a handful of targets
13:18:00amiconn(for armv5)
13:18:51 Join kugel [0] (
13:18:55 Quit kugel (Changing host)
13:18:55 Join kugel [0] (~kugel@rockbox/developer/kugel)
13:19:07kugelstripwax: old: ; new:
13:19:20amiconnOh, and for coldfire too of course
13:19:54stripwaxkugel - cheers - remind me what target that is for?
13:20:15kugelI can't test higher bitrates due to limited ram (and the fact that test_codec doesn't handle it)
13:20:29kugelbut it's at least 3-4MHz faster
13:23:31 Join stripwax_ [0] (
13:24:25 Quit stripwax (Ping timeout: 240 seconds)
13:24:45 Quit anewuser (Quit: =ooo ϢINTER ϾHIP 5iVE is OOON!!)
13:25:14Buschelstripwax: regarding reds caused by iram in atrac3 -> try to remove iconst_attr for several array in atrac3data_fixed.h, but keep window_loopup[].
13:27:05stripwax_Buschel - yep, I can do that, but unfortunately I don't have a target to test/compare. But I'll do whatever is needed for it to continue to build ..
13:27:25Buschelstripwax: if this still isn't sufficient, define the iram usage for imdct in a way that it is only used on targets with large iram (like PP5022)
13:27:27 Quit AlexP (Remote host closed the connection)
13:28:33stripwax_BUschel - I'm (re)using the same define that was already in place for the old mdct (the Tremor one). So it's already defined per-target, etc. I don't want to change that unless I really have to.
13:29:15stripwax_after all, all of the other codecs are fine, and iram is a major speedup for wma for example. so I'd rather just hobble atrac3 rather than hobble any other codecs.
13:29:43stripwax_atrac3 still needs more optimising, anyway
13:30:08Buschelstripwax_: that's right
13:31:33Buschelhmm, it is only red for PP's, right?
13:31:41stripwax_Buschel - building for a new target is so slow (on a netbook, running cygwin..)
13:32:02stripwax_Buschel - I don't know about mrobe, vibe or the samsungs..
13:33:02stripwax_I guess they all have a bit less iram than the 5022?
13:33:23 Join tomers [0] (
13:33:32Buschel96KB vs. 128KB
13:33:37Buschelquite a lot
13:34:19stripwax_atrac3 only recently got changed to use the new imdct - and I suspect subsequently some ICODE/ICONST got added to atrac3 (when it should have been added to the imdct).
13:34:51stripwax_anyway, when I've got a build that works, I'll commit the changes. I think atrac3 is all mt/saratoga right now
13:34:54Buschelit was :o) I did it :)
13:35:02stripwax_oh! hahahaha
13:35:31BuschelI am also just starting to compile for h10...
13:37:28Buschelbtw, I'm also buidling on a slow notebook... will take 10min
13:40:20Buschelgot it compiling for h10 with a small change
13:41:07 Quit Guest88353 (Read error: Operation timed out)
13:41:41stripwax_ok cool. gain_tab1 looks a bit silly, by the way ..
13:42:26stripwax_it's just 1<<(20-i)
13:42:26Buschelremove ICONST_ATTR for qmf_48tap_half_fix[], matrixCoeffs_fix[], channelWeights0[] and channelWeights1[9
13:42:49stripwax_Buschel - are they likely to have the least impact on speed?
13:42:52 Join AlexP [0] (~ap@rockbox/staff/AlexP)
13:43:00 Quit kugel (Ping timeout: 264 seconds)
13:43:30 Quit cjcopi (Ping timeout: 268 seconds)
13:43:40Buschelonly matrixCoeffs_fix[] could have an impact for JointStereo files. all other are irrelevant
13:44:54Buschelok, matrixCoeffs_fix[] can still be iram'ed for h10
13:44:54CIA-8New commit by stripwax (r24787): Remove ICONST_ATTR from some tables, to fit into PP5020 iram (now that mdct is in iram it's a bit of a squeeze). (per Buschel on irc)
13:45:02stripwax_pah, too late!
13:45:20Buschellet's see what the tables will look like in a few minutes ;)
13:46:05stripwax_I confirmed it built ok for ipod1g/2g
13:46:22 Join Omlet05 [0] (
13:46:23Buschelso, the chance is good
13:46:36stripwax_I'm going to blow away gain_tab1 and replace it with 1<<(20-i) in the code, if no objections?
13:49:07Buschelno objections from my side. delete the array and readd ICONST_ATTR for the matrix-array :)
13:49:25 Quit Omlet (Ping timeout: 276 seconds)
13:49:36 Join Omlet [0] (
13:49:36Buschelgreen :)
13:49:53 Join Guest88353 [0] (
13:51:03 Join efyx_ [0] (
13:51:28 Join cjcopi [0] (
13:51:49stripwax_matrixCoeffs_fix also looks like a bit of a silly table.. does that really need to be a lookup? there's not a lot of information in it...
13:52:16 Quit Omlet05 (Ping timeout: 260 seconds)
13:52:16amiconnWhat's the range of i?
13:53:14amiconnDepending on the stupidity of gcc, it might be better to use (1<<20) >> i
13:53:15 Quit flydutch (Quit: /* empty */)
13:53:42amiconnBut this requires i to be non-negative
13:53:49stripwax_amiconn - I'm actually using (ONE_16<<4)>>i . which I think is consistent with what you said, as ONE_16 is defined as 1<<16
13:54:03stripwax_i is definitely no-negative (given that it was an index into a table?)
13:54:13gevaertsamiconn: actually, the D2 does have read-only ftl code that works fine for most people. It just doesn't have write support
13:54:32stripwax_range is 0<=i<=15
13:55:30 Join TucknDar [0] (
13:55:38CIA-8New commit by stripwax (r24788): Reinstate ICONST_ATTR for matrixCoeffs_fix ; remove (silly) gain_tab1 and replace with a simple bitshift in the code
13:56:05Buschelstripwax_: as JointStereo decoding is still buggy I would like too keep those lookups until the code has been debugged. otherwise it becomes harder to find the error...
13:56:14stripwax_which lookups?
13:56:31amiconnstripwax_: Yes it is
13:56:39stripwax_Buschel - gain_tab1?
13:56:55Buschelstripwax_: the silly matrixCoeffs_fix[]
13:57:24amiconnA table index may be negative, depending on the table. Not common in C, but I'm using this in the SH1 bitswap
13:57:30stripwax_Buschel - oh yeah, no problem with leaving that in place. That one at least has *some* information in it compared to gain_tab1 :)
13:57:59 Join GodEater [0] (~bibble@rockbox/staff/GodEater)
13:58:04stripwax_amiconn - the code was gain_tab1[ index ] and the code is now (1<<20)>>index
14:02:07stripwax_amiconn - (I'm still struggling to see how a table index can be negative −− unless your table 'pointer' is something other than the start of the table, e.g. the middle)
14:02:18stripwax_I guess I could read the SH1 bitswap to find out :)
14:02:34amiconnYes, the table base is in the centre
14:03:14amiconnThis is because SH1 sign-extends by default. Making the table work this way saves a separate extend-unsigned instruction
14:04:05 Nick fxb__ is now known as fxb (
14:04:58amiconnThis table is defined in asm, but also used from C code on Ondio, and then the table index is an explicit 'signed char'
14:05:29teruJdGordon: ping
14:06:01gevaertsamiconn: that will need watching with new gcc versions
14:10:32 Join stripwax [0] (
14:13:23 Quit stripwax_ (Ping timeout: 265 seconds)
14:13:34 Join ecio [0] (
14:14:56 Quit Guest88353 (Ping timeout: 268 seconds)
14:16:58 Join Guest88353 [0] (
14:24:44 Join GeekShadow [0] (~Antoine@reactos/tester/GeekShadow)
14:25:31stripwaxTheSeven - did you get a chance to do test_codec on vorbis/wma ?
14:25:57 Quit Buschel (Ping timeout: 265 seconds)
14:26:52CIA-8New commit by tomers (r24789): WPS: Use helper local variable instead of its value (no functional change)
14:28:12 Join Horscht [0] (~Horscht2@xbmc/user/horscht)
14:28:16 Quit Frampis (Ping timeout: 240 seconds)
14:28:21AlexPtomers: Did you see the comments about your commit message yesterday?
14:28:30 Quit Horschti (Ping timeout: 252 seconds)
14:28:39 Join Frampis [0] (
14:29:11AlexPtomers: "Some changes for brickmania" even with the flyspray number isn't very descriptive. It would be nice to describe what those changes were in the commit message
14:29:28 Quit shaggy-h (Ping timeout: 240 seconds)
14:33:15 Join robin0800 [0] (
14:34:13TucknDargevaerts, thanks for the pointers about increasing the thickness of the CF card in my iPod mini, yesterday. Finally got it working and I can now enjoy a 2g 16gb ipod mini with Rockbox! :)
14:34:26 Join TheSeven|Mobile [0] (~theseven@rockbox/developer/TheSeven)
14:35:54 Quit Guest88353 (Ping timeout: 268 seconds)
14:36:00 Join dfkt [0] (dfkt@unaffiliated/dfkt)
14:36:48TucknDarluckily I had two minis, cause I managed to break one :p
14:36:59TucknDarbut I only need one anyway ;) Thanks again.
14:37:23 Quit killan_ (Quit: ( :: NoNameScript 4.22 :: ))
14:39:05 Join killan [0] (
14:39:15tomersAlexP: I agree. Sorry for that
14:39:26AlexPno worries :)
14:40:42 Join Guest88353 [0] (
14:42:25 Quit stripwax (Quit:
14:42:42 Join funman [0] (~fun@rockbox/developer/funman)
14:44:43funmanUnhelpful: thanks for the test, the IRAM might be disabled on clipv2, can you add a CGU_PERI |= (1<<25) before the test ?
14:45:07Unhelpfulfunman: i can later. :)
14:45:45funmanthe datasheet says it should be default enabled but perhaps they disable it and then enable it only in some use case
14:47:12funmanstill weird that nothing happens, i'd think the data abort mode would be entered
14:47:47 Join mirak_ [0] (
14:49:07 Nick fxb is now known as fxb__ (
14:54:25 Quit Guest88353 (Ping timeout: 268 seconds)
14:56:14 Join Guest88353 [0] (
14:57:17***Saving seen data "./dancer.seen"
15:00:59 Join kugel [0] (~kugel@rockbox/developer/kugel)
15:01:13 Quit kugel (Read error: Connection reset by peer)
15:01:23 Join kugel_ [0] (
15:01:29 Nick kugel_ is now known as kugel (
15:01:41 Quit kugel (Changing host)
15:01:41 Join kugel [0] (~kugel@rockbox/developer/kugel)
15:05:31 Quit Guest88353 (Ping timeout: 268 seconds)
15:06:37 Quit robin0800 (Remote host closed the connection)
15:06:56 Join Guest88353 [0] (
15:11:56 Quit funman (Quit: free(random());)
15:14:19*kugel grumbles at FS #11041
15:18:14kugelit happened to me yesterday too, but not in the week of testing prior to committing
15:20:40 Join webguest [0] (
15:20:59Tornethere seem to be more people than usual who can't reset their ipods :(
15:21:19Tornewould it be controversial to suggest we enable bootloader usb on ipod?
15:22:03kugelon all pp please then :)
15:22:29kugelTorne: I don't think there would be objections, it just wasn't needed yet
15:22:32Tornei guess the only reason not to is size
15:22:36Torneand sure, its' not technically needed
15:22:43Tornebut i think it would be easier for some people to recover from mistakes that way
15:23:01kugelbut with rockbox usb and the ability to completely replace the of (including bootloaders) there should be a bootloader recovery mode
15:23:04 Quit TheSeven|Mobile (Ping timeout: 240 seconds)
15:23:05gevaertsis size really an issue on the usual pp targets?
15:23:11Tornegevaerts: don't think so
15:23:21Tornealso, does bootloader usb always go into usb mode if the cable is inserted?
15:23:27Tornebecause we probably don't want that on ipod
15:23:34Tornefor the same reason that usb isn't enabled in release builds
15:23:48Torneit would be nice if it was just if the main binary was not loadable
15:23:54kugelit happened to me once that I trashed my rockbox installation so that I needed e200tool to get it running again. that wouldnt be needed with bootloader usb
15:24:30Tornethere just seem to be a lot of people lately who can't reset their ipods
15:24:42gevaertsyes, it's weird
15:24:46Torneand i'm wondering if the reset keys are really as bulletproof as we claim
15:26:00kugelso I definitely vote for it
15:27:06Tornei assume that it does always go into usb mode, though
15:27:19Tornewhich could be undesirable
15:27:54gevaertsyes, that's the main problem
15:28:27Tornei guess it shouldn't be too hard to arrange for it to only go into usb mode when it would otherwise display a boot failure message
15:28:36Torneand maybe on a button press? dunno.
15:28:45kugelboth would be nice
15:30:11 Join Schmogel [0] (
15:31:03Torneactually i guess you'd just have to patch out the check for usb connected on start?
15:31:17Torneand leave usb insertion as it is
15:34:30 Quit tmzt (Ping timeout: 268 seconds)
15:36:39 Quit webguest (Quit: CGI:IRC)
15:43:46kugelit would surprise me if my buffering commit had any impact on a 64MB device (FS #11041)
15:45:16 Quit Strife89 (Read error: Connection reset by peer)
15:54:38TheSeven[15:21]<kugel>but with rockbox usb and the ability to completely replace the of (including bootloaders) there should be a bootloader recovery mode
15:54:49TheSevenyou mean flashing the nor on older ipods? do we do that?
15:55:48kugelhrm, I think ipodpatcher offers some advanced installation procedure overcoming the of but it's not the default, sansapatcher has something similar
15:55:55Tornebut if you drop rockbox into OSOS then turninghold on to boot the OF won't get you the disk mode either
15:56:25TheSevenholding the button combo should do so
15:56:51Torneit would, except there seem to be quite a few people on the forums who can't reset their ipods
15:56:56Torneand have great trouble as a result
15:57:23TheSevenhow would bootloader usb help there, if the ipod is stuck in a state where it's locked up and you can't reset it?
15:57:33TheSevenor do you mean that the reset fails when the bootloader fails to load rockbox?
15:57:35Tornebecause it's not stuck in such a state
15:57:43Tornethey just cocked up the install and don't ahve .rockbox on there
15:57:50Tornebut they now can't reset to get into disk mode
15:58:05Torneone guy got it sorted by letting the battery run right down then turning hold on and connecting usb
15:58:59TheSevenhas anyone ever investigated how that reset works?
15:59:03Tornedon't know.
15:59:09Tornewe have always assumed it's "hardware"
15:59:16Tornewhich is a great non-answer
15:59:26Tornemaybe someone knows, but i haven't heard :)
15:59:34TheSevenon the nano2g it seems to be 100% bullet-proof, the clickwheel seems to signal the pmu that it should shut off the cpu's power for half a second
15:59:53Torneas far as my experience goes it's 100% bullet proof on all the others as well
15:59:58TheSevenbut on the nano4g, there seems to be a way to inhibit a reset, ipod is doing that during a restore
16:00:06Tornewell, all the PP ipods
16:00:17Tornebut there seem to be people having trouble anyway
16:00:26Tornepreviously they mostly went "oh right" when we told them to hold it down for longer
16:00:38Tornebut the last two or three occurrences on teh forums they just can't make it work
16:02:21Torneanyway, my point was that there are cases wher bootloader usb might help, and i don't think there's much reason *not* to add it :)
16:02:54 Join tmzt [0] (~tmzt@
16:03:35*kugel thinks the use of size_t in buffering.c is generally wrong
16:05:02 Quit teru (Quit: Quit)
16:08:36 Quit tomers (Ping timeout: 252 seconds)
16:13:55CIA-8New commit by kugel (r24790): Use a helpfer function to avoid ugly casting and correct some data types (no functional change).
16:18:17 Nick Guest88353 is now known as martian67 (
16:18:30 Quit martian67 (Changing host)
16:18:30 Join martian67 [0] (~martian67@about/linux/regular/martian67)
16:31:02S_a_i_n_tTheSeven: Torne: I have *once* had an issue (nano2g) where the OF crashed and I couldn't reset it using menu+select, had to let the battery drain completely also (which didn't take too long actually as it froze on a full white screen with the backlight on ;) ).
16:31:22S_a_i_n_tapart from that *one* time, the reset key combo has worked perfectly.
16:31:53TheSevenhm, can one shut off the clickwheel power in the pmu?
16:32:06TheSeveni haven't managed to do that yet at least...
16:36:10 Quit stoffel (Ping timeout: 256 seconds)
16:43:19 Join Blue_Dude [0] (~chatzilla@
16:45:17Blue_DudeCan anyone confirm that r24782 fixed FS #11038? It seems to be OK now.
16:47:53 Join mt [0] (~mtee@rockbox/developer/mt)
16:49:15S_a_i_n_tRE: Nano 1/2g Themes, the original of my Radiance theme had to go (copyright issues), so I'm guessing the guy that mixed it up and used all the art from the original's theme has to go too? I hadn't looked at the theme site in ages.
16:51:24 Quit petur (Quit: time!)
16:57:20***Saving seen data "./dancer.seen"
17:01:47 Join stripwax [0] (
17:05:30 Quit dmb (Quit: No Ping reply in 180 seconds.)
17:07:03 Quit piotrekm (Quit: piotrekm)
17:07:07 Join liar [0] (
17:09:49 Join dmb [0] (~dmb@unaffiliated/dmb)
17:19:28 Quit kugel (Ping timeout: 260 seconds)
17:21:44 Quit S_a_i_n_t (Quit: "PC Restart")
17:22:13 Join stripwax_ [0] (
17:23:54 Quit stripwax (Ping timeout: 245 seconds)
17:25:05 Join marcol [0] (
17:25:51 Quit marcol (Client Quit)
17:25:54 Join marcol [0] (
17:26:45 Join tomers [0] (
17:26:52 Quit Blue_Dude (Quit: ChatZilla 0.9.86 [Firefox 3.5.8/20100202165920])
17:28:36marcolhi there
17:30:39liarTheSeven: r24725 breaks rockboy(stkov main)
17:31:50 Join S_a_i_n_t [0] (S_a_i_n_t@
17:36:01 Quit simabeis (Quit: leaving)
17:36:08 Join simabeis [0] (
17:37:30 Join CaptainKewl [0] (
17:37:56 Join CaptainKwel [0] (
17:38:34 Quit TucknDar (Quit: Miranda IM! Smaller, Faster, Easier.
17:39:29 Join FOAD_ [0] (
17:42:21 Quit CaptainKewl (Ping timeout: 272 seconds)
17:42:48 Quit FOAD (Ping timeout: 260 seconds)
17:42:48 Nick FOAD_ is now known as FOAD (
17:43:54 Quit liar (Ping timeout: 245 seconds)
17:49:58 Join ecio1 [0] (
17:51:49 Quit ecio (Ping timeout: 245 seconds)
17:54:05 Quit marcol (Quit: CGI:IRC)
17:54:58 Quit stripwax_ (Quit:
17:57:20TheSevenliar: i already wondered whether that would happen, but as it didn't break pictureflow which is also prone to stkov's and i don't have any roms, i assumed it should work
17:57:53*TheSeven wonders why rockboy doesn't just allocate it's own stack from plugin ram if it needs more stack than everything else in rockbox
18:03:37 Quit fyre^OS (Quit: Ur skills' fireproof like a wooden panel -- U got feds talking leet on your IRC channel!)
18:05:39TheSevenliar: could you have a look into this?
18:06:31 Join toffe82 [0] (
18:07:04 Join kugel [0] (~kugel@rockbox/developer/kugel)
18:07:20kugelTheSeven: I think it's because plugins run within the main thread
18:07:58TheSeveni can see no reason why it should need that much stack. it just *must* be possible to move something from the stack to the heap to fix this
18:08:15TheSeveni really don't want to waste several kilobytes of iram because of rockboy
18:15:47 Join fleebailey33 [0] (fleebailey@unaffiliated/fleebailey33)
18:17:29TheSevenbtw: 4MHz improvement on vorbis, 11MHz on WMA
18:18:09 Quit Xerion (Ping timeout: 246 seconds)
18:34:12 Quit martian67 (Remote host closed the connection)
18:37:06 Join saratoga_ [0] (~463f90ed@gateway/web/freenode/x-gesbelhyftqbzrbe)
18:41:32 Join Adnyxo [0] (
18:44:29 Join martian67 [0] (~martian67@about/linux/regular/martian67)
18:44:45 Join CGL [0] (~CGL@
18:48:23 Join toffe82_ [0] (
18:50:51 Quit toffe82 (Ping timeout: 260 seconds)
18:53:52 Join MaadMan [0] (
18:57:23***Saving seen data "./dancer.seen"
18:58:29 Quit AlexP (Remote host closed the connection)
18:58:56 Join AlexP [0] (~ap@rockbox/staff/AlexP)
18:59:48 Quit AlexP (Remote host closed the connection)
19:03:23 Join AlexP [0] (~ap@rockbox/staff/AlexP)
19:04:26 Quit toffe82_ (Remote host closed the connection)
19:08:43 Join flydutch [0] (
19:10:17saratoga_TheSeven: the nano2g has a lot more IRAM we could use for codecs right?
19:14:03 Quit CGL (Remote host closed the connection)
19:18:24 Join CGL [0] (~CGL@
19:24:42 Join flyback [0] (
19:25:38 Join phanboy4 [0] (
19:26:33 Join sudoman [0] (~d8ecfceb@gateway/web/freenode/x-lsjuxyvrzcudjtft)
19:28:08 Quit tomers (Quit: ChatZilla 0.9.86 [Firefox 3.5.8/20100214235838])
19:28:15 Join toffe82 [0] (
19:28:29TheSevensaratoga_: With some optimizations, we could give up to 128K to the codec
19:29:26saratoga_TheSeven: that would be nice
19:29:33saratoga_how much is available right now?
19:30:12TheSevencurrently it's a 64/96 split, but I don't know which way round
19:30:22TheSeventhe core easily fits into 64k though
19:30:47TheSevenand some time ago i managed to squeeze it into 48k as well
19:32:24saratoga_i got a nano2g now
19:32:34saratoga_i should look at some codecs and see how much is really useful
19:33:00saratoga_more then 80KB would certainly be nice, though i don't know if we need the full 128k
19:36:35 Join jordan` [0] (~jordan@
19:47:27 Join Xerion [0] (
19:49:05kugelcan someone check if builds fine (a manual patch)?
19:53:06 Join grndslm [0] (
19:53:36 Join shai [0] (
19:54:39*kugel isn't able to install half a GB of latex stuff right now
19:55:14kugelAlexP, pixelma. mc2739 ^ ?
19:57:46 Quit saratoga_ (Quit: Page closed)
20:01:03TheSevenkugel: checking...
20:02:06TheSeveneverything ok :-)
20:02:42 Quit moos (Quit: ChatZilla 0.9.86 [Firefox 3.6/20100115144158])
20:03:46kugelgreat, which manual did you build?
20:04:44TheSevennano2g, pm'ed you a link to the pdf
20:04:44kugelthe change was under "4.3 While Playing Screen" in the e200 manual (4.3.3 to be precise)
20:05:01kugelah nice
20:05:31TheSevenis this specific to e200?
20:06:42CIA-8New commit by kugel (r24791): Playlist Viewer Changes to bring consistency: ...
20:06:45CIA-8New commit by kugel (r24792): Update the manual according to the playlist viewer changes
20:06:49kugelno, it's alright :)
20:07:09kugelTheSeven: the numbers might be different on other manuals though
20:08:56pixelmahmm, is things like moving tracks now in the same submenu as the playlist viewer settings?
20:09:19kugelthe location of the playlist viewer settings changed
20:09:49kugelit's now in the context menu (along with moving etc)
20:10:44pixelmaI'm trying to imagine if I would expect it there
20:11:33kugelwhere should it else be if not in the context menu?
20:11:46kugelon the button which goes to the main menu everywhere else?
20:11:49 Join SeismicMike [0] (
20:12:35pixelmathe context menu in the playlist viewer just dealt with the track you invoked it on, the viewer's settings are more general
20:12:38TheSevenbtw. does anybody know what the "m" (manual) option in configure is doing?
20:12:56TheSeveni can't find any difference to "n" (normal)
20:13:08pixelmaTheSeven: it's an old relict
20:13:14TheSevenmake will always do a normal build, and make manual will always build a manual...
20:13:16SeismicMikeI'm eyeballing this Coby 4GB mp3 video player at for $30... does anyone know if Rockbox will run on it?
20:13:26 Join Buschel [0] (
20:13:26TheSevenprobably not
20:13:34TheSevensee the list of supported devices on the front page
20:14:33pixelmathere was a time where you didn't need "make manual" and could configure a manual target and just "make". It doesn't work anymore since the parsing of features.txt to options in the manual
20:15:12 Join HBK [0] (
20:16:05kugelI think the reason is rather the make remake
20:16:25TheSevenbtw, is there anyone who wants to start a nano4g port?
20:16:29pixelmait's not
20:16:57kugelmake always builds the first target in a make file. prior to the remake the makefile was generated, now it's not anymore (just a file that sets a few variables)
20:17:21kugelmaybe it stopped working before that though
20:17:33 Quit KBH (Ping timeout: 268 seconds)
20:17:54kugelTheSeven: heh, yes! If you have time and a nano4g for me :p
20:18:07TheSeven/data/rockbox-trunk/apps/plugins/lua/loslib.c: In function ‘os_exit’: /data/rockbox-trunk/apps/plugins/lua/loslib.c:176: warning: no return statement in function returning non-void
20:18:24 Quit MethoS- (Remote host closed the connection)
20:18:24TheSevenwhy doesn't this yellow show up in the build table?
20:19:19 Quit flydutch (Quit: /* empty */)
20:19:32kugelTheSeven: you're building eabi with a newer compiler, right? I've seen this warning too but not under the default toolchain
20:19:47TheSevenoh yes
20:20:14kugelpixelma: ah, interesting. thanks
20:20:16 Quit Torne (Ping timeout: 240 seconds)
20:20:35 Nick fxb__ is now known as fxb (
20:21:16 Quit amiconn (Disconnected by services)
20:21:17 Join amiconn_ [0] (quassel@rockbox/developer/amiconn)
20:21:26 Join pixelma_ [0] (quassel@rockbox/staff/pixelma)
20:21:27 Quit pixelma (Disconnected by services)
20:21:41 Nick amiconn_ is now known as amiconn (quassel@rockbox/developer/amiconn)
20:21:44 Nick pixelma_ is now known as pixelma (quassel@rockbox/staff/pixelma)
20:22:03 Join panni_ [0] (
20:29:06 Join stripwax [0] (
20:30:48 Quit Schmogel (Ping timeout: 272 seconds)
20:33:23 Join Torne [0] (torne@rockbox/developer/Torne)
20:33:25 Join Schmogel [0] (
20:35:11 Quit Buschel (Ping timeout: 264 seconds)
20:38:04 Quit Torne (Ping timeout: 245 seconds)
20:40:21CIA-8New commit by kugel (r24793): Remove a few unused defines
20:41:02 Quit stripwax (Quit:
20:42:02 Nick fxb is now known as fxb__ (
20:44:13 Join stoffel [0] (
20:47:24 Join Torne [0] (torne@rockbox/developer/Torne)
20:48:00CIA-8New commit by kugel (r24794): Fix up Fuze's radio keymap a bit.
20:49:20 Quit sudoman (Quit: Page closed)
20:49:39 Quit bmbl (Quit: Bye!)
20:53:08 Quit Torne (Ping timeout: 252 seconds)
20:56:28kugelhaha, I know see what a mess radio.c is
20:56:37kugelit's really disturbing
20:57:17 Join toffe82_ [0] (
20:57:19 Quit toffe82 (Ping timeout: 260 seconds)
20:57:25***Saving seen data "./dancer.seen"
21:12:55leavittxIt seems to me that rockbox just killed my beloved Cowon I7 :(((
21:14:33leavittxI made patched firmware with mktccboot, it updated, after that my cowon doesn't switches on at all
21:15:00leavittx/sorry for offtopic
21:15:34kugelleavittx: yea, I can imagine that happens
21:15:42 Join bmbl [0] (~Miranda@unaffiliated/bmbl)
21:16:02kugelnobody is working on that port
21:24:13leavittxkugel: then what this target is doing in tools/configure?
21:24:33leavittxAnd is there any way to re-alive my i7? :)
21:24:54 Quit Kitar|st ()
21:25:10kugelit worked at some time
21:25:31 Join robin0800 [0] (
21:25:52 Quit stoffel (Remote host closed the connection)
21:26:20kugelno idea how you could unbrick it
21:26:33leavittxbut broken now. heavy broken
21:26:57kugelleavittx: it's under "unusable", you shouldn't have been installing it in the first place
21:27:16leavittxI think that it will be good to remove this from tools/configure
21:27:29kugelI don't see why
21:27:38kugelit's clearly not our fault
21:27:57leavittxis there any info about this port?
21:28:31kugelhave you visited our site at all?
21:30:26 Join phanboy_iv [0] (
21:30:36leavittxSad, sad, sad
21:30:41saratogayou can run rockbox on the i7 without installing it using ttctool
21:30:47saratogai think thats what you are supposed to do
21:31:00saratogayou could probably use that now to put the OF back and fix your player
21:31:14saratogabut i don't know how, or even if anyone has done it
21:31:35TheSeven"Flashing: The OF does the flashing, so a bad flash will ruin the player until we figure out how to do the USB boot mode properly."
21:31:54saratogai suggest reading through the i7 and d2 info
21:32:17kugelTheSeven: he hasn't read it prior to installing
21:32:31leavittxdamn on me
21:32:58leavittxrb on my cowon d2 works so perfectly, so...
21:33:54 Quit phanboy4 (Ping timeout: 245 seconds)
21:35:14saratogai still think you should be able to fix it with tcctool
21:35:35saratogathe dfu mode thing probably doesn't need an intact OF, otherwise it would be pretty useless
21:36:40 Quit toffe82_ (Ping timeout: 272 seconds)
21:37:08saratogaTheSeven: hilariously the sandisk connector fits nicely into a nano2g
21:37:11saratogaalthough it doesn't break it
21:37:29saratogaat least not if you only do it for a couple seconds :)
21:38:01TheSevenhave you checked the pinout?
21:38:05TheSevenmight it be even compatible? ;-)
21:39:11saratogano they're definitely not compatible
21:39:21saratogaputting a sansa into an ipod cable bricks it
21:40:21 Join Kitar|st [0] (
21:41:57leavittxsaratoga: i tried to run ./tcctool -d iaudio7 ../../i7_boot/I7_FW.BIN
21:42:00leavittxnot helped
21:42:12leavittxit says Ensure your TCC device is in USB boot mode and run tcctool again.
21:42:20leavittxhow do i do this?
21:43:00TheSeven"USB Boot Mode: The iAudio7 has a USB boot mode, you can enter it by holding down the 'mode' button while plugging in the USB connector. The Vendor/Product id changes to a TCC specific pair (Vendor=140e ProdID=b021), it is then possible to upload code to the device via tcctool (source available in svn tree: ). "
21:43:03TheSevensee the wiki...
21:44:55 Join m3dlg [0] (~m3dlg@
21:47:45 Join mitk [0] (
21:48:04saratogaleavittx: like I said I have no idea how to do this . . .
21:48:36 Join hebz0rl [0] (
21:49:31 Quit S_a_i_n_t (Quit: There are 10 types of people, those who understand binary, and those who don't.)
21:51:10TheSevendamn, my nano's lcd controller is dying
21:51:22TheSevenit's getting more and more bad GRAM cells
21:51:26 Quit m3dlg (Ping timeout: 252 seconds)
21:52:38 Join m3dlg [0] (~m3dlg@
21:53:14CIA-8New commit by kugel (r24795): Quickscreen for the radio screen. I added a keymap for almost all targets. I couldn't find a nice one (i.e. one that's consistent with the wps/menu ...
21:58:50saratogaTheSeven: time to work on the 4g port ;)
21:59:53saratogaheh we have no recent PP ipod battery benchs
22:00:30leavittxTheSeven: I feel like an absolut noob, BUT what is the 'mode' button?
22:02:13kugelleavittx: everybody here is as clueless as you are...
22:02:36TheSevenleavittx: I've never even seen an i7, i just read what the wiki says
22:02:40saratogayou're probably the only one in this channel thats ever seen an i7
22:03:25leavittxbut there is no such button
22:03:45 Quit robin0800 (Remote host closed the connection)
22:04:19leavittxthere are power+hold button, menu, +, -, and some sensor buttons. Also reset.
22:05:22kugeltry all
22:06:45saratogadid you check google?
22:06:45leavittxalready ;)
22:06:52saratogatheres lots of guides to fixing bad flashes
22:06:55saratoganone of them worked?
22:07:07leavittxno, I tried only tcctool
22:07:33leavittxI can't get to USB boot mode
22:07:44saratogathis one says to use the "mode/menu" button
22:07:47saratogadid you try that
22:08:06leavittxhmm, I think, yes
22:08:33saratogaapparently a few other people have fixed this problem, maybe you could try asking one of them
22:08:38kugelI wonder if it makes sense to transform the preset list into a playlist
22:08:53kugelto enable saving, moving things around, other funny stuff
22:09:00saratogaalso, what is the issue with supporting 240GB players in standard builds?
22:09:14TheSevensaratoga: I guess binsize...
22:09:39kugeldidn't zagor and amiconn talk about it recently? i remember seeing it in the logs
22:09:41saratogasurely thats not a serious problem on the ipod video or gigabeats
22:10:21TheSevensaratoga: you know that some rockbox devs just get crazy when someone pulls their binsize trigger :-)
22:11:19saratogabinsize is important but it needs to be considered in context
22:14:44saratogashouldn't the recent font stuff go into MajorChanges
22:18:37 Quit Omlet (Read error: Connection reset by peer)
22:20:26 Join Omlet [0] (
22:22:48 Quit m3dlg (Ping timeout: 260 seconds)
22:25:28TheSeveneabi seems to have a rather weird impact...
22:26:30leavittxsaratoga: thanks, this helped me a little. Now what do i do after this message from tcctool: "[INFO] Patching application uploaded successfully!" ? How to says "copy 3 firmware files onto your i7 root folder" but how do i access root forder?
22:27:11saratogalook i have no idea
22:27:13saratogastop asking me
22:27:59saratogaTheSeven: do you have a script for parsing the test_codec files?
22:28:02saratogai was just about to write one
22:31:07TheSevensaratoga: a trivial 5-liner
22:31:17TheSeveni should write a proper one
22:31:22TheSeventhe rest was openoffice magic :-)
22:32:06saratogaok i'll do it now then
22:32:12saratogathis is easy in perl
22:32:33CIA-8New commit by kugel (r24796): Factor out some drawing code.
22:32:53kugelTheSeven: I don't think eabi has much impact
22:32:59kugelI would blame the compiler version
22:33:11kugelit has been reported that 4.4.x doesn't work well on arm
22:33:32TheSevenit isn't always worse, it's also better on some codecs
22:33:40TheSevenape is funny ;-)
22:33:46saratogathe results aren't much different except for mpc which is a little better
22:33:47TheSevensorted it properly:
22:34:00 Quit CGL (Remote host closed the connection)
22:34:37TheSeventhe lower ape compression levels are faster on eabi, the higher compression levels are slower
22:35:05saratogado we actually want to show this: 175906 of 175906 | Decode time - 36.43s | File duration - 175.90s
22:35:11saratogaseems like we could cut that out of the wiki
22:35:13kugellow is better?
22:35:27TheSevenyes, it's the mhz value
22:35:43kugelyea, expected w.r.t to the eabi toolchain
22:36:06TheSevenseems so
22:36:08kugelgenerally: flac performs much worse, others vary within +-2%
22:36:33kugelI really want to try 4.4.3 and the last 4.3.x release
22:37:00kugeland some other compilers also :)
22:37:45 Quit phanboy_iv (Read error: Connection reset by peer)
22:38:03*TheSeven wonders if it would be worth the effort to 1) add the device, build type and revision to the test-codec output, 2) make a website where one can upload the logs, 3) make an analyzer tool for the gathered results :-)
22:38:13kugelfeel free to add your results to CodecPerformanceComparison
22:38:33saratogakugel: the code sourcery gcc branch is supposed to be better for arm then stock
22:38:43kugelI heard that too
22:39:20kugelI would also like to try that other compiler that blog (forgot the names) mentioned
22:39:23TheSevenkugel: I think that many results get lost because most people are too lazy to work out how this weird wiki works...
22:39:44kugelI wonder how much effort you need to compile rockbox with a non-gcc-like compiler
22:39:56TheSevenand if you do a full test_codec, copying and pasting all the values will take quite a lot of time
22:39:57saratogaTheSeven: almost got a perl script together for that
22:40:46kugelespecially msvc for targeting windows mobile with raap
22:40:53 Join toffe82 [0] (
22:41:51kugelmight not even be worth trying looking at windows phone 7 (I heard it only accepts managed .net apps)
22:42:25 Join pamaury [0] (
22:42:26kugelTheSeven: the nano2g is pretty slow it seems? samsas have better results
22:42:32saratogawindows moble is a hard target because of all the annoyances of working with gnu stuff on windows
22:47:15 Join Torne [0] (~torne@rockbox/developer/Torne)
22:47:59TheSevenkugel: not much cache, slow IRAM
22:48:24TheSevenuncached iram accesses seem to need 2 cycles, even if the result is just thrown away
22:48:31kugelI think the samsas have less cache and even slower iram
22:48:50 Quit AlexP (Remote host closed the connection)
22:48:56TheSevenand test_codec is of course running at the full cpu speed
22:49:01kugel2x8k and the iram is basicallly as fast as the dram
22:49:04TheSevenit might be more efficient when it isn't boostig
22:49:36TheSeven(because it will run in fastbus mode then, while it is running in asynchronous mode when it needs to hurry)
22:49:42kugelwhat cpu is in the nano2g?
22:49:54 Join AsaelReiter [0] (
22:50:14saratogatakes 1 or 2 test codec results and will either compare them or just print one, either way formatted for the wiki
22:50:45TheSevenso it's 2x4k cache only
22:51:21kugelTheSeven: we use synchronous mode, that might make a difference. arm docs say asynchronous imposes performance penalties for synchronizing
22:51:50kugelok, that's less cache, I guess that's it
22:52:26TheSeveni couldn't make out any real differences in the docs between sync and async mode, the penalties seemed to rougly the same if it's running at an int multiple anyways
22:52:27kugela ldr should still only take 1c on all arm9 IIUC
22:52:34kugel(with result delay)
22:52:43TheSeventhe problem is just that it immediately locks up if i switch it to sync mode, no matter what
22:53:02TheSevenhaven't yet found out why (1:4 factor between the clocks)
22:53:17saratogaif the arm bus and the memory bus don't have an integer relationship, then there pretty much has to be 1 clock delay
22:53:51TheSevenfiring semi-random ldrs one after another, 256 of them in a loop takes ~520 clocks with caches disabled and the accesses hitting iram
22:54:16TheSeven(CPU at 192MHz, bus at 96MHz, async mode)
22:54:32kugelah we run at 248, another reason
22:54:43kugelbus at 62
22:54:45TheSeventhat shouldn't affect the mhz values though
22:55:19TheSevenactually your bus to cpu ratio is worse, so the ipod should be better
22:57:27***Saving seen data "./dancer.seen"
22:57:38saratogathe IRAM on the AMS players is really slow
22:57:43saratogaits basically just DRAM
22:57:49saratogaso performance is more like the gigabeat F
22:58:23kugelTheSeven: every target does worse it seems. PP is so fast compared to other arm targets looking at codec performance
22:58:55kugelPP needs 10MHz less than samsa for vorbis, for example
22:59:12kugelI think that's because the iram is as fast as the caches
23:00:08 Quit mitk (Quit: Ex-Chat)
23:01:15saratogaPP's custom IRAM implementation makes a huge difference
23:01:25saratogasince it makes memory accesses basically free for most codecs
23:03:10saratogaD2 should probably be faster then PP if it really has SRAM and armv5e
23:03:10 Quit AsaelReiter (Quit: CGI:IRC (EOF))
23:03:17saratogathough i heard we don't use the IRAM yet
23:09:02 Join bluebrother [0] (~dom@rockbox/developer/bluebrother)
23:10:16 Join AlexP [0] (~ap@rockbox/staff/AlexP)
23:10:37amiconnsaratoga: The main problem with supporting those 240GB disks is the large sector mess
23:11:07CIA-8New commit by bluebrother (r24797): If the Ipod has been recognized as MacPod always consider no bootloader installed.
23:11:09CIA-8New commit by bluebrother (r24798): Update an outdated comment.
23:11:15CIA-8New commit by bluebrother (r24799): Recognize and handle MacPods during autodetection. ...
23:11:51amiconnTorne: The PP based ipods *can* apparently reach a state where reset doesn't work, but this is rare
23:12:04amiconnI had this happen once on my Mini G2
23:12:14 Join stripwax [0] (
23:12:19 Quit bluebroth3r (Ping timeout: 276 seconds)
23:12:24Tornethat shouldn't be happening for people who have just forgotten to put .rockbox on the player though
23:12:41amiconnLetting the battery run down helped (took a few days)
23:13:07 Quit AlexP (Remote host closed the connection)
23:13:14amiconnThis only happened once, among probably hundreds of times I had to reset the ipod for various reasons
23:14:20 Join m3dlg [0] (
23:14:47 Join robin0800 [0] (
23:14:47amiconnNormally, if the reset doesn't work first time, it helps to flip the hold switch, flip it back, and try again
23:15:29 Quit hebz0rl (Read error: Connection reset by peer)
23:15:55 Join AlexP [0] (~ap@rockbox/staff/AlexP)
23:15:57 Join anewuser [0] (anewuser@unaffiliated/anewuser)
23:17:38UnhelpfulTheSeven: hrm, ldr on arm9tdmi is supposed to be 1c...
23:19:39kugelit looks like the build system got slower after attaching my client
23:19:40TheSevenpossibly not if the bus is slower and it's getting swamped with them? (every single instruction in that loop was an ldr)
23:19:42Unhelpfulalthough perhaps when async it can't actually fetch every cycle?
23:20:56 Quit robin0800 (Remote host closed the connection)
23:22:36 Join Buschel [0] (
23:22:40UnhelpfulTheSeven: hrm, what if you interleave w/ nops? i bet it stays about the same...
23:23:10 Join CGL [0] (~CGL@
23:26:46CIA-8New commit by kugel (r24800): Correct ActionFMPreset manual entry for the Fuze.
23:29:09kugelgah, we really need to generate those button maps from the actualy source files :(
23:32:15saratogaUnhelpful: the ARM9TDMI load/store unit adds one cycle of latency
23:32:28saratogathat doesn't mean the bus and DRAM accesses are free however
23:33:05saratoga(if thats what you were asking)
23:33:39 Join petur [0] (~petur@rockbox/developer/petur)
23:37:24 Quit kugel (Remote host closed the connection)
23:40:23 Quit pamaury (Quit: abort();)
23:40:30 Join hebz0rl [0] (
23:42:42 Quit AlexP (Ping timeout: 240 seconds)
23:43:57amiconnTheSeven: Did your back-to-back ldr instructions use different destination registers?
23:44:54 Join AlexP [0] (~ap@rockbox/staff/AlexP)
23:46:57CIA-8New commit by bluebrother (r24801): Add missing class prefix to logging call.
23:47:00CIA-8New commit by bluebrother (r24802): Be more strict when when resolving devices and allow hfs too. ...
23:48:44TheSevenamiconn: not sure, probably not
23:48:52TheSevenamiconn: not sure, probably not
23:51:24 Quit m3dlg (Ping timeout: 245 seconds)
23:52:05amiconnI'm not sure, but using the same destination register might also cause an interlock. Alternating between two destination registers should be enough avoid that
23:54:56 Quit Buschel ()

Previous day | Next day