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 2019-08-08

00:00:34BilgusI think the backend is so complex you'd have to re run benchmarks for every code change
00:00:54ulmutulBilgus: Yes, I know, but for other games (or other things) it could be good estimation what's possible and what not (e.g. numnber of simultanous voices for midi)
00:01:16ulmutulDo you think result would vary so much?
00:02:06speachyfor the slower targets, yeah. for the faster targets, it probably doesn't matter.
00:02:33Bilgusevery time there was less buffer it would push those slower targets further and further
00:02:43Bilgusfast targets yeah no problem
00:03:12speachythen you have IRAM-constrained targets, where even a tiny change can have a big effect if it means hot code gets put in or evicted.
00:03:58Bilgusnow some kind of sensible benchmark that runs at install maybe
00:04:30Bilguslike do a system test with some bitcoin mining /s
00:04:58*speachy chuckles.
00:05:26speachythe benchmark can generate a number that is "how many times faster we are than the apollo guidance computer"
00:05:31Bilgussomething that you could get some scores on cpu mem maybe disk
00:05:55Bilgusint overflow.
00:07:08ulmutulBut this wouldn't make too much sense, because it would mean that we must keep everything variable (e.g. max midi voices). And if it's variable, you could set it by hand if you have problems with playback.
00:07:46Bilgusvariable = no go been there tried that
00:09:03Bilgusbest bet is probably just a good error message like sorry your device doesn't support this
00:10:08speachyyep. granted, I'm sure that there are some substantial optimization opportunities available for the midi and mikmod codebases but I don't personally think optimizing that is worth the effort
00:10:13Bilgusnot to say the idea wouldn't work with your midi voices
00:10:17speachyversus hardwre enablement tinkering
00:10:39Bilguswe need a midi guy iow
00:11:51speachymore like algorighm rewriting for the CPU's capabilities including handrolled asm
00:12:21speachythe midi side of things is comparitively simple
00:12:34 Join krabador [0] (~krabador@unaffiliated/krabador)
00:12:59Bilguslua could be used to patch binaries
00:13:54speachydon't know how much of my code remains in scummvm today, but I did write quite a lot of the sci0 music code..
00:14:06Bilgusgiven a nice magic set of bytes to look for it wouldn't be too hard to roll a byte patcher to change some configs
00:16:55Bilgusludwig von speachy?
00:17:16speachywell, since the midi and mikmod plugins are now using the mixer infrastructure, it would be relatively straightforward to change the sample rate on the fly. midi voice count could be optimized to ignore unused voices or somthing like that.
00:21:29speachyor, looking at the code, making the upper voice cap dynamic would be pretty easy. Not sure how we'd detect that we're running too slow though.
00:22:24Bilgusthats the problem with being in the matrix
00:24:09speachyand on that note, time to bail from here.
00:25:35 Quit speachy (Remote host closed the connection)
00:27:28 Quit ZincAlloy (Quit: Leaving.)
00:30:07*ulmutul is going to bed now.
00:30:15 Quit ulmutul (Quit: Leaving)
00:35:05 Join Saratoga [0] (
00:35:20SaratogaPP does use the co processor for some things, mp3 for instance splits decoding across cores
00:35:37SaratogaApi is there if anyone wants to use it
00:38:36SaratogaRewriting critical loops in asm tends to help arm7/9 performance a lot, so profiling is a good idea in case there is some trivial function using a lot of time
00:39:15 Quit Saratoga (Remote host closed the connection)
00:39:29BilgusIf you have one to test that is
00:46:22***Saving seen data "./dancer.seen"
01:00:12 Join bremner [0] (~bremner@debian/developer/bremner)
01:11:01 Join mendelmunkis [0] (
01:17:42 Quit mendelmunkis (Ping timeout: 272 seconds)
01:35:27 Quit Acou_Bass (Ping timeout: 258 seconds)
01:45:08 Join Acou_Bass [0] (
02:43:46 Join deevious [0] (~Thunderbi@
02:46:24***Saving seen data "./dancer.seen"
03:11:50 Join mendelmunkis [0] (
03:24:17 Quit mendelmunkis (Ping timeout: 248 seconds)
03:34:27 Join speachy [0] (d102414d@
03:34:29 Join mendelmunkis [0] (
03:38:46 Quit krabador (Remote host closed the connection)
03:54:46 Quit mendelmunkis (Ping timeout: 272 seconds)
04:37:12 Quit yosafbridge (Quit: Leaving)
04:46:28***Saving seen data "./dancer.seen"
04:47:28 Join yosafbridge [0] (
04:54:43__builtinYep, ipod6g #2 is officially dead
04:54:48*__builtin is back to #1
04:56:38speachysomewhere I have an ipod video that I inherited from an ex girlfriend. hard drive's defective. not sure if it's worth the trouble to do a flashmod to it.
04:57:01 Quit yosafbridge (Ping timeout: 244 seconds)
04:57:04speachyone of the many things I have yet to find after my recent move..
04:58:17*__builtin wouldn't mind some more hardware :)
04:58:19 Join mendelmunkis [0] (
04:58:58 Join yosafbridge [0] (
05:01:16speachyit has the CF-like connector on the hard drive.
05:05:22speachyit probably eloped with my busted-screen Fuze+
05:05:50speachyJust found a Sony PSP, minus a power supply but with a Sims 2 disc in it.
05:06:20speachynow that would be a fun rockbox target. :P
05:07:03mendelmunkiswhat about my DS phat :P
05:11:25 Quit mendelmunkis (Ping timeout: 246 seconds)
05:13:02speachything is I found the hard drive a couple of days ago.
05:13:08speachyand now I can't figure out wher eI put it.
05:19:50speachyjust found its battery.
05:30:28 Quit TheSeven (Disconnected by services)
05:30:37 Join [7] [0] (~quassel@rockbox/developer/TheSeven)
05:31:34Bilgusiirc the DS had a woefully slow cpu
05:31:55__builtinfast enough to run Quake, apparently :)
05:32:41speachyhmm. the rocker's keymap for mikmod is all jacked.
05:33:02Bilgus67 mhz with a 33mhz co processor I guess the co processor might be the saving grace
05:33:54Bilgusthe keymaps for the rocker were all funky probably no one recognized mikmod's
05:34:28Bilgusthe psp would be a good target for the sdl app
05:34:46BilgusI think somebody exposed / ran linux on it
05:35:09speachyplus with actual controls it would be fun to play
05:35:50speachyhuh, hit an out of memory error on a couple of midi files.
05:36:12Bilgushell yeah minus the proprietary mem stick card duo thingie
05:37:48Bilguswonder what the audio hardware was like in the PSP
05:39:28speachyit's basically a portable PS2, so it's pretty capable.
05:39:52speachynow whether or not anyone knows how to drive anything other than basic PCM out of it is another matter. :)
05:40:16BilgusWolfson Microelectronics WM8973G
05:41:03speachyhih, there's a native PSP port of mikmod
05:41:34 Join eevan [0] (
05:42:38speachymikmod doesn't have any ARM asm. but it has ALTIVEC and SSE2 stuff
05:50:17 Quit tchan (Quit: WeeChat 2.4)
05:52:26Bilgusyou'd have thought arm would be more common than altivec well I suppose that shows its age..
05:53:04speachyyeah. guess nobody thought it was worth optimizing by the time android make arm relevant..
05:53:52speachyhmm. wonder what it will take to give the midi plugin more RAM on the Rocker.
05:55:46__builtindoes it grab the audio buffer?
05:56:24speachyI haven't dug that far into how that stuff bolts together.
05:57:43speachythis thing has plenty of RAM. time to bump it. :)
06:01:46Bilgushow much ram does the rocker have?
06:01:58speachywonder if that will help the SDL allocation failures too
06:03:45__builtinthe SDL code could use some serious cleanup...
06:05:03__builtinthere's some ickery with overlay loading into the audio buffer that I haven't documented that well
06:07:21 Join tchan [0] (
06:07:21 Quit tchan (Changing host)
06:07:21 Join tchan [0] (~tchan@lunar-linux/developer/tchan)
06:07:25__builtinah, I remember how it works now
06:07:27speachyRocker has 32MB, with about 9MB free for OS buffers/etc.
06:07:36*__builtin should probably document it while he remembers it
06:09:23speachywolf3d still hangs up hard, with whatever it's doing triggering lots of -ENOMEM errors.
06:09:42__builtindo duke/quake work?
06:09:44speachyplenty of OS memory free (~6MB)
06:10:27speachyoh the midi still runs out of memory, but this time when it's loading drums. so it's definitely a matter of patches being too large for the available space.
06:10:40speachyduke hangs too
06:10:56fs-bluebotBuild Server message: New build round started. Revision 1dabca6, 280 builds, 10 clients.
06:11:21speachyquake too.
06:11:35__builtinall because of OOM?
06:12:04Bilgushow are you grabbing the audio buffer are you specifying a fixed(non movable) allocation?
06:12:58speachyif I believe strace the system is trying to set signal masks and getting -ENOMEM.
06:13:31speachyso it might be the SDL init code that inserts its SIGSEGV etc handlers into the fray.
06:13:52__builtinBilgus: it grabs the audio buffer subtracts it from the plugin load address (in audio buffer) to get the remaining buffer size
06:14:22Bilgusso using the plugin.h function
06:14:33__builtinwhich function do you mean?
06:14:55__builtinit's doing buffer = rb->plugin_get_audio_buffer()
06:15:19Bilgusvoid* (*plugin_get_audio_buffer)(size_t *buffer_size);
06:15:20__builtinand buffer_len = plugin_start_addr - buffer
06:15:42__builtinyes, it's using that, but it must subtract afterwards or else corruption might occur
06:16:11Bilgusthat should do it auytomatically?
06:16:45__builtindoes it?
06:17:05BilgusAs far as I can tell it does
06:17:26Bilgusplugin_get_buffer is the one that needs all the reword
06:17:37Bilguson the pointers
06:17:43*__builtin doesn't know... all the low-level stuff is essentially voodoo to him
06:20:00Bilgusyeah see here this is wodz' doing
06:21:09__builtinso no subtraction?
06:21:59__builtinbut is lua loaded as an overlay?
06:22:08Bilgusnot generally
06:22:14Bilgusbut is on some targets
06:22:36__builtinI think subtracting must be done when a plugin is an overlay + grabs audio buffer
06:23:09Bilgusif so lua is in for a world of hurt on some targets
06:23:22Bilgusand that code is OLDDDD
06:23:42BilgusI think we'd have heard something by now
06:24:03__builtinPictureflow does the subtraction too, and plugins/lib/overlay.c hints at it
06:25:26Bilgushmmm that could be bad
06:25:47__builtin"If the overlayed plugin *does* use the audiobuffer itself, it needs to make sure not to overwrite itself."
06:26:37Bilgusyeah just read that I wonder if it still is true
06:26:56__builtinsome stuff has happened since that was written, I'm guessing
06:27:02__builtinthere is a way we can test it, though
06:27:22__builtinload an overlay, grab audio buffer, and memset the whole thing with bogus
06:27:24BilgusI think the buflib stuff was after 2006\
06:27:50__builtinif buflib/whatever isn't adjusting for the overlay then we'll get a crash
06:28:19 Quit Jinx (Ping timeout: 248 seconds)
06:30:01Bilgusit would almost have to not I mean how would it decide that something was loaded as an ovl?
06:30:25__builtinbuflib *magic*?
06:30:33fs-bluebotBuild Server message: Build round completed after 1177 seconds.
06:30:35fs-bluebotBuild Server message: Revision 1dabca6 result: All green
06:32:40Bilgusbuflib is magical but it isn't that magical lol
06:35:22*__builtin really needs some sleep
06:38:13 Join Jinx [0] (~Jinx@unaffiliated/jinx)
06:46:30***Saving seen data "./dancer.seen"
07:11:32fs-bluebotBuild Server message: New build round started. Revision d61ea6c, 280 builds, 11 clients.
07:26:36fs-bluebotBuild Server message: Build round completed after 904 seconds.
07:26:37fs-bluebotBuild Server message: Revision d61ea6c result: All green
08:21:10 Quit flabrus (Quit: #InstallGentoo)
08:23:08 Join flabrus [0] (
08:23:08 Quit flabrus (Changing host)
08:23:08 Join flabrus [0] (~beard@unaffiliated/flabrus)
08:26:27 Quit [7] (Ping timeout: 264 seconds)
08:28:48 Join TheSeven [0] (~quassel@rockbox/developer/TheSeven)
08:39:03 Quit TheSeven (Ping timeout: 264 seconds)
08:40:31 Join TheSeven [0] (~quassel@rockbox/developer/TheSeven)
08:46:32***Saving seen data "./dancer.seen"
09:04:38 Join wodz [0] (
09:05:32 Join petur [0] (~petur@
09:05:32 Quit petur (Changing host)
09:05:32 Join petur [0] (~petur@rockbox/developer/petur)
10:31:45 Quit wodz (Ping timeout: 258 seconds)
10:46:35***Saving seen data "./dancer.seen"
12:46:37***No seen item changed, no save performed.
13:45:30 Join wodz [0] (
14:28:46 Quit dys (Ping timeout: 245 seconds)
14:32:06 Join massiveH [0] (
14:40:42 Quit deevious (Quit: deevious)
14:41:29fs-bluebotBuild Server message: New build round started. Revision 8d77ec8, 280 builds, 11 clients.
14:46:39***Saving seen data "./dancer.seen"
14:58:15fs-bluebotBuild Server message: Build round completed after 1006 seconds.
14:58:16fs-bluebotBuild Server message: Revision 8d77ec8 result: 545 errors 0 warnings
15:05:40speachyevery time we're almost fully green, something comes along and goes BOOM.
15:06:23speachythis time it wasn't me! :)
15:08:41speachyone-line fix
15:09:20fs-bluebotBuild Server message: New build round started. Revision a430e27, 280 builds, 11 clients.
15:14:20 Quit speachy (Remote host closed the connection)
15:23:27 Join deevious [0] (~Thunderbi@
15:24:32fs-bluebotBuild Server message: Build round completed after 911 seconds.
15:24:33fs-bluebotBuild Server message: Revision a430e27 result: All green
15:45:18 Join speachy [0] (40eebded@
15:50:27 Quit wodz (Ping timeout: 272 seconds)
16:04:36 Quit massiveH (Quit: Leaving)
16:08:02 Quit mikroflops (Ping timeout: 248 seconds)
16:46:41***Saving seen data "./dancer.seen"
16:48:38__builtinspeachy: Bilgus: have either of you received your stickers?
16:51:02speachynot as of yesterday but since it has to be forwarded it will be a few extra days
16:56:26 Quit petur (Quit: Connection reset by beer)
17:03:50 Join mikroflops [0] (~yogurt@
17:04:07speachyfinally got my Model M keyboard plugged in at $dayjob
17:05:29speachyit's been with me my whole career. :)
18:41:48 Join dys [0] (
18:45:58 Join ZincAlloy [0] (
18:46:44***Saving seen data "./dancer.seen"
18:51:46 Quit Jinx (Ping timeout: 248 seconds)
19:25:49 Join lebellium [0] (
19:32:18Bilgus__builtin I'll look tonight
20:14:19 Join MrZeus [0] (~MrZeus@2a02:c7f:70d4:cd00:d87c:7c80:5c8e:1cdb)
20:36:19 Quit dys (Ping timeout: 268 seconds)
20:46:45***Saving seen data "./dancer.seen"
21:18:55 Join dys [0] (
21:50:20 Join ulmutul [0] (~ulmutul@rockbox/developer/ulmutul)
21:51:30 Quit pixelma (Quit: .)
21:51:30 Quit amiconn (Quit: - Chat comfortably. Anywhere.)
21:52:43 Join amiconn [0] (jens@rockbox/developer/amiconn)
21:52:43 Join pixelma [0] (marianne@rockbox/staff/pixelma)
21:58:39ulmutulspeachy: g#2214 looks good so far exept for the typo in midiutil.h line 51. I haven't tried it out however.
21:58:41fs-bluebotGerrit review #2214 at : Introduce HW_SAMPR_MIN_GE_22 macro by Solomon Peachy
21:59:36speachywhoops. missed that one..
21:59:46speachyI didn't build test every permutation
22:00:50speachynew version uploaded
22:01:52ulmutulI've tried out several mikmod files and all played flawless on my PP target. However it's hard to find "the most challenging one"(TM)
22:03:00ulmutulI also repeated the midi frequency analysis test with the old code. 22kHz looks now exactly like expected. Let me upload the picture...
22:04:38ulmutulIt's here:
22:05:00speachysome of the tracker files out of deux ex ought to be stressful
22:05:06speachybut.. they probably are too large to load. :)
22:11:25ulmutulI'll give it a whirl.
22:11:52speachythat new 22K target looks like I'd expect, yeah
22:12:01speachythat dip on the 44K target is interesting
22:12:44speachyis it a function of the particular midi file or instruments? I wonder if a HPF is warranted..
22:19:49ulmutulMaybe related to the original sampling frequency of the patch set. If I try another MIDI files it looks similar, but not as symmetric. I think this is because the displayed one is drums only, and these samples are always played as-is and not pitched higher or lower.
22:21:29speachywell, one of these 26-channel 2MB IT files doesn't bring the rocker to its knees.
22:22:44speachyappears to suck down about 75% of the system resources
22:23:13speachywith reverb & interpolation @ 44K.
22:24:27speachybut this thing has a 1GHz MIPS core in it.
22:24:35speachyprobably the most powerful DAP rockbox runs on..
22:29:04ulmutulHm, can't complain. Deus Ex title song takes ages to load, but plays fine (1.8MB and 23 channels max).
22:29:21ulmutulIs "channels" in IT files the number of voices?
22:30:30speachyyeah, number of active/max voices.
22:31:14speachyhuh, the rocker has a dbus broker running
22:31:38speachyand hte filesystem has mkfs.ext2/3/4 but not the library they depend on.
22:35:08speachywifi and airport support too
22:37:40fs-bluebotBuild Server message: New build round started. Revision 5b23c9e, 280 builds, 12 clients.
22:38:37speachymerged the patch in. Now I'll find if I missed any permutation...
22:46:47***Saving seen data "./dancer.seen"
22:57:10fs-bluebotBuild Server message: Build round completed after 1170 seconds.
22:57:11fs-bluebotBuild Server message: Revision 5b23c9e result: 157 errors 0 warnings
22:57:30speachyoh jeez.
23:04:50fs-bluebotBuild Server message: New build round started. Revision 7737327, 280 builds, 12 clients.
23:07:29 Quit speachy (Remote host closed the connection)
23:11:55 Quit lebellium (Quit: Leaving)
23:16:13 Quit ZincAlloy (Quit: Leaving.)
23:21:16fs-bluebotBuild Server message: Build round completed after 986 seconds.
23:21:17fs-bluebotBuild Server message: Revision 7737327 result: All green

Previous day | Next day