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 2005-10-23

00:08:13 Join XavierGr_ [0] (
00:08:18 Quit XavierGr_ (Client Quit)
00:08:49 Join XavierGr_ [0] (
00:16:23linuxstb__I've now got the ffmpeg FLAC decoder working in Rockbox. It seems to be happy at 45MHz :)
00:16:38***Saving seen data "./dancer.seen"
00:17:35amiconnWow.. nice :)
00:18:57amiconnSlasheri: Voice working fine again.
00:19:20linuxstb__There is the very occasional boost, but it's almost always at 45MHz.
00:19:22amiconnThe old clip preemption problem is still there though
00:19:50linuxstb__I haven't implemented seeking yet, and it still needs lots of yields inserting into it. But apart from that, it's working well.
00:19:53amiconnlinuxstb__: Is that with files compressed with -8 ?
00:21:00linuxstb__I'm not sure - I only have one album on my iriver at the moment, and I can't remember the compression settings. I'll do some more tests.
00:30:35 Join DrMoos [0] (
00:32:36 Quit XavierGr (Read error: 110 (Connection timed out))
00:34:55 Quit ender` (Read error: 104 (Connection reset by peer))
00:39:52 Quit JoeBorn ("")
00:40:43linuxstb__amiconn: Yes, it works at the same speed with some -8 files.
00:41:09linuxstb__I still have about 12KB of IRAM to use in some way as well.
00:41:27linuxstb__And this is without any EMAC routines either.
00:43:38XavierGr_wow 45MHz only?
00:43:48 Join arkascha [0] (
00:43:53XavierGr_What about WAV, how much juice does it needs?
00:44:50 Quit Moos (Read error: 110 (Connection timed out))
00:45:07linuxstb__I don't think anyone has every bothered to benchmark WAV.
00:45:43XavierGr_I mean WAV should be the most easy and less consuming right?
00:46:08amiconnaiff would be even easier if it was implemented
00:46:23amiconn(no byteswap on coldfire since it's big endian)
00:46:38linuxstb__Ironically, WAV would be better on the ipod though...
00:47:56XavierGr_maybe it will be good to build a percent table of all codecs (realtime percent). It would be fun to see this.
00:48:16linuxstb__XavierGr_: We have no way to measure that at the moment.
00:48:35linuxstb__All we have is the "boost ratio" - which isn't very useful.
00:48:55XavierGr_what about xxx2wav
00:49:24linuxstb__It hasn't existed for a long time (and was finally removed from CVS today).
00:49:34XavierGr_oh yes
00:50:21XavierGr_then we could make a little utility that calculates how much time the CPU spend in each frequency level.
00:50:58 Quit DrMoos (Read error: 110 (Connection timed out))
00:51:54XavierGr_but i guess that would be difficult, to measure this while the pll changes
00:54:31linuxstb__We need a better measurement than that though - e.g. if two codecs are both working realtime at 45MHz, we don't know which is faster.
00:54:57amiconnMeasuring the CPU time spent at each frequency level would be relatively simple, but that still wouldn't tell us much about codec performance
00:55:15linuxstb__Similarly, if we try to optimise a codec that's already at 45MHz, we don't know how much a change improves it.
00:55:17amiconnWhen there's nothing to do, the CPU goes to sleep
01:10:19 Join Moos [0] (
01:15:09preglowlinuxstb__: did anyone answer your dsp question?
01:15:27linuxstb__About 16-bit ints in 32-bit arrays?
01:15:27preglowlinuxstb__: if not, you set SAMPLE_DEPTH to 16, and use pcmbug_insert_split
01:16:15linuxstb__How does the DSP know I'm using 32-bit arrays, and not 16-bit arrays for each channel?
01:16:23preglowhmm, that's a problem
01:16:33preglowit'll probably assume 16 bit array values :/
01:16:40linuxstb__Slasheri suggested adding an extra config option for array size.
01:16:46preglowyes, so do i
01:18:05linuxstb__Does the DSP make use of the input buffers when processing? i.e. does it work in-place?
01:21:46preglowit depends
01:21:57preglowif the format is deinterleaved, then yes
01:21:59preglowi think it does
01:22:32preglowif it's interleaved, i believe it deinterleaves the data to a temp buffer
01:23:00 Quit arkascha (Read error: 104 (Connection reset by peer))
01:24:03amiconnlinuxstb__: What's the problem with 16 bit ints in 32 bit arrays?
01:24:37preglowamiconn: problem is dsp assumes sixteen bit data type size if you say sample size is sixtee nbits
01:25:21amiconnHmm. I thought data fed to the dsp is always 32 bit, and you only set the number of bits used?
01:26:36amiconnOtoh, why would one place 16 bit ints in 32 bit arrays? Sounds like a space waste...
01:27:34preglowbecause 32 bits is the internal format
01:27:37linuxstb__FLAC uses 32-bit ints internally. In my ffmpeg-based decoder, I've kept it.
01:27:40preglowif it's not 32 bits, it needs to be converted internally
01:28:11CtcpIgnored 2 channel CTCP requests in 8 minutes and 30 seconds at the last flood
01:28:11*HCl stretches and looks around
01:28:31XavierGr_hi hcl
01:28:59linuxstb__preglow: Would it be much work to add support for 24-bit audio. Or is it already there?
01:29:06preglowlike how?
01:29:06*amiconn still doesn't understand
01:29:09linuxstb__I'm thinking about 24-bit FLAC files.
01:29:14preglowahh, like that
01:29:18preglowsupport is already there
01:29:32amiconnIf 32 bit is the internal format, where do the 16 bit ints originate?
01:29:34preglowi believe my emac optimised lpc routine even supports it
01:29:41preglowamiconn: from lossless codecs
01:30:16amiconnNow I'm completely confused
01:30:40preglowinternal format is 32 bit ints
01:30:51preglowthe data may be placed however it wants inside those 32 bits
01:31:06amiconnI think the dsp supports both 16 and 32 bits... so one would use 32 bits with a codec that delivers 32 bits and 16 bitswith a codec that delivers 16 bits
01:31:08preglowlibmad uses only 24, liba52 30, flac 16
01:31:22amiconnI don't see the problem here
01:31:40preglowproblem is the dsp routines would have to be coded twice
01:31:47preglowonce for 16 bit ints, once for 32 bit ints
01:31:50preglowbut i need to go for a while
01:32:07amiconnWhy do the routines need to be coded twice ???
01:48:49preglowcan you code an eq that works equally well on 16 bit array data as 32 bit array data?
01:48:53preglowi can't
01:49:14preglownot without bogging it down with ifs
01:51:29amiconnYou wouldn't need ifs... but then you would read the memory twice for 16 bit data
01:52:13amiconnHow is this handled today?
01:52:35preglowit's not, all data is converted to 32 bit data
01:52:40linuxstb__If anyone is interested in trying out my new FLAC decoder, I've uploaded a binary here:
01:52:46linuxstb__I'll upload the source if anyone's interested.
01:53:13amiconnpreglow: Ah, but then I don't understand linuxstb__'s question
01:53:34 Nick linuxstb__ is now known as linuxstb (
01:54:32linuxstbI think the problem is simply that the DSP assumes 16-bit input arrays if the sample depth is 16-bits.
01:54:44preglowyes, and then converts that to 32 bit data internally
01:54:56preglowproblem is linuxstb has 16 bit data in 32 bit ints, which is what the dsp would have ended up with anyway
01:55:15linuxstbDoes it end up in the high or low 16 bits?
01:55:30linuxstbExactly what I've got then.
01:55:40amiconnYes, and why does the dsp take both 16 and 32 bit input if it converts it to 32 bit internally anyway, and spits out 16 bit?
01:56:04amiconnA bit too many conversions for my taste...
01:56:12preglowsure, we'll remove the worst cases some day
01:57:58amiconnWhat I mean with reading the memory twice is that you can read the 16 bit ints as 32 bits, use the half you want, and then advance the pointer by 2 bytes instead of 4
01:58:15amiconnAt least this is possible on coldfre, but it's somewhat ugly
01:58:53preglowbut yeah, 16 bit intermediate data is a no go
01:58:57preglowwe want 32 bit data for dsp
01:59:01preglowfor precision
01:59:05amiconnlinuxstb: Can I just plug flac.codec into my installation, or does it require core changes?
01:59:27linuxstbI think that should work.
02:00:38linuxstbThings left to do are: Seeking, more internal yields and checking for (and recovering from) errors in the input stream.
02:08:34amiconnwow... That's what I call performance improvement
02:08:52amiconn...and codec size is almost halved
02:09:16linuxstbAnd not a malloc in sight...
02:11:11amiconnBoost ratio with my test album (encoded with -8) around 8% atm
02:12:37linuxstbYes, that's about what I'm getting.
02:13:09linuxstbDoes the backlight boost the CPU still?
02:13:41amiconnBacklight fading does, but I have that disabled atm
02:13:54linuxstbThat's slowing my tests down then...
02:16:08linuxstbDid just replacing the flac.codec work? If so, I'll just upload that so others can test.
02:16:39***Saving seen data "./dancer.seen"
02:17:05linuxstbOK. I've now uploaded just that file:
02:17:37linuxstbIt would be useful if people could test it with their FLAC files before we replace libFLAC with it.
02:22:07linuxstbThe main limitation is that I've set a maximum limit on the block size to 4608. This is so that the buffers will fit in IRAM. Every FLAC file I've got has a block size of 4608 (it's the default for the reference flac encoder).
02:23:14linuxstb4608 pairs of stereo samples @ 32-bits per sample = 36864 bytes.
02:24:12 Quit actionshrimp ("a bird in the bush is worth two in your house")
02:45:42 Join webguest68 [0] (
02:46:29 Quit webguest68 (Client Quit)
02:51:38 Part XavierGr_
02:55:03 Join ashridah [0] (
03:09:16 Quit merbanan (Read error: 104 (Connection reset by peer))
03:10:46 Join XavierGr [0] (
03:14:34 Quit matsl (Remote closed the connection)
03:36:25 Nick ashridah is now known as Lost-ash (
03:55:29 Join _user_ [0] (
03:55:43 Quit _user_ (Client Quit)
04:05:49 Join _Vladoman [0] (
04:16:41***Saving seen data "./dancer.seen"
04:23:10 Quit Vlad0man (Read error: 110 (Connection timed out))
04:23:28 Quit cYmen ("zZz")
06:10:09 Join Tri-NitroToluene [0] (
06:10:19 Quit Tri-NitroToluene ()
06:16:42***Saving seen data "./dancer.seen"
06:23:20 Quit Moos (Read error: 110 (Connection timed out))
06:34:28 Quit RotAtoR ()
06:50:45 Quit Rick ("I don't need to be here.")
07:00:06 Join amiconn_ [0] (
07:15:32 Join Rick [0] (
07:18:33 Quit amiconn (Read error: 110 (Connection timed out))
07:18:33 Nick amiconn_ is now known as amiconn (
07:41:36 Nick Lost-ash is now known as ashridah (
08:16:45***Saving seen data "./dancer.seen"
08:18:01 Join _FireFly_ [0] (
08:53:01 Join DangerousDan [0] (
08:56:31 Join Lost-ash [0] (
08:58:08 Quit ashridah (Nick collision from services.)
08:58:18 Nick Lost-ash is now known as ashridah (
09:10:58 Join solexx_ [0] (
09:14:30 Join Moos [0] (
09:25:13 Quit solexx (Read error: 110 (Connection timed out))
09:34:59 Quit Moos (Read error: 104 (Connection reset by peer))
10:16:48***Saving seen data "./dancer.seen"
10:24:57 Join paugh [0] (n=kickback@2001:5c0:8fff:ffff:8000:0:3e03:6822)
10:57:36 Quit XavierGr (Read error: 110 (Connection timed out))
11:02:04 Join cYmen [0] (
11:07:15 Join cYmen_ [0] (
11:12:27 Join cYmen__ [0] (
11:12:34 Join merbanan [0] (
11:17:40 Join cYmen___ [0] (
11:17:40***Alert Mode level 1
11:17:40DBUGEnqueued KICK cYmen
11:17:40DBUGEnqueued KICK cYmen_
11:17:40***Alert Mode level 2
11:17:40DBUGEnqueued KICK cYmen__
11:17:40DBUGEnqueued KICK cYmen___
11:17:40***Alert Mode level 3
11:19:51 Quit cYmen (Connection timed out)
11:22:49 Join cYmen [0] (
11:22:49***Alert Mode level 4
11:22:49***Alert Mode level 5
11:22:49DBUGEnqueued KICK cYmen
11:22:49***Alert Mode level 6
11:22:49***Alert Mode level 7
11:22:49***Alert Mode level 8
11:24:09 Join matsl [0] (
11:24:58 Quit cYmen_ (Connection timed out)
11:27:59 Join cYmen_ [0] (
11:27:59***Alert Mode level 9
11:27:59***Alert Mode level 10
11:27:59DBUGEnqueued KICK cYmen_
11:27:59***Alert Mode level 11
11:27:59***Alert Mode level 12
11:27:59***Alert Mode level 13
11:30:00 Quit cYmen__ (Connection timed out)
11:30:48 Part amiconn
11:33:11 Join cYmen__ [0] (
11:33:11***Alert Mode level 14
11:33:11***Alert Mode level 15
11:33:11DBUGEnqueued KICK cYmen__
11:33:11***Alert Mode level 16
11:33:11***Alert Mode level 17
11:33:11***Alert Mode level 18
11:35:13 Quit cYmen___ (Connection timed out)
11:35:39linuxstbBagder: Is there a reason that plugins/Makefile still links everything against CODECLIBS ?
11:38:24 Join cYmen___ [0] (
11:38:24***Alert Mode level 19
11:38:24***Alert Mode level 20
11:38:24DBUGEnqueued KICK cYmen___
11:38:24***Alert Mode level 21
11:38:24***Alert Mode level 22
11:38:24***Alert Mode level 23
11:40:51 Quit cYmen (Connection timed out)
11:43:33 Join cYmen [0] (
11:43:33***Alert Mode level 24
11:43:33***Alert Mode level 25
11:43:33DBUGEnqueued KICK cYmen
11:43:33***Alert Mode level 26
11:43:33***Alert Mode level 27
11:43:33***Alert Mode level 28
11:45:39 Quit cYmen_ (Connection timed out)
11:48:43 Join cYmen_ [0] (
11:48:43***Alert Mode level 29
11:48:43***Alert Mode level 30
11:48:43DBUGEnqueued KICK cYmen_
11:48:43***Alert Mode level 31
11:48:43***Alert Mode level 32
11:48:43***Alert Mode level 33
11:50:43 Quit cYmen__ (Connection timed out)
11:53:00 Quit matsl (Remote closed the connection)
11:53:55 Join cYmen__ [0] (
11:53:55***Alert Mode level 34
11:53:55***Alert Mode level 35
11:53:55DBUGEnqueued KICK cYmen__
11:53:55***Alert Mode level 36
11:53:55***Alert Mode level 37
11:53:55***Alert Mode level 38
11:55:17 Part winjer
11:55:52 Quit cYmen___ (Connection timed out)
11:59:07 Join cYmen___ [0] (
11:59:07***Alert Mode level 39
11:59:07***Alert Mode level 40
11:59:07DBUGEnqueued KICK cYmen___
11:59:07***Alert Mode level 41
11:59:07***Alert Mode level 42
11:59:07***Alert Mode level 43
11:59:16 Quit linuxstb (Remote closed the connection)
12:01:25 Quit cYmen (Connection timed out)
12:04:21 Join cYmen [0] (
12:04:21***Alert Mode level 44
12:04:21***Alert Mode level 45
12:04:21DBUGEnqueued KICK cYmen
12:04:21***Alert Mode level 46
12:04:21***Alert Mode level 47
12:04:21***Alert Mode level 48
12:06:21 Quit cYmen_ (Connection timed out)
12:09:11 Join ender` [0] (i=ychat@
12:11:30 Quit cYmen__ (Connection timed out)
12:14:22***Alert Mode OFF
12:16:36 Quit cYmen___ (Connection timed out)
12:16:51***Saving seen data "./dancer.seen"
12:21:54 Quit paugh (Read error: 104 (Connection reset by peer))
12:22:40 Join linuxstb [0] (
12:29:21 Join San [0] (n=Test@
12:52:35 Join markun [0] (
12:52:54markunhi San
13:01:35markunAre you working on rockbox?
13:02:07 Join amiconn [0] (
13:03:42Sanwould like though
13:04:08markunany ideas what you would like to do?
13:05:25 Join arkascha [0] (
13:10:05linuxstbSan: which player do you have?
13:12:35_FireFly_linuxstb: could it be true, that your "new" flac decoder has a bit louder output then the "old" one
13:13:32linuxstb_FireFly_: The decoders themselves are identical - they have to be.
13:13:55linuxstbIt could be some difference in the DSP processing though.
13:14:10_FireFly_hmm either the output is louder or the sound is better :)
13:14:32linuxstbUnless there was a bug in the old decoder. I've confirmed the new one gives bit-perfect output (at least on my PC).
13:15:11 Join paugh [0] (n=kickback@2001:5c0:8fff:ffff:8000:0:3e03:6822)
13:17:58linuxstbWe really do need a WAV writer in Rockbox - so we can check the accuracy of the decoders.
13:36:06 Join Febs [0] (
13:47:42preglowyes :/
13:49:30preglowlinuxstb: is there no better use for iram than the pcm buffer in the flac decoder?
13:50:34linuxstbI don't think so. Most of the processing happens in-place in those buffers. The only other major buffer is the input data - but that's too big to go into IRAM as well.
13:50:54linuxstbDo you want to look at my source?
13:51:19linuxstbIt could be useful to have some functions in IRAM though. Is that possible?
13:51:43preglowyeah, i'd like that
13:51:45linuxstbThe entire decoder is only about 8300 bytes
13:51:54preglowfunctions in iram is possible, but not that much to be gained
13:51:56preglowholy shit
13:52:06preglowwhy the flaming hell is libFLAC the behemoth it is, then?
13:52:07linuxstbThat's the size of flac.codec if I remove the large buffers.
13:52:09 Join _DangerousDan [0] (
13:53:00linuxstbI still need to add seeking support, but that won't be that much code.
13:53:53linuxstbI'm currently using about 38KB of IRAM. So there should be room for the entire codec to be there as well.
13:55:18preglowyou have the code hanging around somewhere?
13:55:28linuxstbI can quickly zip it up for you.
13:55:46CtcpIgnored 13 channel CTCP requests in 46 minutes and 41 seconds at the last flood
13:55:46*amiconn is planning to do some iram section work on codecs
13:57:20linuxstbThis contains a new "flac.c" and "libffmpegFLAC". If you replace libFLAC with "libffmpegFLAC", then you shouldn't have to change anything else to compile it.
13:58:03linuxstbamiconn: Do you think putting the codec itself in IRAM would be a useful optimisation?
13:59:07preglowit will help, yes
13:59:11preglowi've done it before
14:00:03amiconnIt's better to put data in iram than code
14:00:26preglowsure, but when there's no more data
14:00:39 Join muesli- [0] (
14:00:52amiconnSome functions can profit from iram, especially those that are called often, but from all over the place
14:01:08preglowisn't rockbox code supposed to use a four space indent?
14:01:11amiconn...and not several times in sequence
14:01:17amiconnpreglow: yes
14:01:25*linuxstb hides
14:01:38preglowwhere are those rules again? i can't find them in the wiki
14:02:30linuxstbIn the source - docs/CONTRIBUTING
14:02:45preglowlinuxstb: and i think you should access the mem* functions through the codec api pointers, and not depend on the functions in codeclib.c
14:03:32amiconnpreglow: I think we need to correct the mem* function mess at some point
14:03:46preglowamiconn: yes
14:04:35amiconngcc requires that a freestanding environment provides 4 functions: memcpy(), memmove(), memset() and memcmp()
14:05:02amiconnWe need to have a libmem that is linked to the core and to codecs and plugins
14:05:37Sandoes anyone know how to set up an apache server on XP?
14:05:52linuxstbpreglow: I do, but memmove isn't in the codec API struct.
14:06:05 Quit Maxime ()
14:06:16amiconnlinuxstb: Yes, because rockbox doesn't have memmove() yet
14:07:24preglowi think all codec plugins should enable the dsp at all times
14:07:38preglowthis will ease the work for us when we remove the enable dsp option
14:08:03preglowit'll also spur us to fix the dsp layer if something is wrong with it
14:08:26 Quit DangerousDan (Read error: 110 (Connection timed out))
14:09:48linuxstbAny suggestions about how to get the new FLAC decoder into CVS? I don't really want to remove the old decoder until the new one has proven itself and has seeking support.
14:10:49preglowif you post it on the forums or something, i
14:10:55preglowif you post it on the forums or something, i'm sure a lot of people will try it
14:11:03linuxstbYes, I've done that.
14:11:39preglowthe volume is louder?
14:11:43preglowhow can that be?
14:12:15_FireFly_i have the feeling that the volume is louder
14:12:27_FireFly_but it could also be that the quality is better
14:12:42_FireFly_and so i have the feeling that the volume is ouder
14:13:05preglowit's lossless
14:13:10preglowso i don't quite see how any of them can be true
14:13:17preglowwe need a wav writer!
14:13:33linuxstbOne difference is that the new FLAC decoder doesn't always enable the DSP.
14:14:22preglowdsp shouldn't affect the audio unless it's requested
14:14:27preglowif it does, it needs to be fixed
14:14:58preglowbut it's not exactly easy to discover as is
14:16:53***Saving seen data "./dancer.seen"
14:18:41_FireFly_thats only my personal feeling i might be wrong
14:19:50preglowthere might be something behind it
14:19:55preglowand i want to find out, but currently it's not easy
14:27:56_FireFly_i could record a wav with the song over my pc one when the "new" and one when the "old" flac-codec is use
14:29:36_FireFly_i don't have an optical cable
14:29:59preglowthen there's not much of a point
14:30:13preglowunless the differences are huge
14:30:19preglowmight as well try if you want to, of course
14:41:24linuxstbI'll change the encoder to always enable the DSP - that's the only possible cause of any difference I can think of. Unless there are bugs in our libFLAC decoder.
14:41:35linuxstbBut got to go now. Back later.
14:45:19linuxstbI've just uploaded "dspflac.codec" - same location as the other. I haven't tested it, but it should be OK. The DSP is always enabled, in the same way as the existing FLAC decoder.
14:46:37_FireFly_hmm it seems that the output is louder of the "new" codec because the wavrec prog has an peakmeter while recording and there where some clibbing
14:46:46 Quit arkascha (Read error: 104 (Connection reset by peer))
14:48:56 Nick solexx_ is now known as solexx (
14:53:20preglowby how much?
14:53:43_FireFly_i will upload a image
14:55:54CoCoLUSi just updated to the latest daily build
14:56:00CoCoLUSand now rockbox just displays "no files"
14:56:15CoCoLUSand doesn't react to any buttons...
14:56:17CoCoLUSany ideas?
14:56:33_FireFly_preglow: i had startet the recording when i pressed the select button to play the file
14:58:43CoCoLUSis there a way to manually reset the configuration?
15:00:22 Join arkascha [0] (
15:06:16_FireFly_preglow: with the dspflac.codec file it seems that it's the same as the "old" codec
15:07:08preglow_FireFly_: now really
15:07:16preglowthere's a one bit error some place
15:07:41preglowi'll have a look at it in a sec
15:08:11preglowthe audio is almost double the amplitude
15:08:17preglowwhich means dsp some place manages to scale it one bit down
15:08:52preglowbut now for a shower
15:08:56_FireFly_i have just made a record with the dspflac.codec file from linuxstb
15:09:21_FireFly_and now it seems that the amplitude isn't doubled anymore
15:10:58 Join Lost-ash [0] (
15:11:47_FireFly_preglow: and pause works also with the dspflac.codec with the other one pause didn't work
15:11:51CoCoLUSin which sector does rockbox save the configuration?
15:12:38amiconnCoCoLUS: start_of_first_partition - 2, usually sector 61 on harddisks
15:13:54 Quit ashridah (Read error: 110 (Connection timed out))
15:15:11CoCoLUSgreat. found it
15:15:16CoCoLUSnow, how much do i have to delete? :)
15:15:45preglowwe need to remove this DSP_ENABLE stuff
15:15:49preglowdsp should always be enabled
15:16:36preglowit just shouldn't do anything unless it's needed
15:16:37amiconnCoCoLUS: Just this one sector
15:19:24 Nick Lost-ash is now known as ashridah (
15:21:54CoCoLUSa great
15:22:01CoCoLUSoverwriting that sector with zeros helped :)
15:33:08Slasheribtw, why does the rockbox save the configuration to a sector instead of a config file? Better performance?
15:33:09 Quit paugh (Remote closed the connection)
15:34:33SlasheriIt's true that the config structure makes it possible to use eeprom but we could also save that structure to a file so user could delete it
15:35:06crwlmaybe to survive corrupted/emptied filesystems?
15:35:30Slashericrwl: if the filesystem is badly corrupted, rockbox can't boot either..
15:35:40preglowSlasheri: got any thoughts on this flac dsp volume business?
15:35:40Slasheri(unless it's flashed)
15:35:53Slasheripreglow: haven't had yet time to test
15:44:09 Part amiconn
15:54:13 Quit ashridah ("Leaving")
16:09:40 Quit arkascha (Read error: 104 (Connection reset by peer))
16:15:35 Join XavierGr [0] (
16:16:54***Saving seen data "./dancer.seen"
16:24:09 Join XavierGr_ [0] (
16:34:48 Join arkascha [0] (
16:47:08 Quit XavierGr (Read error: 113 (No route to host))
16:56:45 Quit ansivirus (Read error: 110 (Connection timed out))
16:57:05 Quit XavierGr_ (Read error: 110 (Connection timed out))
17:00:04 Quit muesli- (Read error: 113 (No route to host))
17:12:05linuxstbUnless I'm misreading it, line 450 of apps/dsp.c seems to be the culprit. In the case of 16-bit audio with no processing, "scale" is being set to 1, and the data is then being shifted right by scale bits.
17:12:29linuxstbThereby halving the amplitute in all cases.
17:23:21_FireFly_linuxstb: pause doesn't work with your new flac.codec but with dspflac.codec
17:23:38 Quit arkascha (Read error: 104 (Connection reset by peer))
17:26:06 Join paugh [0] (n=kickback@2001:5c0:8fff:ffff:8000:0:3e03:6822)
17:48:21TiMiDI saw there was some changes for X11simulator on today's CVS logh
17:48:40TiMiDbut the remote screen is still 1 row too small :D
17:49:01TiMiD(this doesn't happen under windows)
17:58:58 Join linuxstb_ [0] (
17:59:11 Join Moos [0] (
18:00:16linuxstb__FireFly_: Yes, for some reason pause doesn't seem to work without DSP. But it's not a problem as the plan seems to be to always enable the DSP.
18:00:42 Quit San (Read error: 110 (Connection timed out))
18:16:55***Saving seen data "./dancer.seen"
18:21:37 Join Lear [0] (
18:25:14preglowyes, dsp should always be enabled
18:25:20preglowdithering, etc, should be a user choice, not a codec choice
18:25:40preglowas should dsp
18:27:33linuxstb_preglow: Did you read my comment about line 450 of apps/dsp.c ?
18:27:41preglowyes, checking out now
18:29:42preglowi don't get the + 1
18:30:05preglowSlasheri: comments?
18:30:46Slasheripreglow: Hmm, sounds good :)
18:33:00Learpreglow: I think fracbits is without sign, so to include the sign bit, you need +1. That's how I interpret it anyway (I didn't write the original code...)
18:33:30linuxstb_So fracbits should be set to sample_depth -1 ?
18:33:49linuxstb_In the case of integers
18:34:50_FireFly_to add a sign then you need a multiply with +1 or -1 or i'm wrong
18:36:18preglowi wish this had been documented if that's the case
18:36:32Learlinuxstb_: that's what vorbis.c does. When Tremor does scaling to 16-bits, it shifts right 9 bits, indicating the internal format is 25 bits. Vorbis.c sets depth to 24...
18:37:00linuxstb_As preglow says, where is this documented :)
18:37:11preglowSET_SAMPLE_DEPTH is a very, very bad name for a setting in this case
18:37:12linuxstb_I'm sure most of the codecs do that wrongly then.
18:38:07_FireFly_hmm maybe this is the reason why many users thinks that the sound under iriver fw is better then under rbx
18:39:10LearVorbis and mpa does things in the same way, I think (assuming MAD_F_FRACBITS has the same meaning as in dsp.c :)
18:39:16preglow_FireFly_: doubt it
18:39:49preglow_FireFly_: errors in this setting will result in massive volume changes, and we would have discovered that before now
18:40:14linuxstb_But there WERE errors - the current FLAC decoder for example.
18:40:19preglowin the case of sublte audio wuality issues, that is
18:40:26preglowlinuxstb_: yes, but these people weren't discussing flac
18:41:06linuxstb_What were they discussing? Vorbis and MP3?
18:41:48_FireFly_but maybe only the flac-codec shows the problem better then the other codecs
18:43:05linuxstb_Doing a "grep SET_SAMPLE_DEPTH" *.c in the codecs directory shows that almost all the codecs do the wrong thing.
18:44:00LearFlac sets depth to 16, or "unscaled", so it does the right thing, sort of. It doesn't hurt, at least. :)
18:44:54linuxstb_But the current FLAC output is seems to be half the volume compared with bypassing the DSP
18:45:38linuxstb_Which implies that it does hurt. frac_bits is 16, NATIVE_DEPTH is 16, so scale is 1
18:45:49linuxstb_(jn line 450 of dsp.c)
18:46:20preglownot only does it hurt, it's outright wrong
18:46:22Slasherioh, so there is a bug in the dsp
18:46:36preglowSlasheri: no, we only wonder what SET_SAMPLE_DEPTH argument means
18:46:48preglowSlasheri: i thought it was size in bits, but it seems it is size in bits minus sign
18:47:09Slasheriyes, i also didn't though about that sign bit..
18:47:17Learlinuxstb_: If you set "sample depth" to 16, frac_bits is actually set to 27, so your calculation is likely off. Could still be a problem though. :)
18:47:21Slasherii think we should fix that on the dsp
18:47:35preglowok, but then vorbis.c and mpa.c needs fixing
18:48:12Slasherihmm, is the mpa currently being scaled wrong (volume too high)?
18:48:18LearI don't know how mpc.c and a52.c does it now...
18:49:25linuxstb_Lear: Yes, you're right. I really don't understand dsp.c
18:49:52 Join dpassen1 [0] (
18:50:58LearScaling for Flac look right to me (shift left 12 bits on "dsp entry", then scaled back 12 bits at end).
18:52:38preglowSlasheri: mpa is correct, vorbis is correct, mpc isn't, but that you already know
18:52:43preglowflac isn't correct
18:52:59Slasheripreglow: ah, ok :/
18:53:11Slasherii hoped that mpa and vorbis wouldn't be correct but mpc would ;)
18:53:12Learpreglow: but dsp.c corrects it...
18:54:00LearLine 633: if (dsp->sample_depth <= NATIVE_DEPTH)
18:54:10LearNative depth is 16...
18:55:26LearAnyway, I agree that dsp should probably be cleared up that way (either rename DSP_SET_SAMPLE_DEPTH or change its meaning)...
19:11:45linuxstb__FireFly_: Do you have replaygain enabled?
19:12:55linuxstb_I can't explain the differences your recordings are showing then.
19:13:34linuxstb_I agree with Lear that dsp.c looks correct - it's shifting left 12 bits and then shifting right 12 bits. So the output should be identical.
19:13:35_FireFly_one was made with the "old" flac codec the other with your "new" flac codec
19:13:54linuxstb_Have you compared by new flac.codec and dspflac.codec?
19:14:08_FireFly_if i use the dspflac.codec file then it is the same as the "old" one
19:14:45linuxstb_The only difference is that I have enabled DSP in the second one. In the first, DSP is only enabled if resampling is needed.
19:15:36_FireFly_yepp and the first one has the aplitude doupbled as you can see in the image but the recording was made from the analog-line-out
19:22:58preglowwell, should be quick to verify this
19:23:07preglowmake a flac playing at 44.1, and the same at 48
19:23:22preglowif the dsp screws up, you should hear a definite volume change
19:32:39 Join Mindship-02 [0] (
19:33:47Mindship-02What dows my Linux kernel (2.6.13) need to mount my iRiver iHP140? USB_STORAGE ofcourse, but what about I2C?
19:34:11Mindship-02btw: does rockbox support custom boot screens? ie: my name...
19:34:20 Join RotAtoR [0] (
19:34:39_FireFly_Mindship-02: you only need usb-storage scsi-disk support and vfat
19:34:45_FireFly_to mount the iriver
19:35:22Mindship-02okay, thanks _FireFly_ I was a bit scared because there are a couple of host controllers listed in the kernel-config, but Cyprus is not listed...
19:36:47Mindship-02I just bought a (second hand) iHP 140... is Rockbox stable enough to use as a default 'OS'?
19:37:19Slasheriyes, and you can always dual boot/revert to the original firmware
19:38:06Mindship-02okay Slasheri, I'll give it a try! I'm scared to death though to destroy this expensive machine...
19:38:29linuxstb_preglow: I did your test, and couldn't notice any difference in the volume. The peakmeter was at the same levels as well.
19:38:35Slasherihehe, nobody has yet succeeded in destroying it :)
19:38:45Slasherijust make sure the battery is fully charged before flashing..
19:39:57Mindship-02will do this week
19:40:04Mindship-02need any developers/testers?
19:40:15Mindship-02NB: I won't open it!
19:41:53linuxstb_All contributions are welcome. And don't worry, you don't need to open it.
19:43:19_FireFly_hmm i can't hear any differences in the volume with your test preglow if i use the falc-codec from linuxstb_ where dsp is only used when resampling is needed
19:44:09_FireFly_but i might hear a lower volume if i use either the original flac-codec or the new from linuxstb_ where dsp is always aktivated
19:44:20_FireFly_by both flac-files
19:44:33linuxstb_Can you notice any difference in the peak-meter readings?
19:44:48_FireFly_i don't have a peak-meter running :)
19:44:52_FireFly_so had to test it
19:49:05Mindship-02How many megs should a Rockbox developper think of as wasted? (for test-progs, the rockbox arch, etc.)
19:54:06_FireFly_if i compare the peak-meter values the two files (one 44.1khz the other 48khz) with the falc-codec-file from linuxstb(where dsp isn't aktivated all the time) then there is only a very small diffenrence
19:55:13_FireFly_but if i compare the two codec-files with one flac-file(44.1khz) then there is a greater difference between them
19:56:01_FireFly_the one with has all the time dsp aktivated has a lower value then the other(where dsp is only aktivated for resampling if needed)
20:03:24preglowlinuxstb_: then what the hell triggered firefly's volume change?
20:03:47_FireFly_hmm strange i have mad a falc with −−sample-rate=48000 but metaflac say that the sample-rate is 44100
20:05:27preglowthere will always be a slight difference when resampling from 48khz using linear interpolation
20:05:35preglowwe're looking for a huge difference
20:05:38 Join XavierGr_ [0] (
20:07:55_FireFly_the peak-meter peak-bars stays in one part in the file under the "d" in (no id3) in following text-line "567kBit avg (no id3)"
20:08:25_FireFly_this is for the codec-file which has only dsp aktivated for resampling if needed
20:08:39_FireFly_am i mean if dso is always aktivated
20:09:24_FireFly_if dsp is only aktivated for resampling when needed then the peak-bars stays under the ")" of "(no id3)" in the same text-line
20:09:37_FireFly_i think that is a big difference
20:15:19 Join XavierGr2 [0] (
20:15:28 Nick XavierGr2 is now known as XavierGr (
20:16:29XavierGrwtf? my connection is giving up on me....
20:16:59***No seen item changed, no save performed.
20:18:55 Join muesli- [0] (
20:20:07 Quit linuxstb_ ("CGI:IRC")
20:22:37 Quit paugh ("Leaving")
20:24:45 Join webguest21 [0] (
20:26:01webguest21anyone know about "Save failed No partition?"
20:29:20linuxstbI think it means that for some reason the settings couldn't be saved. Is your battery very low?
20:29:35 Quit XavierGr_ (Read error: 110 (Connection timed out))
20:29:44webguest21Battery is good.
20:30:34linuxstbI'm not sure then. Which player do you have?
20:31:35webguest21I used rockbox on my Archos Recorder V2 for 2 years & loved it. I dropped it a few months ago & just got it repaired by Archos. Now I'm trying to reload the new Rockbox version but every time I play a song the display says "Save Failed No Partition?"
20:31:55webguest21The songs play fine though
20:33:02 Quit Mindship-02 ()
20:37:33linuxstbpreglow, _FireFly_ : I can't reproduce any volume differences with the difference FLAC decoders. So I don't know what's going on.
20:38:44 Quit webguest21 ("CGI:IRC (EOF)")
20:43:10Learwebguest21: have you formated the drive since you got it?
20:49:10linuxstbI've just fixed a few small bugs and omissions in my new flac.codec. I've replaced the version here:
20:49:18linuxstbThis has the DSP enabled by default.
21:48:21 Quit XavierGr (Read error: 110 (Connection timed out))
21:58:58 Join einhirn [0] (
22:01:32 Join XavierGr [0] (
22:04:45 Quit _FireFly_ ("Leaving")
22:15:37 Join tvelocity [0] (n=tony@
22:17:00***Saving seen data "./dancer.seen"
22:32:19 Join amiconn [0] (
22:51:26 Quit XavierGr (Read error: 110 (Connection timed out))
22:58:22 Join matsl [0] (
23:05:36 Quit markun ("leaving")
23:06:37 Quit Lear ("Chatzilla [Firefox 1.4.1/undefined]")
23:08:04 Join webguest23 [0] (
23:09:13webguest23hi all
23:30:06 Quit muesli- (Read error: 113 (No route to host))
23:35:47 Join markun [0] (
23:40:30 Quit ender` (Read error: 104 (Connection reset by peer))
23:43:53linuxstbThe latest ipodlinux CVS commit looks promising:
23:44:12linuxstbAll I have to do is wait for Sourceforge's anon cvs servers to catch up
23:45:31linuxstbSeems the Nano is now supported:
23:51:42preglowi am so getting an ipod if there is any possibility of that being supported by rockbox
23:52:09preglowwhat flash based players do rockbox already run on? ondio?
23:52:20linuxstbThere is definitely a possibility. All it needs is time.
23:52:31linuxstbYes - I think just the Ondio.
23:52:33preglowi'd be happy to port it myself
23:52:45linuxstbGo shopping then :)
23:52:56preglowwhen i get money
23:53:06linuxstbGo working then :)
23:53:13Mooswaiting for Rockbox sponsoring :)
23:53:17preglowand i wonder if they're releasing another version
23:53:23amiconnYes, the Ondio FM and SP are the only flash players supported by rockbox now.
23:53:30preglowamiconn: how does rockbox treat the flash?
23:53:34preglowamiconn: just like the hard drive?
23:53:58preglowhow many write cycles can modern flash take?
23:54:02amiconnSome Ondio specific features are there though
23:54:21amiconnFirst, we have to support FAT16 as well as FAT32
23:54:52amiconnThen the Ondio can control two "disks" - the builtin 128MB flash chip and a pluggable MMC
23:55:41amiconnRockbox supports accessing both at once, by a unix like approach. The internal flash is the "root", conatining the .rockbox folder etc
23:56:08amiconnThe MMC is "mounted" under /<MMC1> Rockbox supports hot plugging the MMC
23:56:53amiconnFlash can take many write cycles today, but nevertheless we kept some write-cycle-reducing features from disk-based rockbox
23:57:22amiconnWhat's good to reduce disk spinups for HDs is also useful for reducing write cycles for flash
23:57:47amiconnNamley the delayed config sector writing

Previous day | Next day