Active Topics
-
Error after installing Leste on sdcard (17)
to Maemo 7 / Leste by Arno_11 - 1 hr, 1 min ago -
Maemo 20th Anniversary (4)
to Brainstorm by teroyk - 1 day, 6 hrs ago -
Nokia Booklet 3G (RX-75) Drivers (64)
to Alternatives by teroyk - 1 day, 6 hrs ago -
Porting apps to Leste (46)
to Maemo 7 / Leste by teroyk - 3 days, 10 hrs ago - more...
Serge |
2008-01-08
, 23:16
|
|
Posts: 503 |
Thanked: 267 times |
Joined on Jul 2006
@ Helsinki
|
#61
|
Could anyone test A2DP with oprofile and identify/confirm what exactly loads cpu so high? It is quite easy to setup oprofile in OS2008 by following the instructions from http://maemo.org/development/tools/doc/oprofile
|
|
2008-01-09
, 16:59
|
|
Posts: 6 |
Thanked: 0 times |
Joined on Mar 2006
|
#62
|
Originally Posted by Serge
The code is written with inline functions all over the place so without making a debug build you'll just see a lot of utilization in sbc_encode iirc.
Could anyone test A2DP with oprofile and identify/confirm what exactly loads cpu so high? It is quite easy to setup oprofile in OS2008 by following the instructions from http://maemo.org/development/tools/doc/oprofile
|
|
2008-01-10
, 00:06
|
|
Posts: 503 |
Thanked: 267 times |
Joined on Jul 2006
@ Helsinki
|
#63
|
Originally Posted by bmidgley
Just benchmarked sbcenc from bluez-utils-3.24 on N800:
As far as CPU utilization, I suspect it's due to cache misses. A pxa270 with identical size cache and association size 32 runs the encoding at about 4% of the cpu; however the 2420, with its 4 associations and similar speed needs around 50%. There is a section of code that moves memory around on every pass so we'll target that first.
Code:
# time ./sbcenc test.snd > /dev/null real 0m 24.57s user 0m 23.23s sys 0m 0.89s
Oprofile shows the following statistics (with '-fno-inline' compilation option added).
Code:
opreport -l ./sbcenc CPU: ARM V6 PMU, speed 0 MHz (estimated) Counted CPU_CYCLES events (clock cycles counter) with a unit mask of 0x00 (No unit mask) count 100000 samples % symbol name 42423 48.6681 _sbc_analyze_eight 23466 26.9204 sbc_pack_frame 14057 16.1263 sbc_analyze_eight 4017 4.6083 sbc_encode 2492 2.8588 sbc_calculate_bits 414 0.4749 sbc_analyze_audio 118 0.1354 sbc_crc8 78 0.0895 encode 49 0.0562 __write 33 0.0379 __read 21 0.0241 .plt
Code:
time ./sbcenc-fp test.snd > /dev/null real 0m 56.69s user 0m 54.75s sys 0m 1.03s
Code:
opreport -l ./sbcenc-fp CPU: ARM V6 PMU, speed 0 MHz (estimated) Counted CPU_CYCLES events (clock cycles counter) with a unit mask of 0x00 (No unit mask) count 100000 samples % symbol name 139072 67.4416 sbc_analyze_eight 57776 28.0179 sbc_pack_frame 6560 3.1812 sbc_encode 1704 0.8263 sbc_calculate_bits 678 0.3288 sbc_analyze_audio 113 0.0548 .plt 101 0.0490 sbc_crc8 64 0.0310 __aeabi_idiv 59 0.0286 encode 46 0.0223 __read 38 0.0184 __write
Performance of vector_fmul function can become ~13x faster with assembly optimizations. And function 'sbc_analyze_eight' from sbc encoder contains a lot of very similar code which could also use VFP optimizations.
Last edited by Serge; 2008-01-10 at 00:09.
| The Following User Says Thank You to Serge For This Useful Post: | ||
 |
|
2008-01-10
, 01:07
|
|
Posts: 643 |
Thanked: 628 times |
Joined on Mar 2007
@ Seattle (or thereabouts)
|
#64
|
It's great to see people who know what they're talking about working on this. I'm afraid that I waded in to this whole thing, then realized that I was out of my depth. 
At this point, to me at least, it looks like there is something strange about ALSA included with 2008OS. I guess my big question to solve before I give up and call it good enough is: Could a problem with buffers in the ALSA libs on the system be causing mplayer (or xmms) to waste lots of CPU time when outputting to A2DP?
Last edited by Johnx; 2008-01-10 at 01:11.
At this point, to me at least, it looks like there is something strange about ALSA included with 2008OS. I guess my big question to solve before I give up and call it good enough is: Could a problem with buffers in the ALSA libs on the system be causing mplayer (or xmms) to waste lots of CPU time when outputting to A2DP?
Last edited by Johnx; 2008-01-10 at 01:11.
|
|
2008-01-10
, 01:48
|
|
Posts: 47 |
Thanked: 10 times |
Joined on Dec 2007
|
#65
|
Originally Posted by w4csc
Or get them here for $38:
If you're looking for LOUD from a BT tablet headset, it's called MotoROKR S9 and Radio Shack, of all places, recently had them on sale for $79! If you turn up the tablet volume controls and the volume on the headset, in just one earphone so far, the sound is PAINFULLY LOUD!...(c;!
http://www.dealextreme.com/details.dx/sku.2354
|
|
2008-01-11
, 16:29
|
|
Posts: 5 |
Thanked: 0 times |
Joined on Nov 2007
|
#66
|
I just got my Sony DR-BT30Q headphones through from Play.com (only 12.99 ukp delivered) and after some initial problems creating the .asoundrc file, I got mplayer outputting choppy stereo audio, so I'm happy for the moment..
|
|
2008-01-11
, 18:24
|
|
Posts: 643 |
Thanked: 628 times |
Joined on Mar 2007
@ Seattle (or thereabouts)
|
#67
|
OK, I've been hacking on this more. Don't think I've given up or something. I have madplay and aplay compiled and I've been testing them. Piping from madplay to aplay uses about 50% CPU. aplay playing a .wav takes up 20%-30% and no matter what aplay (which claims to set the period size) still can't change the period size from 128. 
I'm going to try something else to track this down. Stay tuned...
I'm going to try something else to track this down. Stay tuned...
|
|
2008-01-11
, 20:49
|
|
Posts: 2 |
Thanked: 0 times |
Joined on Jan 2008
|
#68
|
Appreciate your work on this, I'm a noob as far as the N800 goes (and Linux in general) but I'll help with testing in any way I can. I have the Kyocera wireless headphones. Looking forward to the one-click install package.
Thanks again!
Thanks again!
|
|
2008-01-11
, 22:16
|
|
Posts: 326 |
Thanked: 39 times |
Joined on Jul 2007
|
#69
|
Yeah, I think we would all love a .deb. (I'm not really lazy, I'm just a little ignorant in the ways of programming/Linux in general) I know, I should try and learn something...
|
|
2008-01-12
, 07:06
|
|
Posts: 643 |
Thanked: 628 times |
Joined on Mar 2007
@ Seattle (or thereabouts)
|
#70
|
I've resisted doing a .deb until the situation is a little better, but looking at things now that might take a while. I'll consider doing a beta .deb soon with the understanding that it is *VERY BETA*.
-John
-John
| The Following User Says Thank You to Johnx For This Useful Post: | ||