« Deepen an existing shallow clone with git | Home | Solving “Only Content controls are allowed directly in a content page that contains Content controls” »
Android codec benchmarks
By admin | February 1, 2012
VLC for Android pre-alpha benchmarks (Feb 1, 2012) for various audio codecs.
Skip to
Specs
Motorola XT300 – Android 2.1, 500 Mhz, 256 MB, rooted with busybox, Qualcomm MSM7225 (ARM1136EJ-S core, ARMv6TEJ architecture, no NEON extensions)
/proc/cpuinfo output
Processor : ARMv6-compatible processor rev 2 (v6l) BogoMIPS : 525.92 Features : swp half thumb fastmult edsp java CPU implementer : 0x41 CPU architecture: 6TEJ CPU variant : 0x1 CPU part : 0xb36 CPU revision : 2 Hardware : QCT MSM7x25 SURF Revision : 0000 Serial : 0000000000000000
Method
adb shell "/system/xbin/busybox ps aux | grep vlc" adb shell "/system/xbin/busybox top -b | grep <pid of VLC as above>"
Ogg files
Comparing three popular codecs for decoding OGG vorbis files, libavcodec from libav, vorbis and tremor.
Results
PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND avcodec: 5266 956 10072 S 163m 88.4 0 25.0 org.videolan.vlc.android vorbis: 5331 956 10072 R 166m 89.8 0 41.0 org.videolan.vlc.android tremor: 5174 956 10072 S 156m 84.5 0 32.4 org.videolan.vlc.android
So, it seems avcodec is the clear winner, followed by tremor and lagging in last place libvorbis.
AC3 files
Test sample
First 30 seconds of http://people.videolan.org/~jb/ac3_the_other_side_44khz.wav
Results – Inconclusive?
| liba52 Mean: 31.51667% Median: 27.15% |
libavcodec Mean: 28.86667% Median: 29.25% |
ac3 with liba52
-A dummy --codec=a52tofloat32 D/vlc ( 6768): using decoder module "a52" Over 30 seconds PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 6768 956 10072 R 160m 86.7 0 52.6 org.videolan.vlc.android 6768 956 10072 S 159m 86.1 0 30.5 org.videolan.vlc.android 6768 956 10072 R 156m 84.5 0 25.6 org.videolan.vlc.android 6768 956 10072 S 156m 84.5 0 27.4 org.videolan.vlc.android 6768 956 10072 S 156m 84.5 0 26.9 org.videolan.vlc.android 6768 956 10072 S 156m 84.5 0 26.1 org.videolan.vlc.android Mean: 31.51667 Median: 27.15 Mode: N/A
ac3 with libavcodec
-A dummy --codec=avcodec D/vlc ( 6829): using decoder module "avcodec" Over 30 seconds PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 6829 956 10072 S 158m 85.5 0 32.3 org.videolan.vlc.android 6829 956 10072 R 155m 83.8 0 30.0 org.videolan.vlc.android 6829 956 10072 S 155m 83.8 0 26.3 org.videolan.vlc.android 6829 956 10072 S 155m 83.8 0 30.0 org.videolan.vlc.android 6829 956 10072 S 155m 83.8 0 28.5 org.videolan.vlc.android 6829 956 10072 S 155m 83.8 0 26.1 org.videolan.vlc.android Mean: 28.86667 Median: 29.25 Mode: 30.0
DTS files
Test sample
First 35 seconds of http://people.videolan.org/~jb/dts_the_other_side_44khz.wav
Results – Also inconclusive?
| libdca Mean: 19.66% Median: 18.85% Mode: 22.60%, 16.80% |
libavcodec Mean: 19.13% Median: 18.25% Mode: 19.3%, 20.3%, 17.6% |
dts with libdca
-A dummy --codec dts Mean: 19.66 Median: 18.85 Mode: 22.60,16.80 Pass 1: D/vlc ( 2461): using decoder module "dts" PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 2461 951 10072 S 159m 86.3 0 21.7 org.videolan.vlc.android 2461 951 10072 S 159m 86.3 0 22.6 org.videolan.vlc.android 2461 951 10072 R 159m 86.3 0 21.3 org.videolan.vlc.android 2461 951 10072 S 159m 86.3 0 24.4 org.videolan.vlc.android 2461 951 10072 S 159m 86.3 0 20.4 org.videolan.vlc.android 2461 951 10072 S 159m 86.3 0 25.6 org.videolan.vlc.android Pass 2: D/vlc ( 2510): using decoder module "dts" PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 2510 951 10072 S 158m 85.6 0 18.0 org.videolan.vlc.android 2510 951 10072 S 158m 85.6 0 22.6 org.videolan.vlc.android 2510 951 10072 S 158m 85.5 0 16.5 org.videolan.vlc.android 2510 951 10072 S 158m 85.5 0 16.8 org.videolan.vlc.android 2510 951 10072 S 158m 85.6 0 16.8 org.videolan.vlc.android 2510 951 10072 S 158m 85.6 0 18.7 org.videolan.vlc.android Pass 3: D/vlc ( 2536): using decoder module "dts" PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 2536 951 10072 S 159m 86.3 0 19.0 org.videolan.vlc.android 2536 951 10072 S 159m 86.3 0 21.5 org.videolan.vlc.android 2536 951 10072 S 159m 86.3 0 16.0 org.videolan.vlc.android 2536 951 10072 R 159m 86.3 0 16.2 org.videolan.vlc.android 2536 951 10072 S 159m 86.3 0 17.3 org.videolan.vlc.android 2536 951 10072 S 159m 86.3 0 18.6 org.videolan.vlc.android
dts with libavcodec
-A dummy --codec avcodec Mean: 19.13 Median: 18.25 Mode: 19.3, 20.3, 17.6 Pass 1: D/vlc ( 2583): using decoder module "avcodec" PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 2583 951 10072 S 162m 87.5 0 34.6 org.videolan.vlc.android 2583 951 10072 S 157m 85.4 0 20.0 org.videolan.vlc.android 2583 951 10072 S 158m 85.4 0 14.6 org.videolan.vlc.android 2583 951 10072 R 158m 85.4 0 15.8 org.videolan.vlc.android 2583 951 10072 S 158m 85.4 0 19.3 org.videolan.vlc.android 2583 951 10072 S 157m 85.4 0 20.3 org.videolan.vlc.android Pass 2: D/vlc ( 2607): using decoder module "avcodec" PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 2607 951 10072 S 159m 86.4 0 25.5 org.videolan.vlc.android 2607 951 10072 S 156m 84.8 0 16.6 org.videolan.vlc.android 2607 951 10072 S 156m 84.8 0 12.8 org.videolan.vlc.android 2607 951 10072 S 156m 84.8 0 14.2 org.videolan.vlc.android 2607 951 10072 S 156m 84.8 0 16.9 org.videolan.vlc.android 2607 951 10072 R 157m 84.8 0 13.4 org.videolan.vlc.android Pass 3: D/vlc ( 2633): using decoder module "avcodec" PID PPID USER STAT VSZ %MEM CPU %CPU COMMAND 2633 951 10072 S 158m 85.8 0 26.8 org.videolan.vlc.android 2633 951 10072 S 158m 85.8 0 20.3 org.videolan.vlc.android 2633 951 10072 R 158m 85.8 0 17.6 org.videolan.vlc.android 2633 951 10072 S 158m 85.8 0 18.9 org.videolan.vlc.android 2633 951 10072 S 158m 85.8 0 17.6 org.videolan.vlc.android 2633 951 10072 S 158m 85.8 0 19.3 org.videolan.vlc.android
If you found this article helpful or interesting, please help Compdigitec spread the word. Don’t forget to subscribe to Compdigitec Labs for more useful and interesting articles!
Topics: Mobile | 2 Comments »
October 4th, 2012 at 18:00
[…] Benchmarks using the latest VLC beta and benching whether a native build solely for NEON is faster or a hybrid build. The expectation is that the native build should be very slightly faster than the hybrid build. This is done on a TI OMAP 4460 (Galaxy Nexus). Method is same as previous benchmarks. […]
January 4th, 2013 at 00:21
[…] previous audio benchmark for […]