LAME optimization tests

I did some tests last night using —alt-preset standard on 3.90.2, 3.93.1, and a new compile of 3.93.1 I did with several compiler optimization settings:

-faltivec -mcpu=7450 -mdynamic-no-pic -funroll-loops -ffast-math -fstrict-aliasing -O3

Here are the results of the tests I did. I wouldn’t call them scientific, but I did use the same file for each, and didn’t do anything else while it was encoding.

3.90.2 (download):

[Funktanium-Powerbook:~]coj% lame --nohist --alt-preset standard ~/Desktop/machine_deadagent.wav
LAME version 3.90.2  (http://www.mp3dev.org/)
Using polyphase lowpass  filter, transition band: 18671 Hz - 19205 Hz
Encoding /Users/coj/Desktop/machine_deadagent.wav
     to /Users/coj/Desktop/machine_deadagent.wav.mp3
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.3x) qval=2
   Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA
12471/12474 (100%)|    0:26/    0:26|    5:20/    5:20|   12.345x|    0:00
average: 232.3 kbps   LR: 4079 (32.70%)   MS: 8395 (67.30%)

Writing LAME Tag...done

3.93.1:

[Funktanium-Powerbook:~]coj% ~/Documents/RealBasic/LameBrain/bin/lame --nohist --alt-preset standard ~/Desktop/machine_deadagent.wav
LAME version 3.93  (http://www.mp3dev.org/)
Using polyphase lowpass  filter, transition band: 18671 Hz - 19205 Hz
Encoding /Users/coj/Desktop/machine_deadagent.wav
     to /Users/coj/Desktop/machine_deadagent.wav.mp3
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.3x) qval=2
   Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA
12471/12474 (100%)|    0:24/    0:24|    4:59/    4:59|   13.254x|    0:00
average: 231.8 kbps   LR: 4075 (32.67%)   MS: 8399 (67.33%)

Writing LAME Tag...done

3.93.1 with optimizations (download):

[Funktanium-Powerbook:~]coj% /usr/local/lame_optimized/bin/lame --nohist --alt-preset standard ~/Desktop/machine_deadagent.wav
LAME version 3.93  (http://www.mp3dev.org/)
Using polyphase lowpass  filter, transition band: 18671 Hz - 19205 Hz
Encoding /Users/coj/Desktop/machine_deadagent.wav
     to /Users/coj/Desktop/machine_deadagent.wav.mp3
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.3x) qval=2
   Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA
12471/12474 (100%)|    0:16/    0:16|    4:32/    4:32|   19.578x|    0:00
average: 231.8 kbps   LR: 4075 (32.67%)   MS: 8399 (67.33%)

Writing LAME Tag...done

Note that I did NOT do any listening tests — this is just comparing speed. It looks to me that the plain 3.93.1 is faster than 3.90.2, and the optimized 3.93.1 is significantly faster than 3.93.1 standard. The optimizations I did might keep the binary from working on G3s and older G4s (< 500Mhz, I think), so I want to run some tests on it before including it as standard in LameBrain. If you are willing to give it a shot, please download the binary and let me know the results.

  • yoko
    05/18/2003 03:05:46 PM

    Hi, I tested the optimized binary on my Quicksilver 867 and I get about 1.2 - 1.4x… Any idea why? Thanks

    //yoko z.

  • Ed Finkler
    http://www.funkatron.com
    05/18/2003 03:36:14 PM

    Ya know, I’m getting some bad times now as well… I’m a bit confused about it. I may go back to using the standard compile in LameBrain.

  • Inaki
    05/19/2003 08:52:53 AM

    Hi,

    I too have tested the optimized binary on my Flatpannel iMac/800. I’ve got 2.4913x with the standard compile and 2.8805x with the optimized version. (ripping from cd, r3mix profile)

    Not a substancial speed gain, but faster anyway.