- milestone: --> 1.2-dev
- priority: 5 --> 7
- assigned_to: nobody --> kalyanjk
I'm working on a jpeg project and want to beef up the performance. I have no problem to link with FW libraries.
I use IJG (Independent Jpeg Group) as the standard. The platforms are linux 2.4 with 4GB memory each, but with different CPU. The sample photo is of 800x640 true-color. I measured the time of encoding it 1000 times. Here is the result (time relative to IJG, i.e. IJG is 1 and the lower the value the faster, averaged results based on 10 more tests):
CPU IPP FrameWave
P4(SSE3) 0.24 1.07
Amd(SSE3) 0.28 1.15
IPP, as expected, beefs up the performance. However, FrameWave behaves weird.
The Jpeg codec is based on IJG. Since IPP and FW have over 99.9% similarity in APIs, the two binaries actually come from the same source code with different build flags and header/linker sources.
Beyond auto-cpu detection (which was always SSE3 on my two platforms, and that is true), I also tested different cpu types by manually set cpu types. The same result.
Anybody have tested the performance before?
I expect kind helps.
Thank you very much.
Jason
jzou@opentext.com