Intel® Integrated Performance Primitives
Community support and discussions relating to developing high-performance vision, signal, security, and storage applications.
Announcements
This community is designed for sharing of public information. Please do not share Intel or third-party confidential information here.

JPEG decoder showdown : UIC vs IJL

Gaiger_Chen
New Contributor I
131 Views
HI

Below figure is my benchmark in jpeg decoder :: IUL vs UIC

the input is an image captured from WOW, 1440*900.

I encoded it as YUV420 for each quality, then I used UIC and IJL to decode the encoded image for 100 times, in order to find the better decoder.

as you see, the IJL is much better than UIC, BOTH serial and parallel version.

I thought the result is not reasonalbe: thought IJL is a excellent good code for jpeg codec, but it is too old

to fellow up hardware (Not necessarily, maybe the ipp kernal works very well). So maybe the UIC actually could win IJL.

May I ask, if the result(which I upload) be true? if not, what wrong I makes? like preprocessor define, compiler optimized selection...etc..

test platform :


Intel Pentium Dual Core E6500 @ 2.93GHz
windows XP sp3
VS 2005
intel compiler 10.1
(icc and vs8 got the same result.)
IPP 6.1

thank you.
32306-uic_ijl_jpeg_decoder.jpg
0 Kudos
3 Replies
Vladimir_Dudnik
Employee
131 Views
Hello,

accurate measurement of performance might require careful consideration of how you build performance test.

There is also known difference beween IJL and UIC - UIC codec provides better computational accuracy. So, if you also compare decoded images agaisnt original one, you will find that UIC codec introduce less coding artifacts, what cost some additional processor time.

Regards,
Vladiir
Gaiger_Chen
New Contributor I
131 Views
Hi
So, it is not simple way to fast up the UIC jpeg decider???

If it is, thank you.



Vladimir_Dudnik
Employee
131 Views

UIC JPEG codec was intentionally implemented as a more robust and more precise JPEGcode than IJL. That of course have some associated cost in performance.

Vladimir

Reply