JPEG2000 Performance

[jeffreyke81]

Hi Group,

We are testing the decode performance between Jasper and IPP on our jpeg2000 images (6500px X 8700px). We do see some improvements, but we are not able to duplicate the performace described in the IPP application note (200+ times faster for decoding).

Can anyone please give us some suggestions?
Here are the details of our tests.

OS:
32 bits Centos5

Software:
Jasper: 1.900.1
IPP: l_ipp_ia32_p_5.3.043.tgz
IPP-sample-code: l_ipp-samples_p_5.3.072.tgz

Hardware:
Dual Intel Xeon CPU 5140 @ 2.33GHz
3GB ECC DDR2

Performance:
Jasper (time bin/jasper -f sample_images/0001.jp2 -F out.pnm -T pnm)
real 0m9.715s
user 0m8.803s
sys 0m0.880s
IPP g++ (time ./j2kit -i ~/JP2000/jp2-script/sample_images/0001.jp2 -o out.pnm)
real 0m6.588s
user 0m6.460s
sys 0m1.214s
IPP intel (time ./j2kit -i ~/JP2000/jp2-script/sample_images/0001.jp2 -o out.pnm)
real 0m5.331s
user 0m7.827s
sys 0m1.255s

Is it possible to make it decode faster?
Thanks,
Jeffrey,

-- 
DL Consulting
Greenstone Digital Library and Digitisation Specialists
jeffrey@dlconsulting.com
www.dlconsulting.com

[Mikhail_Kulikov__Int]

Jeffrey,

do you see parallelization remarks in build log (in case of Intel compiler build), like the following:
.srcjp2codestreamebcotdprecinct.h(60) : (col. 1) remark: OpenMP DEFINED REGION WAS PARALLELIZED ?
Please check it to be sure, that you build multi-threaded version.
If you see such a remarks, please try the "set OMP_NUM_THREADS=1" and "set OMP_NUM_THREADS=2" (or 4 for quad-core) and compare performance for images which were coded lossless, generally, it should be noticable.
If you do not see performance gain for right build in lossless mode, please submit issue to the Intel Premier Support (where you register your IPP copy, including evaluation one) and attach materials to reproduce (image file, command options etc.).
Please, also do it if you still unsatisfied with performance comparision.
Note, you published some data without sufficient specification: number of channels (Gray, RGB?), bit depth etc. Also note, that your condition is different from appliction notes, which you probably reffer in many aspects (image size, application environment). Formany codingoptions, even image data itself has ifluence on performance. So it's hard to reproduce or compare as well as to provide exact help for you without exact data.

Best Regards,
Mikhail

[jeffreyke]

Hi Mikhail,

Thank you for the response.

I do see parallelization remarks in the build log and will try to change the number of threads shortly.
You are right, our test images are different from the one used in the application note. We have been trying to find the sample image used in the application note, and by doing this we can find out whether the speed issue is caused by the differences between hardware.

However, we have not yet been able to find the sample image used in the application note.

According to the note:
The JPEG2000 images used in this application note are publicly available from the following URL http://premier.intel.com/ under "Intel IPP Sample" product

I have followed the link, but what I get is:
Welcome to Intel Premier Support.

We were unable to authenticate your access to the Intel Premier Support web site. Please check that your login ID and password were entered correctly and that the URL used was "https://premier.intel.com".

If you have forgotten your login or password, the fastest method to gain access to the system is to use the automated login and password links Forgot your password or Forgot your Login ID on the login page.

If you continue to have problems, please contact Intel Customer Support via email at quadsupport@mailbox.intel.com.

What do I need to do to obtain the sample image?
Thanks,
Jeffrey

[Mikhail_Kulikov__Int]

Jeffrey,

try to click the link "https://premier.intel.com",on the page you quoted, it actually may reffer to other address, depending on your location.It will ask you for login. If the problem persists try to use https://registrationcenter.intel.com/support/.

BestRegards,
Mikhail