IPP under QNX

[guybrushthreepwood]

Hi there,

has somebody experiences in running the ipp under QNX? Is there a chance to compile it and get it work?

Greetings ... Guybrush

[Vladimir_Dudnik]

Hi,

that's interesting question!

You know, we do not validate IPP on QNX, but does somebody tried to do something with IPP on QNX?

Regards,

Vladimir

[guybrushthreepwood]

Yes, of course I know. But I hope that QNX is similar enough to Linux to get the IPP work...

[Vladimir_Dudnik]

I believe yes. Could you try and say us about your results?

[guybrushthreepwood]

Hi again,

I already have problems with the installation of the ipp. After executing "install.sh" I get the following message:

./linux/bin/ruby: can't access shared library. An unknown error occurred ... falling back to manual installation.

After trying the manual installation a similar error occure:

./install: Can't access shared library.

Something must be different under QNX. Any ideas?

Thanks a lot ... Guybrush

[Vladimir_Dudnik]

Hi, yes problems with installation may happens, but I not think there will be problems with IPP libs. Could you install IPP on linux system and to move libs to QNX system, just to try?

Regards,

Vladimir

[tjkelman]

[bump]

I'm wondering if anyone has successfully used IPP and/or MKL on QNX since this thread died (with no real conclusion it seems)?


Many thanks,
-todd

[Vladimir_Dudnik]

Hi Todd,

seems no one checks, can you?

Regards,
Vladimir

[tjkelman]

Ok, we've now got a x86 QNX6.3.0 SP2 setup, and we're willing to play guinea-pig (i.e. test) both IPP and the MKL. Do you guys (Intel forum moderators) have access to test suites for both products so that we can truly convince ourselves that the libraries are functioning properly under QNX (and not simply linking without errors)?

Feel free to give me a call to follow-up.

Many thanks,
-todd

Todd Kelman, Engineering Consultant
Volcano Corporation
2870 Kilgore Rd.
Rancho Cordova, CA 95670
(916) 231-4476
tkelman@volcanocorp.com

[tjkelman]

Status thus far:

The Intel IPP installation application seg-faults under QNX, so I first had to install the libraries in a true linux environment. I brought the libraries over to QNX and attempted a build with the static IPP libraries, which works (as long as you get the library order correct).

Use of the shared libraries is, unfortunately, a completely different story. The shared libraries are expecting a typical linux installation with glibc (GNU's standard C library, including libm.so.6, libc.so.6, libpthread.so.0, libdl.so.2). Worse, the references to Standard C & POSIX routines inside the IPP libraries are name-mangled only to resolve to glibc's version of them, not to QNX's. Therefore, use of the shared libraries is a no-go, which means that multi-threading (i.e. OpenMP) is a no-go (threading is only available in the shared version of the IPP libraries).

Considering that Intel and QNX are partnering on a number of levels, it doesn't seem like too much to ask to recompile these libraries on a QNX system. A QNX specific Intel compiler is already available, so asking for QNX specific math libraries isn't a stretch and shouldn't take much effort - all the necessary library routines are present in QNX, the source code simply needs to be rebuilt under the QNX environment.

Who do we have to contact at Intel to get the ball rolling on this? A prompt reply would be greatly appreciated.

-todd

Todd Kelman, Engineering Consultant
Volcano Corporation
2870 Kilgore Rd.
Rancho Cordova, CA 95670
tkelman@volcanocorp.com

[Vladimir_Dudnik]

Hi Todd,

thank you for interesting information. As shared library represent complete executable code it probably should be expected that it can't be run on different platform? But seems you can use static libraries this also should be expectable as both systems use the same ELF object format. Could you please submit your request to Intel Premier Support, where we can analize it further.

Regards,
Vladimir

[cultzero]

Hi, Are there any changes in situation with Intel IPP on QNX? As I understood, static version works fine. But it would be very nice if the shared version would be accesible for QNX too.

[Vladimir_Dudnik]

Hello, we do not have shared libraries compiled for QNX. I'll submit your request to Intel Premier Support, so you will have notification if this will be implemented in future releases.

Regards,
Vladimir

[cultzero]

Thanks a lot!

[Amanda_S_Intel]

Hi,

I just wanted to give everyone an update on QNX* support for Intel IPP. As you already know, Intel IPP does not currently validate on QNX and as of right now there is no plan to support QNX. As a product group, we are aware that some customers are interested in it and we are constantly evaluating how we can support this platform properly, based on our available resources.

Regards,

--Amanda S.

Intel IPP Support

[kostadin_vardin]

Does anybody know what is the correct order of the libraries for the linkage.

I bought Intel IPP and transfered them to QNX632 and desparately trying to link them staticly, but no success. For any function I got the damned "unresolved reference ..."

Please for help!

[Vladimir_Dudnik]

Remember, that on Unix-like systems the order or linking is important. try to re-order libraries in your link command

I assume you are using static link with IPP static not threaded libraries.

Regards,
Vladimir

[kostadin_vardin]

Hi Vladimir,

Finally I succeeded to link and use Color Conversion functions from the library under QNX 632. I found very usefull conversion function from Bayer8 to RGB888 - ippiCFAToRGB_8u_C1C3R() and it works great.

To boost the performance of rendering I would need direct conversion to RGB565.

For now I am passing RGB888 data to the ippiRGBToRGB565_8u16u_C3R, but this is additional processing. I am wondering how difficult would be to implementdirect conversion from Bayer8 to RGB565.

[Vladimir_Dudnik]

Hello,

I would recommend you to submit your request to Intel Premier Support so it will be considered at the next IPP version planning stage. From my perspective, 555 or 565 formats are going out from the scene, why do you use them? With modern graphics hardware it is probably no visible difference in performance between true color resolution and these old formats?

Regards,
Vladimir

[bcor]

hi all
i read the topic and have been thinking maybe there are some changes now.
i am considering to buy intel mkl in order to use on qnx.
i wonder if i can use it or not.
or you have any other advices to use (some complex) mathematical functions on qnx?

[Naveen_G_Intel]

Hi,

If you are looking for complex mathematical functions, then Intel MKL is perfect library.

MKL forum is right place to discuss any information about usage of MKL with experts/developers.

Regards,

Naveen Gv