OpenCL* for CPU
Ask questions and share information on Intel® SDK for OpenCL™ Applications and OpenCL™ implementations for Intel® CPU.
Announcements
This forum covers OpenCL* for CPU only. OpenCL* for GPU questions can be asked in the GPU Compute Software forum. Intel® FPGA SDK for OpenCL™ questions can be ask in the FPGA Intel® High Level Design forum.
1719 Discussions

Unable to fix a bug, '32 bits int' problems ?

Polar01
Beginner
447 Views
Hi,
I'm working on the CLPP OpenCL Library and can't find a bug, so I request some engineering support from the OpenCL Intel Team... of course if possible.
The goal is to develop a radix-sort algorithm, the problem is that I'm unable to sort on 16 bits and 32 bits.
(All others are working fine).
So, my questions are
1) does the OpenCL 'int' representation is different than the 'Win 7 64 bits' int ? (I think no !)
2) do I correctly generate random 'int' values for OpenCL ?
3) can you help me to fix it ? I'm searching to fix it since a long time !
BTW: The code is herehttp://code.google.com/p/clpp/
You can change the "bit" sorting parameter in 'benchmark.cpp'
Krys
0 Kudos
3 Replies
ARNON_P_Intel
Employee
447 Views
We continue to provide support through this forum. Any report is commited, screening, and prioritized.
We will solve bugs related to our implementation, but for that, you shall provide us with specific code that generates the bug and description of the bug. We will not debug a general OpenCL app as we have many of those.
Regards,
0 Kudos
Eli_Bendersky__Intel
447 Views
Hello Polar01,

Answering your numbered questions:

1) Note that according to the OpenCL standard, an 'int' type is always "A signed twos complement 32-bit integer.", on all platforms that run OpenCL. You can also use the same type in your host application as "cl_int".
2) Similarly to any random-number-generation technique implemented in C. You can find any number of algorithms for doing this, in books and online. Usually RNGs depend on some OS-specific seeding (time, PIDs, user actions) - these can be provided from the host to the kernels. Alternatively a whole buffer of pre-generated random numbers can be provided from the host to the kernels. It all depends on the specific needs of your application.
3) It is not clear what it is that you want us to help you fix. Can you be more specific?

In general, wherever the behavior of the Intel OpenCL SDK doesn't match your expectations, please provide a minimal code sample and describe the behavior you expect vs. the one you're seeing.

Good luck
0 Kudos
Polar01
Beginner
447 Views
Thanks a lot for your replies,

I completely agree with you but to be honnest I have find no way to debug this problem.
I generate my 'int' like I generate normal C++ int, so it should be the same I think ! (I'm on Win7 64bits)
I have try several methods to generate my random numbers, but even with 'scoping' my values ( something like X=max(min(0, X), MAXVALUE) ) I still got wrong results ! (I have even use the way NVidia generates its numbers for their samples).
So, maybe there is a problem in the way I generate my numbers... but I still think that OpenCL int are like C++ int ! So, I don't see the difference !
What is strange, is that I have 2 sort algorithms, one for CPU and one for GPU, but I have the same problem with both ! There is not a lot of common stuffs... mainly the way I generate numbers.
I don't know where the problem is coming from ... and because there is no OpenCL Debugger on the CPU I don't know how to process (except printf !).
I don't know "what" can be "fixed" because I have no idea of the problem. Maybe it is a bug "I" have do and I just need some councils :-P
BTW: to debug, you don't need to test the whole program, just the CPU radix-sort in the "benchmark.cpp" (This class is a unit testing class, so you can enable/disable each algorithm)
Thanks
Krys
0 Kudos
Reply