Hi,
I was trying to run SDK radix sort on one of the CUDA servers which supposedly has an installation of CUDA 2.3 SDK but I kept getting segmentation fault on a default run of the executable while a “fail” on other runs.
[codebox]
[shibdas@tesla1 cuda-sdk]$ uname -a
Linux tesla1.ufhpc 2.6.18-128.7.1.el5 #1 SMP Mon Aug 24 08:21:56 EDT 2009 x86_64 x86_64 x86_64 GNU/Linux
[shibdas@tesla1 cuda-sdk]$ nvcc --version
nvcc: NVIDIA ® Cuda compiler driver
Copyright © 2005-2009 NVIDIA Corporation
Built on Thu_Jul_30_09:24:36_PDT_2009
Cuda compilation tools, release 2.3, V0.2.1221
[shibdas@tesla1 cuda-sdk]$ bin/linux/release/radixSort -n=100000
Using device 0: Tesla T10 Processor
Sorting 100000 32-bit unsigned int keys and values
Unordered key[15]: 19341 > key[16]: 18076
Incorrectly sorted value[0] (73477): 2613016277 != 1041
Sorting : FAIL
Press ENTER to exit…
[shibdas@tesla1 cuda-sdk]$ bin/linux/release/radixSort
Using device 0: Tesla T10 Processor
Sorting 1048576 32-bit unsigned int keys and values
Unordered key[0]: 17248 > key[1]: 2949
Segmentation fault
[/codebox]
Is there a problem with the code with new version of the SDK or am I missing something here? I would assume the code would have been tested with the SDK. I’ve tried to compile the source code of the project but keep getting the same result. The code however runs fine with 65536 keys (and values).
I’m wondering if anyone has faced similar problem before starting to hack the code.
Thanks
Shibdas