Problem running radix sort in SDK 2.3

[shibdas]

Hi,
I'm trying to run the radix sort on a 64-bit linux system with a
Tesla C1060 card. After installing the SDK, if I try to run the radix
sort executable, I keep getting segmentation fault on default run of
1048576 key/value pairs

[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]$ bin/linux/release/radixSort
Using device 0: Tesla T10 Processor

Sorting 1048576 32-bit unsigned int keys and values

Unordered key[2]: 893203 > key[3]: 10792
Segmentation fault

Now, the segmentation fault seems to happening because of some random
values in the keys array which make the indexing the values array (in
the sample run) go crazy. If I run the sort with keys only it succeeds
for some number of keys but fails as the number of keys increases.

[shibdas@tesla1 cuda-sdk]$ bin/linux/release/radixSort -n=10000 -
keysonly
Using device 0: Tesla T10 Processor

Sorting 10000 32-bit unsigned int keys (only)

Sorting : PASS
: elements GPUms
:: 10000 0.35879444

[shibdas@tesla1 cuda-sdk]$ bin/linux/release/radixSort -n=1000000 -
keysonly
Using device 0: Tesla T10 Processor

Sorting 1000000 32-bit unsigned int keys (only)

Unordered key[15]: 329812 > key[16]: 52260
Sorting : FAIL

Debugging further it seems when the code sorts an entire block-load of
data using a set of bits, the sorted order is broken by some random
bit values in between in some cases which propagate through the
following phases.

Has anyone seen this before? I think I might be doing something wrong
here as the radix sort code have been tested to work with the SDK.

Thanks in advance!
Shibdas

[Mark Harris]

Hi,

Can you please report this issue at http:://code.google.com/p/cudpp/
issues?

That way we can track it and fix the bug if one exists.

For now, can you try using the upcoming CUDPP 1.1.1 branch or the
trunk from SVN (the latter will require some code changes for you, but
the former won't)?

To check out CUDPP 1.1.1 use something like the below:

svn checkout http://cudpp.googlecode.com/svn/branches/1.1.1
cudpp-1.1.1

To check out the trunk:

svn checkout http://cudpp.googlecode.com/svn/trunk/ cudpp-trunk

Thanks,
Mark

[sash]

hi

I have often seen radix sort giving wrong answers or throwing
seg faults when you don't have the latest driver of nvidia installed.
I don't really get the connection but it has worked for me once or
twice. So do check if you have the latest drivers.

thanks

s

On Feb 12, 5:35 am, Mark Harris <harr...@gmail.com> wrote:
> Hi,
>
> Can you please report this issue at http:://code.google.com/p/cudpp/
> issues?
>
> That way we can track it and fix the bug if one exists.
>
> For now, can you try using the upcoming CUDPP 1.1.1 branch or the
> trunk from SVN (the latter will require some code changes for you, but
> the former won't)?
>
> To check out CUDPP 1.1.1 use something like the below:
>

> svn checkouthttp://cudpp.googlecode.com/svn/branches/1.1.1

[Shibdas Bandyopadhyay]

Thanks a lot for the responses and sorry for not replying as I thought the problem is with the system having old drivers or incorrect installation as the code seems to run fine on my Macbook using 9400M.

Thanks

Shibdas

--
You received this message because you are subscribed to the Google Groups "CUDPP" group.
To post to this group, send email to cu...@googlegroups.com.
To unsubscribe from this group, send email to cudpp+un...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/cudpp?hl=en.

[Mark Harris]

Shibdas,

If there is still an issue, please do report it: http:://code.google.com/p/cudpp/issues

Thanks!

Mark