We've been wanting to switch to Radix sort for a while for our own SPH project, and came across the source for the SPH webinar:
http://developer.amd.com/zones/OpenCLZone/Events/pages/OpenCLWebinars.aspx
I'm wondering what the license is, as the source doesn't contain any kind of licensing language. Can we use it in our own open source project?
More on our project:
IANAL if there is no licence then you have no permison to do anything.
I have rewritten a better implementation of the radix sort that performs well for AMD CPU and GPU
http://code.google.com/p/ocl-radix-sort/
I would be also very proud if it was included in the AMD SDK...
Hi Vic,
Have you done performance comparison between your implementation and SDK implementation. Can you post some numbers?
Originally posted by: himanshu.gautam Hi Vic,
Have you done performance comparison between your implementation and SDK implementation. Can you post some numbers?
I have no access to amd devices at office. But I am sure that my implementation is better because in the amd implementation that I know, the sorted list is transferred to the host at each pass and the scan of the histgram is done on the cpu !
Some benchmarks are given here http://hal.archives-ouvertes.fr/hal-00596730
(I am no more anonymous...)
Originally posted by: himanshu.gautam Hi Vic,
Have you done performance comparison between your implementation and SDK implementation. Can you post some numbers?
More precisely my implementation is better than the SDK implementation but does not use some tricks of the Satish algorithm. So it is probably less efficient that the NVidia and the AMD SPH webinar implementations.
Would it be possible to improve the SPH webinar radix sort and put it in a future release of the AMD SDK ?