CUDPP streaming interface

[GoogleCodeExporter]

We're interested in using cudpp in the middle of an algorithm that 
works with fairly small arrays of different sizes.  We'd like to use 
separate streams to process this data to use all of the SMs, but it 
seems that cudpp doesn't expose a streaming interface.  Are there any 
plans to expose a streaming interface in CUDPP, maybe something like: 

CUDPPResult 
cudppSetStream( CUDPPHandle planHandle, cudaStream_t stream ); 

This would be consistent with with CUFFT and CUBLAS libraries. 

Original issue reported on code.google.com by mike.sa...@gmail.com on 9 May 2011 at 3:51

[GoogleCodeExporter]

Original comment by jow...@gmail.com on 9 May 2011 at 4:03

• Added labels: Type-Enhancement
• Removed labels: Type-Defect

[GoogleCodeExporter]

Pushing this back to 2.1 or later.

Original comment by harr...@gmail.com on 6 Jul 2011 at 2:10

• Added labels: Milestone-Release2.1

[GoogleCodeExporter]

Issue 104 has been merged into this issue.

Original comment by harr...@gmail.com on 9 Aug 2011 at 4:29

[GoogleCodeExporter]

Issue 102 has been merged into this issue.

Original comment by harr...@gmail.com on 9 Aug 2011 at 4:29

[GoogleCodeExporter]

Issue 103 has been merged into this issue.

Original comment by harr...@gmail.com on 9 Aug 2011 at 4:29

[GoogleCodeExporter]

Issue 105 has been merged into this issue.

Original comment by harr...@gmail.com on 9 Aug 2011 at 4:29

[GoogleCodeExporter]

I'm also interested in this functionality, and I'm willing to implement it.

Is the suggested interface the right choice? CUFFT and CUBLAS only seem to 
allow setting the stream globally, not per-plan as suggested here, but maybe 
per-plan is better?

In either case, the implementation should be straightforward:
- Add a cudaStream_t member in either CUDPPPlan or CUDPPManager, initialized to 
0 by default.
- Add the cudppSetStream function.
- Use the stream in all kernel invocations.

Any feedback on this?

Original comment by m0b...@gmail.com on 9 May 2012 at 12:22