Details
-
Enabler
-
Must have
-
None
-
Data Processing
-
-
-
8
-
8
-
8
-
1
-
Team_HIPPO, Team_NZAPP
-
Sprint 5
-
-
-
-
15.1
-
Stories Completed, Outcomes Reviewed, Demonstrated, Accepted by FO
-
-
SPO-1592
Description
Who?
- Goal is to provide a solution that is useful to developers working on imaging pipelines - such as RASCIL, but also new pipeline developments using facettting or distributed FFTs
What?
- Gridding / degridding added following newest Processing Function Library standards
- Consider function decomposition & data structures
- Python bindings
- Add testing and test data (ideally check against ground truth - i.e. DFT)
- Provide both CPU and GPU versions
- Interface progressed towards parity with I/O test functions
- Define how to work on partial visibilities (critical for distribution across nodes - per-baseline?)
- Handle partial views of the grid (needed for reducing memory consumption when distributing across nodes - just need an offset)
- Stretch: Handle translated facets (needed for facetting / reducing w-values and therefore number of w-planes)
- at minimum facet offset (i.e. in-plane, WSClean-style)
- shear transformation works better for 3D gridding (w-snapshot-style, see Tim's paper)
- Demonstrate use for imaging?
Why?
- The function needed to demonstrate a SKA processing function library-based imaging pipeline that we need to develop and evolve to demonstrate performance and scaling for ~AA2 size data sets by ~PI18 as part of a set of development of a set of core SDP processing workflows.
- Demonstrate CPU / GPU interoperability
Notes:
- NZAPP's GPU, I/O test CPU, other?
- Adapted to HIPPO's C interface
- Python bindings developed
- TDB: Demonstrate use in Dask based pipeline (possibly from RASCIL)