This project includes the CUDA programs for processing data from a telescope. It is developed as part of the master thesis of Nina Herrmann.
The GPU implementation is separated into two main parts: low-level implementations and high-level implementations. Complementary a CPU program is part of the repository.
The implementation idea is taken from 1 and the following conferences/proceedings: 2,3,4,5,6
However, for using the programm inside a real system you just need the CriticalPolyphaseFilterbank.cu/h files.
This part of the project includes programs written in the CUDA API. The considered architectures are:
| GPU | GeForce930MX | GTX 1080 | K80 (2× GK210) |
|---|---|---|---|
| Purpose in Project | GPU used for local testing | Targeted architecture of the MPIfR | Performance Testing for HPC Taurus |
| Generation | Maxwell | Pascal | Kepler |
| Compute Capability | 5.0 | 6.0 | 3.7 |
The considered frameworks for the high-level implementation will be Musket (Muenster Skeleton Tool). Unfortunately the framework is currently not publicly available. Skeletons provide an additional step of abstraction by including program structures for repeating problems.