Collaborate with IDG expert to further analyse and investigate performance anomalies/limitations IDG

In previous PIs, IDG has been benchmarked (https://confluence.skatelescope.org/pages/viewpage.action?spaceKey=SE&title=IDG+benchmarks+and+profiling). Based on the complexity analysis of IDG, benchmarks tests with different SKA-like parameters have been done. A performance anomaly has been detected: important performance differences between gridding and degridding. We have noticed that the performance difference may vary depending on various factors: configurations, architectures, etc.

To explore the impact of different configurations on performance differences between gridding and degridding, additional benchmark tests will be carried out. The latest version of IDG will be used. We could additionally use a newly implemented tuning test in IDG to get quick performance reports for different settings for Cuda gridder and degridder kernels. Certain specific configurations which have significant performance differences may require further analysis and profiling.

In collaboration with IDG expert, we will explore the relationship between configurations and performance, try to comprehend the cause of important performance differences between gridding and degridding, and identify if it’s a performance anomaly or limitation.