Further Profiling

[wilfchun]

Hi All,
I have done some further profiling of CME (using a compiler optimization flag (-O3)).
This has been profiled using instruments on Mac. The results are shown in the screenshot below.

Interesting to note that although GDAL isn't directly using too much resource, it appears to be starting multiple threaded tasks in the background.
These threaded tasks are consuming approximately 60% of my brief test case.
This appears to be in varying out a Grid Nearest Neighbor Algorithm, done using quad trees.

The good news is that if we can get around this, might have to build our own algorithm from scratch, we could see significant performance increases.

When I get some time, I will delve deeper into this and see exactly which of CME's GDAL calls is spawning these threaded workers.

[MikeWalkden]

During an an offline chat, Wilf explained that these GDAL operations appear to be used to interpolate wave conditions across the raster from the (CSHORE) profiles.

I wonder whether this is needed, except perhaps in special cases? Also, it may only be needed in local areas, e.g. for the calculation of wave overtopping at a structure. Wave interpolation could perhaps be optional, i.e. a default in which wave heights are only quantified at the profiles (and the raster cells through which those profiles pass) and an option to interpolate across the (wet) raster where and when this is needed?