Increased the initial hash size for speedup#1208
Conversation
… size of hash to speed-up the code
gmao-rreichle
left a comment
There was a problem hiding this comment.
@atrayano : Many thanks! Do we know that this change fixes make_bcs for c5760? It would be good to add a note in the PR's introductory comment to that effect. It'll be helpful to record the motivation for the PR for later reference.
|
@gmao-rreichle This PR alone does not fix the issues with make_bcs for C5760. We had to separately fix a bug in MAPL (PR #4381) which already has been merged with develop. However, in the process of troubleshooting the bug, we discovered the 2 tables (iTable and rTable) were not properly initialized after memory allocation. We also changed the hash size to improve performance. So, to produce correct results with make_bcs at C5760, you need to use MAPL from develop (until Matt releases a new tag for MAPL). Using this version of CombineRasters.F90 will help performance. Rasterization of C5760 (quite possibly C2880 and C3072) might need a bigger (finer) raster size ("true" 10 arc-seconds). For example, a grid box near the equator at C5760 is "covered" by only 2x2 raster pixels at our typical 30 arc-second raster 43200x21600). This topic requires a lot of considerations and merits a separate discussion. |
|
I've run BCs package for C5760 and C2880 with:
I can confirmed with this change c2880 and c5760 are zero diff to what we produced for v12 set in the past. Now, I need tag from @mathomp4 for MAPL + to merge this branch so I can update GEOSldas and we can run for v13 set. |
Good to hear! As for MAPL, I did release MAPL 2.66.0. I should probably make PRs to GEOSgcm for that. ETA: I made a PR for MAPL into v11 here: GEOS-ESM/GEOSgcm#1034 |
In raster file processing for tile file generation:
These changes, in combination with a fix in MAPL GEOS-ESM/MAPL#4381, fix errors in the tile file generation for c5760 (make_bcs).