Re: Performance issue in coordinate transformations with Gdal 2.3.2
Hi Even, Hi Thomas,
Thanks for your response and sorry for the late reply. School holidays and sickness generated quite some delays across my team.
Anyway, meanwhile we have improved our methodology for measuring the performance, by using JMH (https://openjdk.java.net/projects/code-tools/jmh/). What we found is that coordinate transformations have change their performance by roughly these margins:
GDAL 2.3.3 over GDAL 2.3.1, both with PROJ 4.8.0 (without ETMERC):
+50% WGS84 -> UTM projection and vice versa
+30% between different UTM projections
-10% UTM projection -> DHDN projection
GDAL 2.3.3 with PROJ 5.2.0 over GDAL 2.3.3 with PROJ 4.8.0 (with ETMERC):
+48% WGS84 -> UTM projection
+23% UTM projection -> WGS84
+60% between different UTM projections
+23% UTM projection -> DHDN projection
GDAL 2.4 adds another small margin across the board.
Overall, some of the transformations have indeed doubled their computation time.
Hope this is useful. If anyone is interested in the test sources, do let me know.
I bet you also upgraded your PROJ version in the process ? I can't think of a reason in GDAL itself for a performance change. But on the PROJ side, yes, there might be some explanation
The main one I can see is that in PROJ 4.9.3, the Extended Transverse Mercator is used by default for UTM, which has probably a higher computation time. If you define the GDAL configuration option / environmenet variable OSR_USE_ETMERC to NO, then the older Transverse Mercator method will be used.
PROJ 5.0 has also undergone major changes that might have increased a bit the computation time, but I wouldn't expect them to cause a 2x slowdown.