Am Donnerstag, 16. Februar 2017, 22:04:37 CET schrieb Matthias Bodenbinder: > Hi, > > since I have a new CPU and mainboard I did a lot of benchmarking. What I am > basically doing is executing > > darktable-cli bench.SRW test.jpg --core -d perf -d opencl > > I got bench.SRW and bench.SRW.xmp somewhere from a darktable group. > > Anyhow, I am comparing the DT processing times shown in the log file like > this: > > [dev_process_export] pixel pipeline processing took 22,524 secs (173,824 > CPU) > > Today I compared DT 2.2.1 with 2.2.3. I did each benchmark 10 times and > compared the results: > > for i in $(seq 1 10); do echo $i; sh bench-script.sh | grep "pipeline > processing took"; done > > Interesting finding: > With DT 2.2.1 the processing time is around 22 seconds while with DT 2.2.3 > the processing time is around 14 seconds. I briefly crosschecked with a > testfile from a Canon 5Ds and the difference is 58 seconds to 34 seconds. > > This is an amazing speed up! Most of the time is saved in the profiled > denoise (from 33 s to 25 s) and equalizer module (22 s to 7 s). I assume > somebody did a really good job in optimizing the code of these modules.
Neither module was changed between those two releases. Did you compile both versions yourself and are you sure that the compile flags are identical? > Good job! Thank you! > > Matthias Tobias
signature.asc
Description: This is a digitally signed message part.
