Skip to content

Draft: Fix timing following 21.9

This MR is created following a study done to retrieve the timing performance of the default seeder of 21.9 while using the default tracking configuration of the seeder intended for master (in 21.9). The MR port the result of the study into master.

Before the MR, the seeding timing of the default 21.9 seeder and the one intended for master is found to be: timing_vs_Nseeds.pdf

One can notice that the timing of the seeder intended for master is worse than the default seeder. The problem is raised during the pixel seeding: timing_vs_Nseeds_pixel.pdf

This MR pinpoint the code responsible for this timing increase: an additional if loop is add in the pixel seed production, where the size of a container is updated if needed (which is never the case) hence the issue. By replacing the few lines by what is done in the default seeder of 21.9, no segmentation fault is seen, and no impact on the performance is found (the exact same seed are found). The situation in the pixel seeding with this fix is now: timing_vs_Nseeds_pixel_FIX.pdf

Giving a total seeding time of: timing_vs_Nseeds_FIX.pdf

(Note that all plots are done with 21.9 in order to use the same sample)

(@tstreble @pagessin @lgagnon)

Edited by Charline Rougier

Merge request reports