increase maxTriples and Nmax_seed_xz_per_part
maxTriplet and Nmax_seed_xz_per_part limits cause asymmetric qop and phi distributions at high mu on data. Increasing the values seems to fix the issue with negligible throughput cost (at least on data at mu~5). Our current theory is that:
- When the
maxTriplet
limit is reached, the binary search keeps only triplets on one side of the SciFi (from left-right) - If we only increase the
maxTriplet
limit, we then start having too many combinations and then we start reaching the Nmax_seed_xz_per_part. This then starts cutting off the bottom part. Increasing Nmax_seed_xz_per_part fixes that.
This MR temporarily increases these limits. In the future we can perhaps think of a way to also ensure that things stay symmetric in case these limits are exceeded. We should also monitor these cases.
Should resolve #537 (closed)
Edited by Christina Agapopoulou