Closed damb closed 3 years ago
Note that this performance improvements are not relevant anymore since the cross-correlation is going to be implemented as an InplaceFilter
. As a consequence, there won't be any branches within the cross-correlation loop, anyway.
This is a reminder for myself.
As @luca-s pointed out, there is still a way to improve the cross-correlation performance in order to allow the CPU to properly make use of vectorization:
References: