__syncthreads_or() predicate evaluation: before syncing or after syncing?

When using __syncthreads_or(int p), is predicate evaluation happening right after all the threads in the block have arrived and synchronized? I had this thought but below post made me doubt:

https://devtalk.nvidia.com/default/topic/521875/cuda-programming-and-performance/a-__syncthreads_or-predicate-question-when-is-60-predicate-62-evaluated-/post/3703450/#3703450