delete some redundant logic using cg::this_grid().sync();

[zhoutianzi666]

code seems redundant,
I delete some logic using cg::this_grid().sync() instead ?

[alpha-baby]

maybe reduce performance when directly use cg::this_grid().sync();

[sphish]

The original code logic was intended to allow the NIC to start working as early as possible, thereby reducing latency. This is especially beneficial when the batch size is small. Have you tested the performance impact of your changes?

[zhoutianzi666]

The original code logic was intended to allow the NIC to start working as early as possible, thereby reducing latency. This is especially beneficial when the batch size is small. Have you tested the performance impact of your changes?

I tested, it does not hurt performance.