Parallelization of the Vectorized Reduction

PE0

PE1

Parallel

Execution

Compute

Global Sum

Huge

Overhead

Usually, the global computation should be done once after many steps of particle computation.

This can not be expressed in HPF right now!