Lecture 24: High Performance Image Processing & Halide (71)
RichardChen9
Would a sliding window or box in 2D still have no redundant computation?
GitMerlin
I feel like in 2D cases, if you would like to keep an entire row/column in memory, then you ends up with no redundant computation (just like 1D scenario). However, if you want a memory limit of one (or, a few) box, then you get a schedule with high locality, high parallelism, but a decent amount of redundant work.
Would a sliding window or box in 2D still have no redundant computation?
I feel like in 2D cases, if you would like to keep an entire row/column in memory, then you ends up with no redundant computation (just like 1D scenario). However, if you want a memory limit of one (or, a few) box, then you get a schedule with high locality, high parallelism, but a decent amount of redundant work.