arberg
commented
2 years ago I have merged your chances to my branch, and made some performance tests. See the other pull-request. I've made the code multi-threaded, but the blur is incorrect when running in multiple theads. I left the multi-thread code but disabled it. See my branch performance-test for performance-tests, but to work on the multi-threading stay on main so you don't run into trouble committing it to main.
What does this change?
The original stack blur algorithm with its optimized version.
What is the value of this and can you measure success?
The optimized stack blur or you can say multi-threaded stack blur. In this version, the horizontal & the vertical pixels are processed at the same time instead of the nested loop which is enough to provide a good performance benefit.
Considering this (above) the original stack algorithm skips 1-2 frame at the start (not much noticeable if done asynchronously). Since the new algorithm process rows & columns simultaneously this is not seen so far. I've kept the original implementation for you to compare the benchmarks.
Comparison
On average the optimized version takes 200ms less than the original one. I can't provide you a good metric but this is what I got when tested on Android API 21, 23, 27 & 29.
It is necessary that you first compare the two methods on your device as well. Since this project has no instrumentation tests it would be hard to automate it.