Going back to bubble sort, O(n^2) algorithms and Median filtering as an example, these are the practical real-world execution times vs number of stacked images. Trivial SIMD optimized bubble sort is orders of magnitude faster on a smaller, practical/typical dataset size \ While it's O(n^2), but it has very linear, predictable and cache friendly memory access and is easy to SIMD when running across parallel independent streams. A simple naive wikipedia Bubble Sort won't be that fast, slower than qsort. However using mix/max makes it a bit faster. \

Real-world metal cross polarized examples. Nothing is ever pure, it's always a mixture. Left to right: actual photo, specular and diffuse only lit components. It is not uncommon to have differently colored diffuse and specular especially for metals. I hope Metallic PBR dies soon. by "Metallic" I mean BaseColor/Metallic to DiffuseColor/SpecF0 conversion that's not a big issue in offline rendering where you might work with pure materials and mix as many BSDFs as you want at nearly 0 cost. Here is an example of gradual gold deposition on top of blue plastic