I take that to mean that when the blend is at 1 or 0, the unused branch of the graph is not calculated
This explains why they use the switch blend rather than switch nodes in so many of the built in graphs ( you get the same performance but it's easier to debug)
Noticed that switch blend takes 1 ms less at 2048x2048 + non binary alpha. So why everybody uses copy blend then ? Is it pre-multiplied alpha vs source alpha ? That said I don't understand this alpha choice either . See no difference in 32 bit mode I usually do my substances due to more predictable and unrestricted height blending .
premultiplied is useful if you have black in the transparent pixels - not something you see that often these days cos we have image formats that aren't shit
I'd imagine people use copy because they don't know what switch does :D also I'd say 1ms is well within margin of error
Replies
I was curious so I also checked the docs.
I take that to mean that when the blend is at 1 or 0, the unused branch of the graph is not calculated
This explains why they use the switch blend rather than switch nodes in so many of the built in graphs ( you get the same performance but it's easier to debug)
Noticed that switch blend takes 1 ms less at 2048x2048 + non binary alpha. So why everybody uses copy blend then ? Is it pre-multiplied alpha vs source alpha ? That said I don't understand this alpha choice either . See no difference in 32 bit mode I usually do my substances due to more predictable and unrestricted height blending .
premultiplied is useful if you have black in the transparent pixels - not something you see that often these days cos we have image formats that aren't shit
I'd imagine people use copy because they don't know what switch does :D also I'd say 1ms is well within margin of error