I'm sitting in the Eurostar from rustweek to Paris, running down the track at 300km/h, with a 30min delay of course... and finally found some time for something I wanted to add right after GSoC but never came around to it :D

GPU blend nodes

• add support for N-many image inputs for nodes, not just one
  • enables blend GPU node with 2 inputs
    • I could not get the CPU-based blend node to work, it'll always error out, so can't compare it's outputs
  • Can you think of other nodes this could enable?
• TODO support for 0 input textures
  • the output texture and format is determined from the first input image and format, matching behavior of blend node writing it's result into the first input texture
  • with no first input texture, there's no way to determine the output texture resolution
  • mandelbrot node's output image is created based on viewport and some weird code computing something?
• Makes multiple assumptions on how raster images should behave. I feel like a universal definition on how raster textures should handle things would make this a lot easier:
  • size of the output texture? Assumption: based on first input texture
  • format of the output texture? Assumption: currently first input texture, and upload uses RGBA8 so every texture is that
    • changing away from RGBA8 requires a pass to convert ton RGBA8 before using it with vello
  • 2 input textures with mismatched input sizes: couldn't verify CPU impl, currently just places image in the top left and matches pixel by pixel, no scaling