Geometry is first rendered with blending into a multisampled texture, then the texture is rendered onto a full-window plain. The resulting frames differ from one another: there is some color variation where the edges of the geometry are. Color differences aren’t visible to the eye.
In this example we enable blending, draw a cylinder into a texture, then draw all of the 4 samples of the texture onto separate areas of the window. Recorded frames differ.
I’ve got a code of a minimal SDL/GL example (blend-msaa) and nv-nsight-generated code from it. I’ve captured two frames that are supposed to be identical but they aren’t.
Am I using GL properly? Should blending/msaa always give the same result?
(the NULL-pointer segfault that’s in the logs happened only once, seems unrelated to the issue)