ok I think this bug was related Bug: Cannot use SSAO and Waterfilter together when multisampling is enabled
And I’ve been too quick to apply the given patch.
Filter output should never be MS because the first pass will resolve MS from the back buffer.
This patch though prevented to have RESOLVE_DEPTH_MS and RESOLVE_MS have different values, whereas it’s a totaly valid case.
The issue is in how we declare the functions with this type. I guess we shouldn’t make the test and always decalre the functions with both types. the selection will be done by the compiler depending on the defines COLORTEXTURE and DEPTHTEXTURE.