+ /* Some OpenGL driver versions require a depth format for sampler comparison,
I think that's the correct/expected behaviour. I think there are two fairly straightforward ways to address that though: - Declare the format for the SRV as something like "format r32 float shadow" in the test, and then use either GL_DEPTH_COMPONENT32F or GL_R32F depending on the "shadow" flag. - Simply always use GL_DEPTH_COMPONENT32F for the "r32 float" format. Depth formats are valid texture formats in OpenGL. We wouldn't then be able to create e.g. a "r32 uint" view on that texture, but at least for the moment that's not a concern for the shader runner.