02:50 <anarsoul> enunes: re !8549 I couldn't get rendering to r and rg working, that's why we don't export the extension

02:51 <anarsoul> technically there's no r and rg render targers, but IIRC there's b and bg

02:52 <anarsoul> oh, not rendering, reload didn't work

02:55 <anarsoul> so exposing and r and rg formats breaks some deqp tests

11:55 <enunes> anarsoul: thanks, I just wanted to push something for the gstreamer users to try for now. do you remember if fixing the r and rg render targets would be a lot of work? like, was it really hard to do, or just not too important or no time for it back then?

11:57 <enunes> it's unfortunate that even if applications just want to read from rg textures, apparently we need to expose rg render target even if they dont need it

17:32 <anarsoul> enunes: I didn't figure out why reload doesn't work for r and rg

17:33 <anarsoul> or rather b and gb

17:33 <anarsoul> it shouldn't be difficult

17:34 <anarsoul> likely we need a swizzle in reload shader

17:35 <anarsoul> since currently we sample it as I which results in vec4(I, I, I, I)

17:35 <anarsoul> it's fixed up in regular fragment shaders, but not in reload

17:37 <anarsoul> since rendering and sampling without reload works correctly I think that reload messes up alpha channel

18:06 <anarsoul> enunes: so I think we need to check swizzle for texel format and if it's not identity - use custom reload shader to apply swizzle

18:07 <anarsoul> there's no need to over-complicate it, so two more reload shaders - one with (x, 0, 0, 1) swizzle for R, another with (x, w, 0, 1) for RG would do

18:54 <enunes> anarsoul: makes sense, I'll look into that