13:03 <zmike> git log kusma/gallium-move-stride

13:03 <zmike> > hack is massive

13:03 <zmike> yeah this sounds right

13:03 <kusma> I warned ya!

13:04 <zmike> less massive than I hoped

13:05 <kusma> Well, the change isn't massive. The hackyness is ;)

13:36 <zmike> I'm immediately seeing why you didn't get further...

14:39 <zmike> and I think I've managed to talk myself out of doing it at all

14:39 <zmike> which is great because this is so much worse than I imagined it would be

17:36 <anholt> zmike: so cool that you need special labels on vk-gl-cts changes to get khronos to look at them, when you already have to supply test names in "affected tests"

17:37 <zmike> so so so cool yes

17:53 <zmike> anholt: do we have any webgl aquarium traces in ci?

17:54 <zmike> it's the only thing I know of that uses EXT_multisample_render_to_texture, so imo it'd be great to have a few frames in there

17:54 <zmike> (technically dEQP-GLES2.functional.multisampled_render_to_texture.readpixels uses it but I'm not sure how thorough that is)

17:56 <anholt> nope. traces of browser behavior would be pretty great to have, though.

17:57 <zmike> yeah

18:23 <zmike> might try making one for this since it's the thing that keeps me up at night any time I tinker with rendering

21:02 <zmike> this image layout VU got me like https://i.imgflip.com/4/28this.jpg

21:03 <zmike> trying to figure out how to get barrier jit working with rasterization discard + (fake) feedback loop in a way that doesn't end up splitting renderpasses, and it's not going well