07:37 jani: mlankhorst: I have no idea :/
07:41 jani: mlankhorst: looks like it dates back to 2010 and 85345517fe6d ("drm/i915: Retire any pending operations on the old scanout when switching")
08:24 dolphin: dj-death: That discussion has been repeated quite a few times and the short answer is that the pros outweight the cons for feature detection
08:27 dolphin: See the mail thread ZEkQi6Zrb6lR4DEm@phenom.ffwll.local
08:28 dolphin: adixit1: there is no such possibility of knowing apriori, different features are backported to different versions
08:28 dolphin: would not make any sense to try to maintain a list of features drm-tip (or any other kernel has) in the UMD, when you can just check from the KMD underneath
08:43 dj-death: dolphin: well I'll just repeat myself, it's super annoying
08:43 dj-death: dolphin: it's so easy to detect features that even people who write the kernel code can't get it right : https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8064/diffs?commit_id=6c0e41aeaee395ecec93bdd7a7be6386c0b12d24
08:43 dolphin: it should be write-once code which you'll never have to touch again
08:45 dolphin: Well, PXP is used as an example in all the discussions. Maybe the PXP parameter implementation did not go as well as it could have, but doesn't mean we should change direction when the method has been proven in dozens of other cases.
08:50 dj-death: yeah definitely not convinced by the write once :)
10:03 mlankhorst: jani: should probably be removed then
21:36 erlehmann: could it be that on i945 backface culling culls front faces in some contexts?
21:37 erlehmann: like in minetest, where wireframe debug view shows only backfaces if undersampling is on for me: https://github.com/minetest/minetest/issues/13921
21:37 erlehmann: i have this GPU: Intel Corporation Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller (rev 03)
22:03 erlehmann: someone here has an i945 and knows if front face / back face culling order is inverted if rendering to an offscreen target?