00:01zmike: robclark: I have similar things in zink and my usage tracking is all on the internal bo, which has had no issues with tc
00:15robclark: yeah, I guess tracking on the bo itself would make more sense.. the current design dates back to before we pulled libdrm_freedreno into mesa, so at the time tracking at the bo level would have been impracticable but now it makes more sense
00:28zmike: never too late to throw out all the old code, as jekstrand always says
00:37robclark: throwing it out is easy.. replacing it with something at least not as bad sometimes takes more work :-P
00:39zmike: I haven't figured that part out yet
01:10jekstrand: Throw it all out!
01:11alyssa: jekstrand: Recycle it?
01:14dcbaker: Nah, reduce() it
01:26alyssa: mapreduce your code
01:26alyssa: see i can buzzword too
01:26LordKitsuna: Reduce, rustify, recompile
01:26LordKitsuna: The three Rs of programming
01:37alyssa: Are we still doing Mesa RiiR? :P
01:38orbea: mesa needs in on this :P https://blogs.gentoo.org/mgorny/2021/02/19/the-modern-packagers-security-nightmare/
01:45alyssa: orbea: Uh oh, I just read this blog post about why Go, Rust, and Python are insecure.
01:45alyssa: I guess I need to remove all three from my system so I don't get hacked.
01:46dcbaker: Don't forget that c and c++ are also insecure. I guess we just need to write everything in haskell
01:55alyssa: dcbaker: I heard Haskell requires category theory and that sounds hards....
02:33dcbaker: I've been teaching myself haskell for great good. It definitely requires a different mindset fire solving problems, sometimes that makes things hard
02:39kisak: burning everything down with fire does make things hard sometimes
02:40alyssa: This is fine.
02:41dcbaker: Lol, I swear autocorrect gets worse the longer i use it
02:45kisak: "I swear officer, I read a piece of code and it was so bad the whole house needed to go."
02:55dcbaker:has written code that bad
03:49imirkin: does anyone remember offhand what happens if you do imageStore() with a value of 0x100 into a r8ui image? does it get clamped? similar question for r8i.
03:53imirkin: and what happens if you try to store e.g. 2.0 into a r8 (unorm) image?
04:03imirkin: i can't seem to find any reference to it in ARB_shader_image_load_store =/
04:10HdkR: Smells like undefined
04:13imirkin: normally hw handles this, but i'm trying to implement this on extra-awesome hw
12:16alyssa: IIRC Mali clamps and seems to pass the tests, but I don't know if it's defined
12:34alyssa: Equivalently it clamps for int16 varyings, which is a problem since the CTS explicitly checks that wraps.
15:15mareko: image stores clamp
16:32imirkin: mareko: thanks!
23:56airlied: zmike: uggh that srgb pbo test shows that relying on bind kinda sucks
23:56airlied: I know it's come up a few times in the past, and it usually comes down to relying on bind is a bad idea