#asahi-gpu · 2023-06-16 14:11
alyssalina: follow up from wednesday https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/88#asahi-gpu · 2023-06-15 20:31
alyssamany hacks involved but passed the KHR-GLES tests with compression + preambles disabled#asahi-gpu · 2023-06-15 18:42
alyssafragment shader side effects + sample shading = 🤯#asahi-gpu · 2023-06-15 18:41
alyssaOh uffffffffff#asahi-gpu · 2023-06-15 18:33
alyssahit#asahi-gpu · 2023-06-15 18:33
alyssafor the desktop class stuff, possibly we can eat the it#asahi-gpu · 2023-06-15 18:33
alyssafor iPhone SoCs, it'd be a BIG deal for perf there#asahi-gpu · 2023-06-15 18:32
alyssathe only thing I know that uses it is WebGL in Chromium#asahi-gpu · 2023-06-15 18:32
alyssaso yet another reason to defer that until we have a use case#asahi-gpu · 2023-06-15 18:32
alyssalina: I am now observing that MSRTT makes eMRT significantly more annoying#asahi-gpu · 2023-06-15 18:15
alyssalooks like bit 125#asahi-gpu · 2023-06-15 18:15
alyssaOoh tasty new bit in the PBE descriptor for sRGB#asahi-gpu · 2023-06-15 17:48
alyssa(aka "my Apple M1 is an immediate mode renderer now!")#asahi-gpu · 2023-06-15 17:47
alyssagood enough to play supertuxkart with all render targets spilled to memory#asahi-gpu · 2023-06-15 17:47
alyssabut fewer at least#asahi-gpu · 2023-06-15 17:47
alyssastill seeing some HSR related issues..#asahi-gpu · 2023-06-15 17:42
alyssafixed#asahi-gpu · 2023-06-15 17:39
alyssawhich should be trivial#asahi-gpu · 2023-06-15 17:39
alyssanext up, fix partial renders#asahi-gpu · 2023-06-15 17:35
alyssayep ok blend works now#asahi-gpu · 2023-06-15 17:35
alyssayeah ok lol#asahi-gpu · 2023-06-15 17:34
alyssabut actually the problem is HSR, i think#asahi-gpu · 2023-06-15 17:33
alyssaand called flush_memory_to_texture#asahi-gpu · 2023-06-15 17:33
alyssa0, 0, 4 is the barrier I had discovered by bruteforce#asahi-gpu · 2023-06-15 17:33
alyssaOoh#asahi-gpu · 2023-06-15 17:29
alyssaflush the write to other invocations, I guess#asahi-gpu · 2023-06-15 17:29
alyssathe barriers are only inserted when raster order groups are used#asahi-gpu · 2023-06-15 17:28
alyssaalso never did figure out what wait_pix 512, 3 was about#asahi-gpu · 2023-06-15 17:28
alyssaalso, sample_mask 255, 1 at the top of the program#asahi-gpu · 2023-06-15 17:28
alyssaalso interesting that it's inserting these after the write, implicitly#asahi-gpu · 2023-06-15 17:27
alyssathe 0, 0, 4 is notably different here#asahi-gpu · 2023-06-15 17:27
alyssa2, 1, 10#asahi-gpu · 2023-06-15 17:27
alyssa0, 0, 4#asahi-gpu · 2023-06-15 17:27
alyssa3, 2, 10#asahi-gpu · 2023-06-15 17:27
alyssa2, 2, 10#asahi-gpu · 2023-06-15 17:26
alyssaMORE spicy barriers? ah bah oui!#asahi-gpu · 2023-06-15 17:26
alyssa-----#asahi-gpu · 2023-06-15 16:23
alyssaThe bigger issue is DXVK/VKD3D#asahi-gpu · 2023-06-15 16:23
alyssaeasy to patch Zink to do what we need#asahi-gpu · 2023-06-15 16:23
alyssai509vcb: zink doesn't really require customBorderColorWithoutFormat, it already doesn't use it on turnip for perf reasons#asahi-gpu · 2023-06-15 14:54
alyssa(This functionality is definitely supposed to work)#asahi-gpu · 2023-06-15 14:53
alyssaI guess I'll look at how rasterizer order groups work in metal#asahi-gpu · 2023-06-15 14:53
alyssai'm putting wait_pix's in but that doesn't seem to be strong enough#asahi-gpu · 2023-06-15 14:52
alyssadraw #1 writes to an image, draw #2 reads it#asahi-gpu · 2023-06-15 14:52
alyssaok.. my blending is not working with multiple draws together#asahi-gpu · 2023-06-15 13:44
alyssa(For development and CI, anholt's deqp-runner shards the CTS across all available CPUs, which scales almost linearly. But that's not valid for official CTS submissions.)#asahi-gpu · 2023-06-15 13:43
alyssa(CTS is mostly CPU-bound. and the "real" CTS runs are basically single-threaded.)#asahi-gpu · 2023-06-15 13:42
alyssayour M1 Max will not have that problem :-p#asahi-gpu · 2023-06-15 13:42
alyssaGLES3.1 on Mali took ~12h if I remember#asahi-gpu · 2023-06-15 13:42
alyssaThe most annoying part is that the CTS is slow