#asahi-gpu · 2021-04-28 00:33
bloomchrisf: "Because we can" isn't good enough?#asahi-gpu · 2021-04-28 00:33
bloomdougall: Oops :-p#asahi-gpu · 2021-04-27 22:28
bloom;P#asahi-gpu · 2021-04-27 22:28
bloomchrisf: are you sure about that#asahi-gpu · 2021-04-27 22:25
bloomfor extra fun the varying is flipped (reads back wz instead of zw)#asahi-gpu · 2021-04-27 22:24
bloom[1] https://gitlab.freedesktop.org/mesa/mesa/-/blob/master/src/panfrost/bifrost/bifrost_compile.c#L887-913#asahi-gpu · 2021-04-27 22:24
bloomThis seems awfully familiar [1]#asahi-gpu · 2021-04-27 22:23
bloombest of both worlds or something#asahi-gpu · 2021-04-27 22:23
bloomgl_FragCoord.xy is derived from get_sr (sr80/81, u2f32, add 0.5) but gl_FragCoord.zw are loaded as a special varying#asahi-gpu · 2021-04-27 20:34
bloomwhere 0 -> 4 and 1,2,3 -> 1,2,3#asahi-gpu · 2021-04-27 20:34
bloomthinking it's just the upper two bits which act as a # of regs#asahi-gpu · 2021-04-27 20:25
bloomI can workaround in the compiler but oops#asahi-gpu · 2021-04-27 20:25
bloom(this is my fault I know :-p)#asahi-gpu · 2021-04-27 20:25
bloomdougall: the `mask` field on ld_var is definitely wrong#asahi-gpu · 2021-04-27 19:00
bloombalrog: yes, but I don't think it's XQuartz's fault, seems like an XNU weirdness#asahi-gpu · 2021-04-27 17:26
bloomNote to self `ipcrm -m $ID` will clear the segment seen in `ipcs`#asahi-gpu · 2021-04-27 03:30
bloomPossibly user error though#asahi-gpu · 2021-04-27 03:30
bloomThe bigger issue is how unreliable XShm is on macOS.#asahi-gpu · 2021-04-27 01:30
bloomHnngh#asahi-gpu · 2021-04-27 01:30
bloom...and then Metal spends just as many retiling it for on-screen composite.#asahi-gpu · 2021-04-27 01:30
bloomTerribly ironic -- I spend a ton of cycles detiling the framebuffer in software for present...#asahi-gpu · 2021-04-23 22:59
bloomI'll upstream when the curves intersect.#asahi-gpu · 2021-04-23 22:59
bloomBut right now the extra overhead of upstream dev for an early project is more than the overhead of rebasing things.#asahi-gpu · 2021-04-23 22:58
bloomIf I wanted I could push the code now and nobody would complain.#asahi-gpu · 2021-04-23 22:58
bloomUpstreaming into Mesa isn't a big milestone for me, I guess#asahi-gpu · 2021-04-23 22:55
bloomhttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10152 was a week ago :-p#asahi-gpu · 2021-04-23 22:54
bloom(Actually, there was a gap. Panfrost was a mailing list hold out. https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1120 was apparently my frist.)#asahi-gpu · 2021-04-23 22:52
bloomI made my first MR to mesa when Mesa first moved to GitLab a few years back :-p#asahi-gpu · 2021-04-23 22:50
bloomdemo#asahi-gpu · 2021-04-23 22:50
bloomTrying to wire textures into the spinning cube dmeo#asahi-gpu · 2021-04-23 17:04
bloomand it's not in system RAM, it's a dedicated register file#asahi-gpu · 2021-04-23 17:03
bloomunfortunately not a number we can *confirm* but it's a solid guess#asahi-gpu · 2021-04-23 17:03
bloomsven: is right#asahi-gpu · 2021-04-23 03:45
bloomat least I know how formats work#asahi-gpu · 2021-04-23 03:45
bloommy "Render target" descriptor looks like a compact flavour of the texture descriptor#asahi-gpu · 2021-04-23 02:55
bloomrn im more concerned about framebuffers#asahi-gpu · 2021-04-23 02:54
bloomthat won't be too hard but I'll wait till the ISA side is known#asahi-gpu · 2021-04-23 02:54
bloomnot piped into the demo yet#asahi-gpu · 2021-04-23 02:49
bloomvery much incomplete but good enough to get started#asahi-gpu · 2021-04-23 02:49
bloomok, checked in cmdstream defs for 2d textures and samplers#asahi-gpu · 2021-04-21 01:55
bloom(Yes, I mean that ironically :p)#asahi-gpu · 2021-04-21 01:54
bloomDifferential analysis still works, though :)#asahi-gpu · 2021-04-21 01:54
bloomThat's good for efficiency but makes it harder to grep for what I'm looking for in a 67k line hexdump!#asahi-gpu · 2021-04-21 01:54
bloomWas unnecessarily hard... the texture descriptor in particular is tightly packed#asahi-gpu · 2021-04-21 01:53
bloomFound the texture/sampler descriptors.#asahi-gpu · 2021-04-20 23:54
bloomnice 👍#asahi-gpu · 2021-04-20 23:54
bloomalthough I would be entirely unsurpirsed if the TODO.writeout instructions functioned as barriers#asahi-gpu · 2021-04-20 23:53
bloomnope.#asahi-gpu · 2021-04-20 23:24
bloompikachu#asahi-gpu · 2021-04-20 23:24
bloomsurprised_pickachu.jpg