#asahi-dev · 2023-01-03 11:03
marcanthe PCIe subsystem doesn't have an understanding of runtime PM#asahi-dev · 2023-01-02 15:16
marcanoh god and my reply flew right over the T2 guy's head... *sigh*#asahi-dev · 2023-01-02 14:04
marcanbecause y'all from the actual company don't bother answering questions about how to do actually do this *correctly*, that's why#asahi-dev · 2023-01-02 14:04
marcanlol @ the broadcom guy ignoring my original email then replying to the silly-but-safer proposal from the T2 guy asking WTF#asahi-dev · 2023-01-02 13:57
marcanit's kind of useless#asahi-dev · 2023-01-02 13:57
marcanthough I don't know why anyone would want to use that format anyway#asahi-dev · 2023-01-02 13:57
marcancorrect, we shouldn't#asahi-dev · 2023-01-02 13:00
marcanso yes, this means we can support xrgb8888 in the bottommost plane only#asahi-dev · 2023-01-02 12:59
marcanit's only xrgb8888 that is missing#asahi-dev · 2023-01-02 12:59
marcanand most yuv formats are already alphaless#asahi-dev · 2023-01-02 12:58
marcanit's x2#asahi-dev · 2023-01-02 12:58
marcanthere is no support for a2rgb10#asahi-dev · 2023-01-02 12:58
marcanjannau: 10-bit is already xrgb#asahi-dev · 2023-01-02 05:06
marcanI think the bits in swap_completed may actually enable swap presentation timestamping or something like that?#asahi-dev · 2023-01-02 05:04
marcanjannau: the 0xff000000 bg_color is just setting it, and bit 31 of the swap_enabled stuff isn't clear, it's background set enable#asahi-dev · 2023-01-02 04:57
marcanor something like that#asahi-dev · 2023-01-02 04:57
marcanand then 2 is probably meant as a YUV plane (I think it should be an underlay, not an overlay, to allow UI element compositing on top) and 1 as an overlay that can do more advanced blending#asahi-dev · 2023-01-02 04:56
marcanI bet they added 3 for HDR or something like that#asahi-dev · 2023-01-02 04:55
marcanthat's why we have 4 when hardware only supports 2 or 3#asahi-dev · 2023-01-02 04:55
marcanthe surfaces probably have specific meanings, they aren't just arbitrary layers#asahi-dev · 2023-01-02 04:55
marcanjannau: I think the blend order is BG, 2, 0, 1, 3? And some hardware can support only 2 layers enabled at once, some 3#asahi-dev · 2023-01-01 23:40
marcanso then you're expected to use surfaces 1/2 for compositing#asahi-dev · 2023-01-01 23:40
marcanjannau: I'm getting the feeling surface 0 is a special "passthrough/optimized" surface and that's why it doesn't do full colorspace conversion on t6k#asahi-dev · 2023-01-01 16:47
marcanaw, I guess I was giving apple too much credit#asahi-dev · 2023-01-01 16:42
marcanbut not windowed?#asahi-dev · 2023-01-01 16:41
marcansomeone benchmarked it as having the GPU completely off while playing video#asahi-dev · 2023-01-01 16:41
marcanjannau: safari is supposed to (at least with the right codecs)#asahi-dev · 2023-01-01 16:40
marcankettenis: we know at least uncore is clocked in shallow WFI mode since it burns significantly more power on higher pstates, so I still expect more savings since it should at the very least stop the clock to the cluster entirely if all cores are in deep idle#asahi-dev · 2023-01-01 16:30
marcanjannau: can't find anything :(#asahi-dev · 2023-01-01 16:28
marcanthe sysfs interface is deprecated and no longer buildable without CONFIG_EXPERT#asahi-dev · 2023-01-01 16:28
marcanChaosPrincess: use the chardev interface#asahi-dev · 2023-01-01 16:00
marcanunk_10c in the swap_rec is bg_color, which if nothing else is useful to test the alpha#asahi-dev · 2023-01-01 16:00
marcannow I'm just trying random stuff#asahi-dev · 2023-01-01 15:59
marcanthose are just aliases as far as the pixel format is concerned#asahi-dev · 2023-01-01 15:59
marcanno idea#asahi-dev · 2023-01-01 15:39
marcanit could still be somewhere other than the IOSurface struct though#asahi-dev · 2023-01-01 15:36
marcanjannau: https://gist.github.com/marcan/d93ebdffce5b36be9aaa06079a60503f tried to work out more of the DCP formats. I don't see an alpha enable nor any non-alpha 8bit RGB formats.#asahi-dev · 2023-01-01 15:33
marcanit wouldn't lose state if it didn't actively shut down power to something#asahi-dev · 2023-01-01 15:33
marcankettenis: I expect the cluster to actually power down in deep WFI (that's why it loses state)#asahi-dev · 2023-01-01 12:43
marcanI'm pretty sure deep WFI should actually power down the cluster#asahi-dev · 2023-01-01 12:43
marcanand also dynamic power on the uncore areas#asahi-dev · 2023-01-01 12:42
marcanwe are currently burning power on static leakage in the CPU clusters#asahi-dev · 2023-01-01 12:42
marcankettenis: putting the p-cluster in deep WFI should save quite a lot more than 0.1 W#asahi-dev · 2023-01-01 10:44
marcanjannau: confirmed, it's a premultiplied flag#asahi-dev · 2023-01-01 09:10
marcanI think there is a color component thing that might let us implement a color swizzle?#asahi-dev · 2023-01-01 09:10
marcanbut we need to find another way of implementing XRGB formats#asahi-dev · 2023-01-01 09:10
marcanwe should still clear surface[0] of course, since even if we make [1] opaque that still wastes power#asahi-dev · 2023-01-01 09:09
marcanonly for argb formats of course, with a=0, since opaque formats like 10-bit would implicitly have a=1#asahi-dev · 2023-01-01 09:08
marcanbut it would show through the Apple logo-shaped remnant of surface[0] as an additive blend when alpha=0 (which I think is what happens in SDDM) for a surface[1] above it#asahi-dev · 2023-01-01 09:07
marcanthat would mean that for a black background, alpha is effectively irrelevant