#asahi-re · 2021-08-30 03:11
alyssa`but#asahi-re · 2021-08-30 03:11
marcanprobably because the bit or something? this is calling out to some other RTBandwidth service#asahi-re · 2021-08-30 03:11
alyssawhy can't the dcp use sr_map_device_memory for that#asahi-re · 2021-08-30 03:11
alyssathough err#asahi-re · 2021-08-30 03:10
marcanI expect the bit will change for dispext, need to see where that comes from#asahi-re · 2021-08-30 03:10
alyssarock on !#asahi-re · 2021-08-30 03:10
marcanit's just telling DCP about some registers#asahi-re · 2021-08-30 03:10
marcanturns out it was 90% uninitialized memory bullshit#asahi-re · 2021-08-30 03:10
marcanalyssa: https://github.com/AsahiLinux/m1n1/commit/219950235c5f95be8e7c2f69c0039317fdeaf895 enjoy#asahi-re · 2021-08-30 03:04
alyssaFun#asahi-re · 2021-08-30 03:03
marcanlooks like it's just telling it about some IRQ or something#asahi-re · 2021-08-30 03:03
marcanthis is good, I thought this was going to be some mess involving Math™#asahi-re · 2021-08-30 03:00
marcanI think one is actually reg + bit#asahi-re · 2021-08-30 02:59
marcanbut there are three physmem pointers returned, to registers#asahi-re · 2021-08-30 02:59
marcanthen I think most of the structure is actually just uninitialized garbage#asahi-re · 2021-08-30 02:59
marcanalyssa: so actually what that rt bandwidth thing does is call function-bw_req_interrupt0 from the ADT (which is a link to elsewhere)#asahi-re · 2021-08-29 16:07
alyssaOh, like that. Expose a primary, a cursor, and an overlay and reject request sto enable all 3 at once#asahi-re · 2021-08-29 16:07
alyssa"Checking that virtualized resources exported to userspace are not oversubscribed. For various reasons it can make sense to expose more planes, crtcs or encoders than which are physically there"#asahi-re · 2021-08-29 16:04
alyssa(in particular to support broken compositors that expect a cursor plane but ignore overlay planes)#asahi-re · 2021-08-29 16:04
alyssaI'm more wondering how to model this for linux#asahi-re · 2021-08-29 16:02
alyssa🤷#asahi-re · 2021-08-29 16:00
marcanmaybe it's just Not Done Yet#asahi-re · 2021-08-29 16:00
marcanthat's so confusing, if they increased the ABI count one would presume they expect to release hardware with more slots, but then they don't even support the existing 3?#asahi-re · 2021-08-29 16:00
marcanyeah but the DCP ABI used to have 3 slots, now it has 4... yet it limits it to 2... but the hardware apparently supports 3?#asahi-re · 2021-08-29 15:59
alyssaAlthough I guess it's not super surprising given this was designed for iPhones where you only ever have a primary and an overlay#asahi-re · 2021-08-29 15:59
alyssamarcan: baffling#asahi-re · 2021-08-29 15:54
marcanthat is so bizarre#asahi-re · 2021-08-29 15:54
marcanit really is just max 2 surfaces at one time#asahi-re · 2021-08-29 15:54
marcanalyssa: looks like that limitation isn't for blending though#asahi-re · 2021-08-29 15:50
marcanthere was one with a non-text code in the past, but I don't recall seeing any of those in DCP#asahi-re · 2021-08-29 15:49
marcanit doesn't look like there's XBGR#asahi-re · 2021-08-29 15:49
marcanalyssa: and yeah, looking at the format codes it sounds like there should be some flag to disable alpha blending?#asahi-re · 2021-08-29 13:34
alyssa(or some opaque flag enabled)#asahi-re · 2021-08-29 13:34
alyssaunless maybe it's still ARGB but with some alpha flpag disabled somewhere#asahi-re · 2021-08-29 13:33
alyssa(*is* there a fourcc for it?)#asahi-re · 2021-08-29 13:32
alyssado we know the fourcc for XBGR, though?#asahi-re · 2021-08-29 13:28
alyssaalpha is possibly, yeah#asahi-re · 2021-08-29 11:20
marcanshould be easy to work out with monitor rotation on macos#asahi-re · 2021-08-29 11:20
marcanre flips, there have to be flags for that somewhere...#asahi-re · 2021-08-29 11:19
marcanif you're using ARGB for everything that could explain that#asahi-re · 2021-08-29 11:19
marcanthat also makes sense, since you'd have two layers with alpha on top of one without#asahi-re · 2021-08-29 11:19
marcanalyssa: maybe "blend" means specificially formats with alpha#asahi-re · 2021-08-29 04:05
alyssa[Also, the use of %d instead of %u on scaling syslog implies that dcp rect is actually s32's, but the DCP rejects negative values. Which is a shame, it would've been a lovely way to indicate X/Y flips. Which I know the hardware can do since I know the actual register for it...]#asahi-re · 2021-08-29 04:03
alyssanini#asahi-re · 2021-08-29 04:03
alyssano dice there either. 🤷#asahi-re · 2021-08-29 03:58
alyssaor it could be that it's computed on the fly, maybe based on what's in use or maybe resoution-- let me try at 1080p#asahi-re · 2021-08-29 03:58
alyssait could be the firmware wants to be generic to different devices (...but it hardcodes H13P in the name lol)#asahi-re · 2021-08-29 03:57
alyssathe relevant string int he firmware has a %d format specifier, it's not hardcoded to 2#asahi-re · 2021-08-26 06:50
svenyup#asahi-re · 2021-08-26 06:48
marcanwell engineered, this