07:29mripard: rodrigovivi: ack
07:42pq: sima, I had never heard that userspace should explicitly program a CRTC to be used *only* with a writeback connector. It never even occurred to me, because obviously one wants to use writeback to capture the output of another, physical connector. :-o
07:44pq: In fact, I might have even suspected that not including a physical connector might make writeback fail.
07:47sima: pq, maybe I totally misremember how they all work, but I thought this is how it goes for the ones we have
07:47pq: the email replies say, that several hardware support writeback combined with real connector
07:47sima: yeah I guess then maybe I'm a bit wrong
07:47emersion: yeah, the writeback connector needs a CRTC, sometimes can be shared with a physical connector sometime snot
07:48emersion: depends on possible_crtcs
07:49pq: It's ok if it doesn't work, but physical and writeback connectors on the same CRTC seemed like the main use case to me.
07:49sima: I thought encoder->possible_clones was what limited writeback+real connector
07:49sima: emersion, ^^
07:49emersion: hm right, maybe…
07:50sima: at least haven't heard of a writeback-only crtc yet, but maybe those exist too
07:52emersion: sima, example https://drmdb.emersion.fr/snapshots/dcbe13191df8
07:52emersion: the writeback connector can only be enabled with "Encoder 3", "CRTC 1"
07:52emersion: these can't be used with a physical connector
07:54sima: hm interesting
07:54sima: but yeah vaguely remember that vc4 was one of the best effort/idle bandwidth writeback connectors
07:56mripard: sima: vc4 has a CRTC only connected to the writeback connector
07:57emersion: mripard: linked example above :)
07:57mripard: ah :)
07:57mripard: my bad
10:04lumag: mripard, do you possibly have a feedback for the 'HDMI codec' patchset? For msm/hdmi bridge the driver has the corresponding drm_connector, since it has created it. For non-root bridges there is no connector => no way to use drm_atomic_helper_connector_hdmi_update_audio_infoframe
10:05lumag: I toyed with the idea of notifying bridges when the connector is created, but it probably will not play well with the dynamic paths, etc.
10:10mripard: lumag: I did a first pass but would like to do a second one to make up my mind
10:10mripard: I'll try to do it today
10:10lumag: mripard, thank you! if it's on your radar, then there is no hurry
17:11Ristovski: zmike: What's the state on that Xnine vs DXVK bench? I assume they're a lot closer now these days?
17:11zmike: no clue, I haven't gone back to that in years
17:12Ristovski: Asking because I tried it with TF2, but tfw "Failed to create D3D9!" from Source engine side (I see Nine init from mesa just fine beforehand)
17:13Ristovski: I've been having cursed bugs these couple of days, so will absolutely NOT try to debug this
17:13Ristovski:clearly under influence of some sort of computer bad luck deity
19:54Company: today's question from the srgb-is-so-much-fun department:
19:54Company: What's the BGR version of GL_SRGB?
19:55Company: or more exactly: What do I need to put so I get the same format as the default Wayland format, but with srgb?
19:56Company: the equivalent of VK_FORMAT_B8G8R8A8_SRGB
20:08Company: it seems the answer is "can't have that"
20:09Company: at least if I'm reading the format mapping code right
20:35reutersnews: Ok so the powers 65(adder in)+63(addend out) - 64 is 64 first bit in, 65(adder in) + 64(addend in) -64 is 65 first bit (one bit) and second bit (two bit) correspondingly, now 66(adder in)+64(addend out)-65 is 65 , 66(adder in)+65(addend in) -65 is 66 ... so second (2 bit) and third (4 bit) .so 64+66 is 1 +4=5 64+65=1+2=3 65+66=2+4=6 65+65=130-64=66 that but how can one know when to 65 65
20:35reutersnews: is 130 to remove 64? whenever the addend is 128 not 127 or less not 129 and adder is 131 not less and if adder is below 127 it needs to remove 126 and if 128 it needs to remove 63. But yesterdays method was resulting to the same(maybe this is more illustrative though), so now you have this lookup table based of polderan: 128+131is accessing 64 129+131 yields 0 127+131 yields 0 128+128
20:35reutersnews: forwards 63 128+126 is 126, you agree that there is no point in doing such access? you better do 64+65-64 is 65 65+65 -64 is 66 66+65-63 is 68 66+64-63 is 67 now theres fewer states. 65+66 is 131 accessing 65+66 aka 1+2 66+67 is 2+2 would forward 67 and is absent ..... 66+68 2+4 would yield 66+67 so is absent also 63+64 is 0 , 65+64 accesses 1, 66+64 2 and 68 is absent from first
20:35reutersnews: selection bank. now 6 states per 128digit polderan lookup table for 2 first bits 1&2, 256*6*32 easily fits into 65536 max digits of 16bit controller. now bank2 for single selection from bits 4to8, 69 70 71 72 like so: now we are accounting with carry propagates 68+69 is 4+4 is 68 so 65 was 1 , 66 was 2 , 67 was 4, we have 68 is 8, 68+70 is 8+4 and is accessed, 68+71 is 4+8 is also
20:35reutersnews: stored, 68+72 is 16+4 is absent then 69 is 4, 70+61 is 8, 71+62 is 8, 72 is absent. 6 states again but in bank 6, only one is selected. And this way it repeats, in the register file 6 states per bank per every 6 bits. I am not stupid, it's very complex to get bugs off the implementation so it takes alot of time to get stable, but is possible yeah...16bit machine can run 64 bit code yeah.
20:35reutersnews: Every minor bug is catastrophically difficult to find.
20:48DemiMarie: dwfreed: banhammer?
20:49dwfreed: DemiMarie: thanks
20:49DemiMarie: dwfreed: you're welcome
21:11KarenTheDorf: Company: Unfortunately the way GL talks about SRGB is an abstraction layer too high I think.
21:11KarenTheDorf: Also, what just happpened?
21:14Company: KarenTheDorf: I don't quite understand how that works, because apparently I can turn it on/off in a sampler, so it's pretty high-level, but I can't seem to select it when picking the format - it's all weird
21:15Company: and it's a bit tricky because this format is the default in Cairo and Wayland
21:16Company: time for a new extension that adds GL_BGRS and GL_AHPLA_BGRS!
21:17KarenTheDorf: I think you'd need to do this at the EGL/GLX/etc. level, if we're talking about the main framebuffer.
21:17KarenTheDorf: For non-default buffers, IDK, I don't think there's a way to express "GL_BGR but also SRGB" in GL.
21:17Company: the main framebuffer is already in that format
21:18Company: that's how I noticed it being busted - I had no matching internal format to assign to it
21:19Company: the case where it matters is when Cairo hands me a buffer and I want to glTexImage2D() it into GLES
21:19Company: because I think I can't set GL_SRGB_ALPHA as internal format and GL_BGRA as format there
21:31KarenTheDorf: Just import it as RGBA and swizzle it somehow? *shrug*
21:33Company: probably the best choice
21:33Company: but it means I can then not easily draw to the texture
21:34Company: makes my global atlas more complicated to manage, but oh well