00:00 Company: it should get about 7-10fps, at least that's what it gets on Linux
00:01 Company: but it's getting the video in a dmabuf/d3d12resource, so it might (not) do some tricks with either
01:17 DemiMarie: Company: is lavapipe menat to actually be used?
01:38 Company: DemiMarie: depends on your definition of "be used"
01:38 DemiMarie: Company: what is yours?
01:40 Company: DemiMarie: I use it for 3 reasons that are all somewhat valid IMO: (1) fallback - if things go wrong (driver broken for example), there's always lavapipe (2) comparison - if my code doesn't work, I can check with lavapipe, and if it's not broken there it might be a driver issue and (3) CI - lavapipe is portable, predictable (ie I can rely on its bit-exactness) and doesn't require a GPU
01:43 Company: oh, also (4) I can compare work with other people - "can you try on lavapipe" is a good way to get the GPU out of the equation when I'm debugging Debian vs Fedora issues for example
01:44 Company: TL;DR might be: nobody should ever use lavapipe, but everybody should always be able to use lavapipe
01:44 DemiMarie:wonders how long it will be until Qubes OS has to deprecate the GTK templates
01:47 Company: there's also another reason: lavapipe supports almost everything
01:47 Company: so if I want to check something that my GPU can't do, I can test it with lavapipe
01:48 Company: (apart from multiplane dmabuf, booo!)
01:52 Company: DemiMarie: I don't expect GTK to get much slower anywhere - we're fully GPU by now
01:52 Company: the things that might happen is more dynamic apps that take advantage of the GPU more
01:53 Company: but that doesn't influence apps like text editors or settings apps
01:54 DemiMarie: what about font rendering?
02:06 Company: DemiMarie: font rendering might change, but I think it isn't that interesting because text is usually static
02:07 DemiMarie: Company: does that mean that the results will be cached and the pipeline won’t be rerun over and over and over?
02:09 Company: I would expect so
02:10 Company: it depends a bit on the quality gains of exact placement of glyphs vs quantizing to cacheable positions
02:13 Company: and of course, with fractional scaling, glyphs are all over the place anyway these days
11:17 pq: zmike, ooh, thanks! I'll try to get to testing that.
12:19 alyssa: zmike: jenatali Company: smh not shipping radeonsi on windows (:
12:35 pq: what is "smh"?
12:36 zmike: https://www.urbandictionary.com/define.php?term=smh
12:37 pq: indeed
14:45 zmike: can I get an ack on https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34654
15:01 zmike: ty
15:15 mareko: can somebody please review nir_def_bits_used changes: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34489 ?
15:17 jenatali: Company: Your descriptions are the exact reasons we have WARP for D3D (and why I'm advocating for it so much these days)
15:26 glehmann: mareko: I will take a look
15:36 Company: jenatali: it's kind of a thankless job I bet because nobody really needs it
15:37 jenatali: I've been surprised how much feedback I've heard about it
15:37 Company: jenatali: do you know if games studios use Warp for testing or do they use real GPUs?
15:37 jenatali: Some do use it
15:37 jenatali: Mostly for CI-style testing
15:38 Company: yeah, I'd expect people to use it for ensuring the baseline is solid
15:40 jenatali: But it's also what all Windows VMs use for rendering, so it's pretty important
15:42 Company: I was wondering - is rendering of core Windows apps done via DirectX these days or is this still primarily software like Skia/Cairo is?
15:42 Company: File Explorer or Settings or Notepad I mean
15:43 jenatali: It's either GDI or it's DirectX/DComp, depending on which app it is
15:45 Company: because it is definitely quite a bit slower to go through a GPU abstraction layer when it's all software rendering
15:46 jenatali: I'd say it depends
15:48 Company: I guess a layer like DComp can help where you can implement things like texture/glyph blits directly instead of having to set up a shader (that said, I have no idea if DComp uses DirectX internally or has straight driver calls)
16:56 bluetail: so I got DP MST on amdgpu to work. But I could only get 4 displays work on my Navi 33 GPU at a time.
16:57 bluetail: It was on a w7500. Is this expected to be this way? I thought I could use 6 displays on a 4 port gpu
17:11 dwfreed: it only has 4 CRTCs
17:13 dwfreed: You can only have as many unique displays as you have CRTCs
18:32 Fancy2209: Hello, is the code for the Cell B/E Gallium Driver hosted somewhere still?
18:33 Fancy2209: i hope the message didn't go twice, I had forgotten to register my nick
18:34 dwfreed: it didn't get repeated
18:35 Fancy2209: cool thanks
18:37 Fancy2209: > I recently got Linux 6.12 up and running on my PS3 and wanted to mess around with the driver, that's why I am asking about this driver
18:37 Fancy2209: s/this driver/it
18:39 DodoGTA: Is there a real reason for the nvidia-drm GBM backend? I don't see it loaded when running KWin (or eglgears with PRIME render offload) on my hybrid GPU laptop
18:41 DodoGTA: Fancy2209: I see an old Phoronix post about that driver being dropped back in 2011 (and I doubt anyone has updated it for any semi-modern Mesa versions)
18:41 Fancy2209: I see, so I'd have to either try to update it or try to get an an 2011 version of mesa to not blow up on a modern distro?