02:27gfxstrand[d]: I suspect that's because sixteen ends up working okay for both planes because of clamping.
04:53tiredchiku[d]: we don't do dcc yet, right?
07:20blisto[d]: One does not simply
07:41tiredchiku[d]: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9629
07:41tiredchiku[d]: closure candidate
07:42tiredchiku[d]: asdqueerfromeu[d]: pls do the honors :3
07:49tiredchiku[d]: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9644
07:49tiredchiku[d]: another
07:50tiredchiku[d]: https://gitlab.freedesktop.org/mesa/mesa/-/issues/10330
07:56tiredchiku[d]: another
07:56tiredchiku[d]: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9644
11:09mohamexiety[d]: tiredchiku[d]: nope. WIP
14:31dwlsalmeida[d]: gfxstrand[d]: I'll just push these hacks, really, once the time to merge this comes around we can reevaluate what to do
14:31dwlsalmeida[d]: we gotta get something to the igalia people to test on, they're the ones working on the gstreamer part
14:48gfxstrand[d]: tiredchiku[d]: Not quite. There's still one feature bit missing.
14:50gfxstrand[d]: dwlsalmeida[d]: Yeah, carrying hacks in the dev branch is fine. We just need to figure out the multi-plane stuff properly before it all lands.
14:51tiredchiku[d]: gfxstrand[d]: huh, what's missing?
14:54asdqueerfromeu[d]: tiredchiku[d]: I'll go out of online mode for some days so I'll try to do that later today (or tomorrow)
14:58esdrastarsis[d]: tiredchiku[d]: Shared atomics, I think
15:04redsheep[d]: tiredchiku[d]: https://discord.com/channels/1033216351990456371/1034184951790305330/1290784600595107891
15:07tiredchiku[d]: I'm confused
15:07tiredchiku[d]: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/nouveau/vulkan/nvk_physical_device.c?ref_type=heads#L169
15:07tiredchiku[d]: is VK_KHR_shader_atomic_int64 not the same as this?
15:08tiredchiku[d]: ..ah nvm, I see it
15:08tiredchiku[d]: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/nouveau/vulkan/nvk_physical_device.c?ref_type=heads#L378
15:08tiredchiku[d]: confusing!
15:10gfxstrand[d]: Yeah, we're missing 64bit atomics on shared memory.
15:10gfxstrand[d]: Which is a separate feature bit in the extension.
15:12tiredchiku[d]: interesting
15:12tiredchiku[d]: would that still be a decent thing to work on for someone who's new to the compiler? :D
16:11dwlsalmeida[d]: my Rust stuff takes forever to run
16:11dwlsalmeida[d]: ok, I am 100% puzzled now
16:12dwlsalmeida[d]: like 7s vs 85s for the entire video test suite
16:13dwlsalmeida[d]: (vs the C version)
16:15mohamexiety[d]: a bit silly but try release rather than debug. Rust debug can be quite slow
17:08gfxstrand[d]: tiredchiku[d]: Yeah, it's a decent beginner compiler task. It's not trivial but it's not particularly hard as compiler tasks go.
17:09gfxstrand[d]: mohamexiety[d]: Very true. I'm a bit surprised we'd be hitting CPU overhead on command streamer stuff but I can believe it, I suppose.
17:15gfxstrand[d]: I think it's more likely that the rust code is splitting pushbufs aggressively or something.
17:32rinlovesyou[d]: mohamexiety[d]: it is not silly at all lol
17:32rinlovesyou[d]: made a simple frame parser and it went from tens of milliseconds to nanoseconds depending on the video
17:51mohamexiety[d]: yeah I am just not entirely sure this is the cause for it being this much slower in this case
19:04gfxstrand[d]: We may also want to fiddle with the pushbuf stuff so it constant folds/inlines better.
21:03dwlsalmeida[d]: Ok guys, rebased on top of the latest master: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31867
21:03dwlsalmeida[d]: includes instructions on how to test
21:03dwlsalmeida[d]: 123/135, better than the 121/135 that ffmpeg gets using VA-API \o/
21:04dwlsalmeida[d]: avhe[d]: you received credit for your tracer work
21:04dwlsalmeida[d]: thank you, that was absolutely needed to get this going
21:09dwlsalmeida[d]: airlied[d]: this needs to be upstreamed too fyi:
21:09dwlsalmeida[d]: From d703f503fbd5934bbaf1b520c2973db6722cb069 Mon Sep 17 00:00:00 2001
21:09dwlsalmeida[d]: From: Dave Airlie <airlied@gmail.com>
21:09dwlsalmeida[d]: Date: Fri, 19 Jan 2024 08:49:38 +1000
21:09dwlsalmeida[d]: Subject: [PATCH] nouveau: enable nvdec engines magic
21:09dwlsalmeida[d]: ---
21:09dwlsalmeida[d]: drivers/gpu/drm/nouveau/nouveau_abi16.c | 1 +
21:09dwlsalmeida[d]: 1 file changed, 1 insertion(+)
21:09dwlsalmeida[d]: diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c
21:09dwlsalmeida[d]: index 2edd7bb13fae..14c02fcc507a 100644
21:09dwlsalmeida[d]: --- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
21:09dwlsalmeida[d]: +++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
21:09dwlsalmeida[d]: @@ -313,6 +313,7 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
21:09dwlsalmeida[d]: case 0x04: engine = NV_DEVICE_HOST_RUNLIST_ENGINES_MSPPP ; break;
21:09dwlsalmeida[d]: case 0x08: engine = NV_DEVICE_HOST_RUNLIST_ENGINES_MSVLD ; break;
21:09dwlsalmeida[d]: case 0x30: engine = NV_DEVICE_HOST_RUNLIST_ENGINES_CE ; break;
21:09dwlsalmeida[d]: + case 0x300: engine = NV_DEVICE_HOST_RUNLIST_ENGINES_NVDEC; break;
21:09dwlsalmeida[d]: default:
21:09dwlsalmeida[d]: return nouveau_abi16_put(abi16, -ENOSYS);
21:09dwlsalmeida[d]: }
21:09dwlsalmeida[d]: --
21:09dwlsalmeida[d]: 2.43.0
21:15redsheep[d]: I haven't been following too closely, have things progressed with the zink side well enough that this can start to make vaapi work too?
21:18dwlsalmeida[d]: I think Dave has been pushing this
21:24avhe[d]: dwlsalmeida[d]: thanks and congrats on getting this working!
21:28redsheep[d]: dwlsalmeida[d]: Looks like that other MR is still a draft. Still, exciting progress.
21:28redsheep[d]: Having video working will definitely help move nvk closer to being a good daily driver... er, driver