19:17 Microlinux: Anyone here?
19:17 Microlinux: Ahh yeah
19:17 Microlinux: I found the right one
19:18 Microlinux: List time I used this on real was on the late 90s with mIRC
19:19 Microlinux: Making apitraces for nine is easier to me than making them for linux
19:20 Microlinux: On linux they are always broken, something happened, a PITA.
19:25 DavidHeidelberg[m]: Sure :)
19:41 alyssa: yo i heard this is where all the cool kids are hanging out now
19:43 Microlinux: Hahah
19:44 Microlinux: Watching oblivion running on s905y2 was rad, with wine3d I wouldn't pass 2 fps, not with little cores
19:44 alyssa: wine3d being the gl backend?
19:45 Microlinux: Yep
19:45 alyssa: I'm.. surprised there'd be such a difference
19:45 Microlinux: Super intense on cpu also
19:45 alyssa: but, shrug
19:45 Microlinux: While emulating, it seems like night and day
19:45 Microlinux: Well, i use wine 8.0
19:46 Microlinux: So, latest wine3d improved a bit
19:46 Microlinux: But not that much
19:46 alyssa:figures if DXVK isn't going to be practical on Midgard/Bifrost then having nine work is probably a good idea
19:46 Microlinux: Yeah, also bc even dxvk will be, at least on paper, more cpu demanding
19:46 Microlinux: On A311D may not matter, but on little cores. You are done
19:46 alyssa: (and it isn't practical... there are fundamental obstacles to efficient Vulkan drivers for Bifrost and older)
19:47 Microlinux: Yep, I assumed that
19:47 Microlinux: Hopefully valhall is better
19:47 alyssa: yeah, looks like it
19:48 alyssa: there are a bunch of poor architectural decisions in Midgard/Bifrost, they're not a huge deal for GLES but they're showstoppers for VK
19:48 Microlinux: Also, a lot of the textures compression supported only on the t860 are on valhall too, so, good.
19:48 alyssa: it's my understanding the DDK has some really awful hacks to pass conformance on VK
19:48 alyssa: but I am not convinced the result is something anybody would actually want to use after all that effort
19:48 alyssa: so, Valhall.
19:49 Microlinux: Sure, at the end, opengl will not die entirely on the near future. It will, but it will survive a bunch of years.
19:50 alyssa: Valhall (finally) fixes a lot of the fundamental mistakes such that "proper" VK should be doable (i.e. VK won't be slower than GLES)
19:50 Microlinux: Good
19:52 Microlinux: Even then, RK3588 is awful
19:52 Microlinux: 3.14 kernel, rebranded 5.10
19:52 alyssa: Yeah...
19:53 Microlinux: It kinda works, but every hw you plug may work or not... and things that work on linux do not on that kernel
19:57 alyssa: ok, humour me, how do I run these nine tests
19:59 alyssa: funny piglit branch ok
20:04 alyssa: oh, the ninetests are literally imported into piglit, lol that's kinda cute. got it
20:06 q4a: I hope, that it's same tests: https://github.com/iXit/nine-tests
20:07 alyssa: ok, I have everything built I think
20:08 alyssa: uh no needed to enable in piglit
20:09 alyssa: packages d3d/adapter9 required by virtual:world not found
20:09 alyssa: I don't want to do a systemwide install...
20:10 alyssa: oh I just need to somehow get it to use the prefix. um
20:12 Microlinux: You need to install libd3dadapter
20:13 Microlinux: I also had to set my python version, that's a piglit problem, not related
20:13 q4a: or build from source and then use meson devenv =)
20:14 Microlinux: Oh yeah
20:14 Microlinux: But I use oibaf, so that's close to main
20:14 Microlinux: Latest*
20:15 Microlinux: Piglit took like 10 minutes to compile on my zero haha super slow
20:15 Microlinux: It took me back to my rpi3 era
20:20 Microlinux: I will need to learn how to crosscompile to armhf with meson down the road
20:20 alyssa: DavidHeidelberg[m]: libxcb-dri3-dev is required by ninetests but is not required by cmakelists, please fix that
20:20 alyssa: and libxcb-present-dev
20:21 alyssa: although even there getting linking errors for undefined reference to xcb_present_id etc
20:21 alyssa: uh
20:21 Microlinux: Oh, I had all those installed
20:22 Microlinux: And it worked
20:22 Microlinux: But debian rolling release here
20:22 alyssa:writes the easy cmake patch
20:24 alyssa: ok, nine tests are built
20:24 alyssa: unknown result for "sanity" is not a good sign at all
20:26 DavidHeidelberg[m]: alyssa: I'll add, weird, "worked for me" :D
20:27 alyssa: DavidHeidelberg[m]: I mean the cmakelists are wrong regardles
20:29 Microlinux: Also python exe was needed to be set on my end to pick python3-numpy dep. But that's probably fixed on main piglit
20:29 alyssa: DavidHeidelberg[m]: Why do no tests return results?
20:29 alyssa: That's not going to work
20:29 alyssa: On a correct driver, that means whether a test passes or fails is undefined behaviour
20:29 alyssa: Possibly it works by chance on x86 systems
20:30 alyssa: It's not enough to sed the return types from void, you actually need to return PASS at the end of things
20:35 alyssa: but yeah, the tests as integrated don't work
20:37 DavidHeidelberg[m]: I havent tried on arm, but on x86 it stops on first failure
20:37 DavidHeidelberg[m]: Sadly design of wine tests isn't perfectly compat with piglit
20:38 alyssa: stopping on first failure doesn't help when you're returning undefined results
20:38 alyssa: compiling this file gives zillions of warnings. they're completely correct warnings.
20:38 alyssa: the code as-is does not work
20:44 alyssa: I've built the upstream ninetests. Not fully passing but much better shape
20:54 alyssa: ----
20:55 alyssa: Looking at upstream ninetests results, there are a bunch of bugs failing that are IMHO nine bugs but you'll say are driver bugs
20:55 alyssa: alpha testing, for example
20:55 alyssa: Mali does not support alpha testing in hardware from Bifrost onwards, so Panfrost does not set PIPE_CAP_ALPHA_TEST
20:55 alyssa: Accordingly, it's invalid for frontends to set alpha_enable (and expect it to work)
20:56 alyssa: In OpenGL, if an application calls glEnable(GL_ALPHA_TEST) on Panfrost, mesa/st will create a shader variant with nir_lower_alpha_test
20:56 alyssa: Nine has no such variants, which means the alpha test portion of ninetests fails on Bifrost+
20:58 alyssa: I haven't looked through the fails in enough detail but I expect a big chunk fall into that category (nine using Gallium features without checking whether the driver actually supports the feature)