09:57 dmitryK: Hi, all! Could you please help me with clarifying one thing, when the user uses proprietary drivers, like NVidia, do they provide their own EGL/GLX libraries/implementations, or they are somehow forwarded via Mesa/DRI3?
10:01 pq: dmitryK, the is a dispatcher layer called glvnd, it's its own project. The implementations of EGL, GLX, GL are then either Mesa or proprietary.
10:02 pq: Before glvnd was a thing, people indeed needed to replace all those application-facing libraries.
10:03 dmitryK: pq: is it like levelzero API in Intel proucts?
10:03 pq: I've no idea what that is.
10:03 dmitryK: I remember they have some form or "loaders"
10:04 pq: see https://gitlab.freedesktop.org/glvnd/libglvnd
10:04 dmitryK:is reading...
10:09 dmitryK: pq: do I understand it right that from the client side this glvnd comes in a form of libGLX.so, libGL.so and libEGL.so, which basically dispatch all the requests to the implementation registered by the driver and selected by the user based on EGLDisplay?
10:12 dmitryK: it seems like "yes": https://www.x.org/wiki/Events/XDC2017/brenneman_GLVND.pdf
10:19 dolphin: dmitryK: the dispatcher/loader library to forward to different implementations is rather generic mechanism
10:20 dmitryK: is the same thing happens when requesting ELG from Wayland versus GLX? Or it is a different beast?
10:22 MrCooper: that's different front-end libraries (libEGL vs. libGLX / libGL)
10:23 dmitryK: MrCooper: okay, thank you!
10:24 dmitryK: MrCooper: btw, that swap-interval resetting issue for implicit GLX surfaces cannot be considered as a bug in Mesa, can it?
10:24 MrCooper: it is
10:25 dmitryK: MrCooper: hm... should I report that then? Isn't "implicit surface" thing a "gray zone"?
10:27 MrCooper: hmm, maybe it is indeed
10:28 MrCooper: I wouldn't hold my breath for it getting fixed anyway
10:43 dmitryK: Hm.. the specification is rather vague on that:
10:43 dmitryK: "For backwards compatibility with GLX versions 1.2 and earlier, a rendering context can also be used to render into a Window. Thus, a GLXDrawable is the union {GLXWindow, GLXPixmap, GLXPbuffer, Window}"
10:45 dmitryK: so, theoretically, it could be read that Mesa is required to render to Window only for 1.2 apps, but swap-buffers is an extension for 1.3 :)
10:47 dmitryK: MrCooper: is it possible to correlate the version of Mesa with the usage of DRI3 (that presumably broke the thing?)?
11:07 MrCooper: don't think so; one indicator is GLX_EXT_buffer_age, which couldn't be supported before DRI3
12:07 dmitryK: MrCooper: and this problem shouldn't appear on systems before DRI3, right?
13:01 MrCooper: right
15:04 dmitryK: MrCooper: thanks a lot for your help! :)
15:05 MrCooper: no worries
18:42 robclark: karolherbst: what's the deal w/ cts test_spir? AFAIU this extension is deprecated? But it reports as failing..
18:42 robclark: https://www.irccloud.com/pastebin/g2HFR6NW/
22:28 karolherbst: robclark: yeah.. it shouldn't fail, but you also don't have to run it for a submission, soo.. it doesn't really matter
22:28 karolherbst: but yeah, should probably be fixed
22:52 robclark: karolherbst: hmm, but it seems to be listed in opencl_conformance_tests_full.csv? I assume that a the csv that you are supposed to run for conformance submissions?
22:53 karolherbst: well it's more of a suggestion. The conformance process doc doesn't mention anything in that regards specifically or what csvs to use, just mentions specific tests you need to run
22:53 karolherbst: but it could also be that it regressed
22:54 karolherbst: getting the same error code on intels driver as well
22:56 karolherbst: ohh looks like it does mention "opencl_conformance_tests_full.csv" specifically.. oops
22:56 robclark: hmm
22:57 karolherbst: but
22:57 karolherbst: it doesn't say you have to use the runner
22:57 robclark: hmm, ok
22:57 karolherbst: mhh I thought it was listing the sub-tests..
23:00 karolherbst: anyway, it's worth reading through the doc, because the csv file itself isn't enough to meet the passing criteria anyway
23:00 karolherbst: like because rusticl supports SPIR-V, each of those tests also have to be run in spir-v mode
23:00 karolherbst: if the compiler path is mostly identical to the CLC path, you can run either (online or spirv) as wimpy
23:01 karolherbst: if the driver lowers vectors to scalar, you can also run all those math tests as wimpy for non scalars
23:03 robclark: ok, got a few more things remaining to fix (but getting close)... but I'll read the instructions
23:03 karolherbst: I had some CSV file somewhere...
23:06 karolherbst: guess I accidentally deleted those...