01:51 alyssa: jenatali: I'm not sure lower_tex_shadow is correct
01:52 alyssa: shadow comparison is supposed to happen before filtering, not after
01:53 jenatali: alyssa: I'd believe that
01:53 alyssa: although the GL spec is seriously loose about this so you're probably off the hook, cheers
01:53 alyssa: :p
01:54 alyssa: lot of "should"s.
01:54 jenatali: Yeah
01:55 alyssa: and I guess dozen doesn't use it so.. yeah ok
01:57 jenatali: alyssa: Yeah dzn couldn't, samplers can be bindless so there's no reasonable way to apply the lowering
01:58 jenatali: We're about to ship new DXIL support that will let the GL driver not use it
01:58 alyssa: Nice!
09:49 passimoto: https://openmath.org/cd/fns2 https://openmath.org/cd/interval_types https://raw.githubusercontent.com/numerateweb/openmath-cd/master/openmath.ttl
09:49 passimoto: <http://www.openmath.org/cd/list1#map> math-meta:description """This symbol represents a mapping function which may be used to
09:49 passimoto: construct lists; it takes as arguments a function from X to Y and a
09:49 passimoto: list over X in that order. The value that is returned is a list of
09:49 passimoto: values in Y. The argument list may be a set or an integer_interval.""" ;
09:49 passimoto: math-meta:example _:n421
09:49 passimoto: that function seems to read out as very interesting for the earlier offered compressor
09:50 passimoto: I've been looking into rdf and w3 docs in this weekend, gotta say its very large set of languages .
10:03 passimoto: in other words you can add a function to all of the list, and can select the interval of interest, it's very succinct but very expressive language but it is beneficial to get into some IR from that.
18:03 ity: Hi, I am a bit confused about the architecture of gallium, is there docs somewhere I can read about what does what? Just like, a high level overview of what's what in src/gallium/
18:07 airlied: ity: https://docs.mesa3d.org/gallium/index.html
18:08 airlied: has some info on what is in there
19:22 FLHerne: airlied/ity: the first section of that being a long section on TGSI doesn't bode well for up-to-dateness
19:26 ccr: :) "TGSI is the only intermediate representation used by all drivers."
19:28 ity: That seems to mostly be for some specific parts...? Unless I am misreading... It doesn't seem to describe eg what frontends or targets are
19:48 Company: gallium feels like one of those abstractions that drifts rather quickly in its semantics - because it's an abstraction inside Mesa for common things in drivers - and the moment you add a new driver or even just a new generation of hardware to an existing driver, things that were common suddenly aren't anymore or vice versa
19:49 alyssa: this is a feature
19:49 Company: yeah
19:49 Company: I'm pointing that out because it's hard to document
19:49 alyssa: ah yes well
19:49 Company: and documentation quickly ends up out of tdate
19:49 alyssa: "just copy iris"
19:50 Company: like, a blog post from 2022 might describe a gallium feature that doesn't exist anymore
19:51 Company:far from an expert, he just tried to understand the YUV handling/GLES external texture stuff once
20:06 Ristovski: Huh, why are {MC,SDMA,VCN,JPEG,IH} clock gatings disabled on gfx90c/renoir when they are clearly enabled for GC 9.3.0 (which is present)? Peculiar
20:08 Ristovski: Unless of course they are enabled but the mechanism that enables them isn't even present :) (ala `soc15_common_set_powergating_state` being "todo")
20:48 airlied: ity: yeah stuff has move around since then, but what's in src/gallium vs what is src/gallium/frontends are different questions
20:48 airlied: you might need to figure out a bit more specific questions to ask
20:49 airlied: I doubt anyone understands what is src/gallium in its entirety and nobody should need to know that to work on t
20:49 airlied: on it
20:50 ity: I am not really sure what I am looking at there effectively. Like, for example, what about OpenGL drivers? What's the path there? There's also a software rasterizer, and also Zink, right...? I am really confused as to like, what's the structure, where to look for what. I don't have specific questions rn, I am just trying to orient myself
20:52 airlied: ity: opengl is under src/mesa
20:53 airlied: ity: otherwise gallium has frontends and drivers and the include files bind them together
20:53 airlied: and targets are the final linkage off the pieces
20:54 airlied: winsys is part of drivers that connect to the kernel drivers and is meant to be an abstraction (but not every driver does it)
22:20 jenatali: ity: Right, GL is split into two pieces. The main bits are not under src/gallium since until recently, they weren't strictly a gallium frontend. The other GL bits are in the frontends folder and are for GLX/EGL/WGL
22:21 jenatali: But yeah specific question about "what is..." or "where is..." are easier. Gallium is really just an interface which looks a lot like D3D10 (or at least used to) which frontends translate to, and drivers translate from
22:41 mareko: do you remember when we called frontends "state trackers"
22:47 Ristovski: heh
23:26 alyssa: mareko: those were the days (-:
23:27 alyssa: ity: https://docs.mesa3d.org/sourcetree.html
23:30 Lynne: airlied: ping on current queue issues/av1
23:33 airlied: okay I think I can play with av1, remind of the queue one
23:39 ity: alyssa: O, that is what I was looking for, thx!
23:40 ity: Does any frontend work for any driver? Eg, can lavapipe work on top of i915?
23:40 airlied: no
23:40 airlied: that's what targets tie together
23:41 ity: OOOH
23:41 ity: Where are the main pieces for GL?
23:42 airlied: src/mesa/
23:44 ity: Wait, the GL drivers are under src/mesa/? And the vendor-specific dirs are for Vulkan? Then there's extra stuff in Gallium for EGL/GLX?
23:45 airlied: no the GL frontend is in there
23:45 airlied: OpenGL has a quite large non-driver surface
23:45 airlied: the vendor specific dirs have vulkan drivers but also compilers and other stuff
23:46 airlied: EGL/GLX/GBM is a mess that infests many things
23:46 airlied: if you want to work on them I recommend specific dives into those topics
23:46 airlied: they aren't really something you want to learn about unless you have a reason to