00:35 <alyssa----> The code generated for framebuffer fetch is weiiiird

00:37 <HdkR> :D

00:42 <alyssa----> And by weird I mean long and impossible to follow :P

00:42 <alyssa----> Another day

00:43 <alyssa----> (It reminds me of the code generated for pixel local storage, which should be no surprise)

02:15 <alyssa----> Cubemaps why don't you want to work

02:25 <alyssa----> Hey, sweet!

02:26 <alyssa----> We're not faulting anymore!

02:26 <alyssa----> That's an improvement

02:27 <alyssa----> ...and we're faulting again, dangit

02:33 <alyssa----> Okay, faulting fixed, but we're still broken

02:46 <alyssa----> Okay, cubemaps are getting a lot closer but still failing tests

04:24 <alyssa----> Cubemaps work!

04:26 <alyssa----> Well, linearly

04:34 <tlwoerner> X.Org has been accepted into GSoC 2019!

04:35 <tlwoerner> if anyone has any ideas for, is interested in mentoring for, or knows of any students who might want to participate in this year's GSoC, please have a look at https://lists.freedesktop.org/archives/wayland-devel/2019-March/040351.html

04:35 <tlwoerner> students have until April 9, 1400EDT to submit their proposals

04:41 <HdkR> Woo cubes!

04:42 <alyssa----> tlwoerner: Hmm.. any students want to implement pixel_local_storage? =p

04:42 <tlwoerner> alyssa----: are you volunteering to mentor? ;-)

04:43 <tlwoerner> alyssa----: i'd be happy to add it to the list of project ideas

04:44 <alyssa----> tlwoerner: I mean... =P

04:45 <alyssa----> tlwoerner: Interpret as you will.

09:50 <bbrezillon> Looks like we need to support PIPE_TEXTURE_RECT support to get gallium-hud on panfrost

09:52 <bbrezillon> I looked at the doc and it seems TEXTURE_RECT and TEXTURE_2D are similar except for the normalized vs unnormalized coordinates

09:53 <HdkR> Correct

09:53 <HdkR> Since NPOT isn't an issue anymore in this hardware anyway

09:53 <bbrezillon> so I just modify the assert() in the driver to let _RECT texture go through

09:53 <bbrezillon> but I now have to support the texop_txs instruction

09:54 <HdkR> right

09:54 <bbrezillon> any idea how this should be done?

09:54 <HdkR> Depends on if the hardware supports an instruction for RECT texture coordinates or not

09:55 <HdkR> Worst case you have to consume some uniform slots to put width/height in the shader and do it directly in the shader

09:56 <HdkR> Although there is a textureSize function in GLSL. Maybe the hardware has an instruction for pulling the bound texture data in a special way

09:56 <bbrezillon> ok, I guess I should try with ARM's driver and see how the generated instructions

09:56 <bbrezillon> s/how//

09:57 <HdkR> Changing gallium_hud to not use texture rectangles could be more sane as well :P

10:16 <bbrezillon> I tried using TEXTURE_2D instead of TEXTURE_RECT but I still have a texop_txd instruction

10:16 <bbrezillon> so it probably has to do with the gallium-hud shaders

10:24 <HdkR> yes

10:24 <HdkR> You need to change both the shader side and the API facing side

11:58 <austriancoder> bbrezillon: you could handle txs like seen in nir_to_vir.c - https://cgit.freedesktop.org/mesa/mesa/tree/src/broadcom/compiler/nir_to_vir.c#n433

12:05 <HdkR> Why does gallium hud use rectangle textures anyway?

12:07 <tomeu> guess it's something we want to support regardless of that?

12:08 <HdkR> It's part of core GL, just not ES

12:10 <tomeu> well, but there's lots of big GL sw that people run these days on linux desktops

12:11 <HdkR> Sure. I'm just curious why anyone would choose to use them for gallium hud :P

12:12 <tomeu> ah, gotcha

15:06 <bbrezillon> austriancoder: so, if I omit the various optims done in V3D (which we'll have to do at some point), the idea is to add an implicit uniform that contains the texture size info and use use a store op to retrieve it

15:07 <bbrezillon> I'm not entirely sure how it can be done in panfrost though

15:08 <bbrezillon> (by implicit uniform I mean a uniform that is not explicitly declared in the shader)

15:13 <austriancoder> bbrezillon: yeah thats the idea...all you need to do at draw time is to keep the immediate/compiler used uniforms up-to-date - etna does the same for RECT

15:17 <alyssa----> bbrezillon: We already have a little bit of this for viewports

15:18 <alyssa----> I'm planning to overhaul this at some point soon so we can actually implement this stuff

15:21 <bbrezillon> alyssa----: great!

15:21 <alyssa----> bbrezillon: I don't know if this is a blocker for you..?

15:29 <bbrezillon> alyssa----: nothing really urgent, but I'd love to have HW counters exposed through gallium-hud :)

15:30 <bbrezillon> there's also the solution proposed by HdkR (use TEXTURE_2D) which I'm investigating

15:30 <alyssa----> bbrezillon: Nah, might as well get it right the first time

15:30 <alyssa----> Since we *will* need to support txs for ES3

15:35 <bbrezillon> yep, that's why I went on supporting txs first, but I'm completely new to all of this :)

15:57 <alyssa----> bbrezillon: MM, yeah. I don't know what parts of the stack you're comfortable/interested in, so I'm not sure how to help best

15:57 <alyssa----> LeAdErShIp (TM)

16:01 <bbrezillon> the kernel side of things is okay-ish, but I'm interested in learning more on the mesa side of things (absolutely not up to the task right now, so it's more a long term goal not something I hope so achieved in a near future)

17:01 <HdkR> alyssa----: txs for es3?

17:02 <HdkR> oh, texturesize?

17:50 <alyssa----> HdkR: Yeah

17:53 <HdkR> I was thinking just tex_rect :P

19:20 <alyssa----> Blorp

19:21 <HdkR> blurp