Time Nick Message 03:43 lhofhansl Going to merge #16555 soon. (It's fairly trivial) 03:44 ShadowBot https://github.com/luanti-org/luanti/issues/16555 -- Allow FXAA to be used together with FSAA or SSAA by lhofhansl 03:50 lhofhansl And... Done. 03:51 lhofhansl If anybody wants to try out #16569, I'd be grateful :) 03:51 ShadowBot https://github.com/luanti-org/luanti/issues/16569 -- Add smoothing to content_cao by lhofhansl 12:46 [MatrxMT] Hey modders, any usecase for https://github.com/luanti-org/luanti/issues/16382#issuecomment-3161721293 ? 15:29 sfan5 cheapie: #16574 15:29 ShadowBot https://github.com/luanti-org/luanti/issues/16574 -- [no squash] Introduce array textures for node rendering by sfan5 15:33 cheapie :D 15:33 cheapie I just woke up and I'm busy testing some other stuff (unrelated to Luanti) but I'll give it a try later today 16:19 cheapie sfan5: Seems to work, but I don't have any performance numbers for you at the moment since I have some other CPU-intensive stuff running 16:56 sfan5 does ve-c take >60s for you to load too? 17:00 MTDiscord Why would not take and finish the atlas PR which was already tested multiple times in the past instead of writing the whole new graphic feature? # 16574 17:00 sfan5 because I think it's the wrong approach 17:01 MTDiscord Why would not take and finish the atlas PR which was already tested multiple times in the past instead of writing the whole new graphic feature? #16574 17:01 ShadowBot https://github.com/luanti-org/luanti/issues/16574 -- [no squash] Introduce array textures for node rendering by sfan5 17:01 cheapie sfan5: No, really I didn't see much change aside from the debug spew in the chat 17:04 sfan5 I see, weird 17:04 sfan5 because it does for me and after getting rid of a potential inefficiency point it still does 17:06 cheapie I'm not seeing the broken crack textures mentioned in the PR either 17:09 sfan5 uh 17:09 sfan5 that would mean array textures are not working for you 17:10 cheapie hmm 17:10 cheapie It's set for opengl3, is there some other setting I need to change? 17:12 sfan5 you need to have opengl 3.0 or later in practice 17:12 sfan5 other than that, no 17:12 cheapie "4.6 (Compatibility Profile) Mesa 25.2.4-1" is what it's reporting 17:13 sfan5 sounds correct 17:17 cheapie I get the line the PR talks about though, "pre: 7653" 17:17 sfan5 and if you punch a node it doesn't turn black? 17:17 cheapie Nope, I get the normal crack animation 17:18 cheapie But I'm also not sure I'm seeing as much of a performance improvement as I should (hard to tell for sure with the CPU load I mentioned) 17:19 sfan5 well that can happen but you should find that crack doesn't work especially on common nodes like stone, dirt, ... 17:20 cheapie The inventory item images do seem brighter than usual, FWIW 17:20 cheapie Just tried some stone, dirt, grass, etc., cracks are working on all of them 17:22 sfan5 🤷 17:26 sfan5 uh wait I found a bug 17:33 MTDiscord The atlases are actually more efficient approach for decreasing the drawcalls and bindings count. Firstly, it uses the simple 2D textures which are supported on all platforms which Luanti supports itself. Secondly, one atlas allows to pack inside itself millions of textures and of various size, so there's almost no a practical limitation even if use hundreds mods each creating and caching tens of images. Thirdly, there's no need to 17:33 MTDiscord use 3d uvws (third w component is a number of the current array layer) like the texture arrays require, so you still can use 2d 17:35 Krock how do texture atlases avoid color bleeding caused by interpolation? solved by not using interpolation? 17:36 sfan5 how do you handle GL_REPEAT with a texture atlas? 17:36 Krock 🤷 17:36 MTDiscord The texture arrays are most suitable if you use only (or predominantly) textures of one size, say 16x16, 32x32 and etc 17:37 sfan5 but I think we can skip this discussion because this was already brought up when I announced working on this feature a while ago 17:37 MTDiscord but luanti supports even not non power-of-two textures 17:37 Krock it does support those 17:38 MTDiscord issues with the filtering and mipmapping? there are solutions which I used in my PR and they work more or less, just need still fixing 17:39 MTDiscord yes and with the texture arrays approach it creates the significant problem 17:41 pgimeno "not non power-of-two" means "power of two", right? 17:42 MTDiscord it is solvable too, just shift the uvs to [0-1] range which get beyond it 17:43 MTDiscord mistake 17:44 pgimeno "how do you handle GL_REPEAT with a texture atlas?" - you don't 17:45 MTDiscord what "you don't"? 17:46 pgimeno you can't have GL_REPEAT on an image that comes from an atlas, or more precisely, it won't do what you expect 17:47 pgimeno GL_REPEAT applies to texture coordinates < 0 or > 1, and in an image from an atlas they are a fraction between 0 and 1 17:48 pgimeno you could maybe do some trickery with extra data per vertex and a fragment shader, but it would get pretty ugly and prone to rounding errors 17:52 MTDiscord yes, therefore those texture coordinates should be clamped to [0-1] by shifting them by the integer uv space units numbers 17:56 sfan5 you can't do that if the triangle covers more than [0-1] in uv space 18:01 MTDiscord this appears only with the mesh drawtype and as I figured out during my atlas testing with using many texture-filled mods generally it doesn't bring a big problem as actually very few textures for meshes get beyond [0-1] 18:02 sfan5 pretty sure it applies to nodeboxes too 18:02 sfan5 cheapie: you may test again 18:02 cheapie one moment 18:07 MTDiscord the nodeboxes uvs were fixed entirely in my PR 18:07 cheapie sfan5: I still don't see it take any longer to load, but cracks are indeed broken now 18:08 sfan5 cheapie: good, that means it's really working this time 18:11 cheapie I was able to pause at least most of that other CPU load I had going on, and right now I'm certainly getting much better performance at VE-C's spawn than I used to 18:12 cheapie (flying around to load stuff now) 18:13 cheapie Yeah, high 50s with a view range of 300, standing in that same place I was using for testing before 18:39 cheapie sfan5: I can't remember if I ever mentioned, but if it helps testing - on VE-C you can click on the gear icon near the bottom of the inventory menu and change, among other things, your physics override settings 21:22 MTDiscord Does the server cache player meta whenever they join the game? Or does it read the player db every time it needs to access player meta?