00:05 < bridge> lmfao 00:06 < bridge> wait sry what is ddnet-rs :justatest: 00:06 < bridge> a different ddnet rust project 00:06 < bridge> or renamed & moved 00:08 < bridge> it's ddpg 00:36 < bridge> That sounds like you are shooting at deen the full time database engineer 00:40 < bridge> There are code bases that people complain about. And there are code bases nobody uses. 00:55 < bridge> i wish i wasnt so stupid so i could actually make the updates i want 00:56 < bridge> you can only make updates you dont want? 01:05 < bridge> not what i meant :nouis: 02:29 < bridge> there is a toggle in a PR. but that might take longer, we can remove the spec cursor and spec cam feature for 18.9 upon release for now. 02:30 < bridge> it's either removed or there will be a toggle before release 03:09 < bridge> what do you think of subprocess in binds 03:10 < bridge> so we don't have to create many files for toggles 04:58 < bridge> rip 05:12 < bridge> are there any good .HLSL tutorials around? i want to write something for a project i'm doing but have no single clue what anything means 05:13 < bridge> what are you writing the shader for 05:14 < bridge> https://learn.microsoft.com/en-us/windows/win32/direct3dgetstarted/work-with-shaders-and-shader-resources 05:15 < bridge> wait that sucks 05:15 < bridge> https://learn.microsoft.com/en-us/windows/win32/direct3d11/d3d11-graphics-reference-d3d11-shader 05:15 < bridge> https://learn.microsoft.com/en-us/windows/win32/direct3d11/d3d11-graphics-reference-d3d11-shader 05:16 < bridge> There's a cool program, called Tooll3, which is sort of like TouchDesigner, but free and open source. 05:16 < bridge> You can literally create your own functions and connect them to make cool visuals. Though what I plan on doing is to try and make a certain type. 05:16 < bridge> To make what I want optimally it involves multi-thread computing, and hlsl is the best bet for it 05:16 < bridge> I don't even know if I'm saying any of this right, or if this is the answer. 05:16 < bridge> hlsl is a shader language so you first need to figure out what kind of interface this hlsl is even giving you 05:18 < bridge> and multi threading is not necessarily made more optimal by hlsl it's just that it's your way of leveraging the gpu so you have hella parallelization 05:18 < bridge> but if it's i/o bound at all or anything like that then idk 05:48 < bridge> https://github.com/tooll3/t3/wiki/help.ShaderDevelopmentExample 05:49 < bridge> Tooll3 supports hsls compute shaders which I'm assuming is what you want 05:49 < bridge> Tooll3 supports hlsl compute shaders which I'm assuming is what you want 05:49 < bridge> yeah thats what i want 05:50 < bridge> hold on lemme send the uh, video of what im doing 05:50 < bridge> https://cdn.discordapp.com/attachments/293493549758939136/1325687851433857024/2025-01-03_01-33-49.mp4?ex=677cb285&is=677b6105&hm=3f4c6cd096f61389afc63c56543dbcd92a779a0da98dfeeaec742fbefdb060bc& 05:50 < bridge> mmm that's nice node spaghetti 05:51 < bridge> its all math too! but its math that has to like, store a list and constantly look back and forth 05:52 < bridge> the main thing you do with compute shaders is turn GPU memory into other kinds of GPU memory 05:52 < bridge> and you can write to buffers 05:52 < bridge> so you'd probably want to start backwards and figure out how to do what you want under the contrains of the compute shader 05:52 < bridge> so you'd probably want to start backwards and figure out how to do what you want under the contraints of the compute shader 05:52 < bridge> so you'd probably want to start backwards and figure out how to do what you want under the constraints of the compute shader 05:54 < bridge> it feels like the completely wrong way to do it, but for context: 05:54 < bridge> I'm writing a tool to visualize rounded logos, using circles and lines. I first have to go through all the points, figuring out which tangent line between both circles should be rendered (which is a lot of math), and then I have to draw arcs, connecting one end of one line, to the other end of the previous line, so it looks seamless. The problem with THAT is that right now, I have to do the tangent calculation 3 TIMES in order to get the output, 05:57 < bridge> why do you need more performance 05:59 < bridge> Because I plan on using it for real-time motion graphics, which will use, presumably around 15 unique shapes. 05:59 < bridge> are the shapes also procedural 06:00 < bridge> Well, yeah. Each and every point has its own movement, as shown in this video 06:00 < bridge> but the shape before the movement 06:00 < bridge> is that procedural 06:00 < bridge> you can apply the movement effect for super cheap with a vertex shader 06:01 < bridge> Oh, let me look that up- 06:05 < bridge> Ohhhh I was looking at COMPUTE shaders before...! 06:05 < bridge> Yeah, I should DEFINTELY look into vertex shaders, then. 06:06 < bridge> I mean compute shaders are kinda like the swiss army knife of shaders, they can do the job of most of the other kinds 06:07 < bridge> but in this case you only need a vertex shader if you just want to procedurally animate the vertecies 06:07 < bridge> but in this case you only need a vertex shader if you just want to procedurally animate the vertices 06:09 < bridge> ohhh my head hurts LOL 06:17 < bridge> they have a discord btw you can ask in there 06:17 < bridge> i AM in there 06:17 < bridge> o ok 06:18 < bridge> for now i just added a debug mode thats like 3x faster somehow (it barely does any calculations) 06:18 < bridge> https://cdn.discordapp.com/attachments/293493549758939136/1325695024331231242/image.png?ex=677cb934&is=677b67b4&hm=337c4855b24cccba53c071fc667f0bb49aadf657eb69a0ea85874cf5c8b51af0& 06:29 < bridge> GPU bottleneck, too much data is being fed to the GPU, causing freeziness 06:30 < bridge> I know it's GPU problem, cuz the calculations are pre-done after the next frame, however data to the GPU are sometimes discarded cuz it cannot be rendered in time 06:30 < bridge> otherwise, it would look really slow, while also going at 10 fps 06:31 < bridge> it's probably not a gpu bottleneck 06:31 < bridge> Well, care to reiterate? 06:31 < bridge> My first thought was that 06:32 < bridge> he already explained the reason it's slow is because of the procedural shape generation that needs to be done every frame. 06:32 < bridge> the calculations are done every frame tho 06:32 < bridge> the gpu can render about 1000x more triangles than he is showing in the window without issue 06:33 < bridge> There is one exception where the GPU cannot handle that many 06:33 < bridge> computers can generate realistic scenes in realtime 06:33 < bridge> im just a dumbass :nouis: 06:34 < bridge> Yeah through optimized data caching and optimized compressed data sizes 06:35 < bridge> it all comes down to how much work the CPU should do and how much data the GPU should have 06:35 < bridge> the less, the better 06:36 < bridge> he is using a fully procedural framework to make the graphics, it's not the same as regular scenes that get loaded into memory and renders static triangles 06:37 < bridge> Does that kind of framework rely on SIMD? 06:38 < bridge> I did not write it 06:38 < bridge> but likely not 06:38 < bridge> @hectavoxel I haven't asked yet, but what have it caused bottleneck? 06:39 < bridge> if you may have any idea on where went wrong 06:40 < bridge> the CPU shouldn't be it, cuz the positioning for all of the generative triangles are in place, based on the amount of skipping steps from the next frame rendered 06:40 < bridge> either the Shader used are insanely unoptimized, or the GPU data feed is larger, than expected 06:42 < bridge> that's not true at all 06:43 < bridge> just because the framerate is low does not mean the gpu is bottlenecking 06:50 < bridge> i dont know 09:16 < bridge> Why close @tsfreddie 09:18 < bridge> got other things to do, just bite more than i can chew, it just need some limit and some testing probably. someone can pick it up. 09:19 < bridge> Mh OK. I think that was actually a useful PR. Saner than our byte limit approach right now 09:21 < bridge> heinrich had a rather good point that you can fit kind of an entire sentence in 16 chinese codepoints 09:21 < bridge> ye, that is fair but I already spent way too much time on this so maybe someone can figure out the design. code should still stand if someone wants to takeover. 09:21 < bridge> But does that matter? As long as it's not longer visually 09:22 < bridge> kinda yeah? 09:22 < bridge> Why 09:22 < bridge> it's not longer than "WWWWWWWWWWWWWWWW" but only to that one. 09:22 < bridge> https://cdn.discordapp.com/attachments/293493549758939136/1325741286585532537/image.png?ex=677ce449&is=677b92c9&hm=745b3be55888384629c1a4b1f3c7a7986e5dd4a8dd2e2c6ad703a11a55eec721& 09:22 < bridge> it's not longer than "WWWWWWWWWWWWWWWW" but only not to that one. 09:22 < bridge> this fits 09:23 < bridge> And so? You can name yourself a sentence 09:23 < bridge> I can also make a Latin language that could do that 09:24 < bridge> If I name myself MF, then you know what I mean 09:24 < bridge> usually chinese games does have a 2:1 character weighting, chinese are 2, latins are 1 09:25 < bridge> it sorta bypasses the mute feature because we cannot mute usernames 09:25 < bridge> and you can add clan 09:25 < bridge> actually that's already a problem anyway 09:25 < bridge> in chinese 09:25 < bridge> Yeah but other languages like Russian would profit too 09:25 < bridge> yeah but it's mitigated by the low charecter count 09:25 < bridge> not really. you can just use letters and it reads 09:26 < bridge> I think caring for sentences is over thinning 09:26 < bridge> Thinking 09:26 < bridge> GLYWCNMGDSB is a full sentence 09:26 < bridge> in chinese 09:26 < bridge> in english I can't think of anything that someone could say to me in 16 charecters that would insult me personally enough to care 09:28 < bridge> and in chinese you can already do that 09:28 < bridge> without chinese characters 09:29 < bridge> and if someone wants to insult you with a full sentence but they can't do it in english. if they do it in chinese you probably can't even tell it is happening. 09:29 < bridge> anyway, i think heinrich is more worried about infomation density than only insults 09:30 < bridge> He wants graphemes, but that's not realistic 09:30 < bridge> isn't chinese graphemes still only one character? 09:30 < bridge> I was thinking that 09:31 < bridge> i also doubt heinrich would want english graphemes to reach 16 too 09:31 < bridge> idk how graphemes fixes anything 09:31 < bridge> Is it? Are there no like variants in different parts of China or smth? Xd 09:32 < bridge> chinese is always one character per character. there is no words in chinese. i mean there is but it is not defined 09:32 < bridge> https://www.reddit.com/r/ChineseLanguage/comments/1e4ji8i/what_is_a_grapheme_in_the_context_of_character/ 09:33 < bridge> anyway, chinese is pretty solvable just by using the 2:1 weighting 09:33 < bridge> so if anyone wants to do that they can just add it. 09:33 < bridge> Well they are the Unicode standard for characters 09:33 < bridge> but we display unicode 09:33 < bridge> But many of these letters still have a single Unicode codepoint anyway 09:34 < bridge> Yes but we could not display emojis 09:34 < bridge> Not without a custom format 09:34 < bridge> idk the weighting thing seems fine 09:34 < bridge> isn't the problem too long? instead of too short? fit 16 unicode graphemes can be longer than 16 codepoints 09:35 < bridge> ye but utf8 is already similar to the weighting thing for chinese 09:35 < bridge> Yeah but that sounds like u want to parse the content 09:35 < bridge> going from 5 to 7 is not really anything and i think chinese people can cope with 5 09:35 < bridge> isn't it 4:1 instead of 2:1? 09:35 < bridge> Ofc it is 09:35 < bridge> or no? 09:35 < bridge> 3:1 mostly 09:35 < bridge> some can go up to 4 09:35 < bridge> but mostly 3 09:35 < bridge> It's a annoying design imo 09:35 < bridge> but other languages are hurt more than chinese right? 09:35 < bridge> like russian? 09:36 < bridge> yes 09:36 < bridge> but i'm not even russian how am i gonna assess how useful my implementation is 09:36 < bridge> 😬 09:37 < bridge> Fwiw I vaguely remember the unicode consortium actually suggesting a byte limit 09:37 < bridge> if cryllics or linguistic people wants to pick up that PR they can do that. 09:37 < bridge> i don't even remember why i wanted to implement this. it was like 8 years ago when i first complained about it 09:37 < bridge> With which reasoning? 09:37 < bridge> i just remembered i wanted it but it seems like a wrong motive 09:38 < bridge> so i dropped it 09:38 < bridge> I’ll try to dig it back up 09:39 < bridge> одааа колобки одаааа 09:40 < bridge> I mean if traffic is important or similar OK. Utf8 hides this information. But considering a code point is only Max four bytes rn.I dunno 09:40 < bridge> i made this lol 09:40 < bridge> https://cdn.discordapp.com/attachments/293493549758939136/1325745837392793660/image.png?ex=677ce886&is=677b9706&hm=31fded501d14febebc162db2b3badd141418c750744e1356cf3082f34d303315& 09:40 < bridge> Or six bytes 09:41 < bridge> What is that 09:41 < bridge> In game? 09:41 < bridge> just a website 09:42 < bridge> I guess heinrich never provided a source for it 09:42 < bridge> He was about graphemes 09:43 < bridge> 0.7 has a codepoint limit, not a grapheme limit 09:43 < bridge> But again graphemes are a completely different complexity 09:43 < bridge> That is what he dislikes 09:44 < bridge> You think the consortium suggests grapheme limits instead? 09:44 < bridge> Yes 09:44 < bridge> I don't see why graphemes are significantly better than codepoints 09:44 < bridge> anyone knows what the white styling is in a