kb

And oh btw, a few weeks ago I would really liked to have a screen recording tool that doesn't suck (eg. can't keep up a steady 60 fps) Do yo need to do _everything_ yourself? really? github.com/kebby/Capturin…

@iquilezles I checked out vcpkg once and never looked back. It‘s still somewhat ridiculous to watch it pull 500 dependencies just for building one simple library but it always builds and so far everything worked. Total lifesaver for me.

The reason I've never been able to use any open source library is that "you only need python, npm, node, yarn, make, ninja and ssh, but only some pecific versions, and then type these things in the command line like it's still the 90s, only they never work, fuck this shit".

@NickChapmn @iquilezles @SebAaltonen There is a boolean variable (gl_FrontFacing in GLSL or SV_IsFrontFace in HLSL) for fragment/pixel shaders.

@kebby @iquilezles @SebAaltonen How do you determine whether to flip in the shader?

Bug report: I want to scale an object to be flat. I set the X-axis scale to zero. Lighting breaks! When the X-axis in the scale matrix is zero, that matrix can't be inverted. Divide by zero is not legal. So, we can't calculate the inverse-transpose matrix for normal transform.

@NickChapmn @iquilezles @SebAaltonen Then do culling :) . What I tried to say is that "don't flip in the matrix, but do in the shader" seems to always get you the proper result even with zero or negative scale :)

@kebby @iquilezles @SebAaltonen but I want culling, it makes rendering faster :)

@NickChapmn @iquilezles @SebAaltonen I experimented with this today, and turns out: For a normal triangle renderer, not having the normal flip actually fixes double sided lighting (using VFACE, etc) w.r.t flipped world matrices. Adjugate, no culling, flip backface normals, and everything's always correct. :)

@iquilezles @SebAaltonen I also realised you need to compute the determinant anyway, in order to decide to do front face or back face culling.

Fernsehen (oder Events) produzieren ist toll. Gleichzeitig so tun als wäre alles spontan und nicht geplant, aber bei jedem Fuckup als sollte das gerade so. 😀 #24hJK

🤖 Our Devs will join one of the world’s largest computer art festivals: @revision_party! We are proud to be home to some legends of the demoscene: – Chaos of Farbrausch – kb of Farbrausch – BoyC of Conspiracy #Ventuz #Revision2024 ©Photos by T. Findeisen

We‘re excited to kick off the #evoke2024 season with our first organizer meeting - let’s make some ✨good✨ things happen!
#demoscene

@deiruch @mkristensen It would honestly help if that symbol server did more than 10Mbits/sec.

@kebby This one is for @mkristensen :)

I made this, and VS is still going.

Just in case anyone responsible for ITU-T H.273 is (still) on here: Why did you include the DCI-P3 primaries but forgot to include a 2.6 gamma in the TCS? I want answers.

Our invitation is live and won the first place at Nordlicht! A Farbrausch + Slay Bells production. 🫶🏻 pouet.net/prod.php?which…

@conspiracyhu @DefenceForceOrg @Artoria2e5 @Jonathan_Blow Er, ok. I stand corrected. In that case - yeah, it doesn't really make sense, neither from a size nor a security perspective.

@kebby @DefenceForceOrg @Artoria2e5 @Jonathan_Blow

Your "lightweight" browser starts with a 760MB(!) installer. That's 12163 64k intros, which is roughly 36489 minutes (~25 days) of HD audio/video content. @Jonathan_Blow was right: Computing has lost the plot.

@DefenceForceOrg @Artoria2e5 @conspiracyhu @Jonathan_Blow So yeah, the irony is: It's 875MB and the browser core isn't even part of it :D

@DefenceForceOrg @Artoria2e5 @conspiracyhu @Jonathan_Blow Wait. At least in your screenshot there isn't any copy of WebView2. It's a lot of stuff but most of it is what I'd expect from a default deployment of a .NET 6 app using SkiaSharp for UI (which probably brings all the winforms/presentationcore deps)

@leonard_coder Tbh it probably was since all the engines switched to HDR internally, but it certainly didn’t hurt :)

@kebby thanks to A.I hype, fp16 is first class citizen now :)

If you're looking for a good GPU related rabbit hole this weekend, "perf differences between UN16 and FP16 formats" is a quite interesting one. Just saying.

(measurements taken with a cheap and unoptimized gaussian blur shader so there's some ALU involved, the actual differences are possibly even bigger)

So, er, the gist is "use fp16 instead of un16/sn16 if the precision is good enough for you, and load instead of sample if you don't need filtering". Kinda makes sense if you consider where the optimized paths in modern GPUs might lie, but the outliers show it's not that easy :)