The Hornet statuette and Fixer statue in Silksong is probably one of my all time favorite pieces of video game storytelling. The rollercoaster of emotions getting the statuette with its description first and then walking past the statue. I love this medium.

Graffathon happened again and this time I went for an executable graphics looping animation thingy instead of a full demo. There has to be a less messy way to create SDF character rigs, right? www.shadertoy.com/view/tfdSzf

I keep wondering what Blue Prince would be like if the room RNG was deterministic and seeded by the existing floor plan. What if the biggest shortcoming of the game was yet another layer to the puzzle?

How it started: Let’s have concrete SoA for entity archetypes because that’s simple!

How it’s going: Two layers of CRTP for generic typesafe iterators.

Maybe it’s not worth this level of metaslop just to prevent someone from constructing broken accessors at compile time 🫠

I added a quick traditional multi-resolution separated blur implementation for bloom, because why not have that too for reference. I of course added the lens flare gather to one of the bloom mips and ended up finding a cool, more intricate flare pattern in the process 😄

Good thing my vintage Sponza has a glint problem as I can use it for Bloom dev 😄 Perf is still bad, but at least it's consistently bad in a way only FFT convolution can get you.

1000xResist really tackles a lot of hard topics, huh? Even the narrative structure is ambitious. Kudos to the team for taking the swing and even more for succeeding.

Vulkan specialization constants are neat, but my TAA now creates 127 pipelines 😄 I wonder if it’s a bad idea to not guide their use with a hard limit on the number of pipeline variants.

Vulkan is tricky to recommend for hobby/learning purposes. It can be very counterproductive for the time investment if you measure by rendering features. At the same time, it’s a pretty complex system with a couple of concurrent programming models. Getting comfortable with it can teach a lot.

Anyone happen to know literature sources for efficient FFT convolution of multi-channel real 2D signals? A UE4 stream suggests that you can run RG/BA through FFT as "complex", perform convolution and somehow invert the original (re,re)->(re, im) fudge within the horizontal IFFT, but...

It feels good to have my toy renderer architecture in a low-friction state. All that time spent not making the pixels prettier means I can now make them prettier faster! It took me a bit over an hour to add on-demand CS bloom kernel generation.

Apparently the automatic Ubuntu upgrade is still not foolproof and can corrupt the OS. I had also forgotten how to correctly reinstall without nuking the /home partition. On the plus side, my OS setup script and configs were about 90% complete so I only lost a couple hours getting my 24.04 on 😄

It's all fun and games until someone has to open CMakeLists.txt

Tinkering with 2D FFT is fun and all but I did spend an hour comparing my Python prototype to my Vulkan implementation for differences. Turns out I wasn't looking at a corrupted/incorrectly permutated inverse output but the noise floor for a mostly black image🤦

There's nothing quite like freeform hardcore when I need to induce a "get shit done" flow state.

It's interesting how differently existing choices appear when you rebuild parts of a codebase. Even some bits that appeared mundane start to stick out as strange and unnecessary without the dense web around them. It would be real nice to develop better intuition to spot these in situ.

You know what's better than two microblogging platforms?
*Three* microblogging platforms 🚀