After 2 years of sitting on it, I've finally released my unfinished port (only 1 level) of Crystal Castles to the Playdate. Even though the repo is a bit of a mess, I hope someone finds value in it

https://github.com/DanB91/Crystal-Castles-for-Playdate

Was busy, but finally got catmull-clark going 😄

I know I have been relatively quiet and not posting much over the last few months, but there's a good reason for it:
After last year's BSC, a few guys (@sanjacobs, @moritzfalk , @victor.caraulan , @thesandvichmaker and I) started working on a handmade project that has been getting us too excited not to talk about it already.
We're working on a video editor, and we're making it good. We've written it from scratch in Odin, it runs natively on Windows and macOS, (Linux on the plan as well) and it's coming out real soon!
If this interest you, you can join our Discord server for updates: https://discord.gg/AVHVF8sjm4
We currently are running a closed beta and we're adding people periodically.
You can watch a small demo of it here that Sander put together:

Made a small physics engine (in C++) that is decently stable. It is not much, but it is something
It supports 2d capsules and circles as colliders (with friction support), and objects can have multiple colliders (the center of mass that is used as origin for rotation is calculated from each of the colliders' surface area).

After very complicated struggles on piece-wise gregory patch for G1 continuity and found out the twist compatibility condition is still a hit and miss, I decide to rework the evaluation to using iterative catmul-clark... fingers crossed, it should be actually easier

I implemented a machine learning based upscaler from scratch. To do this, I created my own custom CNN (and GAN) framework (implementing forward/backward passes, dataset batching, etc) using SYCL (C++) for wide compute device support such as GPUs (GPGPU).
The gallery shows a CNN that has 300k parameters upscaling various game screenshots/images, sometimes 2x and sometimes even 4x is used because my model can handle arbitrary upscaling factors.
The CNN is trained with a 120k-sample dataset, compiled and preprocessed by my code, using my custom framework to train of course. The architecture uses a Pixel Shuffle layer to do the tensor reshaping necessary for upscaling, and 7 layers of 3x3 convolutions to do feature extraction, as well as a Sobel edge-detection texture passed in as an input channel.
Text is also upscaled well (well, you guys should be judge for that)
Trained completely using my framework (no external ML libraries used like tensorflow/torch.)

I recommend clicking on each of the images to fully see the upscaled details

Added array accessors to my calculator language, in the video i initialize an array, access the indexes, then change the value at index 2, and then show accessing them again to initialize a new array.

Latest: https://youtu.be/05WEqLTpHwY

https://cdn.discordapp.com/attachments/1403474329081086023/1499057677425446972/image.png?ex=69f369c1&is=69f21841&hm=51e506c437ede507c3639fd138b351e716a351d3e8a629fa1b3e3c9d2c45eb4a
A little teaser of what I've been working on....

Silhouette POM on dynamic rounded corners with seamless tiling:
https://youtu.be/yphiknZNO9E

&zvideotrimmer Went a bit overtime but calling it there for the jam. Couldnt quite get encoding working well. But will definitely finish this up post-jam. Learned a bunch about the Media Foundation api and video encoding. Also that multithreading is super important in this domain as video decoding is computationally VERY expensive, even with modern hardware.
Also noticed how bad almost all video players scrub video, even high quality ones like MPV. I did not realize that pre-jam. Feel like its certainly doable to do better job and get close to what proper video editors like Davinci offer with scrubbing/trimming in a much lighter weight package. Even though I'm not encoding well, it is close, and the program is only 149kb with zero dependencies on windows. So even though this jam project was a fail, I really enjoyed diving into this problem domain, and look forward to improving it. Thanks very much Handmade Network admin, this jam was awesome.

Sadly, I didn't make much past this yesterday. The only progress I made yesterday was an internal architecture decision on how to interact with Odin's nbio. I guess I didn't have the time I thought I would, and spent too long hemming and hawing on some design decisions. All I have at the end of the jam for &send-files

• A mostly non-functional UI layout
• Able to copy and paste contacts
• A rough idea of how the protocol will work
• A rough implementation of the first step of this protocol in the form of a unit test

I do plan to continue on working this in some capacity after the jam though. I like where I am going with it

I pervasively can’t use &foo with these fixed-cap dynamic arrays

Great! Really my only complaints are a couple bugs (?) with the &foo syntax in loops

Our final Jam submission video - had to upload to youtube as the video was too big for this site:

https://www.youtube.com/watch?v=be4MmWPEgRs

&Cacophontonic coming along nicely

definitely picking fun / silliness over professionalism due to it being a jam project!

Parallax Occlusion Mapping with Silhouette on columns with proper curvature:

At first I managed to get really fast scrubbing, but then I realized that was because only key frames were being previewed by IMFSourceReader_SetCurrentPosition. Things slowed significantly once I started decoding all the frames around the cursor. It made me realize why scrubbing is generally slow in most video players, frames need to be decoded in real time when you seek, as all the frame information isn't actually present in the video file. Mainly key frames and diffs.
But that made me think of a slight optimization, outside of caching all frames, you could just display key frames while scrubbing quickly, then decode all frames if scrubbing really slow. Practically, human speed of mouse cursor movement should be enough information to do this well. If I dont manage to actually finish the trimming part (handling video using the MediaFoundation api is a bit trickier than I realized), I'll try implementing the seek optimization. &ZVideoTrimmer

&send-files progress:

• Got copying and pasting contact information working (attached)
• Wrote a test that will reflect the flow for sending a "file send request" and "accept file send request" packets

Yea... there's no way I am getting anything remotely close to actually transferring files by tomorrow lol

After 2 years of sitting on it, I've finally released my unfinished port (only 1 level) of Crystal Castles to the Playdate. Even though the repo is a bit of a mess, I hope someone finds value in it

https://github.com/DanB91/Crystal-Castles-for-Playdate

Was busy, but finally got catmull-clark going 😄

I know I have been relatively quiet and not posting much over the last few months, but there's a good reason for it:
After last year's BSC, a few guys (@sanjacobs, @moritzfalk , @victor.caraulan , @thesandvichmaker and I) started working on a handmade project that has been getting us too excited not to talk about it already.
We're working on a video editor, and we're making it good. We've written it from scratch in Odin, it runs natively on Windows and macOS, (Linux on the plan as well) and it's coming out real soon!
If this interest you, you can join our Discord server for updates: https://discord.gg/AVHVF8sjm4
We currently are running a closed beta and we're adding people periodically.
You can watch a small demo of it here that Sander put together:

I've published a public domain Odin package for colour space conversions targeting UI development. It implements OKLab, OKLCh, sRGB and Display P3 transform pipelines with piecewise gamma encoding and decoding. Based on Björn Ottosson's perceptual colour spaces. Gamut clipping, OKHSL and OKHSV are going to be included in the future releases. https://github.com/heavyrain266/colourspace

Porting Alula's zuma deluxe decompilation project to other platforms, right now it's running on macOS and Android
(game data not included, you need to buy it on steam)
https://ksylvestre.itch.io/zuma-portable

Made a small physics engine (in C++) that is decently stable. It is not much, but it is something
It supports 2d capsules and circles as colliders (with friction support), and objects can have multiple colliders (the center of mass that is used as origin for rotation is calculated from each of the colliders' surface area).

After very complicated struggles on piece-wise gregory patch for G1 continuity and found out the twist compatibility condition is still a hit and miss, I decide to rework the evaluation to using iterative catmul-clark... fingers crossed, it should be actually easier

Color pickers are now here with my own ui and rendering for it all.

Have been working on pencil since jam. Got d3d rendering for ui, turned off for the application itself, working on shaders to be able to have ui running. Here is a color picker.

I implemented a machine learning based upscaler from scratch. To do this, I created my own custom CNN (and GAN) framework (implementing forward/backward passes, dataset batching, etc) using SYCL (C++) for wide compute device support such as GPUs (GPGPU).
The gallery shows a CNN that has 300k parameters upscaling various game screenshots/images, sometimes 2x and sometimes even 4x is used because my model can handle arbitrary upscaling factors.
The CNN is trained with a 120k-sample dataset, compiled and preprocessed by my code, using my custom framework to train of course. The architecture uses a Pixel Shuffle layer to do the tensor reshaping necessary for upscaling, and 7 layers of 3x3 convolutions to do feature extraction, as well as a Sobel edge-detection texture passed in as an input channel.
Text is also upscaled well (well, you guys should be judge for that)
Trained completely using my framework (no external ML libraries used like tensorflow/torch.)

I recommend clicking on each of the images to fully see the upscaled details

Did another largish refactor so that all the vertices are in a single buffer so that I can render everything with multidraw indirect. Then I added a second pass for foliage. The change significantly improved performance, it's pretty insane how fast computers are

I just added the ability to diff directories in gap. It makes it so much more usable via git difftool -d since now you can view all the diffs in a single invocation of gap.

Added array accessors to my calculator language, in the video i initialize an array, access the indexes, then change the value at index 2, and then show accessing them again to initialize a new array.