Mattias Gustavsson

I am excited to share this new single-header C library I have been working on for a while now: vecmath.h, a comprehensive vector/matrix math library for graphics/games/3d.
It allows you to write vector math code in C that looks like this:

I tried my hand at making a video codec. I didn't know the first thing about video codecs when I started, so was quite fun to learn about the concepts and how to implement them.

I packaged the result in a stb-style, single-file header-only lib, < 2500 loc with no dependencies.

what are the worst things about the modern day internet?

YouTube video by Psyraven DOSBox Pure Unleashed

Got back into using the video editing features of Blender to make a trailer for some of my open source shenanigans - Announcing: DOSBox Pure Unleashed 🥳 www.youtube.com/watch?v=t4q8...

The single header lib also includes source code for an encoder utility and a playback app (both which makes use of other single header libs), behind conditional compile switches (ifdefs). Instructions for how to build them can be found at the top of the file.

The code is public domain, and available here github.com/mattiasgusta...
For most uses, you are probably better off using a real codec. But for something small, simple, self-contained with no dependencies, this might fit the bill.

The videos in the examples are compressed with my "default" setting, which makes them quite small, with ok-ish quality. On "max" they are more like dvd in size/quality. I focused on avoiding temporal aliasing, and I think i pulled that off quite well.

All in all, it seems like my quality/compression is maybe around dvd quality (?), so not in the same league as modern codecs (but attempting that would start to infringe on patents, so I don't particularly want to go down that route).

Here are two sample video snippets:
mattiasgustavsson.com/wasm/fmv.html
mattiasgustavsson.com/wasm/fmvdl.h...
(both are displayed with my CRT filter, because that's how I am most likely to be using this myself, but it can be toggled with F9).

QOI, and, by the looks of it, QOV (cool, I didn't know about that one) are both lossless, which is super cool, and such wonderfully simple formats too. Mine is a very lossy mpeg-like encoding, doing DCT compression and motion vectors and such things, so quite different, but still video compression 🙂

The single header lib also includes source code for an encoder utility and a playback app (both which makes use of other single header libs), behind conditional compile switches (ifdefs). Instructions for how to build them can be found at the top of the file.

The code is public domain, and available here github.com/mattiasgusta...
For most uses, you are probably better off using a real codec. But for something small, simple, self-contained with no dependencies, this might fit the bill.

The videos in the examples are compressed with my "default" setting, which makes them quite small, with ok-ish quality. On "max" they are more like dvd in size/quality. I focused on avoiding temporal aliasing, and I think i pulled that off quite well.

All in all, it seems like my quality/compression is maybe around dvd quality (?), so not in the same league as modern codecs (but attempting that would start to infringe on patents, so I don't particularly want to go down that route).

Here are two sample video snippets:
mattiasgustavsson.com/wasm/fmv.html
mattiasgustavsson.com/wasm/fmvdl.h...
(both are displayed with my CRT filter, because that's how I am most likely to be using this myself, but it can be toggled with F9).

I tried my hand at making a video codec. I didn't know the first thing about video codecs when I started, so was quite fun to learn about the concepts and how to implement them.

I packaged the result in a stb-style, single-file header-only lib, < 2500 loc with no dependencies.

pow2_ceil.h GitHub Gist: instantly share code, notes, and snippets.

Round a uint32 up to the nearest power-of-two value:
gist.github.com/mattiasgusta...

I found that on the latest compiler versions, my previous intrinsics-based implementations tripped some undefined behavior optimizations. fixed it now

pow2_ceil.h GitHub Gist: instantly share code, notes, and snippets.

Round a uint32 up to the nearest power-of-two value:
gist.github.com/mattiasgusta...

that'd be cool :) the current vecmath.h is already very large, but I tried to limit its features to those found in HLSL and D3DX, to have as a baseline. There's many convenience functions and higher level functionality that could be added, so I'm considering a "vecmath_util.h" for all of those

If you're looking for a vector math lib for C (or C++) check it out

github.com/mattiasgusta...

If you do take a look, please let me know what you think about it - getting feedback is a big part of the reward releasing libraries like this, and I spent a lot of time and effort on it

It's been tested on windows, mac, linux and web-assembly, using msvc, clang, gcc and tcc (tiny c compiler).

It is not trying to be the most optimized library, not using any SIMD intrinsics or assembly code, just straightforward, fully portable, simple implementations of a large number of math functions.

It has a large set of functions, as it attempts to implement most of what you get in shader languages like HLSL and GLSL, and also most of the utility functions from libraries like D3DX in DirectX.

Even though it is primarily designed as a C library, I have also tried to make it fairly nice to use from C++, conditionally adding operator overloads and constructors, but it is not trying to be idiomatic "modern C++" in any way.

Here is the library: github.com/mattiasgusta...

Now, you get the best ergonomics if you use it with clang compiler in C11 or later, as it makes (optional) use of C11 _Generics and clang vector type extension, but it doesn't require either to work.

I see those kind of operations as very much outside the scope of this library, which focus more on basic operations matching the features of HLSL and D3DX. I think for any user that needs special case functions like that, it's easy for them to build out of the primitives provided in the lib

No, I don't use any optimizations like that, it's just pretty much the simplest implementation of textbook Cramer’s rule, for generic inversion

yeah, definitely give clang a go! i find that for my use cases, it compiles faster AND generates more performant code than both msvc and gcc. and, it got that taste vector extension :)