When creating graphic libraries you most likely end up dealing with points and rectangles. If you're particularly unlucky, you may end up dealing with affine matrices and 2D transformations. If you're writing a graphic library with 3D transformations, though, you are going to hit the jackpot: 4x4 matrices, projections, transformations, vectors, and quaternions.

Most of this stuff exists, in various forms, in other libraries, but it has the major drawback of coming along with the rest of those libraries, which may or may not be what you want. Those libraries are also available in various languages, as long as those languages are C++; again, it may or may not be something you want.

For this reason, I decided to write the thinnest, smallest possible layer needed to write a canvas library; given its relative size, and the propensity for graphics libraries to have a pun in their name, I decided to call it Graphene.

This library provides types and their relative API; it does not deal with windowing system surfaces, drawing, scene graphs, or input. You're supposed to do that yourself, in your own canvas implementation, which is the whole point of writing the library in the first place.

Graphene is a C library. For this reason alone, it depends on GLib, as GLib gives us a fairly decent, portable, and maintained layer to depend on, instead of requiring me to re-implement a ton of code.

Graphene does not use the GObject type system, as of yet, but it may grow an optional dependency on it through a separate shared object, especially if in the future I'm going to generate introspection data for the benefit of GObject-based language bindings.

Graphene contains optimizations for speeding up vector operations; those optimizations are optional, and used only if both Graphene was compiled with support for them and if the system you're running on has them. Currently, Graphene supports the following platform-specific fast paths:

• Streaming SIMD Extensions (SSE) 2
• ARM NEON
• GCC vector extensions

In the remote case in which none of these optimizations are available, Graphene will fall back to a naive scalar implementation.

In order to build and install Graphene you will need development tools and the headers of the dependencies.

First of all, clone the Git repository:

$ git clone git://github.com/ebassi/graphene
$ cd graphene

Then run:

$ ./autogen.sh
$ make
$ make check
# make install

It is possible, when building Graphene, to disable specific optimizations by passing options to the configure script:

• --disable-sse2 - will disable the SSE2 fast paths
• --disable-arm-neon - will disable the ARM NEON fast paths
• --disable-gcc-vector - will disable the GCC vector intrinsics

See the output of configure --help for more information.

If you found a bug (which I'm sure there will be plenty), or if you want to add your own pet feature, then follow these steps:

1. Fork the ebassi/graphene repo
2. Fix bugs or add new features and push them to your clone
3. Open a new issue
4. Open a pull request
5. Wait for me to give feedback on the pull request
6. Celebrate when your code gets merged

That's pretty much it.

Graphene provides common types needed to handle 3D transformations:

• 2D points
• 3D points
• rectangles
• quads
• quaternions
• vectors (2, 3, or 4-sized)
• matrices

All types can be placed on the stack, but provide allocation/free functions for working on the heap as well. The contents of all structure types, except for points and rectangles, should be considered private, and should never be accessed directly.

• Segfault in the scalar code path for matrix interpolation with clang. It seems that compiling with clang and -O2 causes a segmentation fault in the graphene_matrix_t interpolation code in an assignment between two SIMD4f types, when using the scalar implementation. Tracked in issue 1.

Graphene is released under the terms of the MIT/X11 license.

See the LICENSE file for more details.