Yocto/GL: Tiny C++ Libraries for Data-Oriented Physically-based Graphics

Yocto/GL is a collection of small C++17 libraries for building physically-based graphics algorithms released under the MIT license. Yocto/GL is written in a deliberately data-oriented style for ease of development and use. Yocto/GL is split into small libraries to make code navigation easier. See each header file for documentation.

• yocto/yocto_math.{h}: fixed-size vectors, matrices, rigid frames, rays, bounding boxes, transforms
• yocto/yocto_color.{h}: color conversion, color adjustment, tone mapping functions, Perlin noise, shading and integration utilities
• yocto/yocto_geometry.{h}: geometry functions, ray-primitive intersection, point-primitive overlap
• yocto/yocto_noise.{h}: Perlin noise
• yocto/yocto_sampling.{h}: random number generation, generation of points and directions, Monte Carlo utilities
• yocto/yocto_shading.{h}: evaluation and sampling of fresnel functions, bsdf lobes, transmittance lobes, phase functions
• yocto/yocto_shape.{h,cpp}: various utilities for manipulating triangle meshes, quads meshes and line sets, computation of normals and tangents, linear and Catmull-Clark subdivision, mesh loading and saving, procedural shapes generation, ray intersection and closest point queries
• yocto/yocto_mesh.{h,cpp}: computational geometry utilities for triangle meshes, mesh geodesic, mesh cutting, mesh loading and saving
• yocto/yocto_bvh.{h,cpp}: ray intersection and closest point queries of triangle meshes, quads meshes, line sets and instances scenes using a two-level bounding volume hierarchy
• yocto/yocto_image.{h,cpp}: simple image data type, image resizing, tone mapping, color correction, image loading and saving, procedural images, procedural sun-sky, advanced color conversion utilities
• yocto/yocto_sceneio.{h,cpp}: simple scene and scene loading and saving of Ply/Obj/Pbrt/glTF and a custom and scalable Json format
• yocto/yocto_trace.{h,cpp}: path tracing of surfaces and hairs supporting area and environment illumination, microfacet GGX and subsurface scattering, multiple importance sampling
• yocto/yocto_modelio.{h,cpp}: parsing and writing for Ply/Obj/Pbrt formats
• yocto/yocto_commonio.h: printing utilities, file io utilities, command line parsing
• yocto/yocto_common.h: container and iterator utilities
• yocto/yocto_parallel.h: concurrency utilities

You can see Yocto/GL in action in the following applications written to test the library:

• apps/yscenetrace.cpp: command-line path-tracer
• apps/ysceneitrace.cpp: interactive path-tracer
• apps/ysceneitraces.cpp: simpler version of apps/ysceneitrace.cpp for demos
• apps/ysceneproc.cpp: command-line scene manipulation and conversion
• apps/yshapeproc.cpp: command-line mesh manipulation and conversion
• apps/yimageview.cpp: Hdr/Ldr image viewer with tonemapping and color grading
• apps/yimageviews.cpp: simpler version of apps/yimageview.cpp for demos
• apps/yimageproc.cpp: command-line image manipulation
• apps/yimagedenoise.cpp: command-line image denoiser that uses Intel Open Image Denoise
• apps/ysceneview.cpp: simple OpenGL viewer

Here are some test images rendered with the path tracer. More images are included in the project site.

Design Considerations

Yocto/GL follows a "data-oriented programming model" that makes data explicit. Data is stored in simple structs and accessed with free functions or directly. All data is public, so we make no attempt at encapsulation. Most objects is Yocto/GL have value semantic, while large data structures use reference semnatic with strict ownership. This means that everything can be trivially serialized and there is no need for memory management.

We do this since this makes Yocto/GL easier to extend and quicker to learn, with a more explicit data flow that is easier when writing parallel code. Since Yocto/GL is mainly used for research and teaching, explicit data is both more hackable and easier to understand.

In terms of code style we prefer a functional approach rather than an object oriented one, favoring free functions to class methods. All functions and data are defined in sibling namespaces contained in the yocto namespace so libraries can call all others, but have to do so explicitly.

The use of templates in Yocto was the reason for many refactoring, going from no template to heavy template use. At this point, Yocto uses some templates for readability. In the future, we will increase the use of templates in math code, while keeping many APIs explicitly typed.

We do not use exception for error reporting, but only to report "programmers" errors. For example, IO operations use boolean flags and error strings for human readable errors, while exceptions are used when preconditions or post conditions are violated in functions.

The current version of the library (2.x) is a major refactoring of the previous library versions (1.x) in three main aspects. First, we now allow the use of reference semantic via pointers and adopt it for all large objects, while keeping value semantic for all others. We did this to avoid erroneous copies that cannot detected and avoided at compile time. Second, we had trouble interacting with C libraries that mostly use reference semantic. Third, we reduce the use of exceptions, again for better integration with external code.

Credits

Main contributors:

• Fabio Pellacini (lead developer): web, github
• Edoardo Carra: github
• Giacomo Nazzaro: github

This library includes code from the PCG random number generator, boost hash_combine, and public domain code from github.com/sgorsten/linalg, gist.github.com/badboy/6267743 and github.com/nothings/stb_perlin.h. Other external libraries are included with their own license.

Compilation

This library requires a C++17 compiler and is know to compiled on OsX (Xcode >= 11), Windows (MSVC 2019) and Linux (gcc >= 9, clang >= 9).

You can build the example applications using CMake with mkdir build; cd build; cmake ..; cmake --build .

Yocto/GL depends on stb_image.h, stb_image_write.h, stb_image_resize.h and tinyexr.h for image loading, saving and resizing, cgltf.h and json.hpp for glTF and JSON support, and filesystem.hpp to support C++17 filesystem API when missing. All dependencies are included in the distribution.

Yocto/GL optionally supports building OpenGL demos, which are handled by including glad, GLFW, ImGui as dependencies in apps. OpenGL support might eventually become part of the Yocto/GL libraries. OpenGL support is enabled by defining the cmake option YOCTO_OPENGL and contained in the yocto_gui library.

Yocto/GL optionally supports the use of Intel's Embree for ray casting. At this point, we rely on prebuilt binaries distributed by Intel. See the main CMake file for how to link to it. Embree support is enabled by defining the cmake option YOCTO_EMBREE.

Yocto/GL optionally supports the use of Intel's Open Image Denoise for denoising renders. At this point, we rely on prebuilt binaries distributed by Intel. See the main CMake file for how to link to it. Open Image Denoise support is enabled by defining the cmake option YOCTO_DENOISE. See apps/yimagedenoise for a demonstration.