RaytracerChallenge

Progress

1. Tuples, Points, and Vectors
2. Drawing on a canvas
3. Matrices
4. Matrix Transformations (WIP)
   1. Translation
   2. Scaling
   3. Rotation
   4. Shearing
   5. Chaining Transformations
   6. Putting it Together
   7. All tests
5. Ray-Sphere Intersections
6. Light and Shading
7. Making a Scene
8. Shadows
9. Planes
10. Patterns
11. Reflection and Refraction
12. Cubes
13. Cylinders
14. Groups
15. Triangles
16. Constructive Solid Geometry (CSG)
17. Next Steps

TODO

1. Replace MVector with non heap-alloc'd array types. Consider
   • SVector (disadvantage: immutable, in-place modification not possible) Consider trying this in another source file and writing benchmarks to compare the two
   • SIMD.jl vector types (*(::SMatrix{4,4,T}, ::Vec{4, T}) needs to be implemented).
2. Create benchmarks
3. Refactor tests
4. Make README.md more presentable

Ideas

• Implement an SoA type for storing Vec3s and Points whose interface is similar to AoS (e.g. a regular Vector{Vec3}).
• Multithreading
  • Load-balancing
• GPU acceleration
  • CUDA kernels
  • Vendor-agnostic kernels

Choice of data type for Vec3 and Point

Heap-alloc'd data types are out of the question. While I am concerned about the immutability of SVectors, I do not know for certain if mutating them is a common case or not. If "Ray Tracing in One Weekend" is anything to go by, it most definitely is as we compute a running average. The alternative would be the a much more memory-heavy allocation of an array of Vectors of and subsequent reduction.

Note to self: When in doubt, just measure and benchmark.