Skip to content

Latest commit

 

History

History
51 lines (33 loc) · 2.6 KB

README.md

File metadata and controls

51 lines (33 loc) · 2.6 KB

RayTracer

Reverse Ray Tracer for rendering graphics using CPU or GPU via OpenCL.

TODO:

  • Add BVH structure support in advancedRayTracer - kernel, potentialy improving performance from n to log(n) with BVH

    • BVH for object/mesh lookup in advancedRayTracer - kernel
    • BVH for triangle lookup per object/mesh in advancedRayTracer - kernel

  • Rename "Object" struct to "Mesh"

  • Make sure there is CPU-compute support and remove depricated "DebugRayTracer"

  • Fix the window resize-functionality, currently crashing when using advancedRender

  • Add support for multiple AABBs per mesh so there will be fewer triangles(expensive) to check for intersection per AABB(cheap) to improve performance.

  • If possible, increase number of threads per Mesh in "vertex-shader"-kernel. There is currently only 1 thread / Mesh.

  • Consider removing last run of the treeTraverser kernel with a modified version of the colorToPixel kernel.

  • Consider moving AABB step into "vertex-shader" if beneficial

  • Consider adding frustum culling step after AABB step

  • Consider adding sorting step before Raytracer to improve performance

  • Consider "chaining" OpenCL kernels to free up some CPU-time and to help making other calculations on CPU without it having to launch kernel steps during rendering of image.

  • Look into what needs to be changed to support not only Windows but also Linux and Mac.

    • OpenGL - CL interop
    • get working directory for program.build()
    • system("pause")
    • etc.

  • Look into improving how the pixel color are computed. Good information at this site: http://www.flipcode.com/archives/Raytracing_Topics_Techniques-Part_2_Phong_Mirrors_and_Shadows.shtml

  • Optimize and clean up kernels

    • Add read_only,write_only,const and similar properties
    • Minimize reads and writes to global memory, for example by making a privat copy when accessing multiple times
    • Look into the possibility of optimizing traceBruteForceColor()
      • Look into if possible and beneficial to: Save the objects whose AABB were hit and only run intersectsTriangle() on the closest ones in some way.

  • Optimize and clean up host code

    • When possible/beneficial change flags for cl::Buffers to as conservative ones as possible such as CL_MEM_READ_ONLY or CL_MEM_WRITE_ONLY.
    • Remove unnecessary reads/writes to cl:Buffers

WIP

  • Add support for reflection, refrection and light source(s) in some recursive-like fashion but implemented iterative through multiple kernels calls. One kernel for tracing rays and one for producing reflection/refraction rays on impacts. Repetedly call these multiple times and then collect the result through a third kernel.