CUDA Playground

This project provides basic templates of a cuda programs that can be modified and executed at runtime by saving the cuda kernel file (with ctrl+s). It's meant to help with developing algorithms by providing a near-instant feedback loop.

Several example programs are provided, one using the console and others using OpenGL interop to draw an image with CUDA and display the resulting image with OpenGL.

CUDA Triangle Rasterizer

Main: main_rasterize_triangles.cpp
Kernel: rasterize.cu

Rasterizes 25 instances of the spot model.

First allocates bytes for an interleaved depth&color buffer
Clears it with all launched threads
Generates and draws a ground plane, colored by triangle index
Then draws the spot model with rasterizeTriangles()
- Each cuda block grabs a triangle
- Block projects it to screen
- Block computes the screen-space bounding box
- Iterates through all fragments, utilizing one block-thread per fragment
- Each thread stores the fragment via a 64bit atomicMin that encodes depth and color
If colored by normalized time, computes the max time.
Finally transfers the results from our custom buffer to an OpenGL texture

Random Numbers

Kernel: randomNumbers.cu

Generates a list of random numbers and then computes the average value. Results are printed directly by the CUDA kernel. Cooperative groups, specifically grid.sync(), are used to sync all GPU threads between different passes.

CUDA Rasterizer

Kernel: rasterize.cu

Draws a parametric plane and sphere into a custom cuda framebuffer, and then transfers the result into an OpenGL texture. At each frame, the kernel

allocates a framebuffer from a byte buffer
clears it with infinite depth and a background value
samples points on the surface, projects them to pixels
stores the closest samples by using atomicMin with a 64 bit value that contains both, depth and color.
Then extracts the color values from the interleaved depth&color buffer, and stores the result in an OpenGL texture,

CUDA Seascape

Kernel: seascape.cu

A CUDA port of the shadertoy demo "Seascape" (from Alexander Alekseev aka TDM). Instead of projecting points to pixels as the rasterizer does, this demo traces rays from pixels to compute a randering of an ocean.

Getting Started

Open build/CudaPlayground.sln in Visual Studio 2022
Specify your desired startup project by right clicking and selecting "Set as startup project".
Press ctrl+f5 to compile and run the project
Open the respective cuda file (randomNumbers.cu, rasterize.cu, seascape.cu) and try to change some code. Changes are instantly applied by saving the file.

Notes

"--gpu-architecture=compute_75" is hardcoded in CudaModularProgram.h. You may want to change this to access newer features.

Citing

This project extracts a minimal viable cuda framework from our research on software rasterization of point clouds (see github project compute_rasterizer).

@article{SCHUETZ-2022-PCC,
  title =      "Software Rasterization of 2 Billion Points in Real Time",
  author =     "Markus Sch\"{u}tz and Bernhard Kerbl and Michael Wimmer",
  year =       "2022",
  month =      jul,
  journal =    "Proc. ACM Comput. Graph. Interact. Tech.",
  volume =     "5",
  pages =      "1--16",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2022/SCHUETZ-2022-PCC/",
}

@article{SCHUETZ-2021-PCC,
  title =      "Rendering Point Clouds with Compute Shaders and Vertex Order Optimization",
  author =     "Markus Sch\"{u}tz and Bernhard Kerbl and Michael Wimmer",
  year =       "2021",
  month =      jul,
  doi =        "10.1111/cgf.14345",
  journal =    "Computer Graphics Forum",
  number =     "4",
  volume =     "40",
  pages =      "115--126",
  keywords =   "point-based rendering, compute shader, real-time rendering",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2021/SCHUETZ-2021-PCC/",
}

adakri / CudaPlayground