A simple ray tracer written in pure CMake. Inspired by raytracer.hpp. More information can be found at my blog.

The ray tracer writes its output to stderr, so you can use it with:

cmake -Dimage_width=64 -Dimage_height=64 -Dnum_procs=4 -P CMakeLists.txt 2> image.ppm

Which renders using 4 CPU cores and writes the output to image.ppm. Then use an image viewer capable of opening PPM files (or this) to view.

Alternatively, the ray tracer can directly write PNG files (thanks to @benmcmorran for contributing this). The PNG encoder is also implemented in pure CMake, but either PowerShell or Python is required with this method since CMake is unable to write binary files directly. Set the use_png variable to set the name of the PNG file (without the extension). The example below will write output to render.png:

cmake -Dimage_width=64 -Dimage_height=64 -Dnum_procs=4 -Duse_png=render -P CMakeLists.txt

num_procs controls the number of worker processes spawned. It is recommended to set this to a value no greater than the number of cores in your CPU, for maximum performance. If not provided, it will be automatically detected based on the number of available cores.

To keep the code simple, the image_width, image_height and num_procs must be powers of 2, otherwise the image will not be fully formed. If not specified, these arguments default to the values shown above.

Using cmake 3.19.2 on Linux 5.4 on a i5-10210U (4 cores, 8 threads), running this command:

for X in 1 2 4 8 16 32 64 128 256 512 ; do echo SIZE $X ; time cmake -Dimage_width=$X -Dimage_height=$X -Dnum_procs=8 -P CMakeLists.txt 2> image_size_${X}.ppm  ; done

PowerShell alternative:

1, 2, 4, 8, 16, 32, 64, 128, 256, 512 | foreach {Write-Host $_; (Measure-Command { cmake "-Dimage_width=$_" "-Dimage_height=$_" "-Dnum_procs=$Env:NUMBER_OF_PROCESSORS" -P CMakeLists.txt 2> image_size_$_.ppm }).TotalSeconds}

Figures reported by time command below (reformatted). As usual, the "real" time is the wall clock time. The others are summed on all processors.

All contributions (issue, PRs) are welcome. This project is licensed under the MIT license.