Ray Tracing in One Weekend in Crystal (Lang)

This is my first crystal program; implementing the simple raytracer from this excellent book:

https://raytracing.github.io/books/RayTracingInOneWeekend.html

(version 3.2.3)

Git Branches

• main - the latest whatever I'm working on
• in-one-weekend - the code as of the end of the first book

Building & Running

$ cd rtiow
$ shards build --release
$ ./bin/inOneWeekend --help

Tasks

• Build naïve functionality, writing things basically in rubyish Crystal
• Do some of the tips from the Crystal "Performance" page
• Reorganize as a well-structured crystal app repo
• Add options processing for render parameters (and scenes?)
• Port "Raytracing: The Next Week"
• Add tests
• Add forking(?) parallelism to use more threads for parallel scanlines

Notes

I found Crystal surprisingly easy to work with. I have no point of reference for performance (I should download the C++ baseline for comparison), but it seems pretty zippy. The compiler errors are very helpful and clear.

Update: performance is amazing, particularly for how high-level the code writing can be

Mac tip:

$ crystal run --release src/main.cr -- -w 75 -h 75 -s 10 > out.ppm \
  && qlmanage -p out.ppm >/dev/null 2>&1

Performance

All times are for rendering the cover graphic:

• at 1200x675 resolution
• with 10 samples per pixel
• on a 2.3 GHz 8-core Intel Core i9 in a 2019 MacBook Pro

Results

• Baseline C++ compiled with defaults from raytracing.github.io as of 2021-02-03:

  ./build/inOneWeekend > out.ppm  401.43s user 0.42s system 99% cpu 6:42.37 total
  

• Naïve Crystal version with no structs or other code optimizations, but built with --release:

  ./inOneWeekend > out.ppm  313.17s user 677.08s system 307% cpu 5:21.90 total
  

• Replaced key class keywords with struct (5 mins work):

  ./inOneWeekend > out.ppm  41.70s user 2.08s system 99% cpu 43.894 total
  

One more data point: as above but with 500 samples-per-pixel took 34m47s