shovon / raytracing-vlang

Just me practicing V by writing a CPU-bound ray-tracer

Ray Tracing in One Weekend in V

This is a ray tracing implementation in V, pretty much a re-implementation of an earlier attempt in Go

All implementation was translated from the minibook Ray Tracing In One Weekend. The book opted for C++, but I figured a good way to try out V.

Running the ray tracer

Be sure to have V installed. Then, from this directory, just run:

v . -prod
./raytracing-vlang run . > img.ppm

The program outputs portable pixmap format to the console/stdout, and so, the > symbol writes the output to a file.

Rendering a whole 1024 by 576 image takes around an 3 minutes, 10 seconds on my 2020 M1 MacBook Pro. If you want something rendered quicker, change the nx and ny variables in main(), in main.v to something smaller than 1024 by 576 (perhaps 200 and 100, respectively).

Even more optimization

You can use link-time optimization (LTO) and profiler-guided optimization (PGO) to make things even faster. I managed to shave the render time down from ~3:10, down to ~2:50, on the same 2020 M1 MacBook Pro.

Here are the steps.

1. Transpile (rather than compile) the program to C

v . -o raytracing-vlang.c

2. Compile the program, with PGO instrumentation added to the binary (NOT THE FINAL BUILD!)

clang \
  -D_DEFAULT_SOURCE \
  -DNDEBUG \
  -O3 \
  -flto \
  -fprofile-instr-generate \
  raytracing-vlang.c -o raytracing-vlang

3. Run the program a bunch of times

I ran it 4 times, but the more the merrier

LLVM_PROFILE_FILE="raytracing-vlang-%p.profraw" ./raytracing-vlang > /dev/null

4. Merge the profiler data

xcrun \
  llvm-profdata merge \
  -output=raytracing-vlang.profdata \
  raytracing-vlang-*.profraw

5. Tell clang to optimize according to whatever profiler data that was collected

clang \
 -D_DEFAULT_SOURCE \
 -DNDEBUG \
 -O3 \
 -flto \
 -fprofile-instr-use=raytracing-vlang.profdata \
 raytracing-vlang.c -o raytracing-vlang

6. Run the program like usual

./raytracing-vlang run . > img.ppm