C vs Swift: Reprise

Mar 12, 2016 · 1 minute read
swift

I revisit a post by David Steuber, which benchmarked C and Swift generating a 2880x2880 pixel Mandelbrot with 4000 iterations per pixel. C killed Swift.

In C vs Swift 1.2, David concludes

The major take away on this is that Swift is not C or C++. For high performance computing, code has to be structured differently.

David acknowledges the Swift version is quite different from the C implemetation, which allocates an array for each pixel. Worse still, this array grows dynamically as it iterates 4000 times, per pixel.

Baseline

The C version will serve as our baseline. Using -Ofast¹ it completed in about 20.5 seconds.

orbitals  20.50s user 0.03s system 99% cpu 20.541 total

Swift

I restructured the Swift code so it was semantically equivalent to the C implementation, meaning that it used fixed arrays and a single core. Using the default compiler optimizations, it completed in just under 43 seconds.

orbitals-swift  42.67s user 0.12s system 100% cpu 42.768 total

Enabling -whole-module-optimization dropped the runtime to 37.5 seconds. Adding -Ounchecked reduced it a further 9 seconds to complete in 28.5 seconds.

orbitals-swift  28.35s user 0.11s system 100% cpu 28.446 total

---

1 -flto hurt performance, hence limiting to the -Ofast option ↩