Somewhat related, ispc also uses the mandelbrot set as a motivating example [0], they get a 5x speedup from writing the code in almost the same way and using ispc. C compilers aren't good at autovectorizing outer loops, which is required here.

I recently learend however that intels icx c compiler, based on llvm, can now autovectorize the outer loop [1]. Upstream clang sadly still can't, same goes for gcc.

[0] https://ispc.github.io/example.html

[1] https://godbolt.org/z/jbn1robKz