InOneWeekend: multi-threaded camera render

[dimitry-ishenko]

This is not for the faint-hearted, but the camera rendering function can be made multi-threaded by replacing inner-most loop:

color pixel_color(0,0,0);
for (int sample = 0; sample < samples_per_pixel; ++sample) {
    ray r = get_ray(i, j);
    pixel_color += ray_color(r, max_depth, world);
}

with parallel version of the std::transform_reduce() function:

const std::vector<int> count(samples_per_pixel);
color pixel_color = std::transform_reduce(std::execution::par_unseq, count.begin(), count.end(), color(0,0,0), std::plus(), [=, &world](auto) {
    ray r = get_ray(i, j);
    return ray_color(r, max_depth, world);
});

You will need to include a few additional headers at the top:

#include <execution>
#include <numeric>
#include <vector>

And, on Linux you will need to install and link against TBB, ie:

g++ -std=c++17 main.cc -ltbb

You will also see more significant improvement if you increase samples_per_pixel to about 50. On my machine I was able to cut render time in half using this method.

Share and enjoy.