Use tiles and queue of them for rendering.

wtracer.cpp

@@ -1,11 +1,14 @@
 #include <iostream>
 #include <iomanip>
+#include <atomic>
 #include <memory>
 #include <cstdio>
 
 #define LODEPNG_NO_COMPILE_DECODER
 #include "lodepng.h"
 
+#include <boost/lockfree/queue.hpp>
+
 #ifdef _OPENMP
 #include <omp.h>
 #endif
@@ -36,11 +39,9 @@ Color ray_color(const Ray& r, const Hittable& world, int depth) {
     return (1.0 - t) * Color(1.0, 1.0, 1.0) + t * Color(0.5, 0.7, 1.0);
 }
 
-Hittable_list setup_random_scene() {
+Hittable_list setup_random_scene(const int sph_i) {
     Hittable_list world;
 
-    const int sph_i = 3;
-
     auto ground_material = std::make_shared<Lambertian>(Color(0.5, 0.5, 0.5));
     world.add(std::make_shared<Sphere>(Point3(0, -1000, 0), 1000, ground_material));
     for (int a = -sph_i; a<sph_i; a++) {
@@ -50,7 +51,7 @@ Hittable_list setup_random_scene() {
 
             if ((center - Point3(4, 0.2, 0)).length() > 0.9) {
                 std::shared_ptr<Material> sphere_material;
-                if (choose_mat < 0.8) {
+                if (choose_mat < 0.7) {
                     // diffuse
                     auto albedo = Color::random() * Color::random();
                     sphere_material = std::make_shared<Lambertian>(albedo);
@@ -92,6 +93,13 @@ struct render_params {
     Camera& cam;
 };
 
+struct render_tile {
+    int start_x;
+    int start_y;
+    int end_x;
+    int end_y;
+};
+
 void split_line(const int& line_length, int& start, int& end) {
 #ifdef _OPENMP
     const int line_length_thread = static_cast<int>(line_length/omp_get_num_threads());
@@ -103,38 +111,51 @@ void split_line(const int& line_length, int& start, int& end) {
 #endif
 }
 
-void render(const render_params& params, std::vector<Color>& image) {
+void render(boost::lockfree::queue<render_tile>& queue,
-    int line_start, line_end;
+        std::atomic<int>& queue_counter,
-    split_line(params.image_width, line_start, line_end);
+        const render_params& params, std::vector<Color>& image) {
-    for (int i = params.image_height-1; i>=0; --i) {
+    render_tile rt;
-        #ifdef _OPENMP
+    while (queue.pop(rt)) {
-        if (omp_get_thread_num() == 0)
+        --queue_counter;
-        #endif
+        int qc = queue_counter;
-        std::cerr << "\rScanline remaining: " << std::setw(4) << i << std::flush;
+        printf("\rTiles remaining: %4d", qc);
-        for (int j = line_start; j<line_end; ++j) {
+        std::cout << std::flush;
-            Color pixel_color(0, 0, 0);
+        for (int i = rt.start_x; i<rt.end_x; ++i) {
-            for (int s = 0; s<params.samples_per_pixel; ++s) {
+            for (int j = rt.start_y; j<rt.end_y; ++j) {
-                auto u = double(j + random_double(-0.5, 0.5)) / (params.image_width - 1);
+                Color pixel_color(0, 0, 0);
-                auto v = double(i + random_double(-0.5, 0.5)) / (params.image_height - 1);
+                for (int s = 0; s<params.samples_per_pixel; ++s) {
-                Ray r = params.cam.get_ray(u, v);
+                    auto u = double(j + random_double(-0.5, 0.5)) / (params.image_width - 1);
-                pixel_color += ray_color(r, params.world, params.max_depth);
+                    auto v = double(i + random_double(-0.5, 0.5)) / (params.image_height - 1);
+                    Ray r = params.cam.get_ray(u, v);
+                    pixel_color += ray_color(r, params.world, params.max_depth);
+                }
+                image[i*params.image_width+j] = pixel_color;
             }
-            image[i*params.image_width+j] = pixel_color;
         }
     }
 }
 
 int main() {
     const auto aspect_ratio = 16.0 / 9.0;
+    //const int image_width = 1920;
     //const int image_width = 1280;
     //const int image_width = 768;
     const int image_width = 384;
     const int image_height = static_cast<int>(image_width / aspect_ratio);
+    //const int samples_per_pixel = 2000;
     //const int samples_per_pixel = 1000;
     const int samples_per_pixel = 400;
     const int max_depth = 50;
 
-    auto world = setup_random_scene();
+    //const int tile_size = 128;
+    const int tile_size = 64;
+
+    const int sph_i = 3;
+    //const int sph_i = 5;
+    //const int sph_i = 7;
+    //const int sph_i = 11;
+
+    auto world = setup_random_scene(sph_i);
 
     Point3 lookfrom(13, 2, 3);
     Point3 lookat(0, 0, 0);
@@ -153,11 +174,27 @@ int main() {
         cam,
     };
 
+    std::atomic<int> queue_counter;
+    boost::lockfree::queue<render_tile> queue(0);
+
+    for (int i = 0; i<std::ceil(double(image_height)/double(tile_size)); i++) {
+        for (int j = 0; j<std::ceil(double(image_width)/double(tile_size)); j++) {
+            render_tile rt = {
+                i*tile_size,
+                j*tile_size,
+                std::min((i+1)*tile_size, image_height),
+                std::min((j+1)*tile_size, image_width)
+            };
+            queue.push(rt);
+            ++queue_counter;
+        }
+    }
+
     std::vector<Color> image(image_width * image_height);
 
     #pragma omp parallel
     {
-        render(params, image);
+        render(queue, queue_counter, params, image);
     }
 
     std::cerr << "\nAssembling image.\n";