From 25ee3cfafea166f2dd155c07b5d43ba90d5bd994 Mon Sep 17 00:00:00 2001
From: Keuin <keuinx@gmail.com>
Date: Wed, 13 Apr 2022 16:25:17 +0800
Subject: Threaded antialiasing.

---
 aa.h | 48 +++++++++++++++++++++++++++++++++++-------------
 1 file changed, 35 insertions(+), 13 deletions(-)

diff --git a/aa.h b/aa.h
index d226fda..368a14a 100644
--- a/aa.h
+++ b/aa.h
@@ -9,33 +9,55 @@
 #include "viewport.h"
 #include <vector>
 #include <thread>
-
+#include <algorithm>
 
 // Antialiasing viewport
 template<typename T>
 class aa_viewport : public viewport<T> {
     unsigned samples;
-    std::vector<basic_viewport<T>> subviews;
+    std::vector<basic_viewport<T>> *subviews;
 
 public:
     aa_viewport(double width, double height, vec3d viewport_center, unsigned samples) : samples(samples) {
         assert(samples >= 1);
-        subviews = std::vector<basic_viewport<T>>{samples, {width, height, viewport_center}};
+        subviews = new std::vector<basic_viewport<T>>{samples, {width, height, viewport_center}};
+    }
+
+    ~aa_viewport() {
+        delete subviews;
     }
 
     virtual bitmap<T> render(const hitlist<T> &world, vec3d viewpoint, uint16_t image_width, uint16_t image_height) {
-//        // TODO threaded rendering
-//        std::vector<std::thread> workers{samples};
-//        for (size_t i = 0; i < samples; ++i) {
-//            const auto &th = workers[i];
-//            const auto &vw = subviews[i];
-//        }
+        const unsigned hwcc = std::thread::hardware_concurrency();
+        std::cerr << "Rendering with " << hwcc << " thread(s)." << std::endl;
+
         const auto seed = 123456789012345678ULL;
         std::mt19937_64 seedgen{seed}; // generates seeds for workers
-        std::vector<bitmap<T>> images{};
-        for (const auto &view: subviews) {
-            bias_ctx bc{seedgen()};
-            images.push_back(view.render(world, viewpoint, image_width, image_height, bc));
+        std::vector<bitmap<T>> images{samples, {1, 1}};
+        std::thread t;
+
+        std::vector<std::thread> workers;
+        unsigned remaining = samples; // tasks remaining
+        size_t base = 0;
+        while (remaining > 0) {
+            const unsigned n = std::min(hwcc, remaining); // threads in current batch
+            remaining -= n;
+            for (unsigned i = 0; i < n; ++i) {
+                workers.emplace_back(std::thread{
+                        [&](int tid, uint64_t seed, std::vector<basic_viewport<T>> *subs, vec3d viewpoint,
+                            uint16_t image_width, uint16_t image_height) {
+                            bias_ctx bc{seed};
+                            auto image = (*subs)[tid].render(world, viewpoint, image_width, image_height, bc);
+                            images[base + tid] = image;
+                        },
+                        i, seedgen(), subviews, viewpoint, image_width, image_height
+                });
+            }
+            for (auto &th: workers) {
+                th.join();
+            }
+            workers.clear();
+            base += n;
         }
         return bitmap<T>::average(images);
     }
-- 
cgit v1.2.3