Perform per-pixel parallelism instead of per frame.

This sets up the renderer to have other sampling algorithms.
Progress bar is now percentage based.

rtiow/src/renderer.rs

@@ -177,21 +177,38 @@ fn trace_pixel(x: usize, y: usize, scene: &Scene) -> Vec3 {
     )
 }
 
+struct PixelRequest {
+    x: usize,
+    y: usize,
+}
+
+struct PixelResponse {
+    x: usize,
+    y: usize,
+    pixel: Vec3,
+}
+
 fn render_worker(
     tid: usize,
     scene: &Scene,
-    input_chan: channel::Receiver<usize>,
-    output_chan: &channel::Sender<(usize, Vec<Vec3>)>,
+    input_chan: channel::Receiver<PixelRequest>,
+    output_chan: &channel::Sender<PixelResponse>,
 ) {
-    for subsample in input_chan {
-        let mut pixel_data: Vec<Vec3> = Vec::with_capacity(scene.width * scene.height);
-        for y in 0..scene.height {
-            for x in 0..scene.width {
-                let p = trace_pixel(x, y, scene);
-                pixel_data.push(p);
-            }
+    for req in input_chan {
+        let mut pixel: Vec3 = Default::default();
+        for _ in 0..scene.subsamples {
+            pixel = pixel + trace_pixel(req.x, req.y, scene);
         }
-        output_chan.send((subsample, pixel_data));
+        pixel = pixel / scene.subsamples as f32;
+        // Gamma correct, use gamma 2 correction, which is 1/gamma where gamma=2 which is 1/2 or
+        // sqrt.
+        let pixel = Vec3::new(pixel[0].sqrt(), pixel[1].sqrt(), pixel[2].sqrt());
+
+        output_chan.send(PixelResponse {
+            x: req.x,
+            y: req.y,
+            pixel,
+        });
     }
     trace!(target: "renderer", "Shutting down worker {}", tid);
 }
@@ -201,65 +218,52 @@ pub fn render(
     output_dir: &Path,
     store_intermediate: bool,
 ) -> std::result::Result<(), std::io::Error> {
-    let (seq_tx, seq_rx) = channel::unbounded();
-    let (pixel_data_tx, pixel_data_rx) = channel::unbounded();
+    let (pixel_req_tx, pixel_req_rx) = channel::unbounded();
+    let (pixel_resp_tx, pixel_resp_rx) = channel::unbounded();
 
     let scene = sync::Arc::new(scene);
     for i in 0..num_cpus::get() {
         let s = sync::Arc::clone(&scene);
-        let seq_rx = seq_rx.clone();
-        let pixel_data_tx = pixel_data_tx.clone();
+        let pixel_req_rx = pixel_req_rx.clone();
+        let pixel_resp_tx = pixel_resp_tx.clone();
         thread::spawn(move || {
-            render_worker(i, &s, seq_rx, &pixel_data_tx);
+            render_worker(i, &s, pixel_req_rx, &pixel_resp_tx);
         });
     }
-    drop(seq_rx);
-    drop(pixel_data_tx);
+    drop(pixel_req_rx);
+    drop(pixel_resp_tx);
 
-    (1..=scene.subsamples).for_each(|idx| seq_tx.send(idx));
-    drop(seq_tx);
-
-    let mut acc_count = 0;
-    let mut acc: Vec<Vec3> = Vec::with_capacity(scene.width * scene.height);
-    for _ in 0..(scene.width * scene.height) {
-        acc.push(Default::default());
+    for y in 0..scene.height {
+        for x in 0..scene.width {
+            pixel_req_tx.send(PixelRequest { x, y });
+        }
     }
+    drop(pixel_req_tx);
 
     println!("Rendering with {} subsamples", scene.subsamples);
     let mut img = RgbImage::new(scene.width as u32, scene.height as u32);
-    for (_subsample, pixel_data) in pixel_data_rx {
-        acc_count += 1;
-        pixel_data.iter().enumerate().for_each(|(idx, p)| {
-            let x = idx % scene.width;
-            let y = idx / scene.width;
-            let y_inv = scene.height - y - 1;
-            acc[idx] = acc[idx] + *p;
-
-            // Gamma correct, use gamma 2 correction, which is 1/gamma where gamma=2 which is 1/2 or
-            // sqrt.
-            let col = acc[idx] / acc_count as f32;
-            let col = Vec3::new(col[0].sqrt(), col[1].sqrt(), col[2].sqrt());
-            img.put_pixel(
-                x as u32,
-                y_inv as u32,
-                image::Rgb([
-                    (col[0] * 255.).min(255.) as u8,
-                    (col[1] * 255.).min(255.) as u8,
-                    (col[2] * 255.).min(255.) as u8,
-                ]),
-            );
-        });
-        if acc_count % 10 == 0 {
-            print!("{}", acc_count);
-        } else {
-            print!(".");
-        }
-        io::stdout().flush().unwrap();
-        if store_intermediate {
-            let path = output_dir.join(format!("iteration{:05}.png", acc_count));
-            trace!(target: "renderer", "Saving {}", path.display());
-            img.save(&path)
-                .unwrap_or_else(|_| panic!("Failed save {}", path.display()));
+    let total = scene.width * scene.height;
+    let mut last_progress = 1000;
+    for (i, resp) in pixel_resp_rx.enumerate() {
+        let y_inv = scene.height - resp.y - 1;
+        img.put_pixel(
+            resp.x as u32,
+            y_inv as u32,
+            image::Rgb([
+                (resp.pixel[0] * 255.).min(255.) as u8,
+                (resp.pixel[1] * 255.).min(255.) as u8,
+                (resp.pixel[2] * 255.).min(255.) as u8,
+            ]),
+        );
+        let progress = 100 * i / total;
+        if progress != last_progress {
+            last_progress = progress;
+            if progress % 10 == 0 {
+                print!("{}%", progress);
+            } else {
+                print!(".");
+            }
+            io::stdout().flush().unwrap();
         }
     }
     println!();