Add missing barrier

Add barrier for write-after-read hazard in coarse. The loop in question processes 64k draw objects at a time, so the barrier only gets invoked when that limit is exceeded.

Also move new test scene so it isn't the first.

examples/scenes/src/test_scenes.rs

diff --git a/examples/scenes/src/test_scenes.rs b/examples/scenes/src/test_scenes.rs
index fa6f717..2e36a39 100644
--- a/examples/scenes/src/test_scenes.rs
+++ b/examples/scenes/src/test_scenes.rs
@@ -33,7 +33,6 @@
 
 pub fn test_scenes() -> SceneSet {
     let scenes = vec![
-        scene!(many_draw),
         scene!(splash_with_tiger(), "splash_with_tiger", false),
         scene!(funky_paths),
         scene!(stroke_styles(Affine::IDENTITY), "stroke_styles", false),
@@ -63,6 +62,7 @@
         scene!(longpathdash(Cap::Butt), "longpathdash (butt caps)", false),
         scene!(longpathdash(Cap::Round), "longpathdash (round caps)", false),
         scene!(crate::mmark::MMark::new(80_000), "mmark", false),
+        scene!(many_draw_objects),
     ];
 
     SceneSet { scenes }
@@ -70,21 +70,6 @@
 
 // Scenes
 
-fn many_draw(scene: &mut Scene, _: &mut SceneParams) {
-    const N_WIDE: usize = 300;
-    const N_HIGH: usize = 300;
-    const SCENE_WIDTH: f64 = 2000.0;
-    const SCENE_HEIGHT: f64 = 1500.0;
-    for j in 0..N_HIGH {
-        let y = (j as f64 + 0.5) * (SCENE_HEIGHT / N_HIGH as f64);
-        for i in 0..N_WIDE {
-            let x = (i as f64 + 0.5) * (SCENE_WIDTH / N_WIDE as f64);
-            let c = Circle::new((x, y), 3.0);
-            scene.fill(Fill::NonZero, Affine::IDENTITY, Color::YELLOW, None, &c);
-        }
-    }
-}
-
 fn funky_paths(scene: &mut Scene, _: &mut SceneParams) {
     use PathEl::*;
     let missing_movetos = [
@@ -1538,6 +1523,21 @@
     ]
 }
 
+fn many_draw_objects(scene: &mut Scene, _: &mut SceneParams) {
+    const N_WIDE: usize = 300;
+    const N_HIGH: usize = 300;
+    const SCENE_WIDTH: f64 = 2000.0;
+    const SCENE_HEIGHT: f64 = 1500.0;
+    for j in 0..N_HIGH {
+        let y = (j as f64 + 0.5) * (SCENE_HEIGHT / N_HIGH as f64);
+        for i in 0..N_WIDE {
+            let x = (i as f64 + 0.5) * (SCENE_WIDTH / N_WIDE as f64);
+            let c = Circle::new((x, y), 3.0);
+            scene.fill(Fill::NonZero, Affine::IDENTITY, Color::YELLOW, None, &c);
+        }
+    }
+}
+
 fn splash_screen(scene: &mut Scene, params: &mut SceneParams) {
     let strings = [
         "Vello test",

shader/coarse.wgsl

diff --git a/shader/coarse.wgsl b/shader/coarse.wgsl
index 6caa35a..758244c 100644
--- a/shader/coarse.wgsl
+++ b/shader/coarse.wgsl
@@ -236,6 +236,7 @@
             if wr_ix - rd_ix >= N_TILE || (wr_ix >= ready_ix && partition_ix >= n_partitions) {
                 break;
             }
+            workgroupBarrier();
         }
         // At this point, sh_drawobj_ix[0.. wr_ix - rd_ix] contains merged binning results.
         var tag = DRAWTAG_NOP;