Fix error in clip reduction (#659)

One bit of logic from the GLSL (piet-gpu) version of clip_reduce didn't
get copied over.

Fixes #648

examples/scenes/src/test_scenes.rs

@@ -68,6 +68,7 @@ export_scenes!(
     brush_transform(brush_transform: animated),
     blend_grid(blend_grid),
     deep_blend(deep_blend),
+    many_clips(many_clips),
     conflation_artifacts(conflation_artifacts),
     labyrinth(labyrinth),
     robust_paths(robust_paths),
@@ -82,7 +83,11 @@ export_scenes!(
 /// Implementations for the test scenes.
 /// In a module because the exported [`ExampleScene`](crate::ExampleScene) creation functions use the same names.
 mod impls {
+    use std::f64::consts::PI;
+
     use crate::SceneParams;
+    use rand::Rng;
+    use rand::{rngs::StdRng, SeedableRng};
     use vello::kurbo::{
         Affine, BezPath, Cap, Circle, Ellipse, Join, PathEl, Point, Rect, Shape, Stroke, Vec2,
     };
@@ -1094,6 +1099,32 @@ mod impls {
         }
     }
 
+    pub(super) fn many_clips(scene: &mut Scene, params: &mut SceneParams) {
+        params.resolution = Some(Vec2::new(1000., 1000.));
+        let mut rng = StdRng::seed_from_u64(42);
+        let mut base_tri = BezPath::new();
+        base_tri.move_to((-50.0, 0.0));
+        base_tri.line_to((25.0, -43.3));
+        base_tri.line_to((25.0, 43.3));
+        for y in 0..10 {
+            for x in 0..10 {
+                let translate =
+                    Affine::translate((100. * (x as f64 + 0.5), 100. * (y as f64 + 0.5)));
+                const CLIPS_PER_FILL: usize = 3;
+                for _ in 0..CLIPS_PER_FILL {
+                    let rot = Affine::rotate(rng.gen_range(0.0..PI));
+                    scene.push_layer(Mix::Clip, 1.0, translate * rot, &base_tri);
+                }
+                let rot = Affine::rotate(rng.gen_range(0.0..PI));
+                let color = Color::rgb(rng.gen(), rng.gen(), rng.gen());
+                scene.fill(Fill::NonZero, translate * rot, color, None, &base_tri);
+                for _ in 0..CLIPS_PER_FILL {
+                    scene.pop_layer();
+                }
+            }
+        }
+    }
+
     // Support functions
 
     pub(super) fn render_cardioid(scene: &mut Scene) {

vello_shaders/shader/clip_reduce.wgsl

@@ -47,6 +47,9 @@ fn main(
     workgroupBarrier();
     let size = sh_bic[0].b;
     bic = Bic();
+    if local_id.x + 1u < WG_SIZE {
+        bic = sh_bic[local_id.x + 1u];
+    }
     if is_push && bic.a == 0u {
         let local_ix = size - bic.b - 1u;
         sh_parent[local_ix] = local_id.x;

vello_shaders/src/cpu/clip_reduce.rs

@@ -24,10 +24,10 @@ fn clip_reduce_main(
             let inp = clip_inp[global_ix].path_ix;
             let is_push = inp >= 0;
             let bic = ClipBic::new(1 - is_push as u32, is_push as u32);
-            bic_reduced = bic.combine(bic_reduced);
             if is_push && bic_reduced.a == 0 {
                 scratch.push(global_ix as u32);
             }
+            bic_reduced = bic.combine(bic_reduced);
         }
         reduced[wg_ix as usize] = bic_reduced;
         for (i, parent_ix) in scratch.iter().rev().enumerate() {

vello_tests/tests/snapshots.rs

@@ -82,3 +82,11 @@ fn snapshot_deep_blend() {
     let params = TestParams::new("deep_blend", 200, 200);
     snapshot_test_scene(test_scene, params);
 }
+
+#[test]
+#[cfg_attr(skip_gpu_tests, ignore)]
+fn snapshot_many_clips() {
+    let test_scene = test_scenes::many_clips();
+    let params = TestParams::new("many_clips", 200, 200);
+    snapshot_test_scene(test_scene, params);
+}