Move multiple-object logic out of BVH4.

This allows each part of Psychopath tp handle the logic in the
best way, instead of a one-size-fits-all approach.

src/accel/bvh4.rs

@@ -96,14 +96,9 @@ impl<'a> BVH4<'a> {
         self.depth
     }
 
-    pub fn traverse<T, F>(
-        &self,
-        rays: &mut RayBatch,
-        ray_stack: &mut RayStack,
-        objects: &[T],
-        mut obj_ray_test: F,
-    ) where
-        F: FnMut(&T, &mut RayBatch, &mut RayStack),
+    pub fn traverse<F>(&self, rays: &mut RayBatch, ray_stack: &mut RayStack, mut obj_ray_test: F)
+    where
+        F: FnMut(std::ops::Range<usize>, &mut RayBatch, &mut RayStack),
     {
         if self.root.is_none() {
             return;
@@ -170,16 +165,8 @@ impl<'a> BVH4<'a> {
                 &BVH4Node::Leaf { object_range } => {
                     trav_time += timer.tick() as f64;
 
-                    // Set up the tasks for each object.
-                    let obj_count = object_range.1 - object_range.0;
-                    for _ in 0..(obj_count - 1) {
-                        ray_stack.duplicate_next_task();
-                    }
-
                     // Do the ray tests.
-                    for obj in &objects[object_range.0..object_range.1] {
-                        obj_ray_test(obj, rays, ray_stack);
-                    }
+                    obj_ray_test(object_range.0..object_range.1, rays, ray_stack);
 
                     timer.tick();
 

src/ray.rs

@@ -292,24 +292,37 @@ impl RayStack {
         self.lanes[l].end_len = self.lanes[l].idxs.len();
     }
 
-    /// Pops the next task off the stack, and executes the provided closure for
-    /// each ray index in the task.
-    pub fn pop_do_next_task<F>(&mut self, mut handle_ray: F)
+    // Pops the next task off the stack.
+    pub fn pop_task(&mut self) {
+        let task = self.tasks.pop().unwrap();
+        self.lanes[task.lane].end_len = task.start_idx;
+        self.lanes[task.lane].idxs.truncate(task.start_idx);
+    }
+
+    // Executes a task without popping it from the task stack.
+    pub fn do_next_task<F>(&mut self, mut handle_ray: F)
     where
         F: FnMut(usize),
     {
-        // Pop the task and do necessary bookkeeping.
-        let task = self.tasks.pop().unwrap();
+        let task = self.tasks.last().unwrap();
         let task_range = (task.start_idx, self.lanes[task.lane].end_len);
-        self.lanes[task.lane].end_len = task.start_idx;
 
         // Execute task.
         for i in task_range.0..task_range.1 {
             let ray_idx = self.lanes[task.lane].idxs[i];
             handle_ray(ray_idx as usize);
         }
+    }
 
-        self.lanes[task.lane].idxs.truncate(task_range.0);
+    /// Pops the next task off the stack, and executes the provided closure for
+    /// each ray index in the task.
+    #[inline(always)]
+    pub fn pop_do_next_task<F>(&mut self, handle_ray: F)
+    where
+        F: FnMut(usize),
+    {
+        self.do_next_task(handle_ray);
+        self.pop_task();
     }
 
     /// Pops the next task off the stack, executes the provided closure for

src/surface/triangle_mesh.rs

@@ -130,11 +130,11 @@ impl<'a> Surface for TriangleMesh<'a> {
             Matrix4x4::new()
         };
 
-        self.accel.traverse(
-            rays,
-            ray_stack,
-            self.indices,
-            |tri_indices, rays, ray_stack| {
+        self.accel
+            .traverse(rays, ray_stack, |idx_range, rays, ray_stack| {
+                for tri_idx in idx_range {
+                    let tri_indices = self.indices[tri_idx];
+
                     // For static triangles with static transforms, cache them.
                     let is_cached = self.time_sample_count == 1 && space.len() <= 1;
                     let mut tri = if is_cached {
@@ -157,7 +157,7 @@ impl<'a> Surface for TriangleMesh<'a> {
                     };
 
                     // Test each ray against the current triangle.
-                ray_stack.pop_do_next_task(|ray_idx| {
+                    ray_stack.do_next_task(|ray_idx| {
                         let ray_idx = ray_idx as usize;
                         let ray_time = rays.time(ray_idx);
 
@@ -276,7 +276,8 @@ impl<'a> Surface for TriangleMesh<'a> {
                             }
                         }
                     });
-            },
-        );
+                }
+                ray_stack.pop_task();
+            });
     }
 }

src/tracer.rs

@@ -85,11 +85,11 @@ impl<'a> TracerInner<'a> {
         rays: &mut RayBatch,
         ray_stack: &mut RayStack,
     ) {
-        assembly.object_accel.traverse(
-            rays,
-            ray_stack,
-            &assembly.instances[..],
-            |inst, rays, ray_stack| {
+        assembly
+            .object_accel
+            .traverse(rays, ray_stack, |idx_range, rays, ray_stack| {
+                let inst = &assembly.instances[idx_range.start];
+
                 // Transform rays if needed
                 if let Some((xstart, xend)) = inst.transform_indices {
                     // Push transforms to stack
@@ -98,12 +98,11 @@ impl<'a> TracerInner<'a> {
                     // Do transforms
                     // TODO: re-divide rays based on direction (maybe?).
                     let xforms = self.xform_stack.top();
-                    ray_stack.pop_do_next_task_and_push_rays(2, |ray_idx| {
+                    ray_stack.do_next_task(|ray_idx| {
                         let t = rays.time(ray_idx);
                         rays.update_local(ray_idx, &lerp_slice(xforms, t));
-                        (Bool4::new(true, true, false, false), 2)
                     });
-                    ray_stack.push_lanes_to_tasks(&[0, 1]);
+                    ray_stack.duplicate_next_task();
                 }
 
                 // Trace rays
@@ -142,8 +141,7 @@ impl<'a> TracerInner<'a> {
                         });
                     }
                 }
-            },
-        );
+            });
     }
 
     fn trace_object<'b>(