Avoid floating point precision issues causing TileIterator to return

empty tiles.

webrender/src/image.rs

@@ -320,8 +320,25 @@ pub fn tiles(
         },
     );
 
-    let x_count = f32::ceil((visible_rect.max_x() - prim_rect.origin.x) / layer_tile_size.width) as u32 - t0.x;
-    let y_count = f32::ceil((visible_rect.max_y() - prim_rect.origin.y) / layer_tile_size.height) as u32 - t0.y;
+    // Since we're working in layer space, we can end up computing leftover tiles with an empty
+    // size due to floating point precision issues. Detect this case so that we don't return
+    // tiles with an empty size.
+    let x_count = {
+        let result = f32::ceil((visible_rect.max_x() - prim_rect.origin.x) / layer_tile_size.width) as u32 - t0.x;
+        if result == leftover_offset.x + 1 && leftover_layer_size.width == 0.0f32 {
+            leftover_offset.x
+        } else {
+            result
+        }
+    };
+    let y_count = {
+        let result = f32::ceil((visible_rect.max_y() - prim_rect.origin.y) / layer_tile_size.height) as u32 - t0.y;
+        if result == leftover_offset.y + 1 && leftover_layer_size.height == 0.0f32 {
+            leftover_offset.y
+        } else {
+            result
+        }
+    };
 
     let mut row_flags = EdgeAaSegmentMask::TOP;
     if y_count == 1 {