diff --git a/frontend/src/debug.cljs b/frontend/src/debug.cljs
index 65ac296895..4ae8be8b17 100644
--- a/frontend/src/debug.cljs
+++ b/frontend/src/debug.cljs
@@ -135,6 +135,14 @@
       (wasm.mem/free)
       text)))
 
+(defn ^:export wasmRenderStats
+  []
+  (let [module wasm/internal-module
+        f      (when module (unchecked-get module "_render_stats"))]
+    (if (fn? f)
+      (wasm.h/call module "_render_stats")
+      (js/console.warn "[debug] render-wasm module not ready or missing _render_stats"))))
+
 (defn ^:export wasmAtlasConsole
   "Logs the current render-wasm atlas as an image in the JS console (if present)."
   []
diff --git a/render-wasm/src/main.rs b/render-wasm/src/main.rs
index 9f298c3900..3ab9853005 100644
--- a/render-wasm/src/main.rs
+++ b/render-wasm/src/main.rs
@@ -1035,6 +1035,13 @@ pub extern "C" fn render_shape_pixels(
     })
 }
 
+#[no_mangle]
+pub extern "C" fn render_stats() {
+    with_state!(state, {
+        state.render_state.print_stats();
+    })
+}
+
 fn main() {
     #[cfg(target_arch = "wasm32")]
     init_gl!();
diff --git a/render-wasm/src/render.rs b/render-wasm/src/render.rs
index 6272d8d9a3..52da6d8344 100644
--- a/render-wasm/src/render.rs
+++ b/render-wasm/src/render.rs
@@ -16,6 +16,7 @@ mod ui;
 use skia_safe::{self as skia, Matrix, RRect, Rect};
 use std::borrow::Cow;
 use std::collections::HashSet;
+use std::collections::HashMap;
 
 use gpu_state::GpuState;
 
@@ -289,9 +290,46 @@ fn sort_z_index(tree: ShapesPoolRef, element: &Shape, children_ids: Vec<Uuid>) -
     }
 }
 
+struct RenderStats {
+    pub counts: HashMap<Uuid, i32>,
+}
+
+impl RenderStats {
+    pub fn new() -> Self {
+        Self {
+            counts: HashMap::new()
+        }
+    }
+
+    fn count(&mut self, id: Uuid) -> i32 {
+        let counter = self.counts.entry(id).or_insert(0);
+        *counter += 1;
+        *counter
+    }
+
+    fn clear(&mut self) {
+        self.counts.clear();
+    }
+
+    #[allow(dead_code)]
+    fn get(&self, id: &Uuid) -> Option<&i32> {
+        self.counts.get(id)
+    }
+
+    fn print(&self) {
+        let mut sum: i32 = 0;
+        for (&id, &count) in self.counts.iter() {
+            println!("{}: {}", id, count);
+            sum += count;
+        }
+        println!("{}: {}", self.counts.len(), sum);
+    }
+}
+
 pub(crate) struct RenderState {
     gpu_state: GpuState,
     pub options: RenderOptions,
+    stats: RenderStats,
     pub surfaces: Surfaces,
     pub fonts: FontStore,
     pub viewbox: Viewbox,
@@ -377,6 +415,7 @@ impl RenderState {
         Ok(RenderState {
             gpu_state: gpu_state.clone(),
             options,
+            stats: RenderStats::new(),
             surfaces,
             fonts,
             viewbox,
@@ -851,6 +890,8 @@ impl RenderState {
         outset: Option<f32>,
         target_surface: SurfaceId,
     ) -> Result<()> {
+        self.stats.count(shape.id);
+
         let surface_ids = fills_surface_id as u32
             | strokes_surface_id as u32
             | innershadows_surface_id as u32
@@ -1662,6 +1703,7 @@ impl RenderState {
         timestamp: i32,
         sync_render: bool,
     ) -> Result<()> {
+        self.stats.clear();
         let _start = performance::begin_timed_log!("start_render_loop");
         let scale = self.get_scale();
 
@@ -3338,6 +3380,10 @@ impl RenderState {
     pub fn set_view(&mut self, zoom: f32, x: f32, y: f32) {
         self.viewbox.set_all(zoom, x, y);
     }
+
+    pub fn print_stats(&self) {
+        self.stats.print();
+    }
 }
 
 impl Drop for RenderState {
diff --git a/render-wasm/src/render/gpu_state.rs b/render-wasm/src/render/gpu_state.rs
index aa62b83817..e1a1fbc02f 100644
--- a/render-wasm/src/render/gpu_state.rs
+++ b/render-wasm/src/render/gpu_state.rs
@@ -1,5 +1,5 @@
 use crate::error::{Error, Result};
-use skia_safe::gpu::{self, gl::FramebufferInfo, gl::TextureInfo, DirectContext};
+use skia_safe::gpu::{self, ganesh::context_options::Enable, gl::FramebufferInfo, gl::TextureInfo, DirectContext, ContextOptions};
 use skia_safe::{self as skia, ISize};
 
 #[derive(Debug, Clone)]
@@ -13,7 +13,18 @@ impl GpuState {
         let interface = gpu::gl::Interface::new_native().ok_or(Error::CriticalError(
             "Failed to create GL interface".to_string(),
         ))?;
-        let context = gpu::direct_contexts::make_gl(interface, None).ok_or(
+
+        // We tweak some options to enhance performance.
+        let mut context_options = ContextOptions::default();
+        println!("context_options {:?}", context_options);
+        // context_options.reduce_ops_task_splitting = Enable::Yes;
+        context_options.skip_gl_error_checks = Enable::Yes;
+        // context_options.runtime_program_cache_size = 1024;
+        // context_options.allow_multiple_glyph_cache_textures = Enable::Yes;
+        // context_options.allow_path_mask_caching = false;
+        println!("context_options {:?}", context_options);
+
+        let context = gpu::direct_contexts::make_gl(interface, Some(&context_options)).ok_or(
             Error::CriticalError("Failed to create GL context".to_string()),
         )?;
         let framebuffer_info = {
diff --git a/render-wasm/src/render/options.rs b/render-wasm/src/render/options.rs
index 40a3125ccd..81cf2da36d 100644
--- a/render-wasm/src/render/options.rs
+++ b/render-wasm/src/render/options.rs
@@ -7,7 +7,7 @@ const SHOW_WASM_INFO: u32 = 0x08;
 // Render performance options
 // This is the extra area used for tile rendering (tiles beyond viewport).
 // Higher values pre-render more tiles, reducing empty squares during pan but using more memory.
-const VIEWPORT_INTEREST_AREA_THRESHOLD: i32 = 3;
+const VIEWPORT_INTEREST_AREA_THRESHOLD: i32 = 1;
 const MAX_BLOCKING_TIME_MS: i32 = 32;
 const NODE_BATCH_THRESHOLD: i32 = 3;
 const BLUR_DOWNSCALE_THRESHOLD: f32 = 8.0;
diff --git a/render-wasm/src/render/strokes.rs b/render-wasm/src/render/strokes.rs
index c5a3a26bf1..0a1baa4ed2 100644
--- a/render-wasm/src/render/strokes.rs
+++ b/render-wasm/src/render/strokes.rs
@@ -10,6 +10,9 @@ use crate::error::{Error, Result};
 use crate::render::filters::compose_filters;
 use crate::render::{get_dest_rect, get_source_rect};
 
+#[cfg(target_arch = "wasm32")]
+use crate::run_script;
+
 #[allow(clippy::too_many_arguments)]
 fn draw_stroke_on_rect(
     canvas: &skia::Canvas,
@@ -229,21 +232,21 @@ fn draw_stroke_on_path(
     if let Some(pt) = path_transform {
         canvas.concat(pt);
     }
-    let skia_path = path.to_skia_path();
 
+    let skia_path = path.to_skia_path();
     match stroke.render_kind(is_open) {
         StrokeKind::Inner => {
-            draw_inner_stroke_path(canvas, &skia_path, &draw_paint, blur, antialias);
+            draw_inner_stroke_path(canvas, skia_path, &draw_paint, blur, antialias);
         }
         StrokeKind::Center => {
-            canvas.draw_path(&skia_path, &draw_paint);
+            canvas.draw_path(skia_path, &draw_paint);
         }
         StrokeKind::Outer => {
-            draw_outer_stroke_path(canvas, &skia_path, &draw_paint, blur, antialias);
+            draw_outer_stroke_path(canvas, skia_path, &draw_paint, blur, antialias);
         }
     }
 
-    handle_stroke_caps(&skia_path, stroke, canvas, is_open, paint, blur, antialias);
+    handle_stroke_caps(skia_path, stroke, canvas, is_open, paint, blur, antialias);
 
     canvas.restore_to_count(save_count);
 }
diff --git a/render-wasm/src/render/surfaces.rs b/render-wasm/src/render/surfaces.rs
index ca7f2a3ef2..1524f73df1 100644
--- a/render-wasm/src/render/surfaces.rs
+++ b/render-wasm/src/render/surfaces.rs
@@ -463,6 +463,7 @@ impl Surfaces {
     }
 
     pub fn flush_and_submit(&mut self, gpu_state: &mut GpuState, id: SurfaceId) {
+        println!("flush_and_submit");
         let surface = self.get_mut(id);
         gpu_state.context.flush_and_submit_surface(surface, None);
     }
diff --git a/render-wasm/src/shapes.rs b/render-wasm/src/shapes.rs
index 9e17de41ee..cd05bcb46d 100644
--- a/render-wasm/src/shapes.rs
+++ b/render-wasm/src/shapes.rs
@@ -2,6 +2,9 @@ use skia_safe::{self as skia};
 
 use indexmap::IndexSet;
 
+#[cfg(target_arch = "wasm32")]
+use crate::run_script;
+
 use crate::uuid::Uuid;
 use std::borrow::Cow;
 use std::cell::{OnceCell, RefCell};
@@ -1349,7 +1352,7 @@ impl Shape {
 
     pub fn get_skia_path(&self) -> Option<skia::Path> {
         if let Some(path) = self.shape_type.path() {
-            let mut skia_path = path.to_skia_path();
+            let mut skia_path = path.to_skia_path().clone();
             if let Some(path_transform) = self.to_path_transform() {
                 skia_path = skia_path.make_transform(&path_transform);
             }
@@ -1392,6 +1395,8 @@ impl Shape {
 
         if let shape_type @ (Type::Path(_) | Type::Bool(_)) = &mut self.shape_type {
             if let Some(path) = shape_type.path_mut() {
+                #[cfg(target_arch = "wasm32")]
+                run_script!("console.count('path.transform')");
                 path.transform(transform);
             }
         } else if let Type::Text(text) = &mut self.shape_type {
diff --git a/render-wasm/src/shapes/paths.rs b/render-wasm/src/shapes/paths.rs
index 2debf44958..4d5c285cb2 100644
--- a/render-wasm/src/shapes/paths.rs
+++ b/render-wasm/src/shapes/paths.rs
@@ -1,7 +1,10 @@
-use skia_safe::{self as skia, Matrix};
-
+use std::{ops::Deref, rc::Rc};
+use skia_safe::{self as skia, Matrix, wrapper::ValueWrapper};
 use crate::math;
 
+#[cfg(target_arch = "wasm32")]
+use crate::run_script;
+
 mod subpaths;
 
 type Point = (f32, f32);
@@ -19,7 +22,7 @@ impl Segment {}
 #[derive(Debug, Clone, PartialEq)]
 pub struct Path {
     segments: Vec<Segment>,
-    skia_path: skia::Path,
+    pub skia_path: skia::Path,
     open: bool,
 }
 
@@ -217,8 +220,8 @@ impl Path {
         Path::new(segments)
     }
 
-    pub fn to_skia_path(&self) -> skia::Path {
-        self.skia_path.snapshot()
+    pub fn to_skia_path(&self) -> &skia::Path {
+        &self.skia_path
     }
 
     pub fn contains(&self, p: skia::Point) -> bool {