Disable the async pipeline by default (#500)

* Handle profiling in the non-async pipeline * Make the async pipeline optional * Profile the blit stage * Apply review suggestions Co-Authored-By: "Spencer C. Imbleau" <[email protected]> --------- Co-authored-by: "Spencer C. Imbleau" <[email protected]>
linebender · Mar 5, 2024 · ec5c8a4 · ec5c8a4
1 parent e39e340
commit ec5c8a4
Show file tree

Hide file tree

Showing 3 changed files with 55 additions and 23 deletions.
diff --git a/examples/with_winit/src/lib.rs b/examples/with_winit/src/lib.rs
@@ -55,6 +55,16 @@ struct Args {
     /// Use `0` for an automatic choice
     #[arg(long, default_value_t=default_threads())]
     num_init_threads: usize,
+    /// Use the asynchronous pipeline (if available) for rendering
+    ///
+    /// The asynchronous pipeline is one approach for robust memory - see
+    /// <https://github.com/linebender/vello/issues/366>
+    ///
+    /// However, it also has potential latency issues, especially for
+    /// accessibility technology, as it (currently) blocks the main thread for
+    /// extended periods
+    #[arg(long)]
+    async_pipeline: bool,
 }
 
 fn default_threads() -> usize {
@@ -451,8 +461,10 @@ fn run(
                             .surface
                             .get_current_texture()
                             .expect("failed to get surface texture");
-                        #[cfg(not(target_arch = "wasm32"))]
-                        {
+                        // Note: we don't run the async/"robust" pipeline, as
+                        // it requires more async wiring for the readback. See
+                        // [#gpu > async on wasm](https://xi.zulipchat.com/#narrow/stream/197075-gpu/topic/async.20on.20wasm)
+                        if args.async_pipeline && cfg!(not(target_arch = "wasm32")) {
                             scene_complexity = vello::block_on_wgpu(
                                 &device_handle.device,
                                 renderers[render_state.surface.dev_id]
@@ -467,21 +479,19 @@ fn run(
                                     ),
                             )
                             .expect("failed to render to surface");
+                        } else {
+                            renderers[render_state.surface.dev_id]
+                                .as_mut()
+                                .unwrap()
+                                .render_to_surface(
+                                    &device_handle.device,
+                                    &device_handle.queue,
+                                    &scene,
+                                    &surface_texture,
+                                    &render_params,
+                                )
+                                .expect("failed to render to surface");
                         }
-                        // Note: in the wasm case, we're currently not running the robust
-                        // pipeline, as it requires more async wiring for the readback.
-                        #[cfg(target_arch = "wasm32")]
-                        renderers[render_state.surface.dev_id]
-                            .as_mut()
-                            .unwrap()
-                            .render_to_surface(
-                                &device_handle.device,
-                                &device_handle.queue,
-                                &scene,
-                                &surface_texture,
-                                &render_params,
-                            )
-                            .expect("failed to render to surface");
                         surface_texture.present();
                         device_handle.device.poll(wgpu::Maintain::Poll);
 

diff --git a/src/lib.rs b/src/lib.rs
@@ -362,12 +362,27 @@ impl Renderer {
                 occlusion_query_set: None,
                 timestamp_writes: None,
             });
+            let mut render_pass = self
+                .profiler
+                .scope("blit to surface", &mut render_pass, device);
             render_pass.set_pipeline(&blit.pipeline);
             render_pass.set_bind_group(0, &bind_group, &[]);
             render_pass.draw(0..6, 0..1);
         }
+        #[cfg(feature = "wgpu-profiler")]
+        self.profiler.resolve_queries(&mut encoder);
         queue.submit(Some(encoder.finish()));
         self.target = Some(target);
+        #[cfg(feature = "wgpu-profiler")]
+        {
+            self.profiler.end_frame().unwrap();
+            if let Some(result) = self
+                .profiler
+                .process_finished_frame(queue.get_timestamp_period())
+            {
+                self.profile_result = Some(result);
+            }
+        }
         Ok(())
     }
 
@@ -510,6 +525,9 @@ impl Renderer {
                 timestamp_writes: None,
                 occlusion_query_set: None,
             });
+            let mut render_pass = self
+                .profiler
+                .scope("blit to surface", &mut render_pass, device);
             render_pass.set_pipeline(&blit.pipeline);
             render_pass.set_bind_group(0, &bind_group, &[]);
             render_pass.draw(0..6, 0..1);
@@ -519,13 +537,14 @@ impl Renderer {
         queue.submit(Some(encoder.finish()));
         self.target = Some(target);
         #[cfg(feature = "wgpu-profiler")]
-        self.profiler.end_frame().unwrap();
-        #[cfg(feature = "wgpu-profiler")]
-        if let Some(result) = self
-            .profiler
-            .process_finished_frame(queue.get_timestamp_period())
         {
-            self.profile_result = Some(result);
+            self.profiler.end_frame().unwrap();
+            if let Some(result) = self
+                .profiler
+                .process_finished_frame(queue.get_timestamp_period())
+            {
+                self.profile_result = Some(result);
+            }
         }
         Ok(bump)
     }

diff --git a/src/util.rs b/src/util.rs
@@ -166,10 +166,13 @@ impl std::task::Wake for NullWake {
 ///
 /// This will deadlock if the future is awaiting anything other than GPU progress.
 pub fn block_on_wgpu<F: Future>(device: &Device, mut fut: F) -> F::Output {
+    if cfg!(target_arch = "wasm32") {
+        panic!("Blocking can't work on WASM, so");
+    }
     let waker = std::task::Waker::from(std::sync::Arc::new(NullWake));
     let mut context = std::task::Context::from_waker(&waker);
     // Same logic as `pin_mut!` macro from `pin_utils`.
-    let mut fut = unsafe { std::pin::Pin::new_unchecked(&mut fut) };
+    let mut fut = std::pin::pin!(fut);
     loop {
         match fut.as_mut().poll(&mut context) {
             std::task::Poll::Pending => {