Add support for profiling using tracy

Cargo.lock

@@ -761,6 +761,7 @@ dependencies = [
  "once_cell",
  "ordered-float",
  "png",
+ "profiling",
  "puffin",
  "puffin_egui",
  "regex",
@@ -1888,6 +1889,19 @@ dependencies = [
  "libc",
 ]
 
+[[package]]
+name = "generator"
+version = "0.7.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5cc16584ff22b460a382b7feec54b23d2908d858152e5739a120b949293bd74e"
+dependencies = [
+ "cc",
+ "libc",
+ "log",
+ "rustversion",
+ "windows 0.48.0",
+]
+
 [[package]]
 name = "generic-array"
 version = "0.14.7"
@@ -2038,7 +2052,7 @@ dependencies = [
  "presser",
  "thiserror",
  "winapi",
- "windows",
+ "windows 0.51.1",
 ]
 
 [[package]]
@@ -2883,6 +2897,19 @@ dependencies = [
  "log",
 ]
 
+[[package]]
+name = "loom"
+version = "0.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7e045d70ddfbc984eacfa964ded019534e8f6cbf36f6410aee0ed5cefa5a9175"
+dependencies = [
+ "cfg-if",
+ "generator",
+ "scoped-tls",
+ "tracing",
+ "tracing-subscriber",
+]
+
 [[package]]
 name = "lru"
 version = "0.12.1"
@@ -3802,18 +3829,19 @@ dependencies = [
 
 [[package]]
 name = "profiling"
-version = "1.0.13"
+version = "1.0.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d135ede8821cf6376eb7a64148901e1690b788c11ae94dc297ae917dbc91dc0e"
+checksum = "43d84d1d7a6ac92673717f9f6d1518374ef257669c24ebc5ac25d5033828be58"
 dependencies = [
  "profiling-procmacros",
+ "tracy-client",
 ]
 
 [[package]]
 name = "profiling-procmacros"
-version = "1.0.13"
+version = "1.0.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4b322d7d65c1ab449be3c890fcbd0db6e1092d0dd05d79dba2dd28032cebeb05"
+checksum = "8021cf59c8ec9c432cfc2526ac6b8aa508ecaf29cd415f271b8406c1b851c3fd"
 dependencies = [
  "quote",
  "syn 2.0.48",
@@ -4220,6 +4248,12 @@ dependencies = [
  "windows-sys 0.52.0",
 ]
 
+[[package]]
+name = "rustversion"
+version = "1.0.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7ffc183a10b4478d04cbbbfc96d0873219d962dd5accaff2ffbd4ceb7df837f4"
+
 [[package]]
 name = "rustybuzz"
 version = "0.12.1"
@@ -5017,6 +5051,26 @@ dependencies = [
  "tracing-log",
 ]
 
+[[package]]
+name = "tracy-client"
+version = "0.17.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "59fb931a64ff88984f86d3e9bcd1ae8843aa7fe44dd0f8097527bc172351741d"
+dependencies = [
+ "loom",
+ "once_cell",
+ "tracy-client-sys",
+]
+
+[[package]]
+name = "tracy-client-sys"
+version = "0.22.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9d104d610dfa9dd154535102cc9c6164ae1fa37842bc2d9e83f9ac82b0ae0882"
+dependencies = [
+ "cc",
+]
+
 [[package]]
 name = "ttf-parser"
 version = "0.20.0"
@@ -5710,6 +5764,15 @@ version = "0.4.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "712e227841d057c1ee1cd2fb22fa7e5a5461ae8e48fa2ca79ec42cfc1931183f"
 
+[[package]]
+name = "windows"
+version = "0.48.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e686886bc078bc1b0b600cac0147aadb815089b6e4da64016cbd754b6342700f"
+dependencies = [
+ "windows-targets 0.48.5",
+]
+
 [[package]]
 name = "windows"
 version = "0.51.1"

Cargo.toml

@@ -58,6 +58,7 @@ xdg = "^2.1"
 xdg-user = "0.2.1"
 xkbcommon = "0.7"
 zbus = "3.15.0"
+profiling = { version = "1.0" }
 
 [dependencies.id_tree]
 branch = "feature/copy_clone"
@@ -96,6 +97,7 @@ rev = "cdc652e0"
 debug = ["egui", "egui_plot", "smithay-egui", "renderdoc", "puffin", "puffin_egui", "anyhow/backtrace"]
 default = ["systemd"]
 systemd = ["libsystemd"]
+profile-with-tracy = ["profiling/profile-with-tracy"]
 
 [profile.dev]
 lto = "thin"

src/backend/kms/mod.rs

@@ -1153,6 +1153,7 @@ fn get_surface_dmabuf_feedback(
 }
 
 impl Surface {
+    #[profiling::function]
     pub fn render_output(
         &mut self,
         api: &mut GpuManager<GbmGlesBackend<GlowRenderer, DrmDeviceFd>>,
@@ -1359,6 +1360,7 @@ impl KmsState {
         self.session.change_vt(num).map_err(Into::into)
     }
 
+    #[profiling::function]
     pub fn apply_config_for_output(
         &mut self,
         output: &Output,
@@ -1587,6 +1589,7 @@ impl KmsState {
         Err(last_err)
     }
 
+    #[profiling::function]
     pub fn schedule_render(
         &mut self,
         loop_handle: &LoopHandle<'_, State>,
@@ -1626,6 +1629,7 @@ impl KmsState {
                         )
                     },
                     move |_time, _, state| {
+                        profiling::scope!("render_timer");
                         let backend = state.backend.kms();
                         let (mut device, mut other) = backend
                             .devices
@@ -1666,6 +1670,8 @@ impl KmsState {
                                 )
                             };
 
+                            profiling::finish_frame!();
+
                             match result {
                                 Ok(_) => {
                                     trace!(?crtc, "Frame pending");

src/backend/render/cursor.rs

@@ -184,6 +184,7 @@ where
     )
 }
 
+#[profiling::function]
 pub fn draw_dnd_icon<R>(
     renderer: &mut R,
     surface: &wl_surface::WlSurface,
@@ -301,6 +302,7 @@ impl Default for CursorState {
     }
 }
 
+#[profiling::function]
 pub fn draw_cursor<R>(
     renderer: &mut R,
     seat: &Seat<State>,

src/backend/render/mod.rs

@@ -389,6 +389,7 @@ pub enum CursorMode {
     All,
 }
 
+#[profiling::function]
 pub fn cursor_elements<'frame, E, R>(
     renderer: &mut R,
     state: &Common,
@@ -464,6 +465,7 @@ where
     elements
 }
 
+#[profiling::function]
 pub fn workspace_elements<R>(
     _gpu: Option<&DrmNode>,
     renderer: &mut R,
@@ -965,6 +967,7 @@ where
     }
 }
 
+#[profiling::function]
 pub fn render_output<R, Target, OffTarget, Source>(
     gpu: Option<&DrmNode>,
     renderer: &mut R,
@@ -1020,6 +1023,7 @@ where
     result
 }
 
+#[profiling::function]
 pub fn render_workspace<R, Target, OffTarget, Source>(
     gpu: Option<&DrmNode>,
     renderer: &mut R,

src/backend/winit.rs

@@ -51,6 +51,7 @@ pub struct WinitState {
 }
 
 impl WinitState {
+    #[profiling::function]
     pub fn render_output(&mut self, state: &mut Common) -> Result<()> {
         self.backend
             .bind()
@@ -197,6 +198,7 @@ pub fn init_backend(
                     error!(?err, "Failed to render frame.");
                     render_ping.ping();
                 }
+                profiling::finish_frame!();
             })
             .map_err(|_| anyhow::anyhow!("Failed to init eventloop timer for winit"))?,
     );

src/main.rs

@@ -38,6 +38,11 @@ pub mod utils;
 pub mod wayland;
 pub mod xwayland;
 
+#[cfg(feature = "profile-with-tracy")]
+#[global_allocator]
+static GLOBAL: profiling::tracy_client::ProfiledAllocator<std::alloc::System> =
+    profiling::tracy_client::ProfiledAllocator::new(std::alloc::System, 10);
+
 fn main() -> Result<()> {
     // setup logger
     logger::init_logger()?;
@@ -80,6 +85,11 @@ fn main() -> Result<()> {
         warn!(?err, "Failed to watch theme");
     }
 
+    #[cfg(feature = "profile-with-tracy")]
+    profiling::tracy_client::Client::start();
+
+    profiling::register_thread!("Main Thread");
+
     // run the event loop
     event_loop.run(None, &mut state, |state| {
         // shall we shut down?

src/shell/element/mod.rs

@@ -847,6 +847,7 @@ impl SpaceElement for CosmicMapped {
     fn z_index(&self) -> u8 {
         SpaceElement::z_index(&self.element)
     }
+    #[profiling::function]
     fn refresh(&self) {
         SpaceElement::refresh(&self.element)
     }

src/shell/element/surface.rs

@@ -641,6 +641,7 @@ impl SpaceElement for CosmicSurface {
         SpaceElement::output_leave(&self.0, output)
     }
 
+    #[profiling::function]
     fn refresh(&self) {
         SpaceElement::refresh(&self.0)
     }

src/shell/element/window.rs

@@ -490,11 +490,13 @@ impl SpaceElement for CosmicWindow {
             });
         }
     }
+    #[profiling::function]
     fn output_enter(&self, output: &Output, overlap: Rectangle<i32, Logical>) {
         SpaceElement::output_enter(&self.0, output, overlap);
         self.0
             .with_program(|p| SpaceElement::output_enter(&p.window, output, overlap));
     }
+    #[profiling::function]
     fn output_leave(&self, output: &Output) {
         SpaceElement::output_leave(&self.0, output);
         self.0
@@ -512,6 +514,7 @@ impl SpaceElement for CosmicWindow {
     fn z_index(&self) -> u8 {
         self.0.with_program(|p| SpaceElement::z_index(&p.window))
     }
+    #[profiling::function]
     fn refresh(&self) {
         SpaceElement::refresh(&self.0);
         if self.0.with_program(|p| {

src/shell/grabs/moving.rs

@@ -61,6 +61,7 @@ pub struct MoveGrabState {
 }
 
 impl MoveGrabState {
+    #[profiling::function]
     pub fn render<I, R>(
         &self,
         renderer: &mut R,

src/shell/layout/floating/mod.rs

@@ -1095,6 +1095,7 @@ impl FloatingLayout {
         self.mapped().flat_map(|e| e.windows().map(|(w, _)| w))
     }
 
+    #[profiling::function]
     pub fn refresh(&mut self) {
         #[cfg(feature = "debug")]
         puffin::profile_function!();
@@ -1150,6 +1151,7 @@ impl FloatingLayout {
         self.refresh(); //fixup any out of bounds elements
     }
 
+    #[profiling::function]
     pub fn render<R>(
         &self,
         renderer: &mut R,

src/shell/layout/tiling/mod.rs

@@ -2303,6 +2303,7 @@ impl TilingLayout {
         self.queue.push_tree(tree, ANIMATION_DURATION, blocker);
     }
 
+    #[profiling::function]
     pub fn refresh(&mut self) {
         #[cfg(feature = "debug")]
         puffin::profile_function!();
@@ -2934,6 +2935,7 @@ impl TilingLayout {
         }
     }
 
+    #[profiling::function]
     fn update_positions(
         output: &Output,
         tree: &mut Tree<Data>,
@@ -3801,6 +3803,7 @@ impl TilingLayout {
         }
     }
 
+    #[profiling::function]
     pub fn render<R>(
         &self,
         renderer: &mut R,

src/shell/mod.rs

@@ -1515,6 +1515,7 @@ impl Shell {
         }
     }
 
+    #[profiling::function]
     pub fn refresh(&mut self) {
         #[cfg(feature = "debug")]
         puffin::profile_function!();

src/shell/workspace.rs

@@ -286,6 +286,7 @@ impl Workspace {
         }
     }
 
+    #[profiling::function]
     pub fn refresh(&mut self, xdg_activation_state: &XdgActivationState) {
         #[cfg(feature = "debug")]
         puffin::profile_function!();
@@ -1059,6 +1060,7 @@ impl Workspace {
         }
     }
 
+    #[profiling::function]
     pub fn render<'a, R>(
         &self,
         renderer: &mut R,

src/state.rs

@@ -353,7 +353,15 @@ impl State {
             .with_context(|| "Failed to load languages")
             .unwrap();
 
+        #[cfg(feature = "profile-with-tracy")]
+        unsafe {
+            time::util::local_offset::set_soundness(time::util::local_offset::Soundness::Unsound);
+        }
         let local_offset = UtcOffset::current_local_offset().expect("No yet multithreaded");
+        #[cfg(feature = "profile-with-tracy")]
+        unsafe {
+            time::util::local_offset::set_soundness(time::util::local_offset::Soundness::Sound);
+        }
         let clock = Clock::new();
         let config = Config::load(&handle);
         let compositor_state = CompositorState::new::<Self>(dh);

src/utils/iced.rs

@@ -358,6 +358,7 @@ impl<P: Program + Send + 'static + Clone> IcedElement<P> {
 }
 
 impl<P: Program + Send + 'static> IcedElementInternal<P> {
+    #[profiling::function]
     fn update(&mut self, mut force: bool) -> Vec<Action<<P as Program>::Message>> {
         while let Ok(message) = self.rx.try_recv() {
             self.state.queue_message(message);
@@ -691,6 +692,7 @@ impl<P: Program + Send + 'static> SpaceElement for IcedElement<P> {
         RenderZindex::Shell as u8
     }
 
+    #[profiling::function]
     fn refresh(&self) {
         let mut internal = self.0.lock().unwrap();
         // makes partial borrows easier

src/wayland/handlers/screencopy.rs

@@ -593,6 +593,7 @@ where
     Ok(())
 }
 
+#[profiling::function]
 pub fn render_session<F, R>(
     node: Option<DrmNode>,
     renderer: &mut R,
@@ -637,6 +638,7 @@ where
     }
 }
 
+#[profiling::function]
 pub fn render_output_to_buffer(
     state: &mut State,
     session: &Session,
@@ -767,6 +769,7 @@ pub fn render_output_to_buffer(
     }
 }
 
+#[profiling::function]
 pub fn render_workspace_to_buffer(
     state: &mut State,
     session: &Session,
@@ -916,6 +919,7 @@ smithay::render_elements! {
     CursorElement=cursor::CursorRenderElement<R>,
 }
 
+#[profiling::function]
 pub fn render_window_to_buffer(
     state: &mut State,
     session: &Session,