Switched to futures_lite for blocking

Cargo.toml

@@ -27,7 +27,6 @@ serde_json = { version = "1.0.119", default-features = false }
 dashmap = "5.5.3"
 hashbrown = "0.14.5"
 spin = { version = "0.9.8", features = ["mutex", "spin_mutex"] }
-rayon = "1"
 
 getrandom = { version = "0.2.15", default-features = false }
 rand = { version = "0.8.5", default-features = false, features = [
@@ -38,6 +37,7 @@ async-channel = "2.3"
 dirs = "5.0.1"
 md5 = "0.7.0"
 sanitize-filename = "0.5"
+wasm-bindgen = "0.2"
 wasm-bindgen-futures = "0.4.45"
 weak-table = "0.3"
 web-time = "1.1.0"
@@ -74,7 +74,6 @@ pretty_assertions = "1.4"
 # Async
 embassy-futures = { version = "0.1.1" }                        # for no-std
 futures-lite = { version = "2.3.0", default-features = false }
-futures = "0.3.31"
 
 [profile.dev]
 opt-level = 2

crates/cubecl-wgpu/Cargo.toml

@@ -39,15 +39,15 @@ wgpu = { version = "22.0.0", features = ["fragile-send-sync-non-atomic-wasm"] }
 
 async-channel = { workspace = true }
 derive-new = { workspace = true }
+futures-lite = { workspace = true }
 hashbrown = { workspace = true }
 log = { workspace = true }
 web-time = { workspace = true }
 
 cfg-if = { workspace = true }
 
 [target.'cfg(all(target_arch = "wasm32", target_feature = "atomics"))'.dependencies]
-futures = { workspace = true }
-rayon = { workspace = true }
+wasm-bindgen = { workspace = true }
 
 [dev-dependencies]
 cubecl-core = { path = "../cubecl-core", version = "0.4.0", features = [

crates/cubecl-wgpu/src/compute/server.rs

@@ -25,6 +25,7 @@ pub struct WgpuServer<C: WgpuCompiler> {
     memory_management: MemoryManagement<WgpuStorage>,
     pub(crate) device: Pdrc<wgpu::Device>,
     queue: Pdrc<wgpu::Queue>,
+    pub(crate) adapter: Pdrc<wgpu::Adapter>,
     encoder: CommandEncoder,
     current_pass: Option<ComputePass<'static>>,
     tasks_count: usize,
@@ -90,6 +91,7 @@ impl<C: WgpuCompiler> WgpuServer<C> {
         memory_management: MemoryManagement<WgpuStorage>,
         device: Pdrc<wgpu::Device>,
         queue: Pdrc<wgpu::Queue>,
+        adapter: Pdrc<wgpu::Adapter>,
         tasks_max: usize,
     ) -> Self {
         let logger = DebugLogger::default();
@@ -99,18 +101,22 @@ impl<C: WgpuCompiler> WgpuServer<C> {
             timestamps.enable(&device);
         }
 
+        let encoder = create_encoder(&device);
+        let poll = WgpuPoll::new(device.clone());
+
         Self {
             memory_management,
-            device: device.clone(),
-            queue: queue.clone(),
-            encoder: create_encoder(&device),
+            device,
+            queue,
+            adapter,
+            encoder,
             current_pass: None,
             tasks_count: 0,
             storage_locked: MemoryLock::default(),
             pipelines: HashMap::new(),
             tasks_max,
             logger,
-            poll: WgpuPoll::new(device.clone()),
+            poll,
             duration_profiled: None,
             timestamps,
             _compiler: PhantomData,