Chore: update cubecl + fix no-std (tracel-ai#3771)

* Update rev and loader changes

* Update rev

Author: laggui

Cargo.lock

@@ -169,9 +169,9 @@ portable-atomic = { version = "1.11.1" }
 portable-atomic-util = { version = "0.2.4", features = ["alloc"] }
 
 ### For the main burn branch. ###
-cubecl = { git = "https://github.com/tracel-ai/cubecl", default-features = false, rev = "ad03f7892f4069d3406f936fd9fa3ac3b090c1fe" }
-cubecl-common = { git = "https://github.com/tracel-ai/cubecl", default-features = false, rev = "ad03f7892f4069d3406f936fd9fa3ac3b090c1fe" }
-cubecl-quant = { git = "https://github.com/tracel-ai/cubecl", default-features = false, rev = "ad03f7892f4069d3406f936fd9fa3ac3b090c1fe" }
+cubecl = { git = "https://github.com/tracel-ai/cubecl", default-features = false, rev = "98eb9d27624375fb11d4a97febd364b049933fdc" }
+cubecl-common = { git = "https://github.com/tracel-ai/cubecl", default-features = false, rev = "98eb9d27624375fb11d4a97febd364b049933fdc" }
+cubecl-quant = { git = "https://github.com/tracel-ai/cubecl", default-features = false, rev = "98eb9d27624375fb11d4a97febd364b049933fdc" }
 ### For local development. ###
 # cubecl = { path = "../cubecl/crates/cubecl", default-features = false }
 # cubecl-common = { path = "../cubecl/crates/cubecl-common", default-features = false }

Cargo.toml

@@ -22,7 +22,7 @@ use burn_ir::TensorIr;
 use cubecl::features::TypeUsage;
 use cubecl::matmul::components::AccG;
 use cubecl::matmul::components::AccS;
-use cubecl::matmul::components::tile::loader::Filled;
+use cubecl::matmul::components::tile::reader::Filled;
 use cubecl::matmul::kernels::layered::Selection;
 use cubecl::matmul::kernels::layered::double_buffering::CyclicDoubleBufferingAlgorithm;
 use cubecl::matmul::kernels::layered::double_buffering::DoubleBufferingArgs;

crates/burn-cubecl-fusion/src/matmul/optimization.rs

@@ -248,8 +248,7 @@ pub fn input_as_linear_view<C: CubePrimitive>(
     #[comptime] pos: u32,
 ) -> View<Line<C>, u32> {
     let slice = input_as_slice::<Line<C>>(inputs, pos);
-    let line_size = slice.line_size();
-    let layout = LinearLayout::new_Plain(PlainLayout::new(slice.len(), comptime!(line_size as u8)));
+    let layout = LinearLayout::new_Plain(PlainLayout::new(slice.len()));
     View::new::<Slice<Line<C>>, u32>(&slice, layout)
 }
 

crates/burn-cubecl-fusion/src/shared/io.rs

@@ -1,7 +1,7 @@
 use burn_tensor::DType;
 use cubecl::{
     matmul::{
-        MatmulInputHandleRef, Strategy, SyncLoadingStrategy, SyncPartialLoadingStrategy,
+        MatmulInputHandleRef, Strategy, SyncPartialReadingStrategy, SyncReadingStrategy,
         components::{AccG, MatmulKind},
         kernels::layered::{
             Selection, TileSizeSelection, double_buffering::DoubleBufferingArgs,
@@ -161,7 +161,7 @@ fn matmul_simple<R: CubeRuntime, E: MatmulElement>(
 ) -> Result<(), String> {
     cubecl::matmul::launch_ref::<R, E>(
         &Strategy::Simple(
-            SyncLoadingStrategy::Cyclic,
+            SyncReadingStrategy::Cyclic,
             Selection::Inferred(SimpleArgs { multi_rows: false }),
         ),
         &lhs.client,
@@ -179,7 +179,7 @@ fn matmul_simple_multi_rows<R: CubeRuntime, E: MatmulElement>(
 ) -> Result<(), String> {
     cubecl::matmul::launch_ref::<R, E>(
         &Strategy::Simple(
-            SyncLoadingStrategy::Cyclic,
+            SyncReadingStrategy::Cyclic,
             Selection::Inferred(SimpleArgs { multi_rows: true }),
         ),
         &lhs.client,
@@ -197,7 +197,7 @@ fn matmul_double_buffering<R: CubeRuntime, E: MatmulElement>(
 ) -> Result<(), String> {
     cubecl::matmul::launch_ref::<R, E>(
         &Strategy::DoubleBuffering(
-            SyncPartialLoadingStrategy::Tilewise,
+            SyncPartialReadingStrategy::Tilewise,
             Selection::Inferred(DoubleBufferingArgs { specialized: false }),
         ),
         &lhs.client,
@@ -215,7 +215,7 @@ fn matmul_double_buffering_specialized<R: CubeRuntime, E: MatmulElement>(
 ) -> Result<(), String> {
     cubecl::matmul::launch_ref::<R, E>(
         &Strategy::DoubleBuffering(
-            SyncPartialLoadingStrategy::Tilewise,
+            SyncPartialReadingStrategy::Tilewise,
             Selection::Inferred(DoubleBufferingArgs { specialized: true }),
         ),
         &lhs.client,