remove is_padded check (#988)

Author: louisfd

crates/cubecl-matmul/src/tests/naive/macros.rs

@@ -21,6 +21,13 @@ macro_rules! testgen_matmul_simple {
                 )
             }
 
+            #[test]
+            pub fn test_odd() {
+                cubecl_matmul::tests::naive::tests::test_odd::<TestRuntime, FloatT>(
+                    &Default::default(),
+                )
+            }
+
             #[test]
             pub fn test_simple_matmul_large() {
                 cubecl_matmul::tests::naive::tests::test_large::<TestRuntime, FloatT>(

crates/cubecl-matmul/src/tests/naive/tests.rs

@@ -23,6 +23,17 @@ pub fn test_small<R: Runtime, F: Float + CubeElement + Display + Sample>(device:
     test_simple::<R, F>(case, device);
 }
 
+pub fn test_odd<R: Runtime, F: Float + CubeElement + Display + Sample>(device: &R::Device) {
+    let case = MatmulTestCase {
+        m: 1,
+        k: 101,
+        n: 255,
+        batch: 1,
+    };
+
+    test_simple::<R, F>(case, device);
+}
+
 pub fn test_large<R: Runtime, F: Float + CubeElement + Display + Sample>(device: &R::Device) {
     let case = MatmulTestCase {
         m: 256,

crates/cubecl-std/src/tensor/contiguous.rs

@@ -332,10 +332,9 @@ pub fn into_contiguous_ref<R: Runtime, E: CubePrimitive>(
     let mut num_elems_per_unit = line_size as u32 * elems_per_unit;
 
     let last_dim = output.shape[rank - 1];
-    let is_padded = rank > 1 && last_dim != output.strides[rank - 2];
 
     // If tensor is strided, elems_per_unit must be compatible with last dim
-    while is_padded && !last_dim.is_multiple_of(num_elems_per_unit as usize) {
+    while !last_dim.is_multiple_of(num_elems_per_unit as usize) {
         elems_per_unit /= 2;
         num_elems_per_unit /= 2;
     }