Fix TF32 feature registration

tracel-ai · Nov 25, 2024 · 1474ce5 · 1474ce5
1 parent 7398dec
commit 1474ce5
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/crates/cubecl-cpp/src/cuda/wmma/cuda_compiler.rs b/crates/cubecl-cpp/src/cuda/wmma/cuda_compiler.rs
@@ -91,7 +91,7 @@ impl WmmaCompiler<CudaDialect<Self>> for CudaWmmaCompiler {
                 gpu::Elem::Float(gpu::FloatKind::TF32),
                 gpu::Elem::Float(gpu::FloatKind::TF32),
                 gpu::Elem::Float(gpu::FloatKind::F32),
-                vec![(16, 8, 16)],
+                vec![(16, 16, 8)],
             ));
         }
         result