diff --git a/dsv4/kernels/cuda/blackwell_swizzle.cu b/dsv4/kernels/cuda/blackwell_swizzle.cu
index 1fb77914..5c6f1479 100644
--- a/dsv4/kernels/cuda/blackwell_swizzle.cu
+++ b/dsv4/kernels/cuda/blackwell_swizzle.cu
@@ -11,6 +11,7 @@
  */
 
 #include <cuda_runtime.h>
+#include <c10/cuda/CUDAStream.h>
 #include <cstdint>
 #include <torch/extension.h>  // For pybind11 bindings