diff --git a/control/common.h b/control/common.h
index f08f2712c91a98fe18790d40331168d06b325f71..ab1a98867bfed1be80fd09e0db14969773b93e6b 100644
--- a/control/common.h
+++ b/control/common.h
@@ -54,6 +54,7 @@
 #include <cuda_runtime.h>
 #include <cuda_runtime_api.h>
 #if defined(CHAMELEON_USE_CUBLAS_V2)
+#include <cublas.h>
 #include <cublas_v2.h>
 #else
 #include <cublas.h>
diff --git a/runtime/starpu/include/morse_starpu.h b/runtime/starpu/include/morse_starpu.h
index a98c893e520c98e4c3802d458f2baa5fc6528bcf..f0649aeaf585e3f45fc4f5a548b8ab0a5d5c8401 100644
--- a/runtime/starpu/include/morse_starpu.h
+++ b/runtime/starpu/include/morse_starpu.h
@@ -45,6 +45,13 @@
 #if defined(CHAMELEON_USE_CUDA) && !defined(CHAMELEON_SIMULATION)
 #include <starpu_scheduler.h>
 #include <starpu_cuda.h>
+
+#include <cublas.h>
+#include <starpu_cublas.h>
+#if defined(CHAMELEON_USE_CUBLAS_V2)
+#include <cublas_v2.h>
+#include <starpu_cublas_v2.h>
+#endif
 #endif
 
 #include "control/common.h"