AthCUDACore: Fix compilation in clang build.

nvcc has problems with the tbb headers in the clang build. Restructure slightly so that we don't need direct access to TBB from cu files.

Control/AthCUDA/AthCUDACore/AthCUDACore/TaskArena.h

@@ -5,24 +5,20 @@
 #ifndef ATHCUDACORE_TASKARENA_H
 #define ATHCUDACORE_TASKARENA_H
 
-// TBB include(s).
-#include <tbb/task_arena.h>
+#include <functional>
 
 namespace AthCUDA {
 
-   /// Function returning the TBB task arena that should be used for CUDA calls
+   /// Enqueue a function to the TBB task arena used for CUDA calls
    ///
    /// Since many (most?) CUDA calls use global locks, in order to avoid
    /// waiting times in the threads, one needs to schedule all CUDA operations
-   /// using this single task arena, which takes care of executing all
-   /// operations one by one. Using the TBB threads allocated for the rest of
+   /// using a single task arena, which takes care of executing all
+   /// operations one by one, using the TBB threads allocated for the rest of
    /// the job.
-   ///
-   /// @return A reference to the @c tbb::task_arena object to use for CUDA
-   ///         calls.
-   ///
-   tbb::task_arena& taskArena();
+   void enqueueTask( std::function<void()> f );
 
 } // namespace AthCUDA
 
+
 #endif // ATHCUDACORE_TASKARENA_H