Add communicator interface

In merge !227 (merged) a test was added to ParameterTest which needs an instant of the class Communicator. Using the concrete class poses two problems, as the communicator depends on CUDA.

1. ParamterTest needs CUDA to compile, even though it is not necessary for the tests themselves
2. The new test from !227 (merged) fails in the pipeline (e.g. https://git.rz.tu-bs.de/irmb/VirtualFluids_dev/-/jobs/137967)

Adding an interface for the communicator fixes both problems. The new interface is named Communicator, while the concrete class was renamed to MpiCommunicator

Note: As mentioned by Sören, the communicator should be reworked and use the existing code in src/mpi (https://git.rz.tu-bs.de/irmb/VirtualFluids_dev/-/issues/45)

src/gpu/VirtualFluids_GPU/Communication/MpiCommunicator.h

@@ -31,35 +31,35 @@ public:
 
     void exchngBottomToTop(float *sbuf, float *rbuf, int count);
     void exchngTopToBottom(float *sbuf, float *rbuf, int count);
-    void waitAll();
+    void waitAll() override;
     void distributeGeometry(unsigned int *dataRoot, unsigned int *dataNode, int dataSizePerNode);
     int getPID() const override;
-    int getNumberOfProcess() const;
+    int getNumberOfProcess() const override;
     int getNeighbourTop();
     int getNeighbourBottom();
     void exchngData(float *sbuf_t, float *rbuf_t, float *sbuf_b, float *rbuf_b, int count);
     void exchngDataNB(float *sbuf_t, int count_st, float *rbuf_t, int count_rt, float *sbuf_b, int count_sb,
                       float *rbuf_b, int count_rb);
     //////////////////////////////////////////////////////////////////////////
-    void exchngDataGPU(real *sbuf, int count_s, real *rbuf, int count_r, int nb_rank);
+    void exchngDataGPU(real *sbuf, int count_s, real *rbuf, int count_r, int nb_rank) override;
     void sendRecvGPU(real *sbuf, int count_s, real *rbuf, int count_r, int nb_rank);
-    void nbRecvDataGPU(real *rbuf, int count_r, int nb_rank);
-    void nbSendDataGPU(real *sbuf, int count_s, int nb_rank);
-    void waitallGPU();
-    void sendDataGPU(real *sbuf, int count_s, int nb_rank);
-    void waitGPU(int id);
-    void resetRequest();
+    void nbRecvDataGPU(real *rbuf, int count_r, int nb_rank) override;
+    void nbSendDataGPU(real *sbuf, int count_s, int nb_rank) override;
+    void waitallGPU() override;
+    void sendDataGPU(real *sbuf, int count_s, int nb_rank) override;
+    void waitGPU(int id) override;
+    void resetRequest() override;
     void barrierGPU();
     void barrier();
     //////////////////////////////////////////////////////////////////////////
     void exchngDataGeo(int *sbuf_t, int *rbuf_t, int *sbuf_b, int *rbuf_b, int count);
     MPI_Comm getMpiCommunicator();
-    void startTimer();
-    void stopTimer();
-    double getTime();
-    int mapCudaDevice(const int &rank, const int &size, const std::vector<unsigned int> &devices, const int &maxdev);
-    std::vector<double> gatherNUPS(double processNups);
-    double sumNups(double processNups);
+    void startTimer() override;
+    void stopTimer() override;
+    double getTime() override;
+    int mapCudaDevice(const int &rank, const int &size, const std::vector<unsigned int> &devices, const int &maxdev) override;
+    std::vector<double> gatherNUPS(double processNups) override;
+    double sumNups(double processNups) override;
     //////////////////////////////////////////////////////////////////////////
     void receive_send(uint *buffer_receive, int size_buffer_recv, int neighbor_rank_recv, uint *buffer_send,
                       int size_buffer_send, int neighbor_rank_send) const override;