New makefile for cuda_mpi and minor changes

Signed-off-by: Alessandro Fanfarillo <fanfarillo@ing.uniroma2.it>

Author: Alessandro Fanfarillo

src/cuda_mpi/Makefile

@@ -0,0 +1,21 @@
+include ../make.inc
+
+MPICC ?= cc
+
+libcaf_cuda.a: mpi_caf.o ../common/caf_auxiliary.o
+	ar rcv $@ mpi_caf.o ../common/caf_auxiliary.o
+	ranlib $@
+
+.c.o:
+	$(MPICC) $(CFLAGS) $(MPI_CFLAGS) -I.. -I/opt/nvidia/cudatoolkit6.5/6.5.14-1.0502.9836.9.1/include/ -L/opt/nvidia/cudatoolkit6.5/6.5.14-1.0502.9836.9.1/lib -lcuda -lcudart -c $< -o $@
+
+mpi_caf.o: mpi_caf.c ../libcaf.h ../libcaf-gfortran-descriptor.h
+
+../common/caf_auxiliary.o:
+	$(MAKE) -C ../common
+
+clean:
+	rm -f mpi_caf.o
+
+distclean: clean
+	rm -f libcaf_mpi.a

src/cuda_mpi/mpi_caf.c

@@ -41,7 +41,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.  */
 #include <unistd.h>
 #include <mpi.h>
 #include <pthread.h>
-
+#include "cuda_runtime_api.h"
 #include "libcaf.h"
 
 /* Define GFC_CAF_CHECK to enable run-time checking.  */
@@ -477,8 +477,8 @@ PREFIX(registernc) (void* mem,size_t mem_size)
   cuda_ierr = cudaHostRegister(mem,mem_size,cudaHostRegisterMapped);
   cudaDeviceSynchronize();
 
-  if (ierr != 0) 
-    call caf_runtime_error ("CUDA allocation failed with code %d", ierr);
+  if (cuda_ierr != 0) 
+    caf_runtime_error ("CUDA allocation failed with code %d", cuda_ierr);
 
   return;