cufftMP working with CUDA and OpenMP

cuda_fft.cu

@@ -5,10 +5,11 @@
 #include <cuda_runtime.h>
 #include <complex.h>
 #include "cuComplex.h"
-#include "w-stacking.h"
+#include "proto.h"
+#include "errcodes.h"
 #include <time.h>
 
-#if defined(CUFFTMP) && !defined(USE_FFTW)
+#if defined(CUFFTMP) && defined(USE_FFTW)
 
 void cuda_fft(
 	      int num_w_planes,
@@ -18,9 +19,22 @@ void cuda_fft(
 	      int yaxis,
 	      double * grid,
 	      double * gridss,
+	      int rank,
 	      MPI_Comm comm)
 {
-#ifdef __CUDACC__
+
+ #if !defined __CUDACC__
+  int ndevices;
+  cudaGetDeviceCount(&ndevices);
+  cudaSetDevice(rank % ndevices);
+
+  if ( rank == 0 ) {
+    if (0 == ndevices) {
+      
+      shutdown_wstacking(NO_ACCELERATORS_FOUND, "No accelerators found", __FILE__, __LINE__ );
+    }
+  }
+ #endif
   
   cudaError_t mmm;
   cufftResult_t status;
@@ -53,8 +67,8 @@ void cuda_fft(
 
 
 
-	long fftwindex = 0;
-	long fftwindex2D = 0;
+  uint fftwindex = 0;
+  uint fftwindex2D = 0;
   double norm = 1.0/(double)(grid_size_x*grid_size_y);
 
 
@@ -64,7 +78,7 @@ void cuda_fft(
 
   for (int iw=0; iw<num_w_planes; iw++)
     {
-                printf("select the %d w-plane to transform\n", iw);
+      //printf("select the %d w-plane to transform\n", iw);
       for (int iv=0; iv<yaxis; iv++)
 	{
 	  for (int iu=0; iu<xaxis; iu++)
@@ -143,6 +157,6 @@ void cuda_fft(
   cudaStreamDestroy(stream);
   cudaDeviceSynchronize();
 
-#endif // __CUDACC__
 }
 #endif 
+