Revert passing of a_dev to redist_band

src/elpa2_bandred_template.X90

@@ -1937,30 +1937,30 @@
      enddo ! istep
 
      if (useGPU) then
-!       ! this is not needed since a_dev is passed along from one subroutine to the other
-!
-!       successCUDA = cuda_memcpy ( &
-!#if REALCASE == 1
-!                                   loc(a),      &
-!#endif
-!#if COMPLEXCASE == 1
-!                                   loc(a(1,1)), &
-!#endif
-!                                   a_dev, lda*na_cols*  &
-!#if REALCASE == 1
-!                                   size_of_PRECISION_real,   &
-!#endif
-!#if COMPLEXCASE ==1
-!                                   size_of_PRECISION_complex,&
-!#endif
-!                                   cudaMemcpyDeviceToHost)
-!       if (.not.(successCUDA)) then
-!         print *,"bandred_&
-!	 &MATH_DATATYPE&
-!	 &: error in cudaMemcpy"
-!         stop
-!       endif
-!
+       ! this is not needed since a_dev is passed along from one subroutine to the other
+
+       successCUDA = cuda_memcpy ( &
+#if REALCASE == 1
+                                   loc(a),      &
+#endif
+#if COMPLEXCASE == 1
+                                   loc(a(1,1)), &
+#endif
+                                   a_dev, lda*na_cols*  &
+#if REALCASE == 1
+                                   size_of_PRECISION_real,   &
+#endif
+#if COMPLEXCASE ==1
+                                   size_of_PRECISION_complex,&
+#endif
+                                   cudaMemcpyDeviceToHost)
+       if (.not.(successCUDA)) then
+         print *,"bandred_&
+	 &MATH_DATATYPE&
+	 &: error in cudaMemcpy"
+         stop
+       endif
+
 !       successCUDA = cuda_free(a_dev)
 !       if (.not.(successCUDA)) then
 !         print *,"bandred_&

src/redist_band.X90

@@ -114,38 +114,38 @@ subroutine redist_band_&
    &PRECISION_SUFFIX &
    )
 
-   if (useGPU) then
-     ! copy a_dev to aMatrix
-     successCUDA = cuda_memcpy ( &
-#if REALCASE == 1
-                                loc(r_a),      &
-#endif
-#if COMPLEXCASE == 1
-                                loc(c_a(1,1)), &
-#endif
-                                a_dev, lda*matrixCols*  &
-#if REALCASE == 1
-#ifdef DOUBLE_PRECISION_REAL
-                                size_of_double_real_datatype,   &
-#else
-                                size_of_single_real_datatype,   &
-#endif
-#endif
-#if COMPLEXCASE ==1
-#ifdef DOUBLE_PRECISION_COMPLEX
-                                size_of_double_complex_datatype,&
-#else
-                                size_of_single_complex_datatype,&
-#endif
-#endif
-                                cudaMemcpyDeviceToHost)
-     if (.not.(successCUDA)) then
-       print *,"redist_band_&
-       &MATH_DATATYPE&
-       &: error in cudaMemcpy"
-       stop
-     endif
-   endif ! useGPU
+!   if (useGPU) then
+!     ! copy a_dev to aMatrix
+!     successCUDA = cuda_memcpy ( &
+!#if REALCASE == 1
+!                                loc(r_a),      &
+!#endif
+!#if COMPLEXCASE == 1
+!                                loc(c_a(1,1)), &
+!#endif
+!                                a_dev, lda*matrixCols*  &
+!#if REALCASE == 1
+!#ifdef DOUBLE_PRECISION_REAL
+!                                size_of_double_real_datatype,   &
+!#else
+!                                size_of_single_real_datatype,   &
+!#endif
+!#endif
+!#if COMPLEXCASE ==1
+!#ifdef DOUBLE_PRECISION_COMPLEX
+!                                size_of_double_complex_datatype,&
+!#else
+!                                size_of_single_complex_datatype,&
+!#endif
+!#endif
+!                                cudaMemcpyDeviceToHost)
+!     if (.not.(successCUDA)) then
+!       print *,"redist_band_&
+!       &MATH_DATATYPE&
+!       &: error in cudaMemcpy"
+!       stop
+!     endif
+!   endif ! useGPU
 
    call timer%start("mpi_communication")
    call mpi_comm_rank(mpi_comm,my_pe,mpierr)