Changeset 2118 for palm/trunk/SOURCE/fft_xy_mod.f90

Timestamp:

Jan 17, 2017 4:38:49 PM (7 years ago)

Author:

raasch

Message:

all OpenACC directives and related parts removed from the code

File:

• palm/trunk/SOURCE/fft_xy_mod.f90 (modified) (12 diffs)

palm/trunk/SOURCE/fft_xy_mod.f90

@@ -20 +20 @@
 ! Current revisions:
 ! -----------------
-! 
+! OpenACC directives and CUDA-fft related code removed
 ! 
 ! Former revisions:
@@ -31 +31 @@
 ! 1850 2016-04-08 13:29:27Z maronga
 ! Module renamed
-! 
 ! 
 ! 1815 2016-04-06 13:49:59Z raasch
@@ -139 +138 @@
         ONLY:  nx, ny, nz
         
-#if defined( __cuda_fft )
-    USE ISO_C_BINDING
-#elif defined( __fftw )
+#if defined( __fftw )
     USE, INTRINSIC ::  ISO_C_BINDING
 #endif
@@ -192 +189 @@
     REAL(wp), DIMENSION(:), ALLOCATABLE, SAVE ::  trig_yf  !<
     
-#elif defined( __cuda_fft )
-    INTEGER(C_INT), SAVE ::  plan_xf  !<
-    INTEGER(C_INT), SAVE ::  plan_xi  !<
-    INTEGER(C_INT), SAVE ::  plan_yf  !<
-    INTEGER(C_INT), SAVE ::  plan_yi  !<
-    
-    INTEGER(iwp), SAVE   ::  total_points_x_transpo  !<
-    INTEGER(iwp), SAVE   ::  total_points_y_transpo  !<
 #endif
 
@@ -261 +250 @@
     SUBROUTINE fft_init
 
-       USE cuda_fft_interfaces
-
        IMPLICIT NONE
 
@@ -338 +325 @@
           CALL ZDFFTM( 0, ny+1, nz1, sqr_dny, work_y, ny+4, work_y, ny+4,      &
                        trig_yb, worky, 0 )
-#elif defined( __cuda_fft )
-          total_points_x_transpo = (nx+1) * (nyn_x-nys_x+1) * (nzt_x-nzb_x+1)
-          total_points_y_transpo = (ny+1) * (nxr_y-nxl_y+1) * (nzt_y-nzb_y+1)
-          CALL CUFFTPLAN1D( plan_xf, nx+1, CUFFT_D2Z, (nyn_x-nys_x+1) * (nzt_x-nzb_x+1) )
-          CALL CUFFTPLAN1D( plan_xi, nx+1, CUFFT_Z2D, (nyn_x-nys_x+1) * (nzt_x-nzb_x+1) )
-          CALL CUFFTPLAN1D( plan_yf, ny+1, CUFFT_D2Z, (nxr_y-nxl_y+1) * (nzt_y-nzb_y+1) )
-          CALL CUFFTPLAN1D( plan_yi, ny+1, CUFFT_Z2D, (nxr_y-nxl_y+1) * (nzt_y-nzb_y+1) )
 #else
           message_string = 'no system-specific fft-call available'
@@ -403 +383 @@
 
 
-       USE cuda_fft_interfaces
-#if defined( __cuda_fft )
-       USE ISO_C_BINDING
-#endif
-
        IMPLICIT NONE
 
@@ -429 +404 @@
 #elif defined( __nec )
        REAL(wp), DIMENSION(6*(nx+1)) ::  work2  !<
-#elif defined( __cuda_fft )
-       COMPLEX(dp), DIMENSION(0:(nx+1)/2,nys_x:nyn_x,nzb_x:nzt_x) ::           &
-          ar_tmp  !<
-       ! following does not work for PGI 14.1 -> to be removed later
-       ! !$acc declare create( ar_tmp )
 #endif
 
@@ -737 +707 @@
 
           ENDIF
-
-#elif defined( __cuda_fft )
-
-          !$acc data create( ar_tmp )
-          IF ( forward_fft )  THEN
-
-             !$acc data present( ar )
-             CALL CUFFTEXECD2Z( plan_xf, ar, ar_tmp )
-
-             !$acc kernels
-             DO  k = nzb_x, nzt_x
-                DO  j = nys_x, nyn_x
-
-                   DO  i = 0, (nx+1)/2
-                      ar(i,j,k)      = REAL( ar_tmp(i,j,k), KIND=wp )  * dnx
-                   ENDDO
-
-                   DO  i = 1, (nx+1)/2 - 1
-                      ar(nx+1-i,j,k) = AIMAG( ar_tmp(i,j,k) ) * dnx
-                   ENDDO
-
-                ENDDO
-             ENDDO
-             !$acc end kernels
-             !$acc end data
-
-          ELSE
-
-             !$acc data present( ar )
-             !$acc kernels
-             DO  k = nzb_x, nzt_x
-                DO  j = nys_x, nyn_x
-
-                   ar_tmp(0,j,k) = CMPLX( ar(0,j,k), 0.0_wp, KIND=wp )
-
-                   DO  i = 1, (nx+1)/2 - 1
-                      ar_tmp(i,j,k) = CMPLX( ar(i,j,k), ar(nx+1-i,j,k),        &
-                                             KIND=wp )
-                   ENDDO
-                   ar_tmp((nx+1)/2,j,k) = CMPLX( ar((nx+1)/2,j,k), 0.0_wp,     &
-                                                 KIND=wp )
-
-                ENDDO
-             ENDDO
-             !$acc end kernels
-
-             CALL CUFFTEXECZ2D( plan_xi, ar_tmp, ar )
-             !$acc end data
-
-          ENDIF
-          !$acc end data
 
 #else
@@ -1052 +971 @@
 
 
-       USE cuda_fft_interfaces
-#if defined( __cuda_fft )
-       USE ISO_C_BINDING
-#endif
-
        IMPLICIT NONE
 
@@ -1082 +996 @@
 #elif defined( __nec )
        REAL(wp), DIMENSION(6*(ny+1)) ::  work2  !<
-#elif defined( __cuda_fft )
-       COMPLEX(dp), DIMENSION(0:(ny+1)/2,nxl_y:nxr_y,nzb_y:nzt_y) ::           &
-          ar_tmp  !<
-       ! following does not work for PGI 14.1 -> to be removed later
-       ! !$acc declare create( ar_tmp )
 #endif
 
@@ -1364 +1273 @@
 
           ENDIF
-#elif defined( __cuda_fft )
-
-          !$acc data create( ar_tmp )
-          IF ( forward_fft )  THEN
-
-             !$acc data present( ar )
-             CALL CUFFTEXECD2Z( plan_yf, ar, ar_tmp )
-
-             !$acc kernels
-             DO  k = nzb_y, nzt_y
-                DO  i = nxl_y, nxr_y
-
-                   DO  j = 0, (ny+1)/2
-                      ar(j,i,k)      = REAL( ar_tmp(j,i,k), KIND=wp )  * dny
-                   ENDDO
-
-                   DO  j = 1, (ny+1)/2 - 1
-                      ar(ny+1-j,i,k) = AIMAG( ar_tmp(j,i,k) ) * dny
-                   ENDDO
-
-                ENDDO
-             ENDDO
-             !$acc end kernels
-             !$acc end data
-
-          ELSE
-
-             !$acc data present( ar )
-             !$acc kernels
-             DO  k = nzb_y, nzt_y
-                DO  i = nxl_y, nxr_y
-
-                   ar_tmp(0,i,k) = CMPLX( ar(0,i,k), 0.0_wp, KIND=wp )
-
-                   DO  j = 1, (ny+1)/2 - 1
-                      ar_tmp(j,i,k) = CMPLX( ar(j,i,k), ar(ny+1-j,i,k),        &
-                                             KIND=wp )
-                   ENDDO
-                   ar_tmp((ny+1)/2,i,k) = CMPLX( ar((ny+1)/2,i,k), 0.0_wp,     &
-                                                 KIND=wp )
-
-                ENDDO
-             ENDDO
-             !$acc end kernels
-
-             CALL CUFFTEXECZ2D( plan_yi, ar_tmp, ar )
-             !$acc end data
-
-          ENDIF
-          !$acc end data
-
 #else
           message_string = 'no system-specific fft-call available'