The definitions of "backward" and "forward" FFT in on-line comments was

not consistent with what FFTs actually do in the code

FFTXlib/fft_fwinv.f90

@@ -11,15 +11,15 @@ SUBROUTINE invfft_y( fft_kind, f, dfft, howmany )
   !! Compute G-space to R-space for a specific grid type
   !! 
   !! **fft_kind = 'Rho'** : 
-  !!   inverse fourier transform of potentials and charge density f
+  !!   inverse (backward) fourier transform of potentials and charge density f
   !!   On output, f is overwritten
   !! 
   !! **fft_kind = 'Wave'** :
-  !!   inverse fourier transform of  wave functions f
+  !!   inverse (backward) fourier transform of  wave functions f
   !!   On output, f is overwritten
   !!
   !! **fft_kind = 'tgWave'** :
-  !!   inverse fourier transform of  wave functions f with task group
+  !!   inverse (backward) fourier transform of  wave functions f with task group
   !!   On output, f is overwritten
   !!
   !! **dfft = FFT grid descriptor**, IMPORTANT NOTICE: grid is specified only by dfft.
@@ -311,15 +311,15 @@ SUBROUTINE invfft_y_gpu( fft_kind, f_d, dfft, howmany, stream )
   !! Compute G-space to R-space for a specific grid type
   !! 
   !! **fft_kind = 'Rho'** : 
-  !!   inverse fourier transform of potentials and charge density f
+  !!   inverse (backward) fourier transform of potentials and charge density f
   !!   On output, f is overwritten
   !! 
   !! **fft_kind = 'Wave'** :
-  !!   inverse fourier transform of  wave functions f
+  !!   inverse (backward) fourier transform of  wave functions f
   !!   On output, f is overwritten
   !!
   !! **fft_kind = 'tgWave'** :
-  !!   inverse fourier transform of  wave functions f with task group
+  !!   inverse (backward) fourier transform of  wave functions f with task group
   !!   On output, f is overwritten
   !!
   !! **dfft = FFT grid descriptor**, IMPORTANT NOTICE: grid is specified only by dfft.

FFTXlib/fft_scalar.DFTI.f90

@@ -56,7 +56,7 @@
 !     (ldz>nz is used on some architectures to reduce memory conflicts)
 !     input  :  c(ldz*nsl)   (complex)
 !     output : cout(ldz*nsl) (complex - NOTA BENE: transform is not in-place!)
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nz, nsl, ldz) are stored and re-used if available
 
@@ -248,7 +248,7 @@
 !     2d array: r2d(ldx, ldy) (x first dimension, y second dimension)
 !     (ldx>nx, ldy>ny used on some architectures to reduce memory conflicts)
 !     pl2ix(nx) (optional) is 1 for columns along y to be transformed
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nx,ny,nzl,ldx) are stored and re-used if available
 

FFTXlib/fft_scalar.ESSL.f90

@@ -61,7 +61,7 @@
 !     (ldz>nz is used on some architectures to reduce memory conflicts)
 !     input  :  c(ldz*nsl)   (complex)
 !     output : cout(ldz*nsl) (complex - NOTA BENE: transform is not in-place!)
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nz, nsl, ldz) are stored and re-used if available
 
@@ -192,7 +192,7 @@
 !     2d array: r2d(ldx, ldy) (x first dimension, y second dimension)
 !     (ldx>nx, ldy>ny used on some architectures to reduce memory conflicts)
 !     pl2ix(nx) (optional) is 1 for columns along y to be transformed
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nx,ny,nzl,ldx) are stored and re-used if available
 

FFTXlib/fft_scalar.FFTW.f90

@@ -49,7 +49,7 @@
 !     (ldz>nz is used on some architectures to reduce memory conflicts)
 !     input  :  c(ldz*nsl)   (complex)
 !     output : cout(ldz*nsl) (complex - NOTA BENE: transform is not in-place!)
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nz, nsl, ldz) are stored and re-used if available
 
@@ -194,7 +194,7 @@
 !     2d array: r2d(ldx, ldy) (x first dimension, y second dimension)
 !     (ldx>nx, ldy>ny used on some architectures to reduce memory conflicts)
 !     pl2ix(nx) (optional) is 1 for columns along y to be transformed
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nx,ny,nzl,ldx) are stored and re-used if available
 

FFTXlib/fft_scalar.FFTW3.f90

@@ -65,7 +65,7 @@
 !     (ldz>nz is used on some architectures to reduce memory conflicts)
 !     input  :  c(ldz*nsl)   (complex)
 !     output : cout(ldz*nsl) (complex - NOTA BENE: transform is not in-place!)
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nz, nsl, ldz) are stored and re-used if available
 
@@ -185,7 +185,7 @@
 !     2d array: r2d(ldx, ldy) (x first dimension, y second dimension)
 !     (ldx>nx, ldy>ny used on some architectures to reduce memory conflicts)
 !     pl2ix(nx) (optional) is 1 for columns along y to be transformed
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nx,ny,nzl,ldx) are stored and re-used if available
 

FFTXlib/fft_scalar.SX6.f90

@@ -47,7 +47,7 @@
 !     (ldz>nz is used on some architectures to reduce memory conflicts)
 !     input  :  c(ldz*nsl)   (complex)
 !     output : cout(ldz*nsl) (complex - NOTA BENE: transform is not in-place!)
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nz, nsl, ldz) are stored and re-used if available
 
@@ -166,7 +166,7 @@
 !     2d array: r2d(ldx, ldy) (x first dimension, y second dimension)
 !     (ldx>nx, ldy>ny used on some architectures to reduce memory conflicts)
 !     pl2ix(nx) (optional) is 1 for columns along y to be transformed
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nx,ny,nzl,ldx) are stored and re-used if available
 

FFTXlib/fft_scalar.cuFFT.f90

@@ -33,8 +33,8 @@
 !     ldz >= nz is the distance between sequences to be transformed
 !     (ldz>nz is used on some architectures to reduce memory conflicts)
 !     input  :  c_d(ldz*nsl)   (complex)
-!     ### GPU VERION IN PLACE!!! #### output : cout_d(ldz*nsl) (complex - NOTA BENE: transform is not in-place!)
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     ### GPU VERSION IN PLACE!!! #### output : cout_d(ldz*nsl) (complex - NOTA BENE: transform is not in-place!)
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nz, nsl, ldz) are stored and re-used if available
 #ifdef TRACK_FLOPS
@@ -178,7 +178,7 @@
 !     2d array: r2d(ldx, ldy) (x first dimension, y second dimension)
 !     (ldx>nx, ldy>ny used on some architectures to reduce memory conflicts)
 !     pl2ix(nx) (optional) is 1 for columns along y to be transformed
-!     isign > 0 : forward (f(G)=>f(R)), isign <0 backward (f(R) => f(G))
+!     isign > 0 : backward (f(G)=>f(R)), isign < 0 : forward (f(R) => f(G))
 !     Up to "ndims" initializations (for different combinations of input
 !     parameters nx,ny,nzl,ldx) are stored and re-used if available
 !#ifdef TRACK_FLOPS