c06pqf provides a simpler interface for both forward and backward transforms. c06ppf returns real sequences in row order at the expense of efficiency.
Old Code
Call c06fqf(m,n,x,init,trig,work,ifail)
New Code
Call c06pqf('B',n,m,x,work,ifail)
where the dimension of work has been extended from ${\mathbf{m}}\times {\mathbf{n}}$ to ${\mathbf{m}}\times {\mathbf{n}}+2\times {\mathbf{n}}$ (to include trig) and the dimension of the array x has been extended from the original ${\mathbf{n}}\times {\mathbf{m}}$ to $({\mathbf{n}}+2)\times {\mathbf{m}}$.
The input values x are stored in a different order with real and imaginary parts of each Hermitian sequence stored contiguously.
The mapping of input elements is as follows:
(Here x begins at element zero ${\mathbf{x}}\left(0\right)$.)
for $\mathit{j}=0,1,\dots ,{\mathbf{m}}-1$
$J=j\times ({\mathbf{n}}+2)$;
${\mathbf{x}}\left(J+2\times \mathit{i}\right)\leftarrow {\mathbf{x}}\left(\mathit{i}\times {\mathbf{m}}+j\right)$, for $\mathit{i}=0,1,\dots ,{\mathbf{n}}/2$ [real parts];
${\mathbf{x}}\left(J+2\times \mathit{i}+1\right)\leftarrow {\mathbf{x}}\left(({\mathbf{n}}-\mathit{i})\times {\mathbf{m}}+j\right)$, for $\mathit{i}=1,2,\dots ,({\mathbf{n}}-1)/2$ [imaginary parts];
${\mathbf{x}}\left(J+1\right)$ must be zero;
${\mathbf{x}}\left(J+{\mathbf{n}}+1\right)$ must zero when n is even.
The output values are stored slightly differently to allow for two extra storage spaces at the end of each sequence.
The mapping of output elements is as follows:
(Here x begins at element zero ${\mathbf{x}}\left(0\right)$.)
for $\mathit{j}=0,1,\dots ,{\mathbf{m}}-1$
$J=j\times ({\mathbf{n}}+2)$;
${\mathbf{x}}\left(J+\mathit{i}\right)\leftarrow {\mathbf{x}}\left(\mathit{i}\times {\mathbf{m}}+j\right)$, for $\mathit{i}=0,1,\dots ,{\mathbf{n}}-1$;
${\mathbf{x}}\left(J+{\mathbf{n}}\right)$
and
${\mathbf{x}}\left(J+{\mathbf{n}}+1\right)$ will be set to zero.
c06fqf computes the discrete Fourier transforms of $m$ Hermitian sequences, each containing $n$ complex data values. This routine is designed to be particularly efficient on vector processors.
The routine may be called by the names c06fqf or nagf_sum_withdraw_fft_hermitian_1d_multi_rfmt.
3Description
Given $m$ Hermitian sequences of $n$ complex data values ${z}_{\mathit{j}}^{\mathit{p}}$, for $\mathit{j}=0,1,\dots ,n-1$ and $\mathit{p}=1,2,\dots ,m$, c06fqf simultaneously calculates the Fourier transforms of all the sequences defined by
To compute this form, this routine should be preceded by forming the complex conjugates of the ${\hat{z}}_{k}^{p}$; that is $x\left(\mathit{k}\right)=-x\left(\mathit{k}\right)$, for $\mathit{k}=(n/2+1)\times m+1,\dots ,m\times n$.
The routine uses a variant of the fast Fourier transform (FFT) algorithm (see Brigham (1974)) known as the Stockham self-sorting algorithm, which is described in Temperton (1983). Special coding is provided for the factors $2$, $3$, $4$, $5$ and $6$. This routine is designed to be particularly efficient on vector processors, and it becomes especially fast as $m$, the number of transforms to be computed in parallel, increases.
4References
Brigham E O (1974) The Fast Fourier Transform Prentice–Hall
Temperton C (1983) Fast mixed-radix real Fourier transforms J. Comput. Phys.52 340–350
5Arguments
1: $\mathbf{m}$ – IntegerInput
On entry: $m$, the number of sequences to be transformed.
Constraint:
${\mathbf{m}}\ge 1$.
2: $\mathbf{n}$ – IntegerInput
On entry: $n$, the number of data values in each sequence.
Constraint:
${\mathbf{n}}\ge 1$.
3: $\mathbf{x}\left({\mathbf{m}}\times {\mathbf{n}}\right)$ – Real (Kind=nag_wp) arrayInput/Output
On entry: the data must be stored in x as if in a two-dimensional array of dimension $(1:{\mathbf{m}},0:{\mathbf{n}}-1)$; each of the $m$ sequences is stored in a row of the array in Hermitian form. If the $n$ data values ${z}_{j}^{p}$ are written as ${x}_{j}^{p}+i{y}_{j}^{p}$, then for $0\le j\le n/2$, ${x}_{j}^{p}$ is contained in ${\mathbf{x}}(p,j)$, and for $1\le j\le (n-1)/2$, ${y}_{j}^{p}$ is contained in ${\mathbf{x}}(p,n-j)$. (See also Section 2.1.2 in the C06 Chapter Introduction.)
On exit: the components of the $m$ discrete Fourier transforms, stored as if in a two-dimensional array of dimension $(1:{\mathbf{m}},0:{\mathbf{n}}-1)$. Each of the $m$ transforms is stored as a row of the array, overwriting the corresponding original sequence.
If the $n$ components of the discrete Fourier transform are denoted by ${\hat{x}}_{\mathit{k}}^{p}$, for $\mathit{k}=0,1,\dots ,n-1$, the $mn$ elements of the array x contain the values
On entry: indicates whether trigonometric coefficients are to be calculated.
${\mathbf{init}}=\text{'I'}$
Calculate the required trigonometric coefficients for the given value of $n$, and store in the array trig.
${\mathbf{init}}=\text{'S'}$ or $\text{'R'}$
The required trigonometric coefficients are assumed to have been calculated and stored in the array trig in a prior call to one of c06fpforc06fqf. The routine performs a simple check that the current value of $n$ is consistent with the values stored in trig.
Constraint:
${\mathbf{init}}=\text{'I'}$, $\text{'S'}$ or $\text{'R'}$.
5: $\mathbf{trig}\left(2\times {\mathbf{n}}\right)$ – Real (Kind=nag_wp) arrayInput/Output
On entry: if ${\mathbf{init}}=\text{'S'}$ or $\text{'R'}$, trig must contain the required trigonometric coefficients that have been previously calculated. Otherwise trig need not be set.
On exit: contains the required coefficients (computed by the routine if ${\mathbf{init}}=\text{'I'}$).
6: $\mathbf{work}\left({\mathbf{m}}\times {\mathbf{n}}\right)$ – Real (Kind=nag_wp) arrayWorkspace
7: $\mathbf{ifail}$ – IntegerInput/Output
On entry: ifail must be set to $0$, $\mathrm{-1}$ or $1$ to set behaviour on detection of an error; these values have no effect when no error is detected.
A value of $0$ causes the printing of an error message and program execution will be halted; otherwise program execution continues. A value of $\mathrm{-1}$ means that an error message is printed while a value of $1$ means that it is not.
If halting is not appropriate, the value $\mathrm{-1}$ or $1$ is recommended. If message printing is undesirable, then the value $1$ is recommended. Otherwise, the value $0$ is recommended. When the value $-\mathbf{1}$ or $\mathbf{1}$ is used it is essential to test the value of ifail on exit.
On exit: ${\mathbf{ifail}}={\mathbf{0}}$ unless the routine detects an error or a warning has been flagged (see Section 6).
6Error Indicators and Warnings
If on entry ${\mathbf{ifail}}=0$ or $\mathrm{-1}$, explanatory error messages are output on the current error message unit (as defined by x04aaf).
Errors or warnings detected by the routine:
${\mathbf{ifail}}=1$
On entry, ${\mathbf{m}}=\u27e8\mathit{\text{value}}\u27e9$.
Constraint: ${\mathbf{m}}\ge 1$.
${\mathbf{ifail}}=2$
On entry, ${\mathbf{n}}=\u27e8\mathit{\text{value}}\u27e9$.
Constraint: ${\mathbf{n}}\ge 1$.
${\mathbf{ifail}}=3$
On entry, ${\mathbf{init}}=\u27e8\mathit{\text{value}}\u27e9$.
Constraint: ${\mathbf{init}}=\text{'I'}$, $\text{'S'}$ or $\text{'R'}$.
${\mathbf{ifail}}=5$
On entry, ${\mathbf{init}}=\u27e8\mathit{\text{value}}\u27e9$ but n and trig array incompatible.
${\mathbf{ifail}}=-99$
An unexpected error has been triggered by this routine. Please
contact NAG.
See Section 7 in the Introduction to the NAG Library FL Interface for further information.
${\mathbf{ifail}}=-399$
Your licence key may have expired or may not have been installed correctly.
See Section 8 in the Introduction to the NAG Library FL Interface for further information.
${\mathbf{ifail}}=-999$
Dynamic memory allocation failed.
See Section 9 in the Introduction to the NAG Library FL Interface for further information.
7Accuracy
Some indication of accuracy can be obtained by performing a subsequent inverse transform and comparing the results with the original sequence (in exact arithmetic they would be identical).
8Parallelism and Performance
Background information to multithreading can be found in the Multithreading documentation.
c06fqf is threaded by NAG for parallel execution in multithreaded implementations of the NAG Library.
c06fqf makes calls to BLAS and/or LAPACK routines, which may be threaded within the vendor library used by this implementation. Consult the documentation for the vendor library for further information.
Please consult the X06 Chapter Introduction for information on how to control and interrogate the OpenMP environment used within this routine. Please also consult the Users' Note for your implementation for any additional implementation-specific information.
9Further Comments
The time taken by c06fqf is approximately proportional to $nm\mathrm{log}\left(n\right)$, but also depends on the factors of $n$. c06fqf is fastest if the only prime factors of $n$ are $2$, $3$ and $5$, and is particularly slow if $n$ is a large prime, or has large prime factors.
10Example
This example reads in sequences of real data values which are assumed to be Hermitian sequences of complex data stored in Hermitian form. The sequences are expanded into full complex form and printed. The discrete Fourier transforms are then computed (using c06fqf) and printed out. Inverse transforms are then calculated by conjugating and calling c06fpf showing that the original sequences are restored.