dcnvcor(3P) Sun Performance Library dcnvcor(3P)NAMEdcnvcor - compute the convolution or correlation of real vectors
SYNOPSIS
SUBROUTINE DCNVCOR(CNVCOR, FOUR, NX, X, IFX, INCX, NY, NPRE, M, Y,
IFY, INC1Y, INC2Y, NZ, K, Z, IFZ, INC1Z, INC2Z, WORK, LWORK)
CHARACTER * 1 CNVCOR, FOUR
INTEGER NX, IFX, INCX, NY, NPRE, M, IFY, INC1Y, INC2Y, NZ, K, IFZ,
INC1Z, INC2Z, LWORK
DOUBLE PRECISION X(*), Y(*), Z(*), WORK(*)
SUBROUTINE DCNVCOR_64(CNVCOR, FOUR, NX, X, IFX, INCX, NY, NPRE, M, Y,
IFY, INC1Y, INC2Y, NZ, K, Z, IFZ, INC1Z, INC2Z, WORK, LWORK)
CHARACTER * 1 CNVCOR, FOUR
INTEGER*8 NX, IFX, INCX, NY, NPRE, M, IFY, INC1Y, INC2Y, NZ, K, IFZ,
INC1Z, INC2Z, LWORK
DOUBLE PRECISION X(*), Y(*), Z(*), WORK(*)
F95 INTERFACE
SUBROUTINE CNVCOR(CNVCOR, FOUR, [NX], X, IFX, [INCX], NY, NPRE, M, Y,
IFY, INC1Y, INC2Y, NZ, K, Z, IFZ, INC1Z, INC2Z, WORK, [LWORK])
CHARACTER(LEN=1) :: CNVCOR, FOUR
INTEGER :: NX, IFX, INCX, NY, NPRE, M, IFY, INC1Y, INC2Y, NZ, K, IFZ,
INC1Z, INC2Z, LWORK
REAL(8), DIMENSION(:) :: X, Y, Z, WORK
SUBROUTINE CNVCOR_64(CNVCOR, FOUR, [NX], X, IFX, [INCX], NY, NPRE, M,
Y, IFY, INC1Y, INC2Y, NZ, K, Z, IFZ, INC1Z, INC2Z, WORK, [LWORK])
CHARACTER(LEN=1) :: CNVCOR, FOUR
INTEGER(8) :: NX, IFX, INCX, NY, NPRE, M, IFY, INC1Y, INC2Y, NZ, K,
IFZ, INC1Z, INC2Z, LWORK
REAL(8), DIMENSION(:) :: X, Y, Z, WORK
C INTERFACE
#include <sunperf.h>
void dcnvcor(char cnvcor, char four, int nx, double *x, int ifx, int
incx, int ny, int npre, int m, double *y, int ify, int inc1y,
int inc2y, int nz, int k, double *z, int ifz, int inc1z, int
inc2z, double *work, int lwork);
void dcnvcor_64(char cnvcor, char four, long nx, double *x, long ifx,
long incx, long ny, long npre, long m, double *y, long ify,
long inc1y, long inc2y, long nz, long k, double *z, long ifz,
long inc1z, long inc2z, double *work, long lwork);
PURPOSEdcnvcor computes the convolution or correlation of real vectors.
ARGUMENTS
CNVCOR (input)
´V' or 'v' if convolution is desired, 'R' or 'r' if correla‐
tion is desired.
FOUR (input)
´T' or 't' if the Fourier transform method is to be used, 'D'
or 'd' if the computation should be done directly from the
definition. The Fourier transform method is generally
faster, but it may introduce noticeable errors into certain
results, notably when both the filter and data vectors con‐
sist entirely of integers or vectors where elements of either
the filter vector or a given data vector differ significantly
in magnitude from the 1-norm of the vector.
NX (input)
Length of the filter vector. NX >= 0. DCNVCOR will return
immediately if NX = 0.
X (input)
Filter vector.
IFX (input)
Index of the first element of X. NX >= IFX >= 1.
INCX (input)
Stride between elements of the filter vector in X. INCX > 0.
NY (input)
Length of the input vectors. NY >= 0. DCNVCOR will return
immediately if NY = 0.
NPRE (input)
The number of implicit zeros prepended to the Y vectors.
NPRE >= 0.
M (input)
Number of input vectors. M >= 0. DCNVCOR will return imme‐
diately if M = 0.
Y (input)
Input vectors.
IFY (input)
Index of the first element of Y. NY >= IFY >= 1.
INC1Y (input)
Stride between elements of the input vectors in Y. INC1Y >
0.
INC2Y (input)
Stride between the input vectors in Y. INC2Y > 0.
NZ (input)
Length of the output vectors. NZ >= 0. DCNVCOR will return
immediately if NZ = 0. See the Notes section below for
information about how this argument interacts with NX and NY
to control circular versus end-off shifting.
K (input)
Number of Z vectors. K >= 0. If K = 0 then DCNVCOR will
return immediately. If K < M then only the first K input
vectors will be processed. If K > M then M input vectors
will be processed.
Z (output)
Result vectors.
IFZ (input)
Index of the first element of Z. NZ >= IFZ >= 1.
INC1Z (input)
Stride between elements of the output vectors in Z. INC1Z >
0.
INC2Z (input)
Stride between the output vectors in Z. INC2Z > 0.
WORK (input/output)
Scratch space. Before the first call to DCNVCOR with partic‐
ular values of the integer arguments the first element of
WORK must be set to zero. If WORK is written between calls
to DCNVCOR or if DCNVCOR is called with different values of
the integer arguments then the first element of WORK must
again be set to zero before each call. If WORK has not been
written and the same values of the integer arguments are used
then the first element of WORK to zero. This can avoid cer‐
tain initializations that store their results into WORK, and
avoiding the initialization can make DCNVCOR run faster.
LWORK (input)
Length of WORK. LWORK >= 4*MAX(NX,NY,NZ)+15.
NOTES
If any vector overlaps a writable vector, either because of argument
aliasing or ill-chosen values of the various INC arguments, the results
are undefined and may vary from one run to the next.
The most common form of the computation, and the case that executes
fastest, is applying a filter vector X to a series of vectors stored in
the columns of Y with the result placed into the columns of Z. In that
case, INCX = 1, INC1Y = 1, INC2Y >= NY, INC1Z = 1, INC2Z >= NZ.
Another common form is applying a filter vector X to a series of vec‐
tors stored in the rows of Y and store the result in the row of Z, in
which case INCX = 1, INC1Y >= NY, INC2Y = 1, INC1Z >= NZ, and INC2Z =
1.
6 Mar 2009 dcnvcor(3P)
