SROTMG(3) BLAS routine of NEC Numeric Library Collection SROTMG(3)
NAME
SROTMG - BLAS Level 1. Constructs a modified Givens plane rotation
SYNOPSIS
SUBROUTINE SROTMG ( d1, d2, b1, b2, rparam )
REAL d1, d2, b1, b2, rparam
DESCRIPTION
SROTMG computes the elements of a modified Givens plane rotation
matrix.
ARGUMENTS
d1 (input and output) REAL
First diagonal element.
On input, this value is the first diagonal element of the scal-
ing matrix D. On the first call to the routine, this value is
typically 1.0. Subsequent calls typically use the value from
the previous call. On output, this value is the first diagonal
element of the updated scaling matrix D'.
d2 (input and output) REAL
Second diagonal element.
On input, this is the second diagonal element of the scaling
matrix D. On the first call to the routine, this value is typ-
ically 1.0. Subsequent calls typically use the value from the
previous call. On output, this value is the second diagonal
element of the updated scaling matrix D'.
b1 (input) REAL
x-coordinate.
on input, this value is the x-coordinate of the vector used to
define the angle of rotation, before scaling (multiplying by
the matrix D). On output, this value is the x-coordinate of
the rotated vector, before scaling (multiplying by the matrix
D').
b2 (input) REAL
y-coordinate.
On input, this value is the y-coordinate of the vector used to
define the angle of rotation, before scaling (multiplying by
the matrix D). It is unchanged on output.
rparam (output) REAL
Real array of dimension 5.
This array contains rotation matrix information. The routine
sets up the computed elements in rparam from inputs d1, d2, b1,
and b2.
RETURN VALUES
Meaning of the Output Values:
The output values are returned through arguments d1, d2, b1, and
rparam.
The scaling factors d1 and d2 are updated with each call to the rou-
tine. Although SROTM does not need the updated factors, they are
needed in two other important contexts:
As input for subsequent calls to this routine. As scaling factors for
rotated but unscaled points (x(i), y(i)), which are output from SROTM.
In this second usage, the actual (scaled) points would be given by
(sqrt(d1)*x(i), sqrt(d2)*y(i)). Doing this operation frequently on all
of your points is counterproductive. The main advantage of the modi-
fied rotation algorithm is to reduce the number of operations. If you
fold in the scaling factors after each rotation, you are performing the
same number of operations as in the standard Givens rotation.
These two uses for the scaling factors are mutually exclusive; that is,
if you fold the scaling factors back into all your points, you no
longer need those factors for this routine. After folding in the scal-
ing factors, and before the next call to SROTMG, reset d1 and d2 to
1.0.
On output, b1 represents the new x-coordinate after rotating (but
before scaling) the rotation vector. Although the y-coordinate of this
vector is 0.0 (see the the previous discussion), the corresponding
value b2 is unchanged on output.
The output array argument rparam specifies the format of matrix H, and
it holds the nonunit values of H. This is the only output of this rou-
tine that SROTM requires. Each element of rparam has a specific mean-
ing, as follows:
rparam(1) a flag parameter that specifies how the matrix is stored
0.0 => Off-diagonal elements of H are units.
1.0 => Diagonal elements of H are units.
-1.0 => Rescaling case (see the following subsection).
-2.0 => H is the identity matrix; no rotation needed.
rparam(2) = h if needed
1,1
rparam(3) = h if needed
2,1
rparam(4) = h if needed
1,2
rparam(5) = h if needed
2,2
SEE ALSO
SROT(3), SROTG(3), SROTM(3)
Gentleman, W. M., "Least Squares Computations by Givens Transformations
Without Square Roots," Journal of the Institute for Mathematical Appli-
cations 12 (1973), pp. 329 - 336.
Lawson, C., Hanson, R., Kincaid, D., and Krogh, F., "Basic Linear Alge-
bra Subprograms for Fortran Usage," ACM Transactions on Mathematical
Software, 5 (1979),
BLAS routine SROTMG(3)