* src/frontend/interp.c: Broken up into individual files and

moved into their own subdirectory: src/maths/poly.

	* src/maths/poly/.cvsignore src/maths/poly/Makefile.am
	src/maths/poly/interpolate.c src/maths/poly/interpolate.h
	src/maths/poly/poly.h src/maths/poly/polyderiv.c
	src/maths/poly/polyderiv.h src/maths/poly/polyeval.c
	src/maths/poly/polyeval.h src/maths/poly/polyfit.c
	src/maths/poly/polyfit.h: The resulting files.

	* src/Makefile.am src/maths/Makefile.am: Updates for the new
	library.
This commit is contained in:
arno 2000-05-13 10:56:58 +00:00
parent be76b6dd44
commit 5492f4cb89
14 changed files with 334 additions and 275 deletions

View File

@ -75,6 +75,7 @@ ngspice_LDADD = \
hlp/libhlp.a \
circuit/libinp.a \
maths/cmaths/libcmaths.a \
maths/poly/libpoly.a \
maths/ni/libni.a \
maths/sparse/libsparse.a \
misc/libmisc.a
@ -97,6 +98,7 @@ nutmeg_LDADD = \
parser/libparser.a \
hlp/libhlp.a \
maths/cmaths/libcmaths.a \
maths/poly/libpoly.a \
misc/libmisc.a

View File

@ -14,271 +14,6 @@ Author: 1985 Wayne A. Christopher, U. C. Berkeley CAD Group
#include "interp.h"
/* Interpolate data from oscale to nscale. data is assumed to be olen long,
* ndata will be nlen long. Returns FALSE if the scales are too strange
* to deal with. Note that we are guaranteed that either both scales are
* strictly increasing or both are strictly decreasing.
*/
/* static declarations */
static int putinterval(double *poly, int degree, double *nvec, int last, double *nscale,
int nlen, double oval, int sign);
bool
ft_interpolate(double *data, double *ndata, double *oscale, int olen, double *nscale, int nlen, int degree)
{
double *result, *scratch, *xdata, *ydata;
int sign, lastone, i, l;
if ((olen < 2) || (nlen < 2)) {
fprintf(cp_err, "Error: lengths too small to interpolate.\n");
return (FALSE);
}
if ((degree < 1) || (degree > olen)) {
fprintf(cp_err, "Error: degree is %d, can't interpolate.\n",
degree);
return (FALSE);
}
if (oscale[1] < oscale[0])
sign = -1;
else
sign = 1;
scratch = (double *) tmalloc((degree + 1) * (degree + 2) *
sizeof (double));
result = (double *) tmalloc((degree + 1) * sizeof (double));
xdata = (double *) tmalloc((degree + 1) * sizeof (double));
ydata = (double *) tmalloc((degree + 1) * sizeof (double));
/* Deal with the first degree pieces. */
bcopy((char *) data, (char *) ydata, (degree + 1) * sizeof (double));
bcopy((char *) oscale, (char *) xdata, (degree + 1) * sizeof (double));
while (!ft_polyfit(xdata, ydata, result, degree, scratch)) {
/* If it doesn't work this time, bump the interpolation
* degree down by one.
*/
if (--degree == 0) {
fprintf(cp_err, "ft_interpolate: Internal Error.\n");
return (FALSE);
}
}
/* Add this part of the curve. What we do is evaluate the polynomial
* at those points between the last one and the one that is greatest,
* without being greater than the leftmost old scale point, or least
* if the scale is decreasing at the end of the interval we are looking
* at.
*/
lastone = -1;
for (i = 0; i < degree; i++) {
lastone = putinterval(result, degree, ndata, lastone,
nscale, nlen, xdata[i], sign);
}
/* Now plot the rest, piece by piece. l is the
* last element under consideration.
*/
for (l = degree + 1; l < olen; l++) {
/* Shift the old stuff by one and get another value. */
for (i = 0; i < degree; i++) {
xdata[i] = xdata[i + 1];
ydata[i] = ydata[i + 1];
}
ydata[i] = data[l];
xdata[i] = oscale[l];
while (!ft_polyfit(xdata, ydata, result, degree, scratch)) {
if (--degree == 0) {
fprintf(cp_err,
"interpolate: Internal Error.\n");
return (FALSE);
}
}
lastone = putinterval(result, degree, ndata, lastone,
nscale, nlen, xdata[i], sign);
}
if (lastone < nlen - 1) /* ??? */
ndata[nlen - 1] = data[olen - 1];
tfree(scratch);
tfree(xdata);
tfree(ydata);
tfree(result);
return (TRUE);
}
/* Takes n = (degree+1) doubles, and fills in result with the n coefficients
* of the polynomial that will fit them. It also takes a pointer to an
* array of n ^ 2 + n doubles to use for scratch -- we want to make this
* fast and avoid doing mallocs for each call.
*/
bool
ft_polyfit(double *xdata, double *ydata, double *result, int degree, double *scratch)
{
register double *mat1 = scratch;
register int l, k, j, i;
register int n = degree + 1;
register double *mat2 = scratch + n * n; /* XXX These guys are hacks! */
double d;
/*
fprintf(cp_err, "n = %d, xdata = ( ", n);
for (i = 0; i < n; i++)
fprintf(cp_err, "%G ", xdata[i]);
fprintf(cp_err, ")\n");
fprintf(cp_err, "ydata = ( ");
for (i = 0; i < n; i++)
fprintf(cp_err, "%G ", ydata[i]);
fprintf(cp_err, ")\n");
*/
bzero((char *) result, n * sizeof(double));
bzero((char *) mat1, n * n * sizeof (double));
bcopy((char *) ydata, (char *) mat2, n * sizeof (double));
/* Fill in the matrix with x^k for 0 <= k <= degree for each point */
l = 0;
for (i = 0; i < n; i++) {
d = 1.0;
for (j = 0; j < n; j++) {
mat1[l] = d;
d *= xdata[i];
l += 1;
}
}
/* Do Gauss-Jordan elimination on mat1. */
for (i = 0; i < n; i++) {
int lindex;
double largest;
/* choose largest pivot */
for (j=i, largest = mat1[i * n + i], lindex = i; j < n; j++) {
if (fabs(mat1[j * n + i]) > largest) {
largest = fabs(mat1[j * n + i]);
lindex = j;
}
}
if (lindex != i) {
/* swap rows i and lindex */
for (k = 0; k < n; k++) {
d = mat1[i * n + k];
mat1[i * n + k] = mat1[lindex * n + k];
mat1[lindex * n + k] = d;
}
d = mat2[i];
mat2[i] = mat2[lindex];
mat2[lindex] = d;
}
/* Make sure we have a non-zero pivot. */
if (mat1[i * n + i] == 0.0) {
/* this should be rotated. */
return (FALSE);
}
for (j = i + 1; j < n; j++) {
d = mat1[j * n + i] / mat1[i * n + i];
for (k = 0; k < n; k++)
mat1[j * n + k] -= d * mat1[i * n + k];
mat2[j] -= d * mat2[i];
}
}
for (i = n - 1; i > 0; i--)
for (j = i - 1; j >= 0; j--) {
d = mat1[j * n + i] / mat1[i * n + i];
for (k = 0; k < n; k++)
mat1[j * n + k] -=
d * mat1[i * n + k];
mat2[j] -= d * mat2[i];
}
/* Now write the stuff into the result vector. */
for (i = 0; i < n; i++) {
result[i] = mat2[i] / mat1[i * n + i];
/* printf(cp_err, "result[%d] = %G\n", i, result[i]);*/
}
#define ABS_TOL 0.001
#define REL_TOL 0.001
/* Let's check and make sure the coefficients are ok. If they aren't,
* just return FALSE. This is not the best way to do it.
*/
for (i = 0; i < n; i++) {
d = ft_peval(xdata[i], result, degree);
if (fabs(d - ydata[i]) > ABS_TOL) {
/*
fprintf(cp_err,
"Error: polyfit: x = %le, y = %le, int = %le\n",
xdata[i], ydata[i], d);
printmat("mat1", mat1, n, n);
printmat("mat2", mat2, n, 1);
*/
return (FALSE);
} else if (fabs(d - ydata[i]) / (fabs(d) > ABS_TOL ? fabs(d) :
ABS_TOL) > REL_TOL) {
/*
fprintf(cp_err,
"Error: polyfit: x = %le, y = %le, int = %le\n",
xdata[i], ydata[i], d);
printmat("mat1", mat1, n, n);
printmat("mat2", mat2, n, 1);
*/
return (FALSE);
}
}
return (TRUE);
}
/* Returns thestrchr of the last element that was calculated. oval is the
* value of the old scale at the end of the interval that is being interpolated
* from, and sign is 1 if the old scale was increasing, and -1 if it was
* decreasing.
*/
static int
putinterval(double *poly, int degree, double *nvec, int last, double *nscale, int nlen, double oval, int sign)
{
int end, i;
/* See how far we have to go. */
for (end = last + 1; end < nlen; end++)
if (nscale[end] * sign > oval * sign)
break;
end--;
for (i = last + 1; i <= end; i++)
nvec[i] = ft_peval(nscale[i], poly, degree);
return (end);
}
double
ft_peval(double x, double *coeffs, int degree)
{
double y;
int i;
if (!coeffs)
return 0.0; /* XXX Should not happen */
y = coeffs[degree]; /* there are (degree+1) coeffs */
for (i = degree - 1; i >= 0; i--) {
y *= x;
y += coeffs[i];
}
return y;
}
void
lincopy(struct dvec *ov, double *newscale, int newlen, struct dvec *oldscale)
{
@ -311,12 +46,3 @@ lincopy(struct dvec *ov, double *newscale, int newlen, struct dvec *oldscale)
return;
}
void
ft_polyderiv(double *coeffs, int degree)
{
int i;
for (i = 0; i < degree; i++) {
coeffs[i] = (i + 1) * coeffs[i + 1];
}
}

View File

@ -1,4 +1,4 @@
# Process this file with automake
SUBDIRS = cmaths ni sparse
SUBDIRS = cmaths ni sparse poly
MAINTAINERCLEANFILES = Makefile.in

View File

@ -0,0 +1,3 @@
Makefile.in
Makefile
.deps

View File

@ -0,0 +1,18 @@
## Process this file with automake to produce Makefile.in
noinst_LIBRARIES = libpoly.a
libpoly_a_SOURCES = \
interpolate.c \
interpolate.h \
polyfit.c \
polyfit.h \
polyderiv.c \
polyderiv.h \
polyeval.c \
polyeval.h
INCLUDES = -I$(top_srcdir)/src/include
MAINTAINERCLEANFILES = Makefile.in

View File

@ -0,0 +1,122 @@
#include <config.h>
#include <ngspice.h>
#include <cpdefs.h>
#include "interpolate.h"
#include "polyeval.h"
#include "polyfit.h"
/* Returns thestrchr of the last element that was calculated. oval is
* the value of the old scale at the end of the interval that is being
* interpolated from, and sign is 1 if the old scale was increasing,
* and -1 if it was decreasing. */
static int
putinterval(double *poly, int degree, double *nvec,
int last, double *nscale, int nlen, double oval, int sign)
{
int end, i;
/* See how far we have to go. */
for (end = last + 1; end < nlen; end++)
if (nscale[end] * sign > oval * sign)
break;
end--;
for (i = last + 1; i <= end; i++)
nvec[i] = ft_peval(nscale[i], poly, degree);
return (end);
}
/* Interpolate data from oscale to nscale. data is assumed to be olen long,
* ndata will be nlen long. Returns FALSE if the scales are too strange
* to deal with. Note that we are guaranteed that either both scales are
* strictly increasing or both are strictly decreasing.
*/
bool
ft_interpolate(double *data, double *ndata, double *oscale, int olen,
double *nscale, int nlen, int degree)
{
double *result, *scratch, *xdata, *ydata;
int sign, lastone, i, l;
if ((olen < 2) || (nlen < 2)) {
fprintf(cp_err, "Error: lengths too small to interpolate.\n");
return (FALSE);
}
if ((degree < 1) || (degree > olen)) {
fprintf(cp_err, "Error: degree is %d, can't interpolate.\n",
degree);
return (FALSE);
}
if (oscale[1] < oscale[0])
sign = -1;
else
sign = 1;
scratch = (double *) tmalloc((degree + 1) * (degree + 2) *
sizeof (double));
result = (double *) tmalloc((degree + 1) * sizeof (double));
xdata = (double *) tmalloc((degree + 1) * sizeof (double));
ydata = (double *) tmalloc((degree + 1) * sizeof (double));
/* Deal with the first degree pieces. */
bcopy((char *) data, (char *) ydata, (degree + 1) * sizeof (double));
bcopy((char *) oscale, (char *) xdata, (degree + 1) * sizeof (double));
while (!ft_polyfit(xdata, ydata, result, degree, scratch)) {
/* If it doesn't work this time, bump the interpolation
* degree down by one.
*/
if (--degree == 0) {
fprintf(cp_err, "ft_interpolate: Internal Error.\n");
return (FALSE);
}
}
/* Add this part of the curve. What we do is evaluate the polynomial
* at those points between the last one and the one that is greatest,
* without being greater than the leftmost old scale point, or least
* if the scale is decreasing at the end of the interval we are looking
* at.
*/
lastone = -1;
for (i = 0; i < degree; i++) {
lastone = putinterval(result, degree, ndata, lastone,
nscale, nlen, xdata[i], sign);
}
/* Now plot the rest, piece by piece. l is the
* last element under consideration.
*/
for (l = degree + 1; l < olen; l++) {
/* Shift the old stuff by one and get another value. */
for (i = 0; i < degree; i++) {
xdata[i] = xdata[i + 1];
ydata[i] = ydata[i + 1];
}
ydata[i] = data[l];
xdata[i] = oscale[l];
while (!ft_polyfit(xdata, ydata, result, degree, scratch)) {
if (--degree == 0) {
fprintf(cp_err,
"interpolate: Internal Error.\n");
return (FALSE);
}
}
lastone = putinterval(result, degree, ndata, lastone,
nscale, nlen, xdata[i], sign);
}
if (lastone < nlen - 1) /* ??? */
ndata[nlen - 1] = data[olen - 1];
tfree(scratch);
tfree(xdata);
tfree(ydata);
tfree(result);
return (TRUE);
}

View File

@ -0,0 +1,8 @@
#ifndef _INTERPOLATE_H
#define _INTERPOLATE_H
#include <bool.h>
bool ft_interpolate(double *data, double *ndata, double *oscale, int olen, double *nscale, int nlen, int degree);
#endif

12
src/maths/poly/poly.h Normal file
View File

@ -0,0 +1,12 @@
/* The public interface to the polygon library. */
#ifndef _POLY_H
#define _POLY_H
#include "interpolate.h"
#include "polyderiv.h"
#include "polyeval.h"
#include "polyfit.h"
#endif

View File

@ -0,0 +1,11 @@
#include "polyderiv.h"
void
ft_polyderiv(double *coeffs, int degree)
{
int i;
for (i = 0; i < degree; i++) {
coeffs[i] = (i + 1) * coeffs[i + 1];
}
}

View File

@ -0,0 +1,6 @@
#ifndef _POLYDERIV_H
#define _POLYDERIV_H
void ft_polyderiv(double *coeffs, int degree);
#endif

20
src/maths/poly/polyeval.c Normal file
View File

@ -0,0 +1,20 @@
#include "polyeval.h"
double
ft_peval(double x, double *coeffs, int degree)
{
double y;
int i;
if (!coeffs)
return 0.0; /* XXX Should not happen */
y = coeffs[degree]; /* there are (degree+1) coeffs */
for (i = degree - 1; i >= 0; i--) {
y *= x;
y += coeffs[i];
}
return y;
}

View File

@ -0,0 +1,6 @@
#ifndef _POLYEVAL_H
#define _POLYEVAL_H
double ft_peval(double x, double *coeffs, int degree);
#endif

116
src/maths/poly/polyfit.c Normal file
View File

@ -0,0 +1,116 @@
#include <math.h>
#include "polyfit.h"
#include "polyeval.h"
/* Takes n = (degree+1) doubles, and fills in result with the n
* coefficients of the polynomial that will fit them. It also takes a
* pointer to an array of n ^ 2 + n doubles to use for scratch -- we
* want to make this fast and avoid doing mallocs for each call. */
bool
ft_polyfit(double *xdata, double *ydata, double *result,
int degree, double *scratch)
{
double *mat1 = scratch;
int l, k, j, i;
int n = degree + 1;
double *mat2 = scratch + n * n; /* XXX These guys are hacks! */
double d;
memset((char *) result, 0, n * sizeof(double));
memset((char *) mat1, 0, n * n * sizeof (double));
memcpy((char *) ydata, (char *) mat2, n * sizeof (double));
/* Fill in the matrix with x^k for 0 <= k <= degree for each point */
l = 0;
for (i = 0; i < n; i++) {
d = 1.0;
for (j = 0; j < n; j++) {
mat1[l] = d;
d *= xdata[i];
l += 1;
}
}
/* Do Gauss-Jordan elimination on mat1. */
for (i = 0; i < n; i++) {
int lindex;
double largest;
/* choose largest pivot */
for (j=i, largest = mat1[i * n + i], lindex = i; j < n; j++) {
if (fabs(mat1[j * n + i]) > largest) {
largest = fabs(mat1[j * n + i]);
lindex = j;
}
}
if (lindex != i) {
/* swap rows i and lindex */
for (k = 0; k < n; k++) {
d = mat1[i * n + k];
mat1[i * n + k] = mat1[lindex * n + k];
mat1[lindex * n + k] = d;
}
d = mat2[i];
mat2[i] = mat2[lindex];
mat2[lindex] = d;
}
/* Make sure we have a non-zero pivot. */
if (mat1[i * n + i] == 0.0) {
/* this should be rotated. */
return (FALSE);
}
for (j = i + 1; j < n; j++) {
d = mat1[j * n + i] / mat1[i * n + i];
for (k = 0; k < n; k++)
mat1[j * n + k] -= d * mat1[i * n + k];
mat2[j] -= d * mat2[i];
}
}
for (i = n - 1; i > 0; i--)
for (j = i - 1; j >= 0; j--) {
d = mat1[j * n + i] / mat1[i * n + i];
for (k = 0; k < n; k++)
mat1[j * n + k] -=
d * mat1[i * n + k];
mat2[j] -= d * mat2[i];
}
/* Now write the stuff into the result vector. */
for (i = 0; i < n; i++) {
result[i] = mat2[i] / mat1[i * n + i];
/* printf(cp_err, "result[%d] = %G\n", i, result[i]);*/
}
#define ABS_TOL 0.001
#define REL_TOL 0.001
/* Let's check and make sure the coefficients are ok. If they aren't,
* just return FALSE. This is not the best way to do it.
*/
for (i = 0; i < n; i++) {
d = ft_peval(xdata[i], result, degree);
if (fabs(d - ydata[i]) > ABS_TOL) {
/*
fprintf(cp_err,
"Error: polyfit: x = %le, y = %le, int = %le\n",
xdata[i], ydata[i], d);
printmat("mat1", mat1, n, n);
printmat("mat2", mat2, n, 1);
*/
return (FALSE);
} else if (fabs(d - ydata[i]) / (fabs(d) > ABS_TOL ? fabs(d) :
ABS_TOL) > REL_TOL) {
/*
fprintf(cp_err,
"Error: polyfit: x = %le, y = %le, int = %le\n",
xdata[i], ydata[i], d);
printmat("mat1", mat1, n, n);
printmat("mat2", mat2, n, 1);
*/
return (FALSE);
}
}
return (TRUE);
}

9
src/maths/poly/polyfit.h Normal file
View File

@ -0,0 +1,9 @@
#ifndef _POLYFIT_H
#define _POLYFIT_H
#include <bool.h>
bool ft_polyfit(double *xdata, double *ydata, double *result,
int degree, double *scratch);
#endif