eccodes/src/grib_iterator_class_gaussian.c

178 lines
5.1 KiB
C

/*
* Copyright 2005-2017 ECMWF.
*
* This software is licensed under the terms of the Apache Licence Version 2.0
* which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
*
* In applying this licence, ECMWF does not waive the privileges and immunities granted to it by
* virtue of its status as an intergovernmental organisation nor does it submit to any jurisdiction.
*/
/**************************************
* Enrico Fucile
**************************************/
#include "grib_api_internal.h"
#include <math.h>
/*
This is used by make_class.pl
START_CLASS_DEF
CLASS = iterator
SUPER = grib_iterator_class_regular
IMPLEMENTS = init
END_CLASS_DEF
*/
/* START_CLASS_IMP */
/*
Don't edit anything between START_CLASS_IMP and END_CLASS_IMP
Instead edit values between START_CLASS_DEF and END_CLASS_DEF
or edit "iterator.class" and rerun ./make_class.pl
*/
static void init_class (grib_iterator_class*);
static int init (grib_iterator* i,grib_handle*,grib_arguments*);
typedef struct grib_iterator_gaussian{
grib_iterator it;
/* Members defined in gen */
long carg;
const char* missingValue;
/* Members defined in regular */
double *las;
double *los;
long nap;
long nam;
long iScansNegatively;
long isRotated;
double angleOfRotation;
double southPoleLat;
double southPoleLon;
/* Members defined in gaussian */
} grib_iterator_gaussian;
extern grib_iterator_class* grib_iterator_class_regular;
static grib_iterator_class _grib_iterator_class_gaussian = {
&grib_iterator_class_regular, /* super */
"gaussian", /* name */
sizeof(grib_iterator_gaussian),/* size of instance */
0, /* inited */
&init_class, /* init_class */
&init, /* constructor */
0, /* destructor */
0, /* Next Value */
0, /* Previous Value */
0, /* Reset the counter */
0, /* has next values */
};
grib_iterator_class* grib_iterator_class_gaussian = &_grib_iterator_class_gaussian;
static void init_class(grib_iterator_class* c)
{
c->next = (*(c->super))->next;
c->previous = (*(c->super))->previous;
c->reset = (*(c->super))->reset;
c->has_next = (*(c->super))->has_next;
}
/* END_CLASS_IMP */
static void binary_search(double xx[], const unsigned long n, double x, unsigned long *j);
static int init(grib_iterator* i,grib_handle* h,grib_arguments *args){
grib_iterator_gaussian* self = (grib_iterator_gaussian*)i;
double *lats;
double laf; /* latitude of first point in degrees */
double lal; /* latitude of last point in degrees */
long trunc; /* number of parallels between a pole and the equator */
long lai;
long jScansPositively=0;
int size;
double start;
unsigned long istart=0;
int ret = GRIB_SUCCESS;
const char* latofirst = grib_arguments_get_name(h,args,self->carg++);
const char* latoflast = grib_arguments_get_name(h,args,self->carg++);
const char* numtrunc = grib_arguments_get_name(h,args,self->carg++);
const char* s_jScansPositively = grib_arguments_get_name(h,args,self->carg++);
if((ret = grib_get_double_internal(h,latofirst, &laf))) return ret;
if((ret = grib_get_double_internal(h,latoflast, &lal))) return ret;
if((ret = grib_get_long_internal(h,numtrunc,&trunc))) return ret;
if((ret = grib_get_long_internal(h,s_jScansPositively,&jScansPositively)))
return ret;
start=laf;
size=trunc*2;
lats = (double*)grib_context_malloc(h->context,size*sizeof(double));
ret = grib_get_gaussian_latitudes(trunc, lats);
if(ret != GRIB_SUCCESS) {
grib_context_log(h->context, GRIB_LOG_ERROR,"error %d calculating gaussian points",ret);
return ret;
}
/*
for(loi=(trunc*2)-1;loi>=0;loi--)
if(fabs(lats[loi] - lal) < glatPrecision) break;
for(j=(trunc*2)-1;j>0;j--) {
if(fabs(lats[j] - laf) < glatPrecision) break;
}
*/
binary_search(lats,size-1,start,&istart);
Assert(istart < size);
if (jScansPositively) {
for(lai=0;lai<self->nam;lai++) {
self->las[lai] = lats[istart--];
/*if (istart<0) istart=size-1; this condition is always FALSE -- 'istart' is unsigned long */
}
} else {
for(lai=0;lai<self->nam;lai++) {
self->las[lai] = lats[istart++];
if (istart>size-1) istart=0;
}
}
grib_context_free(h->context,lats);
return ret;
}
static void binary_search(double xx[], const unsigned long n, double x, unsigned long *j)
{
/*This routine works only on descending ordered arrays*/
#define EPSILON 1e-3
unsigned long ju,jm,jl;
jl=0;
ju=n;
while (ju-jl > 1) {
jm=(ju+jl) >> 1;
if (fabs(x-xx[jm]) < EPSILON) {
/* found something close enough. We're done */
*j=jm;
return;
}
if (x < xx[jm]) jl=jm;
else ju=jm;
}
*j=jl;
}