cosmotool/external/sharp/libsharp/sharp.c

670 lines
19 KiB
C
Raw Normal View History

2012-11-10 13:59:10 +00:00
/*
* This file is part of libsharp.
*
* libsharp is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* libsharp is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with libsharp; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
/*
* libsharp is being developed at the Max-Planck-Institut fuer Astrophysik
* and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt
* (DLR).
*/
/*! \file sharp.c
* Spherical transform library
*
* Copyright (C) 2006-2012 Max-Planck-Society
* \author Martin Reinecke
*/
#include <math.h>
#include "ls_fft.h"
#include "sharp_ylmgen_c.h"
#include "sharp_internal.h"
#include "c_utils.h"
#include "sharp_core.h"
#include "sharp_vecutil.h"
#include "walltime_c.h"
#include "sharp_almhelpers.h"
#include "sharp_geomhelpers.h"
typedef complex double dcmplx;
typedef complex float fcmplx;
static void get_chunk_info (int ndata, int nmult, int *nchunks, int *chunksize)
{
static const int chunksize_min=500, nchunks_max=10;
*chunksize = IMAX(chunksize_min,(ndata+nchunks_max-1)/nchunks_max);
*chunksize = ((*chunksize+nmult-1)/nmult)*nmult;
*nchunks = (ndata+*chunksize-1) / *chunksize;
}
typedef struct
{
double s;
int i;
} idxhelper;
static int idx_compare (const void *xa, const void *xb)
{
const idxhelper *a=xa, *b=xb;
return (a->s > b->s) ? -1 : (a->s < b->s) ? 1 : 0;
}
typedef struct
{
double phi0_;
dcmplx *shiftarr, *work;
int s_shift, s_work;
real_plan plan;
int norot;
} ringhelper;
static void ringhelper_init (ringhelper *self)
{
static ringhelper rh_null = { 0, NULL, NULL, 0, 0, NULL, 0 };
*self = rh_null;
}
static void ringhelper_destroy (ringhelper *self)
{
if (self->plan) kill_real_plan(self->plan);
DEALLOC(self->shiftarr);
DEALLOC(self->work);
ringhelper_init(self);
}
static void ringhelper_update (ringhelper *self, int nph, int mmax, double phi0)
{
self->norot = (fabs(phi0)<1e-14);
if (!(self->norot))
if ((mmax!=self->s_shift-1) || (!FAPPROX(phi0,self->phi0_,1e-12)))
{
RESIZE (self->shiftarr,dcmplx,mmax+1);
self->s_shift = mmax+1;
self->phi0_ = phi0;
for (int m=0; m<=mmax; ++m)
self->shiftarr[m] = cos(m*phi0) + _Complex_I*sin(m*phi0);
}
if (!self->plan) self->plan=make_real_plan(nph);
if (nph!=(int)self->plan->length)
{
kill_real_plan(self->plan);
self->plan=make_real_plan(nph);
}
GROW(self->work,dcmplx,self->s_work,nph);
}
static int ringinfo_compare (const void *xa, const void *xb)
{
const sharp_ringinfo *a=xa, *b=xb;
return (a->sth < b->sth) ? -1 : (a->sth > b->sth) ? 1 : 0;
}
static int ringpair_compare (const void *xa, const void *xb)
{
const sharp_ringpair *a=xa, *b=xb;
if (a->r1.nph==b->r1.nph)
return (a->r1.phi0 < b->r1.phi0) ? -1 :
((a->r1.phi0 > b->r1.phi0) ? 1 :
(a->r1.cth>b->r1.cth ? -1 : 1));
return (a->r1.nph<b->r1.nph) ? -1 : 1;
}
void sharp_make_general_alm_info (int lmax, int nm, int stride, const int *mval,
const ptrdiff_t *mstart, sharp_alm_info **alm_info)
{
sharp_alm_info *info = RALLOC(sharp_alm_info,1);
info->lmax = lmax;
info->nm = nm;
info->mval = RALLOC(int,nm);
info->mvstart = RALLOC(ptrdiff_t,nm);
info->stride = stride;
for (int mi=0; mi<nm; ++mi)
{
info->mval[mi] = mval[mi];
info->mvstart[mi] = mstart[mi];
}
*alm_info = info;
}
void sharp_make_alm_info (int lmax, int mmax, int stride,
const ptrdiff_t *mstart, sharp_alm_info **alm_info)
{
int *mval=RALLOC(int,mmax+1);
for (int i=0; i<=mmax; ++i)
mval[i]=i;
sharp_make_general_alm_info (lmax, mmax+1, stride, mval, mstart, alm_info);
DEALLOC(mval);
}
ptrdiff_t sharp_alm_index (const sharp_alm_info *self, int l, int mi)
{ return self->mvstart[mi]+self->stride*l; }
void sharp_destroy_alm_info (sharp_alm_info *info)
{
DEALLOC (info->mval);
DEALLOC (info->mvstart);
DEALLOC (info);
}
void sharp_make_geom_info (int nrings, const int *nph, const ptrdiff_t *ofs,
const int *stride, const double *phi0, const double *theta,
const double *weight, sharp_geom_info **geom_info)
{
sharp_geom_info *info = RALLOC(sharp_geom_info,1);
sharp_ringinfo *infos = RALLOC(sharp_ringinfo,nrings);
int pos=0;
info->pair=RALLOC(sharp_ringpair,nrings);
info->npairs=0;
*geom_info = info;
for (int m=0; m<nrings; ++m)
{
infos[m].theta = theta[m];
infos[m].cth = cos(theta[m]);
infos[m].sth = sin(theta[m]);
infos[m].weight = weight[m];
infos[m].phi0 = phi0[m];
infos[m].ofs = ofs[m];
infos[m].stride = stride[m];
infos[m].nph = nph[m];
}
qsort(infos,nrings,sizeof(sharp_ringinfo),ringinfo_compare);
while (pos<nrings)
{
info->pair[info->npairs].r1=infos[pos];
if ((pos<nrings-1) && FAPPROX(infos[pos].cth,-infos[pos+1].cth,1e-12))
{
if (infos[pos].cth>0) // make sure northern ring is in r1
info->pair[info->npairs].r2=infos[pos+1];
else
{
info->pair[info->npairs].r1=infos[pos+1];
info->pair[info->npairs].r2=infos[pos];
}
++pos;
}
else
info->pair[info->npairs].r2.nph=-1;
++pos;
++info->npairs;
}
DEALLOC(infos);
qsort(info->pair,info->npairs,sizeof(sharp_ringpair),ringpair_compare);
}
void sharp_destroy_geom_info (sharp_geom_info *geom_info)
{
DEALLOC (geom_info->pair);
DEALLOC (geom_info);
}
static int sharp_get_mmax (int *mval, int nm)
{
int *mcheck=RALLOC(int,nm);
SET_ARRAY(mcheck,0,nm,0);
for (int i=0; i<nm; ++i)
{
int m_cur=mval[i];
UTIL_ASSERT((m_cur>=0) && (m_cur<nm), "m out of range");
UTIL_ASSERT(mcheck[m_cur]==0, "duplicate m value");
mcheck[m_cur]=1;
}
DEALLOC(mcheck);
return nm-1; // FIXME: this looks wrong
}
static void ringhelper_phase2ring (ringhelper *self,
const sharp_ringinfo *info, void *data, int mmax, const dcmplx *phase,
int pstride, sharp_fde fde)
{
int nph = info->nph;
int stride = info->stride;
ringhelper_update (self, nph, mmax, info->phi0);
self->work[0]=phase[0];
SET_ARRAY(self->work,1,nph,0.);
#if 0
if (self->norot)
for (int m=1; m<=mmax; ++m)
{
int idx1 = m%nph;
int idx2 = nph-1-((m-1)%nph);
self->work[idx1]+=phase[m*pstride];
self->work[idx2]+=conj(phase[m*pstride]);
}
else
for (int m=1; m<=mmax; ++m)
{
int idx1 = m%nph;
int idx2 = nph-1-((m-1)%nph);
dcmplx tmp = phase[m*pstride]*self->shiftarr[m];
self->work[idx1]+=tmp;
self->work[idx2]+=conj(tmp);
}
#else
int idx1=1, idx2=nph-1;
for (int m=1; m<=mmax; ++m)
{
dcmplx tmp = phase[m*pstride];
if(!self->norot) tmp*=self->shiftarr[m];
self->work[idx1]+=tmp;
self->work[idx2]+=conj(tmp);
if (++idx1>=nph) idx1=0;
if (--idx2<0) idx2=nph-1;
}
#endif
real_plan_backward_c (self->plan, (double *)(self->work));
if (fde==DOUBLE)
for (int m=0; m<nph; ++m)
((double *)data)[m*stride+info->ofs] += creal(self->work[m]);
else
for (int m=0; m<nph; ++m)
((float *)data)[m*stride+info->ofs] += (float)creal(self->work[m]);
}
static void ringhelper_ring2phase (ringhelper *self,
const sharp_ringinfo *info, const void *data, int mmax, dcmplx *phase,
int pstride, sharp_fde fde)
{
int nph = info->nph;
#if 1
int maxidx = mmax; /* Enable this for traditional Healpix compatibility */
#else
int maxidx = IMIN(nph-1,mmax);
#endif
ringhelper_update (self, nph, mmax, -info->phi0);
if (fde==DOUBLE)
for (int m=0; m<nph; ++m)
self->work[m] = ((double *)data)[info->ofs+m*info->stride]*info->weight;
else
for (int m=0; m<nph; ++m)
self->work[m] = ((float *)data)[info->ofs+m*info->stride]*info->weight;
real_plan_forward_c (self->plan, (double *)self->work);
if (self->norot)
for (int m=0; m<=maxidx; ++m)
phase[m*pstride] = self->work[m%nph];
else
for (int m=0; m<=maxidx; ++m)
phase[m*pstride]=self->work[m%nph]*self->shiftarr[m];
for (int m=maxidx+1;m<=mmax; ++m)
phase[m*pstride]=0.;
}
static void ringhelper_pair2phase (ringhelper *self, int mmax,
const sharp_ringpair *pair, const void *data, dcmplx *phase1, dcmplx *phase2,
int pstride, sharp_fde fde)
{
ringhelper_ring2phase (self, &(pair->r1), data, mmax, phase1, pstride, fde);
if (pair->r2.nph>0)
ringhelper_ring2phase (self, &(pair->r2), data, mmax, phase2, pstride, fde);
}
static void ringhelper_phase2pair (ringhelper *self, int mmax,
const dcmplx *phase1, const dcmplx *phase2, int pstride,
const sharp_ringpair *pair, void *data, sharp_fde fde)
{
ringhelper_phase2ring (self, &(pair->r1), data, mmax, phase1, pstride, fde);
if (pair->r2.nph>0)
ringhelper_phase2ring (self, &(pair->r2), data, mmax, phase2, pstride, fde);
}
static void fill_map (const sharp_geom_info *ginfo, void *map, double value,
sharp_fde fde)
{
for (int j=0;j<ginfo->npairs;++j)
{
if (fde==DOUBLE)
{
for (int i=0;i<ginfo->pair[j].r1.nph;++i)
((double *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride]=value;
for (int i=0;i<ginfo->pair[j].r2.nph;++i)
((double *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride]=value;
}
else
{
for (int i=0;i<ginfo->pair[j].r1.nph;++i)
((float *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride]
=(float)value;
for (int i=0;i<ginfo->pair[j].r2.nph;++i)
((float *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride]
=(float)value;
}
}
}
static void fill_alm (const sharp_alm_info *ainfo, void *alm, dcmplx value,
sharp_fde fde)
{
if (fde==DOUBLE)
for (int mi=0;mi<ainfo->nm;++mi)
for (int l=ainfo->mval[mi];l<=ainfo->lmax;++l)
((dcmplx *)alm)[sharp_alm_index(ainfo,l,mi)] = value;
else
for (int mi=0;mi<ainfo->nm;++mi)
for (int l=ainfo->mval[mi];l<=ainfo->lmax;++l)
((fcmplx *)alm)[sharp_alm_index(ainfo,l,mi)] = (fcmplx)value;
}
static void init_output (sharp_job *job)
{
if (job->add_output) return;
if (job->type == SHARP_MAP2ALM)
for (int i=0; i<job->ntrans*job->nalm; ++i)
fill_alm (job->ainfo,job->alm[i],0.,job->fde);
else
for (int i=0; i<job->ntrans*job->nmaps; ++i)
fill_map (job->ginfo,job->map[i],0.,job->fde);
}
static void alloc_phase (sharp_job *job, int nm, int ntheta)
{ job->phase=RALLOC(dcmplx,2*job->ntrans*job->nmaps*nm*ntheta); }
static void dealloc_phase (sharp_job *job)
{ DEALLOC(job->phase); }
//FIXME: set phase to zero if not SHARP_MAP2ALM?
static void map2phase (sharp_job *job, int mmax, int llim, int ulim)
{
if (job->type != SHARP_MAP2ALM) return;
int pstride = 2*job->ntrans*job->nmaps;
#pragma omp parallel
{
ringhelper helper;
ringhelper_init(&helper);
#pragma omp for schedule(dynamic,1)
for (int ith=llim; ith<ulim; ++ith)
{
int dim2 = pstride*(ith-llim)*(mmax+1);
for (int i=0; i<job->ntrans*job->nmaps; ++i)
ringhelper_pair2phase(&helper,mmax,&job->ginfo->pair[ith], job->map[i],
&job->phase[dim2+2*i], &job->phase[dim2+2*i+1], pstride, job->fde);
}
ringhelper_destroy(&helper);
} /* end of parallel region */
}
static void alloc_almtmp (sharp_job *job, int lmax)
{ job->almtmp=RALLOC(dcmplx,job->ntrans*job->nalm*(lmax+1)); }
static void dealloc_almtmp (sharp_job *job)
{ DEALLOC(job->almtmp); }
static void alm2almtmp (sharp_job *job, int lmax, int mi)
{
if (job->type!=SHARP_MAP2ALM)
for (int l=job->ainfo->mval[mi]; l<=lmax; ++l)
{
ptrdiff_t aidx = sharp_alm_index(job->ainfo,l,mi);
double fct = job->norm_l[l];
for (int i=0; i<job->ntrans*job->nalm; ++i)
if (job->fde==DOUBLE)
job->almtmp[job->ntrans*job->nalm*l+i]
= ((dcmplx *)job->alm[i])[aidx]*fct;
else
job->almtmp[job->ntrans*job->nalm*l+i]
= ((fcmplx *)job->alm[i])[aidx]*fct;
}
else
SET_ARRAY(job->almtmp,job->ntrans*job->nalm*job->ainfo->mval[mi],
job->ntrans*job->nalm*(lmax+1),0.);
}
static void almtmp2alm (sharp_job *job, int lmax, int mi)
{
if (job->type != SHARP_MAP2ALM) return;
for (int l=job->ainfo->mval[mi]; l<=lmax; ++l)
{
ptrdiff_t aidx = sharp_alm_index(job->ainfo,l,mi);
for (int i=0;i<job->ntrans*job->nalm;++i)
if (job->fde==DOUBLE)
((dcmplx *)job->alm[i])[aidx] +=
job->almtmp[job->ntrans*job->nalm*l+i]*job->norm_l[l];
else
((fcmplx *)job->alm[i])[aidx] +=
(fcmplx)(job->almtmp[job->ntrans*job->nalm*l+i]*job->norm_l[l]);
}
}
static void phase2map (sharp_job *job, int mmax, int llim, int ulim)
{
if (job->type == SHARP_MAP2ALM) return;
int pstride = 2*job->ntrans*job->nmaps;
#pragma omp parallel
{
ringhelper helper;
ringhelper_init(&helper);
#pragma omp for schedule(dynamic,1)
for (int ith=llim; ith<ulim; ++ith)
{
int dim2 = pstride*(ith-llim)*(mmax+1);
for (int i=0; i<job->ntrans*job->nmaps; ++i)
ringhelper_phase2pair(&helper,mmax,&job->phase[dim2+2*i],
&job->phase[dim2+2*i+1],pstride,&job->ginfo->pair[ith],job->map[i],
job->fde);
}
ringhelper_destroy(&helper);
} /* end of parallel region */
}
static void sharp_execute_job (sharp_job *job)
{
double timer=wallTime();
job->opcnt=0;
int lmax = job->ainfo->lmax,
mmax=sharp_get_mmax(job->ainfo->mval, job->ainfo->nm);
job->norm_l = (job->type==SHARP_ALM2MAP_DERIV1) ?
sharp_Ylmgen_get_d1norm (lmax) :
sharp_Ylmgen_get_norm (lmax, job->spin);
/* clear output arrays if requested */
init_output (job);
int nchunks, chunksize;
get_chunk_info(job->ginfo->npairs,job->nv*VLEN,&nchunks,&chunksize);
alloc_phase (job,mmax+1,chunksize);
/* chunk loop */
for (int chunk=0; chunk<nchunks; ++chunk)
{
int llim=chunk*chunksize, ulim=IMIN(llim+chunksize,job->ginfo->npairs);
int *ispair = RALLOC(int,ulim-llim);
double *cth = RALLOC(double,ulim-llim), *sth = RALLOC(double,ulim-llim);
idxhelper *stmp = RALLOC(idxhelper,ulim-llim);
for (int i=0; i<ulim-llim; ++i)
{
ispair[i] = job->ginfo->pair[i+llim].r2.nph>0;
cth[i] = job->ginfo->pair[i+llim].r1.cth;
sth[i] = job->ginfo->pair[i+llim].r1.sth;
stmp[i].s=sth[i];
stmp[i].i=i;
}
qsort (stmp,ulim-llim,sizeof(idxhelper),idx_compare);
int *idx = RALLOC(int,ulim-llim);
for (int i=0; i<ulim-llim; ++i)
idx[i]=stmp[i].i;
DEALLOC(stmp);
/* map->phase where necessary */
map2phase (job, mmax, llim, ulim);
#pragma omp parallel
{
sharp_job ljob = *job;
ljob.opcnt=0;
sharp_Ylmgen_C generator;
sharp_Ylmgen_init (&generator,lmax,mmax,ljob.spin);
alloc_almtmp(&ljob,lmax);
#pragma omp for schedule(dynamic,1)
for (int mi=0; mi<job->ainfo->nm; ++mi)
{
/* alm->alm_tmp where necessary */
alm2almtmp (&ljob, lmax, mi);
inner_loop (&ljob, ispair, cth, sth, llim, ulim, &generator, mi, idx);
/* alm_tmp->alm where necessary */
almtmp2alm (&ljob, lmax, mi);
}
sharp_Ylmgen_destroy(&generator);
dealloc_almtmp(&ljob);
#pragma omp critical
job->opcnt+=ljob.opcnt;
} /* end of parallel region */
/* phase->map where necessary */
phase2map (job, mmax, llim, ulim);
DEALLOC(ispair);
DEALLOC(cth);
DEALLOC(sth);
DEALLOC(idx);
} /* end of chunk loop */
DEALLOC(job->norm_l);
dealloc_phase (job);
job->time=wallTime()-timer;
}
static void sharp_build_job_common (sharp_job *job, sharp_jobtype type,
int spin, int add_output, void *alm, void *map,
const sharp_geom_info *geom_info, const sharp_alm_info *alm_info, int ntrans,
int dp, int nv)
{
UTIL_ASSERT((ntrans>0),"bad number of simultaneous transforms");
if (type==SHARP_ALM2MAP_DERIV1) spin=1;
UTIL_ASSERT((spin>=0)&&(spin<=30), "bad spin");
job->type = type;
job->spin = spin;
job->norm_l = NULL;
job->add_output = add_output;
job->nmaps = (type==SHARP_ALM2MAP_DERIV1) ? 2 : ((spin>0) ? 2 : 1);
job->nalm = (type==SHARP_ALM2MAP_DERIV1) ? 1 : ((spin>0) ? 2 : 1);
job->ginfo = geom_info;
job->ainfo = alm_info;
job->nv = (nv==0) ? sharp_nv_oracle (type, spin, ntrans) : nv;
job->time = 0.;
job->opcnt = 0;
job->ntrans = ntrans;
job->alm=alm;
job->map=map;
job->fde=dp ? DOUBLE : FLOAT;
}
void sharp_execute (sharp_jobtype type, int spin, int add_output, void *alm,
void *map, const sharp_geom_info *geom_info, const sharp_alm_info *alm_info,
int ntrans, int dp, int nv, double *time, unsigned long long *opcnt)
{
sharp_job job;
sharp_build_job_common (&job, type, spin, add_output, alm, map, geom_info,
alm_info, ntrans, dp, nv);
sharp_execute_job (&job);
if (time!=NULL) *time = job.time;
if (opcnt!=NULL) *opcnt = job.opcnt;
}
int sharp_get_nv_max (void)
{ return 6; }
static int sharp_oracle (sharp_jobtype type, int spin, int ntrans)
{
int lmax=127;
int mmax=(lmax+1)/2;
int nrings=(lmax+1)/4;
int ppring=1;
ptrdiff_t npix=(ptrdiff_t)nrings*ppring;
sharp_geom_info *tinfo;
sharp_make_gauss_geom_info (nrings, ppring, 1, ppring, &tinfo);
ptrdiff_t nalms = ((mmax+1)*(mmax+2))/2 + (mmax+1)*(lmax-mmax);
int ncomp = ntrans*((spin==0) ? 1 : 2);
double **map;
ALLOC2D(map,double,ncomp,npix);
SET_ARRAY(map[0],0,npix*ncomp,0.);
sharp_alm_info *alms;
sharp_make_triangular_alm_info(lmax,mmax,1,&alms);
dcmplx **alm;
ALLOC2D(alm,dcmplx,ncomp,nalms);
SET_ARRAY(alm[0],0,nalms*ncomp,0.);
double time=1e30;
int nvbest=-1;
for (int nv=1; nv<=sharp_get_nv_max(); ++nv)
{
double time_acc=0.;
double jtime;
int ntries=0;
do
{
sharp_execute(type,spin,0,&alm[0],&map[0],tinfo,alms,ntrans,1,nv,&jtime,
NULL);
if (jtime<time) { time=jtime; nvbest=nv; }
time_acc+=jtime;
++ntries;
}
while ((time_acc<0.02)&&(ntries<2));
}
DEALLOC2D(map);
DEALLOC2D(alm);
sharp_destroy_alm_info(alms);
sharp_destroy_geom_info(tinfo);
return nvbest;
}
int sharp_nv_oracle (sharp_jobtype type, int spin, int ntrans)
{
static const int maxtr = 6;
static int nv_opt[6][2][3] = {
{{0,0,0},{0,0,0}},
{{0,0,0},{0,0,0}},
{{0,0,0},{0,0,0}},
{{0,0,0},{0,0,0}},
{{0,0,0},{0,0,0}},
{{0,0,0},{0,0,0}} };
if (type==SHARP_ALM2MAP_DERIV1) spin=1;
UTIL_ASSERT((ntrans>0),"bad number of simultaneous transforms");
UTIL_ASSERT((spin>=0)&&(spin<=30), "bad spin");
ntrans=IMIN(ntrans,maxtr);
if (nv_opt[ntrans-1][spin!=0][type]==0)
nv_opt[ntrans-1][spin!=0][type]=sharp_oracle(type,spin,ntrans);
return nv_opt[ntrans-1][spin!=0][type];
}
#ifdef USE_MPI
#include "sharp_mpi.c"
#endif