profit.c

  profit->iter = 0;
  memset(dcovar, 0, profit->nparam*profit->nparam*sizeof(double));

/* Perform fit */
  lm_opts[0] = 1.0e-2;
  lm_opts[1] = 1.0e-12;
  lm_opts[2] = 1.0e-12;
  lm_opts[3] = 1.0e-12;
  lm_opts[4] = 1.0e-4;

  profit_boundtounbound(profit, profit->paraminit, dparam, PARAM_ALLPARAMS);

  niter = dlevmar_dif(profit_evaluate, dparam, NULL, profit->nfreeparam,
		profit->nresi, niter, lm_opts, info, NULL, dcovar, profit);

  profit_unboundtobound(profit, dparam, profit->paraminit, PARAM_ALLPARAMS);

/* Convert covariance matrix to bounded space */
  profit_covarunboundtobound(profit, dcovar, profit->covar);

  return niter;
  }


/****** profit_printout *******************************************************
PROTO	void profit_printout(int n_par, float* par, int m_dat, float* fvec,
		void *data, int iflag, int iter, int nfev )
PURPOSE	Provide a function to print out results to lmfit.
INPUT	Number of fitted parameters,
	pointer to the vector of parameters,
	number of data points,
	pointer to the vector of residuals (output),
	pointer to the data structure (unused),
	0 (init) 1 (outer loop) 2(inner loop) -1(terminated),
	outer loop counter,
	number of calls to evaluate().
OUTPUT	-.
NOTES	Input arguments are there only for compatibility purposes (unused)
AUTHOR	E. Bertin (IAP)
VERSION	17/09/2008
 ***/
void	profit_printout(int n_par, float* par, int m_dat, float* fvec,
		void *data, int iflag, int iter, int nfev )
  {
   checkstruct	*check;
   profitstruct	*profit;
   char		filename[256];
   static int	itero;

  profit = (profitstruct *)data;

  if (0 && (iter!=itero || iter<0))
    {
    if (iter<0)
      itero++;
    else
      itero = iter;
    sprintf(filename, "check_%d_%04d.fits", the_gal, itero);
    check=initcheck(filename, CHECK_PROFILES, 0);
    reinitcheck(the_field, check);
    addcheck(check, profit->lmodpix, profit->objnaxisn[0],profit->objnaxisn[1],
		profit->ix,profit->iy, 1.0);

    reendcheck(the_field, check);
    endcheck(check);
    }

  return;
  }


/****** profit_evaluate ******************************************************
PROTO	void profit_evaluate(double *par, double *fvec, int m, int n,
			void *adata)
PURPOSE	Provide a function returning residuals to levmar.
INPUT	Pointer to the vector of parameters,
	pointer to the vector of residuals (output),
	number of model parameters,
	number of data points,
	pointer to a data structure (we use it for the profit structure here).
OUTPUT	-.
NOTES	-.
AUTHOR	E. Bertin (IAP)
VERSION	07/10/2010
 ***/
void	profit_evaluate(double *dpar, double *fvec, int m, int n, void *adata)
  {
   profitstruct		*profit;
   profstruct		**prof;
   double		*dpar0, *dresi;
   float		*modpixt, *profpixt, *resi,
			tparam, val;
   PIXTYPE		*lmodpixt,*lmodpix2t, *objpix,*weight,
			wval;
   int			c,f,i,p,q, fd,pd, jflag,sflag, nprof;

  profit = (profitstruct *)adata;

/* Detect "Jacobian-related" calls */
  jflag = pd = fd = 0;
  dpar0 = profit->dparam;
  if (profit->iter)
    {
    f = q = 0;
    for (p=0; p<profit->nparam; p++)
      {
      if (dpar[f] - dpar0[f] != 0.0)
        {
        pd = p;
        fd = f;
        q++;
        }
      if (profit->freeparam_flag[p])
        f++;
      }
    if (f>0 && q==1)
      jflag = 1;
    }

  if (jflag && !(profit->nprof==1 && profit->prof[0]->code == PROF_DIRAC))
    {
    prof = profit->prof;
    nprof = profit->nprof;

/*-- "Jacobian call" */
    tparam = profit->param[pd];
    profit_unboundtobound(profit, &dpar[fd], &profit->param[pd], pd);
    sflag = 1;
    switch(profit->paramrevindex[pd])
      {
      case PARAM_BACK:
        lmodpixt = profit->lmodpix;
        lmodpix2t = profit->lmodpix2;
        val = (profit->param[pd] - tparam);
        for (i=profit->nobjpix;i--;)
          *(lmodpix2t++) = val;
        break;
      case PARAM_X:
      case PARAM_Y:
        profit_resample(profit, profit->cmodpix, profit->lmodpix2, 1.0);
        lmodpixt = profit->lmodpix;
        lmodpix2t = profit->lmodpix2;
        for (i=profit->nobjpix;i--;)
          *(lmodpix2t++) -= *(lmodpixt++);
        break;
      case PARAM_DIRAC_FLUX:
      case PARAM_SPHEROID_FLUX:
      case PARAM_DISK_FLUX:
      case PARAM_ARMS_FLUX:
      case PARAM_BAR_FLUX:
        if (nprof==1 && tparam != 0.0)
          {
          lmodpixt = profit->lmodpix;
          lmodpix2t = profit->lmodpix2;
          val = (profit->param[pd] - tparam) / tparam;
          for (i=profit->nobjpix;i--;)
            *(lmodpix2t++) = val**(lmodpixt++);
          }
        else
          {
          for (c=0; c<nprof; c++)
            if (prof[c]->flux == &profit->param[pd])
              break;
          memcpy(profit->modpix, prof[c]->pix, profit->nmodpix*sizeof(float));
          profit_convolve(profit, profit->modpix);
          profit_resample(profit, profit->modpix, profit->lmodpix2,
		profit->param[pd] - tparam);
          }
        break;
      case PARAM_SPHEROID_REFF:
      case PARAM_SPHEROID_ASPECT:
      case PARAM_SPHEROID_POSANG:
      case PARAM_SPHEROID_SERSICN:
        sflag = 0;			/* We are in the same switch */
        for (c=0; c<nprof; c++)
          if (prof[c]->code == PROF_SERSIC
		|| prof[c]->code == PROF_DEVAUCOULEURS)
            break; 
      case PARAM_DISK_SCALE:
      case PARAM_DISK_ASPECT:
      case PARAM_DISK_POSANG:
        if (sflag)
          for (c=0; c<nprof; c++)
            if (prof[c]->code == PROF_EXPONENTIAL)
              break; 
        sflag = 0;
      case PARAM_ARMS_QUADFRAC:
      case PARAM_ARMS_SCALE:
      case PARAM_ARMS_START:
      case PARAM_ARMS_POSANG:
      case PARAM_ARMS_PITCH:
      case PARAM_ARMS_PITCHVAR:
      case PARAM_ARMS_WIDTH:
        if (sflag)
          for (c=0; c<nprof; c++)
            if (prof[c]->code == PROF_ARMS)
              break; 
        sflag = 0;
      case PARAM_BAR_ASPECT:
      case PARAM_BAR_POSANG:
        if (sflag)
          for (c=0; c<nprof; c++)
            if (prof[c]->code == PROF_ARMS)
              break; 
        modpixt = profit->modpix;
        profpixt = prof[c]->pix;
        val = -*prof[c]->flux;
        for (i=profit->nmodpix;i--;)
          *(modpixt++) = val**(profpixt++);
        memcpy(profit->modpix2, prof[c]->pix, profit->nmodpix*sizeof(float));
        prof_add(profit, prof[c], 0);
        memcpy(prof[c]->pix, profit->modpix2, profit->nmodpix*sizeof(float));
        profit_convolve(profit, profit->modpix);
        profit_resample(profit, profit->modpix, profit->lmodpix2, 1.0);
        break;
      default:
        error(EXIT_FAILURE, "*Internal Error*: ",
			"unknown parameter index in profit_jacobian()");
        break;
      }
    objpix = profit->objpix;
    weight = profit->objweight;
    lmodpixt = profit->lmodpix;
    lmodpix2t = profit->lmodpix2;
    resi = profit->resi;
    dresi = fvec;
    if (PROFIT_DYNPARAM > 0.0)
      for (i=profit->nobjpix;i--; lmodpixt++, lmodpix2t++)
        {
        val = *(objpix++);
        if ((wval=*(weight++))>0.0)
          *(dresi++) = *(resi++) + *lmodpix2t
		* wval/(1.0+wval*fabs(*lmodpixt - val)/PROFIT_DYNPARAM);
        }
    else
      for (i=profit->nobjpix;i--; lmodpix2t++)
        if ((wval=*(weight++))>0.0)
          *(dresi++) = *(resi++) + *lmodpix2t * wval;
    }
  else
    {
/*-- "Regular call" */
    for (p=0; p<profit->nparam; p++)
      dpar0[p] = dpar[p];
    profit_unboundtobound(profit, dpar, profit->param, PARAM_ALLPARAMS);

    profit_residuals(profit, the_field, the_wfield, PROFIT_DYNPARAM,
	profit->param, profit->resi);

    for (p=0; p<profit->nresi; p++)
      fvec[p] = profit->resi[p];
    }

//  profit_printout(m, par, n, fvec, adata, 0, -1, 0 );
  profit->iter++;

  return;
  }


/****** profit_residuals ******************************************************
PROTO	float *prof_residuals(profitstruct *profit, picstruct *field,
		picstruct *wfield, float dynparam, float *param, float *resi)
PURPOSE	Compute the vector of residuals between the data and the galaxy
	profile model.
INPUT	Profile-fitting structure,
	pointer to the field,
	pointer to the field weight,
	dynamic compression parameter (0=no compression),
	pointer to the model parameters (output),
	pointer to the computed residuals (output).
OUTPUT	Vector of residuals.
NOTES	-.
AUTHOR	E. Bertin (IAP)
VERSION	08/07/2010
 ***/
float	*profit_residuals(profitstruct *profit, picstruct *field,
		picstruct *wfield, float dynparam, float *param, float *resi)
  {
   int		p, nmodpix;

  nmodpix = profit->modnaxisn[0]*profit->modnaxisn[1]*sizeof(float);
  memset(profit->modpix, 0, nmodpix);
  for (p=0; p<profit->nparam; p++)
    profit->param[p] = param[p];
/* Simple PSF shortcut */
  if (profit->nprof == 1 && profit->prof[0]->code == PROF_DIRAC)
    {
    profit_resample(profit, profit->psfpix, profit->lmodpix,
		*profit->prof[0]->flux);
    profit->flux = *profit->prof[0]->flux;
    }
  else
    {
    profit->flux = 0.0;
    for (p=0; p<profit->nprof; p++)
      profit->flux += prof_add(profit, profit->prof[p], 0);
    memcpy(profit->cmodpix, profit->modpix, profit->nmodpix*sizeof(float));
    profit_convolve(profit, profit->cmodpix);
    profit_resample(profit, profit->cmodpix, profit->lmodpix, 1.0);
    }

  if (resi)
    profit_compresi(profit, dynparam, resi);

  return resi;
  }


/****** profit_compresi ******************************************************
PROTO	float *prof_compresi(profitstruct *profit, float dynparam,
			float *resi)
PURPOSE	Compute the vector of residuals between the data and the galaxy
	profile model.
INPUT	Profile-fitting structure,
	dynamic-compression parameter (0=no compression),
	vector of residuals (output).
OUTPUT	Vector of residuals.
NOTES	-.
AUTHOR	E. Bertin (IAP)
VERSION	07/09/2009
 ***/
float	*profit_compresi(profitstruct *profit, float dynparam, float *resi)
  {
   double	error;
   float	*resit;
   PIXTYPE	*objpix, *objweight, *lmodpix,
		val,val2,wval, invsig;
   int		npix, i;
  
/* Compute vector of residuals */
  resit = resi;
  objpix = profit->objpix;
  objweight = profit->objweight;
  lmodpix = profit->lmodpix;
  error = 0.0;
  npix = profit->objnaxisn[0]*profit->objnaxisn[1];
  if (dynparam > 0.0)
    {
    invsig = (PIXTYPE)(1.0/dynparam);
    for (i=npix; i--; lmodpix++)
      {
      val = *(objpix++);
      if ((wval=*(objweight++))>0.0)
        {
        val2 = (*lmodpix - val)*wval*invsig;
        val2 = val2>0.0? logf(1.0+val2) : -logf(1.0-val2);
        *(resit++) = val2*dynparam;
        error += val2*val2;
        }
      }
    profit->chi2 = dynparam*dynparam*error;
    }
  else
    {
    for (i=npix; i--; lmodpix++)
      {
      val = *(objpix++);
      if ((wval=*(objweight++))>0.0)
        {
        val2 = (*lmodpix - val)*wval;
        *(resit++) = val2;
        error += val2*val2;
        }
      }
    profit->chi2 = error;
    }

  return resi;
  }


/****** profit_resample ******************************************************
PROTO	int	prof_resample(profitstruct *profit, float *inpix,
		PIXTYPE *outpix, float factor)
PURPOSE	Resample the current full resolution model to image resolution.
INPUT	Profile-fitting structure,
	pointer to input raster,
	pointer to output raster,
	multiplicating factor.
OUTPUT	RETURN_ERROR if the rasters don't overlap, RETURN_OK otherwise.
NOTES	-.
AUTHOR	E. Bertin (IAP)
VERSION	12/09/2010
 ***/
int	profit_resample(profitstruct *profit, float *inpix, PIXTYPE *outpix,
	float factor)
  {
   PIXTYPE	*pixout,*pixout0;
   float	*pixin,*pixin0, *mask,*maskt, *pixinout, *dpixin,*dpixin0,
		*dpixout,*dpixout0, *dx,*dy,
		xcin,xcout,ycin,ycout, xsin,ysin, xin,yin, x,y, dxm,dym, val,
		invpixstep, norm;
   int		*start,*startt, *nmask,*nmaskt,
		i,j,k,n,t, 
		ixsout,iysout, ixout,iyout, dixout,diyout, nxout,nyout,
		iysina, nyin, hmw,hmh, ix,iy, ixin,iyin;

  invpixstep = profit->subsamp/profit->pixstep;

  xcin = (profit->modnaxisn[0]/2);
  xcout = (float)(profit->objnaxisn[0]/2);
  if ((dx=(profit->paramlist[PARAM_X])))
    xcout += *dx;

  xsin = xcin - xcout*invpixstep;			/* Input start x-coord*/
  if ((int)xsin >= profit->modnaxisn[0])
    return RETURN_ERROR;
  ixsout = 0;				/* Int. part of output start x-coord */
  if (xsin<0.0)
    {
    dixout = (int)(1.0-xsin/invpixstep);
/*-- Simply leave here if the images do not overlap in x */
    if (dixout >= profit->objnaxisn[0])
      return RETURN_ERROR;
    ixsout += dixout;
    xsin += dixout*invpixstep;
    }
  nxout = (int)((profit->modnaxisn[0]-xsin)/invpixstep);/* nb of interpolated
							input pixels along x */
  if (nxout>(ixout=profit->objnaxisn[0]-ixsout))
    nxout = ixout;
  if (!nxout)
    return RETURN_ERROR;

  ycin = (profit->modnaxisn[1]/2);
  ycout = (float)(profit->objnaxisn[1]/2);
  if ((dy=(profit->paramlist[PARAM_Y])))
    ycout += *dy;

  ysin = ycin - ycout*invpixstep;		/* Input start y-coord*/
  if ((int)ysin >= profit->modnaxisn[1])
    return RETURN_ERROR;
  iysout = 0;				/* Int. part of output start y-coord */
  if (ysin<0.0)
    {
    diyout = (int)(1.0-ysin/invpixstep);
/*-- Simply leave here if the images do not overlap in y */
    if (diyout >= profit->objnaxisn[1])
      return RETURN_ERROR;
    iysout += diyout;
    ysin += diyout*invpixstep;
    }
  nyout = (int)((profit->modnaxisn[1]-ysin)/invpixstep);/* nb of interpolated
							input pixels along y */
  if (nyout>(iyout=profit->objnaxisn[1]-iysout))
    nyout = iyout;
  if (!nyout)
    return RETURN_ERROR;

/* Set the yrange for the x-resampling with some margin for interpolation */
  iysina = (int)ysin;	/* Int. part of Input start y-coord with margin */
  hmh = INTERPW/2 - 1;	/* Interpolant start */
  if (iysina<0 || ((iysina -= hmh)< 0))
    iysina = 0;
  nyin = (int)(ysin+nyout*invpixstep)+INTERPW-hmh;/* Interpolated Input y size*/
  if (nyin>profit->modnaxisn[1])						/* with margin */
    nyin = profit->modnaxisn[1];
/* Express everything relative to the effective Input start (with margin) */
  nyin -= iysina;
  ysin -= (float)iysina;

/* Allocate interpolant stuff for the x direction */
  QMALLOC(mask, float, nxout*INTERPW);	/* Interpolation masks */
  QMALLOC(nmask, int, nxout);		/* Interpolation mask sizes */
  QMALLOC(start, int, nxout);		/* Int. part of Input conv starts */
/* Compute the local interpolant and data starting points in x */
  hmw = INTERPW/2 - 1;
  xin = xsin;
  maskt = mask;
  nmaskt = nmask;
  startt = start;
  for (j=nxout; j--; xin+=invpixstep)
    {
    ix = (ixin=(int)xin) - hmw;
    dxm = ixin - xin - hmw;	/* starting point in the interpolation func */
    if (ix < 0)
      {
      n = INTERPW+ix;
      dxm -= (float)ix;
      ix = 0;
      }
    else
      n = INTERPW;
    if (n>(t=profit->modnaxisn[0]-ix))
      n=t;
    *(startt++) = ix;
    *(nmaskt++) = n;
    norm = 0.0;
    for (x=dxm, i=n; i--; x+=1.0)
      norm += (*(maskt++) = INTERPF(x));
    norm = norm>0.0? 1.0/norm : 1.0;
    maskt -= n;
    for (i=n; i--;)
      *(maskt++) *= norm;
    }

  QCALLOC(pixinout, float, nxout*nyin);	/* Intermediary frame-buffer */

/* Make the interpolation in x (this includes transposition) */
  pixin0 = inpix + iysina*profit->modnaxisn[0];
  dpixout0 = pixinout;
  for (k=nyin; k--; pixin0+=profit->modnaxisn[0], dpixout0++)
    {
    maskt = mask;
    nmaskt = nmask;
    startt = start;
    dpixout = dpixout0;
    for (j=nxout; j--; dpixout+=nyin)
      {
      pixin = pixin0+*(startt++);
      val = 0.0; 
      for (i=*(nmaskt++); i--;)
        val += *(maskt++)**(pixin++);
      *dpixout = val;
      }
    }

/* Reallocate interpolant stuff for the y direction */
  QREALLOC(mask, float, nyout*INTERPW);	/* Interpolation masks */
  QREALLOC(nmask, int, nyout);			/* Interpolation mask sizes */
  QREALLOC(start, int, nyout);		/* Int. part of Input conv starts */

/* Compute the local interpolant and data starting points in y */
  hmh = INTERPW/2 - 1;
  yin = ysin;
  maskt = mask;
  nmaskt = nmask;
  startt = start;
  for (j=nyout; j--; yin+=invpixstep)
    {
    iy = (iyin=(int)yin) - hmh;
    dym = iyin - yin - hmh;	/* starting point in the interpolation func */
    if (iy < 0)
      {
      n = INTERPW+iy;
      dym -= (float)iy;
      iy = 0;
      }
    else
      n = INTERPW;
    if (n>(t=nyin-iy))
      n=t;
    *(startt++) = iy;
    *(nmaskt++) = n;
    norm = 0.0;
    for (y=dym, i=n; i--; y+=1.0)
      norm += (*(maskt++) = INTERPF(y));
    norm = norm>0.0? 1.0/norm : 1.0;
    maskt -= n;
    for (i=n; i--;)
      *(maskt++) *= norm;
    }

/* Initialize destination buffer to zero */
  memset(outpix, 0, (size_t)profit->nobjpix*sizeof(PIXTYPE));

/* Make the interpolation in y and transpose once again */
  dpixin0 = pixinout;
  pixout0 = outpix+ixsout+iysout*profit->objnaxisn[0];
  for (k=nxout; k--; dpixin0+=nyin, pixout0++)
    {
    maskt = mask;
    nmaskt = nmask;
    startt = start;
    pixout = pixout0;
    for (j=nyout; j--; pixout+=profit->objnaxisn[0])
      {
      dpixin = dpixin0+*(startt++);
      val = 0.0; 
      for (i=*(nmaskt++); i--;)
        val += *(maskt++)**(dpixin++);
       *pixout = (PIXTYPE)(factor*val);
      }
    }

/* Free memory */
  free(pixinout);
  free(mask);
  free(nmask);
  free(start);

  return RETURN_OK;
  }


/****** profit_convolve *******************************************************
PROTO	void profit_convolve(profitstruct *profit, float *modpix)
PURPOSE	Convolve a model image with the local PSF.
INPUT	Pointer to the profit structure,
	Pointer to the image raster.
OUTPUT	-.
NOTES	-.
AUTHOR	E. Bertin (IAP)
VERSION	15/09/2008
 ***/
void	profit_convolve(profitstruct *profit, float *modpix)
  {
  if (!profit->psfdft)
    profit_makedft(profit);

  fft_conv(modpix, profit->psfdft, profit->modnaxisn);

  return;
  }


/****** profit_makedft *******************************************************
PROTO	void profit_makedft(profitstruct *profit)
PURPOSE	Create the Fourier transform of the descrambled PSF component.
INPUT	Pointer to the profit structure.
OUTPUT	-.
NOTES	-.
AUTHOR	E. Bertin (IAP)
VERSION	22/04/2008
 ***/
void	profit_makedft(profitstruct *profit)
  {
   psfstruct	*psf;
   float      *mask,*maskt, *ppix;
   float       dx,dy, r,r2,rmin,rmin2,rmax,rmax2,rsig,invrsig2;
   int          width,height,npix,offset, psfwidth,psfheight,psfnpix,
                cpwidth, cpheight,hcpwidth,hcpheight, i,j,x,y;

  if (!(psf=profit->psf))
    return;

  psfwidth = profit->modnaxisn[0];
  psfheight = profit->modnaxisn[1];
  psfnpix = psfwidth*psfheight;
  width = profit->modnaxisn[0];
  height = profit->modnaxisn[1];
  npix = width*height;
  QCALLOC(mask, float, npix);
  cpwidth = (width>psfwidth)?psfwidth:width;
  hcpwidth = cpwidth>>1;
  cpwidth = hcpwidth<<1;
  offset = width - cpwidth;
  cpheight = (height>psfheight)?psfheight:height;
  hcpheight = cpheight>>1;
  cpheight = hcpheight<<1;

/* Frame and descramble the PSF data */
  ppix = profit->psfpix + (psfheight/2)*psfwidth + psfwidth/2;
  maskt = mask;
  for (j=hcpheight; j--; ppix+=psfwidth)
    {
    for (i=hcpwidth; i--;)
      *(maskt++) = *(ppix++);      
    ppix -= cpwidth;
    maskt += offset;
    for (i=hcpwidth; i--;)
      *(maskt++) = *(ppix++);      
    }

  ppix = profit->psfpix + ((psfheight/2)-hcpheight)*psfwidth + psfwidth/2;
  maskt += width*(height-cpheight);
  for (j=hcpheight; j--; ppix+=psfwidth)
    {
    for (i=hcpwidth; i--;)
      *(maskt++) = *(ppix++);      
    ppix -= cpwidth;
    maskt += offset;
    for (i=hcpwidth; i--;)
      *(maskt++) = *(ppix++);      
    }

/* Truncate to a disk that has diameter = (box width) */
  rmax = cpwidth - 1.0 - hcpwidth;
  if (rmax > (r=hcpwidth))
    rmax = r;
  if (rmax > (r=cpheight-1.0-hcpheight))
    rmax = r;
  if (rmax > (r=hcpheight))
    rmax = r;
  if (rmax<1.0)
    rmax = 1.0;
  rmax2 = rmax*rmax;
  rsig = psf->fwhm/profit->pixstep;
  invrsig2 = 1/(2*rsig*rsig);
  rmin = rmax - (3*rsig);     /* 3 sigma annulus (almost no aliasing) */
  rmin2 = rmin*rmin;

  maskt = mask;
  dy = 0.0;
  for (y=hcpheight; y--; dy+=1.0)
    {
    dx = 0.0;
    for (x=hcpwidth; x--; dx+=1.0, maskt++)
      if ((r2=dx*dx+dy*dy)>rmin2)
        *maskt *= (r2>rmax2)?0.0:expf((2*rmin*sqrtf(r2)-r2-rmin2)*invrsig2);
    dx = -hcpwidth;
    maskt += offset;
    for (x=hcpwidth; x--; dx+=1.0, maskt++)
      if ((r2=dx*dx+dy*dy)>rmin2)
        *maskt *= (r2>rmax2)?0.0:expf((2*rmin*sqrtf(r2)-r2-rmin2)*invrsig2);
    }
  dy = -hcpheight;
  maskt += width*(height-cpheight);
  for (y=hcpheight; y--; dy+=1.0)
    {
    dx = 0.0;
    for (x=hcpwidth; x--; dx+=1.0, maskt++)
      if ((r2=dx*dx+dy*dy)>rmin2)
        *maskt *= (r2>rmax2)?0.0:expf((2*rmin*sqrtf(r2)-r2-rmin2)*invrsig2);
    dx = -hcpwidth;
    maskt += offset;
    for (x=hcpwidth; x--; dx+=1.0, maskt++)
      if ((r2=dx*dx+dy*dy)>rmin2)
        *maskt *= (r2>rmax2)?0.0:expf((2*rmin*sqrtf(r2)-r2-rmin2)*invrsig2);
    }

/* Finally move to Fourier space */
  profit->psfdft = fft_rtf(mask, profit->modnaxisn);

  free(mask);

  return;
  }


/****** profit_copyobjpix *****************************************************
PROTO	int profit_copyobjpix(profitstruct *profit, picstruct *field,
			picstruct *wfield)
PURPOSE	Copy a piece of the input field image to a profit structure.
INPUT	Pointer to the profit structure,
	Pointer to the field structure,
	Pointer to the field weight structure.
OUTPUT	The number of valid pixels copied.
NOTES	Global preferences are used.
AUTHOR	E. Bertin (IAP)
VERSION	01/12/2009
 ***/
int	profit_copyobjpix(profitstruct *profit, picstruct *field,
			picstruct *wfield)
  {
   float	dx, dy2, dr2, rad2;
   PIXTYPE	*pixin,*spixin, *wpixin,*swpixin, *pixout,*wpixout,
		backnoise2, invgain, satlevel, wthresh, pix,spix, wpix,swpix;
   int		i,x,y, xmin,xmax,ymin,ymax, w,h,dw, npix, off, gainflag,
		badflag, sflag, sx,sy,sn,sw, ix,iy;

/* First put the image background to -BIG */
  pixout = profit->objpix;
  wpixout = profit->objweight;
  for (i=profit->objnaxisn[0]*profit->objnaxisn[1]; i--;)
    {
    *(pixout++) = -BIG;
    *(wpixout++) = 0.0;
    }

/* Don't go further if out of frame!! */
  ix = profit->ix;
  iy = profit->iy;
  if (ix<0 || ix>=field->width || iy<field->ymin || iy>=field->ymax)
    return 0;

  backnoise2 = field->backsig*field->backsig;
  sn = (int)profit->subsamp;
  sflag = (sn>1);
  w = profit->objnaxisn[0]*sn;
  h = profit->objnaxisn[1]*sn;
  if (sflag)
    backnoise2 *= (PIXTYPE)sn;
  invgain = (field->gain > 0.0) ? 1.0/field->gain : 0.0;
  satlevel = field->satur_level - profit->obj->bkg;
  rad2 = h/2.0;
  if (rad2 > w/2.0)
    rad2 = w/2.0;
  rad2 *= rad2;

/* Set the image boundaries */
  pixout = profit->objpix;
  wpixout = profit->objweight;
  ymin = iy-h/2;
  ymax = ymin + h;
  if (ymin<field->ymin)
    {
    off = (field->ymin-ymin-1)/sn + 1;
    pixout += off*profit->objnaxisn[0];
    wpixout += off*profit->objnaxisn[0];
    ymin += off*sn;
    }
  if (ymax>field->ymax)
    ymax -= ((ymax-field->ymax-1)/sn + 1)*sn;

  xmin = ix-w/2;
  xmax = xmin + w;
  dw = 0;
  if (xmax>field->width)
    {
    off = (xmax-field->width-1)/sn + 1;
    dw += off;
    xmax -= off*sn;
    }
  if (xmin<0)
    {
    off = (-xmin-1)/sn + 1;
    pixout += off;
    wpixout += off;
    dw += off;
    xmin += off*sn;
    }
/* Make sure the input frame size is a multiple of the subsampling step */
  if (sflag)
    {
/*
    if (((rem=ymax-ymin)%sn))
      {
      ymin += rem/2;
      ymax -= (rem-rem/2);
      }
    if (((rem=xmax-xmin)%sn))
      {
      xmin += rem/2;
      pixout += rem/2;
      wpixout += rem/2;
      dw += rem;
      xmax -= (rem-rem/2);
      }
*/
    sw = field->width;
    }

/* Copy the right pixels to the destination */
  npix = 0;
  if (wfield)
    {
    wthresh = wfield->weight_thresh;
    gainflag = prefs.weightgain_flag;
    if (sflag)
      {
/*---- Sub-sampling case */
      for (y=ymin; y<ymax; y+=sn, pixout+=dw,wpixout+=dw)
        {
        for (x=xmin; x<xmax; x+=sn)
          {
          pix = wpix = 0.0;
          badflag = 0;
          for (sy=0; sy<sn; sy++)
            {
            dy2 = (y+sy-iy);
            dy2 *= dy2;
            dx = (x-ix);
            spixin = &PIX(field, x, y+sy);
            swpixin = &PIX(wfield, x, y+sy);
            for (sx=sn; sx--;)
              {
              dr2 = dy2 + dx*dx;
              dx++;
              spix = *(spixin++);
              swpix = *(swpixin++);
              if (dr2<rad2 && spix>-BIG && spix<satlevel && swpix<wthresh)
                {
                pix += spix;
                wpix += swpix;
                }
              else
                badflag=1;
              }
            }
          *(pixout++) = pix;
          if (!badflag)	/* A single bad pixel ruins is all (saturation, etc.)*/
            {
            *(wpixout++) = 1.0 / sqrt(wpix+(pix>0.0?
		(gainflag? pix*wpix/backnoise2:pix)*invgain : 0.0));
            npix++;
            }
          else
            *(wpixout++) = 0.0;
          }
        }
      }
    else
      for (y=ymin; y<ymax; y++, pixout+=dw,wpixout+=dw)
        {
        dy2 = y-iy;
        dy2 *= dy2;
        pixin = &PIX(field, xmin, y);
        wpixin = &PIX(wfield, xmin, y);
        for (x=xmin; x<xmax; x++)
          {
          dx = x-ix;
          dr2 = dy2 + dx*dx;
          pix = *(pixin++);
          wpix = *(wpixin++);
          if (dr2<rad2 && pix>-BIG && pix<satlevel && wpix<wthresh)
            {
            *(pixout++) = pix;
            *(wpixout++) = 1.0 / sqrt(wpix+(pix>0.0?
		(gainflag? pix*wpix/backnoise2:pix)*invgain : 0.0));
            npix++;
            }
          else
            *(pixout++) = *(wpixout++) = 0.0;
          }
        }
    }
  else
    {
    if (sflag)
      {
/*---- Sub-sampling case */
      for (y=ymin; y<ymax; y+=sn, pixout+=dw, wpixout+=dw)
        {
        for (x=xmin; x<xmax; x+=sn)
          {
          pix = 0.0;
          badflag = 0;
          for (sy=0; sy<sn; sy++)
            {
            dy2 = y+sy-iy;
            dy2 *= dy2;
            dx = x-ix;
            spixin = &PIX(field, x, y+sy);
            for (sx=sn; sx--;)
              {
              dr2 = dy2 + dx*dx;
              dx++;
              spix = *(spixin++);
              if (dr2<rad2 && spix>-BIG && spix<satlevel)
                pix += spix;
              else
                badflag=1;
              }
            }
          *(pixout++) = pix;
          if (!badflag)	/* A single bad pixel ruins is all (saturation, etc.)*/
            {
            *(wpixout++) = 1.0 / sqrt(backnoise2 + (pix>0.0?pix*invgain:0.0));
            npix++;
            }
          else
            *(wpixout++) = 0.0;
          }
        }
      }
    else
      for (y=ymin; y<ymax; y++, pixout+=dw,wpixout+=dw)
        {
        dy2 = y-iy;
        dy2 *= dy2;
        pixin = &PIX(field, xmin, y);
        for (x=xmin; x<xmax; x++)
          {
          dx = x-ix;
          dr2 = dy2 + dx*dx;
          pix = *(pixin++);
          if (dr2<rad2 && pix>-BIG && pix<satlevel)
            {
            *(pixout++) = pix;
            *(wpixout++) = 1.0 / sqrt(backnoise2 + (pix>0.0?pix*invgain : 0.0));
            npix++;
            }
          else
            *(pixout++) = *(wpixout++) = 0.0;
          }
        }
    }
 
  return npix;
  }


/****** profit_spiralindex ****************************************************
PROTO	float profit_spiralindex(profitstruct *profit)
PURPOSE	Compute the spiral index of a galaxy image (positive for arms
	extending counter-clockwise and negative for arms extending CW, 0 for
	no spiral pattern).
INPUT	Profile-fitting structure.
OUTPUT	Vector of residuals.
NOTES	-.
AUTHOR	E. Bertin (IAP)
VERSION	18/09/2008
 ***/
float profit_spiralindex(profitstruct *profit)
  {
   objstruct	*obj;
   obj2struct	*obj2;
   float	*dx,*dy, *fdx,*fdy, *gdx,*gdy, *gdxt,*gdyt, *pix,
		fwhm, invtwosigma2, hw,hh, ohw,ohh, x,y,xstart, tx,ty,txstart,
		gx,gy, r2, spirindex, invsig, val, sep;
   PIXTYPE	*fpix;
   int		i,j, npix;

  npix = profit->objnaxisn[0]*profit->objnaxisn[1];

  obj = profit->obj;
  obj2 = profit->obj2;
/* Compute simple derivative vectors at a fraction of the object scale */
  fwhm = obj2->hl_radius * 2.0 / 4.0;
  if (fwhm < 2.0)
    fwhm = 2.0;
  sep = 2.0;

  invtwosigma2 = -(2.35*2.35/(2.0*fwhm*fwhm));
  hw = (float)(profit->objnaxisn[0]/2);
  ohw = profit->objnaxisn[0] - hw;
  hh = (float)(profit->objnaxisn[1]/2);
  ohh = profit->objnaxisn[1] - hh;