/* Copyright (C) 1999, 2000, 2001, 2002, Massachusetts Institute of Technology.
 *
 * This program 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.
 *
 * This program 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 this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#include <stdlib.h>
#include <stdio.h>
#include <math.h>

#include "../config.h"
#include <check.h>

#include <mpiglue.h>
#include "maxwell.h"

#define PRECOND_SUBTR_EIGS 0

#define PRECOND_MIN_DENOM 5e-3
#define MIN2(a,b) ((a) < (b) ? (a) : (b))
#define MAX2(a,b) ((a) > (b) ? (a) : (b))

/* used to be: MAX2(x, PRECOND_MIN_DENOM) but now zero k-point is
   handled specially */
#define FIX_DENOM(x) ((x) == 0 ? 1.0 : (x))

void maxwell_simple_precondition(evectmatrix X, void *data, real *eigenvals)
{
     maxwell_data *d = (maxwell_data *) data;
     int i, c, b;
     real *kpGn2 = d->k_plus_G_normsqr;

     for (i = 0; i < X.localN; ++i) {
	  for (c = 0; c < X.c; ++c) {
	       for (b = 0; b < X.p; ++b) {
		    int index = (i * X.c + c) * X.p + b;
		    real scale = kpGn2[i] * d->eps_inv_mean;

#if PRECOND_SUBTR_EIGS
		    if (eigenvals) {
			 scale -= eigenvals[b];
			 scale = 1.0 / (scale + copysign(PRECOND_MIN_DENOM, 
							 scale));
		    }
		    else
#else
		    {
			 scale = 1.0 / FIX_DENOM(scale);
		    }
#endif

		    ASSIGN_SCALAR(X.data[index],
				  scale * SCALAR_RE(X.data[index]),
				  scale * SCALAR_IM(X.data[index]));
	       }
	  }
     }
}

void maxwell_preconditioner(evectmatrix Xin, evectmatrix Xout, void *data,
			    evectmatrix Y, real *eigenvals, sqmatrix YtY)
{
     evectmatrix_XeYS(Xout, Xin, YtY, 1);
     maxwell_simple_precondition(Xout, data, eigenvals);
}

void maxwell_target_preconditioner(evectmatrix Xin, evectmatrix Xout, 
				   void *data,
				   evectmatrix Y, real *eigenvals,
				   sqmatrix YtY)
{
     maxwell_target_data *td = (maxwell_target_data *) data;
     maxwell_data *d = td->d;
     real omega_sqr = td->target_frequency * td->target_frequency;
     int i, c, b;
     real *kpGn2 = d->k_plus_G_normsqr;

     evectmatrix_XeYS(Xout, Xin, YtY, 1);

     for (i = 0; i < Xout.localN; ++i) {
	  for (c = 0; c < Xout.c; ++c) {
	       for (b = 0; b < Xout.p; ++b) {
		    int index = (i * Xout.c + c) * Xout.p + b;
		    real scale = kpGn2[i] * d->eps_inv_mean;

#if PRECOND_SUBTR_EIGS
		    scale -= omega_sqr;
#endif

		    scale = scale * scale;

#if PRECOND_SUBTR_EIGS
		    if (eigenvals) {
			 scale -= eigenvals[b];
			 scale = 1.0 / (scale + copysign(PRECOND_MIN_DENOM,
							 scale));
		    }
		    else
#else
		    {
			 scale = 1.0 / FIX_DENOM(scale);
		    }
#endif

		    ASSIGN_SCALAR(Xout.data[index],
				  scale * SCALAR_RE(Xout.data[index]),
				  scale * SCALAR_IM(Xout.data[index]));
	       }
	  }
     }
}

/**************************************************************************/

/* Fancy preconditioners */

/* Compute 'a' where v = k x a (cross product).

   Here, a = (a[0],a[1],a[2]) and k = (k.kx,k.ky,k.kz) are in
   cartesian coordinates.  (v[0],v[vstride]) is in the transverse basis of
   k.m and k.n. 

   We can't compute 'a' exactly, since there is no way to find the
   component of a parallel to k.  So, we only compute the transverse
   component of 'a'--this is the main approximation in our preconditioner.
*/
static void assign_crossinv_t2c(scalar *a, const k_data k,
				const scalar *v, int vstride)
{
     /* k x v = k x (k x a) = (k*a)k - k^2 a
	      = -(a_transverse) * k^2
        ==> a_transverse = -1/k^2 * k x v     */
     
     /* Thus, we just do the same thing as assign_cross_t2c
	in maxwell_op.c, except that we divide by -k^2: */

     scalar v0 = v[0], v1 = v[vstride];
     real kmag_inv = -1.0 / FIX_DENOM(k.kmag);

     ASSIGN_SCALAR(a[0],
                   (SCALAR_RE(v0)*k.nx - SCALAR_RE(v1)*k.mx) * kmag_inv,
                   (SCALAR_IM(v0)*k.nx - SCALAR_IM(v1)*k.mx) * kmag_inv);
     ASSIGN_SCALAR(a[1],
                   (SCALAR_RE(v0)*k.ny - SCALAR_RE(v1)*k.my) * kmag_inv,
                   (SCALAR_IM(v0)*k.ny - SCALAR_IM(v1)*k.my) * kmag_inv);
     ASSIGN_SCALAR(a[2],
                   (SCALAR_RE(v0)*k.nz - SCALAR_RE(v1)*k.mz) * kmag_inv,
                   (SCALAR_IM(v0)*k.nz - SCALAR_IM(v1)*k.mz) * kmag_inv);
}

/* Compute 'v' * scale, where a = k x v, going from cartesian to transverse
   coordinates.  Since v is tranvserse to k, we can compute this inverse
   exactly. */
static void assign_crossinv_c2t(scalar *v, int vstride,
				const k_data k, const scalar *a,
				real scale)
{
     /* As in assign_crossinv_t2c above, we find:

	v = v_transverse = -1/k^2 * k x a

	So, we do the same thing as in assign_cross_c2t of maxwell_op.c,
	with the additional -1/k^2 factor. */

     scalar a0 = a[0], a1 = a[1], a2 = a[2];
     scalar at0, at1;

     ASSIGN_SCALAR(at0,
          SCALAR_RE(a0)*k.mx + SCALAR_RE(a1)*k.my + SCALAR_RE(a2)*k.mz,
          SCALAR_IM(a0)*k.mx + SCALAR_IM(a1)*k.my + SCALAR_IM(a2)*k.mz);
     ASSIGN_SCALAR(at1,
          SCALAR_RE(a0)*k.nx + SCALAR_RE(a1)*k.ny + SCALAR_RE(a2)*k.nz,
          SCALAR_IM(a0)*k.nx + SCALAR_IM(a1)*k.ny + SCALAR_IM(a2)*k.nz);

     /* combine scale factor and k * (-1/k^2) */
     scale = -scale / FIX_DENOM(k.kmag);
     
     ASSIGN_SCALAR(v[0],
                   - scale * SCALAR_RE(at1),
                   - scale * SCALAR_IM(at1));
     ASSIGN_SCALAR(v[vstride],
                   scale * SCALAR_RE(at0),
                   scale * SCALAR_IM(at0));
}

/* Fancy preconditioner.  This is very similar to maxwell_op, except that
   the steps are (approximately) inverted: */

void maxwell_preconditioner2(evectmatrix Xin, evectmatrix Xout, void *data,
			     evectmatrix Y, real *eigenvals,
			     sqmatrix YtY)
{
     maxwell_data *d = (maxwell_data *) data;
     int cur_band_start;
     scalar *fft_data;
     scalar_complex *cdata;
     real scale;
     int i, j, b;

     CHECK(d, "null maxwell data pointer!");
     CHECK(Xin.c == 2, "fields don't have 2 components!");

     if (Xout.data != Xin.data)
	  evectmatrix_XeYS(Xout, Xin, YtY, 1);

     fft_data = d->fft_data;
     cdata = (scalar_complex *) fft_data;

     scale = -1.0 / Xout.N;  /* scale factor to normalize FFT;
                                negative sign comes from 2 i's from curls */

     for (cur_band_start = 0; cur_band_start < Xout.p;
          cur_band_start += d->num_fft_bands) {
          int cur_num_bands = MIN2(d->num_fft_bands, Xout.p - cur_band_start);

          /********************************************/
	  /* Compute approx. inverse of curl (inverse cross product with k): */

	  for (i = 0; i < d->other_dims; ++i)
	       for (j = 0; j < d->last_dim; ++j) {
		    int ij = i * d->last_dim + j;
		    int ij2 = i * d->last_dim_size + j;
		    k_data cur_k = d->k_plus_G[ij];
		    
		    for (b = 0; b < cur_num_bands; ++b)
			 assign_crossinv_t2c(&fft_data[3 * (ij2*cur_num_bands
							    + b)],
					     cur_k,
					     &Xout.data[ij * 2 * Xout.p +
						      b + cur_band_start],
					     Xout.p);
	       }

	  /********************************************/
	  /* Multiply by epsilon: */
	  
	  /* convert to position space via FFT: */
	  maxwell_compute_fft(+1, d, fft_data,
			      cur_num_bands*3, cur_num_bands*3, 1);

	  /* multiply by epsilon in position space.  Don't bother to
	     invert the whole epsilon-inverse tensor; just take the
	     inverse of the average epsilon-inverse (= trace / 3). */
	  for (i = 0; i < d->fft_output_size; ++i) {
	       symmetric_matrix eps_inv = d->eps_inv[i];
	       real eps = 3.0 / (eps_inv.m00 + eps_inv.m11 + eps_inv.m22);
	       for (b = 0; b < cur_num_bands; ++b) {
		    int ib = 3 * (i * cur_num_bands + b);
		    cdata[ib].re *= eps;
		    cdata[ib].im *= eps;
		    cdata[ib+1].re *= eps;
		    cdata[ib+1].im *= eps;
		    cdata[ib+2].re *= eps;
		    cdata[ib+2].im *= eps;
	       }
	  }

	  /* convert back to Fourier space */
          maxwell_compute_fft(-1, d, fft_data,
			      cur_num_bands*3, cur_num_bands*3, 1);

	  /********************************************/
	  /* Finally, do second inverse curl (inverse cross product with k): */

          for (i = 0; i < d->other_dims; ++i)
               for (j = 0; j < d->last_dim; ++j) {
                    int ij = i * d->last_dim + j;
                    int ij2 = i * d->last_dim_size + j;
                    k_data cur_k = d->k_plus_G[ij];

                    for (b = 0; b < cur_num_bands; ++b)
                         assign_crossinv_c2t(&Xout.data[ij * 2 * Xout.p +
						       b + cur_band_start],
					     Xout.p,
					     cur_k,
					     &fft_data[3 * (ij2*cur_num_bands
							    + b)],
					     scale);
               }

          /********************************************/

     } /* end of cur_band_start loop */
}

void maxwell_target_preconditioner2(evectmatrix Xin, evectmatrix Xout,
				    void *data,
				    evectmatrix Y, real *eigenvals,
				    sqmatrix YtY)
{
     maxwell_target_data *d = (maxwell_target_data *) data;
     maxwell_preconditioner2(Xin, Xout, d->d, Y, eigenvals, YtY);
     maxwell_preconditioner2(Xout, Xout, d->d, Y, eigenvals, YtY);
}


syntax highlighted by Code2HTML, v. 0.9.1