cannam@95: /*
cannam@95:  * Copyright (c) 2003, 2007-11 Matteo Frigo
cannam@95:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
cannam@95:  *
cannam@95:  * This program is free software; you can redistribute it and/or modify
cannam@95:  * it under the terms of the GNU General Public License as published by
cannam@95:  * the Free Software Foundation; either version 2 of the License, or
cannam@95:  * (at your option) any later version.
cannam@95:  *
cannam@95:  * This program is distributed in the hope that it will be useful,
cannam@95:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@95:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@95:  * GNU General Public License for more details.
cannam@95:  *
cannam@95:  * You should have received a copy of the GNU General Public License
cannam@95:  * along with this program; if not, write to the Free Software
cannam@95:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@95:  *
cannam@95:  */
cannam@95: 
cannam@95: /* express a twiddle problem in terms of dft + multiplication by
cannam@95:    twiddle factors */
cannam@95: 
cannam@95: #include "ct.h"
cannam@95: 
cannam@95: typedef ct_solver S;
cannam@95: 
cannam@95: typedef struct {
cannam@95:      plan_dftw super;
cannam@95: 
cannam@95:      INT r, rs, m, mb, me, ms, v, vs;
cannam@95: 
cannam@95:      plan *cld;
cannam@95: 
cannam@95:      twid *td;
cannam@95: 
cannam@95:      const S *slv;
cannam@95:      int dec;
cannam@95: } P;
cannam@95: 
cannam@95: static void mktwiddle(P *ego, enum wakefulness wakefulness)
cannam@95: {
cannam@95:      static const tw_instr tw[] = { { TW_FULL, 0, 0 }, { TW_NEXT, 1, 0 } };
cannam@95: 
cannam@95:      /* note that R and M are swapped, to allow for sequential
cannam@95: 	access both to data and twiddles */
cannam@95:      X(twiddle_awake)(wakefulness, &ego->td, tw,
cannam@95: 		      ego->r * ego->m, ego->m, ego->r);
cannam@95: }
cannam@95: 
cannam@95: static void bytwiddle(const P *ego, R *rio, R *iio)
cannam@95: {
cannam@95:      INT iv, ir, im;
cannam@95:      INT r = ego->r, rs = ego->rs;
cannam@95:      INT m = ego->m, mb = ego->mb, me = ego->me, ms = ego->ms;
cannam@95:      INT v = ego->v, vs = ego->vs;
cannam@95:      const R *W = ego->td->W;
cannam@95: 
cannam@95:      mb += (mb == 0); /* skip m=0 iteration */
cannam@95:      for (iv = 0; iv < v; ++iv) {
cannam@95: 	  for (ir = 1; ir < r; ++ir) {
cannam@95: 	       for (im = mb; im < me; ++im) {
cannam@95: 		    R *pr = rio + ms * im + rs * ir;
cannam@95: 		    R *pi = iio + ms * im + rs * ir;
cannam@95: 		    E xr = *pr;
cannam@95: 		    E xi = *pi;
cannam@95: 		    E wr = W[2 * im + (2 * (m-1)) * ir - 2];
cannam@95: 		    E wi = W[2 * im + (2 * (m-1)) * ir - 1];
cannam@95: 		    *pr = xr * wr + xi * wi;
cannam@95: 		    *pi = xi * wr - xr * wi;
cannam@95: 	       }
cannam@95: 	  }
cannam@95: 	  rio += vs;
cannam@95: 	  iio += vs;
cannam@95:      }
cannam@95: }
cannam@95: 
cannam@95: static int applicable(INT irs, INT ors, INT ivs, INT ovs,
cannam@95: 		      const planner *plnr)
cannam@95: {
cannam@95:      return (1
cannam@95: 	     && irs == ors
cannam@95: 	     && ivs == ovs
cannam@95: 	     && !NO_SLOWP(plnr)
cannam@95: 	  );
cannam@95: }
cannam@95: 
cannam@95: static void apply_dit(const plan *ego_, R *rio, R *iio)
cannam@95: {
cannam@95:      const P *ego = (const P *) ego_;
cannam@95:      plan_dft *cld;
cannam@95:      INT dm = ego->ms * ego->mb;
cannam@95: 
cannam@95:      bytwiddle(ego, rio, iio);
cannam@95: 
cannam@95:      cld = (plan_dft *) ego->cld;
cannam@95:      cld->apply(ego->cld, rio + dm, iio + dm, rio + dm, iio + dm);
cannam@95: }
cannam@95: 
cannam@95: static void apply_dif(const plan *ego_, R *rio, R *iio)
cannam@95: {
cannam@95:      const P *ego = (const P *) ego_;
cannam@95:      plan_dft *cld;
cannam@95:      INT dm = ego->ms * ego->mb;
cannam@95: 
cannam@95:      cld = (plan_dft *) ego->cld;
cannam@95:      cld->apply(ego->cld, rio + dm, iio + dm, rio + dm, iio + dm);
cannam@95: 
cannam@95:      bytwiddle(ego, rio, iio);
cannam@95: }
cannam@95: 
cannam@95: static void awake(plan *ego_, enum wakefulness wakefulness)
cannam@95: {
cannam@95:      P *ego = (P *) ego_;
cannam@95:      X(plan_awake)(ego->cld, wakefulness);
cannam@95:      mktwiddle(ego, wakefulness);
cannam@95: }
cannam@95: 
cannam@95: static void destroy(plan *ego_)
cannam@95: {
cannam@95:      P *ego = (P *) ego_;
cannam@95:      X(plan_destroy_internal)(ego->cld);
cannam@95: }
cannam@95: 
cannam@95: static void print(const plan *ego_, printer *p)
cannam@95: {
cannam@95:      const P *ego = (const P *) ego_;
cannam@95:      p->print(p, "(dftw-generic-%s-%D-%D%v%(%p%))",
cannam@95: 	      ego->dec == DECDIT ? "dit" : "dif",
cannam@95: 	      ego->r, ego->m, ego->v, ego->cld);
cannam@95: }
cannam@95: 
cannam@95: static plan *mkcldw(const ct_solver *ego_,
cannam@95: 		    INT r, INT irs, INT ors,
cannam@95: 		    INT m, INT ms,
cannam@95: 		    INT v, INT ivs, INT ovs,
cannam@95: 		    INT mstart, INT mcount,
cannam@95: 		    R *rio, R *iio,
cannam@95: 		    planner *plnr)
cannam@95: {
cannam@95:      const S *ego = (const S *)ego_;
cannam@95:      P *pln;
cannam@95:      plan *cld = 0;
cannam@95:      INT dm = ms * mstart;
cannam@95: 
cannam@95:      static const plan_adt padt = {
cannam@95: 	  0, awake, print, destroy
cannam@95:      };
cannam@95: 
cannam@95:      A(mstart >= 0 && mstart + mcount <= m);
cannam@95:      if (!applicable(irs, ors, ivs, ovs, plnr))
cannam@95:           return (plan *)0;
cannam@95: 
cannam@95:      cld = X(mkplan_d)(plnr,
cannam@95: 			X(mkproblem_dft_d)(
cannam@95: 			     X(mktensor_1d)(r, irs, irs),
cannam@95: 			     X(mktensor_2d)(mcount, ms, ms, v, ivs, ivs),
cannam@95: 			     rio + dm, iio + dm, rio + dm, iio + dm)
cannam@95: 			);
cannam@95:      if (!cld) goto nada;
cannam@95: 
cannam@95:      pln = MKPLAN_DFTW(P, &padt, ego->dec == DECDIT ? apply_dit : apply_dif);
cannam@95:      pln->slv = ego;
cannam@95:      pln->cld = cld;
cannam@95:      pln->r = r;
cannam@95:      pln->rs = irs;
cannam@95:      pln->m = m;
cannam@95:      pln->ms = ms;
cannam@95:      pln->v = v;
cannam@95:      pln->vs = ivs;
cannam@95:      pln->mb = mstart;
cannam@95:      pln->me = mstart + mcount;
cannam@95:      pln->dec = ego->dec;
cannam@95:      pln->td = 0;
cannam@95: 
cannam@95:      {
cannam@95: 	  double n0 = (r - 1) * (mcount - 1) * v;
cannam@95: 	  pln->super.super.ops = cld->ops;
cannam@95: 	  pln->super.super.ops.mul += 8 * n0;
cannam@95: 	  pln->super.super.ops.add += 4 * n0;
cannam@95: 	  pln->super.super.ops.other += 8 * n0;
cannam@95:      }
cannam@95:      return &(pln->super.super);
cannam@95: 
cannam@95:  nada:
cannam@95:      X(plan_destroy_internal)(cld);
cannam@95:      return (plan *) 0;
cannam@95: }
cannam@95: 
cannam@95: static void regsolver(planner *plnr, INT r, int dec)
cannam@95: {
cannam@95:      S *slv = (S *)X(mksolver_ct)(sizeof(S), r, dec, mkcldw, 0);
cannam@95:      REGISTER_SOLVER(plnr, &(slv->super));
cannam@95:      if (X(mksolver_ct_hook)) {
cannam@95: 	  slv = (S *)X(mksolver_ct_hook)(sizeof(S), r, dec, mkcldw, 0);
cannam@95: 	  REGISTER_SOLVER(plnr, &(slv->super));
cannam@95:      }
cannam@95: }
cannam@95: 
cannam@95: void X(ct_generic_register)(planner *p)
cannam@95: {
cannam@95:      regsolver(p, 0, DECDIT);
cannam@95:      regsolver(p, 0, DECDIF);
cannam@95: }