Chris@10: /*
Chris@10:  * Copyright (c) 2003, 2007-11 Matteo Frigo
Chris@10:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
Chris@10:  *
Chris@10:  * This program is free software; you can redistribute it and/or modify
Chris@10:  * it under the terms of the GNU General Public License as published by
Chris@10:  * the Free Software Foundation; either version 2 of the License, or
Chris@10:  * (at your option) any later version.
Chris@10:  *
Chris@10:  * This program is distributed in the hope that it will be useful,
Chris@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@10:  * GNU General Public License for more details.
Chris@10:  *
Chris@10:  * You should have received a copy of the GNU General Public License
Chris@10:  * along with this program; if not, write to the Free Software
Chris@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@10:  *
Chris@10:  */
Chris@10: 
Chris@10: /* express a twiddle problem in terms of dft + multiplication by
Chris@10:    twiddle factors */
Chris@10: 
Chris@10: #include "ct.h"
Chris@10: 
Chris@10: typedef ct_solver S;
Chris@10: 
Chris@10: typedef struct {
Chris@10:      plan_dftw super;
Chris@10: 
Chris@10:      INT r, rs, m, mb, me, ms, v, vs;
Chris@10: 
Chris@10:      plan *cld;
Chris@10: 
Chris@10:      twid *td;
Chris@10: 
Chris@10:      const S *slv;
Chris@10:      int dec;
Chris@10: } P;
Chris@10: 
Chris@10: static void mktwiddle(P *ego, enum wakefulness wakefulness)
Chris@10: {
Chris@10:      static const tw_instr tw[] = { { TW_FULL, 0, 0 }, { TW_NEXT, 1, 0 } };
Chris@10: 
Chris@10:      /* note that R and M are swapped, to allow for sequential
Chris@10: 	access both to data and twiddles */
Chris@10:      X(twiddle_awake)(wakefulness, &ego->td, tw,
Chris@10: 		      ego->r * ego->m, ego->m, ego->r);
Chris@10: }
Chris@10: 
Chris@10: static void bytwiddle(const P *ego, R *rio, R *iio)
Chris@10: {
Chris@10:      INT iv, ir, im;
Chris@10:      INT r = ego->r, rs = ego->rs;
Chris@10:      INT m = ego->m, mb = ego->mb, me = ego->me, ms = ego->ms;
Chris@10:      INT v = ego->v, vs = ego->vs;
Chris@10:      const R *W = ego->td->W;
Chris@10: 
Chris@10:      mb += (mb == 0); /* skip m=0 iteration */
Chris@10:      for (iv = 0; iv < v; ++iv) {
Chris@10: 	  for (ir = 1; ir < r; ++ir) {
Chris@10: 	       for (im = mb; im < me; ++im) {
Chris@10: 		    R *pr = rio + ms * im + rs * ir;
Chris@10: 		    R *pi = iio + ms * im + rs * ir;
Chris@10: 		    E xr = *pr;
Chris@10: 		    E xi = *pi;
Chris@10: 		    E wr = W[2 * im + (2 * (m-1)) * ir - 2];
Chris@10: 		    E wi = W[2 * im + (2 * (m-1)) * ir - 1];
Chris@10: 		    *pr = xr * wr + xi * wi;
Chris@10: 		    *pi = xi * wr - xr * wi;
Chris@10: 	       }
Chris@10: 	  }
Chris@10: 	  rio += vs;
Chris@10: 	  iio += vs;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static int applicable(INT irs, INT ors, INT ivs, INT ovs,
Chris@10: 		      const planner *plnr)
Chris@10: {
Chris@10:      return (1
Chris@10: 	     && irs == ors
Chris@10: 	     && ivs == ovs
Chris@10: 	     && !NO_SLOWP(plnr)
Chris@10: 	  );
Chris@10: }
Chris@10: 
Chris@10: static void apply_dit(const plan *ego_, R *rio, R *iio)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      plan_dft *cld;
Chris@10:      INT dm = ego->ms * ego->mb;
Chris@10: 
Chris@10:      bytwiddle(ego, rio, iio);
Chris@10: 
Chris@10:      cld = (plan_dft *) ego->cld;
Chris@10:      cld->apply(ego->cld, rio + dm, iio + dm, rio + dm, iio + dm);
Chris@10: }
Chris@10: 
Chris@10: static void apply_dif(const plan *ego_, R *rio, R *iio)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      plan_dft *cld;
Chris@10:      INT dm = ego->ms * ego->mb;
Chris@10: 
Chris@10:      cld = (plan_dft *) ego->cld;
Chris@10:      cld->apply(ego->cld, rio + dm, iio + dm, rio + dm, iio + dm);
Chris@10: 
Chris@10:      bytwiddle(ego, rio, iio);
Chris@10: }
Chris@10: 
Chris@10: static void awake(plan *ego_, enum wakefulness wakefulness)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
Chris@10:      X(plan_awake)(ego->cld, wakefulness);
Chris@10:      mktwiddle(ego, wakefulness);
Chris@10: }
Chris@10: 
Chris@10: static void destroy(plan *ego_)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
Chris@10:      X(plan_destroy_internal)(ego->cld);
Chris@10: }
Chris@10: 
Chris@10: static void print(const plan *ego_, printer *p)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      p->print(p, "(dftw-generic-%s-%D-%D%v%(%p%))",
Chris@10: 	      ego->dec == DECDIT ? "dit" : "dif",
Chris@10: 	      ego->r, ego->m, ego->v, ego->cld);
Chris@10: }
Chris@10: 
Chris@10: static plan *mkcldw(const ct_solver *ego_,
Chris@10: 		    INT r, INT irs, INT ors,
Chris@10: 		    INT m, INT ms,
Chris@10: 		    INT v, INT ivs, INT ovs,
Chris@10: 		    INT mstart, INT mcount,
Chris@10: 		    R *rio, R *iio,
Chris@10: 		    planner *plnr)
Chris@10: {
Chris@10:      const S *ego = (const S *)ego_;
Chris@10:      P *pln;
Chris@10:      plan *cld = 0;
Chris@10:      INT dm = ms * mstart;
Chris@10: 
Chris@10:      static const plan_adt padt = {
Chris@10: 	  0, awake, print, destroy
Chris@10:      };
Chris@10: 
Chris@10:      A(mstart >= 0 && mstart + mcount <= m);
Chris@10:      if (!applicable(irs, ors, ivs, ovs, plnr))
Chris@10:           return (plan *)0;
Chris@10: 
Chris@10:      cld = X(mkplan_d)(plnr,
Chris@10: 			X(mkproblem_dft_d)(
Chris@10: 			     X(mktensor_1d)(r, irs, irs),
Chris@10: 			     X(mktensor_2d)(mcount, ms, ms, v, ivs, ivs),
Chris@10: 			     rio + dm, iio + dm, rio + dm, iio + dm)
Chris@10: 			);
Chris@10:      if (!cld) goto nada;
Chris@10: 
Chris@10:      pln = MKPLAN_DFTW(P, &padt, ego->dec == DECDIT ? apply_dit : apply_dif);
Chris@10:      pln->slv = ego;
Chris@10:      pln->cld = cld;
Chris@10:      pln->r = r;
Chris@10:      pln->rs = irs;
Chris@10:      pln->m = m;
Chris@10:      pln->ms = ms;
Chris@10:      pln->v = v;
Chris@10:      pln->vs = ivs;
Chris@10:      pln->mb = mstart;
Chris@10:      pln->me = mstart + mcount;
Chris@10:      pln->dec = ego->dec;
Chris@10:      pln->td = 0;
Chris@10: 
Chris@10:      {
Chris@10: 	  double n0 = (r - 1) * (mcount - 1) * v;
Chris@10: 	  pln->super.super.ops = cld->ops;
Chris@10: 	  pln->super.super.ops.mul += 8 * n0;
Chris@10: 	  pln->super.super.ops.add += 4 * n0;
Chris@10: 	  pln->super.super.ops.other += 8 * n0;
Chris@10:      }
Chris@10:      return &(pln->super.super);
Chris@10: 
Chris@10:  nada:
Chris@10:      X(plan_destroy_internal)(cld);
Chris@10:      return (plan *) 0;
Chris@10: }
Chris@10: 
Chris@10: static void regsolver(planner *plnr, INT r, int dec)
Chris@10: {
Chris@10:      S *slv = (S *)X(mksolver_ct)(sizeof(S), r, dec, mkcldw, 0);
Chris@10:      REGISTER_SOLVER(plnr, &(slv->super));
Chris@10:      if (X(mksolver_ct_hook)) {
Chris@10: 	  slv = (S *)X(mksolver_ct_hook)(sizeof(S), r, dec, mkcldw, 0);
Chris@10: 	  REGISTER_SOLVER(plnr, &(slv->super));
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: void X(ct_generic_register)(planner *p)
Chris@10: {
Chris@10:      regsolver(p, 0, DECDIT);
Chris@10:      regsolver(p, 0, DECDIF);
Chris@10: }