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: 
Chris@10: /* direct RDFT2 R2HC/HC2R solver, if we have a codelet */
Chris@10: 
Chris@10: #include "rdft.h"
Chris@10: 
Chris@10: typedef struct {
Chris@10:      solver super;
Chris@10:      const kr2c_desc *desc;
Chris@10:      kr2c k;
Chris@10: } S;
Chris@10: 
Chris@10: typedef struct {
Chris@10:      plan_rdft2 super;
Chris@10: 
Chris@10:      stride rs, cs;
Chris@10:      INT vl;
Chris@10:      INT ivs, ovs;
Chris@10:      kr2c k;
Chris@10:      const S *slv;
Chris@10:      INT ilast;
Chris@10: } P;
Chris@10: 
Chris@10: static void apply(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      ASSERT_ALIGNED_DOUBLE;
Chris@10:      ego->k(r0, r1, cr, ci,
Chris@10: 	    ego->rs, ego->cs, ego->cs,
Chris@10: 	    ego->vl, ego->ivs, ego->ovs);
Chris@10: }
Chris@10: 
Chris@10: static void apply_r2hc(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      INT i, vl = ego->vl, ovs = ego->ovs;
Chris@10:      ASSERT_ALIGNED_DOUBLE;
Chris@10:      ego->k(r0, r1, cr, ci,
Chris@10: 	    ego->rs, ego->cs, ego->cs,
Chris@10: 	    vl, ego->ivs, ovs);
Chris@10:      for (i = 0; i < vl; ++i, ci += ovs)
Chris@10: 	  ci[0] = ci[ego->ilast] = 0;
Chris@10: }
Chris@10: 
Chris@10: static void destroy(plan *ego_)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
Chris@10:      X(stride_destroy)(ego->rs);
Chris@10:      X(stride_destroy)(ego->cs);
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:      const S *s = ego->slv;
Chris@10: 
Chris@10:      p->print(p, "(rdft2-%s-direct-%D%v \"%s\")", 
Chris@10: 	      X(rdft_kind_str)(s->desc->genus->kind), s->desc->n, 
Chris@10: 	      ego->vl, s->desc->nam);
Chris@10: }
Chris@10: 
Chris@10: static int applicable(const solver *ego_, const problem *p_)
Chris@10: {
Chris@10:      const S *ego = (const S *) ego_;
Chris@10:      const kr2c_desc *desc = ego->desc;
Chris@10:      const problem_rdft2 *p = (const problem_rdft2 *) p_;
Chris@10:      INT vl;
Chris@10:      INT ivs, ovs;
Chris@10: 
Chris@10:      return (
Chris@10: 	  1
Chris@10: 	  && p->sz->rnk == 1
Chris@10: 	  && p->vecsz->rnk <= 1
Chris@10: 	  && p->sz->dims[0].n == desc->n
Chris@10: 	  && p->kind == desc->genus->kind
Chris@10: 
Chris@10: 	  /* check strides etc */
Chris@10: 	  && X(tensor_tornk1)(p->vecsz, &vl, &ivs, &ovs)
Chris@10: 
Chris@10: 	  && (0
Chris@10: 	      /* can operate out-of-place */
Chris@10: 	      || p->r0 != p->cr
Chris@10: 
Chris@10: 	      /*
Chris@10: 	       * can compute one transform in-place, no matter
Chris@10: 	       * what the strides are.
Chris@10: 	       */
Chris@10: 	      || p->vecsz->rnk == 0
Chris@10: 
Chris@10: 	      /* can operate in-place as long as strides are the same */
Chris@10: 	      || X(rdft2_inplace_strides)(p, RNK_MINFTY)
Chris@10: 	       )
Chris@10: 	  );
Chris@10: }
Chris@10: 
Chris@10: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
Chris@10: {
Chris@10:      const S *ego = (const S *) ego_;
Chris@10:      P *pln;
Chris@10:      const problem_rdft2 *p;
Chris@10:      iodim *d;
Chris@10:      int r2hc_kindp;
Chris@10: 
Chris@10:      static const plan_adt padt = {
Chris@10: 	  X(rdft2_solve), X(null_awake), print, destroy
Chris@10:      };
Chris@10: 
Chris@10:      UNUSED(plnr);
Chris@10: 
Chris@10:      if (!applicable(ego_, p_))
Chris@10:           return (plan *)0;
Chris@10: 
Chris@10:      p = (const problem_rdft2 *) p_;
Chris@10: 
Chris@10:      r2hc_kindp = R2HC_KINDP(p->kind);
Chris@10:      A(r2hc_kindp || HC2R_KINDP(p->kind));
Chris@10: 
Chris@10:      pln = MKPLAN_RDFT2(P, &padt, p->kind == R2HC ? apply_r2hc : apply);
Chris@10: 
Chris@10:      d = p->sz->dims;
Chris@10: 
Chris@10:      pln->k = ego->k;
Chris@10: 
Chris@10:      pln->rs = X(mkstride)(d->n, r2hc_kindp ? d->is : d->os);
Chris@10:      pln->cs = X(mkstride)(d->n, r2hc_kindp ? d->os : d->is);
Chris@10: 
Chris@10:      X(tensor_tornk1)(p->vecsz, &pln->vl, &pln->ivs, &pln->ovs);
Chris@10: 
Chris@10:      /* Nyquist freq., if any */
Chris@10:      pln->ilast = (d->n % 2) ? 0 : (d->n/2) * d->os;
Chris@10: 
Chris@10:      pln->slv = ego;
Chris@10:      X(ops_zero)(&pln->super.super.ops);
Chris@10:      X(ops_madd2)(pln->vl / ego->desc->genus->vl,
Chris@10: 		  &ego->desc->ops,
Chris@10: 		  &pln->super.super.ops);
Chris@10:      if (p->kind == R2HC)
Chris@10: 	  pln->super.super.ops.other += 2 * pln->vl; /* + 2 stores */
Chris@10: 
Chris@10:      pln->super.super.could_prune_now_p = 1;
Chris@10:      return &(pln->super.super);
Chris@10: }
Chris@10: 
Chris@10: /* constructor */
Chris@10: solver *X(mksolver_rdft2_direct)(kr2c k, const kr2c_desc *desc)
Chris@10: {
Chris@10:      static const solver_adt sadt = { PROBLEM_RDFT2, mkplan, 0 };
Chris@10:      S *slv = MKSOLVER(S, &sadt);
Chris@10:      slv->k = k;
Chris@10:      slv->desc = desc;
Chris@10:      return &(slv->super);
Chris@10: }