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: /* plans for rank-0 RDFT2 (copy operations, plus setting 0 imag. parts) */
Chris@10: 
Chris@10: #include "rdft.h"
Chris@10: 
Chris@10: #ifdef HAVE_STRING_H
Chris@10: #include <string.h>		/* for memcpy() */
Chris@10: #endif
Chris@10: 
Chris@10: typedef struct {
Chris@10:      solver super;
Chris@10: } S;
Chris@10: 
Chris@10: typedef struct {
Chris@10:      plan_rdft super;
Chris@10:      INT vl;
Chris@10:      INT ivs, ovs;
Chris@10:      plan *cldcpy;
Chris@10: } P;
Chris@10: 
Chris@10: static int applicable(const problem *p_)
Chris@10: {
Chris@10:      const problem_rdft2 *p = (const problem_rdft2 *) p_;
Chris@10:      return (1
Chris@10: 	     && p->sz->rnk == 0
Chris@10: 	     && (p->kind == HC2R
Chris@10: 		 ||
Chris@10: 		 (1
Chris@10: 		  && p->kind == R2HC
Chris@10: 		
Chris@10: 		  && p->vecsz->rnk <= 1
Chris@10:   
Chris@10: 		  && ((p->r0 != p->cr) 
Chris@10: 		      || 
Chris@10: 		      X(rdft2_inplace_strides)(p, RNK_MINFTY)) ))
Chris@10: 	  );
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;
Chris@10:      INT ivs = ego->ivs, ovs = ego->ovs;
Chris@10: 
Chris@10:      UNUSED(r1); /* rank-0 has no real odd-index elements */
Chris@10: 
Chris@10:      for (i = 4; i <= vl; i += 4) {
Chris@10:           R x0, x1, x2, x3;
Chris@10:           x0 = *r0; r0 += ivs;
Chris@10:           x1 = *r0; r0 += ivs;
Chris@10:           x2 = *r0; r0 += ivs;
Chris@10:           x3 = *r0; r0 += ivs;
Chris@10:           *cr = x0; cr += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10:           *cr = x1; cr += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10:           *cr = x2; cr += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10: 	  *cr = x3; cr += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10:      }
Chris@10:      for (; i < vl + 4; ++i) {
Chris@10:           R x0;
Chris@10:           x0 = *r0; r0 += ivs;
Chris@10:           *cr = x0; cr += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: /* in-place r2hc rank-0: set imaginary parts of output to 0 */
Chris@10: static void apply_r2hc_inplace(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;
Chris@10:      INT ovs = ego->ovs;
Chris@10: 
Chris@10:      UNUSED(r0); UNUSED(r1); UNUSED(cr);
Chris@10: 
Chris@10:      for (i = 4; i <= vl; i += 4) {
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10:      }
Chris@10:      for (; i < vl + 4; ++i) {
Chris@10: 	  *ci = K(0.0); ci += ovs;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: /* a rank-0 HC2R rdft2 problem is just a copy from cr to r0,
Chris@10:    so we can use a rank-0 rdft plan */
Chris@10: static void apply_hc2r(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      plan_rdft *cldcpy = (plan_rdft *) ego->cldcpy;
Chris@10:      UNUSED(ci);
Chris@10:      UNUSED(r1);
Chris@10:      cldcpy->apply((plan *) cldcpy, cr, r0);
Chris@10: }
Chris@10: 
Chris@10: static void awake(plan *ego_, enum wakefulness wakefulness)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
Chris@10:      if (ego->cldcpy)
Chris@10: 	  X(plan_awake)(ego->cldcpy, wakefulness);
Chris@10: }
Chris@10: 
Chris@10: static void destroy(plan *ego_)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
Chris@10:      if (ego->cldcpy)
Chris@10: 	  X(plan_destroy_internal)(ego->cldcpy);
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:      if (ego->cldcpy)
Chris@10: 	  p->print(p, "(rdft2-hc2r-rank0%(%p%))", ego->cldcpy);
Chris@10:      else
Chris@10: 	  p->print(p, "(rdft2-r2hc-rank0%v)", ego->vl);
Chris@10: }
Chris@10: 
Chris@10: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
Chris@10: {
Chris@10:      const problem_rdft2 *p;
Chris@10:      plan *cldcpy = (plan *) 0;
Chris@10:      P *pln;
Chris@10: 
Chris@10:      static const plan_adt padt = {
Chris@10: 	  X(rdft2_solve), awake, print, destroy
Chris@10:      };
Chris@10: 
Chris@10:      UNUSED(ego_);
Chris@10: 
Chris@10:      if (!applicable(p_))
Chris@10:           return (plan *) 0;
Chris@10: 
Chris@10:      p = (const problem_rdft2 *) p_;
Chris@10: 
Chris@10:      if (p->kind == HC2R) {
Chris@10: 	  cldcpy = X(mkplan_d)(plnr,
Chris@10: 			       X(mkproblem_rdft_0_d)(
Chris@10: 				    X(tensor_copy)(p->vecsz),
Chris@10: 				    p->cr, p->r0));
Chris@10: 	  if (!cldcpy) return (plan *) 0;
Chris@10:      }
Chris@10: 
Chris@10:      pln = MKPLAN_RDFT2(P, &padt, 
Chris@10: 			p->kind == R2HC ? 
Chris@10: 			(p->r0 == p->cr ? apply_r2hc_inplace : apply_r2hc) 
Chris@10: 			: apply_hc2r);
Chris@10:      
Chris@10:      if (p->kind == R2HC)
Chris@10: 	  X(tensor_tornk1)(p->vecsz, &pln->vl, &pln->ivs, &pln->ovs);
Chris@10:      pln->cldcpy = cldcpy;
Chris@10: 
Chris@10:      if (p->kind == R2HC) {
Chris@10: 	  /* vl loads, 2*vl stores */
Chris@10: 	  X(ops_other)(3 * pln->vl, &pln->super.super.ops);
Chris@10:      }
Chris@10:      else {
Chris@10: 	  pln->super.super.ops = cldcpy->ops;
Chris@10:      }
Chris@10: 
Chris@10:      return &(pln->super.super);
Chris@10: }
Chris@10: 
Chris@10: static solver *mksolver(void)
Chris@10: {
Chris@10:      static const solver_adt sadt = { PROBLEM_RDFT2, mkplan, 0 };
Chris@10:      S *slv = MKSOLVER(S, &sadt);
Chris@10:      return &(slv->super);
Chris@10: }
Chris@10: 
Chris@10: void X(rdft2_rank0_register)(planner *p)
Chris@10: {
Chris@10:      REGISTER_SOLVER(p, mksolver());
Chris@10: }