cannam@127: /*
cannam@127:  * Copyright (c) 2003, 2007-14 Matteo Frigo
cannam@127:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
cannam@127:  *
cannam@127:  * This program is free software; you can redistribute it and/or modify
cannam@127:  * it under the terms of the GNU General Public License as published by
cannam@127:  * the Free Software Foundation; either version 2 of the License, or
cannam@127:  * (at your option) any later version.
cannam@127:  *
cannam@127:  * This program is distributed in the hope that it will be useful,
cannam@127:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@127:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@127:  * GNU General Public License for more details.
cannam@127:  *
cannam@127:  * You should have received a copy of the GNU General Public License
cannam@127:  * along with this program; if not, write to the Free Software
cannam@127:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@127:  *
cannam@127:  */
cannam@127: 
cannam@127: 
cannam@127: /* plans for rank-0 RDFT2 (copy operations, plus setting 0 imag. parts) */
cannam@127: 
cannam@127: #include "rdft.h"
cannam@127: 
cannam@127: #ifdef HAVE_STRING_H
cannam@127: #include <string.h>		/* for memcpy() */
cannam@127: #endif
cannam@127: 
cannam@127: typedef struct {
cannam@127:      solver super;
cannam@127: } S;
cannam@127: 
cannam@127: typedef struct {
cannam@127:      plan_rdft super;
cannam@127:      INT vl;
cannam@127:      INT ivs, ovs;
cannam@127:      plan *cldcpy;
cannam@127: } P;
cannam@127: 
cannam@127: static int applicable(const problem *p_)
cannam@127: {
cannam@127:      const problem_rdft2 *p = (const problem_rdft2 *) p_;
cannam@127:      return (1
cannam@127: 	     && p->sz->rnk == 0
cannam@127: 	     && (p->kind == HC2R
cannam@127: 		 ||
cannam@127: 		 (1
cannam@127: 		  && p->kind == R2HC
cannam@127: 		
cannam@127: 		  && p->vecsz->rnk <= 1
cannam@127:   
cannam@127: 		  && ((p->r0 != p->cr) 
cannam@127: 		      || 
cannam@127: 		      X(rdft2_inplace_strides)(p, RNK_MINFTY)) ))
cannam@127: 	  );
cannam@127: }
cannam@127: 
cannam@127: static void apply_r2hc(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      INT i, vl = ego->vl;
cannam@127:      INT ivs = ego->ivs, ovs = ego->ovs;
cannam@127: 
cannam@127:      UNUSED(r1); /* rank-0 has no real odd-index elements */
cannam@127: 
cannam@127:      for (i = 4; i <= vl; i += 4) {
cannam@127:           R x0, x1, x2, x3;
cannam@127:           x0 = *r0; r0 += ivs;
cannam@127:           x1 = *r0; r0 += ivs;
cannam@127:           x2 = *r0; r0 += ivs;
cannam@127:           x3 = *r0; r0 += ivs;
cannam@127:           *cr = x0; cr += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127:           *cr = x1; cr += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127:           *cr = x2; cr += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127: 	  *cr = x3; cr += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127:      }
cannam@127:      for (; i < vl + 4; ++i) {
cannam@127:           R x0;
cannam@127:           x0 = *r0; r0 += ivs;
cannam@127:           *cr = x0; cr += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127:      }
cannam@127: }
cannam@127: 
cannam@127: /* in-place r2hc rank-0: set imaginary parts of output to 0 */
cannam@127: static void apply_r2hc_inplace(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      INT i, vl = ego->vl;
cannam@127:      INT ovs = ego->ovs;
cannam@127: 
cannam@127:      UNUSED(r0); UNUSED(r1); UNUSED(cr);
cannam@127: 
cannam@127:      for (i = 4; i <= vl; i += 4) {
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127:      }
cannam@127:      for (; i < vl + 4; ++i) {
cannam@127: 	  *ci = K(0.0); ci += ovs;
cannam@127:      }
cannam@127: }
cannam@127: 
cannam@127: /* a rank-0 HC2R rdft2 problem is just a copy from cr to r0,
cannam@127:    so we can use a rank-0 rdft plan */
cannam@127: static void apply_hc2r(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      plan_rdft *cldcpy = (plan_rdft *) ego->cldcpy;
cannam@127:      UNUSED(ci);
cannam@127:      UNUSED(r1);
cannam@127:      cldcpy->apply((plan *) cldcpy, cr, r0);
cannam@127: }
cannam@127: 
cannam@127: static void awake(plan *ego_, enum wakefulness wakefulness)
cannam@127: {
cannam@127:      P *ego = (P *) ego_;
cannam@127:      if (ego->cldcpy)
cannam@127: 	  X(plan_awake)(ego->cldcpy, wakefulness);
cannam@127: }
cannam@127: 
cannam@127: static void destroy(plan *ego_)
cannam@127: {
cannam@127:      P *ego = (P *) ego_;
cannam@127:      if (ego->cldcpy)
cannam@127: 	  X(plan_destroy_internal)(ego->cldcpy);
cannam@127: }
cannam@127: 
cannam@127: static void print(const plan *ego_, printer *p)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      if (ego->cldcpy)
cannam@127: 	  p->print(p, "(rdft2-hc2r-rank0%(%p%))", ego->cldcpy);
cannam@127:      else
cannam@127: 	  p->print(p, "(rdft2-r2hc-rank0%v)", ego->vl);
cannam@127: }
cannam@127: 
cannam@127: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
cannam@127: {
cannam@127:      const problem_rdft2 *p;
cannam@127:      plan *cldcpy = (plan *) 0;
cannam@127:      P *pln;
cannam@127: 
cannam@127:      static const plan_adt padt = {
cannam@127: 	  X(rdft2_solve), awake, print, destroy
cannam@127:      };
cannam@127: 
cannam@127:      UNUSED(ego_);
cannam@127: 
cannam@127:      if (!applicable(p_))
cannam@127:           return (plan *) 0;
cannam@127: 
cannam@127:      p = (const problem_rdft2 *) p_;
cannam@127: 
cannam@127:      if (p->kind == HC2R) {
cannam@127: 	  cldcpy = X(mkplan_d)(plnr,
cannam@127: 			       X(mkproblem_rdft_0_d)(
cannam@127: 				    X(tensor_copy)(p->vecsz),
cannam@127: 				    p->cr, p->r0));
cannam@127: 	  if (!cldcpy) return (plan *) 0;
cannam@127:      }
cannam@127: 
cannam@127:      pln = MKPLAN_RDFT2(P, &padt, 
cannam@127: 			p->kind == R2HC ? 
cannam@127: 			(p->r0 == p->cr ? apply_r2hc_inplace : apply_r2hc) 
cannam@127: 			: apply_hc2r);
cannam@127:      
cannam@127:      if (p->kind == R2HC)
cannam@127: 	  X(tensor_tornk1)(p->vecsz, &pln->vl, &pln->ivs, &pln->ovs);
cannam@127:      pln->cldcpy = cldcpy;
cannam@127: 
cannam@127:      if (p->kind == R2HC) {
cannam@127: 	  /* vl loads, 2*vl stores */
cannam@127: 	  X(ops_other)(3 * pln->vl, &pln->super.super.ops);
cannam@127:      }
cannam@127:      else {
cannam@127: 	  pln->super.super.ops = cldcpy->ops;
cannam@127:      }
cannam@127: 
cannam@127:      return &(pln->super.super);
cannam@127: }
cannam@127: 
cannam@127: static solver *mksolver(void)
cannam@127: {
cannam@127:      static const solver_adt sadt = { PROBLEM_RDFT2, mkplan, 0 };
cannam@127:      S *slv = MKSOLVER(S, &sadt);
cannam@127:      return &(slv->super);
cannam@127: }
cannam@127: 
cannam@127: void X(rdft2_rank0_register)(planner *p)
cannam@127: {
cannam@127:      REGISTER_SOLVER(p, mksolver());
cannam@127: }