Chris@42: /*
Chris@42:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@42:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@42:  *
Chris@42:  * This program is free software; you can redistribute it and/or modify
Chris@42:  * it under the terms of the GNU General Public License as published by
Chris@42:  * the Free Software Foundation; either version 2 of the License, or
Chris@42:  * (at your option) any later version.
Chris@42:  *
Chris@42:  * This program is distributed in the hope that it will be useful,
Chris@42:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@42:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@42:  * GNU General Public License for more details.
Chris@42:  *
Chris@42:  * You should have received a copy of the GNU General Public License
Chris@42:  * along with this program; if not, write to the Free Software
Chris@42:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@42:  *
Chris@42:  */
Chris@42: 
Chris@42: 
Chris@42: /* plans for rank-0 RDFTs (copy operations) */
Chris@42: 
Chris@42: #include "rdft.h"
Chris@42: 
Chris@42: #ifdef HAVE_STRING_H
Chris@42: #include <string.h>		/* for memcpy() */
Chris@42: #endif
Chris@42: 
Chris@42: #define MAXRNK 32 /* FIXME: should malloc() */
Chris@42: 
Chris@42: typedef struct {
Chris@42:      plan_rdft super;
Chris@42:      INT vl;
Chris@42:      int rnk;
Chris@42:      iodim d[MAXRNK];
Chris@42:      const char *nam;
Chris@42: } P;
Chris@42: 
Chris@42: typedef struct {
Chris@42:      solver super;
Chris@42:      rdftapply apply;
Chris@42:      int (*applicable)(const P *pln, const problem_rdft *p);
Chris@42:      const char *nam;
Chris@42: } S;
Chris@42: 
Chris@42: /* copy up to MAXRNK dimensions from problem into plan.  If a
Chris@42:    contiguous dimension exists, save its length in pln->vl */
Chris@42: static int fill_iodim(P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      int i;
Chris@42:      const tensor *vecsz = p->vecsz;
Chris@42: 
Chris@42:      pln->vl = 1;
Chris@42:      pln->rnk = 0;
Chris@42:      for (i = 0; i < vecsz->rnk; ++i) {
Chris@42: 	  /* extract contiguous dimensions */
Chris@42: 	  if (pln->vl == 1 &&
Chris@42: 	      vecsz->dims[i].is == 1 && vecsz->dims[i].os == 1) 
Chris@42: 	       pln->vl = vecsz->dims[i].n;
Chris@42: 	  else if (pln->rnk == MAXRNK) 
Chris@42: 	       return 0;
Chris@42: 	  else 
Chris@42: 	       pln->d[pln->rnk++] = vecsz->dims[i];
Chris@42:      }
Chris@42: 
Chris@42:      return 1;
Chris@42: }
Chris@42: 
Chris@42: /* generic higher-rank copy routine, calls cpy2d() to do the real work */
Chris@42: static void copy(const iodim *d, int rnk, INT vl,
Chris@42: 		 R *I, R *O,
Chris@42: 		 cpy2d_func cpy2d)
Chris@42: {
Chris@42:      A(rnk >= 2);
Chris@42:      if (rnk == 2)
Chris@42: 	  cpy2d(I, O, d[0].n, d[0].is, d[0].os, d[1].n, d[1].is, d[1].os, vl);
Chris@42:      else {
Chris@42: 	  INT i;
Chris@42: 	  for (i = 0; i < d[0].n; ++i, I += d[0].is, O += d[0].os)
Chris@42: 	       copy(d + 1, rnk - 1, vl, I, O, cpy2d);
Chris@42:      }
Chris@42: }
Chris@42: 
Chris@42: /* FIXME: should be more general */
Chris@42: static int transposep(const P *pln)
Chris@42: {
Chris@42:      int i;
Chris@42: 
Chris@42:      for (i = 0; i < pln->rnk - 2; ++i) 
Chris@42: 	  if (pln->d[i].is != pln->d[i].os)
Chris@42: 	       return 0;
Chris@42:      
Chris@42:      return (pln->d[i].n == pln->d[i+1].n &&
Chris@42: 	     pln->d[i].is == pln->d[i+1].os &&
Chris@42: 	     pln->d[i].os == pln->d[i+1].is);
Chris@42: }
Chris@42: 
Chris@42: /* generic higher-rank transpose routine, calls transpose2d() to do
Chris@42:  * the real work */
Chris@42: static void transpose(const iodim *d, int rnk, INT vl,
Chris@42: 		      R *I,
Chris@42: 		      transpose_func transpose2d)
Chris@42: {
Chris@42:      A(rnk >= 2);
Chris@42:      if (rnk == 2)
Chris@42: 	  transpose2d(I, d[0].n, d[0].is, d[0].os, vl);
Chris@42:      else {
Chris@42: 	  INT i;
Chris@42: 	  for (i = 0; i < d[0].n; ++i, I += d[0].is)
Chris@42: 	       transpose(d + 1, rnk - 1, vl, I, transpose2d);
Chris@42:      }
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* rank 0,1,2, out of place, iterative */
Chris@42: static void apply_iter(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42: 
Chris@42:      switch (ego->rnk) {
Chris@42: 	 case 0: 
Chris@42: 	      X(cpy1d)(I, O, ego->vl, 1, 1, 1);
Chris@42: 	      break;
Chris@42: 	 case 1:
Chris@42: 	      X(cpy1d)(I, O, 
Chris@42: 		       ego->d[0].n, ego->d[0].is, ego->d[0].os, 
Chris@42: 		       ego->vl);
Chris@42: 	      break;
Chris@42: 	 default:
Chris@42: 	      copy(ego->d, ego->rnk, ego->vl, I, O, X(cpy2d_ci));
Chris@42: 	      break;
Chris@42:      }
Chris@42: }
Chris@42: 
Chris@42: static int applicable_iter(const P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      UNUSED(pln);
Chris@42:      return (p->I != p->O);
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* out of place, write contiguous output */
Chris@42: static void apply_cpy2dco(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      copy(ego->d, ego->rnk, ego->vl, I, O, X(cpy2d_co));
Chris@42: }
Chris@42: 
Chris@42: static int applicable_cpy2dco(const P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      int rnk = pln->rnk;
Chris@42:      return (1
Chris@42: 	     && p->I != p->O
Chris@42: 	     && rnk >= 2
Chris@42: 
Chris@42: 	     /* must not duplicate apply_iter */
Chris@42: 	     && (X(iabs)(pln->d[rnk - 2].is) <= X(iabs)(pln->d[rnk - 1].is)
Chris@42: 		 ||
Chris@42: 		 X(iabs)(pln->d[rnk - 2].os) <= X(iabs)(pln->d[rnk - 1].os))
Chris@42: 	  );
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* out of place, tiled, no buffering */
Chris@42: static void apply_tiled(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      copy(ego->d, ego->rnk, ego->vl, I, O, X(cpy2d_tiled));
Chris@42: }
Chris@42: 
Chris@42: static int applicable_tiled(const P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      return (1
Chris@42: 	     && p->I != p->O
Chris@42: 	     && pln->rnk >= 2
Chris@42: 
Chris@42: 	     /* somewhat arbitrary */
Chris@42: 	     && X(compute_tilesz)(pln->vl, 1) > 4
Chris@42: 	  );
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* out of place, tiled, with buffer */
Chris@42: static void apply_tiledbuf(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      copy(ego->d, ego->rnk, ego->vl, I, O, X(cpy2d_tiledbuf));
Chris@42: }
Chris@42: 
Chris@42: #define applicable_tiledbuf applicable_tiled
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* rank 0, out of place, using memcpy */
Chris@42: static void apply_memcpy(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42: 
Chris@42:      A(ego->rnk == 0);
Chris@42:      memcpy(O, I, ego->vl * sizeof(R));
Chris@42: }
Chris@42: 
Chris@42: static int applicable_memcpy(const P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      return (1
Chris@42: 	     && p->I != p->O 
Chris@42: 	     && pln->rnk == 0
Chris@42: 	     && pln->vl > 2 /* do not bother memcpy-ing complex numbers */
Chris@42: 	     );
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* rank > 0 vecloop, out of place, using memcpy (e.g. out-of-place
Chris@42:    transposes of vl-tuples ... for large vl it should be more
Chris@42:    efficient to use memcpy than the tiled stuff). */
Chris@42: 
Chris@42: static void memcpy_loop(size_t cpysz, int rnk, const iodim *d, R *I, R *O)
Chris@42: {
Chris@42:      INT i, n = d->n, is = d->is, os = d->os;
Chris@42:      if (rnk == 1)
Chris@42: 	  for (i = 0; i < n; ++i, I += is, O += os)
Chris@42: 	       memcpy(O, I, cpysz);
Chris@42:      else {
Chris@42: 	  --rnk; ++d;
Chris@42: 	  for (i = 0; i < n; ++i, I += is, O += os)
Chris@42: 	       memcpy_loop(cpysz, rnk, d, I, O);
Chris@42:      }
Chris@42: }
Chris@42: 
Chris@42: static void apply_memcpy_loop(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      memcpy_loop(ego->vl * sizeof(R), ego->rnk, ego->d, I, O);
Chris@42: }
Chris@42: 
Chris@42: static int applicable_memcpy_loop(const P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      return (p->I != p->O
Chris@42: 	     && pln->rnk > 0
Chris@42:              && pln->vl > 2 /* do not bother memcpy-ing complex numbers */);
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* rank 2, in place, square transpose, iterative */
Chris@42: static void apply_ip_sq(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      UNUSED(O);
Chris@42:      transpose(ego->d, ego->rnk, ego->vl, I, X(transpose));
Chris@42: }
Chris@42: 
Chris@42: 
Chris@42: static int applicable_ip_sq(const P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      return (1
Chris@42: 	     && p->I == p->O
Chris@42: 	     && pln->rnk >= 2
Chris@42: 	     && transposep(pln));
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* rank 2, in place, square transpose, tiled */
Chris@42: static void apply_ip_sq_tiled(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      UNUSED(O);
Chris@42:      transpose(ego->d, ego->rnk, ego->vl, I, X(transpose_tiled));
Chris@42: }
Chris@42: 
Chris@42: static int applicable_ip_sq_tiled(const P *pln, const problem_rdft *p)
Chris@42: {
Chris@42:      return (1
Chris@42: 	     && applicable_ip_sq(pln, p)
Chris@42: 
Chris@42: 	     /* somewhat arbitrary */
Chris@42: 	     && X(compute_tilesz)(pln->vl, 2) > 4
Chris@42: 	  );
Chris@42: }
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: /* rank 2, in place, square transpose, tiled, buffered */
Chris@42: static void apply_ip_sq_tiledbuf(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      UNUSED(O);
Chris@42:      transpose(ego->d, ego->rnk, ego->vl, I, X(transpose_tiledbuf));
Chris@42: }
Chris@42: 
Chris@42: #define applicable_ip_sq_tiledbuf applicable_ip_sq_tiled
Chris@42: 
Chris@42: /**************************************************************/
Chris@42: static int applicable(const S *ego, const problem *p_)
Chris@42: {
Chris@42:      const problem_rdft *p = (const problem_rdft *) p_;
Chris@42:      P pln;
Chris@42:      return (1
Chris@42: 	     && p->sz->rnk == 0
Chris@42: 	     && FINITE_RNK(p->vecsz->rnk)
Chris@42: 	     && fill_iodim(&pln, p)
Chris@42: 	     && ego->applicable(&pln, p)
Chris@42: 	  );
Chris@42: }
Chris@42: 
Chris@42: static void print(const plan *ego_, printer *p)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      int i;
Chris@42:      p->print(p, "(%s/%D", ego->nam, ego->vl);
Chris@42:      for (i = 0; i < ego->rnk; ++i)
Chris@42: 	  p->print(p, "%v", ego->d[i].n);
Chris@42:      p->print(p, ")");
Chris@42: }
Chris@42: 
Chris@42: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
Chris@42: {
Chris@42:      const problem_rdft *p;
Chris@42:      const S *ego = (const S *) ego_;
Chris@42:      P *pln;
Chris@42:      int retval;
Chris@42: 
Chris@42:      static const plan_adt padt = {
Chris@42: 	  X(rdft_solve), X(null_awake), print, X(plan_null_destroy)
Chris@42:      };
Chris@42: 
Chris@42:      UNUSED(plnr);
Chris@42: 
Chris@42:      if (!applicable(ego, p_))
Chris@42:           return (plan *) 0;
Chris@42: 
Chris@42:      p = (const problem_rdft *) p_;
Chris@42:      pln = MKPLAN_RDFT(P, &padt, ego->apply);
Chris@42: 
Chris@42:      retval = fill_iodim(pln, p);
Chris@42:      (void)retval; /* UNUSED unless DEBUG */
Chris@42:      A(retval);
Chris@42:      A(pln->vl > 0); /* because FINITE_RNK(p->vecsz->rnk) holds */
Chris@42:      pln->nam = ego->nam;
Chris@42: 
Chris@42:      /* X(tensor_sz)(p->vecsz) loads, X(tensor_sz)(p->vecsz) stores */
Chris@42:      X(ops_other)(2 * X(tensor_sz)(p->vecsz), &pln->super.super.ops);
Chris@42:      return &(pln->super.super);
Chris@42: }
Chris@42: 
Chris@42: 
Chris@42: void X(rdft_rank0_register)(planner *p)
Chris@42: {
Chris@42:      unsigned i;
Chris@42:      static struct {
Chris@42: 	  rdftapply apply;
Chris@42: 	  int (*applicable)(const P *, const problem_rdft *);
Chris@42: 	  const char *nam;
Chris@42:      } tab[] = {
Chris@42: 	  { apply_memcpy,   applicable_memcpy,   "rdft-rank0-memcpy" },
Chris@42: 	  { apply_memcpy_loop,   applicable_memcpy_loop,  
Chris@42: 	    "rdft-rank0-memcpy-loop" },
Chris@42: 	  { apply_iter,     applicable_iter,     "rdft-rank0-iter-ci" },
Chris@42: 	  { apply_cpy2dco,  applicable_cpy2dco,  "rdft-rank0-iter-co" },
Chris@42: 	  { apply_tiled,    applicable_tiled,    "rdft-rank0-tiled" },
Chris@42: 	  { apply_tiledbuf, applicable_tiledbuf, "rdft-rank0-tiledbuf" },
Chris@42: 	  { apply_ip_sq,    applicable_ip_sq,    "rdft-rank0-ip-sq" },
Chris@42: 	  { 
Chris@42: 	       apply_ip_sq_tiled,
Chris@42: 	       applicable_ip_sq_tiled,
Chris@42: 	       "rdft-rank0-ip-sq-tiled" 
Chris@42: 	  },
Chris@42: 	  { 
Chris@42: 	       apply_ip_sq_tiledbuf,
Chris@42: 	       applicable_ip_sq_tiledbuf,
Chris@42: 	       "rdft-rank0-ip-sq-tiledbuf" 
Chris@42: 	  },
Chris@42:      };
Chris@42: 
Chris@42:      for (i = 0; i < sizeof(tab) / sizeof(tab[0]); ++i) {
Chris@42: 	  static const solver_adt sadt = { PROBLEM_RDFT, mkplan, 0 };
Chris@42: 	  S *slv = MKSOLVER(S, &sadt);
Chris@42: 	  slv->apply = tab[i].apply;
Chris@42: 	  slv->applicable = tab[i].applicable;
Chris@42: 	  slv->nam = tab[i].nam;
Chris@42: 	  REGISTER_SOLVER(p, &(slv->super));
Chris@42:      }
Chris@42: }