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: #include "dft.h"
Chris@10: #include "rdft.h"
Chris@10: #include <stddef.h>
Chris@10: 
Chris@10: static void destroy(problem *ego_)
Chris@10: {
Chris@10:      problem_rdft2 *ego = (problem_rdft2 *) ego_;
Chris@10:      X(tensor_destroy2)(ego->vecsz, ego->sz);
Chris@10:      X(ifree)(ego_);
Chris@10: }
Chris@10: 
Chris@10: static void hash(const problem *p_, md5 *m)
Chris@10: {
Chris@10:      const problem_rdft2 *p = (const problem_rdft2 *) p_;
Chris@10:      X(md5puts)(m, "rdft2");
Chris@10:      X(md5int)(m, p->r0 == p->cr);
Chris@10:      X(md5INT)(m, p->r1 - p->r0);
Chris@10:      X(md5INT)(m, p->ci - p->cr);
Chris@10:      X(md5int)(m, X(alignment_of)(p->r0));
Chris@10:      X(md5int)(m, X(alignment_of)(p->r1));
Chris@10:      X(md5int)(m, X(alignment_of)(p->cr)); 
Chris@10:      X(md5int)(m, X(alignment_of)(p->ci)); 
Chris@10:      X(md5int)(m, p->kind);
Chris@10:      X(tensor_md5)(m, p->sz);
Chris@10:      X(tensor_md5)(m, p->vecsz);
Chris@10: }
Chris@10: 
Chris@10: static void print(const problem *ego_, printer *p)
Chris@10: {
Chris@10:      const problem_rdft2 *ego = (const problem_rdft2 *) ego_;
Chris@10:      p->print(p, "(rdft2 %d %d %T %T)", 
Chris@10: 	      (int)(ego->cr == ego->r0), 
Chris@10: 	      (int)(ego->kind),
Chris@10: 	      ego->sz,
Chris@10: 	      ego->vecsz);
Chris@10: }
Chris@10: 
Chris@10: static void recur(const iodim *dims, int rnk, R *I0, R *I1)
Chris@10: {
Chris@10:      if (rnk == RNK_MINFTY)
Chris@10:           return;
Chris@10:      else if (rnk == 0)
Chris@10:           I0[0] = K(0.0);
Chris@10:      else if (rnk > 0) {
Chris@10:           INT i, n = dims[0].n, is = dims[0].is;
Chris@10: 
Chris@10: 	  if (rnk == 1) {
Chris@10: 	       for (i = 0; i < n - 1; i += 2) {
Chris@10: 		    *I0 = *I1 = K(0.0);
Chris@10: 		    I0 += is; I1 += is;
Chris@10: 	       }
Chris@10: 	       if (i < n) 
Chris@10: 		    *I0 = K(0.0);
Chris@10: 	  } else {
Chris@10: 	       for (i = 0; i < n; ++i)
Chris@10: 		    recur(dims + 1, rnk - 1, I0 + i * is, I1 + i * is);
Chris@10: 	  }
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static void vrecur(const iodim *vdims, int vrnk,
Chris@10: 		   const iodim *dims, int rnk, R *I0, R *I1)
Chris@10: {
Chris@10:      if (vrnk == RNK_MINFTY)
Chris@10:           return;
Chris@10:      else if (vrnk == 0)
Chris@10: 	  recur(dims, rnk, I0, I1);
Chris@10:      else if (vrnk > 0) {
Chris@10:           INT i, n = vdims[0].n, is = vdims[0].is;
Chris@10: 
Chris@10: 	  for (i = 0; i < n; ++i)
Chris@10: 	       vrecur(vdims + 1, vrnk - 1, 
Chris@10: 		      dims, rnk, I0 + i * is, I1 + i * is);
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: INT X(rdft2_complex_n)(INT real_n, rdft_kind kind)
Chris@10: {
Chris@10:      switch (kind) {
Chris@10: 	 case R2HC:
Chris@10: 	 case HC2R:
Chris@10: 	      return (real_n / 2) + 1;
Chris@10: 	 case R2HCII:
Chris@10: 	 case HC2RIII:
Chris@10: 	      return (real_n + 1) / 2;
Chris@10: 	 default:
Chris@10: 	      /* can't happen */
Chris@10: 	      A(0);
Chris@10: 	      return 0;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static void zero(const problem *ego_)
Chris@10: {
Chris@10:      const problem_rdft2 *ego = (const problem_rdft2 *) ego_;
Chris@10:      if (R2HC_KINDP(ego->kind)) {
Chris@10: 	  /* FIXME: can we avoid the double recursion somehow? */
Chris@10: 	  vrecur(ego->vecsz->dims, ego->vecsz->rnk, 
Chris@10: 		 ego->sz->dims, ego->sz->rnk, 
Chris@10: 		 UNTAINT(ego->r0), UNTAINT(ego->r1));
Chris@10:      } else {
Chris@10: 	  tensor *sz;
Chris@10: 	  tensor *sz2 = X(tensor_copy)(ego->sz);
Chris@10: 	  int rnk = sz2->rnk;
Chris@10: 	  if (rnk > 0) /* ~half as many complex outputs */
Chris@10: 	       sz2->dims[rnk-1].n = 
Chris@10: 		    X(rdft2_complex_n)(sz2->dims[rnk-1].n, ego->kind);
Chris@10: 	  sz = X(tensor_append)(ego->vecsz, sz2);
Chris@10: 	  X(tensor_destroy)(sz2);
Chris@10: 	  X(dft_zerotens)(sz, UNTAINT(ego->cr), UNTAINT(ego->ci));
Chris@10: 	  X(tensor_destroy)(sz);
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static const problem_adt padt =
Chris@10: {
Chris@10:      PROBLEM_RDFT2,
Chris@10:      hash,
Chris@10:      zero,
Chris@10:      print,
Chris@10:      destroy
Chris@10: };
Chris@10: 
Chris@10: problem *X(mkproblem_rdft2)(const tensor *sz, const tensor *vecsz,
Chris@10: 			    R *r0, R *r1, R *cr, R *ci,
Chris@10: 			    rdft_kind kind)
Chris@10: {
Chris@10:      problem_rdft2 *ego;
Chris@10: 
Chris@10:      A(kind == R2HC || kind == R2HCII || kind == HC2R || kind == HC2RIII);
Chris@10:      A(X(tensor_kosherp)(sz));
Chris@10:      A(X(tensor_kosherp)(vecsz));
Chris@10:      A(FINITE_RNK(sz->rnk));
Chris@10: 
Chris@10:      /* require in-place problems to use r0 == cr */
Chris@10:      if (UNTAINT(r0) == UNTAINT(ci))
Chris@10: 	  return X(mkproblem_unsolvable)();
Chris@10: 
Chris@10:      /* FIXME: should check UNTAINT(r1) == UNTAINT(cr) but
Chris@10: 	only if odd elements exist, which requires compressing the 
Chris@10: 	tensors first */
Chris@10: 
Chris@10:      if (UNTAINT(r0) == UNTAINT(cr)) 
Chris@10: 	  r0 = cr = JOIN_TAINT(r0, cr);
Chris@10: 
Chris@10:      ego = (problem_rdft2 *)X(mkproblem)(sizeof(problem_rdft2), &padt);
Chris@10: 
Chris@10:      if (sz->rnk > 1) { /* have to compress rnk-1 dims separately, ugh */
Chris@10: 	  tensor *szc = X(tensor_copy_except)(sz, sz->rnk - 1);
Chris@10: 	  tensor *szr = X(tensor_copy_sub)(sz, sz->rnk - 1, 1);
Chris@10: 	  tensor *szcc = X(tensor_compress)(szc);
Chris@10: 	  if (szcc->rnk > 0)
Chris@10: 	       ego->sz = X(tensor_append)(szcc, szr);
Chris@10: 	  else
Chris@10: 	       ego->sz = X(tensor_compress)(szr);
Chris@10: 	  X(tensor_destroy2)(szc, szr); X(tensor_destroy)(szcc);
Chris@10:      } else {
Chris@10: 	  ego->sz = X(tensor_compress)(sz);
Chris@10:      }
Chris@10:      ego->vecsz = X(tensor_compress_contiguous)(vecsz);
Chris@10:      ego->r0 = r0;
Chris@10:      ego->r1 = r1;
Chris@10:      ego->cr = cr;
Chris@10:      ego->ci = ci;
Chris@10:      ego->kind = kind;
Chris@10: 
Chris@10:      A(FINITE_RNK(ego->sz->rnk));
Chris@10:      return &(ego->super);
Chris@10: 
Chris@10: }
Chris@10: 
Chris@10: /* Same as X(mkproblem_rdft2), but also destroy input tensors. */
Chris@10: problem *X(mkproblem_rdft2_d)(tensor *sz, tensor *vecsz,
Chris@10: 			      R *r0, R *r1, R *cr, R *ci, rdft_kind kind)
Chris@10: {
Chris@10:      problem *p = X(mkproblem_rdft2)(sz, vecsz, r0, r1, cr, ci, kind);
Chris@10:      X(tensor_destroy2)(vecsz, sz);
Chris@10:      return p;
Chris@10: }
Chris@10: 
Chris@10: /* Same as X(mkproblem_rdft2_d), but with only one R pointer.
Chris@10:    Used by the API. */
Chris@10: problem *X(mkproblem_rdft2_d_3pointers)(tensor *sz, tensor *vecsz,
Chris@10: 					R *r0, R *cr, R *ci, rdft_kind kind)
Chris@10: {
Chris@10:      problem *p;
Chris@10:      int rnk = sz->rnk;
Chris@10:      R *r1;
Chris@10: 
Chris@10:      if (rnk == 0)
Chris@10: 	  r1 = r0;
Chris@10:      else if (R2HC_KINDP(kind)) {
Chris@10: 	  r1 = r0 + sz->dims[rnk-1].is;
Chris@10: 	  sz->dims[rnk-1].is *= 2;
Chris@10:      } else {
Chris@10: 	  r1 = r0 + sz->dims[rnk-1].os;
Chris@10: 	  sz->dims[rnk-1].os *= 2;
Chris@10:      }
Chris@10: 
Chris@10:      p = X(mkproblem_rdft2)(sz, vecsz, r0, r1, cr, ci, kind);
Chris@10:      X(tensor_destroy2)(vecsz, sz);
Chris@10:      return p;
Chris@10: }