1 /*
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 */
20
21
22 #include "dft/dft.h"
23 #include "rdft/rdft.h"
24 #include <stddef.h>
25
destroy(problem * ego_)26 static void destroy(problem *ego_)
27 {
28 problem_rdft2 *ego = (problem_rdft2 *) ego_;
29 X(tensor_destroy2)(ego->vecsz, ego->sz);
30 X(ifree)(ego_);
31 }
32
hash(const problem * p_,md5 * m)33 static void hash(const problem *p_, md5 *m)
34 {
35 const problem_rdft2 *p = (const problem_rdft2 *) p_;
36 X(md5puts)(m, "rdft2");
37 X(md5int)(m, p->r0 == p->cr);
38 X(md5INT)(m, p->r1 - p->r0);
39 X(md5INT)(m, p->ci - p->cr);
40 X(md5int)(m, X(ialignment_of)(p->r0));
41 X(md5int)(m, X(ialignment_of)(p->r1));
42 X(md5int)(m, X(ialignment_of)(p->cr));
43 X(md5int)(m, X(ialignment_of)(p->ci));
44 X(md5int)(m, p->kind);
45 X(tensor_md5)(m, p->sz);
46 X(tensor_md5)(m, p->vecsz);
47 }
48
print(const problem * ego_,printer * p)49 static void print(const problem *ego_, printer *p)
50 {
51 const problem_rdft2 *ego = (const problem_rdft2 *) ego_;
52 p->print(p, "(rdft2 %d %d %T %T)",
53 (int)(ego->cr == ego->r0),
54 (int)(ego->kind),
55 ego->sz,
56 ego->vecsz);
57 }
58
recur(const iodim * dims,int rnk,R * I0,R * I1)59 static void recur(const iodim *dims, int rnk, R *I0, R *I1)
60 {
61 if (rnk == RNK_MINFTY)
62 return;
63 else if (rnk == 0)
64 I0[0] = K(0.0);
65 else if (rnk > 0) {
66 INT i, n = dims[0].n, is = dims[0].is;
67
68 if (rnk == 1) {
69 for (i = 0; i < n - 1; i += 2) {
70 *I0 = *I1 = K(0.0);
71 I0 += is; I1 += is;
72 }
73 if (i < n)
74 *I0 = K(0.0);
75 } else {
76 for (i = 0; i < n; ++i)
77 recur(dims + 1, rnk - 1, I0 + i * is, I1 + i * is);
78 }
79 }
80 }
81
vrecur(const iodim * vdims,int vrnk,const iodim * dims,int rnk,R * I0,R * I1)82 static void vrecur(const iodim *vdims, int vrnk,
83 const iodim *dims, int rnk, R *I0, R *I1)
84 {
85 if (vrnk == RNK_MINFTY)
86 return;
87 else if (vrnk == 0)
88 recur(dims, rnk, I0, I1);
89 else if (vrnk > 0) {
90 INT i, n = vdims[0].n, is = vdims[0].is;
91
92 for (i = 0; i < n; ++i)
93 vrecur(vdims + 1, vrnk - 1,
94 dims, rnk, I0 + i * is, I1 + i * is);
95 }
96 }
97
X(rdft2_complex_n)98 INT X(rdft2_complex_n)(INT real_n, rdft_kind kind)
99 {
100 switch (kind) {
101 case R2HC:
102 case HC2R:
103 return (real_n / 2) + 1;
104 case R2HCII:
105 case HC2RIII:
106 return (real_n + 1) / 2;
107 default:
108 /* can't happen */
109 A(0);
110 return 0;
111 }
112 }
113
zero(const problem * ego_)114 static void zero(const problem *ego_)
115 {
116 const problem_rdft2 *ego = (const problem_rdft2 *) ego_;
117 if (R2HC_KINDP(ego->kind)) {
118 /* FIXME: can we avoid the double recursion somehow? */
119 vrecur(ego->vecsz->dims, ego->vecsz->rnk,
120 ego->sz->dims, ego->sz->rnk,
121 UNTAINT(ego->r0), UNTAINT(ego->r1));
122 } else {
123 tensor *sz;
124 tensor *sz2 = X(tensor_copy)(ego->sz);
125 int rnk = sz2->rnk;
126 if (rnk > 0) /* ~half as many complex outputs */
127 sz2->dims[rnk-1].n =
128 X(rdft2_complex_n)(sz2->dims[rnk-1].n, ego->kind);
129 sz = X(tensor_append)(ego->vecsz, sz2);
130 X(tensor_destroy)(sz2);
131 X(dft_zerotens)(sz, UNTAINT(ego->cr), UNTAINT(ego->ci));
132 X(tensor_destroy)(sz);
133 }
134 }
135
136 static const problem_adt padt =
137 {
138 PROBLEM_RDFT2,
139 hash,
140 zero,
141 print,
142 destroy
143 };
144
X(mkproblem_rdft2)145 problem *X(mkproblem_rdft2)(const tensor *sz, const tensor *vecsz,
146 R *r0, R *r1, R *cr, R *ci,
147 rdft_kind kind)
148 {
149 problem_rdft2 *ego;
150
151 A(kind == R2HC || kind == R2HCII || kind == HC2R || kind == HC2RIII);
152 A(X(tensor_kosherp)(sz));
153 A(X(tensor_kosherp)(vecsz));
154 A(FINITE_RNK(sz->rnk));
155
156 /* require in-place problems to use r0 == cr */
157 if (UNTAINT(r0) == UNTAINT(ci))
158 return X(mkproblem_unsolvable)();
159
160 /* FIXME: should check UNTAINT(r1) == UNTAINT(cr) but
161 only if odd elements exist, which requires compressing the
162 tensors first */
163
164 if (UNTAINT(r0) == UNTAINT(cr))
165 r0 = cr = JOIN_TAINT(r0, cr);
166
167 ego = (problem_rdft2 *)X(mkproblem)(sizeof(problem_rdft2), &padt);
168
169 if (sz->rnk > 1) { /* have to compress rnk-1 dims separately, ugh */
170 tensor *szc = X(tensor_copy_except)(sz, sz->rnk - 1);
171 tensor *szr = X(tensor_copy_sub)(sz, sz->rnk - 1, 1);
172 tensor *szcc = X(tensor_compress)(szc);
173 if (szcc->rnk > 0)
174 ego->sz = X(tensor_append)(szcc, szr);
175 else
176 ego->sz = X(tensor_compress)(szr);
177 X(tensor_destroy2)(szc, szr); X(tensor_destroy)(szcc);
178 } else {
179 ego->sz = X(tensor_compress)(sz);
180 }
181 ego->vecsz = X(tensor_compress_contiguous)(vecsz);
182 ego->r0 = r0;
183 ego->r1 = r1;
184 ego->cr = cr;
185 ego->ci = ci;
186 ego->kind = kind;
187
188 A(FINITE_RNK(ego->sz->rnk));
189 return &(ego->super);
190
191 }
192
193 /* Same as X(mkproblem_rdft2), but also destroy input tensors. */
X(mkproblem_rdft2_d)194 problem *X(mkproblem_rdft2_d)(tensor *sz, tensor *vecsz,
195 R *r0, R *r1, R *cr, R *ci, rdft_kind kind)
196 {
197 problem *p = X(mkproblem_rdft2)(sz, vecsz, r0, r1, cr, ci, kind);
198 X(tensor_destroy2)(vecsz, sz);
199 return p;
200 }
201
202 /* Same as X(mkproblem_rdft2_d), but with only one R pointer.
203 Used by the API. */
X(mkproblem_rdft2_d_3pointers)204 problem *X(mkproblem_rdft2_d_3pointers)(tensor *sz, tensor *vecsz,
205 R *r0, R *cr, R *ci, rdft_kind kind)
206 {
207 problem *p;
208 int rnk = sz->rnk;
209 R *r1;
210
211 if (rnk == 0)
212 r1 = r0;
213 else if (R2HC_KINDP(kind)) {
214 r1 = r0 + sz->dims[rnk-1].is;
215 sz->dims[rnk-1].is *= 2;
216 } else {
217 r1 = r0 + sz->dims[rnk-1].os;
218 sz->dims[rnk-1].os *= 2;
219 }
220
221 p = X(mkproblem_rdft2)(sz, vecsz, r0, r1, cr, ci, kind);
222 X(tensor_destroy2)(vecsz, sz);
223 return p;
224 }
225