• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2008-2020 Stefan Krah. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  *
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  *
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS "AS IS" AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 
29 #include "mpdecimal.h"
30 
31 #include <assert.h>
32 #include <stdio.h>
33 
34 #include "bits.h"
35 #include "constants.h"
36 #include "difradix2.h"
37 #include "numbertheory.h"
38 #include "sixstep.h"
39 #include "transpose.h"
40 #include "umodarith.h"
41 
42 
43 /* Bignum: Cache efficient Matrix Fourier Transform for arrays of the
44    form 2**n (See literature/six-step.txt). */
45 
46 
47 /* forward transform with sign = -1 */
48 int
six_step_fnt(mpd_uint_t * a,mpd_size_t n,int modnum)49 six_step_fnt(mpd_uint_t *a, mpd_size_t n, int modnum)
50 {
51     struct fnt_params *tparams;
52     mpd_size_t log2n, C, R;
53     mpd_uint_t kernel;
54     mpd_uint_t umod;
55 #ifdef PPRO
56     double dmod;
57     uint32_t dinvmod[3];
58 #endif
59     mpd_uint_t *x, w0, w1, wstep;
60     mpd_size_t i, k;
61 
62 
63     assert(ispower2(n));
64     assert(n >= 16);
65     assert(n <= MPD_MAXTRANSFORM_2N);
66 
67     log2n = mpd_bsr(n);
68     C = ((mpd_size_t)1) << (log2n / 2);  /* number of columns */
69     R = ((mpd_size_t)1) << (log2n - (log2n / 2)); /* number of rows */
70 
71 
72     /* Transpose the matrix. */
73     if (!transpose_pow2(a, R, C)) {
74         return 0;
75     }
76 
77     /* Length R transform on the rows. */
78     if ((tparams = _mpd_init_fnt_params(R, -1, modnum)) == NULL) {
79         return 0;
80     }
81     for (x = a; x < a+n; x += R) {
82         fnt_dif2(x, R, tparams);
83     }
84 
85     /* Transpose the matrix. */
86     if (!transpose_pow2(a, C, R)) {
87         mpd_free(tparams);
88         return 0;
89     }
90 
91     /* Multiply each matrix element (addressed by i*C+k) by r**(i*k). */
92     SETMODULUS(modnum);
93     kernel = _mpd_getkernel(n, -1, modnum);
94     for (i = 1; i < R; i++) {
95         w0 = 1;                  /* r**(i*0): initial value for k=0 */
96         w1 = POWMOD(kernel, i);  /* r**(i*1): initial value for k=1 */
97         wstep = MULMOD(w1, w1);  /* r**(2*i) */
98         for (k = 0; k < C; k += 2) {
99             mpd_uint_t x0 = a[i*C+k];
100             mpd_uint_t x1 = a[i*C+k+1];
101             MULMOD2(&x0, w0, &x1, w1);
102             MULMOD2C(&w0, &w1, wstep);  /* r**(i*(k+2)) = r**(i*k) * r**(2*i) */
103             a[i*C+k] = x0;
104             a[i*C+k+1] = x1;
105         }
106     }
107 
108     /* Length C transform on the rows. */
109     if (C != R) {
110         mpd_free(tparams);
111         if ((tparams = _mpd_init_fnt_params(C, -1, modnum)) == NULL) {
112             return 0;
113         }
114     }
115     for (x = a; x < a+n; x += C) {
116         fnt_dif2(x, C, tparams);
117     }
118     mpd_free(tparams);
119 
120 #if 0
121     /* An unordered transform is sufficient for convolution. */
122     /* Transpose the matrix. */
123     if (!transpose_pow2(a, R, C)) {
124         return 0;
125     }
126 #endif
127 
128     return 1;
129 }
130 
131 
132 /* reverse transform, sign = 1 */
133 int
inv_six_step_fnt(mpd_uint_t * a,mpd_size_t n,int modnum)134 inv_six_step_fnt(mpd_uint_t *a, mpd_size_t n, int modnum)
135 {
136     struct fnt_params *tparams;
137     mpd_size_t log2n, C, R;
138     mpd_uint_t kernel;
139     mpd_uint_t umod;
140 #ifdef PPRO
141     double dmod;
142     uint32_t dinvmod[3];
143 #endif
144     mpd_uint_t *x, w0, w1, wstep;
145     mpd_size_t i, k;
146 
147 
148     assert(ispower2(n));
149     assert(n >= 16);
150     assert(n <= MPD_MAXTRANSFORM_2N);
151 
152     log2n = mpd_bsr(n);
153     C = ((mpd_size_t)1) << (log2n / 2); /* number of columns */
154     R = ((mpd_size_t)1) << (log2n - (log2n / 2)); /* number of rows */
155 
156 
157 #if 0
158     /* An unordered transform is sufficient for convolution. */
159     /* Transpose the matrix, producing an R*C matrix. */
160     if (!transpose_pow2(a, C, R)) {
161         return 0;
162     }
163 #endif
164 
165     /* Length C transform on the rows. */
166     if ((tparams = _mpd_init_fnt_params(C, 1, modnum)) == NULL) {
167         return 0;
168     }
169     for (x = a; x < a+n; x += C) {
170         fnt_dif2(x, C, tparams);
171     }
172 
173     /* Multiply each matrix element (addressed by i*C+k) by r**(i*k). */
174     SETMODULUS(modnum);
175     kernel = _mpd_getkernel(n, 1, modnum);
176     for (i = 1; i < R; i++) {
177         w0 = 1;
178         w1 = POWMOD(kernel, i);
179         wstep = MULMOD(w1, w1);
180         for (k = 0; k < C; k += 2) {
181             mpd_uint_t x0 = a[i*C+k];
182             mpd_uint_t x1 = a[i*C+k+1];
183             MULMOD2(&x0, w0, &x1, w1);
184             MULMOD2C(&w0, &w1, wstep);
185             a[i*C+k] = x0;
186             a[i*C+k+1] = x1;
187         }
188     }
189 
190     /* Transpose the matrix. */
191     if (!transpose_pow2(a, R, C)) {
192         mpd_free(tparams);
193         return 0;
194     }
195 
196     /* Length R transform on the rows. */
197     if (R != C) {
198         mpd_free(tparams);
199         if ((tparams = _mpd_init_fnt_params(R, 1, modnum)) == NULL) {
200             return 0;
201         }
202     }
203     for (x = a; x < a+n; x += R) {
204         fnt_dif2(x, R, tparams);
205     }
206     mpd_free(tparams);
207 
208     /* Transpose the matrix. */
209     if (!transpose_pow2(a, C, R)) {
210         return 0;
211     }
212 
213     return 1;
214 }
215