1 | 1 |
deleted file mode 100644 |
... | ... |
@@ -1,276 +0,0 @@ |
1 |
-/* |
|
2 |
- * Template for the Discrete Cosine Transform for 32 samples |
|
3 |
- * Copyright (c) 2001, 2002 Fabrice Bellard |
|
4 |
- * |
|
5 |
- * This file is part of Libav. |
|
6 |
- * |
|
7 |
- * Libav is free software; you can redistribute it and/or |
|
8 |
- * modify it under the terms of the GNU Lesser General Public |
|
9 |
- * License as published by the Free Software Foundation; either |
|
10 |
- * version 2.1 of the License, or (at your option) any later version. |
|
11 |
- * |
|
12 |
- * Libav is distributed in the hope that it will be useful, |
|
13 |
- * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
14 |
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
15 |
- * Lesser General Public License for more details. |
|
16 |
- * |
|
17 |
- * You should have received a copy of the GNU Lesser General Public |
|
18 |
- * License along with Libav; if not, write to the Free Software |
|
19 |
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
20 |
- */ |
|
21 |
- |
|
22 |
-#include "dct32.h" |
|
23 |
-#include "mathops.h" |
|
24 |
- |
|
25 |
-#if DCT32_FLOAT |
|
26 |
-# define dct32 ff_dct32_float |
|
27 |
-# define FIXHR(x) ((float)(x)) |
|
28 |
-# define MULH3(x, y, s) ((s)*(y)*(x)) |
|
29 |
-# define INTFLOAT float |
|
30 |
-#else |
|
31 |
-# define dct32 ff_dct32_fixed |
|
32 |
-# define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5)) |
|
33 |
-# define MULH3(x, y, s) MULH((s)*(x), y) |
|
34 |
-# define INTFLOAT int |
|
35 |
-#endif |
|
36 |
- |
|
37 |
- |
|
38 |
-/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */ |
|
39 |
- |
|
40 |
-/* cos(i*pi/64) */ |
|
41 |
- |
|
42 |
-#define COS0_0 FIXHR(0.50060299823519630134/2) |
|
43 |
-#define COS0_1 FIXHR(0.50547095989754365998/2) |
|
44 |
-#define COS0_2 FIXHR(0.51544730992262454697/2) |
|
45 |
-#define COS0_3 FIXHR(0.53104259108978417447/2) |
|
46 |
-#define COS0_4 FIXHR(0.55310389603444452782/2) |
|
47 |
-#define COS0_5 FIXHR(0.58293496820613387367/2) |
|
48 |
-#define COS0_6 FIXHR(0.62250412303566481615/2) |
|
49 |
-#define COS0_7 FIXHR(0.67480834145500574602/2) |
|
50 |
-#define COS0_8 FIXHR(0.74453627100229844977/2) |
|
51 |
-#define COS0_9 FIXHR(0.83934964541552703873/2) |
|
52 |
-#define COS0_10 FIXHR(0.97256823786196069369/2) |
|
53 |
-#define COS0_11 FIXHR(1.16943993343288495515/4) |
|
54 |
-#define COS0_12 FIXHR(1.48416461631416627724/4) |
|
55 |
-#define COS0_13 FIXHR(2.05778100995341155085/8) |
|
56 |
-#define COS0_14 FIXHR(3.40760841846871878570/8) |
|
57 |
-#define COS0_15 FIXHR(10.19000812354805681150/32) |
|
58 |
- |
|
59 |
-#define COS1_0 FIXHR(0.50241928618815570551/2) |
|
60 |
-#define COS1_1 FIXHR(0.52249861493968888062/2) |
|
61 |
-#define COS1_2 FIXHR(0.56694403481635770368/2) |
|
62 |
-#define COS1_3 FIXHR(0.64682178335999012954/2) |
|
63 |
-#define COS1_4 FIXHR(0.78815462345125022473/2) |
|
64 |
-#define COS1_5 FIXHR(1.06067768599034747134/4) |
|
65 |
-#define COS1_6 FIXHR(1.72244709823833392782/4) |
|
66 |
-#define COS1_7 FIXHR(5.10114861868916385802/16) |
|
67 |
- |
|
68 |
-#define COS2_0 FIXHR(0.50979557910415916894/2) |
|
69 |
-#define COS2_1 FIXHR(0.60134488693504528054/2) |
|
70 |
-#define COS2_2 FIXHR(0.89997622313641570463/2) |
|
71 |
-#define COS2_3 FIXHR(2.56291544774150617881/8) |
|
72 |
- |
|
73 |
-#define COS3_0 FIXHR(0.54119610014619698439/2) |
|
74 |
-#define COS3_1 FIXHR(1.30656296487637652785/4) |
|
75 |
- |
|
76 |
-#define COS4_0 FIXHR(0.70710678118654752439/2) |
|
77 |
- |
|
78 |
-/* butterfly operator */ |
|
79 |
-#define BF(a, b, c, s)\ |
|
80 |
-{\ |
|
81 |
- tmp0 = val##a + val##b;\ |
|
82 |
- tmp1 = val##a - val##b;\ |
|
83 |
- val##a = tmp0;\ |
|
84 |
- val##b = MULH3(tmp1, c, 1<<(s));\ |
|
85 |
-} |
|
86 |
- |
|
87 |
-#define BF0(a, b, c, s)\ |
|
88 |
-{\ |
|
89 |
- tmp0 = tab[a] + tab[b];\ |
|
90 |
- tmp1 = tab[a] - tab[b];\ |
|
91 |
- val##a = tmp0;\ |
|
92 |
- val##b = MULH3(tmp1, c, 1<<(s));\ |
|
93 |
-} |
|
94 |
- |
|
95 |
-#define BF1(a, b, c, d)\ |
|
96 |
-{\ |
|
97 |
- BF(a, b, COS4_0, 1);\ |
|
98 |
- BF(c, d,-COS4_0, 1);\ |
|
99 |
- val##c += val##d;\ |
|
100 |
-} |
|
101 |
- |
|
102 |
-#define BF2(a, b, c, d)\ |
|
103 |
-{\ |
|
104 |
- BF(a, b, COS4_0, 1);\ |
|
105 |
- BF(c, d,-COS4_0, 1);\ |
|
106 |
- val##c += val##d;\ |
|
107 |
- val##a += val##c;\ |
|
108 |
- val##c += val##b;\ |
|
109 |
- val##b += val##d;\ |
|
110 |
-} |
|
111 |
- |
|
112 |
-#define ADD(a, b) val##a += val##b |
|
113 |
- |
|
114 |
-/* DCT32 without 1/sqrt(2) coef zero scaling. */ |
|
115 |
-void dct32(INTFLOAT *out, const INTFLOAT *tab) |
|
116 |
-{ |
|
117 |
- INTFLOAT tmp0, tmp1; |
|
118 |
- |
|
119 |
- INTFLOAT val0 , val1 , val2 , val3 , val4 , val5 , val6 , val7 , |
|
120 |
- val8 , val9 , val10, val11, val12, val13, val14, val15, |
|
121 |
- val16, val17, val18, val19, val20, val21, val22, val23, |
|
122 |
- val24, val25, val26, val27, val28, val29, val30, val31; |
|
123 |
- |
|
124 |
- /* pass 1 */ |
|
125 |
- BF0( 0, 31, COS0_0 , 1); |
|
126 |
- BF0(15, 16, COS0_15, 5); |
|
127 |
- /* pass 2 */ |
|
128 |
- BF( 0, 15, COS1_0 , 1); |
|
129 |
- BF(16, 31,-COS1_0 , 1); |
|
130 |
- /* pass 1 */ |
|
131 |
- BF0( 7, 24, COS0_7 , 1); |
|
132 |
- BF0( 8, 23, COS0_8 , 1); |
|
133 |
- /* pass 2 */ |
|
134 |
- BF( 7, 8, COS1_7 , 4); |
|
135 |
- BF(23, 24,-COS1_7 , 4); |
|
136 |
- /* pass 3 */ |
|
137 |
- BF( 0, 7, COS2_0 , 1); |
|
138 |
- BF( 8, 15,-COS2_0 , 1); |
|
139 |
- BF(16, 23, COS2_0 , 1); |
|
140 |
- BF(24, 31,-COS2_0 , 1); |
|
141 |
- /* pass 1 */ |
|
142 |
- BF0( 3, 28, COS0_3 , 1); |
|
143 |
- BF0(12, 19, COS0_12, 2); |
|
144 |
- /* pass 2 */ |
|
145 |
- BF( 3, 12, COS1_3 , 1); |
|
146 |
- BF(19, 28,-COS1_3 , 1); |
|
147 |
- /* pass 1 */ |
|
148 |
- BF0( 4, 27, COS0_4 , 1); |
|
149 |
- BF0(11, 20, COS0_11, 2); |
|
150 |
- /* pass 2 */ |
|
151 |
- BF( 4, 11, COS1_4 , 1); |
|
152 |
- BF(20, 27,-COS1_4 , 1); |
|
153 |
- /* pass 3 */ |
|
154 |
- BF( 3, 4, COS2_3 , 3); |
|
155 |
- BF(11, 12,-COS2_3 , 3); |
|
156 |
- BF(19, 20, COS2_3 , 3); |
|
157 |
- BF(27, 28,-COS2_3 , 3); |
|
158 |
- /* pass 4 */ |
|
159 |
- BF( 0, 3, COS3_0 , 1); |
|
160 |
- BF( 4, 7,-COS3_0 , 1); |
|
161 |
- BF( 8, 11, COS3_0 , 1); |
|
162 |
- BF(12, 15,-COS3_0 , 1); |
|
163 |
- BF(16, 19, COS3_0 , 1); |
|
164 |
- BF(20, 23,-COS3_0 , 1); |
|
165 |
- BF(24, 27, COS3_0 , 1); |
|
166 |
- BF(28, 31,-COS3_0 , 1); |
|
167 |
- |
|
168 |
- |
|
169 |
- |
|
170 |
- /* pass 1 */ |
|
171 |
- BF0( 1, 30, COS0_1 , 1); |
|
172 |
- BF0(14, 17, COS0_14, 3); |
|
173 |
- /* pass 2 */ |
|
174 |
- BF( 1, 14, COS1_1 , 1); |
|
175 |
- BF(17, 30,-COS1_1 , 1); |
|
176 |
- /* pass 1 */ |
|
177 |
- BF0( 6, 25, COS0_6 , 1); |
|
178 |
- BF0( 9, 22, COS0_9 , 1); |
|
179 |
- /* pass 2 */ |
|
180 |
- BF( 6, 9, COS1_6 , 2); |
|
181 |
- BF(22, 25,-COS1_6 , 2); |
|
182 |
- /* pass 3 */ |
|
183 |
- BF( 1, 6, COS2_1 , 1); |
|
184 |
- BF( 9, 14,-COS2_1 , 1); |
|
185 |
- BF(17, 22, COS2_1 , 1); |
|
186 |
- BF(25, 30,-COS2_1 , 1); |
|
187 |
- |
|
188 |
- /* pass 1 */ |
|
189 |
- BF0( 2, 29, COS0_2 , 1); |
|
190 |
- BF0(13, 18, COS0_13, 3); |
|
191 |
- /* pass 2 */ |
|
192 |
- BF( 2, 13, COS1_2 , 1); |
|
193 |
- BF(18, 29,-COS1_2 , 1); |
|
194 |
- /* pass 1 */ |
|
195 |
- BF0( 5, 26, COS0_5 , 1); |
|
196 |
- BF0(10, 21, COS0_10, 1); |
|
197 |
- /* pass 2 */ |
|
198 |
- BF( 5, 10, COS1_5 , 2); |
|
199 |
- BF(21, 26,-COS1_5 , 2); |
|
200 |
- /* pass 3 */ |
|
201 |
- BF( 2, 5, COS2_2 , 1); |
|
202 |
- BF(10, 13,-COS2_2 , 1); |
|
203 |
- BF(18, 21, COS2_2 , 1); |
|
204 |
- BF(26, 29,-COS2_2 , 1); |
|
205 |
- /* pass 4 */ |
|
206 |
- BF( 1, 2, COS3_1 , 2); |
|
207 |
- BF( 5, 6,-COS3_1 , 2); |
|
208 |
- BF( 9, 10, COS3_1 , 2); |
|
209 |
- BF(13, 14,-COS3_1 , 2); |
|
210 |
- BF(17, 18, COS3_1 , 2); |
|
211 |
- BF(21, 22,-COS3_1 , 2); |
|
212 |
- BF(25, 26, COS3_1 , 2); |
|
213 |
- BF(29, 30,-COS3_1 , 2); |
|
214 |
- |
|
215 |
- /* pass 5 */ |
|
216 |
- BF1( 0, 1, 2, 3); |
|
217 |
- BF2( 4, 5, 6, 7); |
|
218 |
- BF1( 8, 9, 10, 11); |
|
219 |
- BF2(12, 13, 14, 15); |
|
220 |
- BF1(16, 17, 18, 19); |
|
221 |
- BF2(20, 21, 22, 23); |
|
222 |
- BF1(24, 25, 26, 27); |
|
223 |
- BF2(28, 29, 30, 31); |
|
224 |
- |
|
225 |
- /* pass 6 */ |
|
226 |
- |
|
227 |
- ADD( 8, 12); |
|
228 |
- ADD(12, 10); |
|
229 |
- ADD(10, 14); |
|
230 |
- ADD(14, 9); |
|
231 |
- ADD( 9, 13); |
|
232 |
- ADD(13, 11); |
|
233 |
- ADD(11, 15); |
|
234 |
- |
|
235 |
- out[ 0] = val0; |
|
236 |
- out[16] = val1; |
|
237 |
- out[ 8] = val2; |
|
238 |
- out[24] = val3; |
|
239 |
- out[ 4] = val4; |
|
240 |
- out[20] = val5; |
|
241 |
- out[12] = val6; |
|
242 |
- out[28] = val7; |
|
243 |
- out[ 2] = val8; |
|
244 |
- out[18] = val9; |
|
245 |
- out[10] = val10; |
|
246 |
- out[26] = val11; |
|
247 |
- out[ 6] = val12; |
|
248 |
- out[22] = val13; |
|
249 |
- out[14] = val14; |
|
250 |
- out[30] = val15; |
|
251 |
- |
|
252 |
- ADD(24, 28); |
|
253 |
- ADD(28, 26); |
|
254 |
- ADD(26, 30); |
|
255 |
- ADD(30, 25); |
|
256 |
- ADD(25, 29); |
|
257 |
- ADD(29, 27); |
|
258 |
- ADD(27, 31); |
|
259 |
- |
|
260 |
- out[ 1] = val16 + val24; |
|
261 |
- out[17] = val17 + val25; |
|
262 |
- out[ 9] = val18 + val26; |
|
263 |
- out[25] = val19 + val27; |
|
264 |
- out[ 5] = val20 + val28; |
|
265 |
- out[21] = val21 + val29; |
|
266 |
- out[13] = val22 + val30; |
|
267 |
- out[29] = val23 + val31; |
|
268 |
- out[ 3] = val24 + val20; |
|
269 |
- out[19] = val25 + val21; |
|
270 |
- out[11] = val26 + val22; |
|
271 |
- out[27] = val27 + val23; |
|
272 |
- out[ 7] = val28 + val18; |
|
273 |
- out[23] = val29 + val19; |
|
274 |
- out[15] = val30 + val17; |
|
275 |
- out[31] = val31; |
|
276 |
-} |
21 | 21 |
new file mode 100644 |
... | ... |
@@ -0,0 +1,276 @@ |
0 |
+/* |
|
1 |
+ * Template for the Discrete Cosine Transform for 32 samples |
|
2 |
+ * Copyright (c) 2001, 2002 Fabrice Bellard |
|
3 |
+ * |
|
4 |
+ * This file is part of Libav. |
|
5 |
+ * |
|
6 |
+ * Libav is free software; you can redistribute it and/or |
|
7 |
+ * modify it under the terms of the GNU Lesser General Public |
|
8 |
+ * License as published by the Free Software Foundation; either |
|
9 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
10 |
+ * |
|
11 |
+ * Libav is distributed in the hope that it will be useful, |
|
12 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
13 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
14 |
+ * Lesser General Public License for more details. |
|
15 |
+ * |
|
16 |
+ * You should have received a copy of the GNU Lesser General Public |
|
17 |
+ * License along with Libav; if not, write to the Free Software |
|
18 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
19 |
+ */ |
|
20 |
+ |
|
21 |
+#include "dct32.h" |
|
22 |
+#include "mathops.h" |
|
23 |
+ |
|
24 |
+#if DCT32_FLOAT |
|
25 |
+# define dct32 ff_dct32_float |
|
26 |
+# define FIXHR(x) ((float)(x)) |
|
27 |
+# define MULH3(x, y, s) ((s)*(y)*(x)) |
|
28 |
+# define INTFLOAT float |
|
29 |
+#else |
|
30 |
+# define dct32 ff_dct32_fixed |
|
31 |
+# define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5)) |
|
32 |
+# define MULH3(x, y, s) MULH((s)*(x), y) |
|
33 |
+# define INTFLOAT int |
|
34 |
+#endif |
|
35 |
+ |
|
36 |
+ |
|
37 |
+/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */ |
|
38 |
+ |
|
39 |
+/* cos(i*pi/64) */ |
|
40 |
+ |
|
41 |
+#define COS0_0 FIXHR(0.50060299823519630134/2) |
|
42 |
+#define COS0_1 FIXHR(0.50547095989754365998/2) |
|
43 |
+#define COS0_2 FIXHR(0.51544730992262454697/2) |
|
44 |
+#define COS0_3 FIXHR(0.53104259108978417447/2) |
|
45 |
+#define COS0_4 FIXHR(0.55310389603444452782/2) |
|
46 |
+#define COS0_5 FIXHR(0.58293496820613387367/2) |
|
47 |
+#define COS0_6 FIXHR(0.62250412303566481615/2) |
|
48 |
+#define COS0_7 FIXHR(0.67480834145500574602/2) |
|
49 |
+#define COS0_8 FIXHR(0.74453627100229844977/2) |
|
50 |
+#define COS0_9 FIXHR(0.83934964541552703873/2) |
|
51 |
+#define COS0_10 FIXHR(0.97256823786196069369/2) |
|
52 |
+#define COS0_11 FIXHR(1.16943993343288495515/4) |
|
53 |
+#define COS0_12 FIXHR(1.48416461631416627724/4) |
|
54 |
+#define COS0_13 FIXHR(2.05778100995341155085/8) |
|
55 |
+#define COS0_14 FIXHR(3.40760841846871878570/8) |
|
56 |
+#define COS0_15 FIXHR(10.19000812354805681150/32) |
|
57 |
+ |
|
58 |
+#define COS1_0 FIXHR(0.50241928618815570551/2) |
|
59 |
+#define COS1_1 FIXHR(0.52249861493968888062/2) |
|
60 |
+#define COS1_2 FIXHR(0.56694403481635770368/2) |
|
61 |
+#define COS1_3 FIXHR(0.64682178335999012954/2) |
|
62 |
+#define COS1_4 FIXHR(0.78815462345125022473/2) |
|
63 |
+#define COS1_5 FIXHR(1.06067768599034747134/4) |
|
64 |
+#define COS1_6 FIXHR(1.72244709823833392782/4) |
|
65 |
+#define COS1_7 FIXHR(5.10114861868916385802/16) |
|
66 |
+ |
|
67 |
+#define COS2_0 FIXHR(0.50979557910415916894/2) |
|
68 |
+#define COS2_1 FIXHR(0.60134488693504528054/2) |
|
69 |
+#define COS2_2 FIXHR(0.89997622313641570463/2) |
|
70 |
+#define COS2_3 FIXHR(2.56291544774150617881/8) |
|
71 |
+ |
|
72 |
+#define COS3_0 FIXHR(0.54119610014619698439/2) |
|
73 |
+#define COS3_1 FIXHR(1.30656296487637652785/4) |
|
74 |
+ |
|
75 |
+#define COS4_0 FIXHR(0.70710678118654752439/2) |
|
76 |
+ |
|
77 |
+/* butterfly operator */ |
|
78 |
+#define BF(a, b, c, s)\ |
|
79 |
+{\ |
|
80 |
+ tmp0 = val##a + val##b;\ |
|
81 |
+ tmp1 = val##a - val##b;\ |
|
82 |
+ val##a = tmp0;\ |
|
83 |
+ val##b = MULH3(tmp1, c, 1<<(s));\ |
|
84 |
+} |
|
85 |
+ |
|
86 |
+#define BF0(a, b, c, s)\ |
|
87 |
+{\ |
|
88 |
+ tmp0 = tab[a] + tab[b];\ |
|
89 |
+ tmp1 = tab[a] - tab[b];\ |
|
90 |
+ val##a = tmp0;\ |
|
91 |
+ val##b = MULH3(tmp1, c, 1<<(s));\ |
|
92 |
+} |
|
93 |
+ |
|
94 |
+#define BF1(a, b, c, d)\ |
|
95 |
+{\ |
|
96 |
+ BF(a, b, COS4_0, 1);\ |
|
97 |
+ BF(c, d,-COS4_0, 1);\ |
|
98 |
+ val##c += val##d;\ |
|
99 |
+} |
|
100 |
+ |
|
101 |
+#define BF2(a, b, c, d)\ |
|
102 |
+{\ |
|
103 |
+ BF(a, b, COS4_0, 1);\ |
|
104 |
+ BF(c, d,-COS4_0, 1);\ |
|
105 |
+ val##c += val##d;\ |
|
106 |
+ val##a += val##c;\ |
|
107 |
+ val##c += val##b;\ |
|
108 |
+ val##b += val##d;\ |
|
109 |
+} |
|
110 |
+ |
|
111 |
+#define ADD(a, b) val##a += val##b |
|
112 |
+ |
|
113 |
+/* DCT32 without 1/sqrt(2) coef zero scaling. */ |
|
114 |
+void dct32(INTFLOAT *out, const INTFLOAT *tab) |
|
115 |
+{ |
|
116 |
+ INTFLOAT tmp0, tmp1; |
|
117 |
+ |
|
118 |
+ INTFLOAT val0 , val1 , val2 , val3 , val4 , val5 , val6 , val7 , |
|
119 |
+ val8 , val9 , val10, val11, val12, val13, val14, val15, |
|
120 |
+ val16, val17, val18, val19, val20, val21, val22, val23, |
|
121 |
+ val24, val25, val26, val27, val28, val29, val30, val31; |
|
122 |
+ |
|
123 |
+ /* pass 1 */ |
|
124 |
+ BF0( 0, 31, COS0_0 , 1); |
|
125 |
+ BF0(15, 16, COS0_15, 5); |
|
126 |
+ /* pass 2 */ |
|
127 |
+ BF( 0, 15, COS1_0 , 1); |
|
128 |
+ BF(16, 31,-COS1_0 , 1); |
|
129 |
+ /* pass 1 */ |
|
130 |
+ BF0( 7, 24, COS0_7 , 1); |
|
131 |
+ BF0( 8, 23, COS0_8 , 1); |
|
132 |
+ /* pass 2 */ |
|
133 |
+ BF( 7, 8, COS1_7 , 4); |
|
134 |
+ BF(23, 24,-COS1_7 , 4); |
|
135 |
+ /* pass 3 */ |
|
136 |
+ BF( 0, 7, COS2_0 , 1); |
|
137 |
+ BF( 8, 15,-COS2_0 , 1); |
|
138 |
+ BF(16, 23, COS2_0 , 1); |
|
139 |
+ BF(24, 31,-COS2_0 , 1); |
|
140 |
+ /* pass 1 */ |
|
141 |
+ BF0( 3, 28, COS0_3 , 1); |
|
142 |
+ BF0(12, 19, COS0_12, 2); |
|
143 |
+ /* pass 2 */ |
|
144 |
+ BF( 3, 12, COS1_3 , 1); |
|
145 |
+ BF(19, 28,-COS1_3 , 1); |
|
146 |
+ /* pass 1 */ |
|
147 |
+ BF0( 4, 27, COS0_4 , 1); |
|
148 |
+ BF0(11, 20, COS0_11, 2); |
|
149 |
+ /* pass 2 */ |
|
150 |
+ BF( 4, 11, COS1_4 , 1); |
|
151 |
+ BF(20, 27,-COS1_4 , 1); |
|
152 |
+ /* pass 3 */ |
|
153 |
+ BF( 3, 4, COS2_3 , 3); |
|
154 |
+ BF(11, 12,-COS2_3 , 3); |
|
155 |
+ BF(19, 20, COS2_3 , 3); |
|
156 |
+ BF(27, 28,-COS2_3 , 3); |
|
157 |
+ /* pass 4 */ |
|
158 |
+ BF( 0, 3, COS3_0 , 1); |
|
159 |
+ BF( 4, 7,-COS3_0 , 1); |
|
160 |
+ BF( 8, 11, COS3_0 , 1); |
|
161 |
+ BF(12, 15,-COS3_0 , 1); |
|
162 |
+ BF(16, 19, COS3_0 , 1); |
|
163 |
+ BF(20, 23,-COS3_0 , 1); |
|
164 |
+ BF(24, 27, COS3_0 , 1); |
|
165 |
+ BF(28, 31,-COS3_0 , 1); |
|
166 |
+ |
|
167 |
+ |
|
168 |
+ |
|
169 |
+ /* pass 1 */ |
|
170 |
+ BF0( 1, 30, COS0_1 , 1); |
|
171 |
+ BF0(14, 17, COS0_14, 3); |
|
172 |
+ /* pass 2 */ |
|
173 |
+ BF( 1, 14, COS1_1 , 1); |
|
174 |
+ BF(17, 30,-COS1_1 , 1); |
|
175 |
+ /* pass 1 */ |
|
176 |
+ BF0( 6, 25, COS0_6 , 1); |
|
177 |
+ BF0( 9, 22, COS0_9 , 1); |
|
178 |
+ /* pass 2 */ |
|
179 |
+ BF( 6, 9, COS1_6 , 2); |
|
180 |
+ BF(22, 25,-COS1_6 , 2); |
|
181 |
+ /* pass 3 */ |
|
182 |
+ BF( 1, 6, COS2_1 , 1); |
|
183 |
+ BF( 9, 14,-COS2_1 , 1); |
|
184 |
+ BF(17, 22, COS2_1 , 1); |
|
185 |
+ BF(25, 30,-COS2_1 , 1); |
|
186 |
+ |
|
187 |
+ /* pass 1 */ |
|
188 |
+ BF0( 2, 29, COS0_2 , 1); |
|
189 |
+ BF0(13, 18, COS0_13, 3); |
|
190 |
+ /* pass 2 */ |
|
191 |
+ BF( 2, 13, COS1_2 , 1); |
|
192 |
+ BF(18, 29,-COS1_2 , 1); |
|
193 |
+ /* pass 1 */ |
|
194 |
+ BF0( 5, 26, COS0_5 , 1); |
|
195 |
+ BF0(10, 21, COS0_10, 1); |
|
196 |
+ /* pass 2 */ |
|
197 |
+ BF( 5, 10, COS1_5 , 2); |
|
198 |
+ BF(21, 26,-COS1_5 , 2); |
|
199 |
+ /* pass 3 */ |
|
200 |
+ BF( 2, 5, COS2_2 , 1); |
|
201 |
+ BF(10, 13,-COS2_2 , 1); |
|
202 |
+ BF(18, 21, COS2_2 , 1); |
|
203 |
+ BF(26, 29,-COS2_2 , 1); |
|
204 |
+ /* pass 4 */ |
|
205 |
+ BF( 1, 2, COS3_1 , 2); |
|
206 |
+ BF( 5, 6,-COS3_1 , 2); |
|
207 |
+ BF( 9, 10, COS3_1 , 2); |
|
208 |
+ BF(13, 14,-COS3_1 , 2); |
|
209 |
+ BF(17, 18, COS3_1 , 2); |
|
210 |
+ BF(21, 22,-COS3_1 , 2); |
|
211 |
+ BF(25, 26, COS3_1 , 2); |
|
212 |
+ BF(29, 30,-COS3_1 , 2); |
|
213 |
+ |
|
214 |
+ /* pass 5 */ |
|
215 |
+ BF1( 0, 1, 2, 3); |
|
216 |
+ BF2( 4, 5, 6, 7); |
|
217 |
+ BF1( 8, 9, 10, 11); |
|
218 |
+ BF2(12, 13, 14, 15); |
|
219 |
+ BF1(16, 17, 18, 19); |
|
220 |
+ BF2(20, 21, 22, 23); |
|
221 |
+ BF1(24, 25, 26, 27); |
|
222 |
+ BF2(28, 29, 30, 31); |
|
223 |
+ |
|
224 |
+ /* pass 6 */ |
|
225 |
+ |
|
226 |
+ ADD( 8, 12); |
|
227 |
+ ADD(12, 10); |
|
228 |
+ ADD(10, 14); |
|
229 |
+ ADD(14, 9); |
|
230 |
+ ADD( 9, 13); |
|
231 |
+ ADD(13, 11); |
|
232 |
+ ADD(11, 15); |
|
233 |
+ |
|
234 |
+ out[ 0] = val0; |
|
235 |
+ out[16] = val1; |
|
236 |
+ out[ 8] = val2; |
|
237 |
+ out[24] = val3; |
|
238 |
+ out[ 4] = val4; |
|
239 |
+ out[20] = val5; |
|
240 |
+ out[12] = val6; |
|
241 |
+ out[28] = val7; |
|
242 |
+ out[ 2] = val8; |
|
243 |
+ out[18] = val9; |
|
244 |
+ out[10] = val10; |
|
245 |
+ out[26] = val11; |
|
246 |
+ out[ 6] = val12; |
|
247 |
+ out[22] = val13; |
|
248 |
+ out[14] = val14; |
|
249 |
+ out[30] = val15; |
|
250 |
+ |
|
251 |
+ ADD(24, 28); |
|
252 |
+ ADD(28, 26); |
|
253 |
+ ADD(26, 30); |
|
254 |
+ ADD(30, 25); |
|
255 |
+ ADD(25, 29); |
|
256 |
+ ADD(29, 27); |
|
257 |
+ ADD(27, 31); |
|
258 |
+ |
|
259 |
+ out[ 1] = val16 + val24; |
|
260 |
+ out[17] = val17 + val25; |
|
261 |
+ out[ 9] = val18 + val26; |
|
262 |
+ out[25] = val19 + val27; |
|
263 |
+ out[ 5] = val20 + val28; |
|
264 |
+ out[21] = val21 + val29; |
|
265 |
+ out[13] = val22 + val30; |
|
266 |
+ out[29] = val23 + val31; |
|
267 |
+ out[ 3] = val24 + val20; |
|
268 |
+ out[19] = val25 + val21; |
|
269 |
+ out[11] = val26 + val22; |
|
270 |
+ out[27] = val27 + val23; |
|
271 |
+ out[ 7] = val28 + val18; |
|
272 |
+ out[23] = val29 + val19; |
|
273 |
+ out[15] = val30 + val17; |
|
274 |
+ out[31] = val31; |
|
275 |
+} |
0 | 276 |
deleted file mode 100644 |
... | ... |
@@ -1,352 +0,0 @@ |
1 |
-/* |
|
2 |
- * FFT/IFFT transforms |
|
3 |
- * Copyright (c) 2008 Loren Merritt |
|
4 |
- * Copyright (c) 2002 Fabrice Bellard |
|
5 |
- * Partly based on libdjbfft by D. J. Bernstein |
|
6 |
- * |
|
7 |
- * This file is part of Libav. |
|
8 |
- * |
|
9 |
- * Libav is free software; you can redistribute it and/or |
|
10 |
- * modify it under the terms of the GNU Lesser General Public |
|
11 |
- * License as published by the Free Software Foundation; either |
|
12 |
- * version 2.1 of the License, or (at your option) any later version. |
|
13 |
- * |
|
14 |
- * Libav is distributed in the hope that it will be useful, |
|
15 |
- * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
16 |
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
17 |
- * Lesser General Public License for more details. |
|
18 |
- * |
|
19 |
- * You should have received a copy of the GNU Lesser General Public |
|
20 |
- * License along with Libav; if not, write to the Free Software |
|
21 |
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
22 |
- */ |
|
23 |
- |
|
24 |
-/** |
|
25 |
- * @file |
|
26 |
- * FFT/IFFT transforms. |
|
27 |
- */ |
|
28 |
- |
|
29 |
-#include <stdlib.h> |
|
30 |
-#include <string.h> |
|
31 |
-#include "libavutil/mathematics.h" |
|
32 |
-#include "fft.h" |
|
33 |
-#include "fft-internal.h" |
|
34 |
- |
|
35 |
-/* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */ |
|
36 |
-#if !CONFIG_HARDCODED_TABLES |
|
37 |
-COSTABLE(16); |
|
38 |
-COSTABLE(32); |
|
39 |
-COSTABLE(64); |
|
40 |
-COSTABLE(128); |
|
41 |
-COSTABLE(256); |
|
42 |
-COSTABLE(512); |
|
43 |
-COSTABLE(1024); |
|
44 |
-COSTABLE(2048); |
|
45 |
-COSTABLE(4096); |
|
46 |
-COSTABLE(8192); |
|
47 |
-COSTABLE(16384); |
|
48 |
-COSTABLE(32768); |
|
49 |
-COSTABLE(65536); |
|
50 |
-#endif |
|
51 |
-COSTABLE_CONST FFTSample * const FFT_NAME(ff_cos_tabs)[] = { |
|
52 |
- NULL, NULL, NULL, NULL, |
|
53 |
- FFT_NAME(ff_cos_16), |
|
54 |
- FFT_NAME(ff_cos_32), |
|
55 |
- FFT_NAME(ff_cos_64), |
|
56 |
- FFT_NAME(ff_cos_128), |
|
57 |
- FFT_NAME(ff_cos_256), |
|
58 |
- FFT_NAME(ff_cos_512), |
|
59 |
- FFT_NAME(ff_cos_1024), |
|
60 |
- FFT_NAME(ff_cos_2048), |
|
61 |
- FFT_NAME(ff_cos_4096), |
|
62 |
- FFT_NAME(ff_cos_8192), |
|
63 |
- FFT_NAME(ff_cos_16384), |
|
64 |
- FFT_NAME(ff_cos_32768), |
|
65 |
- FFT_NAME(ff_cos_65536), |
|
66 |
-}; |
|
67 |
- |
|
68 |
-static void fft_permute_c(FFTContext *s, FFTComplex *z); |
|
69 |
-static void fft_calc_c(FFTContext *s, FFTComplex *z); |
|
70 |
- |
|
71 |
-static int split_radix_permutation(int i, int n, int inverse) |
|
72 |
-{ |
|
73 |
- int m; |
|
74 |
- if(n <= 2) return i&1; |
|
75 |
- m = n >> 1; |
|
76 |
- if(!(i&m)) return split_radix_permutation(i, m, inverse)*2; |
|
77 |
- m >>= 1; |
|
78 |
- if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1; |
|
79 |
- else return split_radix_permutation(i, m, inverse)*4 - 1; |
|
80 |
-} |
|
81 |
- |
|
82 |
-av_cold void ff_init_ff_cos_tabs(int index) |
|
83 |
-{ |
|
84 |
-#if !CONFIG_HARDCODED_TABLES |
|
85 |
- int i; |
|
86 |
- int m = 1<<index; |
|
87 |
- double freq = 2*M_PI/m; |
|
88 |
- FFTSample *tab = FFT_NAME(ff_cos_tabs)[index]; |
|
89 |
- for(i=0; i<=m/4; i++) |
|
90 |
- tab[i] = FIX15(cos(i*freq)); |
|
91 |
- for(i=1; i<m/4; i++) |
|
92 |
- tab[m/2-i] = tab[i]; |
|
93 |
-#endif |
|
94 |
-} |
|
95 |
- |
|
96 |
-static const int avx_tab[] = { |
|
97 |
- 0, 4, 1, 5, 8, 12, 9, 13, 2, 6, 3, 7, 10, 14, 11, 15 |
|
98 |
-}; |
|
99 |
- |
|
100 |
-static int is_second_half_of_fft32(int i, int n) |
|
101 |
-{ |
|
102 |
- if (n <= 32) |
|
103 |
- return i >= 16; |
|
104 |
- else if (i < n/2) |
|
105 |
- return is_second_half_of_fft32(i, n/2); |
|
106 |
- else if (i < 3*n/4) |
|
107 |
- return is_second_half_of_fft32(i - n/2, n/4); |
|
108 |
- else |
|
109 |
- return is_second_half_of_fft32(i - 3*n/4, n/4); |
|
110 |
-} |
|
111 |
- |
|
112 |
-static av_cold void fft_perm_avx(FFTContext *s) |
|
113 |
-{ |
|
114 |
- int i; |
|
115 |
- int n = 1 << s->nbits; |
|
116 |
- |
|
117 |
- for (i = 0; i < n; i += 16) { |
|
118 |
- int k; |
|
119 |
- if (is_second_half_of_fft32(i, n)) { |
|
120 |
- for (k = 0; k < 16; k++) |
|
121 |
- s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = |
|
122 |
- i + avx_tab[k]; |
|
123 |
- |
|
124 |
- } else { |
|
125 |
- for (k = 0; k < 16; k++) { |
|
126 |
- int j = i + k; |
|
127 |
- j = (j & ~7) | ((j >> 1) & 3) | ((j << 2) & 4); |
|
128 |
- s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = j; |
|
129 |
- } |
|
130 |
- } |
|
131 |
- } |
|
132 |
-} |
|
133 |
- |
|
134 |
-av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse) |
|
135 |
-{ |
|
136 |
- int i, j, n; |
|
137 |
- |
|
138 |
- if (nbits < 2 || nbits > 16) |
|
139 |
- goto fail; |
|
140 |
- s->nbits = nbits; |
|
141 |
- n = 1 << nbits; |
|
142 |
- |
|
143 |
- s->revtab = av_malloc(n * sizeof(uint16_t)); |
|
144 |
- if (!s->revtab) |
|
145 |
- goto fail; |
|
146 |
- s->tmp_buf = av_malloc(n * sizeof(FFTComplex)); |
|
147 |
- if (!s->tmp_buf) |
|
148 |
- goto fail; |
|
149 |
- s->inverse = inverse; |
|
150 |
- s->fft_permutation = FF_FFT_PERM_DEFAULT; |
|
151 |
- |
|
152 |
- s->fft_permute = fft_permute_c; |
|
153 |
- s->fft_calc = fft_calc_c; |
|
154 |
-#if CONFIG_MDCT |
|
155 |
- s->imdct_calc = ff_imdct_calc_c; |
|
156 |
- s->imdct_half = ff_imdct_half_c; |
|
157 |
- s->mdct_calc = ff_mdct_calc_c; |
|
158 |
-#endif |
|
159 |
- |
|
160 |
-#if CONFIG_FFT_FLOAT |
|
161 |
- if (ARCH_ARM) ff_fft_init_arm(s); |
|
162 |
- if (ARCH_PPC) ff_fft_init_ppc(s); |
|
163 |
- if (ARCH_X86) ff_fft_init_x86(s); |
|
164 |
- if (CONFIG_MDCT) s->mdct_calcw = s->mdct_calc; |
|
165 |
-#else |
|
166 |
- if (CONFIG_MDCT) s->mdct_calcw = ff_mdct_calcw_c; |
|
167 |
- if (ARCH_ARM) ff_fft_fixed_init_arm(s); |
|
168 |
-#endif |
|
169 |
- |
|
170 |
- for(j=4; j<=nbits; j++) { |
|
171 |
- ff_init_ff_cos_tabs(j); |
|
172 |
- } |
|
173 |
- |
|
174 |
- if (s->fft_permutation == FF_FFT_PERM_AVX) { |
|
175 |
- fft_perm_avx(s); |
|
176 |
- } else { |
|
177 |
- for(i=0; i<n; i++) { |
|
178 |
- int j = i; |
|
179 |
- if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS) |
|
180 |
- j = (j&~3) | ((j>>1)&1) | ((j<<1)&2); |
|
181 |
- s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = j; |
|
182 |
- } |
|
183 |
- } |
|
184 |
- |
|
185 |
- return 0; |
|
186 |
- fail: |
|
187 |
- av_freep(&s->revtab); |
|
188 |
- av_freep(&s->tmp_buf); |
|
189 |
- return -1; |
|
190 |
-} |
|
191 |
- |
|
192 |
-static void fft_permute_c(FFTContext *s, FFTComplex *z) |
|
193 |
-{ |
|
194 |
- int j, np; |
|
195 |
- const uint16_t *revtab = s->revtab; |
|
196 |
- np = 1 << s->nbits; |
|
197 |
- /* TODO: handle split-radix permute in a more optimal way, probably in-place */ |
|
198 |
- for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j]; |
|
199 |
- memcpy(z, s->tmp_buf, np * sizeof(FFTComplex)); |
|
200 |
-} |
|
201 |
- |
|
202 |
-av_cold void ff_fft_end(FFTContext *s) |
|
203 |
-{ |
|
204 |
- av_freep(&s->revtab); |
|
205 |
- av_freep(&s->tmp_buf); |
|
206 |
-} |
|
207 |
- |
|
208 |
-#define BUTTERFLIES(a0,a1,a2,a3) {\ |
|
209 |
- BF(t3, t5, t5, t1);\ |
|
210 |
- BF(a2.re, a0.re, a0.re, t5);\ |
|
211 |
- BF(a3.im, a1.im, a1.im, t3);\ |
|
212 |
- BF(t4, t6, t2, t6);\ |
|
213 |
- BF(a3.re, a1.re, a1.re, t4);\ |
|
214 |
- BF(a2.im, a0.im, a0.im, t6);\ |
|
215 |
-} |
|
216 |
- |
|
217 |
-// force loading all the inputs before storing any. |
|
218 |
-// this is slightly slower for small data, but avoids store->load aliasing |
|
219 |
-// for addresses separated by large powers of 2. |
|
220 |
-#define BUTTERFLIES_BIG(a0,a1,a2,a3) {\ |
|
221 |
- FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\ |
|
222 |
- BF(t3, t5, t5, t1);\ |
|
223 |
- BF(a2.re, a0.re, r0, t5);\ |
|
224 |
- BF(a3.im, a1.im, i1, t3);\ |
|
225 |
- BF(t4, t6, t2, t6);\ |
|
226 |
- BF(a3.re, a1.re, r1, t4);\ |
|
227 |
- BF(a2.im, a0.im, i0, t6);\ |
|
228 |
-} |
|
229 |
- |
|
230 |
-#define TRANSFORM(a0,a1,a2,a3,wre,wim) {\ |
|
231 |
- CMUL(t1, t2, a2.re, a2.im, wre, -wim);\ |
|
232 |
- CMUL(t5, t6, a3.re, a3.im, wre, wim);\ |
|
233 |
- BUTTERFLIES(a0,a1,a2,a3)\ |
|
234 |
-} |
|
235 |
- |
|
236 |
-#define TRANSFORM_ZERO(a0,a1,a2,a3) {\ |
|
237 |
- t1 = a2.re;\ |
|
238 |
- t2 = a2.im;\ |
|
239 |
- t5 = a3.re;\ |
|
240 |
- t6 = a3.im;\ |
|
241 |
- BUTTERFLIES(a0,a1,a2,a3)\ |
|
242 |
-} |
|
243 |
- |
|
244 |
-/* z[0...8n-1], w[1...2n-1] */ |
|
245 |
-#define PASS(name)\ |
|
246 |
-static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\ |
|
247 |
-{\ |
|
248 |
- FFTDouble t1, t2, t3, t4, t5, t6;\ |
|
249 |
- int o1 = 2*n;\ |
|
250 |
- int o2 = 4*n;\ |
|
251 |
- int o3 = 6*n;\ |
|
252 |
- const FFTSample *wim = wre+o1;\ |
|
253 |
- n--;\ |
|
254 |
-\ |
|
255 |
- TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\ |
|
256 |
- TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\ |
|
257 |
- do {\ |
|
258 |
- z += 2;\ |
|
259 |
- wre += 2;\ |
|
260 |
- wim -= 2;\ |
|
261 |
- TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\ |
|
262 |
- TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\ |
|
263 |
- } while(--n);\ |
|
264 |
-} |
|
265 |
- |
|
266 |
-PASS(pass) |
|
267 |
-#undef BUTTERFLIES |
|
268 |
-#define BUTTERFLIES BUTTERFLIES_BIG |
|
269 |
-PASS(pass_big) |
|
270 |
- |
|
271 |
-#define DECL_FFT(n,n2,n4)\ |
|
272 |
-static void fft##n(FFTComplex *z)\ |
|
273 |
-{\ |
|
274 |
- fft##n2(z);\ |
|
275 |
- fft##n4(z+n4*2);\ |
|
276 |
- fft##n4(z+n4*3);\ |
|
277 |
- pass(z,FFT_NAME(ff_cos_##n),n4/2);\ |
|
278 |
-} |
|
279 |
- |
|
280 |
-static void fft4(FFTComplex *z) |
|
281 |
-{ |
|
282 |
- FFTDouble t1, t2, t3, t4, t5, t6, t7, t8; |
|
283 |
- |
|
284 |
- BF(t3, t1, z[0].re, z[1].re); |
|
285 |
- BF(t8, t6, z[3].re, z[2].re); |
|
286 |
- BF(z[2].re, z[0].re, t1, t6); |
|
287 |
- BF(t4, t2, z[0].im, z[1].im); |
|
288 |
- BF(t7, t5, z[2].im, z[3].im); |
|
289 |
- BF(z[3].im, z[1].im, t4, t8); |
|
290 |
- BF(z[3].re, z[1].re, t3, t7); |
|
291 |
- BF(z[2].im, z[0].im, t2, t5); |
|
292 |
-} |
|
293 |
- |
|
294 |
-static void fft8(FFTComplex *z) |
|
295 |
-{ |
|
296 |
- FFTDouble t1, t2, t3, t4, t5, t6; |
|
297 |
- |
|
298 |
- fft4(z); |
|
299 |
- |
|
300 |
- BF(t1, z[5].re, z[4].re, -z[5].re); |
|
301 |
- BF(t2, z[5].im, z[4].im, -z[5].im); |
|
302 |
- BF(t5, z[7].re, z[6].re, -z[7].re); |
|
303 |
- BF(t6, z[7].im, z[6].im, -z[7].im); |
|
304 |
- |
|
305 |
- BUTTERFLIES(z[0],z[2],z[4],z[6]); |
|
306 |
- TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf); |
|
307 |
-} |
|
308 |
- |
|
309 |
-#if !CONFIG_SMALL |
|
310 |
-static void fft16(FFTComplex *z) |
|
311 |
-{ |
|
312 |
- FFTDouble t1, t2, t3, t4, t5, t6; |
|
313 |
- FFTSample cos_16_1 = FFT_NAME(ff_cos_16)[1]; |
|
314 |
- FFTSample cos_16_3 = FFT_NAME(ff_cos_16)[3]; |
|
315 |
- |
|
316 |
- fft8(z); |
|
317 |
- fft4(z+8); |
|
318 |
- fft4(z+12); |
|
319 |
- |
|
320 |
- TRANSFORM_ZERO(z[0],z[4],z[8],z[12]); |
|
321 |
- TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf); |
|
322 |
- TRANSFORM(z[1],z[5],z[9],z[13],cos_16_1,cos_16_3); |
|
323 |
- TRANSFORM(z[3],z[7],z[11],z[15],cos_16_3,cos_16_1); |
|
324 |
-} |
|
325 |
-#else |
|
326 |
-DECL_FFT(16,8,4) |
|
327 |
-#endif |
|
328 |
-DECL_FFT(32,16,8) |
|
329 |
-DECL_FFT(64,32,16) |
|
330 |
-DECL_FFT(128,64,32) |
|
331 |
-DECL_FFT(256,128,64) |
|
332 |
-DECL_FFT(512,256,128) |
|
333 |
-#if !CONFIG_SMALL |
|
334 |
-#define pass pass_big |
|
335 |
-#endif |
|
336 |
-DECL_FFT(1024,512,256) |
|
337 |
-DECL_FFT(2048,1024,512) |
|
338 |
-DECL_FFT(4096,2048,1024) |
|
339 |
-DECL_FFT(8192,4096,2048) |
|
340 |
-DECL_FFT(16384,8192,4096) |
|
341 |
-DECL_FFT(32768,16384,8192) |
|
342 |
-DECL_FFT(65536,32768,16384) |
|
343 |
- |
|
344 |
-static void (* const fft_dispatch[])(FFTComplex*) = { |
|
345 |
- fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024, |
|
346 |
- fft2048, fft4096, fft8192, fft16384, fft32768, fft65536, |
|
347 |
-}; |
|
348 |
- |
|
349 |
-static void fft_calc_c(FFTContext *s, FFTComplex *z) |
|
350 |
-{ |
|
351 |
- fft_dispatch[s->nbits-2](z); |
|
352 |
-} |
21 | 21 |
new file mode 100644 |
... | ... |
@@ -0,0 +1,352 @@ |
0 |
+/* |
|
1 |
+ * FFT/IFFT transforms |
|
2 |
+ * Copyright (c) 2008 Loren Merritt |
|
3 |
+ * Copyright (c) 2002 Fabrice Bellard |
|
4 |
+ * Partly based on libdjbfft by D. J. Bernstein |
|
5 |
+ * |
|
6 |
+ * This file is part of Libav. |
|
7 |
+ * |
|
8 |
+ * Libav is free software; you can redistribute it and/or |
|
9 |
+ * modify it under the terms of the GNU Lesser General Public |
|
10 |
+ * License as published by the Free Software Foundation; either |
|
11 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
12 |
+ * |
|
13 |
+ * Libav is distributed in the hope that it will be useful, |
|
14 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
15 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
16 |
+ * Lesser General Public License for more details. |
|
17 |
+ * |
|
18 |
+ * You should have received a copy of the GNU Lesser General Public |
|
19 |
+ * License along with Libav; if not, write to the Free Software |
|
20 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
21 |
+ */ |
|
22 |
+ |
|
23 |
+/** |
|
24 |
+ * @file |
|
25 |
+ * FFT/IFFT transforms. |
|
26 |
+ */ |
|
27 |
+ |
|
28 |
+#include <stdlib.h> |
|
29 |
+#include <string.h> |
|
30 |
+#include "libavutil/mathematics.h" |
|
31 |
+#include "fft.h" |
|
32 |
+#include "fft-internal.h" |
|
33 |
+ |
|
34 |
+/* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */ |
|
35 |
+#if !CONFIG_HARDCODED_TABLES |
|
36 |
+COSTABLE(16); |
|
37 |
+COSTABLE(32); |
|
38 |
+COSTABLE(64); |
|
39 |
+COSTABLE(128); |
|
40 |
+COSTABLE(256); |
|
41 |
+COSTABLE(512); |
|
42 |
+COSTABLE(1024); |
|
43 |
+COSTABLE(2048); |
|
44 |
+COSTABLE(4096); |
|
45 |
+COSTABLE(8192); |
|
46 |
+COSTABLE(16384); |
|
47 |
+COSTABLE(32768); |
|
48 |
+COSTABLE(65536); |
|
49 |
+#endif |
|
50 |
+COSTABLE_CONST FFTSample * const FFT_NAME(ff_cos_tabs)[] = { |
|
51 |
+ NULL, NULL, NULL, NULL, |
|
52 |
+ FFT_NAME(ff_cos_16), |
|
53 |
+ FFT_NAME(ff_cos_32), |
|
54 |
+ FFT_NAME(ff_cos_64), |
|
55 |
+ FFT_NAME(ff_cos_128), |
|
56 |
+ FFT_NAME(ff_cos_256), |
|
57 |
+ FFT_NAME(ff_cos_512), |
|
58 |
+ FFT_NAME(ff_cos_1024), |
|
59 |
+ FFT_NAME(ff_cos_2048), |
|
60 |
+ FFT_NAME(ff_cos_4096), |
|
61 |
+ FFT_NAME(ff_cos_8192), |
|
62 |
+ FFT_NAME(ff_cos_16384), |
|
63 |
+ FFT_NAME(ff_cos_32768), |
|
64 |
+ FFT_NAME(ff_cos_65536), |
|
65 |
+}; |
|
66 |
+ |
|
67 |
+static void fft_permute_c(FFTContext *s, FFTComplex *z); |
|
68 |
+static void fft_calc_c(FFTContext *s, FFTComplex *z); |
|
69 |
+ |
|
70 |
+static int split_radix_permutation(int i, int n, int inverse) |
|
71 |
+{ |
|
72 |
+ int m; |
|
73 |
+ if(n <= 2) return i&1; |
|
74 |
+ m = n >> 1; |
|
75 |
+ if(!(i&m)) return split_radix_permutation(i, m, inverse)*2; |
|
76 |
+ m >>= 1; |
|
77 |
+ if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1; |
|
78 |
+ else return split_radix_permutation(i, m, inverse)*4 - 1; |
|
79 |
+} |
|
80 |
+ |
|
81 |
+av_cold void ff_init_ff_cos_tabs(int index) |
|
82 |
+{ |
|
83 |
+#if !CONFIG_HARDCODED_TABLES |
|
84 |
+ int i; |
|
85 |
+ int m = 1<<index; |
|
86 |
+ double freq = 2*M_PI/m; |
|
87 |
+ FFTSample *tab = FFT_NAME(ff_cos_tabs)[index]; |
|
88 |
+ for(i=0; i<=m/4; i++) |
|
89 |
+ tab[i] = FIX15(cos(i*freq)); |
|
90 |
+ for(i=1; i<m/4; i++) |
|
91 |
+ tab[m/2-i] = tab[i]; |
|
92 |
+#endif |
|
93 |
+} |
|
94 |
+ |
|
95 |
+static const int avx_tab[] = { |
|
96 |
+ 0, 4, 1, 5, 8, 12, 9, 13, 2, 6, 3, 7, 10, 14, 11, 15 |
|
97 |
+}; |
|
98 |
+ |
|
99 |
+static int is_second_half_of_fft32(int i, int n) |
|
100 |
+{ |
|
101 |
+ if (n <= 32) |
|
102 |
+ return i >= 16; |
|
103 |
+ else if (i < n/2) |
|
104 |
+ return is_second_half_of_fft32(i, n/2); |
|
105 |
+ else if (i < 3*n/4) |
|
106 |
+ return is_second_half_of_fft32(i - n/2, n/4); |
|
107 |
+ else |
|
108 |
+ return is_second_half_of_fft32(i - 3*n/4, n/4); |
|
109 |
+} |
|
110 |
+ |
|
111 |
+static av_cold void fft_perm_avx(FFTContext *s) |
|
112 |
+{ |
|
113 |
+ int i; |
|
114 |
+ int n = 1 << s->nbits; |
|
115 |
+ |
|
116 |
+ for (i = 0; i < n; i += 16) { |
|
117 |
+ int k; |
|
118 |
+ if (is_second_half_of_fft32(i, n)) { |
|
119 |
+ for (k = 0; k < 16; k++) |
|
120 |
+ s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = |
|
121 |
+ i + avx_tab[k]; |
|
122 |
+ |
|
123 |
+ } else { |
|
124 |
+ for (k = 0; k < 16; k++) { |
|
125 |
+ int j = i + k; |
|
126 |
+ j = (j & ~7) | ((j >> 1) & 3) | ((j << 2) & 4); |
|
127 |
+ s->revtab[-split_radix_permutation(i + k, n, s->inverse) & (n - 1)] = j; |
|
128 |
+ } |
|
129 |
+ } |
|
130 |
+ } |
|
131 |
+} |
|
132 |
+ |
|
133 |
+av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse) |
|
134 |
+{ |
|
135 |
+ int i, j, n; |
|
136 |
+ |
|
137 |
+ if (nbits < 2 || nbits > 16) |
|
138 |
+ goto fail; |
|
139 |
+ s->nbits = nbits; |
|
140 |
+ n = 1 << nbits; |
|
141 |
+ |
|
142 |
+ s->revtab = av_malloc(n * sizeof(uint16_t)); |
|
143 |
+ if (!s->revtab) |
|
144 |
+ goto fail; |
|
145 |
+ s->tmp_buf = av_malloc(n * sizeof(FFTComplex)); |
|
146 |
+ if (!s->tmp_buf) |
|
147 |
+ goto fail; |
|
148 |
+ s->inverse = inverse; |
|
149 |
+ s->fft_permutation = FF_FFT_PERM_DEFAULT; |
|
150 |
+ |
|
151 |
+ s->fft_permute = fft_permute_c; |
|
152 |
+ s->fft_calc = fft_calc_c; |
|
153 |
+#if CONFIG_MDCT |
|
154 |
+ s->imdct_calc = ff_imdct_calc_c; |
|
155 |
+ s->imdct_half = ff_imdct_half_c; |
|
156 |
+ s->mdct_calc = ff_mdct_calc_c; |
|
157 |
+#endif |
|
158 |
+ |
|
159 |
+#if CONFIG_FFT_FLOAT |
|
160 |
+ if (ARCH_ARM) ff_fft_init_arm(s); |
|
161 |
+ if (ARCH_PPC) ff_fft_init_ppc(s); |
|
162 |
+ if (ARCH_X86) ff_fft_init_x86(s); |
|
163 |
+ if (CONFIG_MDCT) s->mdct_calcw = s->mdct_calc; |
|
164 |
+#else |
|
165 |
+ if (CONFIG_MDCT) s->mdct_calcw = ff_mdct_calcw_c; |
|
166 |
+ if (ARCH_ARM) ff_fft_fixed_init_arm(s); |
|
167 |
+#endif |
|
168 |
+ |
|
169 |
+ for(j=4; j<=nbits; j++) { |
|
170 |
+ ff_init_ff_cos_tabs(j); |
|
171 |
+ } |
|
172 |
+ |
|
173 |
+ if (s->fft_permutation == FF_FFT_PERM_AVX) { |
|
174 |
+ fft_perm_avx(s); |
|
175 |
+ } else { |
|
176 |
+ for(i=0; i<n; i++) { |
|
177 |
+ int j = i; |
|
178 |
+ if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS) |
|
179 |
+ j = (j&~3) | ((j>>1)&1) | ((j<<1)&2); |
|
180 |
+ s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = j; |
|
181 |
+ } |
|
182 |
+ } |
|
183 |
+ |
|
184 |
+ return 0; |
|
185 |
+ fail: |
|
186 |
+ av_freep(&s->revtab); |
|
187 |
+ av_freep(&s->tmp_buf); |
|
188 |
+ return -1; |
|
189 |
+} |
|
190 |
+ |
|
191 |
+static void fft_permute_c(FFTContext *s, FFTComplex *z) |
|
192 |
+{ |
|
193 |
+ int j, np; |
|
194 |
+ const uint16_t *revtab = s->revtab; |
|
195 |
+ np = 1 << s->nbits; |
|
196 |
+ /* TODO: handle split-radix permute in a more optimal way, probably in-place */ |
|
197 |
+ for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j]; |
|
198 |
+ memcpy(z, s->tmp_buf, np * sizeof(FFTComplex)); |
|
199 |
+} |
|
200 |
+ |
|
201 |
+av_cold void ff_fft_end(FFTContext *s) |
|
202 |
+{ |
|
203 |
+ av_freep(&s->revtab); |
|
204 |
+ av_freep(&s->tmp_buf); |
|
205 |
+} |
|
206 |
+ |
|
207 |
+#define BUTTERFLIES(a0,a1,a2,a3) {\ |
|
208 |
+ BF(t3, t5, t5, t1);\ |
|
209 |
+ BF(a2.re, a0.re, a0.re, t5);\ |
|
210 |
+ BF(a3.im, a1.im, a1.im, t3);\ |
|
211 |
+ BF(t4, t6, t2, t6);\ |
|
212 |
+ BF(a3.re, a1.re, a1.re, t4);\ |
|
213 |
+ BF(a2.im, a0.im, a0.im, t6);\ |
|
214 |
+} |
|
215 |
+ |
|
216 |
+// force loading all the inputs before storing any. |
|
217 |
+// this is slightly slower for small data, but avoids store->load aliasing |
|
218 |
+// for addresses separated by large powers of 2. |
|
219 |
+#define BUTTERFLIES_BIG(a0,a1,a2,a3) {\ |
|
220 |
+ FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\ |
|
221 |
+ BF(t3, t5, t5, t1);\ |
|
222 |
+ BF(a2.re, a0.re, r0, t5);\ |
|
223 |
+ BF(a3.im, a1.im, i1, t3);\ |
|
224 |
+ BF(t4, t6, t2, t6);\ |
|
225 |
+ BF(a3.re, a1.re, r1, t4);\ |
|
226 |
+ BF(a2.im, a0.im, i0, t6);\ |
|
227 |
+} |
|
228 |
+ |
|
229 |
+#define TRANSFORM(a0,a1,a2,a3,wre,wim) {\ |
|
230 |
+ CMUL(t1, t2, a2.re, a2.im, wre, -wim);\ |
|
231 |
+ CMUL(t5, t6, a3.re, a3.im, wre, wim);\ |
|
232 |
+ BUTTERFLIES(a0,a1,a2,a3)\ |
|
233 |
+} |
|
234 |
+ |
|
235 |
+#define TRANSFORM_ZERO(a0,a1,a2,a3) {\ |
|
236 |
+ t1 = a2.re;\ |
|
237 |
+ t2 = a2.im;\ |
|
238 |
+ t5 = a3.re;\ |
|
239 |
+ t6 = a3.im;\ |
|
240 |
+ BUTTERFLIES(a0,a1,a2,a3)\ |
|
241 |
+} |
|
242 |
+ |
|
243 |
+/* z[0...8n-1], w[1...2n-1] */ |
|
244 |
+#define PASS(name)\ |
|
245 |
+static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\ |
|
246 |
+{\ |
|
247 |
+ FFTDouble t1, t2, t3, t4, t5, t6;\ |
|
248 |
+ int o1 = 2*n;\ |
|
249 |
+ int o2 = 4*n;\ |
|
250 |
+ int o3 = 6*n;\ |
|
251 |
+ const FFTSample *wim = wre+o1;\ |
|
252 |
+ n--;\ |
|
253 |
+\ |
|
254 |
+ TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\ |
|
255 |
+ TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\ |
|
256 |
+ do {\ |
|
257 |
+ z += 2;\ |
|
258 |
+ wre += 2;\ |
|
259 |
+ wim -= 2;\ |
|
260 |
+ TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\ |
|
261 |
+ TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\ |
|
262 |
+ } while(--n);\ |
|
263 |
+} |
|
264 |
+ |
|
265 |
+PASS(pass) |
|
266 |
+#undef BUTTERFLIES |
|
267 |
+#define BUTTERFLIES BUTTERFLIES_BIG |
|
268 |
+PASS(pass_big) |
|
269 |
+ |
|
270 |
+#define DECL_FFT(n,n2,n4)\ |
|
271 |
+static void fft##n(FFTComplex *z)\ |
|
272 |
+{\ |
|
273 |
+ fft##n2(z);\ |
|
274 |
+ fft##n4(z+n4*2);\ |
|
275 |
+ fft##n4(z+n4*3);\ |
|
276 |
+ pass(z,FFT_NAME(ff_cos_##n),n4/2);\ |
|
277 |
+} |
|
278 |
+ |
|
279 |
+static void fft4(FFTComplex *z) |
|
280 |
+{ |
|
281 |
+ FFTDouble t1, t2, t3, t4, t5, t6, t7, t8; |
|
282 |
+ |
|
283 |
+ BF(t3, t1, z[0].re, z[1].re); |
|
284 |
+ BF(t8, t6, z[3].re, z[2].re); |
|
285 |
+ BF(z[2].re, z[0].re, t1, t6); |
|
286 |
+ BF(t4, t2, z[0].im, z[1].im); |
|
287 |
+ BF(t7, t5, z[2].im, z[3].im); |
|
288 |
+ BF(z[3].im, z[1].im, t4, t8); |
|
289 |
+ BF(z[3].re, z[1].re, t3, t7); |
|
290 |
+ BF(z[2].im, z[0].im, t2, t5); |
|
291 |
+} |
|
292 |
+ |
|
293 |
+static void fft8(FFTComplex *z) |
|
294 |
+{ |
|
295 |
+ FFTDouble t1, t2, t3, t4, t5, t6; |
|
296 |
+ |
|
297 |
+ fft4(z); |
|
298 |
+ |
|
299 |
+ BF(t1, z[5].re, z[4].re, -z[5].re); |
|
300 |
+ BF(t2, z[5].im, z[4].im, -z[5].im); |
|
301 |
+ BF(t5, z[7].re, z[6].re, -z[7].re); |
|
302 |
+ BF(t6, z[7].im, z[6].im, -z[7].im); |
|
303 |
+ |
|
304 |
+ BUTTERFLIES(z[0],z[2],z[4],z[6]); |
|
305 |
+ TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf); |
|
306 |
+} |
|
307 |
+ |
|
308 |
+#if !CONFIG_SMALL |
|
309 |
+static void fft16(FFTComplex *z) |
|
310 |
+{ |
|
311 |
+ FFTDouble t1, t2, t3, t4, t5, t6; |
|
312 |
+ FFTSample cos_16_1 = FFT_NAME(ff_cos_16)[1]; |
|
313 |
+ FFTSample cos_16_3 = FFT_NAME(ff_cos_16)[3]; |
|
314 |
+ |
|
315 |
+ fft8(z); |
|
316 |
+ fft4(z+8); |
|
317 |
+ fft4(z+12); |
|
318 |
+ |
|
319 |
+ TRANSFORM_ZERO(z[0],z[4],z[8],z[12]); |
|
320 |
+ TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf); |
|
321 |
+ TRANSFORM(z[1],z[5],z[9],z[13],cos_16_1,cos_16_3); |
|
322 |
+ TRANSFORM(z[3],z[7],z[11],z[15],cos_16_3,cos_16_1); |
|
323 |
+} |
|
324 |
+#else |
|
325 |
+DECL_FFT(16,8,4) |
|
326 |
+#endif |
|
327 |
+DECL_FFT(32,16,8) |
|
328 |
+DECL_FFT(64,32,16) |
|
329 |
+DECL_FFT(128,64,32) |
|
330 |
+DECL_FFT(256,128,64) |
|
331 |
+DECL_FFT(512,256,128) |
|
332 |
+#if !CONFIG_SMALL |
|
333 |
+#define pass pass_big |
|
334 |
+#endif |
|
335 |
+DECL_FFT(1024,512,256) |
|
336 |
+DECL_FFT(2048,1024,512) |
|
337 |
+DECL_FFT(4096,2048,1024) |
|
338 |
+DECL_FFT(8192,4096,2048) |
|
339 |
+DECL_FFT(16384,8192,4096) |
|
340 |
+DECL_FFT(32768,16384,8192) |
|
341 |
+DECL_FFT(65536,32768,16384) |
|
342 |
+ |
|
343 |
+static void (* const fft_dispatch[])(FFTComplex*) = { |
|
344 |
+ fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024, |
|
345 |
+ fft2048, fft4096, fft8192, fft16384, fft32768, fft65536, |
|
346 |
+}; |
|
347 |
+ |
|
348 |
+static void fft_calc_c(FFTContext *s, FFTComplex *z) |
|
349 |
+{ |
|
350 |
+ fft_dispatch[s->nbits-2](z); |
|
351 |
+} |
0 | 352 |
deleted file mode 100644 |
... | ... |
@@ -1,203 +0,0 @@ |
1 |
-/* |
|
2 |
- * MDCT/IMDCT transforms |
|
3 |
- * Copyright (c) 2002 Fabrice Bellard |
|
4 |
- * |
|
5 |
- * This file is part of Libav. |
|
6 |
- * |
|
7 |
- * Libav is free software; you can redistribute it and/or |
|
8 |
- * modify it under the terms of the GNU Lesser General Public |
|
9 |
- * License as published by the Free Software Foundation; either |
|
10 |
- * version 2.1 of the License, or (at your option) any later version. |
|
11 |
- * |
|
12 |
- * Libav is distributed in the hope that it will be useful, |
|
13 |
- * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
14 |
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
15 |
- * Lesser General Public License for more details. |
|
16 |
- * |
|
17 |
- * You should have received a copy of the GNU Lesser General Public |
|
18 |
- * License along with Libav; if not, write to the Free Software |
|
19 |
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
20 |
- */ |
|
21 |
- |
|
22 |
-#include <stdlib.h> |
|
23 |
-#include <string.h> |
|
24 |
-#include "libavutil/common.h" |
|
25 |
-#include "libavutil/mathematics.h" |
|
26 |
-#include "fft.h" |
|
27 |
-#include "fft-internal.h" |
|
28 |
- |
|
29 |
-/** |
|
30 |
- * @file |
|
31 |
- * MDCT/IMDCT transforms. |
|
32 |
- */ |
|
33 |
- |
|
34 |
-#if CONFIG_FFT_FLOAT |
|
35 |
-# define RSCALE(x) (x) |
|
36 |
-#else |
|
37 |
-# define RSCALE(x) ((x) >> 1) |
|
38 |
-#endif |
|
39 |
- |
|
40 |
-/** |
|
41 |
- * init MDCT or IMDCT computation. |
|
42 |
- */ |
|
43 |
-av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale) |
|
44 |
-{ |
|
45 |
- int n, n4, i; |
|
46 |
- double alpha, theta; |
|
47 |
- int tstep; |
|
48 |
- |
|
49 |
- memset(s, 0, sizeof(*s)); |
|
50 |
- n = 1 << nbits; |
|
51 |
- s->mdct_bits = nbits; |
|
52 |
- s->mdct_size = n; |
|
53 |
- n4 = n >> 2; |
|
54 |
- s->mdct_permutation = FF_MDCT_PERM_NONE; |
|
55 |
- |
|
56 |
- if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0) |
|
57 |
- goto fail; |
|
58 |
- |
|
59 |
- s->tcos = av_malloc(n/2 * sizeof(FFTSample)); |
|
60 |
- if (!s->tcos) |
|
61 |
- goto fail; |
|
62 |
- |
|
63 |
- switch (s->mdct_permutation) { |
|
64 |
- case FF_MDCT_PERM_NONE: |
|
65 |
- s->tsin = s->tcos + n4; |
|
66 |
- tstep = 1; |
|
67 |
- break; |
|
68 |
- case FF_MDCT_PERM_INTERLEAVE: |
|
69 |
- s->tsin = s->tcos + 1; |
|
70 |
- tstep = 2; |
|
71 |
- break; |
|
72 |
- default: |
|
73 |
- goto fail; |
|
74 |
- } |
|
75 |
- |
|
76 |
- theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0); |
|
77 |
- scale = sqrt(fabs(scale)); |
|
78 |
- for(i=0;i<n4;i++) { |
|
79 |
- alpha = 2 * M_PI * (i + theta) / n; |
|
80 |
- s->tcos[i*tstep] = FIX15(-cos(alpha) * scale); |
|
81 |
- s->tsin[i*tstep] = FIX15(-sin(alpha) * scale); |
|
82 |
- } |
|
83 |
- return 0; |
|
84 |
- fail: |
|
85 |
- ff_mdct_end(s); |
|
86 |
- return -1; |
|
87 |
-} |
|
88 |
- |
|
89 |
-/** |
|
90 |
- * Compute the middle half of the inverse MDCT of size N = 2^nbits, |
|
91 |
- * thus excluding the parts that can be derived by symmetry |
|
92 |
- * @param output N/2 samples |
|
93 |
- * @param input N/2 samples |
|
94 |
- */ |
|
95 |
-void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
|
96 |
-{ |
|
97 |
- int k, n8, n4, n2, n, j; |
|
98 |
- const uint16_t *revtab = s->revtab; |
|
99 |
- const FFTSample *tcos = s->tcos; |
|
100 |
- const FFTSample *tsin = s->tsin; |
|
101 |
- const FFTSample *in1, *in2; |
|
102 |
- FFTComplex *z = (FFTComplex *)output; |
|
103 |
- |
|
104 |
- n = 1 << s->mdct_bits; |
|
105 |
- n2 = n >> 1; |
|
106 |
- n4 = n >> 2; |
|
107 |
- n8 = n >> 3; |
|
108 |
- |
|
109 |
- /* pre rotation */ |
|
110 |
- in1 = input; |
|
111 |
- in2 = input + n2 - 1; |
|
112 |
- for(k = 0; k < n4; k++) { |
|
113 |
- j=revtab[k]; |
|
114 |
- CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]); |
|
115 |
- in1 += 2; |
|
116 |
- in2 -= 2; |
|
117 |
- } |
|
118 |
- s->fft_calc(s, z); |
|
119 |
- |
|
120 |
- /* post rotation + reordering */ |
|
121 |
- for(k = 0; k < n8; k++) { |
|
122 |
- FFTSample r0, i0, r1, i1; |
|
123 |
- CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]); |
|
124 |
- CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]); |
|
125 |
- z[n8-k-1].re = r0; |
|
126 |
- z[n8-k-1].im = i0; |
|
127 |
- z[n8+k ].re = r1; |
|
128 |
- z[n8+k ].im = i1; |
|
129 |
- } |
|
130 |
-} |
|
131 |
- |
|
132 |
-/** |
|
133 |
- * Compute inverse MDCT of size N = 2^nbits |
|
134 |
- * @param output N samples |
|
135 |
- * @param input N/2 samples |
|
136 |
- */ |
|
137 |
-void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
|
138 |
-{ |
|
139 |
- int k; |
|
140 |
- int n = 1 << s->mdct_bits; |
|
141 |
- int n2 = n >> 1; |
|
142 |
- int n4 = n >> 2; |
|
143 |
- |
|
144 |
- ff_imdct_half_c(s, output+n4, input); |
|
145 |
- |
|
146 |
- for(k = 0; k < n4; k++) { |
|
147 |
- output[k] = -output[n2-k-1]; |
|
148 |
- output[n-k-1] = output[n2+k]; |
|
149 |
- } |
|
150 |
-} |
|
151 |
- |
|
152 |
-/** |
|
153 |
- * Compute MDCT of size N = 2^nbits |
|
154 |
- * @param input N samples |
|
155 |
- * @param out N/2 samples |
|
156 |
- */ |
|
157 |
-void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input) |
|
158 |
-{ |
|
159 |
- int i, j, n, n8, n4, n2, n3; |
|
160 |
- FFTDouble re, im; |
|
161 |
- const uint16_t *revtab = s->revtab; |
|
162 |
- const FFTSample *tcos = s->tcos; |
|
163 |
- const FFTSample *tsin = s->tsin; |
|
164 |
- FFTComplex *x = (FFTComplex *)out; |
|
165 |
- |
|
166 |
- n = 1 << s->mdct_bits; |
|
167 |
- n2 = n >> 1; |
|
168 |
- n4 = n >> 2; |
|
169 |
- n8 = n >> 3; |
|
170 |
- n3 = 3 * n4; |
|
171 |
- |
|
172 |
- /* pre rotation */ |
|
173 |
- for(i=0;i<n8;i++) { |
|
174 |
- re = RSCALE(-input[2*i+n3] - input[n3-1-2*i]); |
|
175 |
- im = RSCALE(-input[n4+2*i] + input[n4-1-2*i]); |
|
176 |
- j = revtab[i]; |
|
177 |
- CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]); |
|
178 |
- |
|
179 |
- re = RSCALE( input[2*i] - input[n2-1-2*i]); |
|
180 |
- im = RSCALE(-input[n2+2*i] - input[ n-1-2*i]); |
|
181 |
- j = revtab[n8 + i]; |
|
182 |
- CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]); |
|
183 |
- } |
|
184 |
- |
|
185 |
- s->fft_calc(s, x); |
|
186 |
- |
|
187 |
- /* post rotation */ |
|
188 |
- for(i=0;i<n8;i++) { |
|
189 |
- FFTSample r0, i0, r1, i1; |
|
190 |
- CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]); |
|
191 |
- CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]); |
|
192 |
- x[n8-i-1].re = r0; |
|
193 |
- x[n8-i-1].im = i0; |
|
194 |
- x[n8+i ].re = r1; |
|
195 |
- x[n8+i ].im = i1; |
|
196 |
- } |
|
197 |
-} |
|
198 |
- |
|
199 |
-av_cold void ff_mdct_end(FFTContext *s) |
|
200 |
-{ |
|
201 |
- av_freep(&s->tcos); |
|
202 |
- ff_fft_end(s); |
|
203 |
-} |
21 | 21 |
new file mode 100644 |
... | ... |
@@ -0,0 +1,203 @@ |
0 |
+/* |
|
1 |
+ * MDCT/IMDCT transforms |
|
2 |
+ * Copyright (c) 2002 Fabrice Bellard |
|
3 |
+ * |
|
4 |
+ * This file is part of Libav. |
|
5 |
+ * |
|
6 |
+ * Libav is free software; you can redistribute it and/or |
|
7 |
+ * modify it under the terms of the GNU Lesser General Public |
|
8 |
+ * License as published by the Free Software Foundation; either |
|
9 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
10 |
+ * |
|
11 |
+ * Libav is distributed in the hope that it will be useful, |
|
12 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
13 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
14 |
+ * Lesser General Public License for more details. |
|
15 |
+ * |
|
16 |
+ * You should have received a copy of the GNU Lesser General Public |
|
17 |
+ * License along with Libav; if not, write to the Free Software |
|
18 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
19 |
+ */ |
|
20 |
+ |
|
21 |
+#include <stdlib.h> |
|
22 |
+#include <string.h> |
|
23 |
+#include "libavutil/common.h" |
|
24 |
+#include "libavutil/mathematics.h" |
|
25 |
+#include "fft.h" |
|
26 |
+#include "fft-internal.h" |
|
27 |
+ |
|
28 |
+/** |
|
29 |
+ * @file |
|
30 |
+ * MDCT/IMDCT transforms. |
|
31 |
+ */ |
|
32 |
+ |
|
33 |
+#if CONFIG_FFT_FLOAT |
|
34 |
+# define RSCALE(x) (x) |
|
35 |
+#else |
|
36 |
+# define RSCALE(x) ((x) >> 1) |
|
37 |
+#endif |
|
38 |
+ |
|
39 |
+/** |
|
40 |
+ * init MDCT or IMDCT computation. |
|
41 |
+ */ |
|
42 |
+av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale) |
|
43 |
+{ |
|
44 |
+ int n, n4, i; |
|
45 |
+ double alpha, theta; |
|
46 |
+ int tstep; |
|
47 |
+ |
|
48 |
+ memset(s, 0, sizeof(*s)); |
|
49 |
+ n = 1 << nbits; |
|
50 |
+ s->mdct_bits = nbits; |
|
51 |
+ s->mdct_size = n; |
|
52 |
+ n4 = n >> 2; |
|
53 |
+ s->mdct_permutation = FF_MDCT_PERM_NONE; |
|
54 |
+ |
|
55 |
+ if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0) |
|
56 |
+ goto fail; |
|
57 |
+ |
|
58 |
+ s->tcos = av_malloc(n/2 * sizeof(FFTSample)); |
|
59 |
+ if (!s->tcos) |
|
60 |
+ goto fail; |
|
61 |
+ |
|
62 |
+ switch (s->mdct_permutation) { |
|
63 |
+ case FF_MDCT_PERM_NONE: |
|
64 |
+ s->tsin = s->tcos + n4; |
|
65 |
+ tstep = 1; |
|
66 |
+ break; |
|
67 |
+ case FF_MDCT_PERM_INTERLEAVE: |
|
68 |
+ s->tsin = s->tcos + 1; |
|
69 |
+ tstep = 2; |
|
70 |
+ break; |
|
71 |
+ default: |
|
72 |
+ goto fail; |
|
73 |
+ } |
|
74 |
+ |
|
75 |
+ theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0); |
|
76 |
+ scale = sqrt(fabs(scale)); |
|
77 |
+ for(i=0;i<n4;i++) { |
|
78 |
+ alpha = 2 * M_PI * (i + theta) / n; |
|
79 |
+ s->tcos[i*tstep] = FIX15(-cos(alpha) * scale); |
|
80 |
+ s->tsin[i*tstep] = FIX15(-sin(alpha) * scale); |
|
81 |
+ } |
|
82 |
+ return 0; |
|
83 |
+ fail: |
|
84 |
+ ff_mdct_end(s); |
|
85 |
+ return -1; |
|
86 |
+} |
|
87 |
+ |
|
88 |
+/** |
|
89 |
+ * Compute the middle half of the inverse MDCT of size N = 2^nbits, |
|
90 |
+ * thus excluding the parts that can be derived by symmetry |
|
91 |
+ * @param output N/2 samples |
|
92 |
+ * @param input N/2 samples |
|
93 |
+ */ |
|
94 |
+void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
|
95 |
+{ |
|
96 |
+ int k, n8, n4, n2, n, j; |
|
97 |
+ const uint16_t *revtab = s->revtab; |
|
98 |
+ const FFTSample *tcos = s->tcos; |
|
99 |
+ const FFTSample *tsin = s->tsin; |
|
100 |
+ const FFTSample *in1, *in2; |
|
101 |
+ FFTComplex *z = (FFTComplex *)output; |
|
102 |
+ |
|
103 |
+ n = 1 << s->mdct_bits; |
|
104 |
+ n2 = n >> 1; |
|
105 |
+ n4 = n >> 2; |
|
106 |
+ n8 = n >> 3; |
|
107 |
+ |
|
108 |
+ /* pre rotation */ |
|
109 |
+ in1 = input; |
|
110 |
+ in2 = input + n2 - 1; |
|
111 |
+ for(k = 0; k < n4; k++) { |
|
112 |
+ j=revtab[k]; |
|
113 |
+ CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]); |
|
114 |
+ in1 += 2; |
|
115 |
+ in2 -= 2; |
|
116 |
+ } |
|
117 |
+ s->fft_calc(s, z); |
|
118 |
+ |
|
119 |
+ /* post rotation + reordering */ |
|
120 |
+ for(k = 0; k < n8; k++) { |
|
121 |
+ FFTSample r0, i0, r1, i1; |
|
122 |
+ CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]); |
|
123 |
+ CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]); |
|
124 |
+ z[n8-k-1].re = r0; |
|
125 |
+ z[n8-k-1].im = i0; |
|
126 |
+ z[n8+k ].re = r1; |
|
127 |
+ z[n8+k ].im = i1; |
|
128 |
+ } |
|
129 |
+} |
|
130 |
+ |
|
131 |
+/** |
|
132 |
+ * Compute inverse MDCT of size N = 2^nbits |
|
133 |
+ * @param output N samples |
|
134 |
+ * @param input N/2 samples |
|
135 |
+ */ |
|
136 |
+void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
|
137 |
+{ |
|
138 |
+ int k; |
|
139 |
+ int n = 1 << s->mdct_bits; |
|
140 |
+ int n2 = n >> 1; |
|
141 |
+ int n4 = n >> 2; |
|
142 |
+ |
|
143 |
+ ff_imdct_half_c(s, output+n4, input); |
|
144 |
+ |
|
145 |
+ for(k = 0; k < n4; k++) { |
|
146 |
+ output[k] = -output[n2-k-1]; |
|
147 |
+ output[n-k-1] = output[n2+k]; |
|
148 |
+ } |
|
149 |
+} |
|
150 |
+ |
|
151 |
+/** |
|
152 |
+ * Compute MDCT of size N = 2^nbits |
|
153 |
+ * @param input N samples |
|
154 |
+ * @param out N/2 samples |
|
155 |
+ */ |
|
156 |
+void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input) |
|
157 |
+{ |
|
158 |
+ int i, j, n, n8, n4, n2, n3; |
|
159 |
+ FFTDouble re, im; |
|
160 |
+ const uint16_t *revtab = s->revtab; |
|
161 |
+ const FFTSample *tcos = s->tcos; |
|
162 |
+ const FFTSample *tsin = s->tsin; |
|
163 |
+ FFTComplex *x = (FFTComplex *)out; |
|
164 |
+ |
|
165 |
+ n = 1 << s->mdct_bits; |
|
166 |
+ n2 = n >> 1; |
|
167 |
+ n4 = n >> 2; |
|
168 |
+ n8 = n >> 3; |
|
169 |
+ n3 = 3 * n4; |
|
170 |
+ |
|
171 |
+ /* pre rotation */ |
|
172 |
+ for(i=0;i<n8;i++) { |
|
173 |
+ re = RSCALE(-input[2*i+n3] - input[n3-1-2*i]); |
|
174 |
+ im = RSCALE(-input[n4+2*i] + input[n4-1-2*i]); |
|
175 |
+ j = revtab[i]; |
|
176 |
+ CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]); |
|
177 |
+ |
|
178 |
+ re = RSCALE( input[2*i] - input[n2-1-2*i]); |
|
179 |
+ im = RSCALE(-input[n2+2*i] - input[ n-1-2*i]); |
|
180 |
+ j = revtab[n8 + i]; |
|
181 |
+ CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]); |
|
182 |
+ } |
|
183 |
+ |
|
184 |
+ s->fft_calc(s, x); |
|
185 |
+ |
|
186 |
+ /* post rotation */ |
|
187 |
+ for(i=0;i<n8;i++) { |
|
188 |
+ FFTSample r0, i0, r1, i1; |
|
189 |
+ CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]); |
|
190 |
+ CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]); |
|
191 |
+ x[n8-i-1].re = r0; |
|
192 |
+ x[n8-i-1].im = i0; |
|
193 |
+ x[n8+i ].re = r1; |
|
194 |
+ x[n8+i ].im = i1; |
|
195 |
+ } |
|
196 |
+} |
|
197 |
+ |
|
198 |
+av_cold void ff_mdct_end(FFTContext *s) |
|
199 |
+{ |
|
200 |
+ av_freep(&s->tcos); |
|
201 |
+ ff_fft_end(s); |
|
202 |
+} |