1 /*
2 * Discrete Cosine Tansform (DCT) for subband synthesis
3 * optimized for machines with no auto-increment.
4 * The performance is highly compiler dependend. Maybe
5 * the dct64.c version for 'normal' processor may be faster
6 * even for Intel processors.
7 */
8
9 /* $Id: dct64_i386.c,v 1.2 2003/08/12 00:02:55 snelg Exp $ */
10
11 #ifdef HAVE_CONFIG_H
12 #include <config.h>
13 #endif
14
15 #include "dct64_i386.h"
16 #include "tabinit.h"
17
18 #ifdef WITH_DMALLOC
19 #include <dmalloc.h>
20 #endif
21
dct64_1(real * out0,real * out1,real * b1,real * b2,real * samples)22 static void dct64_1(real *out0,real *out1,real *b1,real *b2,real *samples)
23 {
24
25 {
26 register real *costab = pnts[0];
27
28 b1[0x00] = samples[0x00] + samples[0x1F];
29 b1[0x1F] = (samples[0x00] - samples[0x1F]) * costab[0x0];
30
31 b1[0x01] = samples[0x01] + samples[0x1E];
32 b1[0x1E] = (samples[0x01] - samples[0x1E]) * costab[0x1];
33
34 b1[0x02] = samples[0x02] + samples[0x1D];
35 b1[0x1D] = (samples[0x02] - samples[0x1D]) * costab[0x2];
36
37 b1[0x03] = samples[0x03] + samples[0x1C];
38 b1[0x1C] = (samples[0x03] - samples[0x1C]) * costab[0x3];
39
40 b1[0x04] = samples[0x04] + samples[0x1B];
41 b1[0x1B] = (samples[0x04] - samples[0x1B]) * costab[0x4];
42
43 b1[0x05] = samples[0x05] + samples[0x1A];
44 b1[0x1A] = (samples[0x05] - samples[0x1A]) * costab[0x5];
45
46 b1[0x06] = samples[0x06] + samples[0x19];
47 b1[0x19] = (samples[0x06] - samples[0x19]) * costab[0x6];
48
49 b1[0x07] = samples[0x07] + samples[0x18];
50 b1[0x18] = (samples[0x07] - samples[0x18]) * costab[0x7];
51
52 b1[0x08] = samples[0x08] + samples[0x17];
53 b1[0x17] = (samples[0x08] - samples[0x17]) * costab[0x8];
54
55 b1[0x09] = samples[0x09] + samples[0x16];
56 b1[0x16] = (samples[0x09] - samples[0x16]) * costab[0x9];
57
58 b1[0x0A] = samples[0x0A] + samples[0x15];
59 b1[0x15] = (samples[0x0A] - samples[0x15]) * costab[0xA];
60
61 b1[0x0B] = samples[0x0B] + samples[0x14];
62 b1[0x14] = (samples[0x0B] - samples[0x14]) * costab[0xB];
63
64 b1[0x0C] = samples[0x0C] + samples[0x13];
65 b1[0x13] = (samples[0x0C] - samples[0x13]) * costab[0xC];
66
67 b1[0x0D] = samples[0x0D] + samples[0x12];
68 b1[0x12] = (samples[0x0D] - samples[0x12]) * costab[0xD];
69
70 b1[0x0E] = samples[0x0E] + samples[0x11];
71 b1[0x11] = (samples[0x0E] - samples[0x11]) * costab[0xE];
72
73 b1[0x0F] = samples[0x0F] + samples[0x10];
74 b1[0x10] = (samples[0x0F] - samples[0x10]) * costab[0xF];
75 }
76
77
78 {
79 register real *costab = pnts[1];
80
81 b2[0x00] = b1[0x00] + b1[0x0F];
82 b2[0x0F] = (b1[0x00] - b1[0x0F]) * costab[0];
83 b2[0x01] = b1[0x01] + b1[0x0E];
84 b2[0x0E] = (b1[0x01] - b1[0x0E]) * costab[1];
85 b2[0x02] = b1[0x02] + b1[0x0D];
86 b2[0x0D] = (b1[0x02] - b1[0x0D]) * costab[2];
87 b2[0x03] = b1[0x03] + b1[0x0C];
88 b2[0x0C] = (b1[0x03] - b1[0x0C]) * costab[3];
89 b2[0x04] = b1[0x04] + b1[0x0B];
90 b2[0x0B] = (b1[0x04] - b1[0x0B]) * costab[4];
91 b2[0x05] = b1[0x05] + b1[0x0A];
92 b2[0x0A] = (b1[0x05] - b1[0x0A]) * costab[5];
93 b2[0x06] = b1[0x06] + b1[0x09];
94 b2[0x09] = (b1[0x06] - b1[0x09]) * costab[6];
95 b2[0x07] = b1[0x07] + b1[0x08];
96 b2[0x08] = (b1[0x07] - b1[0x08]) * costab[7];
97
98 b2[0x10] = b1[0x10] + b1[0x1F];
99 b2[0x1F] = (b1[0x1F] - b1[0x10]) * costab[0];
100 b2[0x11] = b1[0x11] + b1[0x1E];
101 b2[0x1E] = (b1[0x1E] - b1[0x11]) * costab[1];
102 b2[0x12] = b1[0x12] + b1[0x1D];
103 b2[0x1D] = (b1[0x1D] - b1[0x12]) * costab[2];
104 b2[0x13] = b1[0x13] + b1[0x1C];
105 b2[0x1C] = (b1[0x1C] - b1[0x13]) * costab[3];
106 b2[0x14] = b1[0x14] + b1[0x1B];
107 b2[0x1B] = (b1[0x1B] - b1[0x14]) * costab[4];
108 b2[0x15] = b1[0x15] + b1[0x1A];
109 b2[0x1A] = (b1[0x1A] - b1[0x15]) * costab[5];
110 b2[0x16] = b1[0x16] + b1[0x19];
111 b2[0x19] = (b1[0x19] - b1[0x16]) * costab[6];
112 b2[0x17] = b1[0x17] + b1[0x18];
113 b2[0x18] = (b1[0x18] - b1[0x17]) * costab[7];
114 }
115
116 {
117 register real *costab = pnts[2];
118
119 b1[0x00] = b2[0x00] + b2[0x07];
120 b1[0x07] = (b2[0x00] - b2[0x07]) * costab[0];
121 b1[0x01] = b2[0x01] + b2[0x06];
122 b1[0x06] = (b2[0x01] - b2[0x06]) * costab[1];
123 b1[0x02] = b2[0x02] + b2[0x05];
124 b1[0x05] = (b2[0x02] - b2[0x05]) * costab[2];
125 b1[0x03] = b2[0x03] + b2[0x04];
126 b1[0x04] = (b2[0x03] - b2[0x04]) * costab[3];
127
128 b1[0x08] = b2[0x08] + b2[0x0F];
129 b1[0x0F] = (b2[0x0F] - b2[0x08]) * costab[0];
130 b1[0x09] = b2[0x09] + b2[0x0E];
131 b1[0x0E] = (b2[0x0E] - b2[0x09]) * costab[1];
132 b1[0x0A] = b2[0x0A] + b2[0x0D];
133 b1[0x0D] = (b2[0x0D] - b2[0x0A]) * costab[2];
134 b1[0x0B] = b2[0x0B] + b2[0x0C];
135 b1[0x0C] = (b2[0x0C] - b2[0x0B]) * costab[3];
136
137 b1[0x10] = b2[0x10] + b2[0x17];
138 b1[0x17] = (b2[0x10] - b2[0x17]) * costab[0];
139 b1[0x11] = b2[0x11] + b2[0x16];
140 b1[0x16] = (b2[0x11] - b2[0x16]) * costab[1];
141 b1[0x12] = b2[0x12] + b2[0x15];
142 b1[0x15] = (b2[0x12] - b2[0x15]) * costab[2];
143 b1[0x13] = b2[0x13] + b2[0x14];
144 b1[0x14] = (b2[0x13] - b2[0x14]) * costab[3];
145
146 b1[0x18] = b2[0x18] + b2[0x1F];
147 b1[0x1F] = (b2[0x1F] - b2[0x18]) * costab[0];
148 b1[0x19] = b2[0x19] + b2[0x1E];
149 b1[0x1E] = (b2[0x1E] - b2[0x19]) * costab[1];
150 b1[0x1A] = b2[0x1A] + b2[0x1D];
151 b1[0x1D] = (b2[0x1D] - b2[0x1A]) * costab[2];
152 b1[0x1B] = b2[0x1B] + b2[0x1C];
153 b1[0x1C] = (b2[0x1C] - b2[0x1B]) * costab[3];
154 }
155
156 {
157 register real const cos0 = pnts[3][0];
158 register real const cos1 = pnts[3][1];
159
160 b2[0x00] = b1[0x00] + b1[0x03];
161 b2[0x03] = (b1[0x00] - b1[0x03]) * cos0;
162 b2[0x01] = b1[0x01] + b1[0x02];
163 b2[0x02] = (b1[0x01] - b1[0x02]) * cos1;
164
165 b2[0x04] = b1[0x04] + b1[0x07];
166 b2[0x07] = (b1[0x07] - b1[0x04]) * cos0;
167 b2[0x05] = b1[0x05] + b1[0x06];
168 b2[0x06] = (b1[0x06] - b1[0x05]) * cos1;
169
170 b2[0x08] = b1[0x08] + b1[0x0B];
171 b2[0x0B] = (b1[0x08] - b1[0x0B]) * cos0;
172 b2[0x09] = b1[0x09] + b1[0x0A];
173 b2[0x0A] = (b1[0x09] - b1[0x0A]) * cos1;
174
175 b2[0x0C] = b1[0x0C] + b1[0x0F];
176 b2[0x0F] = (b1[0x0F] - b1[0x0C]) * cos0;
177 b2[0x0D] = b1[0x0D] + b1[0x0E];
178 b2[0x0E] = (b1[0x0E] - b1[0x0D]) * cos1;
179
180 b2[0x10] = b1[0x10] + b1[0x13];
181 b2[0x13] = (b1[0x10] - b1[0x13]) * cos0;
182 b2[0x11] = b1[0x11] + b1[0x12];
183 b2[0x12] = (b1[0x11] - b1[0x12]) * cos1;
184
185 b2[0x14] = b1[0x14] + b1[0x17];
186 b2[0x17] = (b1[0x17] - b1[0x14]) * cos0;
187 b2[0x15] = b1[0x15] + b1[0x16];
188 b2[0x16] = (b1[0x16] - b1[0x15]) * cos1;
189
190 b2[0x18] = b1[0x18] + b1[0x1B];
191 b2[0x1B] = (b1[0x18] - b1[0x1B]) * cos0;
192 b2[0x19] = b1[0x19] + b1[0x1A];
193 b2[0x1A] = (b1[0x19] - b1[0x1A]) * cos1;
194
195 b2[0x1C] = b1[0x1C] + b1[0x1F];
196 b2[0x1F] = (b1[0x1F] - b1[0x1C]) * cos0;
197 b2[0x1D] = b1[0x1D] + b1[0x1E];
198 b2[0x1E] = (b1[0x1E] - b1[0x1D]) * cos1;
199 }
200
201 {
202 register real const cos0 = pnts[4][0];
203
204 b1[0x00] = b2[0x00] + b2[0x01];
205 b1[0x01] = (b2[0x00] - b2[0x01]) * cos0;
206 b1[0x02] = b2[0x02] + b2[0x03];
207 b1[0x03] = (b2[0x03] - b2[0x02]) * cos0;
208 b1[0x02] += b1[0x03];
209
210 b1[0x04] = b2[0x04] + b2[0x05];
211 b1[0x05] = (b2[0x04] - b2[0x05]) * cos0;
212 b1[0x06] = b2[0x06] + b2[0x07];
213 b1[0x07] = (b2[0x07] - b2[0x06]) * cos0;
214 b1[0x06] += b1[0x07];
215 b1[0x04] += b1[0x06];
216 b1[0x06] += b1[0x05];
217 b1[0x05] += b1[0x07];
218
219 b1[0x08] = b2[0x08] + b2[0x09];
220 b1[0x09] = (b2[0x08] - b2[0x09]) * cos0;
221 b1[0x0A] = b2[0x0A] + b2[0x0B];
222 b1[0x0B] = (b2[0x0B] - b2[0x0A]) * cos0;
223 b1[0x0A] += b1[0x0B];
224
225 b1[0x0C] = b2[0x0C] + b2[0x0D];
226 b1[0x0D] = (b2[0x0C] - b2[0x0D]) * cos0;
227 b1[0x0E] = b2[0x0E] + b2[0x0F];
228 b1[0x0F] = (b2[0x0F] - b2[0x0E]) * cos0;
229 b1[0x0E] += b1[0x0F];
230 b1[0x0C] += b1[0x0E];
231 b1[0x0E] += b1[0x0D];
232 b1[0x0D] += b1[0x0F];
233
234 b1[0x10] = b2[0x10] + b2[0x11];
235 b1[0x11] = (b2[0x10] - b2[0x11]) * cos0;
236 b1[0x12] = b2[0x12] + b2[0x13];
237 b1[0x13] = (b2[0x13] - b2[0x12]) * cos0;
238 b1[0x12] += b1[0x13];
239
240 b1[0x14] = b2[0x14] + b2[0x15];
241 b1[0x15] = (b2[0x14] - b2[0x15]) * cos0;
242 b1[0x16] = b2[0x16] + b2[0x17];
243 b1[0x17] = (b2[0x17] - b2[0x16]) * cos0;
244 b1[0x16] += b1[0x17];
245 b1[0x14] += b1[0x16];
246 b1[0x16] += b1[0x15];
247 b1[0x15] += b1[0x17];
248
249 b1[0x18] = b2[0x18] + b2[0x19];
250 b1[0x19] = (b2[0x18] - b2[0x19]) * cos0;
251 b1[0x1A] = b2[0x1A] + b2[0x1B];
252 b1[0x1B] = (b2[0x1B] - b2[0x1A]) * cos0;
253 b1[0x1A] += b1[0x1B];
254
255 b1[0x1C] = b2[0x1C] + b2[0x1D];
256 b1[0x1D] = (b2[0x1C] - b2[0x1D]) * cos0;
257 b1[0x1E] = b2[0x1E] + b2[0x1F];
258 b1[0x1F] = (b2[0x1F] - b2[0x1E]) * cos0;
259 b1[0x1E] += b1[0x1F];
260 b1[0x1C] += b1[0x1E];
261 b1[0x1E] += b1[0x1D];
262 b1[0x1D] += b1[0x1F];
263 }
264
265 out0[0x10*16] = b1[0x00];
266 out0[0x10*12] = b1[0x04];
267 out0[0x10* 8] = b1[0x02];
268 out0[0x10* 4] = b1[0x06];
269 out0[0x10* 0] = b1[0x01];
270 out1[0x10* 0] = b1[0x01];
271 out1[0x10* 4] = b1[0x05];
272 out1[0x10* 8] = b1[0x03];
273 out1[0x10*12] = b1[0x07];
274
275 b1[0x08] += b1[0x0C];
276 out0[0x10*14] = b1[0x08];
277 b1[0x0C] += b1[0x0a];
278 out0[0x10*10] = b1[0x0C];
279 b1[0x0A] += b1[0x0E];
280 out0[0x10* 6] = b1[0x0A];
281 b1[0x0E] += b1[0x09];
282 out0[0x10* 2] = b1[0x0E];
283 b1[0x09] += b1[0x0D];
284 out1[0x10* 2] = b1[0x09];
285 b1[0x0D] += b1[0x0B];
286 out1[0x10* 6] = b1[0x0D];
287 b1[0x0B] += b1[0x0F];
288 out1[0x10*10] = b1[0x0B];
289 out1[0x10*14] = b1[0x0F];
290
291 b1[0x18] += b1[0x1C];
292 out0[0x10*15] = b1[0x10] + b1[0x18];
293 out0[0x10*13] = b1[0x18] + b1[0x14];
294 b1[0x1C] += b1[0x1a];
295 out0[0x10*11] = b1[0x14] + b1[0x1C];
296 out0[0x10* 9] = b1[0x1C] + b1[0x12];
297 b1[0x1A] += b1[0x1E];
298 out0[0x10* 7] = b1[0x12] + b1[0x1A];
299 out0[0x10* 5] = b1[0x1A] + b1[0x16];
300 b1[0x1E] += b1[0x19];
301 out0[0x10* 3] = b1[0x16] + b1[0x1E];
302 out0[0x10* 1] = b1[0x1E] + b1[0x11];
303 b1[0x19] += b1[0x1D];
304 out1[0x10* 1] = b1[0x11] + b1[0x19];
305 out1[0x10* 3] = b1[0x19] + b1[0x15];
306 b1[0x1D] += b1[0x1B];
307 out1[0x10* 5] = b1[0x15] + b1[0x1D];
308 out1[0x10* 7] = b1[0x1D] + b1[0x13];
309 b1[0x1B] += b1[0x1F];
310 out1[0x10* 9] = b1[0x13] + b1[0x1B];
311 out1[0x10*11] = b1[0x1B] + b1[0x17];
312 out1[0x10*13] = b1[0x17] + b1[0x1F];
313 out1[0x10*15] = b1[0x1F];
314 }
315
316 /*
317 * the call via dct64 is a trick to force GCC to use
318 * (new) registers for the b1,b2 pointer to the bufs[xx] field
319 */
dct64(real * a,real * b,real * c)320 void dct64( real *a,real *b,real *c)
321 {
322 real bufs[0x40];
323 dct64_1(a,b,bufs,bufs+0x20,c);
324 }
325
326