3
* Discrete Cosine Tansform (DCT) for subband synthesis
4
* optimized for machines with no auto-increment.
5
* The performance is highly compiler dependend. Maybe
6
* the dct64.c version for 'normal' processor may be faster
7
* even for Intel processors.
12
static void dct64_1(real *out0,real *out1,real *b1,real *b2,real *samples)
15
register real *costab = pnts[0];
17
b1[0x00] = samples[0x00] + samples[0x1F];
18
b1[0x01] = samples[0x01] + samples[0x1E];
19
b1[0x1F] = (samples[0x00] - samples[0x1F]) * costab[0x0];
20
b1[0x1E] = (samples[0x01] - samples[0x1E]) * costab[0x1];
22
b1[0x02] = samples[0x02] + samples[0x1D];
23
b1[0x03] = samples[0x03] + samples[0x1C];
24
b1[0x1D] = (samples[0x02] - samples[0x1D]) * costab[0x2];
25
b1[0x1C] = (samples[0x03] - samples[0x1C]) * costab[0x3];
27
b1[0x04] = samples[0x04] + samples[0x1B];
28
b1[0x05] = samples[0x05] + samples[0x1A];
29
b1[0x1B] = (samples[0x04] - samples[0x1B]) * costab[0x4];
30
b1[0x1A] = (samples[0x05] - samples[0x1A]) * costab[0x5];
32
b1[0x06] = samples[0x06] + samples[0x19];
33
b1[0x07] = samples[0x07] + samples[0x18];
34
b1[0x19] = (samples[0x06] - samples[0x19]) * costab[0x6];
35
b1[0x18] = (samples[0x07] - samples[0x18]) * costab[0x7];
37
b1[0x08] = samples[0x08] + samples[0x17];
38
b1[0x09] = samples[0x09] + samples[0x16];
39
b1[0x17] = (samples[0x08] - samples[0x17]) * costab[0x8];
40
b1[0x16] = (samples[0x09] - samples[0x16]) * costab[0x9];
42
b1[0x0A] = samples[0x0A] + samples[0x15];
43
b1[0x0B] = samples[0x0B] + samples[0x14];
44
b1[0x15] = (samples[0x0A] - samples[0x15]) * costab[0xA];
45
b1[0x14] = (samples[0x0B] - samples[0x14]) * costab[0xB];
47
b1[0x0C] = samples[0x0C] + samples[0x13];
48
b1[0x0D] = samples[0x0D] + samples[0x12];
49
b1[0x13] = (samples[0x0C] - samples[0x13]) * costab[0xC];
50
b1[0x12] = (samples[0x0D] - samples[0x12]) * costab[0xD];
52
b1[0x0E] = samples[0x0E] + samples[0x11];
53
b1[0x0F] = samples[0x0F] + samples[0x10];
54
b1[0x11] = (samples[0x0E] - samples[0x11]) * costab[0xE];
55
b1[0x10] = (samples[0x0F] - samples[0x10]) * costab[0xF];
61
register real *costab = pnts[1];
63
b2[0x00] = b1[0x00] + b1[0x0F];
64
b2[0x01] = b1[0x01] + b1[0x0E];
65
b2[0x0F] = (b1[0x00] - b1[0x0F]) * costab[0];
66
b2[0x0E] = (b1[0x01] - b1[0x0E]) * costab[1];
68
b2[0x02] = b1[0x02] + b1[0x0D];
69
b2[0x03] = b1[0x03] + b1[0x0C];
70
b2[0x0D] = (b1[0x02] - b1[0x0D]) * costab[2];
71
b2[0x0C] = (b1[0x03] - b1[0x0C]) * costab[3];
73
b2[0x04] = b1[0x04] + b1[0x0B];
74
b2[0x05] = b1[0x05] + b1[0x0A];
75
b2[0x0B] = (b1[0x04] - b1[0x0B]) * costab[4];
76
b2[0x0A] = (b1[0x05] - b1[0x0A]) * costab[5];
78
b2[0x06] = b1[0x06] + b1[0x09];
79
b2[0x07] = b1[0x07] + b1[0x08];
80
b2[0x09] = (b1[0x06] - b1[0x09]) * costab[6];
81
b2[0x08] = (b1[0x07] - b1[0x08]) * costab[7];
85
b2[0x10] = b1[0x10] + b1[0x1F];
86
b2[0x11] = b1[0x11] + b1[0x1E];
87
b2[0x1F] = (b1[0x1F] - b1[0x10]) * costab[0];
88
b2[0x1E] = (b1[0x1E] - b1[0x11]) * costab[1];
90
b2[0x12] = b1[0x12] + b1[0x1D];
91
b2[0x13] = b1[0x13] + b1[0x1C];
92
b2[0x1D] = (b1[0x1D] - b1[0x12]) * costab[2];
93
b2[0x1C] = (b1[0x1C] - b1[0x13]) * costab[3];
95
b2[0x14] = b1[0x14] + b1[0x1B];
96
b2[0x15] = b1[0x15] + b1[0x1A];
97
b2[0x1B] = (b1[0x1B] - b1[0x14]) * costab[4];
98
b2[0x1A] = (b1[0x1A] - b1[0x15]) * costab[5];
100
b2[0x16] = b1[0x16] + b1[0x19];
101
b2[0x17] = b1[0x17] + b1[0x18];
102
b2[0x19] = (b1[0x19] - b1[0x16]) * costab[6];
103
b2[0x18] = (b1[0x18] - b1[0x17]) * costab[7];
107
register real *costab = pnts[2];
109
b1[0x00] = b2[0x00] + b2[0x07];
110
b1[0x07] = (b2[0x00] - b2[0x07]) * costab[0];
111
b1[0x01] = b2[0x01] + b2[0x06];
112
b1[0x06] = (b2[0x01] - b2[0x06]) * costab[1];
113
b1[0x02] = b2[0x02] + b2[0x05];
114
b1[0x05] = (b2[0x02] - b2[0x05]) * costab[2];
115
b1[0x03] = b2[0x03] + b2[0x04];
116
b1[0x04] = (b2[0x03] - b2[0x04]) * costab[3];
118
b1[0x08] = b2[0x08] + b2[0x0F];
119
b1[0x0F] = (b2[0x0F] - b2[0x08]) * costab[0];
120
b1[0x09] = b2[0x09] + b2[0x0E];
121
b1[0x0E] = (b2[0x0E] - b2[0x09]) * costab[1];
122
b1[0x0A] = b2[0x0A] + b2[0x0D];
123
b1[0x0D] = (b2[0x0D] - b2[0x0A]) * costab[2];
124
b1[0x0B] = b2[0x0B] + b2[0x0C];
125
b1[0x0C] = (b2[0x0C] - b2[0x0B]) * costab[3];
127
b1[0x10] = b2[0x10] + b2[0x17];
128
b1[0x17] = (b2[0x10] - b2[0x17]) * costab[0];
129
b1[0x11] = b2[0x11] + b2[0x16];
130
b1[0x16] = (b2[0x11] - b2[0x16]) * costab[1];
131
b1[0x12] = b2[0x12] + b2[0x15];
132
b1[0x15] = (b2[0x12] - b2[0x15]) * costab[2];
133
b1[0x13] = b2[0x13] + b2[0x14];
134
b1[0x14] = (b2[0x13] - b2[0x14]) * costab[3];
136
b1[0x18] = b2[0x18] + b2[0x1F];
137
b1[0x1F] = (b2[0x1F] - b2[0x18]) * costab[0];
138
b1[0x19] = b2[0x19] + b2[0x1E];
139
b1[0x1E] = (b2[0x1E] - b2[0x19]) * costab[1];
140
b1[0x1A] = b2[0x1A] + b2[0x1D];
141
b1[0x1D] = (b2[0x1D] - b2[0x1A]) * costab[2];
142
b1[0x1B] = b2[0x1B] + b2[0x1C];
143
b1[0x1C] = (b2[0x1C] - b2[0x1B]) * costab[3];
147
register real const cos0 = pnts[3][0];
148
register real const cos1 = pnts[3][1];
150
b2[0x00] = b1[0x00] + b1[0x03];
151
b2[0x03] = (b1[0x00] - b1[0x03]) * cos0;
152
b2[0x01] = b1[0x01] + b1[0x02];
153
b2[0x02] = (b1[0x01] - b1[0x02]) * cos1;
155
b2[0x04] = b1[0x04] + b1[0x07];
156
b2[0x07] = (b1[0x07] - b1[0x04]) * cos0;
157
b2[0x05] = b1[0x05] + b1[0x06];
158
b2[0x06] = (b1[0x06] - b1[0x05]) * cos1;
160
b2[0x08] = b1[0x08] + b1[0x0B];
161
b2[0x0B] = (b1[0x08] - b1[0x0B]) * cos0;
162
b2[0x09] = b1[0x09] + b1[0x0A];
163
b2[0x0A] = (b1[0x09] - b1[0x0A]) * cos1;
165
b2[0x0C] = b1[0x0C] + b1[0x0F];
166
b2[0x0F] = (b1[0x0F] - b1[0x0C]) * cos0;
167
b2[0x0D] = b1[0x0D] + b1[0x0E];
168
b2[0x0E] = (b1[0x0E] - b1[0x0D]) * cos1;
170
b2[0x10] = b1[0x10] + b1[0x13];
171
b2[0x13] = (b1[0x10] - b1[0x13]) * cos0;
172
b2[0x11] = b1[0x11] + b1[0x12];
173
b2[0x12] = (b1[0x11] - b1[0x12]) * cos1;
175
b2[0x14] = b1[0x14] + b1[0x17];
176
b2[0x17] = (b1[0x17] - b1[0x14]) * cos0;
177
b2[0x15] = b1[0x15] + b1[0x16];
178
b2[0x16] = (b1[0x16] - b1[0x15]) * cos1;
180
b2[0x18] = b1[0x18] + b1[0x1B];
181
b2[0x1B] = (b1[0x18] - b1[0x1B]) * cos0;
182
b2[0x19] = b1[0x19] + b1[0x1A];
183
b2[0x1A] = (b1[0x19] - b1[0x1A]) * cos1;
185
b2[0x1C] = b1[0x1C] + b1[0x1F];
186
b2[0x1F] = (b1[0x1F] - b1[0x1C]) * cos0;
187
b2[0x1D] = b1[0x1D] + b1[0x1E];
188
b2[0x1E] = (b1[0x1E] - b1[0x1D]) * cos1;
192
register real const cos0 = pnts[4][0];
194
b1[0x00] = b2[0x00] + b2[0x01];
195
b1[0x01] = (b2[0x00] - b2[0x01]) * cos0;
196
b1[0x02] = b2[0x02] + b2[0x03];
197
b1[0x03] = (b2[0x03] - b2[0x02]) * cos0;
198
b1[0x02] += b1[0x03];
200
b1[0x04] = b2[0x04] + b2[0x05];
201
b1[0x05] = (b2[0x04] - b2[0x05]) * cos0;
202
b1[0x06] = b2[0x06] + b2[0x07];
203
b1[0x07] = (b2[0x07] - b2[0x06]) * cos0;
204
b1[0x06] += b1[0x07];
205
b1[0x04] += b1[0x06];
206
b1[0x06] += b1[0x05];
207
b1[0x05] += b1[0x07];
209
b1[0x08] = b2[0x08] + b2[0x09];
210
b1[0x09] = (b2[0x08] - b2[0x09]) * cos0;
211
b1[0x0A] = b2[0x0A] + b2[0x0B];
212
b1[0x0B] = (b2[0x0B] - b2[0x0A]) * cos0;
213
b1[0x0A] += b1[0x0B];
215
b1[0x0C] = b2[0x0C] + b2[0x0D];
216
b1[0x0D] = (b2[0x0C] - b2[0x0D]) * cos0;
217
b1[0x0E] = b2[0x0E] + b2[0x0F];
218
b1[0x0F] = (b2[0x0F] - b2[0x0E]) * cos0;
219
b1[0x0E] += b1[0x0F];
220
b1[0x0C] += b1[0x0E];
221
b1[0x0E] += b1[0x0D];
222
b1[0x0D] += b1[0x0F];
224
b1[0x10] = b2[0x10] + b2[0x11];
225
b1[0x11] = (b2[0x10] - b2[0x11]) * cos0;
226
b1[0x12] = b2[0x12] + b2[0x13];
227
b1[0x13] = (b2[0x13] - b2[0x12]) * cos0;
228
b1[0x12] += b1[0x13];
230
b1[0x14] = b2[0x14] + b2[0x15];
231
b1[0x15] = (b2[0x14] - b2[0x15]) * cos0;
232
b1[0x16] = b2[0x16] + b2[0x17];
233
b1[0x17] = (b2[0x17] - b2[0x16]) * cos0;
234
b1[0x16] += b1[0x17];
235
b1[0x14] += b1[0x16];
236
b1[0x16] += b1[0x15];
237
b1[0x15] += b1[0x17];
239
b1[0x18] = b2[0x18] + b2[0x19];
240
b1[0x19] = (b2[0x18] - b2[0x19]) * cos0;
241
b1[0x1A] = b2[0x1A] + b2[0x1B];
242
b1[0x1B] = (b2[0x1B] - b2[0x1A]) * cos0;
243
b1[0x1A] += b1[0x1B];
245
b1[0x1C] = b2[0x1C] + b2[0x1D];
246
b1[0x1D] = (b2[0x1C] - b2[0x1D]) * cos0;
247
b1[0x1E] = b2[0x1E] + b2[0x1F];
248
b1[0x1F] = (b2[0x1F] - b2[0x1E]) * cos0;
249
b1[0x1E] += b1[0x1F];
250
b1[0x1C] += b1[0x1E];
251
b1[0x1E] += b1[0x1D];
252
b1[0x1D] += b1[0x1F];
255
out0[0x10*16] = b1[0x00];
256
out0[0x10*12] = b1[0x04];
257
out0[0x10* 8] = b1[0x02];
258
out0[0x10* 4] = b1[0x06];
259
out0[0x10* 0] = b1[0x01];
260
out1[0x10* 0] = b1[0x01];
261
out1[0x10* 4] = b1[0x05];
262
out1[0x10* 8] = b1[0x03];
263
out1[0x10*12] = b1[0x07];
266
out0[0x10*14] = b1[0x08] + b1[0x0C];
267
out0[0x10*10] = b1[0x0C] + b1[0x0a];
268
out0[0x10* 6] = b1[0x0A] + b1[0x0E];
269
out0[0x10* 2] = b1[0x0E] + b1[0x09];
270
out1[0x10* 2] = b1[0x09] + b1[0x0D];
271
out1[0x10* 6] = b1[0x0D] + b1[0x0B];
272
out1[0x10*10] = b1[0x0B] + b1[0x0F];
273
out1[0x10*14] = b1[0x0F];
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];
294
tmp = b1[0x18] + b1[0x1C];
295
out0[0x10*15] = tmp + b1[0x10];
296
out0[0x10*13] = tmp + b1[0x14];
297
tmp = b1[0x1C] + b1[0x1A];
298
out0[0x10*11] = tmp + b1[0x14];
299
out0[0x10* 9] = tmp + b1[0x12];
300
tmp = b1[0x1A] + b1[0x1E];
301
out0[0x10* 7] = tmp + b1[0x12];
302
out0[0x10* 5] = tmp + b1[0x16];
303
tmp = b1[0x1E] + b1[0x19];
304
out0[0x10* 3] = tmp + b1[0x16];
305
out0[0x10* 1] = tmp + b1[0x11];
306
tmp = b1[0x19] + b1[0x1D];
307
out1[0x10* 1] = tmp + b1[0x11];
308
out1[0x10* 3] = tmp + b1[0x15];
309
tmp = b1[0x1D] + b1[0x1B];
310
out1[0x10* 5] = tmp + b1[0x15];
311
out1[0x10* 7] = tmp + b1[0x13];
312
tmp = b1[0x1B] + b1[0x1F];
313
out1[0x10* 9] = tmp + b1[0x13];
314
out1[0x10*11] = tmp + b1[0x17];
315
out1[0x10*13] = b1[0x17] + b1[0x1F];
316
out1[0x10*15] = b1[0x1F];
321
* the call via dct64 is a trick to force GCC to use
322
* (new) registers for the b1,b2 pointer to the bufs[xx] field
324
void dct64(real *a,real *b,real *c)
327
dct64_1(a,b,bufs,bufs+0x20,c);
added
removed
dct64_i386.c
