3394
|
1 /*
|
|
2 * imdct.c
|
|
3 * Copyright (C) 2000-2001 Michel Lespinasse <walken@zoy.org>
|
|
4 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
|
|
5 *
|
|
6 * This file is part of a52dec, a free ATSC A-52 stream decoder.
|
|
7 * See http://liba52.sourceforge.net/ for updates.
|
|
8 *
|
|
9 * a52dec is free software; you can redistribute it and/or modify
|
|
10 * it under the terms of the GNU General Public License as published by
|
|
11 * the Free Software Foundation; either version 2 of the License, or
|
|
12 * (at your option) any later version.
|
|
13 *
|
|
14 * a52dec 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
|
|
17 * GNU General Public License for more details.
|
|
18 *
|
|
19 * You should have received a copy of the GNU General Public License
|
|
20 * along with this program; if not, write to the Free Software
|
|
21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
3579
|
22 *
|
|
23 * SSE optimizations from Michael Niedermayer (michaelni@gmx.at)
|
3394
|
24 */
|
|
25
|
|
26 #include "config.h"
|
3579
|
27 #include "../cpudetect.h"
|
3394
|
28
|
|
29 #include <math.h>
|
|
30 #include <stdio.h>
|
|
31 #ifndef M_PI
|
|
32 #define M_PI 3.1415926535897932384626433832795029
|
|
33 #endif
|
|
34 #include <inttypes.h>
|
|
35
|
|
36 #include "a52.h"
|
|
37 #include "a52_internal.h"
|
|
38 #include "mm_accel.h"
|
|
39
|
|
40 void (* imdct_256) (sample_t data[], sample_t delay[], sample_t bias);
|
|
41 void (* imdct_512) (sample_t data[], sample_t delay[], sample_t bias);
|
|
42
|
|
43 typedef struct complex_s {
|
|
44 sample_t real;
|
|
45 sample_t imag;
|
|
46 } complex_t;
|
|
47
|
|
48
|
|
49 /* 128 point bit-reverse LUT */
|
|
50 static uint8_t bit_reverse_512[] = {
|
|
51 0x00, 0x40, 0x20, 0x60, 0x10, 0x50, 0x30, 0x70,
|
|
52 0x08, 0x48, 0x28, 0x68, 0x18, 0x58, 0x38, 0x78,
|
|
53 0x04, 0x44, 0x24, 0x64, 0x14, 0x54, 0x34, 0x74,
|
|
54 0x0c, 0x4c, 0x2c, 0x6c, 0x1c, 0x5c, 0x3c, 0x7c,
|
|
55 0x02, 0x42, 0x22, 0x62, 0x12, 0x52, 0x32, 0x72,
|
|
56 0x0a, 0x4a, 0x2a, 0x6a, 0x1a, 0x5a, 0x3a, 0x7a,
|
|
57 0x06, 0x46, 0x26, 0x66, 0x16, 0x56, 0x36, 0x76,
|
|
58 0x0e, 0x4e, 0x2e, 0x6e, 0x1e, 0x5e, 0x3e, 0x7e,
|
|
59 0x01, 0x41, 0x21, 0x61, 0x11, 0x51, 0x31, 0x71,
|
|
60 0x09, 0x49, 0x29, 0x69, 0x19, 0x59, 0x39, 0x79,
|
|
61 0x05, 0x45, 0x25, 0x65, 0x15, 0x55, 0x35, 0x75,
|
|
62 0x0d, 0x4d, 0x2d, 0x6d, 0x1d, 0x5d, 0x3d, 0x7d,
|
|
63 0x03, 0x43, 0x23, 0x63, 0x13, 0x53, 0x33, 0x73,
|
|
64 0x0b, 0x4b, 0x2b, 0x6b, 0x1b, 0x5b, 0x3b, 0x7b,
|
|
65 0x07, 0x47, 0x27, 0x67, 0x17, 0x57, 0x37, 0x77,
|
|
66 0x0f, 0x4f, 0x2f, 0x6f, 0x1f, 0x5f, 0x3f, 0x7f};
|
|
67
|
|
68 static uint8_t bit_reverse_256[] = {
|
|
69 0x00, 0x20, 0x10, 0x30, 0x08, 0x28, 0x18, 0x38,
|
|
70 0x04, 0x24, 0x14, 0x34, 0x0c, 0x2c, 0x1c, 0x3c,
|
|
71 0x02, 0x22, 0x12, 0x32, 0x0a, 0x2a, 0x1a, 0x3a,
|
|
72 0x06, 0x26, 0x16, 0x36, 0x0e, 0x2e, 0x1e, 0x3e,
|
|
73 0x01, 0x21, 0x11, 0x31, 0x09, 0x29, 0x19, 0x39,
|
|
74 0x05, 0x25, 0x15, 0x35, 0x0d, 0x2d, 0x1d, 0x3d,
|
|
75 0x03, 0x23, 0x13, 0x33, 0x0b, 0x2b, 0x1b, 0x3b,
|
|
76 0x07, 0x27, 0x17, 0x37, 0x0f, 0x2f, 0x1f, 0x3f};
|
|
77
|
3579
|
78 #ifdef ARCH_X86
|
3508
|
79 // NOTE: SSE needs 16byte alignment or it will segfault
|
3581
|
80 //
|
3508
|
81 static complex_t __attribute__((aligned(16))) buf[128];
|
3581
|
82 static float __attribute__((aligned(16))) sseSinCos1c[256];
|
|
83 static float __attribute__((aligned(16))) sseSinCos1d[256];
|
3512
|
84 static float __attribute__((aligned(16))) ps111_1[4]={1,1,1,-1};
|
3534
|
85 //static float __attribute__((aligned(16))) sseW0[4];
|
|
86 static float __attribute__((aligned(16))) sseW1[8];
|
|
87 static float __attribute__((aligned(16))) sseW2[16];
|
|
88 static float __attribute__((aligned(16))) sseW3[32];
|
|
89 static float __attribute__((aligned(16))) sseW4[64];
|
|
90 static float __attribute__((aligned(16))) sseW5[128];
|
|
91 static float __attribute__((aligned(16))) sseW6[256];
|
|
92 static float __attribute__((aligned(16))) *sseW[7]=
|
|
93 {NULL /*sseW0*/,sseW1,sseW2,sseW3,sseW4,sseW5,sseW6};
|
3553
|
94 static float __attribute__((aligned(16))) sseWindow[512];
|
3508
|
95 #else
|
3394
|
96 static complex_t buf[128];
|
3508
|
97 #endif
|
3394
|
98
|
|
99 /* Twiddle factor LUT */
|
|
100 static complex_t w_1[1];
|
|
101 static complex_t w_2[2];
|
|
102 static complex_t w_4[4];
|
|
103 static complex_t w_8[8];
|
|
104 static complex_t w_16[16];
|
|
105 static complex_t w_32[32];
|
|
106 static complex_t w_64[64];
|
|
107 static complex_t * w[7] = {w_1, w_2, w_4, w_8, w_16, w_32, w_64};
|
|
108
|
|
109 /* Twiddle factors for IMDCT */
|
|
110 static sample_t xcos1[128];
|
|
111 static sample_t xsin1[128];
|
|
112 static sample_t xcos2[64];
|
|
113 static sample_t xsin2[64];
|
|
114
|
|
115 /* Windowing function for Modified DCT - Thank you acroread */
|
|
116 sample_t imdct_window[] = {
|
|
117 0.00014, 0.00024, 0.00037, 0.00051, 0.00067, 0.00086, 0.00107, 0.00130,
|
|
118 0.00157, 0.00187, 0.00220, 0.00256, 0.00297, 0.00341, 0.00390, 0.00443,
|
|
119 0.00501, 0.00564, 0.00632, 0.00706, 0.00785, 0.00871, 0.00962, 0.01061,
|
|
120 0.01166, 0.01279, 0.01399, 0.01526, 0.01662, 0.01806, 0.01959, 0.02121,
|
|
121 0.02292, 0.02472, 0.02662, 0.02863, 0.03073, 0.03294, 0.03527, 0.03770,
|
|
122 0.04025, 0.04292, 0.04571, 0.04862, 0.05165, 0.05481, 0.05810, 0.06153,
|
|
123 0.06508, 0.06878, 0.07261, 0.07658, 0.08069, 0.08495, 0.08935, 0.09389,
|
|
124 0.09859, 0.10343, 0.10842, 0.11356, 0.11885, 0.12429, 0.12988, 0.13563,
|
|
125 0.14152, 0.14757, 0.15376, 0.16011, 0.16661, 0.17325, 0.18005, 0.18699,
|
|
126 0.19407, 0.20130, 0.20867, 0.21618, 0.22382, 0.23161, 0.23952, 0.24757,
|
|
127 0.25574, 0.26404, 0.27246, 0.28100, 0.28965, 0.29841, 0.30729, 0.31626,
|
|
128 0.32533, 0.33450, 0.34376, 0.35311, 0.36253, 0.37204, 0.38161, 0.39126,
|
|
129 0.40096, 0.41072, 0.42054, 0.43040, 0.44030, 0.45023, 0.46020, 0.47019,
|
|
130 0.48020, 0.49022, 0.50025, 0.51028, 0.52031, 0.53033, 0.54033, 0.55031,
|
|
131 0.56026, 0.57019, 0.58007, 0.58991, 0.59970, 0.60944, 0.61912, 0.62873,
|
|
132 0.63827, 0.64774, 0.65713, 0.66643, 0.67564, 0.68476, 0.69377, 0.70269,
|
|
133 0.71150, 0.72019, 0.72877, 0.73723, 0.74557, 0.75378, 0.76186, 0.76981,
|
|
134 0.77762, 0.78530, 0.79283, 0.80022, 0.80747, 0.81457, 0.82151, 0.82831,
|
|
135 0.83496, 0.84145, 0.84779, 0.85398, 0.86001, 0.86588, 0.87160, 0.87716,
|
|
136 0.88257, 0.88782, 0.89291, 0.89785, 0.90264, 0.90728, 0.91176, 0.91610,
|
|
137 0.92028, 0.92432, 0.92822, 0.93197, 0.93558, 0.93906, 0.94240, 0.94560,
|
|
138 0.94867, 0.95162, 0.95444, 0.95713, 0.95971, 0.96217, 0.96451, 0.96674,
|
|
139 0.96887, 0.97089, 0.97281, 0.97463, 0.97635, 0.97799, 0.97953, 0.98099,
|
|
140 0.98236, 0.98366, 0.98488, 0.98602, 0.98710, 0.98811, 0.98905, 0.98994,
|
|
141 0.99076, 0.99153, 0.99225, 0.99291, 0.99353, 0.99411, 0.99464, 0.99513,
|
|
142 0.99558, 0.99600, 0.99639, 0.99674, 0.99706, 0.99736, 0.99763, 0.99788,
|
|
143 0.99811, 0.99831, 0.99850, 0.99867, 0.99882, 0.99895, 0.99908, 0.99919,
|
|
144 0.99929, 0.99938, 0.99946, 0.99953, 0.99959, 0.99965, 0.99969, 0.99974,
|
|
145 0.99978, 0.99981, 0.99984, 0.99986, 0.99988, 0.99990, 0.99992, 0.99993,
|
|
146 0.99994, 0.99995, 0.99996, 0.99997, 0.99998, 0.99998, 0.99998, 0.99999,
|
|
147 0.99999, 0.99999, 0.99999, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000,
|
|
148 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000 };
|
|
149
|
|
150
|
|
151 static inline void swap_cmplx(complex_t *a, complex_t *b)
|
|
152 {
|
|
153 complex_t tmp;
|
|
154
|
|
155 tmp = *a;
|
|
156 *a = *b;
|
|
157 *b = tmp;
|
|
158 }
|
|
159
|
|
160
|
|
161
|
|
162 static inline complex_t cmplx_mult(complex_t a, complex_t b)
|
|
163 {
|
|
164 complex_t ret;
|
|
165
|
|
166 ret.real = a.real * b.real - a.imag * b.imag;
|
|
167 ret.imag = a.real * b.imag + a.imag * b.real;
|
|
168
|
|
169 return ret;
|
|
170 }
|
|
171
|
|
172 void
|
|
173 imdct_do_512(sample_t data[],sample_t delay[], sample_t bias)
|
|
174 {
|
|
175 int i,k;
|
|
176 int p,q;
|
|
177 int m;
|
|
178 int two_m;
|
|
179 int two_m_plus_one;
|
|
180
|
|
181 sample_t tmp_a_i;
|
|
182 sample_t tmp_a_r;
|
|
183 sample_t tmp_b_i;
|
|
184 sample_t tmp_b_r;
|
|
185
|
|
186 sample_t *data_ptr;
|
|
187 sample_t *delay_ptr;
|
|
188 sample_t *window_ptr;
|
|
189
|
|
190 /* 512 IMDCT with source and dest data in 'data' */
|
|
191
|
3529
|
192 /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
|
3394
|
193 for( i=0; i < 128; i++) {
|
|
194 /* z[i] = (X[256-2*i-1] + j * X[2*i]) * (xcos1[i] + j * xsin1[i]) ; */
|
|
195 buf[i].real = (data[256-2*i-1] * xcos1[i]) - (data[2*i] * xsin1[i]);
|
|
196 buf[i].imag = -1.0 * ((data[2*i] * xcos1[i]) + (data[256-2*i-1] * xsin1[i]));
|
|
197 }
|
|
198
|
|
199 /* Bit reversed shuffling */
|
|
200 for(i=0; i<128; i++) {
|
|
201 k = bit_reverse_512[i];
|
|
202 if (k < i)
|
|
203 swap_cmplx(&buf[i],&buf[k]);
|
|
204 }
|
3529
|
205
|
3394
|
206
|
|
207 /* FFT Merge */
|
3549
|
208 /* unoptimized variant
|
|
209 for (m=1; m < 7; m++) {
|
|
210 if(m)
|
|
211 two_m = (1 << m);
|
|
212 else
|
|
213 two_m = 1;
|
|
214
|
|
215 two_m_plus_one = (1 << (m+1));
|
|
216
|
|
217 for(i = 0; i < 128; i += two_m_plus_one) {
|
|
218 for(k = 0; k < two_m; k++) {
|
|
219 p = k + i;
|
|
220 q = p + two_m;
|
3508
|
221 tmp_a_r = buf[p].real;
|
|
222 tmp_a_i = buf[p].imag;
|
3549
|
223 tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
|
|
224 tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
|
3508
|
225 buf[p].real = tmp_a_r + tmp_b_r;
|
|
226 buf[p].imag = tmp_a_i + tmp_b_i;
|
|
227 buf[q].real = tmp_a_r - tmp_b_r;
|
|
228 buf[q].imag = tmp_a_i - tmp_b_i;
|
3549
|
229 }
|
|
230 }
|
|
231 }
|
|
232 */
|
|
233
|
3579
|
234 for(i = 0; i < 128; i += 2) {
|
|
235 tmp_a_r = buf[i].real;
|
|
236 tmp_a_i = buf[i].imag;
|
|
237 tmp_b_r = buf[i+1].real;
|
|
238 tmp_b_i = buf[i+1].imag;
|
|
239 buf[i].real = tmp_a_r + tmp_b_r;
|
|
240 buf[i].imag = tmp_a_i + tmp_b_i;
|
|
241 buf[i+1].real = tmp_a_r - tmp_b_r;
|
|
242 buf[i+1].imag = tmp_a_i - tmp_b_i;
|
|
243 }
|
|
244
|
|
245 // Note w[1]={{1,0}, {0,-1}}
|
|
246 for(i = 0; i < 128; i += 4) {
|
|
247 tmp_a_r = buf[i].real;
|
|
248 tmp_a_i = buf[i].imag;
|
|
249 tmp_b_r = buf[i+2].real;
|
|
250 tmp_b_i = buf[i+2].imag;
|
|
251 buf[i].real = tmp_a_r + tmp_b_r;
|
|
252 buf[i].imag = tmp_a_i + tmp_b_i;
|
|
253 buf[i+2].real = tmp_a_r - tmp_b_r;
|
|
254 buf[i+2].imag = tmp_a_i - tmp_b_i;
|
|
255 tmp_a_r = buf[i+1].real;
|
|
256 tmp_a_i = buf[i+1].imag;
|
|
257 tmp_b_r = buf[i+3].imag;
|
|
258 tmp_b_i = buf[i+3].real;
|
|
259 buf[i+1].real = tmp_a_r + tmp_b_r;
|
|
260 buf[i+1].imag = tmp_a_i - tmp_b_i;
|
|
261 buf[i+3].real = tmp_a_r - tmp_b_r;
|
|
262 buf[i+3].imag = tmp_a_i + tmp_b_i;
|
|
263 }
|
|
264
|
|
265 for(i = 0; i < 128; i += 8) {
|
|
266 tmp_a_r = buf[i].real;
|
|
267 tmp_a_i = buf[i].imag;
|
|
268 tmp_b_r = buf[i+4].real;
|
|
269 tmp_b_i = buf[i+4].imag;
|
|
270 buf[i].real = tmp_a_r + tmp_b_r;
|
|
271 buf[i].imag = tmp_a_i + tmp_b_i;
|
|
272 buf[i+4].real = tmp_a_r - tmp_b_r;
|
|
273 buf[i+4].imag = tmp_a_i - tmp_b_i;
|
|
274 tmp_a_r = buf[1+i].real;
|
|
275 tmp_a_i = buf[1+i].imag;
|
|
276 tmp_b_r = (buf[i+5].real + buf[i+5].imag) * w[2][1].real;
|
|
277 tmp_b_i = (buf[i+5].imag - buf[i+5].real) * w[2][1].real;
|
|
278 buf[1+i].real = tmp_a_r + tmp_b_r;
|
|
279 buf[1+i].imag = tmp_a_i + tmp_b_i;
|
|
280 buf[i+5].real = tmp_a_r - tmp_b_r;
|
|
281 buf[i+5].imag = tmp_a_i - tmp_b_i;
|
|
282 tmp_a_r = buf[i+2].real;
|
|
283 tmp_a_i = buf[i+2].imag;
|
|
284 tmp_b_r = buf[i+6].imag;
|
|
285 tmp_b_i = - buf[i+6].real;
|
|
286 buf[i+2].real = tmp_a_r + tmp_b_r;
|
|
287 buf[i+2].imag = tmp_a_i + tmp_b_i;
|
|
288 buf[i+6].real = tmp_a_r - tmp_b_r;
|
|
289 buf[i+6].imag = tmp_a_i - tmp_b_i;
|
|
290 tmp_a_r = buf[i+3].real;
|
|
291 tmp_a_i = buf[i+3].imag;
|
|
292 tmp_b_r = (buf[i+7].real - buf[i+7].imag) * w[2][3].imag;
|
|
293 tmp_b_i = (buf[i+7].imag + buf[i+7].real) * w[2][3].imag;
|
|
294 buf[i+3].real = tmp_a_r + tmp_b_r;
|
|
295 buf[i+3].imag = tmp_a_i + tmp_b_i;
|
|
296 buf[i+7].real = tmp_a_r - tmp_b_r;
|
|
297 buf[i+7].imag = tmp_a_i - tmp_b_i;
|
|
298 }
|
|
299
|
|
300 for (m=3; m < 7; m++) {
|
|
301 two_m = (1 << m);
|
|
302
|
|
303 two_m_plus_one = two_m<<1;
|
|
304
|
|
305 for(i = 0; i < 128; i += two_m_plus_one) {
|
|
306 for(k = 0; k < two_m; k++) {
|
|
307 int p = k + i;
|
|
308 int q = p + two_m;
|
|
309 tmp_a_r = buf[p].real;
|
|
310 tmp_a_i = buf[p].imag;
|
|
311 tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
|
|
312 tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
|
|
313 buf[p].real = tmp_a_r + tmp_b_r;
|
|
314 buf[p].imag = tmp_a_i + tmp_b_i;
|
|
315 buf[q].real = tmp_a_r - tmp_b_r;
|
|
316 buf[q].imag = tmp_a_i - tmp_b_i;
|
|
317 }
|
|
318 }
|
|
319 }
|
|
320
|
|
321 /* Post IFFT complex multiply plus IFFT complex conjugate*/
|
|
322 for( i=0; i < 128; i++) {
|
|
323 /* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
|
|
324 tmp_a_r = buf[i].real;
|
|
325 tmp_a_i = -1.0 * buf[i].imag;
|
|
326 buf[i].real =(tmp_a_r * xcos1[i]) - (tmp_a_i * xsin1[i]);
|
|
327 buf[i].imag =(tmp_a_r * xsin1[i]) + (tmp_a_i * xcos1[i]);
|
|
328 }
|
|
329
|
|
330 data_ptr = data;
|
|
331 delay_ptr = delay;
|
|
332 window_ptr = imdct_window;
|
|
333
|
|
334 /* Window and convert to real valued signal */
|
|
335 for(i=0; i< 64; i++) {
|
|
336 *data_ptr++ = -buf[64+i].imag * *window_ptr++ + *delay_ptr++ + bias;
|
|
337 *data_ptr++ = buf[64-i-1].real * *window_ptr++ + *delay_ptr++ + bias;
|
|
338 }
|
|
339
|
|
340 for(i=0; i< 64; i++) {
|
|
341 *data_ptr++ = -buf[i].real * *window_ptr++ + *delay_ptr++ + bias;
|
|
342 *data_ptr++ = buf[128-i-1].imag * *window_ptr++ + *delay_ptr++ + bias;
|
|
343 }
|
|
344
|
|
345 /* The trailing edge of the window goes into the delay line */
|
|
346 delay_ptr = delay;
|
|
347
|
|
348 for(i=0; i< 64; i++) {
|
|
349 *delay_ptr++ = -buf[64+i].real * *--window_ptr;
|
|
350 *delay_ptr++ = buf[64-i-1].imag * *--window_ptr;
|
|
351 }
|
|
352
|
|
353 for(i=0; i<64; i++) {
|
|
354 *delay_ptr++ = buf[i].imag * *--window_ptr;
|
|
355 *delay_ptr++ = -buf[128-i-1].real * *--window_ptr;
|
|
356 }
|
|
357 }
|
|
358
|
|
359 #ifdef ARCH_X86
|
|
360 void
|
|
361 imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
|
|
362 {
|
|
363 int i,k;
|
|
364 int p,q;
|
|
365 int m;
|
|
366 int two_m;
|
|
367 int two_m_plus_one;
|
|
368
|
|
369 sample_t tmp_a_i;
|
|
370 sample_t tmp_a_r;
|
|
371 sample_t tmp_b_i;
|
|
372 sample_t tmp_b_r;
|
|
373
|
|
374 sample_t *data_ptr;
|
|
375 sample_t *delay_ptr;
|
|
376 sample_t *window_ptr;
|
|
377
|
|
378 /* 512 IMDCT with source and dest data in 'data' */
|
|
379
|
|
380 /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
|
|
381 /* Bit reversed shuffling */
|
|
382 asm volatile(
|
|
383 "xorl %%esi, %%esi \n\t"
|
|
384 "leal bit_reverse_512, %%eax \n\t"
|
|
385 "movl $1008, %%edi \n\t"
|
|
386 "pushl %%ebp \n\t" //use ebp without telling gcc
|
|
387 ".balign 16 \n\t"
|
|
388 "1: \n\t"
|
3584
|
389 "movlps (%0, %%esi), %%xmm0 \n\t" // XXXI
|
|
390 "movhps 8(%0, %%edi), %%xmm0 \n\t" // RXXI
|
|
391 "movlps 8(%0, %%esi), %%xmm1 \n\t" // XXXi
|
|
392 "movhps (%0, %%edi), %%xmm1 \n\t" // rXXi
|
|
393 "shufps $0x33, %%xmm1, %%xmm0 \n\t" // irIR
|
|
394 "movaps sseSinCos1c(%%esi), %%xmm2 \n\t"
|
|
395 "mulps %%xmm0, %%xmm2 \n\t"
|
|
396 "shufps $0xB1, %%xmm0, %%xmm0 \n\t" // riRI
|
|
397 "mulps sseSinCos1d(%%esi), %%xmm0 \n\t"
|
|
398 "subps %%xmm0, %%xmm2 \n\t"
|
3579
|
399 "movzbl (%%eax), %%edx \n\t"
|
|
400 "movzbl 1(%%eax), %%ebp \n\t"
|
3584
|
401 "movlps %%xmm2, (%1, %%edx,8) \n\t"
|
|
402 "movhps %%xmm2, (%1, %%ebp,8) \n\t"
|
3579
|
403 "addl $16, %%esi \n\t"
|
|
404 "addl $2, %%eax \n\t" // avoid complex addressing for P4 crap
|
|
405 "subl $16, %%edi \n\t"
|
|
406 " jnc 1b \n\t"
|
|
407 "popl %%ebp \n\t"//no we didnt touch ebp *g*
|
|
408 :: "b" (data), "c" (buf)
|
|
409 : "%esi", "%edi", "%eax", "%edx"
|
|
410 );
|
|
411
|
|
412
|
|
413 /* FFT Merge */
|
|
414 /* unoptimized variant
|
|
415 for (m=1; m < 7; m++) {
|
|
416 if(m)
|
|
417 two_m = (1 << m);
|
|
418 else
|
|
419 two_m = 1;
|
|
420
|
|
421 two_m_plus_one = (1 << (m+1));
|
|
422
|
|
423 for(i = 0; i < 128; i += two_m_plus_one) {
|
|
424 for(k = 0; k < two_m; k++) {
|
|
425 p = k + i;
|
|
426 q = p + two_m;
|
|
427 tmp_a_r = buf[p].real;
|
|
428 tmp_a_i = buf[p].imag;
|
|
429 tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
|
|
430 tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
|
|
431 buf[p].real = tmp_a_r + tmp_b_r;
|
|
432 buf[p].imag = tmp_a_i + tmp_b_i;
|
|
433 buf[q].real = tmp_a_r - tmp_b_r;
|
|
434 buf[q].imag = tmp_a_i - tmp_b_i;
|
|
435 }
|
|
436 }
|
|
437 }
|
|
438 */
|
|
439
|
3549
|
440 // Note w[0][0]={1,0}
|
3508
|
441 asm volatile(
|
|
442 "xorps %%xmm1, %%xmm1 \n\t"
|
|
443 "xorps %%xmm2, %%xmm2 \n\t"
|
|
444 "movl %0, %%esi \n\t"
|
3529
|
445 ".balign 16 \n\t"
|
3508
|
446 "1: \n\t"
|
|
447 "movlps (%%esi), %%xmm0 \n\t" //buf[p]
|
|
448 "movlps 8(%%esi), %%xmm1\n\t" //buf[q]
|
|
449 "movhps (%%esi), %%xmm0 \n\t" //buf[p]
|
|
450 "movhps 8(%%esi), %%xmm2\n\t" //buf[q]
|
|
451 "addps %%xmm1, %%xmm0 \n\t"
|
|
452 "subps %%xmm2, %%xmm0 \n\t"
|
|
453 "movaps %%xmm0, (%%esi) \n\t"
|
|
454 "addl $16, %%esi \n\t"
|
|
455 "cmpl %1, %%esi \n\t"
|
|
456 " jb 1b \n\t"
|
|
457 :: "g" (buf), "r" (buf + 128)
|
|
458 : "%esi"
|
|
459 );
|
3549
|
460
|
3512
|
461 // Note w[1]={{1,0}, {0,-1}}
|
|
462 asm volatile(
|
|
463 "movaps ps111_1, %%xmm7 \n\t" // 1,1,1,-1
|
|
464 "movl %0, %%esi \n\t"
|
3529
|
465 ".balign 16 \n\t"
|
3512
|
466 "1: \n\t"
|
|
467 "movaps 16(%%esi), %%xmm2 \n\t" //r2,i2,r3,i3
|
|
468 "shufps $0xB4, %%xmm2, %%xmm2 \n\t" //r2,i2,i3,r3
|
|
469 "mulps %%xmm7, %%xmm2 \n\t" //r2,i2,i3,-r3
|
|
470 "movaps (%%esi), %%xmm0 \n\t" //r0,i0,r1,i1
|
|
471 "movaps (%%esi), %%xmm1 \n\t" //r0,i0,r1,i1
|
|
472 "addps %%xmm2, %%xmm0 \n\t"
|
|
473 "subps %%xmm2, %%xmm1 \n\t"
|
|
474 "movaps %%xmm0, (%%esi) \n\t"
|
|
475 "movaps %%xmm1, 16(%%esi) \n\t"
|
|
476 "addl $32, %%esi \n\t"
|
|
477 "cmpl %1, %%esi \n\t"
|
|
478 " jb 1b \n\t"
|
|
479 :: "g" (buf), "r" (buf + 128)
|
|
480 : "%esi"
|
|
481 );
|
3549
|
482
|
3534
|
483 /*
|
|
484 Note sseW2+0={1,1,sqrt(2),sqrt(2))
|
|
485 Note sseW2+16={0,0,sqrt(2),-sqrt(2))
|
|
486 Note sseW2+32={0,0,-sqrt(2),-sqrt(2))
|
|
487 Note sseW2+48={1,-1,sqrt(2),-sqrt(2))
|
|
488 */
|
|
489 asm volatile(
|
3537
|
490 "movaps 48+sseW2, %%xmm6 \n\t"
|
3534
|
491 "movaps 16+sseW2, %%xmm7 \n\t"
|
|
492 "xorps %%xmm5, %%xmm5 \n\t"
|
|
493 "xorps %%xmm2, %%xmm2 \n\t"
|
|
494 "movl %0, %%esi \n\t"
|
|
495 ".balign 16 \n\t"
|
|
496 "1: \n\t"
|
3537
|
497 "movaps 32(%%esi), %%xmm2 \n\t" //r4,i4,r5,i5
|
3534
|
498 "movaps 48(%%esi), %%xmm3 \n\t" //r6,i6,r7,i7
|
3537
|
499 "movaps sseW2, %%xmm4 \n\t" //r4,i4,r5,i5
|
|
500 "movaps 32+sseW2, %%xmm5 \n\t" //r6,i6,r7,i7
|
|
501 "mulps %%xmm2, %%xmm4 \n\t"
|
|
502 "mulps %%xmm3, %%xmm5 \n\t"
|
3534
|
503 "shufps $0xB1, %%xmm2, %%xmm2 \n\t" //i4,r4,i5,r5
|
|
504 "shufps $0xB1, %%xmm3, %%xmm3 \n\t" //i6,r6,i7,r7
|
3537
|
505 "mulps %%xmm6, %%xmm3 \n\t"
|
3534
|
506 "mulps %%xmm7, %%xmm2 \n\t"
|
|
507 "movaps (%%esi), %%xmm0 \n\t" //r0,i0,r1,i1
|
|
508 "movaps 16(%%esi), %%xmm1 \n\t" //r2,i2,r3,i3
|
|
509 "addps %%xmm4, %%xmm2 \n\t"
|
|
510 "addps %%xmm5, %%xmm3 \n\t"
|
|
511 "movaps %%xmm2, %%xmm4 \n\t"
|
|
512 "movaps %%xmm3, %%xmm5 \n\t"
|
|
513 "addps %%xmm0, %%xmm2 \n\t"
|
|
514 "addps %%xmm1, %%xmm3 \n\t"
|
|
515 "subps %%xmm4, %%xmm0 \n\t"
|
|
516 "subps %%xmm5, %%xmm1 \n\t"
|
|
517 "movaps %%xmm2, (%%esi) \n\t"
|
|
518 "movaps %%xmm3, 16(%%esi) \n\t"
|
|
519 "movaps %%xmm0, 32(%%esi) \n\t"
|
|
520 "movaps %%xmm1, 48(%%esi) \n\t"
|
|
521 "addl $64, %%esi \n\t"
|
|
522 "cmpl %1, %%esi \n\t"
|
|
523 " jb 1b \n\t"
|
|
524 :: "g" (buf), "r" (buf + 128)
|
|
525 : "%esi"
|
|
526 );
|
3508
|
527
|
3546
|
528 for (m=3; m < 7; m++) {
|
|
529 two_m = (1 << m);
|
|
530 two_m_plus_one = two_m<<1;
|
|
531 asm volatile(
|
|
532 "movl %0, %%esi \n\t"
|
|
533 ".balign 16 \n\t"
|
|
534 "1: \n\t"
|
|
535 "xorl %%edi, %%edi \n\t" // k
|
|
536 "leal (%%esi, %3), %%edx \n\t"
|
|
537 "2: \n\t"
|
|
538 "movaps (%%edx, %%edi), %%xmm1 \n\t"
|
|
539 "movaps (%4, %%edi, 2), %%xmm2 \n\t"
|
|
540 "mulps %%xmm1, %%xmm2 \n\t"
|
|
541 "shufps $0xB1, %%xmm1, %%xmm1 \n\t"
|
|
542 "mulps 16(%4, %%edi, 2), %%xmm1 \n\t"
|
|
543 "movaps (%%esi, %%edi), %%xmm0 \n\t"
|
|
544 "addps %%xmm2, %%xmm1 \n\t"
|
|
545 "movaps %%xmm1, %%xmm2 \n\t"
|
|
546 "addps %%xmm0, %%xmm1 \n\t"
|
|
547 "subps %%xmm2, %%xmm0 \n\t"
|
|
548 "movaps %%xmm1, (%%esi, %%edi) \n\t"
|
|
549 "movaps %%xmm0, (%%edx, %%edi) \n\t"
|
|
550 "addl $16, %%edi \n\t"
|
|
551 "cmpl %3, %%edi \n\t" //FIXME (opt) count against 0
|
|
552 " jb 2b \n\t"
|
|
553 "addl %2, %%esi \n\t"
|
|
554 "cmpl %1, %%esi \n\t"
|
|
555 " jb 1b \n\t"
|
|
556 :: "g" (buf), "m" (buf+128), "m" (two_m_plus_one<<3), "r" (two_m<<3),
|
|
557 "r" (sseW[m])
|
|
558 : "%esi", "%edi", "%edx"
|
|
559 );
|
|
560 }
|
|
561
|
3581
|
562 asm volatile(
|
|
563 "movl $-1024, %%esi \n\t"
|
|
564 ".balign 16 \n\t"
|
|
565 "1: \n\t"
|
|
566 "movaps (%0, %%esi), %%xmm0 \n\t"
|
|
567 "movaps (%0, %%esi), %%xmm1 \n\t"
|
|
568 "shufps $0xB1, %%xmm0, %%xmm0 \n\t"
|
|
569 "mulps 1024+sseSinCos1c(%%esi), %%xmm1 \n\t"
|
|
570 "mulps 1024+sseSinCos1d(%%esi), %%xmm0 \n\t"
|
|
571 "addps %%xmm1, %%xmm0 \n\t"
|
|
572 "movaps %%xmm0, (%0, %%esi) \n\t"
|
|
573 "addl $16, %%esi \n\t"
|
|
574 " jnz 1b \n\t"
|
|
575 :: "r" (buf+128)
|
|
576 : "%esi"
|
|
577 );
|
|
578
|
3394
|
579
|
|
580 data_ptr = data;
|
|
581 delay_ptr = delay;
|
|
582 window_ptr = imdct_window;
|
|
583
|
|
584 /* Window and convert to real valued signal */
|
3552
|
585 asm volatile(
|
|
586 "xorl %%edi, %%edi \n\t" // 0
|
|
587 "xorl %%esi, %%esi \n\t" // 0
|
|
588 "movss %3, %%xmm2 \n\t" // bias
|
|
589 "shufps $0x00, %%xmm2, %%xmm2 \n\t" // bias, bias, ...
|
|
590 ".balign 16 \n\t"
|
|
591 "1: \n\t"
|
|
592 "movlps (%0, %%esi), %%xmm0 \n\t" // ? ? A ?
|
|
593 "movlps 8(%0, %%esi), %%xmm1 \n\t" // ? ? C ?
|
|
594 "movhps -16(%0, %%edi), %%xmm1 \n\t" // ? D C ?
|
|
595 "movhps -8(%0, %%edi), %%xmm0 \n\t" // ? B A ?
|
|
596 "shufps $0x99, %%xmm1, %%xmm0 \n\t" // D C B A
|
|
597 "mulps sseWindow(%%esi), %%xmm0 \n\t"
|
|
598 "addps (%2, %%esi), %%xmm0 \n\t"
|
|
599 "addps %%xmm2, %%xmm0 \n\t"
|
|
600 "movaps %%xmm0, (%1, %%esi) \n\t"
|
|
601 "addl $16, %%esi \n\t"
|
|
602 "subl $16, %%edi \n\t"
|
|
603 "cmpl $512, %%esi \n\t"
|
|
604 " jb 1b \n\t"
|
|
605 :: "r" (buf+64), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
|
|
606 : "%esi", "%edi"
|
|
607 );
|
|
608 data_ptr+=128;
|
|
609 delay_ptr+=128;
|
3553
|
610 // window_ptr+=128;
|
3579
|
611
|
3552
|
612 asm volatile(
|
|
613 "movl $1024, %%edi \n\t" // 512
|
|
614 "xorl %%esi, %%esi \n\t" // 0
|
|
615 "movss %3, %%xmm2 \n\t" // bias
|
|
616 "shufps $0x00, %%xmm2, %%xmm2 \n\t" // bias, bias, ...
|
|
617 ".balign 16 \n\t"
|
|
618 "1: \n\t"
|
|
619 "movlps (%0, %%esi), %%xmm0 \n\t" // ? ? ? A
|
|
620 "movlps 8(%0, %%esi), %%xmm1 \n\t" // ? ? ? C
|
|
621 "movhps -16(%0, %%edi), %%xmm1 \n\t" // D ? ? C
|
|
622 "movhps -8(%0, %%edi), %%xmm0 \n\t" // B ? ? A
|
|
623 "shufps $0xCC, %%xmm1, %%xmm0 \n\t" // D C B A
|
|
624 "mulps 512+sseWindow(%%esi), %%xmm0 \n\t"
|
|
625 "addps (%2, %%esi), %%xmm0 \n\t"
|
|
626 "addps %%xmm2, %%xmm0 \n\t"
|
|
627 "movaps %%xmm0, (%1, %%esi) \n\t"
|
|
628 "addl $16, %%esi \n\t"
|
|
629 "subl $16, %%edi \n\t"
|
|
630 "cmpl $512, %%esi \n\t"
|
|
631 " jb 1b \n\t"
|
|
632 :: "r" (buf), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
|
|
633 : "%esi", "%edi"
|
|
634 );
|
|
635 data_ptr+=128;
|
3553
|
636 // window_ptr+=128;
|
3394
|
637
|
|
638 /* The trailing edge of the window goes into the delay line */
|
|
639 delay_ptr = delay;
|
|
640
|
3553
|
641 asm volatile(
|
|
642 "xorl %%edi, %%edi \n\t" // 0
|
|
643 "xorl %%esi, %%esi \n\t" // 0
|
|
644 ".balign 16 \n\t"
|
|
645 "1: \n\t"
|
|
646 "movlps (%0, %%esi), %%xmm0 \n\t" // ? ? ? A
|
|
647 "movlps 8(%0, %%esi), %%xmm1 \n\t" // ? ? ? C
|
|
648 "movhps -16(%0, %%edi), %%xmm1 \n\t" // D ? ? C
|
|
649 "movhps -8(%0, %%edi), %%xmm0 \n\t" // B ? ? A
|
|
650 "shufps $0xCC, %%xmm1, %%xmm0 \n\t" // D C B A
|
|
651 "mulps 1024+sseWindow(%%esi), %%xmm0 \n\t"
|
|
652 "movaps %%xmm0, (%1, %%esi) \n\t"
|
|
653 "addl $16, %%esi \n\t"
|
|
654 "subl $16, %%edi \n\t"
|
|
655 "cmpl $512, %%esi \n\t"
|
|
656 " jb 1b \n\t"
|
|
657 :: "r" (buf+64), "r" (delay_ptr)
|
|
658 : "%esi", "%edi"
|
|
659 );
|
|
660 delay_ptr+=128;
|
|
661 // window_ptr-=128;
|
3579
|
662
|
3553
|
663 asm volatile(
|
|
664 "movl $1024, %%edi \n\t" // 1024
|
|
665 "xorl %%esi, %%esi \n\t" // 0
|
|
666 ".balign 16 \n\t"
|
|
667 "1: \n\t"
|
|
668 "movlps (%0, %%esi), %%xmm0 \n\t" // ? ? A ?
|
|
669 "movlps 8(%0, %%esi), %%xmm1 \n\t" // ? ? C ?
|
|
670 "movhps -16(%0, %%edi), %%xmm1 \n\t" // ? D C ?
|
|
671 "movhps -8(%0, %%edi), %%xmm0 \n\t" // ? B A ?
|
|
672 "shufps $0x99, %%xmm1, %%xmm0 \n\t" // D C B A
|
|
673 "mulps 1536+sseWindow(%%esi), %%xmm0 \n\t"
|
|
674 "movaps %%xmm0, (%1, %%esi) \n\t"
|
|
675 "addl $16, %%esi \n\t"
|
|
676 "subl $16, %%edi \n\t"
|
|
677 "cmpl $512, %%esi \n\t"
|
|
678 " jb 1b \n\t"
|
|
679 :: "r" (buf), "r" (delay_ptr)
|
|
680 : "%esi", "%edi"
|
|
681 );
|
3394
|
682 }
|
3579
|
683 #endif //arch_x86
|
3394
|
684
|
|
685 void
|
|
686 imdct_do_256(sample_t data[],sample_t delay[],sample_t bias)
|
|
687 {
|
|
688 int i,k;
|
|
689 int p,q;
|
|
690 int m;
|
|
691 int two_m;
|
|
692 int two_m_plus_one;
|
|
693
|
|
694 sample_t tmp_a_i;
|
|
695 sample_t tmp_a_r;
|
|
696 sample_t tmp_b_i;
|
|
697 sample_t tmp_b_r;
|
|
698
|
|
699 sample_t *data_ptr;
|
|
700 sample_t *delay_ptr;
|
|
701 sample_t *window_ptr;
|
|
702
|
|
703 complex_t *buf_1, *buf_2;
|
|
704
|
|
705 buf_1 = &buf[0];
|
|
706 buf_2 = &buf[64];
|
|
707
|
|
708 /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
|
|
709 for(k=0; k<64; k++) {
|
|
710 /* X1[k] = X[2*k] */
|
|
711 /* X2[k] = X[2*k+1] */
|
|
712
|
|
713 p = 2 * (128-2*k-1);
|
|
714 q = 2 * (2 * k);
|
|
715
|
|
716 /* Z1[k] = (X1[128-2*k-1] + j * X1[2*k]) * (xcos2[k] + j * xsin2[k]); */
|
|
717 buf_1[k].real = data[p] * xcos2[k] - data[q] * xsin2[k];
|
|
718 buf_1[k].imag = -1.0f * (data[q] * xcos2[k] + data[p] * xsin2[k]);
|
|
719 /* Z2[k] = (X2[128-2*k-1] + j * X2[2*k]) * (xcos2[k] + j * xsin2[k]); */
|
|
720 buf_2[k].real = data[p + 1] * xcos2[k] - data[q + 1] * xsin2[k];
|
|
721 buf_2[k].imag = -1.0f * ( data[q + 1] * xcos2[k] + data[p + 1] * xsin2[k]);
|
|
722 }
|
|
723
|
|
724 /* IFFT Bit reversed shuffling */
|
|
725 for(i=0; i<64; i++) {
|
|
726 k = bit_reverse_256[i];
|
|
727 if (k < i) {
|
|
728 swap_cmplx(&buf_1[i],&buf_1[k]);
|
|
729 swap_cmplx(&buf_2[i],&buf_2[k]);
|
|
730 }
|
|
731 }
|
|
732
|
|
733 /* FFT Merge */
|
|
734 for (m=0; m < 6; m++) {
|
|
735 two_m = (1 << m);
|
|
736 two_m_plus_one = (1 << (m+1));
|
|
737
|
|
738 /* FIXME */
|
|
739 if(m)
|
|
740 two_m = (1 << m);
|
|
741 else
|
|
742 two_m = 1;
|
|
743
|
|
744 for(k = 0; k < two_m; k++) {
|
|
745 for(i = 0; i < 64; i += two_m_plus_one) {
|
|
746 p = k + i;
|
|
747 q = p + two_m;
|
|
748 /* Do block 1 */
|
|
749 tmp_a_r = buf_1[p].real;
|
|
750 tmp_a_i = buf_1[p].imag;
|
|
751 tmp_b_r = buf_1[q].real * w[m][k].real - buf_1[q].imag * w[m][k].imag;
|
|
752 tmp_b_i = buf_1[q].imag * w[m][k].real + buf_1[q].real * w[m][k].imag;
|
|
753 buf_1[p].real = tmp_a_r + tmp_b_r;
|
|
754 buf_1[p].imag = tmp_a_i + tmp_b_i;
|
|
755 buf_1[q].real = tmp_a_r - tmp_b_r;
|
|
756 buf_1[q].imag = tmp_a_i - tmp_b_i;
|
|
757
|
|
758 /* Do block 2 */
|
|
759 tmp_a_r = buf_2[p].real;
|
|
760 tmp_a_i = buf_2[p].imag;
|
|
761 tmp_b_r = buf_2[q].real * w[m][k].real - buf_2[q].imag * w[m][k].imag;
|
|
762 tmp_b_i = buf_2[q].imag * w[m][k].real + buf_2[q].real * w[m][k].imag;
|
|
763 buf_2[p].real = tmp_a_r + tmp_b_r;
|
|
764 buf_2[p].imag = tmp_a_i + tmp_b_i;
|
|
765 buf_2[q].real = tmp_a_r - tmp_b_r;
|
|
766 buf_2[q].imag = tmp_a_i - tmp_b_i;
|
|
767 }
|
|
768 }
|
|
769 }
|
|
770
|
|
771 /* Post IFFT complex multiply */
|
|
772 for( i=0; i < 64; i++) {
|
|
773 /* y1[n] = z1[n] * (xcos2[n] + j * xs in2[n]) ; */
|
|
774 tmp_a_r = buf_1[i].real;
|
|
775 tmp_a_i = -buf_1[i].imag;
|
|
776 buf_1[i].real =(tmp_a_r * xcos2[i]) - (tmp_a_i * xsin2[i]);
|
|
777 buf_1[i].imag =(tmp_a_r * xsin2[i]) + (tmp_a_i * xcos2[i]);
|
|
778 /* y2[n] = z2[n] * (xcos2[n] + j * xsin2[n]) ; */
|
|
779 tmp_a_r = buf_2[i].real;
|
|
780 tmp_a_i = -buf_2[i].imag;
|
|
781 buf_2[i].real =(tmp_a_r * xcos2[i]) - (tmp_a_i * xsin2[i]);
|
|
782 buf_2[i].imag =(tmp_a_r * xsin2[i]) + (tmp_a_i * xcos2[i]);
|
|
783 }
|
|
784
|
|
785 data_ptr = data;
|
|
786 delay_ptr = delay;
|
|
787 window_ptr = imdct_window;
|
|
788
|
|
789 /* Window and convert to real valued signal */
|
|
790 for(i=0; i< 64; i++) {
|
|
791 *data_ptr++ = -buf_1[i].imag * *window_ptr++ + *delay_ptr++ + bias;
|
|
792 *data_ptr++ = buf_1[64-i-1].real * *window_ptr++ + *delay_ptr++ + bias;
|
|
793 }
|
|
794
|
|
795 for(i=0; i< 64; i++) {
|
|
796 *data_ptr++ = -buf_1[i].real * *window_ptr++ + *delay_ptr++ + bias;
|
|
797 *data_ptr++ = buf_1[64-i-1].imag * *window_ptr++ + *delay_ptr++ + bias;
|
|
798 }
|
|
799
|
|
800 delay_ptr = delay;
|
|
801
|
|
802 for(i=0; i< 64; i++) {
|
|
803 *delay_ptr++ = -buf_2[i].real * *--window_ptr;
|
|
804 *delay_ptr++ = buf_2[64-i-1].imag * *--window_ptr;
|
|
805 }
|
|
806
|
|
807 for(i=0; i< 64; i++) {
|
|
808 *delay_ptr++ = buf_2[i].imag * *--window_ptr;
|
|
809 *delay_ptr++ = -buf_2[64-i-1].real * *--window_ptr;
|
|
810 }
|
|
811 }
|
|
812
|
|
813 void imdct_init (uint32_t mm_accel)
|
|
814 {
|
|
815 #ifdef LIBA52_MLIB
|
|
816 if (mm_accel & MM_ACCEL_MLIB) {
|
|
817 fprintf (stderr, "Using mlib for IMDCT transform\n");
|
|
818 imdct_512 = imdct_do_512_mlib;
|
|
819 imdct_256 = imdct_do_256_mlib;
|
|
820 } else
|
|
821 #endif
|
|
822 {
|
|
823 int i, j, k;
|
|
824
|
3579
|
825 if(gCpuCaps.hasSSE) fprintf (stderr, "Using SSE optimized IMDCT transform\n");
|
|
826 else fprintf (stderr, "No accelerated IMDCT transform found\n");
|
3394
|
827
|
|
828 /* Twiddle factors to turn IFFT into IMDCT */
|
|
829 for (i = 0; i < 128; i++) {
|
|
830 xcos1[i] = -cos ((M_PI / 2048) * (8 * i + 1));
|
|
831 xsin1[i] = -sin ((M_PI / 2048) * (8 * i + 1));
|
|
832 }
|
3579
|
833 #ifdef ARCH_X86
|
3527
|
834 for (i = 0; i < 128; i++) {
|
3581
|
835 sseSinCos1c[2*i+0]= xcos1[i];
|
|
836 sseSinCos1c[2*i+1]= -xcos1[i];
|
|
837 sseSinCos1d[2*i+0]= xsin1[i];
|
|
838 sseSinCos1d[2*i+1]= xsin1[i];
|
3527
|
839 }
|
|
840 #endif
|
3394
|
841
|
|
842 /* More twiddle factors to turn IFFT into IMDCT */
|
|
843 for (i = 0; i < 64; i++) {
|
|
844 xcos2[i] = -cos ((M_PI / 1024) * (8 * i + 1));
|
|
845 xsin2[i] = -sin ((M_PI / 1024) * (8 * i + 1));
|
|
846 }
|
|
847
|
|
848 for (i = 0; i < 7; i++) {
|
|
849 j = 1 << i;
|
|
850 for (k = 0; k < j; k++) {
|
|
851 w[i][k].real = cos (-M_PI * k / j);
|
|
852 w[i][k].imag = sin (-M_PI * k / j);
|
|
853 }
|
|
854 }
|
3579
|
855 #ifdef ARCH_X86
|
3534
|
856 for (i = 1; i < 7; i++) {
|
|
857 j = 1 << i;
|
|
858 for (k = 0; k < j; k+=2) {
|
|
859
|
|
860 sseW[i][4*k + 0] = w[i][k+0].real;
|
|
861 sseW[i][4*k + 1] = w[i][k+0].real;
|
|
862 sseW[i][4*k + 2] = w[i][k+1].real;
|
|
863 sseW[i][4*k + 3] = w[i][k+1].real;
|
|
864
|
|
865 sseW[i][4*k + 4] = -w[i][k+0].imag;
|
|
866 sseW[i][4*k + 5] = w[i][k+0].imag;
|
|
867 sseW[i][4*k + 6] = -w[i][k+1].imag;
|
|
868 sseW[i][4*k + 7] = w[i][k+1].imag;
|
|
869
|
|
870 //we multiply more or less uninitalized numbers so we need to use exactly 0.0
|
|
871 if(k==0)
|
|
872 {
|
|
873 // sseW[i][4*k + 0]= sseW[i][4*k + 1]= 1.0;
|
|
874 sseW[i][4*k + 4]= sseW[i][4*k + 5]= 0.0;
|
|
875 }
|
|
876
|
|
877 if(2*k == j)
|
|
878 {
|
|
879 sseW[i][4*k + 0]= sseW[i][4*k + 1]= 0.0;
|
|
880 // sseW[i][4*k + 4]= -(sseW[i][4*k + 5]= -1.0);
|
|
881 }
|
|
882 }
|
|
883 }
|
3552
|
884
|
|
885 for(i=0; i<128; i++)
|
|
886 {
|
|
887 sseWindow[2*i+0]= -imdct_window[2*i+0];
|
3553
|
888 sseWindow[2*i+1]= imdct_window[2*i+1];
|
3552
|
889 }
|
3553
|
890
|
|
891 for(i=0; i<64; i++)
|
|
892 {
|
|
893 sseWindow[256 + 2*i+0]= -imdct_window[254 - 2*i+1];
|
|
894 sseWindow[256 + 2*i+1]= imdct_window[254 - 2*i+0];
|
|
895 sseWindow[384 + 2*i+0]= imdct_window[126 - 2*i+1];
|
|
896 sseWindow[384 + 2*i+1]= -imdct_window[126 - 2*i+0];
|
|
897 }
|
3579
|
898 #endif // arch_x86
|
|
899
|
|
900 if(gCpuCaps.hasSSE) imdct_512 = imdct_do_512_sse;
|
|
901 else imdct_512 = imdct_do_512;
|
3394
|
902 imdct_256 = imdct_do_256;
|
|
903 }
|
|
904 }
|