4 #include "ac3_decoder.h"
5 #include "ac3_internal.h"
7 void imdct_do_256(float x[],float y[],float delay[]);
8 void imdct_do_512(float x[],float y[],float delay[]);
10 typedef struct complex_s {
17 static complex_t buf[N/4];
19 /* 128 point bit-reverse LUT */
20 static u8 bit_reverse_512[] = {
21 0x00, 0x40, 0x20, 0x60, 0x10, 0x50, 0x30, 0x70,
22 0x08, 0x48, 0x28, 0x68, 0x18, 0x58, 0x38, 0x78,
23 0x04, 0x44, 0x24, 0x64, 0x14, 0x54, 0x34, 0x74,
24 0x0c, 0x4c, 0x2c, 0x6c, 0x1c, 0x5c, 0x3c, 0x7c,
25 0x02, 0x42, 0x22, 0x62, 0x12, 0x52, 0x32, 0x72,
26 0x0a, 0x4a, 0x2a, 0x6a, 0x1a, 0x5a, 0x3a, 0x7a,
27 0x06, 0x46, 0x26, 0x66, 0x16, 0x56, 0x36, 0x76,
28 0x0e, 0x4e, 0x2e, 0x6e, 0x1e, 0x5e, 0x3e, 0x7e,
29 0x01, 0x41, 0x21, 0x61, 0x11, 0x51, 0x31, 0x71,
30 0x09, 0x49, 0x29, 0x69, 0x19, 0x59, 0x39, 0x79,
31 0x05, 0x45, 0x25, 0x65, 0x15, 0x55, 0x35, 0x75,
32 0x0d, 0x4d, 0x2d, 0x6d, 0x1d, 0x5d, 0x3d, 0x7d,
33 0x03, 0x43, 0x23, 0x63, 0x13, 0x53, 0x33, 0x73,
34 0x0b, 0x4b, 0x2b, 0x6b, 0x1b, 0x5b, 0x3b, 0x7b,
35 0x07, 0x47, 0x27, 0x67, 0x17, 0x57, 0x37, 0x77,
36 0x0f, 0x4f, 0x2f, 0x6f, 0x1f, 0x5f, 0x3f, 0x7f};
38 static u8 bit_reverse_256[] = {
39 0x00, 0x20, 0x10, 0x30, 0x08, 0x28, 0x18, 0x38,
40 0x04, 0x24, 0x14, 0x34, 0x0c, 0x2c, 0x1c, 0x3c,
41 0x02, 0x22, 0x12, 0x32, 0x0a, 0x2a, 0x1a, 0x3a,
42 0x06, 0x26, 0x16, 0x36, 0x0e, 0x2e, 0x1e, 0x3e,
43 0x01, 0x21, 0x11, 0x31, 0x09, 0x29, 0x19, 0x39,
44 0x05, 0x25, 0x15, 0x35, 0x0d, 0x2d, 0x1d, 0x3d,
45 0x03, 0x23, 0x13, 0x33, 0x0b, 0x2b, 0x1b, 0x3b,
46 0x07, 0x27, 0x17, 0x37, 0x0f, 0x2f, 0x1f, 0x3f};
48 /* Twiddle factor LUT */
49 static complex_t *w[7];
50 static complex_t w_1[1];
51 static complex_t w_2[2];
52 static complex_t w_4[4];
53 static complex_t w_8[8];
54 static complex_t w_16[16];
55 static complex_t w_32[32];
56 static complex_t w_64[64];
58 /* Twiddle factors for IMDCT */
59 static float xcos1[N/4];
60 static float xsin1[N/4];
61 static float xcos2[N/8];
62 static float xsin2[N/8];
64 /* Delay buffer for time domain interleaving */
65 static float delay[6][256];
67 /* Windowing function for Modified DCT - Thank you acroread */
68 static float window[] = {
69 0.00014, 0.00024, 0.00037, 0.00051, 0.00067, 0.00086, 0.00107, 0.00130,
70 0.00157, 0.00187, 0.00220, 0.00256, 0.00297, 0.00341, 0.00390, 0.00443,
71 0.00501, 0.00564, 0.00632, 0.00706, 0.00785, 0.00871, 0.00962, 0.01061,
72 0.01166, 0.01279, 0.01399, 0.01526, 0.01662, 0.01806, 0.01959, 0.02121,
73 0.02292, 0.02472, 0.02662, 0.02863, 0.03073, 0.03294, 0.03527, 0.03770,
74 0.04025, 0.04292, 0.04571, 0.04862, 0.05165, 0.05481, 0.05810, 0.06153,
75 0.06508, 0.06878, 0.07261, 0.07658, 0.08069, 0.08495, 0.08935, 0.09389,
76 0.09859, 0.10343, 0.10842, 0.11356, 0.11885, 0.12429, 0.12988, 0.13563,
77 0.14152, 0.14757, 0.15376, 0.16011, 0.16661, 0.17325, 0.18005, 0.18699,
78 0.19407, 0.20130, 0.20867, 0.21618, 0.22382, 0.23161, 0.23952, 0.24757,
79 0.25574, 0.26404, 0.27246, 0.28100, 0.28965, 0.29841, 0.30729, 0.31626,
80 0.32533, 0.33450, 0.34376, 0.35311, 0.36253, 0.37204, 0.38161, 0.39126,
81 0.40096, 0.41072, 0.42054, 0.43040, 0.44030, 0.45023, 0.46020, 0.47019,
82 0.48020, 0.49022, 0.50025, 0.51028, 0.52031, 0.53033, 0.54033, 0.55031,
83 0.56026, 0.57019, 0.58007, 0.58991, 0.59970, 0.60944, 0.61912, 0.62873,
84 0.63827, 0.64774, 0.65713, 0.66643, 0.67564, 0.68476, 0.69377, 0.70269,
85 0.71150, 0.72019, 0.72877, 0.73723, 0.74557, 0.75378, 0.76186, 0.76981,
86 0.77762, 0.78530, 0.79283, 0.80022, 0.80747, 0.81457, 0.82151, 0.82831,
87 0.83496, 0.84145, 0.84779, 0.85398, 0.86001, 0.86588, 0.87160, 0.87716,
88 0.88257, 0.88782, 0.89291, 0.89785, 0.90264, 0.90728, 0.91176, 0.91610,
89 0.92028, 0.92432, 0.92822, 0.93197, 0.93558, 0.93906, 0.94240, 0.94560,
90 0.94867, 0.95162, 0.95444, 0.95713, 0.95971, 0.96217, 0.96451, 0.96674,
91 0.96887, 0.97089, 0.97281, 0.97463, 0.97635, 0.97799, 0.97953, 0.98099,
92 0.98236, 0.98366, 0.98488, 0.98602, 0.98710, 0.98811, 0.98905, 0.98994,
93 0.99076, 0.99153, 0.99225, 0.99291, 0.99353, 0.99411, 0.99464, 0.99513,
94 0.99558, 0.99600, 0.99639, 0.99674, 0.99706, 0.99736, 0.99763, 0.99788,
95 0.99811, 0.99831, 0.99850, 0.99867, 0.99882, 0.99895, 0.99908, 0.99919,
96 0.99929, 0.99938, 0.99946, 0.99953, 0.99959, 0.99965, 0.99969, 0.99974,
97 0.99978, 0.99981, 0.99984, 0.99986, 0.99988, 0.99990, 0.99992, 0.99993,
98 0.99994, 0.99995, 0.99996, 0.99997, 0.99998, 0.99998, 0.99998, 0.99999,
99 0.99999, 0.99999, 0.99999, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000,
100 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000 };
102 static __inline__ void swap_cmplx(complex_t *a, complex_t *b)
111 static __inline__ complex_t cmplx_mult(complex_t a, complex_t b)
115 ret.real = a.real * b.real - a.imag * b.imag;
116 ret.imag = a.real * b.imag + a.imag * b.real;
121 static void imdct_init(void) __attribute__ ((__constructor__));
122 static void imdct_init(void)
125 complex_t angle_step;
126 complex_t current_angle;
128 /* Twiddle factors to turn IFFT into IMDCT */
129 for (i=0; i < N/4; i++) {
130 xcos1[i] = -cos(2 * M_PI * (8*i+1)/(8*N)) ;
131 xsin1[i] = -sin(2 * M_PI * (8*i+1)/(8*N)) ;
134 /* More twiddle factors to turn IFFT into IMDCT */
135 for (i=0; i < N/8; i++) {
136 xcos2[i] = -cos(2 * M_PI * (8*i+1)/(4*N)) ;
137 xsin2[i] = -sin(2 * M_PI * (8*i+1)/(4*N)) ;
140 /* Canonical twiddle factors for FFT */
149 for (i = 0; i < 7; i++) {
150 angle_step.real = cos(-2.0f * M_PI / (1 << (i+1)));
151 angle_step.imag = sin(-2.0f * M_PI / (1 << (i+1)));
153 current_angle.real = 1.0f;
154 current_angle.imag = 0.0f;
156 for (k = 0; k < 1 << i; k++) {
157 w[i][k] = current_angle;
158 current_angle = cmplx_mult(current_angle,angle_step);
163 void imdct (ac3dec_t * p_ac3dec)
167 for (i=0; i<p_ac3dec->bsi.nfchans;i++) {
168 if (p_ac3dec->audblk.blksw[i])
169 imdct_do_256(p_ac3dec->coeffs.fbw[i],p_ac3dec->samples.channel[i],delay[i]);
171 imdct_do_512(p_ac3dec->coeffs.fbw[i],p_ac3dec->samples.channel[i],delay[i]);
174 /* XXX?? We don't bother with the IMDCT for the LFE as it's currently
177 // imdct_do_512(coeffs->lfe,samples->channel[5],delay[5]);
181 imdct_do_512(float x[],float y[],float delay[])
199 /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
200 for (i=0; i < N/4; i++) {
201 /* z[i] = (X[N/2-2*i-1] + j * X[2*i]) * (xcos1[i] + j * xsin1[i]) ; */
202 buf[i].real = (x[N/2-2*i-1] * xcos1[i]) - (x[2*i] * xsin1[i]);
203 buf[i].imag = -((x[2*i] * xcos1[i]) + (x[N/2-2*i-1] * xsin1[i]));
206 /* Bit reversed shuffling */
207 for (i=0; i<N/4; i++) {
208 k = bit_reverse_512[i];
210 swap_cmplx(&buf[i],&buf[k]);
214 for (m=0; m < 7; m++) {
216 two_m_plus_one = (1 << (m+1));
218 for (k = 0; k < two_m; k++) {
219 for (i = 0; i < 128; i += two_m_plus_one) {
222 tmp_a_r = buf[p].real;
223 tmp_a_i = buf[p].imag;
224 tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
225 tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
226 buf[p].real = tmp_a_r + tmp_b_r;
227 buf[p].imag = tmp_a_i + tmp_b_i;
228 buf[q].real = tmp_a_r - tmp_b_r;
229 buf[q].imag = tmp_a_i - tmp_b_i;
234 /* Post IFFT complex multiply plus IFFT complex conjugate*/
235 for (i=0; i < N/4; i++) {
236 /* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
237 tmp_a_r = buf[i].real;
238 tmp_a_i = - buf[i].imag;
239 buf[i].real =(tmp_a_r * xcos1[i]) - (tmp_a_i * xsin1[i]);
240 buf[i].imag =(tmp_a_r * xsin1[i]) + (tmp_a_i * xcos1[i]);
246 /* Window and convert to real valued signal */
247 for (i=0; i<N/8; i++) {
248 *y_ptr++ = 2.0f * (-buf[N/8+i].imag * *window_ptr++ + *delay_ptr++);
249 *y_ptr++ = 2.0f * (buf[N/8-i-1].real * *window_ptr++ + *delay_ptr++);
252 for (i=0; i<N/8; i++) {
253 *y_ptr++ = 2.0f * (-buf[i].real * *window_ptr++ + *delay_ptr++);
254 *y_ptr++ = 2.0f * (buf[N/4-i-1].imag * *window_ptr++ + *delay_ptr++);
257 /* The trailing edge of the window goes into the delay line */
260 for (i=0; i<N/8; i++) {
261 *delay_ptr++ = -buf[N/8+i].real * *--window_ptr;
262 *delay_ptr++ = buf[N/8-i-1].imag * *--window_ptr;
265 for (i=0; i<N/8; i++) {
266 *delay_ptr++ = buf[i].imag * *--window_ptr;
267 *delay_ptr++ = -buf[N/4-i-1].real * *--window_ptr;
272 imdct_do_256(float x[],float y[],float delay[])
285 complex_t *buf_1, *buf_2;
290 /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
291 for (k=0; k<N/8; k++) {
293 /* X2[k] = X[2*k+1] */
298 /* Z1[k] = (X1[N/4-2*k-1] + j * X1[2*k]) * (xcos2[k] + j * xsin2[k]); */
299 buf_1[k].real = x[p] * xcos2[k] - x[q] * xsin2[k];
300 buf_1[k].imag = - (x[q] * xcos2[k] + x[p] * xsin2[k]);
301 /* Z2[k] = (X2[N/4-2*k-1] + j * X2[2*k]) * (xcos2[k] + j * xsin2[k]); */
302 buf_2[k].real = x[p + 1] * xcos2[k] - x[q + 1] * xsin2[k];
303 buf_2[k].imag = - (x[q + 1] * xcos2[k] + x[p + 1] * xsin2[k]);
306 /* IFFT Bit reversed shuffling */
307 for (i=0; i<N/8; i++) {
308 k = bit_reverse_256[i];
310 swap_cmplx(&buf_1[i],&buf_1[k]);
311 swap_cmplx(&buf_2[i],&buf_2[k]);
316 for (m=0; m < 6; m++) {
318 two_m_plus_one = (1 << (m+1));
320 for (k = 0; k < two_m; k++) {
321 for (i = 0; i < 64; i += two_m_plus_one) {
325 tmp_a_r = buf_1[p].real;
326 tmp_a_i = buf_1[p].imag;
327 tmp_b_r = buf_1[q].real * w[m][k].real - buf_1[q].imag * w[m][k].imag;
328 tmp_b_i = buf_1[q].imag * w[m][k].real + buf_1[q].real * w[m][k].imag;
329 buf_1[p].real = tmp_a_r + tmp_b_r;
330 buf_1[p].imag = tmp_a_i + tmp_b_i;
331 buf_1[q].real = tmp_a_r - tmp_b_r;
332 buf_1[q].imag = tmp_a_i - tmp_b_i;
335 tmp_a_r = buf_2[p].real;
336 tmp_a_i = buf_2[p].imag;
337 tmp_b_r = buf_2[q].real * w[m][k].real - buf_2[q].imag * w[m][k].imag;
338 tmp_b_i = buf_2[q].imag * w[m][k].real + buf_2[q].real * w[m][k].imag;
339 buf_2[p].real = tmp_a_r + tmp_b_r;
340 buf_2[p].imag = tmp_a_i + tmp_b_i;
341 buf_2[q].real = tmp_a_r - tmp_b_r;
342 buf_2[q].imag = tmp_a_i - tmp_b_i;
347 /* Post IFFT complex multiply */
348 for (i=0; i < N/8; i++) {
349 /* y1[n] = z1[n] * (xcos2[n] + j * xs in2[n]) ; */
350 tmp_a_r = buf_1[i].real;
351 tmp_a_i = - buf_1[i].imag;
352 buf_1[i].real =(tmp_a_r * xcos2[i]) - (tmp_a_i * xsin2[i]);
353 buf_1[i].imag =(tmp_a_r * xsin2[i]) + (tmp_a_i * xcos2[i]);
354 /* y2[n] = z2[n] * (xcos2[n] + j * xsin2[n]) ; */
355 tmp_a_r = buf_2[i].real;
356 tmp_a_i = - buf_2[i].imag;
357 buf_2[i].real =(tmp_a_r * xcos2[i]) - (tmp_a_i * xsin2[i]);
358 buf_2[i].imag =(tmp_a_r * xsin2[i]) + (tmp_a_i * xcos2[i]);
361 /* Window and convert to real valued signal */
362 for (i=0; i<N/8; i++) {
363 y[2*i] = -buf_1[i].imag * window[2*i];
364 y[2*i+1] = buf_1[N/8-i-1].real * window[2*i+1];
365 y[N/4+2*i] = -buf_1[i].real * window[N/4+2*i];
366 y[N/4+2*i+1] = buf_1[N/8-i-1].imag * window[N/4+2*i+1];
367 y[N/2+2*i] = -buf_2[i].real * window[N/2-2*i-1];
368 y[N/2+2*i+1] = buf_2[N/8-i-1].imag * window[N/2-2*i-2];
369 y[3*N/4+2*i] = buf_2[i].imag * window[N/4-2*i-1];
370 y[3*N/4+2*i+1] = -buf_2[N/8-i-1].real * window[N/4-2*i-2];
373 /* Overlap and add */
374 for (i=0; i<N/2; i++) {
375 y[i] = 2 * (y[i] + delay[i]);