1/*
2 * AC-3 DSP functions
3 * Copyright (c) 2011 Justin Ruggles
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22#include "libavutil/avassert.h"
23#include "avcodec.h"
24#include "ac3.h"
25#include "ac3dsp.h"
26#include "mathops.h"
27
28static void ac3_exponent_min_c(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
29{
30    int blk, i;
31
32    if (!num_reuse_blocks)
33        return;
34
35    for (i = 0; i < nb_coefs; i++) {
36        uint8_t min_exp = *exp;
37        uint8_t *exp1 = exp + 256;
38        for (blk = 0; blk < num_reuse_blocks; blk++) {
39            uint8_t next_exp = *exp1;
40            if (next_exp < min_exp)
41                min_exp = next_exp;
42            exp1 += 256;
43        }
44        *exp++ = min_exp;
45    }
46}
47
48static int ac3_max_msb_abs_int16_c(const int16_t *src, int len)
49{
50    int i, v = 0;
51    for (i = 0; i < len; i++)
52        v |= abs(src[i]);
53    return v;
54}
55
56static void ac3_lshift_int16_c(int16_t *src, unsigned int len,
57                               unsigned int shift)
58{
59    uint32_t *src32 = (uint32_t *)src;
60    const uint32_t mask = ~(((1 << shift) - 1) << 16);
61    int i;
62    len >>= 1;
63    for (i = 0; i < len; i += 8) {
64        src32[i  ] = (src32[i  ] << shift) & mask;
65        src32[i+1] = (src32[i+1] << shift) & mask;
66        src32[i+2] = (src32[i+2] << shift) & mask;
67        src32[i+3] = (src32[i+3] << shift) & mask;
68        src32[i+4] = (src32[i+4] << shift) & mask;
69        src32[i+5] = (src32[i+5] << shift) & mask;
70        src32[i+6] = (src32[i+6] << shift) & mask;
71        src32[i+7] = (src32[i+7] << shift) & mask;
72    }
73}
74
75static void ac3_rshift_int32_c(int32_t *src, unsigned int len,
76                               unsigned int shift)
77{
78    do {
79        *src++ >>= shift;
80        *src++ >>= shift;
81        *src++ >>= shift;
82        *src++ >>= shift;
83        *src++ >>= shift;
84        *src++ >>= shift;
85        *src++ >>= shift;
86        *src++ >>= shift;
87        len -= 8;
88    } while (len > 0);
89}
90
91static void float_to_fixed24_c(int32_t *dst, const float *src, unsigned int len)
92{
93    const float scale = 1 << 24;
94    do {
95        *dst++ = lrintf(*src++ * scale);
96        *dst++ = lrintf(*src++ * scale);
97        *dst++ = lrintf(*src++ * scale);
98        *dst++ = lrintf(*src++ * scale);
99        *dst++ = lrintf(*src++ * scale);
100        *dst++ = lrintf(*src++ * scale);
101        *dst++ = lrintf(*src++ * scale);
102        *dst++ = lrintf(*src++ * scale);
103        len -= 8;
104    } while (len > 0);
105}
106
107static void ac3_bit_alloc_calc_bap_c(int16_t *mask, int16_t *psd,
108                                     int start, int end,
109                                     int snr_offset, int floor,
110                                     const uint8_t *bap_tab, uint8_t *bap)
111{
112    int bin, band, band_end;
113
114    /* special case, if snr offset is -960, set all bap's to zero */
115    if (snr_offset == -960) {
116        memset(bap, 0, AC3_MAX_COEFS);
117        return;
118    }
119
120    bin  = start;
121    band = ff_ac3_bin_to_band_tab[start];
122    do {
123        int m = (FFMAX(mask[band] - snr_offset - floor, 0) & 0x1FE0) + floor;
124        band_end = ff_ac3_band_start_tab[++band];
125        band_end = FFMIN(band_end, end);
126
127        for (; bin < band_end; bin++) {
128            int address = av_clip((psd[bin] - m) >> 5, 0, 63);
129            bap[bin] = bap_tab[address];
130        }
131    } while (end > band_end);
132}
133
134static void ac3_update_bap_counts_c(uint16_t mant_cnt[16], uint8_t *bap,
135                                    int len)
136{
137    while (len-- > 0)
138        mant_cnt[bap[len]]++;
139}
140
141DECLARE_ALIGNED(16, const uint16_t, ff_ac3_bap_bits)[16] = {
142    0,  0,  0,  3,  0,  4,  5,  6,  7,  8,  9, 10, 11, 12, 14, 16
143};
144
145static int ac3_compute_mantissa_size_c(uint16_t mant_cnt[6][16])
146{
147    int blk, bap;
148    int bits = 0;
149
150    for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
151        // bap=1 : 3 mantissas in 5 bits
152        bits += (mant_cnt[blk][1] / 3) * 5;
153        // bap=2 : 3 mantissas in 7 bits
154        // bap=4 : 2 mantissas in 7 bits
155        bits += ((mant_cnt[blk][2] / 3) + (mant_cnt[blk][4] >> 1)) * 7;
156        // bap=3 : 1 mantissa in 3 bits
157        bits += mant_cnt[blk][3] * 3;
158        // bap=5 to 15 : get bits per mantissa from table
159        for (bap = 5; bap < 16; bap++)
160            bits += mant_cnt[blk][bap] * ff_ac3_bap_bits[bap];
161    }
162    return bits;
163}
164
165static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs)
166{
167    int i;
168
169    for (i = 0; i < nb_coefs; i++) {
170        int v = abs(coef[i]);
171        exp[i] = v ? 23 - av_log2(v) : 24;
172    }
173}
174
175static void ac3_sum_square_butterfly_int32_c(int64_t sum[4],
176                                             const int32_t *coef0,
177                                             const int32_t *coef1,
178                                             int len)
179{
180    int i;
181
182    sum[0] = sum[1] = sum[2] = sum[3] = 0;
183
184    for (i = 0; i < len; i++) {
185        int lt = coef0[i];
186        int rt = coef1[i];
187        int md = lt + rt;
188        int sd = lt - rt;
189        MAC64(sum[0], lt, lt);
190        MAC64(sum[1], rt, rt);
191        MAC64(sum[2], md, md);
192        MAC64(sum[3], sd, sd);
193    }
194}
195
196static void ac3_sum_square_butterfly_float_c(float sum[4],
197                                             const float *coef0,
198                                             const float *coef1,
199                                             int len)
200{
201    int i;
202
203    sum[0] = sum[1] = sum[2] = sum[3] = 0;
204
205    for (i = 0; i < len; i++) {
206        float lt = coef0[i];
207        float rt = coef1[i];
208        float md = lt + rt;
209        float sd = lt - rt;
210        sum[0] += lt * lt;
211        sum[1] += rt * rt;
212        sum[2] += md * md;
213        sum[3] += sd * sd;
214    }
215}
216
217static void ac3_downmix_c(float **samples, float (*matrix)[2],
218                          int out_ch, int in_ch, int len)
219{
220    int i, j;
221    float v0, v1;
222    if (out_ch == 2) {
223        for (i = 0; i < len; i++) {
224            v0 = v1 = 0.0f;
225            for (j = 0; j < in_ch; j++) {
226                v0 += samples[j][i] * matrix[j][0];
227                v1 += samples[j][i] * matrix[j][1];
228            }
229            samples[0][i] = v0;
230            samples[1][i] = v1;
231        }
232    } else if (out_ch == 1) {
233        for (i = 0; i < len; i++) {
234            v0 = 0.0f;
235            for (j = 0; j < in_ch; j++)
236                v0 += samples[j][i] * matrix[j][0];
237            samples[0][i] = v0;
238        }
239    }
240}
241
242static void ac3_downmix_c_fixed(int32_t **samples, int16_t (*matrix)[2],
243                                int out_ch, int in_ch, int len)
244{
245    int i, j;
246    int64_t v0, v1;
247    if (out_ch == 2) {
248        for (i = 0; i < len; i++) {
249            v0 = v1 = 0;
250            for (j = 0; j < in_ch; j++) {
251                v0 += (int64_t)samples[j][i] * matrix[j][0];
252                v1 += (int64_t)samples[j][i] * matrix[j][1];
253            }
254            samples[0][i] = (v0+2048)>>12;
255            samples[1][i] = (v1+2048)>>12;
256        }
257    } else if (out_ch == 1) {
258        for (i = 0; i < len; i++) {
259            v0 = 0;
260            for (j = 0; j < in_ch; j++)
261                v0 += (int64_t)samples[j][i] * matrix[j][0];
262            samples[0][i] = (v0+2048)>>12;
263        }
264    }
265}
266
267static void apply_window_int16_c(int16_t *output, const int16_t *input,
268                                 const int16_t *window, unsigned int len)
269{
270    int i;
271    int len2 = len >> 1;
272
273    for (i = 0; i < len2; i++) {
274        int16_t w       = window[i];
275        output[i]       = (MUL16(input[i],       w) + (1 << 14)) >> 15;
276        output[len-i-1] = (MUL16(input[len-i-1], w) + (1 << 14)) >> 15;
277    }
278}
279
280av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact)
281{
282    c->ac3_exponent_min = ac3_exponent_min_c;
283    c->ac3_max_msb_abs_int16 = ac3_max_msb_abs_int16_c;
284    c->ac3_lshift_int16 = ac3_lshift_int16_c;
285    c->ac3_rshift_int32 = ac3_rshift_int32_c;
286    c->float_to_fixed24 = float_to_fixed24_c;
287    c->bit_alloc_calc_bap = ac3_bit_alloc_calc_bap_c;
288    c->update_bap_counts = ac3_update_bap_counts_c;
289    c->compute_mantissa_size = ac3_compute_mantissa_size_c;
290    c->extract_exponents = ac3_extract_exponents_c;
291    c->sum_square_butterfly_int32 = ac3_sum_square_butterfly_int32_c;
292    c->sum_square_butterfly_float = ac3_sum_square_butterfly_float_c;
293    c->downmix = ac3_downmix_c;
294    c->downmix_fixed = ac3_downmix_c_fixed;
295    c->apply_window_int16 = apply_window_int16_c;
296
297    if (ARCH_ARM)
298        ff_ac3dsp_init_arm(c, bit_exact);
299    if (ARCH_X86)
300        ff_ac3dsp_init_x86(c, bit_exact);
301    if (ARCH_MIPS)
302        ff_ac3dsp_init_mips(c, bit_exact);
303}
304