blob: bef4c103f382a663a07678ce592ce25bc9a815cf
1 | /* |
2 | * AAC encoder utilities |
3 | * Copyright (C) 2015 Rostislav Pehlivanov |
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 | /** |
23 | * @file |
24 | * AAC encoder utilities |
25 | * @author Rostislav Pehlivanov ( atomnuker gmail com ) |
26 | */ |
27 | |
28 | #ifndef AVCODEC_AACENC_UTILS_H |
29 | #define AVCODEC_AACENC_UTILS_H |
30 | |
31 | #include "libavutil/ffmath.h" |
32 | #include "aac.h" |
33 | #include "aacenctab.h" |
34 | #include "aactab.h" |
35 | |
36 | #define ROUND_STANDARD 0.4054f |
37 | #define ROUND_TO_ZERO 0.1054f |
38 | #define C_QUANT 0.4054f |
39 | |
40 | static inline void abs_pow34_v(float *out, const float *in, const int size) |
41 | { |
42 | int i; |
43 | for (i = 0; i < size; i++) { |
44 | float a = fabsf(in[i]); |
45 | out[i] = sqrtf(a * sqrtf(a)); |
46 | } |
47 | } |
48 | |
49 | static inline float pos_pow34(float a) |
50 | { |
51 | return sqrtf(a * sqrtf(a)); |
52 | } |
53 | |
54 | /** |
55 | * Quantize one coefficient. |
56 | * @return absolute value of the quantized coefficient |
57 | * @see 3GPP TS26.403 5.6.2 "Scalefactor determination" |
58 | */ |
59 | static inline int quant(float coef, const float Q, const float rounding) |
60 | { |
61 | float a = coef * Q; |
62 | return sqrtf(a * sqrtf(a)) + rounding; |
63 | } |
64 | |
65 | static inline void quantize_bands(int *out, const float *in, const float *scaled, |
66 | int size, int is_signed, int maxval, const float Q34, |
67 | const float rounding) |
68 | { |
69 | int i; |
70 | for (i = 0; i < size; i++) { |
71 | float qc = scaled[i] * Q34; |
72 | int tmp = (int)FFMIN(qc + rounding, (float)maxval); |
73 | if (is_signed && in[i] < 0.0f) { |
74 | tmp = -tmp; |
75 | } |
76 | out[i] = tmp; |
77 | } |
78 | } |
79 | |
80 | static inline float find_max_val(int group_len, int swb_size, const float *scaled) |
81 | { |
82 | float maxval = 0.0f; |
83 | int w2, i; |
84 | for (w2 = 0; w2 < group_len; w2++) { |
85 | for (i = 0; i < swb_size; i++) { |
86 | maxval = FFMAX(maxval, scaled[w2*128+i]); |
87 | } |
88 | } |
89 | return maxval; |
90 | } |
91 | |
92 | static inline int find_min_book(float maxval, int sf) |
93 | { |
94 | float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - sf + SCALE_ONE_POS - SCALE_DIV_512]; |
95 | int qmaxval, cb; |
96 | qmaxval = maxval * Q34 + C_QUANT; |
97 | if (qmaxval >= (FF_ARRAY_ELEMS(aac_maxval_cb))) |
98 | cb = 11; |
99 | else |
100 | cb = aac_maxval_cb[qmaxval]; |
101 | return cb; |
102 | } |
103 | |
104 | static inline float find_form_factor(int group_len, int swb_size, float thresh, |
105 | const float *scaled, float nzslope) { |
106 | const float iswb_size = 1.0f / swb_size; |
107 | const float iswb_sizem1 = 1.0f / (swb_size - 1); |
108 | const float ethresh = thresh; |
109 | float form = 0.0f, weight = 0.0f; |
110 | int w2, i; |
111 | for (w2 = 0; w2 < group_len; w2++) { |
112 | float e = 0.0f, e2 = 0.0f, var = 0.0f, maxval = 0.0f; |
113 | float nzl = 0; |
114 | for (i = 0; i < swb_size; i++) { |
115 | float s = fabsf(scaled[w2*128+i]); |
116 | maxval = FFMAX(maxval, s); |
117 | e += s; |
118 | e2 += s *= s; |
119 | /* We really don't want a hard non-zero-line count, since |
120 | * even below-threshold lines do add up towards band spectral power. |
121 | * So, fall steeply towards zero, but smoothly |
122 | */ |
123 | if (s >= ethresh) { |
124 | nzl += 1.0f; |
125 | } else { |
126 | if (nzslope == 2.f) |
127 | nzl += (s / ethresh) * (s / ethresh); |
128 | else |
129 | nzl += ff_fast_powf(s / ethresh, nzslope); |
130 | } |
131 | } |
132 | if (e2 > thresh) { |
133 | float frm; |
134 | e *= iswb_size; |
135 | |
136 | /** compute variance */ |
137 | for (i = 0; i < swb_size; i++) { |
138 | float d = fabsf(scaled[w2*128+i]) - e; |
139 | var += d*d; |
140 | } |
141 | var = sqrtf(var * iswb_sizem1); |
142 | |
143 | e2 *= iswb_size; |
144 | frm = e / FFMIN(e+4*var,maxval); |
145 | form += e2 * sqrtf(frm) / FFMAX(0.5f,nzl); |
146 | weight += e2; |
147 | } |
148 | } |
149 | if (weight > 0) { |
150 | return form / weight; |
151 | } else { |
152 | return 1.0f; |
153 | } |
154 | } |
155 | |
156 | /** Return the minimum scalefactor where the quantized coef does not clip. */ |
157 | static inline uint8_t coef2minsf(float coef) |
158 | { |
159 | return av_clip_uint8(log2f(coef)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512); |
160 | } |
161 | |
162 | /** Return the maximum scalefactor where the quantized coef is not zero. */ |
163 | static inline uint8_t coef2maxsf(float coef) |
164 | { |
165 | return av_clip_uint8(log2f(coef)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512); |
166 | } |
167 | |
168 | /* |
169 | * Returns the closest possible index to an array of float values, given a value. |
170 | */ |
171 | static inline int quant_array_idx(const float val, const float *arr, const int num) |
172 | { |
173 | int i, index = 0; |
174 | float quant_min_err = INFINITY; |
175 | for (i = 0; i < num; i++) { |
176 | float error = (val - arr[i])*(val - arr[i]); |
177 | if (error < quant_min_err) { |
178 | quant_min_err = error; |
179 | index = i; |
180 | } |
181 | } |
182 | return index; |
183 | } |
184 | |
185 | /** |
186 | * approximates exp10f(-3.0f*(0.5f + 0.5f * cosf(FFMIN(b,15.5f) / 15.5f))) |
187 | */ |
188 | static av_always_inline float bval2bmax(float b) |
189 | { |
190 | return 0.001f + 0.0035f * (b*b*b) / (15.5f*15.5f*15.5f); |
191 | } |
192 | |
193 | /* |
194 | * Compute a nextband map to be used with SF delta constraint utilities. |
195 | * The nextband array should contain 128 elements, and positions that don't |
196 | * map to valid, nonzero bands of the form w*16+g (with w being the initial |
197 | * window of the window group, only) are left indetermined. |
198 | */ |
199 | static inline void ff_init_nextband_map(const SingleChannelElement *sce, uint8_t *nextband) |
200 | { |
201 | unsigned char prevband = 0; |
202 | int w, g; |
203 | /** Just a safe default */ |
204 | for (g = 0; g < 128; g++) |
205 | nextband[g] = g; |
206 | |
207 | /** Now really navigate the nonzero band chain */ |
208 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { |
209 | for (g = 0; g < sce->ics.num_swb; g++) { |
210 | if (!sce->zeroes[w*16+g] && sce->band_type[w*16+g] < RESERVED_BT) |
211 | prevband = nextband[prevband] = w*16+g; |
212 | } |
213 | } |
214 | nextband[prevband] = prevband; /* terminate */ |
215 | } |
216 | |
217 | /* |
218 | * Updates nextband to reflect a removed band (equivalent to |
219 | * calling ff_init_nextband_map after marking a band as zero) |
220 | */ |
221 | static inline void ff_nextband_remove(uint8_t *nextband, int prevband, int band) |
222 | { |
223 | nextband[prevband] = nextband[band]; |
224 | } |
225 | |
226 | /* |
227 | * Checks whether the specified band could be removed without inducing |
228 | * scalefactor delta that violates SF delta encoding constraints. |
229 | * prev_sf has to be the scalefactor of the previous nonzero, nonspecial |
230 | * band, in encoding order, or negative if there was no such band. |
231 | */ |
232 | static inline int ff_sfdelta_can_remove_band(const SingleChannelElement *sce, |
233 | const uint8_t *nextband, int prev_sf, int band) |
234 | { |
235 | return prev_sf >= 0 |
236 | && sce->sf_idx[nextband[band]] >= (prev_sf - SCALE_MAX_DIFF) |
237 | && sce->sf_idx[nextband[band]] <= (prev_sf + SCALE_MAX_DIFF); |
238 | } |
239 | |
240 | /* |
241 | * Checks whether the specified band's scalefactor could be replaced |
242 | * with another one without violating SF delta encoding constraints. |
243 | * prev_sf has to be the scalefactor of the previous nonzero, nonsepcial |
244 | * band, in encoding order, or negative if there was no such band. |
245 | */ |
246 | static inline int ff_sfdelta_can_replace(const SingleChannelElement *sce, |
247 | const uint8_t *nextband, int prev_sf, int new_sf, int band) |
248 | { |
249 | return new_sf >= (prev_sf - SCALE_MAX_DIFF) |
250 | && new_sf <= (prev_sf + SCALE_MAX_DIFF) |
251 | && sce->sf_idx[nextband[band]] >= (new_sf - SCALE_MAX_DIFF) |
252 | && sce->sf_idx[nextband[band]] <= (new_sf + SCALE_MAX_DIFF); |
253 | } |
254 | |
255 | /** |
256 | * linear congruential pseudorandom number generator |
257 | * |
258 | * @param previous_val pointer to the current state of the generator |
259 | * |
260 | * @return Returns a 32-bit pseudorandom integer |
261 | */ |
262 | static av_always_inline int lcg_random(unsigned previous_val) |
263 | { |
264 | union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 }; |
265 | return v.s; |
266 | } |
267 | |
268 | #define ERROR_IF(cond, ...) \ |
269 | if (cond) { \ |
270 | av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \ |
271 | return AVERROR(EINVAL); \ |
272 | } |
273 | |
274 | #define WARN_IF(cond, ...) \ |
275 | if (cond) { \ |
276 | av_log(avctx, AV_LOG_WARNING, __VA_ARGS__); \ |
277 | } |
278 | |
279 | #endif /* AVCODEC_AACENC_UTILS_H */ |
280 |