2 * AAC encoder utilities
3 * Copyright (C) 2015 Rostislav Pehlivanov
5 * This file is part of FFmpeg.
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.
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.
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
24 * AAC encoder utilities
25 * @author Rostislav Pehlivanov ( atomnuker gmail com )
28 #ifndef AVCODEC_AACENC_UTILS_H
29 #define AVCODEC_AACENC_UTILS_H
32 #include "aac_tablegen_decl.h"
33 #include "aacenctab.h"
35 #define ROUND_STANDARD 0.4054f
36 #define ROUND_TO_ZERO 0.1054f
37 #define C_QUANT 0.4054f
39 static inline void abs_pow34_v(float *out, const float *in, const int size)
42 for (i = 0; i < size; i++) {
43 float a = fabsf(in[i]);
44 out[i] = sqrtf(a * sqrtf(a));
49 * Quantize one coefficient.
50 * @return absolute value of the quantized coefficient
51 * @see 3GPP TS26.403 5.6.2 "Scalefactor determination"
53 static inline int quant(float coef, const float Q, const float rounding)
56 return sqrtf(a * sqrtf(a)) + rounding;
59 static inline void quantize_bands(int *out, const float *in, const float *scaled,
60 int size, float Q34, int is_signed, int maxval,
65 for (i = 0; i < size; i++) {
67 out[i] = (int)FFMIN(qc + rounding, (double)maxval);
68 if (is_signed && in[i] < 0.0f) {
74 static inline float find_max_val(int group_len, int swb_size, const float *scaled)
78 for (w2 = 0; w2 < group_len; w2++) {
79 for (i = 0; i < swb_size; i++) {
80 maxval = FFMAX(maxval, scaled[w2*128+i]);
86 static inline int find_min_book(float maxval, int sf)
88 float Q = ff_aac_pow2sf_tab[POW_SF2_ZERO - sf + SCALE_ONE_POS - SCALE_DIV_512];
89 float Q34 = sqrtf(Q * sqrtf(Q));
91 qmaxval = maxval * Q34 + C_QUANT;
92 if (qmaxval >= (FF_ARRAY_ELEMS(aac_maxval_cb)))
95 cb = aac_maxval_cb[qmaxval];
99 static inline float find_form_factor(int group_len, int swb_size, float thresh,
100 const float *scaled, float nzslope) {
101 const float iswb_size = 1.0f / swb_size;
102 const float iswb_sizem1 = 1.0f / (swb_size - 1);
103 const float ethresh = thresh;
104 float form = 0.0f, weight = 0.0f;
106 for (w2 = 0; w2 < group_len; w2++) {
107 float e = 0.0f, e2 = 0.0f, var = 0.0f, maxval = 0.0f;
109 for (i = 0; i < swb_size; i++) {
110 float s = fabsf(scaled[w2*128+i]);
111 maxval = FFMAX(maxval, s);
114 /* We really don't want a hard non-zero-line count, since
115 * even below-threshold lines do add up towards band spectral power.
116 * So, fall steeply towards zero, but smoothly
121 nzl += powf(s / ethresh, nzslope);
128 /** compute variance */
129 for (i = 0; i < swb_size; i++) {
130 float d = fabsf(scaled[w2*128+i]) - e;
133 var = sqrtf(var * iswb_sizem1);
136 frm = e / FFMIN(e+4*var,maxval);
137 form += e2 * sqrtf(frm) / FFMAX(0.5f,nzl);
142 return form / weight;
148 /** Return the minimum scalefactor where the quantized coef does not clip. */
149 static inline uint8_t coef2minsf(float coef)
151 return av_clip_uint8(log2f(coef)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512);
154 /** Return the maximum scalefactor where the quantized coef is not zero. */
155 static inline uint8_t coef2maxsf(float coef)
157 return av_clip_uint8(log2f(coef)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512);
161 * Returns the closest possible index to an array of float values, given a value.
163 static inline int quant_array_idx(const float val, const float *arr, const int num)
166 float quant_min_err = INFINITY;
167 for (i = 0; i < num; i++) {
168 float error = (val - arr[i])*(val - arr[i]);
169 if (error < quant_min_err) {
170 quant_min_err = error;
178 * approximates exp10f(-3.0f*(0.5f + 0.5f * cosf(FFMIN(b,15.5f) / 15.5f)))
180 static av_always_inline float bval2bmax(float b)
182 return 0.001f + 0.0035f * (b*b*b) / (15.5f*15.5f*15.5f);
186 * linear congruential pseudorandom number generator, copied from the decoder
188 static inline int lcg_random(unsigned previous_val)
190 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
194 #define ERROR_IF(cond, ...) \
196 av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
197 return AVERROR(EINVAL); \
200 #define WARN_IF(cond, ...) \
202 av_log(avctx, AV_LOG_WARNING, __VA_ARGS__); \
205 #endif /* AVCODEC_AACENC_UTILS_H */