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 "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));
48 static inline float pos_pow34(float a)
50 return sqrtf(a * sqrtf(a));
54 * Quantize one coefficient.
55 * @return absolute value of the quantized coefficient
56 * @see 3GPP TS26.403 5.6.2 "Scalefactor determination"
58 static inline int quant(float coef, const float Q, const float rounding)
61 return sqrtf(a * sqrtf(a)) + rounding;
64 static inline void quantize_bands(int *out, const float *in, const float *scaled,
65 int size, float Q34, int is_signed, int maxval,
70 for (i = 0; i < size; i++) {
72 out[i] = (int)FFMIN(qc + rounding, (double)maxval);
73 if (is_signed && in[i] < 0.0f) {
79 static inline float find_max_val(int group_len, int swb_size, const float *scaled)
83 for (w2 = 0; w2 < group_len; w2++) {
84 for (i = 0; i < swb_size; i++) {
85 maxval = FFMAX(maxval, scaled[w2*128+i]);
91 static inline int find_min_book(float maxval, int sf)
93 float Q = ff_aac_pow2sf_tab[POW_SF2_ZERO - sf + SCALE_ONE_POS - SCALE_DIV_512];
94 float Q34 = sqrtf(Q * sqrtf(Q));
96 qmaxval = maxval * Q34 + C_QUANT;
97 if (qmaxval >= (FF_ARRAY_ELEMS(aac_maxval_cb)))
100 cb = aac_maxval_cb[qmaxval];
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;
111 for (w2 = 0; w2 < group_len; w2++) {
112 float e = 0.0f, e2 = 0.0f, var = 0.0f, maxval = 0.0f;
114 for (i = 0; i < swb_size; i++) {
115 float s = fabsf(scaled[w2*128+i]);
116 maxval = FFMAX(maxval, 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
126 nzl += powf(s / ethresh, nzslope);
133 /** compute variance */
134 for (i = 0; i < swb_size; i++) {
135 float d = fabsf(scaled[w2*128+i]) - e;
138 var = sqrtf(var * iswb_sizem1);
141 frm = e / FFMIN(e+4*var,maxval);
142 form += e2 * sqrtf(frm) / FFMAX(0.5f,nzl);
147 return form / weight;
153 /** Return the minimum scalefactor where the quantized coef does not clip. */
154 static inline uint8_t coef2minsf(float coef)
156 return av_clip_uint8(log2f(coef)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512);
159 /** Return the maximum scalefactor where the quantized coef is not zero. */
160 static inline uint8_t coef2maxsf(float coef)
162 return av_clip_uint8(log2f(coef)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512);
166 * Returns the closest possible index to an array of float values, given a value.
168 static inline int quant_array_idx(const float val, const float *arr, const int num)
171 float quant_min_err = INFINITY;
172 for (i = 0; i < num; i++) {
173 float error = (val - arr[i])*(val - arr[i]);
174 if (error < quant_min_err) {
175 quant_min_err = error;
183 * approximates exp10f(-3.0f*(0.5f + 0.5f * cosf(FFMIN(b,15.5f) / 15.5f)))
185 static av_always_inline float bval2bmax(float b)
187 return 0.001f + 0.0035f * (b*b*b) / (15.5f*15.5f*15.5f);
191 * Compute a nextband map to be used with SF delta constraint utilities.
192 * The nextband array should contain 128 elements, and positions that don't
193 * map to valid, nonzero bands of the form w*16+g (with w being the initial
194 * window of the window group, only) are left indetermined.
196 static inline void ff_init_nextband_map(const SingleChannelElement *sce, uint8_t *nextband)
198 unsigned char prevband = 0;
200 /** Just a safe default */
201 for (g = 0; g < 128; g++)
204 /** Now really navigate the nonzero band chain */
205 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
206 for (g = 0; g < sce->ics.num_swb; g++) {
207 if (!sce->zeroes[w*16+g] && sce->band_type[w*16+g] < RESERVED_BT)
208 prevband = nextband[prevband] = w*16+g;
211 nextband[prevband] = prevband; /* terminate */
215 * Updates nextband to reflect a removed band (equivalent to
216 * calling ff_init_nextband_map after marking a band as zero)
218 static inline void ff_nextband_remove(uint8_t *nextband, int prevband, int band)
220 nextband[prevband] = nextband[band];
224 * Checks whether the specified band could be removed without inducing
225 * scalefactor delta that violates SF delta encoding constraints.
226 * prev_sf has to be the scalefactor of the previous nonzero, nonspecial
227 * band, in encoding order, or negative if there was no such band.
229 static inline int ff_sfdelta_can_remove_band(const SingleChannelElement *sce,
230 const uint8_t *nextband, int prev_sf, int band)
233 && sce->sf_idx[nextband[band]] >= (prev_sf - SCALE_MAX_DIFF)
234 && sce->sf_idx[nextband[band]] <= (prev_sf + SCALE_MAX_DIFF);
238 * Checks whether the specified band's scalefactor could be replaced
239 * with another one without violating SF delta encoding constraints.
240 * prev_sf has to be the scalefactor of the previous nonzero, nonsepcial
241 * band, in encoding order, or negative if there was no such band.
243 static inline int ff_sfdelta_can_replace(const SingleChannelElement *sce,
244 const uint8_t *nextband, int prev_sf, int new_sf, int band)
246 return new_sf >= (prev_sf - SCALE_MAX_DIFF)
247 && new_sf <= (prev_sf + SCALE_MAX_DIFF)
248 && sce->sf_idx[nextband[band]] >= (new_sf - SCALE_MAX_DIFF)
249 && sce->sf_idx[nextband[band]] <= (new_sf + SCALE_MAX_DIFF);
252 #define ERROR_IF(cond, ...) \
254 av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
255 return AVERROR(EINVAL); \
258 #define WARN_IF(cond, ...) \
260 av_log(avctx, AV_LOG_WARNING, __VA_ARGS__); \
263 #endif /* AVCODEC_AACENC_UTILS_H */