aacenc_utils.h 8.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279
  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. * @file
  23. * AAC encoder utilities
  24. * @author Rostislav Pehlivanov ( atomnuker gmail com )
  25. */
  26. #ifndef AVCODEC_AACENC_UTILS_H
  27. #define AVCODEC_AACENC_UTILS_H
  28. #include "libavutil/ffmath.h"
  29. #include "aac.h"
  30. #include "aacenctab.h"
  31. #include "aactab.h"
  32. #define ROUND_STANDARD 0.4054f
  33. #define ROUND_TO_ZERO 0.1054f
  34. #define C_QUANT 0.4054f
  35. static inline void abs_pow34_v(float *out, const float *in, const int size)
  36. {
  37. int i;
  38. for (i = 0; i < size; i++) {
  39. float a = fabsf(in[i]);
  40. out[i] = sqrtf(a * sqrtf(a));
  41. }
  42. }
  43. static inline float pos_pow34(float a)
  44. {
  45. return sqrtf(a * sqrtf(a));
  46. }
  47. /**
  48. * Quantize one coefficient.
  49. * @return absolute value of the quantized coefficient
  50. * @see 3GPP TS26.403 5.6.2 "Scalefactor determination"
  51. */
  52. static inline int quant(float coef, const float Q, const float rounding)
  53. {
  54. float a = coef * Q;
  55. return sqrtf(a * sqrtf(a)) + rounding;
  56. }
  57. static inline void quantize_bands(int *out, const float *in, const float *scaled,
  58. int size, int is_signed, int maxval, const float Q34,
  59. const float rounding)
  60. {
  61. int i;
  62. for (i = 0; i < size; i++) {
  63. float qc = scaled[i] * Q34;
  64. int tmp = (int)FFMIN(qc + rounding, (float)maxval);
  65. if (is_signed && in[i] < 0.0f) {
  66. tmp = -tmp;
  67. }
  68. out[i] = tmp;
  69. }
  70. }
  71. static inline float find_max_val(int group_len, int swb_size, const float *scaled)
  72. {
  73. float maxval = 0.0f;
  74. int w2, i;
  75. for (w2 = 0; w2 < group_len; w2++) {
  76. for (i = 0; i < swb_size; i++) {
  77. maxval = FFMAX(maxval, scaled[w2*128+i]);
  78. }
  79. }
  80. return maxval;
  81. }
  82. static inline int find_min_book(float maxval, int sf)
  83. {
  84. float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - sf + SCALE_ONE_POS - SCALE_DIV_512];
  85. int qmaxval, cb;
  86. qmaxval = maxval * Q34 + C_QUANT;
  87. if (qmaxval >= (FF_ARRAY_ELEMS(aac_maxval_cb)))
  88. cb = 11;
  89. else
  90. cb = aac_maxval_cb[qmaxval];
  91. return cb;
  92. }
  93. static inline float find_form_factor(int group_len, int swb_size, float thresh,
  94. const float *scaled, float nzslope) {
  95. const float iswb_size = 1.0f / swb_size;
  96. const float iswb_sizem1 = 1.0f / (swb_size - 1);
  97. const float ethresh = thresh;
  98. float form = 0.0f, weight = 0.0f;
  99. int w2, i;
  100. for (w2 = 0; w2 < group_len; w2++) {
  101. float e = 0.0f, e2 = 0.0f, var = 0.0f, maxval = 0.0f;
  102. float nzl = 0;
  103. for (i = 0; i < swb_size; i++) {
  104. float s = fabsf(scaled[w2*128+i]);
  105. maxval = FFMAX(maxval, s);
  106. e += s;
  107. e2 += s *= s;
  108. /* We really don't want a hard non-zero-line count, since
  109. * even below-threshold lines do add up towards band spectral power.
  110. * So, fall steeply towards zero, but smoothly
  111. */
  112. if (s >= ethresh) {
  113. nzl += 1.0f;
  114. } else {
  115. if (nzslope == 2.f)
  116. nzl += (s / ethresh) * (s / ethresh);
  117. else
  118. nzl += ff_fast_powf(s / ethresh, nzslope);
  119. }
  120. }
  121. if (e2 > thresh) {
  122. float frm;
  123. e *= iswb_size;
  124. /** compute variance */
  125. for (i = 0; i < swb_size; i++) {
  126. float d = fabsf(scaled[w2*128+i]) - e;
  127. var += d*d;
  128. }
  129. var = sqrtf(var * iswb_sizem1);
  130. e2 *= iswb_size;
  131. frm = e / FFMIN(e+4*var,maxval);
  132. form += e2 * sqrtf(frm) / FFMAX(0.5f,nzl);
  133. weight += e2;
  134. }
  135. }
  136. if (weight > 0) {
  137. return form / weight;
  138. } else {
  139. return 1.0f;
  140. }
  141. }
  142. /** Return the minimum scalefactor where the quantized coef does not clip. */
  143. static inline uint8_t coef2minsf(float coef)
  144. {
  145. return av_clip_uint8(log2f(coef)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512);
  146. }
  147. /** Return the maximum scalefactor where the quantized coef is not zero. */
  148. static inline uint8_t coef2maxsf(float coef)
  149. {
  150. return av_clip_uint8(log2f(coef)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512);
  151. }
  152. /*
  153. * Returns the closest possible index to an array of float values, given a value.
  154. */
  155. static inline int quant_array_idx(const float val, const float *arr, const int num)
  156. {
  157. int i, index = 0;
  158. float quant_min_err = INFINITY;
  159. for (i = 0; i < num; i++) {
  160. float error = (val - arr[i])*(val - arr[i]);
  161. if (error < quant_min_err) {
  162. quant_min_err = error;
  163. index = i;
  164. }
  165. }
  166. return index;
  167. }
  168. /**
  169. * approximates exp10f(-3.0f*(0.5f + 0.5f * cosf(FFMIN(b,15.5f) / 15.5f)))
  170. */
  171. static av_always_inline float bval2bmax(float b)
  172. {
  173. return 0.001f + 0.0035f * (b*b*b) / (15.5f*15.5f*15.5f);
  174. }
  175. /*
  176. * Compute a nextband map to be used with SF delta constraint utilities.
  177. * The nextband array should contain 128 elements, and positions that don't
  178. * map to valid, nonzero bands of the form w*16+g (with w being the initial
  179. * window of the window group, only) are left indetermined.
  180. */
  181. static inline void ff_init_nextband_map(const SingleChannelElement *sce, uint8_t *nextband)
  182. {
  183. unsigned char prevband = 0;
  184. int w, g;
  185. /** Just a safe default */
  186. for (g = 0; g < 128; g++)
  187. nextband[g] = g;
  188. /** Now really navigate the nonzero band chain */
  189. for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
  190. for (g = 0; g < sce->ics.num_swb; g++) {
  191. if (!sce->zeroes[w*16+g] && sce->band_type[w*16+g] < RESERVED_BT)
  192. prevband = nextband[prevband] = w*16+g;
  193. }
  194. }
  195. nextband[prevband] = prevband; /* terminate */
  196. }
  197. /*
  198. * Updates nextband to reflect a removed band (equivalent to
  199. * calling ff_init_nextband_map after marking a band as zero)
  200. */
  201. static inline void ff_nextband_remove(uint8_t *nextband, int prevband, int band)
  202. {
  203. nextband[prevband] = nextband[band];
  204. }
  205. /*
  206. * Checks whether the specified band could be removed without inducing
  207. * scalefactor delta that violates SF delta encoding constraints.
  208. * prev_sf has to be the scalefactor of the previous nonzero, nonspecial
  209. * band, in encoding order, or negative if there was no such band.
  210. */
  211. static inline int ff_sfdelta_can_remove_band(const SingleChannelElement *sce,
  212. const uint8_t *nextband, int prev_sf, int band)
  213. {
  214. return prev_sf >= 0
  215. && sce->sf_idx[nextband[band]] >= (prev_sf - SCALE_MAX_DIFF)
  216. && sce->sf_idx[nextband[band]] <= (prev_sf + SCALE_MAX_DIFF);
  217. }
  218. /*
  219. * Checks whether the specified band's scalefactor could be replaced
  220. * with another one without violating SF delta encoding constraints.
  221. * prev_sf has to be the scalefactor of the previous nonzero, nonsepcial
  222. * band, in encoding order, or negative if there was no such band.
  223. */
  224. static inline int ff_sfdelta_can_replace(const SingleChannelElement *sce,
  225. const uint8_t *nextband, int prev_sf, int new_sf, int band)
  226. {
  227. return new_sf >= (prev_sf - SCALE_MAX_DIFF)
  228. && new_sf <= (prev_sf + SCALE_MAX_DIFF)
  229. && sce->sf_idx[nextband[band]] >= (new_sf - SCALE_MAX_DIFF)
  230. && sce->sf_idx[nextband[band]] <= (new_sf + SCALE_MAX_DIFF);
  231. }
  232. /**
  233. * linear congruential pseudorandom number generator
  234. *
  235. * @param previous_val pointer to the current state of the generator
  236. *
  237. * @return Returns a 32-bit pseudorandom integer
  238. */
  239. static av_always_inline int lcg_random(unsigned previous_val)
  240. {
  241. union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
  242. return v.s;
  243. }
  244. #define ERROR_IF(cond, ...) \
  245. if (cond) { \
  246. av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
  247. return AVERROR(EINVAL); \
  248. }
  249. #define WARN_IF(cond, ...) \
  250. if (cond) { \
  251. av_log(avctx, AV_LOG_WARNING, __VA_ARGS__); \
  252. }
  253. #endif /* AVCODEC_AACENC_UTILS_H */