* AAC encoder psychoacoustic model
*/
+#include "libavutil/attributes.h"
#include "avcodec.h"
#include "aactab.h"
#include "psymodel.h"
/**
* LAME psy model preset struct
*/
-typedef struct {
+typedef struct PsyLamePreset {
int quality; ///< Quality to map the rest of the vaules to.
/* This is overloaded to be both kbps per channel in ABR mode, and
* requested quality in constant quality mode.
/**
* LAME psy model specific initialization
*/
-static void lame_window_init(AacPsyContext *ctx, AVCodecContext *avctx) {
+static av_cold void lame_window_init(AacPsyContext *ctx, AVCodecContext *avctx)
+{
int i, j;
for (i = 0; i < avctx->channels; i++) {
AacPsyChannel *pch = &ctx->ch[i];
- if (avctx->flags & CODEC_FLAG_QSCALE)
+ if (avctx->flags & AV_CODEC_FLAG_QSCALE)
pch->attack_threshold = psy_vbr_map[avctx->global_quality / FF_QP2LAMBDA].st_lrm;
else
pch->attack_threshold = lame_calc_attack_threshold(avctx->bit_rate / avctx->channels / 1000);
const float num_bark = calc_bark((float)bandwidth);
ctx->model_priv_data = av_mallocz(sizeof(AacPsyContext));
+ if (!ctx->model_priv_data)
+ return AVERROR(ENOMEM);
pctx = (AacPsyContext*) ctx->model_priv_data;
pctx->chan_bitrate = chan_bitrate;
ctx->bitres.size = 6144 - pctx->frame_bits;
ctx->bitres.size -= ctx->bitres.size % 8;
pctx->fill_level = ctx->bitres.size;
- minath = ath(3410, ATH_ADD);
+ minath = ath(3410 - 0.733 * ATH_ADD, ATH_ADD);
for (j = 0; j < 2; j++) {
AacPsyCoeffs *coeffs = pctx->psy_coef[j];
const uint8_t *band_sizes = ctx->bands[j];
float line_to_frequency = ctx->avctx->sample_rate / (j ? 256.f : 2048.0f);
- float avg_chan_bits = chan_bitrate / ctx->avctx->sample_rate * (j ? 128.0f : 1024.0f);
+ float avg_chan_bits = chan_bitrate * (j ? 128.0f : 1024.0f) / ctx->avctx->sample_rate;
/* reference encoder uses 2.4% here instead of 60% like the spec says */
float bark_pe = 0.024f * PSY_3GPP_BITS_TO_PE(avg_chan_bits) / num_bark;
float en_spread_low = j ? PSY_3GPP_EN_SPREAD_LOW_S : PSY_3GPP_EN_SPREAD_LOW_L;
}
pctx->ch = av_mallocz(sizeof(AacPsyChannel) * ctx->avctx->channels);
+ if (!pctx->ch) {
+ av_freep(&pctx);
+ return AVERROR(ENOMEM);
+ }
lame_window_init(pctx, ctx->avctx);
AacPsyChannel *pch = &pctx->ch[channel];
uint8_t grouping = 0;
int next_type = pch->next_window_seq;
- FFPsyWindowInfo wi;
+ FFPsyWindowInfo wi = { { 0 } };
- memset(&wi, 0, sizeof(wi));
if (la) {
float s[8], v;
int switch_to_eight = 0;
int stay_short = 0;
for (i = 0; i < 8; i++) {
for (j = 0; j < 128; j++) {
- v = iir_filter(la[(i*128+j)*ctx->avctx->channels], pch->iir_state);
+ v = iir_filter(la[i*128+j], pch->iir_state);
sum += v*v;
}
s[i] = sum;
for (w = 0; w < wi->num_windows*16; w += 16) {
AacPsyBand *bands = &pch->band[w];
- //5.4.2.3 "Spreading" & 5.4.3 "Spreaded Energy Calculation"
+ /* 5.4.2.3 "Spreading" & 5.4.3 "Spread Energy Calculation" */
spread_en[0] = bands[0].energy;
for (g = 1; g < num_bands; g++) {
bands[g].thr = FFMAX(bands[g].thr, bands[g-1].thr * coeffs[g].spread_hi[0]);
band->thr = FFMAX(PSY_3GPP_RPEMIN*band->thr, FFMIN(band->thr,
PSY_3GPP_RPELEV*pch->prev_band[w+g].thr_quiet));
- /* 5.6.1.3.1 "Prepatory steps of the perceptual entropy calculation" */
+ /* 5.6.1.3.1 "Preparatory steps of the perceptual entropy calculation" */
pe += calc_pe_3gpp(band);
a += band->pe_const;
active_lines += band->active_lines;
int uselongblock = 1;
int attacks[AAC_NUM_BLOCKS_SHORT + 1] = { 0 };
int i;
- FFPsyWindowInfo wi;
+ FFPsyWindowInfo wi = { { 0 } };
- memset(&wi, 0, sizeof(wi));
if (la) {
float hpfsmpl[AAC_BLOCK_SIZE_LONG];
- float const *pf = hpfsmpl;
+ const float *pf = hpfsmpl;
float attack_intensity[(AAC_NUM_BLOCKS_SHORT + 1) * PSY_LAME_NUM_SUBBLOCKS];
float energy_subshort[(AAC_NUM_BLOCKS_SHORT + 1) * PSY_LAME_NUM_SUBBLOCKS];
float energy_short[AAC_NUM_BLOCKS_SHORT + 1] = { 0 };
- int chans = ctx->avctx->channels;
- const float *firbuf = la + (AAC_BLOCK_SIZE_SHORT/4 - PSY_LAME_FIR_LEN) * chans;
+ const float *firbuf = la + (AAC_BLOCK_SIZE_SHORT/4 - PSY_LAME_FIR_LEN);
int j, att_sum = 0;
/* LAME comment: apply high pass filter of fs/4 */
for (i = 0; i < AAC_BLOCK_SIZE_LONG; i++) {
float sum1, sum2;
- sum1 = firbuf[(i + ((PSY_LAME_FIR_LEN - 1) / 2)) * chans];
+ sum1 = firbuf[i + (PSY_LAME_FIR_LEN - 1) / 2];
sum2 = 0.0;
for (j = 0; j < ((PSY_LAME_FIR_LEN - 1) / 2) - 1; j += 2) {
- sum1 += psy_fir_coeffs[j] * (firbuf[(i + j) * chans] + firbuf[(i + PSY_LAME_FIR_LEN - j) * chans]);
- sum2 += psy_fir_coeffs[j + 1] * (firbuf[(i + j + 1) * chans] + firbuf[(i + PSY_LAME_FIR_LEN - j - 1) * chans]);
+ sum1 += psy_fir_coeffs[j] * (firbuf[i + j] + firbuf[i + PSY_LAME_FIR_LEN - j]);
+ sum2 += psy_fir_coeffs[j + 1] * (firbuf[i + j + 1] + firbuf[i + PSY_LAME_FIR_LEN - j - 1]);
}
- /* NOTE: The LAME psymodel expects it's input in the range -32768 to 32768. Tuning this for normalized floats would be difficult. */
+ /* NOTE: The LAME psymodel expects its input in the range -32768 to
+ * 32768. Tuning this for normalized floats would be difficult. */
hpfsmpl[i] = (sum1 + sum2) * 32768.0f;
}
}
for (i = 0; i < AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS; i++) {
- float const *const pfe = pf + AAC_BLOCK_SIZE_LONG / (AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS);
+ const float *const pfe = pf + AAC_BLOCK_SIZE_LONG / (AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS);
float p = 1.0f;
for (; pf < pfe; pf++)
p = FFMAX(p, fabsf(*pf));
/* GB: tuned (1) to avoid too many short blocks for test sample TRUMPET */
/* RH: tuned (2) to let enough short blocks through for test sample FSOL and SNAPS */
for (i = 1; i < AAC_NUM_BLOCKS_SHORT + 1; i++) {
- float const u = energy_short[i - 1];
- float const v = energy_short[i];
- float const m = FFMAX(u, v);
+ const float u = energy_short[i - 1];
+ const float v = energy_short[i];
+ const float m = FFMAX(u, v);
if (m < 40000) { /* (2) */
if (u < 1.7f * v && v < 1.7f * u) { /* (1) */
if (i == 1 && attacks[0] < attacks[i])