[FFmpeg-cvslog] r15919 - in trunk/libavcodec: aac.c aac.h aactab.c aactab.h
superdump
subversion
Mon Nov 24 01:13:50 CET 2008
Author: superdump
Date: Mon Nov 24 01:13:50 2008
New Revision: 15919
Log:
AAC: Frequency domain prediction and hence Main profile support
Patch by Alex Converse ( alex converse gmail com )
Modified:
trunk/libavcodec/aac.c
trunk/libavcodec/aac.h
trunk/libavcodec/aactab.c
trunk/libavcodec/aactab.h
Modified: trunk/libavcodec/aac.c
==============================================================================
--- trunk/libavcodec/aac.c (original)
+++ trunk/libavcodec/aac.c Mon Nov 24 01:13:50 2008
@@ -41,7 +41,7 @@
* N (code in SoC repo) Long Term Prediction
* Y intensity stereo
* Y channel coupling
- * N frequency domain prediction
+ * Y frequency domain prediction
* Y Perceptual Noise Substitution
* Y Mid/Side stereo
* N Scalable Inverse AAC Quantization
@@ -331,6 +331,7 @@ static int decode_audio_specific_config(
skip_bits_long(&gb, i);
switch (ac->m4ac.object_type) {
+ case AOT_AAC_MAIN:
case AOT_AAC_LC:
if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
return -1;
@@ -354,6 +355,27 @@ static av_always_inline int lcg_random(i
return previous_val * 1664525 + 1013904223;
}
+static void reset_predict_state(PredictorState * ps) {
+ ps->r0 = 0.0f;
+ ps->r1 = 0.0f;
+ ps->cor0 = 0.0f;
+ ps->cor1 = 0.0f;
+ ps->var0 = 1.0f;
+ ps->var1 = 1.0f;
+}
+
+static void reset_all_predictors(PredictorState * ps) {
+ int i;
+ for (i = 0; i < MAX_PREDICTORS; i++)
+ reset_predict_state(&ps[i]);
+}
+
+static void reset_predictor_group(PredictorState * ps, int group_num) {
+ int i;
+ for (i = group_num-1; i < MAX_PREDICTORS; i+=30)
+ reset_predict_state(&ps[i]);
+}
+
static av_cold int aac_decode_init(AVCodecContext * avccontext) {
AACContext * ac = avccontext->priv_data;
int i;
@@ -432,6 +454,21 @@ static void skip_data_stream_element(Get
skip_bits_long(gb, 8 * count);
}
+static int decode_prediction(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb) {
+ int sfb;
+ if (get_bits1(gb)) {
+ ics->predictor_reset_group = get_bits(gb, 5);
+ if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
+ av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
+ return -1;
+ }
+ }
+ for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
+ ics->prediction_used[sfb] = get_bits1(gb);
+ }
+ return 0;
+}
+
/**
* Decode Individual Channel Stream info; reference: table 4.6.
*
@@ -464,16 +501,30 @@ static int decode_ics_info(AACContext *
ics->swb_offset = swb_offset_128[ac->m4ac.sampling_index];
ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
ics->tns_max_bands = tns_max_bands_128[ac->m4ac.sampling_index];
+ ics->predictor_present = 0;
} else {
ics->max_sfb = get_bits(gb, 6);
ics->num_windows = 1;
ics->swb_offset = swb_offset_1024[ac->m4ac.sampling_index];
ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
ics->tns_max_bands = tns_max_bands_1024[ac->m4ac.sampling_index];
- if (get_bits1(gb)) {
+ ics->predictor_present = get_bits1(gb);
+ ics->predictor_reset_group = 0;
+ if (ics->predictor_present) {
+ if (ac->m4ac.object_type == AOT_AAC_MAIN) {
+ if (decode_prediction(ac, ics, gb)) {
+ memset(ics, 0, sizeof(IndividualChannelStream));
+ return -1;
+ }
+ } else if (ac->m4ac.object_type == AOT_AAC_LC) {
+ av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
+ memset(ics, 0, sizeof(IndividualChannelStream));
+ return -1;
+ } else {
av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
memset(ics, 0, sizeof(IndividualChannelStream));
return -1;
+ }
}
}
@@ -786,6 +837,77 @@ static int decode_spectrum_and_dequant(A
return 0;
}
+static av_always_inline float flt16_round(float pf) {
+ int exp;
+ pf = frexpf(pf, &exp);
+ pf = ldexpf(roundf(ldexpf(pf, 8)), exp-8);
+ return pf;
+}
+
+static av_always_inline float flt16_even(float pf) {
+ int exp;
+ pf = frexpf(pf, &exp);
+ pf = ldexpf(rintf(ldexpf(pf, 8)), exp-8);
+ return pf;
+}
+
+static av_always_inline float flt16_trunc(float pf) {
+ int exp;
+ pf = frexpf(pf, &exp);
+ pf = ldexpf(truncf(ldexpf(pf, 8)), exp-8);
+ return pf;
+}
+
+static void predict(AACContext * ac, PredictorState * ps, float* coef, int output_enable) {
+ const float a = 0.953125; // 61.0/64
+ const float alpha = 0.90625; // 29.0/32
+ float e0, e1;
+ float pv;
+ float k1, k2;
+
+ k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
+ k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
+
+ pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
+ if (output_enable)
+ *coef += pv * ac->sf_scale;
+
+ e0 = *coef / ac->sf_scale;
+ e1 = e0 - k1 * ps->r0;
+
+ ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
+ ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
+ ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
+ ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
+
+ ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
+ ps->r0 = flt16_trunc(a * e0);
+}
+
+/**
+ * Apply AAC-Main style frequency domain prediction.
+ */
+static void apply_prediction(AACContext * ac, SingleChannelElement * sce) {
+ int sfb, k;
+
+ if (!sce->ics.predictor_initialized) {
+ reset_all_predictors(sce->ics.predictor_state);
+ sce->ics.predictor_initialized = 1;
+ }
+
+ if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
+ for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
+ for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
+ predict(ac, &sce->ics.predictor_state[k], &sce->coeffs[k],
+ sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
+ }
+ }
+ if (sce->ics.predictor_reset_group)
+ reset_predictor_group(sce->ics.predictor_state, sce->ics.predictor_reset_group);
+ } else
+ reset_all_predictors(sce->ics.predictor_state);
+}
+
/**
* Decode an individual_channel_stream payload; reference: table 4.44.
*
@@ -840,6 +962,10 @@ static int decode_ics(AACContext * ac, S
if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
return -1;
+
+ if(ac->m4ac.object_type == AOT_AAC_MAIN)
+ apply_prediction(ac, sce);
+
return 0;
}
Modified: trunk/libavcodec/aac.h
==============================================================================
--- trunk/libavcodec/aac.h (original)
+++ trunk/libavcodec/aac.h Mon Nov 24 01:13:50 2008
@@ -133,6 +133,20 @@ enum CouplingPoint {
};
/**
+ * Predictor State
+ */
+typedef struct {
+ float cor0;
+ float cor1;
+ float var0;
+ float var1;
+ float r0;
+ float r1;
+} PredictorState;
+
+#define MAX_PREDICTORS 672
+
+/**
* Individual Channel Stream
*/
typedef struct {
@@ -145,6 +159,11 @@ typedef struct {
int num_swb; ///< number of scalefactor window bands
int num_windows;
int tns_max_bands;
+ int predictor_present;
+ int predictor_initialized;
+ int predictor_reset_group;
+ uint8_t prediction_used[41];
+ PredictorState predictor_state[MAX_PREDICTORS];
} IndividualChannelStream;
/**
Modified: trunk/libavcodec/aactab.c
==============================================================================
--- trunk/libavcodec/aactab.c (original)
+++ trunk/libavcodec/aactab.c Mon Nov 24 01:13:50 2008
@@ -43,6 +43,10 @@ const uint8_t ff_aac_num_swb_128[] = {
12, 12, 12, 14, 14, 14, 15, 15, 15, 15, 15, 15
};
+const uint8_t ff_aac_pred_sfb_max[] = {
+ 33, 33, 38, 40, 40, 40, 41, 41, 37, 37, 37, 34
+};
+
const uint32_t ff_aac_scalefactor_code[121] = {
0x3ffe8, 0x3ffe6, 0x3ffe7, 0x3ffe5, 0x7fff5, 0x7fff1, 0x7ffed, 0x7fff6,
0x7ffee, 0x7ffef, 0x7fff0, 0x7fffc, 0x7fffd, 0x7ffff, 0x7fffe, 0x7fff7,
Modified: trunk/libavcodec/aactab.h
==============================================================================
--- trunk/libavcodec/aactab.h (original)
+++ trunk/libavcodec/aactab.h Mon Nov 24 01:13:50 2008
@@ -54,6 +54,8 @@ extern const uint8_t ff_aac_num_swb_1024
extern const uint8_t ff_aac_num_swb_128 [];
// @}
+extern const uint8_t ff_aac_pred_sfb_max [];
+
extern const uint32_t ff_aac_scalefactor_code[121];
extern const uint8_t ff_aac_scalefactor_bits[121];
More information about the ffmpeg-cvslog
mailing list