]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/ac3dec.c
skip unsupported frame types and substream id's
[ffmpeg] / libavcodec / ac3dec.c
index 5a12a46bd955872315545451f2312e6c434b908c..b7941c1623725325b7542c3ab641f43ddbe30342 100644 (file)
 #include <math.h>
 #include <string.h>
 
+#include "libavutil/crc.h"
+#include "libavutil/random.h"
 #include "avcodec.h"
 #include "ac3_parser.h"
 #include "bitstream.h"
 #include "dsputil.h"
-#include "random.h"
+
+/** Maximum possible frame size when the specification limit is ignored */
+#define AC3_MAX_FRAME_SIZE 21695
 
 /**
  * Table of bin locations for rematrixing bands
  */
 static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
 
-/**
- * table for exponent to scale_factor mapping
- * scale_factors[i] = 2 ^ -i
- */
-static float scale_factors[25];
-
 /** table for grouping exponents */
 static uint8_t exp_ungroup_tab[128][3];
 
 
 /** tables for ungrouping mantissas */
-static float b1_mantissas[32][3];
-static float b2_mantissas[128][3];
-static float b3_mantissas[8];
-static float b4_mantissas[128][2];
-static float b5_mantissas[16];
+static int b1_mantissas[32][3];
+static int b2_mantissas[128][3];
+static int b3_mantissas[8];
+static int b4_mantissas[128][2];
+static int b5_mantissas[16];
 
 /**
  * Quantization table: levels for symmetric. bits for asymmetric.
@@ -90,18 +88,6 @@ static const float gain_levels[6] = {
     LEVEL_MINUS_9DB
 };
 
-/**
- * Table for center mix levels
- * reference: Section 5.4.2.4 cmixlev
- */
-static const uint8_t center_levels[4] = { 2, 3, 4, 3 };
-
-/**
- * Table for surround mix levels
- * reference: Section 5.4.2.5 surmixlev
- */
-static const uint8_t surround_levels[4] = { 2, 4, 0, 4 };
-
 /**
  * Table for default stereo downmixing coefficients
  * reference: Section 7.8.2 Downmixing Into Two Channels
@@ -125,6 +111,7 @@ static const uint8_t ac3_default_coeffs[8][5][2] = {
 #define AC3_OUTPUT_LFEON  8
 
 typedef struct {
+    int num_blocks;                         ///< number of audio blocks
     int channel_mode;                       ///< channel mode (acmod)
     int block_switch[AC3_MAX_CHANNELS];     ///< block switch flags
     int dither_flag[AC3_MAX_CHANNELS];      ///< dither flags
@@ -132,8 +119,8 @@ typedef struct {
     int cpl_in_use;                         ///< coupling in use
     int channel_in_cpl[AC3_MAX_CHANNELS];   ///< channel in coupling
     int phase_flags_in_use;                 ///< phase flags in use
+    int phase_flags[18];                    ///< phase flags
     int cpl_band_struct[18];                ///< coupling band structure
-    int rematrixing_strategy;               ///< rematrixing strategy
     int num_rematrixing_bands;              ///< number of rematrixing bands
     int rematrixing_flags[4];               ///< rematrixing flags
     int exp_strategy[AC3_MAX_CHANNELS];     ///< exponent strategies
@@ -145,8 +132,10 @@ typedef struct {
     uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
     uint8_t dba_values[AC3_MAX_CHANNELS][8];  ///< delta values for each segment
 
-    int sampling_rate;                      ///< sample frequency, in Hz
+    int sample_rate;                        ///< sample frequency, in Hz
     int bit_rate;                           ///< stream bit rate, in bits-per-second
+    int frame_type;                         ///< frame type (strmtyp)
+    int substreamid;                        ///< substream identification
     int frame_size;                         ///< current frame size, in bytes
 
     int channels;                           ///< number of total channels
@@ -156,22 +145,28 @@ typedef struct {
     int output_mode;                        ///< output channel configuration
     int out_channels;                       ///< number of output channels
 
+    int center_mix_level;                   ///< Center mix level index
+    int surround_mix_level;                 ///< Surround mix level index
     float downmix_coeffs[AC3_MAX_CHANNELS][2];  ///< stereo downmix coefficients
+    float downmix_coeff_adjust[2];          ///< adjustment needed for each output channel when downmixing
     float dynamic_range[2];                 ///< dynamic range
-    float cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
+    int   cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
     int   num_cpl_bands;                    ///< number of coupling bands
     int   num_cpl_subbands;                 ///< number of coupling sub bands
     int   start_freq[AC3_MAX_CHANNELS];     ///< start frequency bin
     int   end_freq[AC3_MAX_CHANNELS];       ///< end frequency bin
     AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
 
+    int num_exp_groups[AC3_MAX_CHANNELS];   ///< Number of exponent groups
     int8_t  dexps[AC3_MAX_CHANNELS][256];   ///< decoded exponents
     uint8_t bap[AC3_MAX_CHANNELS][256];     ///< bit allocation pointers
     int16_t psd[AC3_MAX_CHANNELS][256];     ///< scaled exponents
     int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
     int16_t mask[AC3_MAX_CHANNELS][50];     ///< masking curve values
 
+    int fixed_coeffs[AC3_MAX_CHANNELS][256];    ///> fixed-point transform coefficients
     DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]);  ///< transform coefficients
+    int downmixed;                              ///< indicates if coeffs are currently downmixed
 
     /* For IMDCT. */
     MDCTContext imdct_512;                  ///< for 512 sample IMDCT
@@ -180,9 +175,9 @@ typedef struct {
     float       add_bias;                   ///< offset for float_to_int16 conversion
     float       mul_bias;                   ///< scaling for float_to_int16 conversion
 
-    DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS-1][256]);     ///< output after imdct transform and windowing
+    DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]);       ///< output after imdct transform and windowing
     DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
-    DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS-1][256]);      ///< delay - added to the next block
+    DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]);        ///< delay - added to the next block
     DECLARE_ALIGNED_16(float, tmp_imdct[256]);                      ///< temporary storage for imdct transform
     DECLARE_ALIGNED_16(float, tmp_output[512]);                     ///< temporary storage for output before windowing
     DECLARE_ALIGNED_16(float, window[256]);                         ///< window coefficients
@@ -191,47 +186,24 @@ typedef struct {
     GetBitContext gbc;                      ///< bitstream reader
     AVRandomState dith_state;               ///< for dither generation
     AVCodecContext *avctx;                  ///< parent context
+    uint8_t *input_buffer;                  ///< temp buffer to prevent overread
 } AC3DecodeContext;
 
-/**
- * Generate a Kaiser-Bessel Derived Window.
- */
-static void ac3_window_init(float *window)
-{
-   int i, j;
-   double sum = 0.0, bessel, tmp;
-   double local_window[256];
-   double alpha2 = (5.0 * M_PI / 256.0) * (5.0 * M_PI / 256.0);
-
-   for (i = 0; i < 256; i++) {
-       tmp = i * (256 - i) * alpha2;
-       bessel = 1.0;
-       for (j = 100; j > 0; j--) /* default to 100 iterations */
-           bessel = bessel * tmp / (j * j) + 1;
-       sum += bessel;
-       local_window[i] = sum;
-   }
-
-   sum++;
-   for (i = 0; i < 256; i++)
-       window[i] = sqrt(local_window[i] / sum);
-}
-
 /**
  * Symmetrical Dequantization
  * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
  *            Tables 7.19 to 7.23
  */
-static inline float
+static inline int
 symmetric_dequant(int code, int levels)
 {
-    return (code - (levels >> 1)) * (2.0f / levels);
+    return ((code - (levels >> 1)) << 24) / levels;
 }
 
 /*
  * Initialize tables at runtime.
  */
-static void ac3_tables_init(void)
+static av_cold void ac3_tables_init(void)
 {
     int i;
 
@@ -271,11 +243,6 @@ static void ac3_tables_init(void)
         dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
     }
 
-    /* generate scale factors for exponents and asymmetrical dequantization
-       reference: Section 7.3.2 Expansion of Mantissas for Asymmetric Quantization */
-    for (i = 0; i < 25; i++)
-        scale_factors[i] = pow(2.0, -i);
-
     /* generate exponent tables
        reference: Section 7.1.3 Exponent Decoding */
     for(i=0; i<128; i++) {
@@ -289,7 +256,7 @@ static void ac3_tables_init(void)
 /**
  * AVCodec initialization
  */
-static int ac3_decode_init(AVCodecContext *avctx)
+static av_cold int ac3_decode_init(AVCodecContext *avctx)
 {
     AC3DecodeContext *s = avctx->priv_data;
     s->avctx = avctx;
@@ -298,7 +265,7 @@ static int ac3_decode_init(AVCodecContext *avctx)
     ac3_tables_init();
     ff_mdct_init(&s->imdct_256, 8, 1);
     ff_mdct_init(&s->imdct_512, 9, 1);
-    ac3_window_init(s->window);
+    ff_kbd_window_init(s->window, 5.0, 256);
     dsputil_init(&s->dsp, avctx);
     av_init_random(0, &s->dith_state);
 
@@ -311,6 +278,21 @@ static int ac3_decode_init(AVCodecContext *avctx)
         s->mul_bias = 32767.0f;
     }
 
+    /* allow downmixing to stereo or mono */
+    if (avctx->channels > 0 && avctx->request_channels > 0 &&
+            avctx->request_channels < avctx->channels &&
+            avctx->request_channels <= 2) {
+        avctx->channels = avctx->request_channels;
+    }
+    s->downmixed = 1;
+
+    /* allocate context input buffer */
+    if (avctx->error_resilience >= FF_ER_CAREFUL) {
+        s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
+        if (!s->input_buffer)
+            return AVERROR_NOMEM;
+    }
+
     return 0;
 }
 
@@ -323,47 +305,38 @@ static int ac3_parse_header(AC3DecodeContext *s)
 {
     AC3HeaderInfo hdr;
     GetBitContext *gbc = &s->gbc;
-    float center_mix_level, surround_mix_level;
     int err, i;
 
-    err = ff_ac3_parse_header(gbc->buffer, &hdr);
+    err = ff_ac3_parse_header(gbc, &hdr);
     if(err)
         return err;
 
+    if(hdr.bitstream_id > 10)
+        return AC3_PARSE_ERROR_BSID;
+
     /* get decoding parameters from header info */
     s->bit_alloc_params.sr_code     = hdr.sr_code;
     s->channel_mode                 = hdr.channel_mode;
-    center_mix_level                  = gain_levels[center_levels[hdr.center_mix_level]];
-    surround_mix_level                = gain_levels[surround_levels[hdr.surround_mix_level]];
-    s->lfe_on                        = hdr.lfe_on;
+    s->lfe_on                       = hdr.lfe_on;
     s->bit_alloc_params.sr_shift    = hdr.sr_shift;
-    s->sampling_rate                = hdr.sample_rate;
+    s->sample_rate                  = hdr.sample_rate;
     s->bit_rate                     = hdr.bit_rate;
     s->channels                     = hdr.channels;
     s->fbw_channels                 = s->channels - s->lfe_on;
     s->lfe_ch                       = s->fbw_channels + 1;
     s->frame_size                   = hdr.frame_size;
-
-    /* set default output to all source channels */
-    s->out_channels = s->channels;
-    s->output_mode = s->channel_mode;
-    if(s->lfe_on)
-        s->output_mode |= AC3_OUTPUT_LFEON;
-
-    /* skip over portion of header which has already been read */
-    skip_bits(gbc, 16); // skip the sync_word
-    skip_bits(gbc, 16); // skip crc1
-    skip_bits(gbc, 8);  // skip fscod and frmsizecod
-    skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
-    if(s->channel_mode == AC3_CHMODE_STEREO) {
-        skip_bits(gbc, 2); // skip dsurmod
-    } else {
-        if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
-            skip_bits(gbc, 2); // skip cmixlev
-        if(s->channel_mode & 4)
-            skip_bits(gbc, 2); // skip surmixlev
+    s->center_mix_level             = hdr.center_mix_level;
+    s->surround_mix_level           = hdr.surround_mix_level;
+    s->num_blocks                   = hdr.num_blocks;
+    s->frame_type                   = hdr.frame_type;
+    s->substreamid                  = hdr.substreamid;
+
+    if(s->lfe_on) {
+        s->start_freq[s->lfe_ch] = 0;
+        s->end_freq[s->lfe_ch] = 7;
+        s->num_exp_groups[s->lfe_ch] = 2;
+        s->channel_in_cpl[s->lfe_ch] = 0;
     }
-    skip_bits1(gbc); // skip lfeon
 
     /* read the rest of the bsi. read twice for dual mono mode. */
     i = !(s->channel_mode);
@@ -394,25 +367,43 @@ static int ac3_parse_header(AC3DecodeContext *s)
         } while(i--);
     }
 
-    /* set stereo downmixing coefficients
-       reference: Section 7.8.2 Downmixing Into Two Channels */
+    return 0;
+}
+
+/**
+ * Set stereo downmixing coefficients based on frame header info.
+ * reference: Section 7.8.2 Downmixing Into Two Channels
+ */
+static void set_downmix_coeffs(AC3DecodeContext *s)
+{
+    int i;
+    float cmix = gain_levels[s->center_mix_level];
+    float smix = gain_levels[s->surround_mix_level];
+
     for(i=0; i<s->fbw_channels; i++) {
         s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
         s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
     }
     if(s->channel_mode > 1 && s->channel_mode & 1) {
-        s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = center_mix_level;
+        s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
     }
     if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
         int nf = s->channel_mode - 2;
-        s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = surround_mix_level * LEVEL_MINUS_3DB;
+        s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
     }
     if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
         int nf = s->channel_mode - 4;
-        s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = surround_mix_level;
+        s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
     }
 
-    return 0;
+    /* calculate adjustment needed for each channel to avoid clipping */
+    s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
+    for(i=0; i<s->fbw_channels; i++) {
+        s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
+        s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
+    }
+    s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
+    s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
 }
 
 /**
@@ -461,8 +452,11 @@ static void uncouple_channels(AC3DecodeContext *s)
             subbnd++;
             for(j=0; j<12; j++) {
                 for(ch=1; ch<=s->fbw_channels; ch++) {
-                    if(s->channel_in_cpl[ch])
-                        s->transform_coeffs[ch][i] = s->transform_coeffs[CPL_CH][i] * s->cpl_coords[ch][bnd] * 8.0f;
+                    if(s->channel_in_cpl[ch]) {
+                        s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
+                        if (ch == 2 && s->phase_flags[bnd])
+                            s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
+                    }
                 }
                 i++;
             }
@@ -474,9 +468,9 @@ static void uncouple_channels(AC3DecodeContext *s)
  * Grouped mantissas for 3-level 5-level and 11-level quantization
  */
 typedef struct {
-    float b1_mant[3];
-    float b2_mant[3];
-    float b4_mant[2];
+    int b1_mant[3];
+    int b2_mant[3];
+    int b4_mant[2];
     int b1ptr;
     int b2ptr;
     int b4ptr;
@@ -486,17 +480,17 @@ typedef struct {
  * Get the transform coefficients for a particular channel
  * reference: Section 7.3 Quantization and Decoding of Mantissas
  */
-static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
+static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
 {
     GetBitContext *gbc = &s->gbc;
     int i, gcode, tbap, start, end;
     uint8_t *exps;
     uint8_t *bap;
-    float *coeffs;
+    int *coeffs;
 
     exps = s->dexps[ch_index];
     bap = s->bap[ch_index];
-    coeffs = s->transform_coeffs[ch_index];
+    coeffs = s->fixed_coeffs[ch_index];
     start = s->start_freq[ch_index];
     end = s->end_freq[ch_index];
 
@@ -504,7 +498,7 @@ static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_group
         tbap = bap[i];
         switch (tbap) {
             case 0:
-                coeffs[i] = ((av_random(&s->dith_state) & 0xFFFF) / 65535.0f) - 0.5f;
+                coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
                 break;
 
             case 1:
@@ -547,15 +541,15 @@ static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_group
                 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
                 break;
 
-            default:
+            default: {
                 /* asymmetric dequantization */
-                coeffs[i] = get_sbits(gbc, quantization_tab[tbap]) * scale_factors[quantization_tab[tbap]-1];
+                int qlevel = quantization_tab[tbap];
+                coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
                 break;
+            }
         }
-        coeffs[i] *= scale_factors[exps[i]];
+        coeffs[i] >>= exps[i];
     }
-
-    return 0;
 }
 
 /**
@@ -565,26 +559,26 @@ static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_group
 static void remove_dithering(AC3DecodeContext *s) {
     int ch, i;
     int end=0;
-    float *coeffs;
+    int *coeffs;
     uint8_t *bap;
 
     for(ch=1; ch<=s->fbw_channels; ch++) {
         if(!s->dither_flag[ch]) {
-            coeffs = s->transform_coeffs[ch];
+            coeffs = s->fixed_coeffs[ch];
             bap = s->bap[ch];
             if(s->channel_in_cpl[ch])
                 end = s->start_freq[CPL_CH];
             else
                 end = s->end_freq[ch];
             for(i=0; i<end; i++) {
-                if(bap[i] == 0)
-                    coeffs[i] = 0.0f;
+                if(!bap[i])
+                    coeffs[i] = 0;
             }
             if(s->channel_in_cpl[ch]) {
                 bap = s->bap[CPL_CH];
                 for(; i<s->end_freq[CPL_CH]; i++) {
-                    if(bap[i] == 0)
-                        coeffs[i] = 0.0f;
+                    if(!bap[i])
+                        coeffs[i] = 0;
                 }
             }
         }
@@ -594,7 +588,7 @@ static void remove_dithering(AC3DecodeContext *s) {
 /**
  * Get the transform coefficients.
  */
-static int get_transform_coeffs(AC3DecodeContext *s)
+static void get_transform_coeffs(AC3DecodeContext *s)
 {
     int ch, end;
     int got_cplchan = 0;
@@ -604,16 +598,12 @@ static int get_transform_coeffs(AC3DecodeContext *s)
 
     for (ch = 1; ch <= s->channels; ch++) {
         /* transform coefficients for full-bandwidth channel */
-        if (get_transform_coeffs_ch(s, ch, &m))
-            return -1;
+        get_transform_coeffs_ch(s, ch, &m);
         /* tranform coefficients for coupling channel come right after the
            coefficients for the first coupled channel*/
         if (s->channel_in_cpl[ch])  {
             if (!got_cplchan) {
-                if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
-                    av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
-                    return -1;
-                }
+                get_transform_coeffs_ch(s, CPL_CH, &m);
                 uncouple_channels(s);
                 got_cplchan = 1;
             }
@@ -622,15 +612,13 @@ static int get_transform_coeffs(AC3DecodeContext *s)
             end = s->end_freq[ch];
         }
         do
-            s->transform_coeffs[ch][end] = 0;
+            s->fixed_coeffs[ch][end] = 0;
         while(++end < 256);
     }
 
     /* if any channel doesn't use dithering, zero appropriate coefficients */
     if(!s->dither_all)
         remove_dithering(s);
-
-    return 0;
 }
 
 /**
@@ -641,7 +629,7 @@ static void do_rematrixing(AC3DecodeContext *s)
 {
     int bnd, i;
     int end, bndend;
-    float tmp0, tmp1;
+    int tmp0, tmp1;
 
     end = FFMIN(s->end_freq[1], s->end_freq[2]);
 
@@ -649,10 +637,10 @@ static void do_rematrixing(AC3DecodeContext *s)
         if(s->rematrixing_flags[bnd]) {
             bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
             for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
-                tmp0 = s->transform_coeffs[1][i];
-                tmp1 = s->transform_coeffs[2][i];
-                s->transform_coeffs[1][i] = tmp0 + tmp1;
-                s->transform_coeffs[2][i] = tmp0 - tmp1;
+                tmp0 = s->fixed_coeffs[1][i];
+                tmp1 = s->fixed_coeffs[2][i];
+                s->fixed_coeffs[1][i] = tmp0 + tmp1;
+                s->fixed_coeffs[2][i] = tmp0 - tmp1;
             }
         }
     }
@@ -704,23 +692,16 @@ static void do_imdct_256(AC3DecodeContext *s, int chindex)
  * Convert frequency domain coefficients to time-domain audio samples.
  * reference: Section 7.9.4 Transformation Equations
  */
-static inline void do_imdct(AC3DecodeContext *s)
+static inline void do_imdct(AC3DecodeContext *s, int channels)
 {
     int ch;
-    int channels;
-
-    /* Don't perform the IMDCT on the LFE channel unless it's used in the output */
-    channels = s->fbw_channels;
-    if(s->output_mode & AC3_OUTPUT_LFEON)
-        channels++;
 
     for (ch=1; ch<=channels; ch++) {
         if (s->block_switch[ch]) {
             do_imdct_256(s, ch);
         } else {
             s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
-                                          s->transform_coeffs[ch],
-                                          s->tmp_imdct);
+                                        s->transform_coeffs[ch], s->tmp_imdct);
         }
         /* For the first half of the block, apply the window, add the delay
            from the previous block, and send to output */
@@ -729,38 +710,64 @@ static inline void do_imdct(AC3DecodeContext *s)
         /* For the second half of the block, apply the window and store the
            samples to delay, to be combined with the next block */
         s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
-                                     s->window, 256);
+                                   s->window, 256);
     }
 }
 
 /**
  * Downmix the output to mono or stereo.
  */
-static void ac3_downmix(float samples[AC3_MAX_CHANNELS][256], int fbw_channels,
-                        int output_mode, float coef[AC3_MAX_CHANNELS][2])
+static void ac3_downmix(AC3DecodeContext *s,
+                        float samples[AC3_MAX_CHANNELS][256], int ch_offset)
 {
     int i, j;
-    float v0, v1, s0, s1;
+    float v0, v1;
 
     for(i=0; i<256; i++) {
-        v0 = v1 = s0 = s1 = 0.0f;
-        for(j=0; j<fbw_channels; j++) {
-            v0 += samples[j][i] * coef[j][0];
-            v1 += samples[j][i] * coef[j][1];
-            s0 += coef[j][0];
-            s1 += coef[j][1];
+        v0 = v1 = 0.0f;
+        for(j=0; j<s->fbw_channels; j++) {
+            v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
+            v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
         }
-        v0 /= s0;
-        v1 /= s1;
-        if(output_mode == AC3_CHMODE_MONO) {
-            samples[0][i] = (v0 + v1) * LEVEL_MINUS_3DB;
-        } else if(output_mode == AC3_CHMODE_STEREO) {
-            samples[0][i] = v0;
-            samples[1][i] = v1;
+        v0 *= s->downmix_coeff_adjust[0];
+        v1 *= s->downmix_coeff_adjust[1];
+        if(s->output_mode == AC3_CHMODE_MONO) {
+            samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
+        } else if(s->output_mode == AC3_CHMODE_STEREO) {
+            samples[  ch_offset][i] = v0;
+            samples[1+ch_offset][i] = v1;
         }
     }
 }
 
+/**
+ * Upmix delay samples from stereo to original channel layout.
+ */
+static void ac3_upmix_delay(AC3DecodeContext *s)
+{
+    int channel_data_size = sizeof(s->delay[0]);
+    switch(s->channel_mode) {
+        case AC3_CHMODE_DUALMONO:
+        case AC3_CHMODE_STEREO:
+            /* upmix mono to stereo */
+            memcpy(s->delay[1], s->delay[0], channel_data_size);
+            break;
+        case AC3_CHMODE_2F2R:
+            memset(s->delay[3], 0, channel_data_size);
+        case AC3_CHMODE_2F1R:
+            memset(s->delay[2], 0, channel_data_size);
+            break;
+        case AC3_CHMODE_3F2R:
+            memset(s->delay[4], 0, channel_data_size);
+        case AC3_CHMODE_3F1R:
+            memset(s->delay[3], 0, channel_data_size);
+        case AC3_CHMODE_3F:
+            memcpy(s->delay[2], s->delay[1], channel_data_size);
+            memset(s->delay[1], 0, channel_data_size);
+            break;
+    }
+}
+
 /**
  * Parse an audio block from AC-3 bitstream.
  */
@@ -769,14 +776,20 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
     int fbw_channels = s->fbw_channels;
     int channel_mode = s->channel_mode;
     int i, bnd, seg, ch;
+    int different_transforms;
+    int downmix_output;
     GetBitContext *gbc = &s->gbc;
     uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
 
     memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
 
     /* block switch flags */
-    for (ch = 1; ch <= fbw_channels; ch++)
+    different_transforms = 0;
+    for (ch = 1; ch <= fbw_channels; ch++) {
         s->block_switch[ch] = get_bits1(gbc);
+        if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
+            different_transforms = 1;
+    }
 
     /* dithering flags */
     s->dither_all = 1;
@@ -791,7 +804,7 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
     do {
         if(get_bits1(gbc)) {
             s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
-                                    s->avctx->drc_scale)+1.0;
+                                  s->avctx->drc_scale)+1.0;
         } else if(blk == 0) {
             s->dynamic_range[i] = 1.0f;
         }
@@ -805,6 +818,11 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
             /* coupling in use */
             int cpl_begin_freq, cpl_end_freq;
 
+            if (channel_mode < AC3_CHMODE_STEREO) {
+                av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
+                return -1;
+            }
+
             /* determine which channels are coupled */
             for (ch = 1; ch <= fbw_channels; ch++)
                 s->channel_in_cpl[ch] = get_bits1(gbc);
@@ -829,11 +847,15 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
                     s->num_cpl_bands--;
                 }
             }
+            s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
         } else {
             /* coupling not in use */
             for (ch = 1; ch <= fbw_channels; ch++)
                 s->channel_in_cpl[ch] = 0;
         }
+    } else if (!blk) {
+        av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
+        return -1;
     }
 
     /* coupling coordinates */
@@ -850,32 +872,36 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
                         cpl_coord_exp = get_bits(gbc, 4);
                         cpl_coord_mant = get_bits(gbc, 4);
                         if (cpl_coord_exp == 15)
-                            s->cpl_coords[ch][bnd] = cpl_coord_mant / 16.0f;
+                            s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
                         else
-                            s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16.0f) / 32.0f;
-                        s->cpl_coords[ch][bnd] *= scale_factors[cpl_coord_exp + master_cpl_coord];
+                            s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
+                        s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
                     }
+                } else if (!blk) {
+                    av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
+                    return -1;
                 }
             }
         }
         /* phase flags */
-        if (channel_mode == AC3_CHMODE_STEREO && s->phase_flags_in_use && cpl_coords_exist) {
+        if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
             for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
-                if (get_bits1(gbc))
-                    s->cpl_coords[2][bnd] = -s->cpl_coords[2][bnd];
+                s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
             }
         }
     }
 
     /* stereo rematrixing strategy and band structure */
     if (channel_mode == AC3_CHMODE_STEREO) {
-        s->rematrixing_strategy = get_bits1(gbc);
-        if (s->rematrixing_strategy) {
+        if (get_bits1(gbc)) {
             s->num_rematrixing_bands = 4;
             if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
                 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
             for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
                 s->rematrixing_flags[bnd] = get_bits1(gbc);
+        } else if (!blk) {
+            av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
+            return -1;
         }
     }
 
@@ -883,10 +909,7 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
     s->exp_strategy[CPL_CH] = EXP_REUSE;
     s->exp_strategy[s->lfe_ch] = EXP_REUSE;
     for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
-        if(ch == s->lfe_ch)
-            s->exp_strategy[ch] = get_bits(gbc, 1);
-        else
-            s->exp_strategy[ch] = get_bits(gbc, 2);
+        s->exp_strategy[ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
         if(s->exp_strategy[ch] != EXP_REUSE)
             bit_alloc_stages[ch] = 3;
     }
@@ -895,6 +918,7 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
     for (ch = 1; ch <= fbw_channels; ch++) {
         s->start_freq[ch] = 0;
         if (s->exp_strategy[ch] != EXP_REUSE) {
+            int group_size;
             int prev = s->end_freq[ch];
             if (s->channel_in_cpl[ch])
                 s->end_freq[ch] = s->start_freq[CPL_CH];
@@ -906,26 +930,23 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
                 }
                 s->end_freq[ch] = bandwidth_code * 3 + 73;
             }
+            group_size = 3 << (s->exp_strategy[ch] - 1);
+            s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
             if(blk > 0 && s->end_freq[ch] != prev)
                 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
         }
     }
-    s->start_freq[s->lfe_ch] = 0;
-    s->end_freq[s->lfe_ch] = 7;
+    if (s->cpl_in_use && s->exp_strategy[CPL_CH] != EXP_REUSE) {
+        s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
+                                    (3 << (s->exp_strategy[CPL_CH] - 1));
+    }
 
     /* decode exponents for each channel */
     for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
         if (s->exp_strategy[ch] != EXP_REUSE) {
-            int group_size, num_groups;
-            group_size = 3 << (s->exp_strategy[ch] - 1);
-            if(ch == CPL_CH)
-                num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
-            else if(ch == s->lfe_ch)
-                num_groups = 2;
-            else
-                num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
             s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
-            decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
+            decode_exponents(gbc, s->exp_strategy[ch],
+                             s->num_exp_groups[ch], s->dexps[ch][0],
                              &s->dexps[ch][s->start_freq[ch]+!!ch]);
             if(ch != CPL_CH && ch != s->lfe_ch)
                 skip_bits(gbc, 2); /* skip gainrng */
@@ -939,9 +960,11 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
         s->bit_alloc_params.slow_gain  = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
         s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
         s->bit_alloc_params.floor  = ff_ac3_floor_tab[get_bits(gbc, 3)];
-        for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
+        for(ch=!s->cpl_in_use; ch<=s->channels; ch++)
             bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
-        }
+    } else if (!blk) {
+        av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
+        return -1;
     }
 
     /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
@@ -953,13 +976,21 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
             s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
         }
         memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
+    } else if (!blk) {
+        av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
+        return -1;
     }
 
     /* coupling leak information */
-    if (s->cpl_in_use && get_bits1(gbc)) {
-        s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
-        s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
-        bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
+    if (s->cpl_in_use) {
+        if (get_bits1(gbc)) {
+            s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
+            s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
+            bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
+        } else if (!blk) {
+            av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
+            return -1;
+        }
     }
 
     /* delta bit allocation information */
@@ -982,6 +1013,8 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
                     s->dba_lengths[ch][seg] = get_bits(gbc, 4);
                     s->dba_values[ch][seg] = get_bits(gbc, 3);
                 }
+                /* run last 2 bit allocation stages if new dba values */
+                bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
             }
         }
     } else if(blk == 0) {
@@ -1027,10 +1060,7 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
 
     /* unpack the transform coefficients
        this also uncouples channels if coupling is in use. */
-    if (get_transform_coeffs(s)) {
-        av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
-        return -1;
-    }
+    get_transform_coeffs(s);
 
     /* recover coefficients if rematrixing is in use */
     if(s->channel_mode == AC3_CHMODE_STEREO)
@@ -1038,24 +1068,47 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
 
     /* apply scaling to coefficients (headroom, dynrng) */
     for(ch=1; ch<=s->channels; ch++) {
-        float gain = 2.0f * s->mul_bias;
+        float gain = s->mul_bias / 4194304.0f;
         if(s->channel_mode == AC3_CHMODE_DUALMONO) {
             gain *= s->dynamic_range[ch-1];
         } else {
             gain *= s->dynamic_range[0];
         }
-        for(i=0; i<s->end_freq[ch]; i++) {
-            s->transform_coeffs[ch][i] *= gain;
+        for(i=0; i<256; i++) {
+            s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
         }
     }
 
-    do_imdct(s);
+    /* downmix and MDCT. order depends on whether block switching is used for
+       any channel in this block. this is because coefficients for the long
+       and short transforms cannot be mixed. */
+    downmix_output = s->channels != s->out_channels &&
+                     !((s->output_mode & AC3_OUTPUT_LFEON) &&
+                     s->fbw_channels == s->out_channels);
+    if(different_transforms) {
+        /* the delay samples have already been downmixed, so we upmix the delay
+           samples in order to reconstruct all channels before downmixing. */
+        if(s->downmixed) {
+            s->downmixed = 0;
+            ac3_upmix_delay(s);
+        }
+
+        do_imdct(s, s->channels);
 
-    /* downmix output if needed */
-    if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
-            s->fbw_channels == s->out_channels)) {
-        ac3_downmix(s->output, s->fbw_channels, s->output_mode,
-                    s->downmix_coeffs);
+        if(downmix_output) {
+            ac3_downmix(s, s->output, 0);
+        }
+    } else {
+        if(downmix_output) {
+            ac3_downmix(s, s->transform_coeffs, 1);
+        }
+
+        if(!s->downmixed) {
+            s->downmixed = 1;
+            ac3_downmix(s, s->delay, 0);
+        }
+
+        do_imdct(s, s->out_channels);
     }
 
     /* convert float to 16-bit integer */
@@ -1072,22 +1125,46 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
 /**
  * Decode a single AC-3 frame.
  */
-static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size, uint8_t *buf, int buf_size)
+static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
+                            const uint8_t *buf, int buf_size)
 {
-    AC3DecodeContext *s = (AC3DecodeContext *)avctx->priv_data;
+    AC3DecodeContext *s = avctx->priv_data;
     int16_t *out_samples = (int16_t *)data;
     int i, blk, ch, err;
 
     /* initialize the GetBitContext with the start of valid AC-3 Frame */
-    init_get_bits(&s->gbc, buf, buf_size * 8);
+    if (s->input_buffer) {
+        /* copy input buffer to decoder context to avoid reading past the end
+           of the buffer, which can be caused by a damaged input stream. */
+        memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
+        init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
+    } else {
+        init_get_bits(&s->gbc, buf, buf_size * 8);
+    }
 
     /* parse the syncinfo */
+    *data_size = 0;
     err = ac3_parse_header(s);
-    if(err) {
+
+    /* check that reported frame size fits in input buffer */
+    if(s->frame_size > buf_size) {
+        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
+        err = AC3_PARSE_ERROR_FRAME_SIZE;
+    }
+
+    /* check for crc mismatch */
+    if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
+        if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
+            av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
+            err = AC3_PARSE_ERROR_CRC;
+        }
+    }
+
+    if(err && err != AC3_PARSE_ERROR_CRC) {
         switch(err) {
             case AC3_PARSE_ERROR_SYNC:
                 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
-                break;
+                return -1;
             case AC3_PARSE_ERROR_BSID:
                 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
                 break;
@@ -1097,55 +1174,76 @@ static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
             case AC3_PARSE_ERROR_FRAME_SIZE:
                 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
                 break;
+            case AC3_PARSE_ERROR_FRAME_TYPE:
+                /* skip frame if CRC is ok. otherwise use error concealment. */
+                /* TODO: add support for substreams and dependent frames */
+                if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
+                    av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
+                    return s->frame_size;
+                } else {
+                av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
+                }
+                break;
             default:
                 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
                 break;
         }
-        return -1;
     }
 
-    avctx->sample_rate = s->sampling_rate;
-    avctx->bit_rate = s->bit_rate;
-
-    /* check that reported frame size fits in input buffer */
-    if(s->frame_size > buf_size) {
-        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
-        return -1;
-    }
+    /* if frame is ok, set audio parameters */
+    if (!err) {
+        avctx->sample_rate = s->sample_rate;
+        avctx->bit_rate = s->bit_rate;
+
+        /* channel config */
+        s->out_channels = s->channels;
+        s->output_mode = s->channel_mode;
+        if(s->lfe_on)
+            s->output_mode |= AC3_OUTPUT_LFEON;
+        if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
+                avctx->request_channels < s->channels) {
+            s->out_channels = avctx->request_channels;
+            s->output_mode  = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
+        }
+        avctx->channels = s->out_channels;
 
-    /* channel config */
-    s->out_channels = s->channels;
-    if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
-        avctx->request_channels < s->channels) {
-        s->out_channels = avctx->request_channels;
-        s->output_mode  = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
+        /* set downmixing coefficients if needed */
+        if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
+                s->fbw_channels == s->out_channels)) {
+            set_downmix_coeffs(s);
+        }
+    } else if (!s->out_channels) {
+        s->out_channels = avctx->channels;
+        if(s->out_channels < s->channels)
+            s->output_mode  = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
     }
-    avctx->channels = s->out_channels;
 
     /* parse the audio blocks */
-    for (blk = 0; blk < NB_BLOCKS; blk++) {
-        if (ac3_parse_audio_block(s, blk)) {
+    for (blk = 0; blk < s->num_blocks; blk++) {
+        if (!err && ac3_parse_audio_block(s, blk)) {
             av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
-            *data_size = 0;
-            return s->frame_size;
         }
+
+        /* interleave output samples */
         for (i = 0; i < 256; i++)
             for (ch = 0; ch < s->out_channels; ch++)
                 *(out_samples++) = s->int_output[ch][i];
     }
-    *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
+    *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
     return s->frame_size;
 }
 
 /**
  * Uninitialize the AC-3 decoder.
  */
-static int ac3_decode_end(AVCodecContext *avctx)
+static av_cold int ac3_decode_end(AVCodecContext *avctx)
 {
-    AC3DecodeContext *s = (AC3DecodeContext *)avctx->priv_data;
+    AC3DecodeContext *s = avctx->priv_data;
     ff_mdct_end(&s->imdct_512);
     ff_mdct_end(&s->imdct_256);
 
+    av_freep(&s->input_buffer);
+
     return 0;
 }
 
@@ -1157,4 +1255,5 @@ AVCodec ac3_decoder = {
     .init = ac3_decode_init,
     .close = ac3_decode_end,
     .decode = ac3_decode_frame,
+    .long_name = "ATSC A/52 / AC-3",
 };