]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/ac3dec.c
part 1 of EAC3 support
[ffmpeg] / libavcodec / ac3dec.c
index 6c725c43903cdfa1997fe93731e1436e9a0a0bc6..fc12f347fb4df550a009747a8c5d016e9281741a 100644 (file)
@@ -39,6 +39,9 @@
 #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
  * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
@@ -154,6 +157,7 @@ typedef struct {
     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
     int   cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
     int   num_cpl_bands;                    ///< number of coupling bands
@@ -170,6 +174,7 @@ typedef struct {
 
     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
@@ -178,9 +183,9 @@ 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
@@ -189,6 +194,7 @@ 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;
 
 /**
@@ -205,7 +211,7 @@ symmetric_dequant(int code, int levels)
 /*
  * Initialize tables at runtime.
  */
-static void ac3_tables_init(void)
+static av_cold void ac3_tables_init(void)
 {
     int i;
 
@@ -258,7 +264,7 @@ 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;
@@ -286,6 +292,14 @@ static int ac3_decode_init(AVCodecContext *avctx)
             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;
 }
@@ -402,6 +416,15 @@ static void set_downmix_coeffs(AC3DecodeContext *s)
         int nf = s->channel_mode - 4;
         s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
     }
+
+    /* 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];
 }
 
 /**
@@ -698,15 +721,9 @@ 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]) {
@@ -729,30 +746,57 @@ static inline void do_imdct(AC3DecodeContext *s)
 /**
  * Downmix the output to mono or stereo.
  */
-static void ac3_downmix(AC3DecodeContext *s)
+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;
+        v0 = v1 = 0.0f;
         for(j=0; j<s->fbw_channels; j++) {
-            v0 += s->output[j][i] * s->downmix_coeffs[j][0];
-            v1 += s->output[j][i] * s->downmix_coeffs[j][1];
-            s0 += s->downmix_coeffs[j][0];
-            s1 += s->downmix_coeffs[j][1];
+            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;
+        v0 *= s->downmix_coeff_adjust[0];
+        v1 *= s->downmix_coeff_adjust[1];
         if(s->output_mode == AC3_CHMODE_MONO) {
-            s->output[0][i] = (v0 + v1) * LEVEL_MINUS_3DB;
+            samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
         } else if(s->output_mode == AC3_CHMODE_STEREO) {
-            s->output[0][i] = v0;
-            s->output[1][i] = v1;
+            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.
  */
@@ -761,14 +805,20 @@ 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;
@@ -1040,12 +1090,36 @@ static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
         }
     }
 
-    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);
+        }
 
-    /* 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);
+        do_imdct(s, s->channels);
+
+        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 */
@@ -1062,14 +1136,22 @@ 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 = 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 */
     err = ac3_parse_header(s);
@@ -1087,6 +1169,9 @@ 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:
+                av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
+                break;
             default:
                 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
                 break;
@@ -1145,12 +1230,14 @@ static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
 /**
  * Uninitialize the AC-3 decoder.
  */
-static int ac3_decode_end(AVCodecContext *avctx)
+static av_cold int ac3_decode_end(AVCodecContext *avctx)
 {
     AC3DecodeContext *s = avctx->priv_data;
     ff_mdct_end(&s->imdct_512);
     ff_mdct_end(&s->imdct_256);
 
+    av_freep(&s->input_buffer);
+
     return 0;
 }