Merge remote-tracking branch 'qatar/master'
[ffmpeg.git] / libavcodec / alac.c
index 942f686..83e0d81 100644 (file)
  */
 
 /**
- * @file alac.c
+ * @file
  * ALAC (Apple Lossless Audio Codec) decoder
  * @author 2005 David Hammerton
+ * @see http://crazney.net/programs/itunes/alac.html
  *
- * For more information on the ALAC format, visit:
- *  http://crazney.net/programs/itunes/alac.html
- *
- * Note: This decoder expects a 36- (0x24-)byte QuickTime atom to be
+ * Note: This decoder expects a 36-byte QuickTime atom to be
  * passed through the extradata[_size] fields. This atom is tacked onto
  * the end of an 'alac' stsd atom and has the following format:
- *  bytes 0-3   atom size (0x24), big-endian
- *  bytes 4-7   atom type ('alac', not the 'alac' tag from start of stsd)
- *  bytes 8-35  data bytes needed by decoder
  *
- * Extradata:
- * 32bit  size
- * 32bit  tag (=alac)
- * 32bit  zero?
- * 32bit  max sample per frame
- *  8bit  ?? (zero?)
+ * 32bit  atom size
+ * 32bit  tag                  ("alac")
+ * 32bit  tag version          (0)
+ * 32bit  samples per frame    (used when not set explicitly in the frames)
+ *  8bit  compatible version   (0)
  *  8bit  sample size
- *  8bit  history mult
- *  8bit  initial history
- *  8bit  kmodifier
- *  8bit  channels?
- * 16bit  ??
- * 32bit  max coded frame size
- * 32bit  bitrate?
+ *  8bit  history mult         (40)
+ *  8bit  initial history      (14)
+ *  8bit  kmodifier            (10)
+ *  8bit  channels
+ * 16bit  maxRun               (255)
+ * 32bit  max coded frame size (0 means unknown)
+ * 32bit  average bitrate      (0 means unknown)
  * 32bit  samplerate
  */
 
 
 #include "avcodec.h"
-#include "bitstream.h"
+#include "get_bits.h"
 #include "bytestream.h"
+#include "unary.h"
+#include "mathops.h"
 
 #define ALAC_EXTRADATA_SIZE 36
 #define MAX_CHANNELS 2
 typedef struct {
 
     AVCodecContext *avctx;
+    AVFrame frame;
     GetBitContext gb;
-    /* init to 0; first frame decode should initialize from extradata and
-     * set this to 1 */
-    int context_initialized;
 
-    int samplesize;
     int numchannels;
-    int bytespersample;
 
     /* buffers */
     int32_t *predicterror_buffer[MAX_CHANNELS];
 
     int32_t *outputsamples_buffer[MAX_CHANNELS];
 
+    int32_t *extra_bits_buffer[MAX_CHANNELS];
+
     /* stuff from setinfo */
     uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */    /* max samples per frame? */
-    uint8_t setinfo_7a; /* 0x00 */
     uint8_t setinfo_sample_size; /* 0x10 */
     uint8_t setinfo_rice_historymult; /* 0x28 */
     uint8_t setinfo_rice_initialhistory; /* 0x0a */
     uint8_t setinfo_rice_kmodifier; /* 0x0e */
-    uint8_t setinfo_7f; /* 0x02 */
-    uint16_t setinfo_80; /* 0x00ff */
-    uint32_t setinfo_82; /* 0x000020e7 */ /* max sample size?? */
-    uint32_t setinfo_86; /* 0x00069fe4 */ /* bit rate (average)?? */
-    uint32_t setinfo_8a_rate; /* 0x0000ac44 */
     /* end setinfo stuff */
 
+    int extra_bits;                         /**< number of extra bits beyond 16-bit */
 } ALACContext;
 
-static void allocate_buffers(ALACContext *alac)
-{
-    int chan;
-    for (chan = 0; chan < MAX_CHANNELS; chan++) {
-        alac->predicterror_buffer[chan] =
-            av_malloc(alac->setinfo_max_samples_per_frame * 4);
+static inline int decode_scalar(GetBitContext *gb, int k, int limit, int readsamplesize){
+    /* read x - number of 1s before 0 represent the rice */
+    int x = get_unary_0_9(gb);
 
-        alac->outputsamples_buffer[chan] =
-            av_malloc(alac->setinfo_max_samples_per_frame * 4);
-    }
-}
+    if (x > 8) { /* RICE THRESHOLD */
+        /* use alternative encoding */
+        x = get_bits(gb, readsamplesize);
+    } else {
+        if (k >= limit)
+            k = limit;
 
-static int alac_set_info(ALACContext *alac)
-{
-    unsigned char *ptr = alac->avctx->extradata;
+        if (k != 1) {
+            int extrabits = show_bits(gb, k);
 
-    ptr += 4; /* size */
-    ptr += 4; /* alac */
-    ptr += 4; /* 0 ? */
+            /* multiply x by 2^k - 1, as part of their strange algorithm */
+            x = (x << k) - x;
 
-    if(AV_RB32(ptr) >= UINT_MAX/4){
-        av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
-        return -1;
+            if (extrabits > 1) {
+                x += extrabits - 1;
+                skip_bits(gb, k);
+            } else
+                skip_bits(gb, k - 1);
+        }
     }
-
-    /* buffer size / 2 ? */
-    alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
-    alac->setinfo_7a                    = *ptr++;
-    alac->setinfo_sample_size           = *ptr++;
-    alac->setinfo_rice_historymult      = *ptr++;
-    alac->setinfo_rice_initialhistory   = *ptr++;
-    alac->setinfo_rice_kmodifier        = *ptr++;
-    /* channels? */
-    alac->setinfo_7f                    = *ptr++;
-    alac->setinfo_80                    = bytestream_get_be16(&ptr);
-    /* max coded frame size */
-    alac->setinfo_82                    = bytestream_get_be32(&ptr);
-    /* bitrate ? */
-    alac->setinfo_86                    = bytestream_get_be32(&ptr);
-    /* samplerate */
-    alac->setinfo_8a_rate               = bytestream_get_be32(&ptr);
-
-    allocate_buffers(alac);
-
-    return 0;
-}
-
-static inline int count_leading_zeros(int32_t input)
-{
-    return 31-av_log2(input);
+    return x;
 }
 
-static void bastardized_rice_decompress(ALACContext *alac,
+static int bastardized_rice_decompress(ALACContext *alac,
                                  int32_t *output_buffer,
                                  int output_size,
                                  int readsamplesize, /* arg_10 */
@@ -159,53 +123,19 @@ static void bastardized_rice_decompress(ALACContext *alac,
     int sign_modifier = 0;
 
     for (output_count = 0; output_count < output_size; output_count++) {
-        int32_t x = 0;
+        int32_t x;
         int32_t x_modified;
         int32_t final_val;
 
-        /* read x - number of 1s before 0 represent the rice */
-        while (x <= 8 && get_bits1(&alac->gb)) {
-            x++;
-        }
-
-
-        if (x > 8) { /* RICE THRESHOLD */
-            /* use alternative encoding */
-            int32_t value;
-
-            value = get_bits(&alac->gb, readsamplesize);
+        /* standard rice encoding */
+        int k; /* size of extra bits */
 
-            /* mask value to readsamplesize size */
-            if (readsamplesize != 32)
-                value &= (0xffffffff >> (32 - readsamplesize));
+        if(get_bits_left(&alac->gb) <= 0)
+            return -1;
 
-            x = value;
-        } else {
-            /* standard rice encoding */
-            int extrabits;
-            int k; /* size of extra bits */
-
-            /* read k, that is bits as is */
-            k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3);
-
-            if (k < 0)
-                k += rice_kmodifier;
-            else
-                k = rice_kmodifier;
-
-            if (k != 1) {
-                extrabits = show_bits(&alac->gb, k);
-
-                /* multiply x by 2^k - 1, as part of their strange algorithm */
-                x = (x << k) - x;
-
-                if (extrabits > 1) {
-                    x += extrabits - 1;
-                    get_bits(&alac->gb, k);
-                } else
-                    get_bits(&alac->gb, k - 1);
-            }
-        }
+        /* read k, that is bits as is */
+        k = av_log2((history >> 9) + 3);
+        x= decode_scalar(&alac->gb, k, rice_kmodifier, readsamplesize);
 
         x_modified = sign_modifier + x;
         final_val = (x_modified + 1) / 2;
@@ -224,39 +154,20 @@ static void bastardized_rice_decompress(ALACContext *alac,
 
         /* special case: there may be compressed blocks of 0 */
         if ((history < 128) && (output_count+1 < output_size)) {
-            int block_size;
+            int k;
+            unsigned int block_size;
 
             sign_modifier = 1;
 
-            x = 0;
-            while (x <= 8 && get_bits1(&alac->gb)) {
-                x++;
-            }
-
-            if (x > 8) {
-                block_size = get_bits(&alac->gb, 16);
-                block_size &= 0xffff;
-            } else {
-                int k;
-                int extrabits;
+            k = 7 - av_log2(history) + ((history + 16) >> 6 /* / 64 */);
 
-                k = count_leading_zeros(history) + ((history + 16) >> 6 /* / 64 */) - 24;
-
-                extrabits = show_bits(&alac->gb, k);
-
-                block_size = (((1 << k) - 1) & rice_kmodifier_mask) * x
-                           + extrabits - 1;
-
-                if (extrabits < 2) {
-                    x = 1 - extrabits;
-                    block_size += x;
-                    get_bits(&alac->gb, k - 1);
-                } else {
-                    get_bits(&alac->gb, k);
-                }
-            }
+            block_size= decode_scalar(&alac->gb, k, rice_kmodifier, 16);
 
             if (block_size > 0) {
+                if(block_size >= output_size - output_count){
+                    av_log(alac->avctx, AV_LOG_ERROR, "invalid zero block size of %d %d %d\n", block_size, output_size, output_count);
+                    block_size= output_size - output_count - 1;
+                }
                 memset(&output_buffer[output_count+1], 0, block_size * 4);
                 output_count += block_size;
             }
@@ -267,14 +178,13 @@ static void bastardized_rice_decompress(ALACContext *alac,
             history = 0;
         }
     }
+    return 0;
 }
 
-#define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits))
-
-#define SIGN_ONLY(v) \
-                     ((v < 0) ? (-1) : \
-                                ((v > 0) ? (1) : \
-                                           (0)))
+static inline int sign_only(int v)
+{
+    return v ? FFSIGN(v) : 0;
+}
 
 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
                                            int32_t *buffer_out,
@@ -310,7 +220,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
             prev_value = buffer_out[i];
             error_value = error_buffer[i+1];
             buffer_out[i+1] =
-                SIGN_EXTENDED32((prev_value + error_value), readsamplesize);
+                sign_extend((prev_value + error_value), readsamplesize);
         }
         return;
     }
@@ -321,24 +231,13 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
             int32_t val;
 
             val = buffer_out[i] + error_buffer[i+1];
-            val = SIGN_EXTENDED32(val, readsamplesize);
+            val = sign_extend(val, readsamplesize);
             buffer_out[i+1] = val;
         }
 
-#if 0
     /* 4 and 8 are very common cases (the only ones i've seen). these
-     * should be unrolled and optimised
+     * should be unrolled and optimized
      */
-    if (predictor_coef_num == 4) {
-        /* FIXME: optimised general case */
-        return;
-    }
-
-    if (predictor_coef_table == 8) {
-        /* FIXME: optimised general case */
-        return;
-    }
-#endif
 
     /* general case */
     if (predictor_coef_num > 0) {
@@ -356,7 +255,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
             outval = (1 << (predictor_quantitization-1)) + sum;
             outval = outval >> predictor_quantitization;
             outval = outval + buffer_out[0] + error_val;
-            outval = SIGN_EXTENDED32(outval, readsamplesize);
+            outval = sign_extend(outval, readsamplesize);
 
             buffer_out[predictor_coef_num+1] = outval;
 
@@ -365,7 +264,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
 
                 while (predictor_num >= 0 && error_val > 0) {
                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
-                    int sign = SIGN_ONLY(val);
+                    int sign = sign_only(val);
 
                     predictor_coef_table[predictor_num] -= sign;
 
@@ -381,7 +280,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
 
                 while (predictor_num >= 0 && error_val < 0) {
                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
-                    int sign = - SIGN_ONLY(val);
+                    int sign = - sign_only(val);
 
                     predictor_coef_table[predictor_num] -= sign;
 
@@ -399,269 +298,399 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer,
     }
 }
 
-static void deinterlace_16(int32_t *buffer[MAX_CHANNELS],
-                           int16_t *buffer_out,
-                           int numchannels, int numsamples,
-                           uint8_t interlacing_shift,
-                           uint8_t interlacing_leftweight)
+static void decorrelate_stereo(int32_t *buffer[MAX_CHANNELS],
+                               int numsamples, uint8_t interlacing_shift,
+                               uint8_t interlacing_leftweight)
 {
     int i;
-    if (numsamples <= 0)
-        return;
 
-    /* weighted interlacing */
-    if (interlacing_leftweight) {
-        for (i = 0; i < numsamples; i++) {
-            int32_t difference, midright;
-            int16_t left;
-            int16_t right;
+    for (i = 0; i < numsamples; i++) {
+        int32_t a, b;
 
-            midright = buffer[0][i];
-            difference = buffer[1][i];
+        a = buffer[0][i];
+        b = buffer[1][i];
 
+        a -= (b * interlacing_leftweight) >> interlacing_shift;
+        b += a;
 
-            right = midright - ((difference * interlacing_leftweight) >> interlacing_shift);
-            left = right + difference;
+        buffer[0][i] = b;
+        buffer[1][i] = a;
+    }
+}
 
-            buffer_out[i*numchannels] = left;
-            buffer_out[i*numchannels + 1] = right;
-        }
+static void append_extra_bits(int32_t *buffer[MAX_CHANNELS],
+                              int32_t *extra_bits_buffer[MAX_CHANNELS],
+                              int extra_bits, int numchannels, int numsamples)
+{
+    int i, ch;
 
-        return;
+    for (ch = 0; ch < numchannels; ch++)
+        for (i = 0; i < numsamples; i++)
+            buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
+}
+
+static void interleave_stereo_16(int32_t *buffer[MAX_CHANNELS],
+                                 int16_t *buffer_out, int numsamples)
+{
+    int i;
+
+    for (i = 0; i < numsamples; i++) {
+        *buffer_out++ = buffer[0][i];
+        *buffer_out++ = buffer[1][i];
     }
+}
+
+static void interleave_stereo_24(int32_t *buffer[MAX_CHANNELS],
+                                 int32_t *buffer_out, int numsamples)
+{
+    int i;
 
-    /* otherwise basic interlacing took place */
     for (i = 0; i < numsamples; i++) {
-        int16_t left, right;
+        *buffer_out++ = buffer[0][i] << 8;
+        *buffer_out++ = buffer[1][i] << 8;
+    }
+}
 
-        left = buffer[0][i];
-        right = buffer[1][i];
+static void interleave_stereo_32(int32_t *buffer[MAX_CHANNELS],
+                                 int32_t *buffer_out, int numsamples)
+{
+    int i;
 
-        buffer_out[i*numchannels] = left;
-        buffer_out[i*numchannels + 1] = right;
+    for (i = 0; i < numsamples; i++) {
+        *buffer_out++ = buffer[0][i];
+        *buffer_out++ = buffer[1][i];
     }
 }
 
-static int alac_decode_frame(AVCodecContext *avctx,
-                             void *outbuffer, int *outputsize,
-                             uint8_t *inbuffer, int input_buffer_size)
+static int alac_decode_frame(AVCodecContext *avctx, void *data,
+                             int *got_frame_ptr, AVPacket *avpkt)
 {
+    const uint8_t *inbuffer = avpkt->data;
+    int input_buffer_size = avpkt->size;
     ALACContext *alac = avctx->priv_data;
 
     int channels;
-    int32_t outputsamples;
+    unsigned int outputsamples;
     int hassize;
-    int readsamplesize;
-    int wasted_bytes;
+    unsigned int readsamplesize;
     int isnotcompressed;
     uint8_t interlacing_shift;
     uint8_t interlacing_leftweight;
-
-    /* short-circuit null buffers */
-    if (!inbuffer || !input_buffer_size)
-        return input_buffer_size;
-
-    /* initialize from the extradata */
-    if (!alac->context_initialized) {
-        if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
-            av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
-                ALAC_EXTRADATA_SIZE);
-            return input_buffer_size;
-        }
-        if (alac_set_info(alac)) {
-            av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
-            return input_buffer_size;
-        }
-        alac->context_initialized = 1;
-    }
+    int i, ch, ret;
 
     init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
 
     channels = get_bits(&alac->gb, 3) + 1;
-    if (channels > MAX_CHANNELS) {
-        av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
-               MAX_CHANNELS);
-        return input_buffer_size;
+    if (channels != avctx->channels) {
+        av_log(avctx, AV_LOG_ERROR, "frame header channel count mismatch\n");
+        return AVERROR_INVALIDDATA;
     }
 
     /* 2^result = something to do with output waiting.
      * perhaps matters if we read > 1 frame in a pass?
      */
-    get_bits(&alac->gb, 4);
+    skip_bits(&alac->gb, 4);
 
-    get_bits(&alac->gb, 12); /* unknown, skip 12 bits */
+    skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */
 
     /* the output sample size is stored soon */
-    hassize = get_bits(&alac->gb, 1);
+    hassize = get_bits1(&alac->gb);
 
-    wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
+    alac->extra_bits = get_bits(&alac->gb, 2) << 3;
 
     /* whether the frame is compressed */
-    isnotcompressed = get_bits(&alac->gb, 1);
+    isnotcompressed = get_bits1(&alac->gb);
 
     if (hassize) {
         /* now read the number of samples as a 32bit integer */
-        outputsamples = get_bits(&alac->gb, 32);
+        outputsamples = get_bits_long(&alac->gb, 32);
+        if(outputsamples > alac->setinfo_max_samples_per_frame){
+            av_log(avctx, AV_LOG_ERROR, "outputsamples %d > %d\n", outputsamples, alac->setinfo_max_samples_per_frame);
+            return -1;
+        }
     } else
         outputsamples = alac->setinfo_max_samples_per_frame;
 
-    *outputsize = outputsamples * alac->bytespersample;
-    readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
+    /* get output buffer */
+    if (outputsamples > INT32_MAX) {
+        av_log(avctx, AV_LOG_ERROR, "unsupported block size: %u\n", outputsamples);
+        return AVERROR_INVALIDDATA;
+    }
+    alac->frame.nb_samples = outputsamples;
+    if ((ret = avctx->get_buffer(avctx, &alac->frame)) < 0) {
+        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+        return ret;
+    }
+
+    readsamplesize = alac->setinfo_sample_size - alac->extra_bits + channels - 1;
+    if (readsamplesize > MIN_CACHE_BITS) {
+        av_log(avctx, AV_LOG_ERROR, "readsamplesize too big (%d)\n", readsamplesize);
+        return -1;
+    }
 
     if (!isnotcompressed) {
         /* so it is compressed */
-        int16_t predictor_coef_table[channels][32];
-        int predictor_coef_num[channels];
-        int prediction_type[channels];
-        int prediction_quantitization[channels];
-        int ricemodifier[channels];
-        int i, chan;
+        int16_t predictor_coef_table[MAX_CHANNELS][32];
+        int predictor_coef_num[MAX_CHANNELS];
+        int prediction_type[MAX_CHANNELS];
+        int prediction_quantitization[MAX_CHANNELS];
+        int ricemodifier[MAX_CHANNELS];
 
         interlacing_shift = get_bits(&alac->gb, 8);
         interlacing_leftweight = get_bits(&alac->gb, 8);
 
-        for (chan = 0; chan < channels; chan++) {
-            prediction_type[chan] = get_bits(&alac->gb, 4);
-            prediction_quantitization[chan] = get_bits(&alac->gb, 4);
+        for (ch = 0; ch < channels; ch++) {
+            prediction_type[ch] = get_bits(&alac->gb, 4);
+            prediction_quantitization[ch] = get_bits(&alac->gb, 4);
 
-            ricemodifier[chan] = get_bits(&alac->gb, 3);
-            predictor_coef_num[chan] = get_bits(&alac->gb, 5);
+            ricemodifier[ch] = get_bits(&alac->gb, 3);
+            predictor_coef_num[ch] = get_bits(&alac->gb, 5);
 
             /* read the predictor table */
-            for (i = 0; i < predictor_coef_num[chan]; i++)
-                predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
+            for (i = 0; i < predictor_coef_num[ch]; i++)
+                predictor_coef_table[ch][i] = (int16_t)get_bits(&alac->gb, 16);
         }
 
-        if (wasted_bytes) {
-            av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
+        if (alac->extra_bits) {
+            for (i = 0; i < outputsamples; i++) {
+                if(get_bits_left(&alac->gb) <= 0)
+                    return -1;
+                for (ch = 0; ch < channels; ch++)
+                    alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
+            }
         }
-
-        for (chan = 0; chan < channels; chan++) {
-            bastardized_rice_decompress(alac,
-                                        alac->predicterror_buffer[chan],
+        for (ch = 0; ch < channels; ch++) {
+            int ret = bastardized_rice_decompress(alac,
+                                        alac->predicterror_buffer[ch],
                                         outputsamples,
                                         readsamplesize,
                                         alac->setinfo_rice_initialhistory,
                                         alac->setinfo_rice_kmodifier,
-                                        ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
+                                        ricemodifier[ch] * alac->setinfo_rice_historymult / 4,
                                         (1 << alac->setinfo_rice_kmodifier) - 1);
-
-            if (prediction_type[chan] == 0) {
-                /* adaptive fir */
-                predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
-                                               alac->outputsamples_buffer[chan],
-                                               outputsamples,
-                                               readsamplesize,
-                                               predictor_coef_table[chan],
-                                               predictor_coef_num[chan],
-                                               prediction_quantitization[chan]);
-            } else {
-                av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
-                /* I think the only other prediction type (or perhaps this is
-                 * just a boolean?) runs adaptive fir twice.. like:
-                 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
-                 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
-                 * little strange..
+            if(ret<0)
+                return ret;
+
+            /* adaptive FIR filter */
+            if (prediction_type[ch] == 15) {
+                /* Prediction type 15 runs the adaptive FIR twice.
+                 * The first pass uses the special-case coef_num = 31, while
+                 * the second pass uses the coefs from the bitstream.
+                 *
+                 * However, this prediction type is not currently used by the
+                 * reference encoder.
                  */
+                predictor_decompress_fir_adapt(alac->predicterror_buffer[ch],
+                                               alac->predicterror_buffer[ch],
+                                               outputsamples, readsamplesize,
+                                               NULL, 31, 0);
+            } else if (prediction_type[ch] > 0) {
+                av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
+                       prediction_type[ch]);
             }
+            predictor_decompress_fir_adapt(alac->predicterror_buffer[ch],
+                                           alac->outputsamples_buffer[ch],
+                                           outputsamples, readsamplesize,
+                                           predictor_coef_table[ch],
+                                           predictor_coef_num[ch],
+                                           prediction_quantitization[ch]);
         }
     } else {
         /* not compressed, easy case */
-        if (alac->setinfo_sample_size <= 16) {
-            int i, chan;
-            for (chan = 0; chan < channels; chan++)
-                for (i = 0; i < outputsamples; i++) {
-                    int32_t audiobits;
-
-                    audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
-                    audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
-
-                    alac->outputsamples_buffer[chan][i] = audiobits;
-                }
-        } else {
-            int i, chan;
-            for (chan = 0; chan < channels; chan++)
-                for (i = 0; i < outputsamples; i++) {
-                    int32_t audiobits;
-
-                    audiobits = get_bits(&alac->gb, 16);
-                    /* special case of sign extension..
-                     * as we'll be ORing the low 16bits into this */
-                    audiobits = audiobits << 16;
-                    audiobits = audiobits >> (32 - alac->setinfo_sample_size);
-                    audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
-
-                    alac->outputsamples_buffer[chan][i] = audiobits;
-                }
+        for (i = 0; i < outputsamples; i++) {
+            if(get_bits_left(&alac->gb) <= 0)
+                return -1;
+            for (ch = 0; ch < channels; ch++) {
+                alac->outputsamples_buffer[ch][i] = get_sbits_long(&alac->gb,
+                                                                   alac->setinfo_sample_size);
+            }
         }
-        /* wasted_bytes = 0; */
+        alac->extra_bits = 0;
         interlacing_shift = 0;
         interlacing_leftweight = 0;
     }
+    if (get_bits(&alac->gb, 3) != 7)
+        av_log(avctx, AV_LOG_ERROR, "Error : Wrong End Of Frame\n");
+
+    if (channels == 2 && interlacing_leftweight) {
+        decorrelate_stereo(alac->outputsamples_buffer, outputsamples,
+                           interlacing_shift, interlacing_leftweight);
+    }
+
+    if (alac->extra_bits) {
+        append_extra_bits(alac->outputsamples_buffer, alac->extra_bits_buffer,
+                          alac->extra_bits, alac->numchannels, outputsamples);
+    }
 
     switch(alac->setinfo_sample_size) {
-    case 16: {
+    case 16:
         if (channels == 2) {
-            deinterlace_16(alac->outputsamples_buffer,
-                           (int16_t*)outbuffer,
-                           alac->numchannels,
-                           outputsamples,
-                           interlacing_shift,
-                           interlacing_leftweight);
+            interleave_stereo_16(alac->outputsamples_buffer,
+                                 (int16_t *)alac->frame.data[0], outputsamples);
         } else {
-            int i;
+            int16_t *outbuffer = (int16_t *)alac->frame.data[0];
             for (i = 0; i < outputsamples; i++) {
-                int16_t sample = alac->outputsamples_buffer[0][i];
-                ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
+                outbuffer[i] = alac->outputsamples_buffer[0][i];
             }
         }
         break;
-    }
-    case 20:
     case 24:
-    case 32:
-        av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
+        if (channels == 2) {
+            interleave_stereo_24(alac->outputsamples_buffer,
+                                 (int32_t *)alac->frame.data[0], outputsamples);
+        } else {
+            int32_t *outbuffer = (int32_t *)alac->frame.data[0];
+            for (i = 0; i < outputsamples; i++)
+                outbuffer[i] = alac->outputsamples_buffer[0][i] << 8;
+        }
         break;
-    default:
+    case 32:
+        if (channels == 2) {
+            interleave_stereo_32(alac->outputsamples_buffer,
+                                 (int32_t *)alac->frame.data[0], outputsamples);
+        } else {
+            int32_t *outbuffer = (int32_t *)alac->frame.data[0];
+            for (i = 0; i < outputsamples; i++)
+                outbuffer[i] = alac->outputsamples_buffer[0][i];
+        }
         break;
     }
 
+    if (input_buffer_size * 8 - get_bits_count(&alac->gb) > 8)
+        av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n", input_buffer_size * 8 - get_bits_count(&alac->gb));
+
+    *got_frame_ptr   = 1;
+    *(AVFrame *)data = alac->frame;
+
     return input_buffer_size;
 }
 
-static int alac_decode_init(AVCodecContext * avctx)
+static av_cold int alac_decode_close(AVCodecContext *avctx)
 {
     ALACContext *alac = avctx->priv_data;
-    alac->avctx = avctx;
-    alac->context_initialized = 0;
 
-    alac->samplesize = alac->avctx->bits_per_sample;
-    alac->numchannels = alac->avctx->channels;
-    alac->bytespersample = (alac->samplesize / 8) * alac->numchannels;
+    int ch;
+    for (ch = 0; ch < alac->numchannels; ch++) {
+        av_freep(&alac->predicterror_buffer[ch]);
+        av_freep(&alac->outputsamples_buffer[ch]);
+        av_freep(&alac->extra_bits_buffer[ch]);
+    }
+
+    return 0;
+}
+
+static int allocate_buffers(ALACContext *alac)
+{
+    int ch;
+    for (ch = 0; ch < alac->numchannels; ch++) {
+        int buf_size = alac->setinfo_max_samples_per_frame * sizeof(int32_t);
+
+        FF_ALLOC_OR_GOTO(alac->avctx, alac->predicterror_buffer[ch],
+                         buf_size, buf_alloc_fail);
+
+        FF_ALLOC_OR_GOTO(alac->avctx, alac->outputsamples_buffer[ch],
+                         buf_size, buf_alloc_fail);
+
+        FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
+                         buf_size, buf_alloc_fail);
+    }
+    return 0;
+buf_alloc_fail:
+    alac_decode_close(alac->avctx);
+    return AVERROR(ENOMEM);
+}
+
+static int alac_set_info(ALACContext *alac)
+{
+    const unsigned char *ptr = alac->avctx->extradata;
+
+    ptr += 4; /* size */
+    ptr += 4; /* alac */
+    ptr += 4; /* version */
+
+    if(AV_RB32(ptr) >= UINT_MAX/4){
+        av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
+        return -1;
+    }
+
+    /* buffer size / 2 ? */
+    alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
+    ptr++;                          /* compatible version */
+    alac->setinfo_sample_size           = *ptr++;
+    alac->setinfo_rice_historymult      = *ptr++;
+    alac->setinfo_rice_initialhistory   = *ptr++;
+    alac->setinfo_rice_kmodifier        = *ptr++;
+    alac->numchannels                   = *ptr++;
+    bytestream_get_be16(&ptr);      /* maxRun */
+    bytestream_get_be32(&ptr);      /* max coded frame size */
+    bytestream_get_be32(&ptr);      /* average bitrate */
+    bytestream_get_be32(&ptr);      /* samplerate */
 
     return 0;
 }
 
-static int alac_decode_close(AVCodecContext *avctx)
+static av_cold int alac_decode_init(AVCodecContext * avctx)
 {
+    int ret;
     ALACContext *alac = avctx->priv_data;
+    alac->avctx = avctx;
+
+    /* initialize from the extradata */
+    if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
+        av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
+            ALAC_EXTRADATA_SIZE);
+        return -1;
+    }
+    if (alac_set_info(alac)) {
+        av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
+        return -1;
+    }
+
+    switch (alac->setinfo_sample_size) {
+    case 16: avctx->sample_fmt    = AV_SAMPLE_FMT_S16;
+             break;
+    case 32:
+    case 24: avctx->sample_fmt    = AV_SAMPLE_FMT_S32;
+             break;
+    default: av_log_ask_for_sample(avctx, "Sample depth %d is not supported.\n",
+                                   alac->setinfo_sample_size);
+             return AVERROR_PATCHWELCOME;
+    }
 
-    int chan;
-    for (chan = 0; chan < MAX_CHANNELS; chan++) {
-        av_free(alac->predicterror_buffer[chan]);
-        av_free(alac->outputsamples_buffer[chan]);
+    if (alac->numchannels < 1) {
+        av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
+        alac->numchannels = avctx->channels;
+    } else {
+        if (alac->numchannels > MAX_CHANNELS)
+            alac->numchannels = avctx->channels;
+        else
+            avctx->channels = alac->numchannels;
+    }
+    if (avctx->channels > MAX_CHANNELS) {
+        av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
+               avctx->channels);
+        return AVERROR_PATCHWELCOME;
     }
 
+    if ((ret = allocate_buffers(alac)) < 0) {
+        av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
+        return ret;
+    }
+
+    avcodec_get_frame_defaults(&alac->frame);
+    avctx->coded_frame = &alac->frame;
+
     return 0;
 }
 
-AVCodec alac_decoder = {
-    "alac",
-    CODEC_TYPE_AUDIO,
-    CODEC_ID_ALAC,
-    sizeof(ALACContext),
-    alac_decode_init,
-    NULL,
-    alac_decode_close,
-    alac_decode_frame,
+AVCodec ff_alac_decoder = {
+    .name           = "alac",
+    .type           = AVMEDIA_TYPE_AUDIO,
+    .id             = CODEC_ID_ALAC,
+    .priv_data_size = sizeof(ALACContext),
+    .init           = alac_decode_init,
+    .close          = alac_decode_close,
+    .decode         = alac_decode_frame,
+    .capabilities   = CODEC_CAP_DR1,
+    .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
 };