]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/shorten.c
celp filters: Do not read earlier than the start of the 'out' vector.
[ffmpeg] / libavcodec / shorten.c
index 638d886c4fd2545b5b143d99ec9d7f0a8ffeb4f0..62b6ae1fe0dd96f81e7d0cbd73815a02150c9c18 100644 (file)
@@ -28,6 +28,7 @@
 
 #include <limits.h>
 #include "avcodec.h"
+#include "bytestream.h"
 #include "get_bits.h"
 #include "golomb.h"
 
@@ -78,12 +79,14 @@ static const uint8_t is_audio_command[10] = { 1, 1, 1, 1, 0, 0, 0, 1, 1, 0 };
 
 typedef struct ShortenContext {
     AVCodecContext *avctx;
+    AVFrame frame;
     GetBitContext gb;
 
     int min_framesize, max_framesize;
     int channels;
 
     int32_t *decoded[MAX_CHANNELS];
+    int32_t *decoded_base[MAX_CHANNELS];
     int32_t *offset[MAX_CHANNELS];
     int *coeffs;
     uint8_t *bitstream;
@@ -101,6 +104,8 @@ typedef struct ShortenContext {
     int blocksize;
     int bitindex;
     int32_t lpcqoffset;
+    int got_header;
+    int got_quit_command;
 } ShortenContext;
 
 static av_cold int shorten_decode_init(AVCodecContext * avctx)
@@ -109,6 +114,9 @@ static av_cold int shorten_decode_init(AVCodecContext * avctx)
     s->avctx = avctx;
     avctx->sample_fmt = AV_SAMPLE_FMT_S16;
 
+    avcodec_get_frame_defaults(&s->frame);
+    avctx->coded_frame = &s->frame;
+
     return 0;
 }
 
@@ -133,13 +141,14 @@ static int allocate_buffers(ShortenContext *s)
             return AVERROR(ENOMEM);
         s->offset[chan] = tmp_ptr;
 
-        tmp_ptr = av_realloc(s->decoded[chan], sizeof(int32_t)*(s->blocksize + s->nwrap));
+        tmp_ptr = av_realloc(s->decoded_base[chan], (s->blocksize + s->nwrap) *
+                             sizeof(s->decoded_base[0][0]));
         if (!tmp_ptr)
             return AVERROR(ENOMEM);
-        s->decoded[chan] = tmp_ptr;
+        s->decoded_base[chan] = tmp_ptr;
         for (i=0; i<s->nwrap; i++)
-            s->decoded[chan][i] = 0;
-        s->decoded[chan] += s->nwrap;
+            s->decoded_base[chan][i] = 0;
+        s->decoded[chan] = s->decoded_base[chan] + s->nwrap;
     }
 
     coeffs = av_realloc(s->coeffs, s->nwrap * sizeof(*s->coeffs));
@@ -169,7 +178,7 @@ static void fix_bitshift(ShortenContext *s, int32_t *buffer)
 }
 
 
-static void init_offset(ShortenContext *s)
+static int init_offset(ShortenContext *s)
 {
     int32_t mean = 0;
     int  chan, i;
@@ -183,55 +192,46 @@ static void init_offset(ShortenContext *s)
             break;
         default:
             av_log(s->avctx, AV_LOG_ERROR, "unknown audio type");
-            abort();
+            return AVERROR_INVALIDDATA;
     }
 
     for (chan = 0; chan < s->channels; chan++)
         for (i = 0; i < nblock; i++)
             s->offset[chan][i] = mean;
+    return 0;
 }
 
-static inline int get_le32(GetBitContext *gb)
-{
-    return av_bswap32(get_bits_long(gb, 32));
-}
-
-static inline short get_le16(GetBitContext *gb)
-{
-    return av_bswap16(get_bits_long(gb, 16));
-}
-
-static int decode_wave_header(AVCodecContext *avctx, uint8_t *header, int header_size)
+static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
+                              int header_size)
 {
-    GetBitContext hb;
     int len;
     short wave_format;
 
-    init_get_bits(&hb, header, header_size*8);
-    if (get_le32(&hb) != MKTAG('R','I','F','F')) {
+
+    if (bytestream_get_le32(&header) != MKTAG('R','I','F','F')) {
         av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
         return -1;
     }
 
-    skip_bits_long(&hb, 32);    /* chunk_size */
+    header += 4; /* chunk size */;
 
-    if (get_le32(&hb) != MKTAG('W','A','V','E')) {
+    if (bytestream_get_le32(&header) != MKTAG('W','A','V','E')) {
         av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
         return -1;
     }
 
-    while (get_le32(&hb) != MKTAG('f','m','t',' ')) {
-        len = get_le32(&hb);
-        skip_bits(&hb, 8*len);
+    while (bytestream_get_le32(&header) != MKTAG('f','m','t',' ')) {
+        len = bytestream_get_le32(&header);
+        header += len;
     }
-    len = get_le32(&hb);
+    len = bytestream_get_le32(&header);
 
     if (len < 16) {
         av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
         return -1;
     }
 
-    wave_format = get_le16(&hb);
+    wave_format = bytestream_get_le16(&header);
 
     switch (wave_format) {
         case WAVE_FORMAT_PCM:
@@ -241,11 +241,11 @@ static int decode_wave_header(AVCodecContext *avctx, uint8_t *header, int header
             return -1;
     }
 
-    skip_bits(&hb, 16); // skip channels    (already got from shorten header)
-    avctx->sample_rate = get_le32(&hb);
-    skip_bits(&hb, 32); // skip bit rate    (represents original uncompressed bit rate)
-    skip_bits(&hb, 16); // skip block align (not needed)
-    avctx->bits_per_coded_sample = get_le16(&hb);
+    header += 2;        // skip channels    (already got from shorten header)
+    avctx->sample_rate = bytestream_get_le32(&header);
+    header += 4;        // skip bit rate    (represents original uncompressed bit rate)
+    header += 2;        // skip block align (not needed)
+    avctx->bits_per_coded_sample = bytestream_get_le16(&header);
 
     if (avctx->bits_per_coded_sample != 16) {
         av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
@@ -259,45 +259,63 @@ static int decode_wave_header(AVCodecContext *avctx, uint8_t *header, int header
     return 0;
 }
 
-static int16_t * interleave_buffer(int16_t *samples, int nchan, int blocksize, int32_t **buffer) {
+static void interleave_buffer(int16_t *samples, int nchan, int blocksize,
+                              int32_t **buffer)
+{
     int i, chan;
     for (i=0; i<blocksize; i++)
         for (chan=0; chan < nchan; chan++)
-            *samples++ = FFMIN(buffer[chan][i], 32768);
-    return samples;
+            *samples++ = av_clip_int16(buffer[chan][i]);
 }
 
-static int decode_subframe_lpc(ShortenContext *s, int channel,
+static const int fixed_coeffs[3][3] = {
+    { 1,  0,  0 },
+    { 2, -1,  0 },
+    { 3, -3,  1 }
+};
+
+static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
                                int residual_size, int32_t coffset)
 {
-    int pred_order, sum, i, j;
-    int *coeffs = s->coeffs;
-
-    /* read/validate prediction order */
-    pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
-    if (pred_order > s->nwrap) {
-        av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n", pred_order);
-        return AVERROR(EINVAL);
+    int pred_order, sum, qshift, init_sum, i, j;
+    const int *coeffs;
+
+    if (command == FN_QLPC) {
+        /* read/validate prediction order */
+        pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
+        if (pred_order > s->nwrap) {
+            av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n", pred_order);
+            return AVERROR(EINVAL);
+        }
+        /* read LPC coefficients */
+        for (i=0; i<pred_order; i++)
+            s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
+        coeffs = s->coeffs;
+
+        qshift = LPCQUANT;
+    } else {
+        /* fixed LPC coeffs */
+        pred_order = command;
+        coeffs     = fixed_coeffs[pred_order-1];
+        qshift     = 0;
     }
-    /* read LPC coefficients */
-    for (i=0; i<pred_order; i++)
-        coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
 
     /* subtract offset from previous samples to use in prediction */
-    if (coffset)
+    if (command == FN_QLPC && coffset)
         for (i = -pred_order; i < 0; i++)
             s->decoded[channel][i] -= coffset;
 
     /* decode residual and do LPC prediction */
+    init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
     for (i=0; i < s->blocksize; i++) {
-        sum = s->lpcqoffset;
+        sum = init_sum;
         for (j=0; j<pred_order; j++)
             sum += coeffs[j] * s->decoded[channel][i-j-1];
-        s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + (sum >> LPCQUANT);
+        s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + (sum >> qshift);
     }
 
     /* add offset to current samples */
-    if (coffset != 0)
+    if (command == FN_QLPC && coffset)
         for (i = 0; i < s->blocksize; i++)
             s->decoded[channel][i] += coffset;
 
@@ -316,7 +334,6 @@ static int read_header(ShortenContext *s)
 
     s->lpcqoffset = 0;
     s->blocksize = DEFAULT_BLOCK_SIZE;
-    s->channels = 1;
     s->nmean = -1;
     s->version = get_bits(&s->gb, 8);
     s->internal_ftype = get_uint(s, TYPESIZE);
@@ -330,8 +347,16 @@ static int read_header(ShortenContext *s)
 
     /* get blocksize if version > 0 */
     if (s->version > 0) {
-        int skip_bytes;
-        s->blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
+        int skip_bytes, blocksize;
+
+        blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
+        if (!blocksize || blocksize > MAX_BLOCKSIZE) {
+            av_log(s->avctx, AV_LOG_ERROR, "invalid or unsupported block size: %d\n",
+                   blocksize);
+            return AVERROR(EINVAL);
+        }
+        s->blocksize = blocksize;
+
         maxnlpc = get_uint(s, LPCQSIZE);
         s->nmean = get_uint(s, 0);
 
@@ -345,7 +370,8 @@ static int read_header(ShortenContext *s)
     if ((ret = allocate_buffers(s)) < 0)
         return ret;
 
-    init_offset(s);
+    if ((ret = init_offset(s)) < 0)
+        return ret;
 
     if (s->version > 1)
         s->lpcqoffset = V2LPCQOFFSET;
@@ -370,18 +396,21 @@ static int read_header(ShortenContext *s)
     s->cur_chan = 0;
     s->bitshift = 0;
 
+    s->got_header = 1;
+
     return 0;
 }
 
-static int shorten_decode_frame(AVCodecContext *avctx,
-        void *data, int *data_size,
-        AVPacket *avpkt)
+static int shorten_decode_frame(AVCodecContext *avctx, void *data,
+                                int *got_frame_ptr, AVPacket *avpkt)
 {
     const uint8_t *buf = avpkt->data;
     int buf_size = avpkt->size;
     ShortenContext *s = avctx->priv_data;
     int i, input_buf_size = 0;
-    int16_t *samples = data;
+    int ret;
+
+    /* allocate internal bitstream buffer */
     if(s->max_framesize == 0){
         void *tmp_ptr;
         s->max_framesize= 1024; // should hopefully be enough for the first header
@@ -394,47 +423,62 @@ static int shorten_decode_frame(AVCodecContext *avctx,
         s->bitstream = tmp_ptr;
     }
 
+    /* append current packet data to bitstream buffer */
     if(1 && s->max_framesize){//FIXME truncated
         buf_size= FFMIN(buf_size, s->max_framesize - s->bitstream_size);
         input_buf_size= buf_size;
 
         if(s->bitstream_index + s->bitstream_size + buf_size > s->allocated_bitstream_size){
-            //                printf("memmove\n");
             memmove(s->bitstream, &s->bitstream[s->bitstream_index], s->bitstream_size);
             s->bitstream_index=0;
         }
-        memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf, buf_size);
+        if (buf)
+            memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf, buf_size);
         buf= &s->bitstream[s->bitstream_index];
         buf_size += s->bitstream_size;
         s->bitstream_size= buf_size;
 
-        if(buf_size < s->max_framesize){
-            *data_size = 0;
+        /* do not decode until buffer has at least max_framesize bytes or
+           the end of the file has been reached */
+        if (buf_size < s->max_framesize && avpkt->data) {
+            *got_frame_ptr = 0;
             return input_buf_size;
         }
     }
+    /* init and position bitstream reader */
     init_get_bits(&s->gb, buf, buf_size*8);
     skip_bits(&s->gb, s->bitindex);
-    if (!s->blocksize)
-    {
-        int ret;
+
+    /* process header or next subblock */
+    if (!s->got_header) {
         if ((ret = read_header(s)) < 0)
             return ret;
-        *data_size = 0;
+        *got_frame_ptr = 0;
+        goto finish_frame;
     }
-    else
-    {
+
+    /* if quit command was read previously, don't decode anything */
+    if (s->got_quit_command) {
+        *got_frame_ptr = 0;
+        return avpkt->size;
+    }
+
+    s->cur_chan = 0;
+    while (s->cur_chan < s->channels) {
         int cmd;
         int len;
+
+        if (get_bits_left(&s->gb) < 3+FNSIZE) {
+            *got_frame_ptr = 0;
+            break;
+        }
+
         cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
 
         if (cmd > FN_VERBATIM) {
             av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
-            if (s->bitstream_size > 0) {
-                s->bitstream_index++;
-                s->bitstream_size--;
-            }
-            return -1;
+            *got_frame_ptr = 0;
+            break;
         }
 
         if (!is_audio_command[cmd]) {
@@ -455,19 +499,29 @@ static int shorten_decode_frame(AVCodecContext *avctx,
                         av_log(avctx, AV_LOG_ERROR, "Increasing block size is not supported\n");
                         return AVERROR_PATCHWELCOME;
                     }
+                    if (!blocksize || blocksize > MAX_BLOCKSIZE) {
+                        av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
+                               "block size: %d\n", blocksize);
+                        return AVERROR(EINVAL);
+                    }
                     s->blocksize = blocksize;
                     break;
                 }
                 case FN_QUIT:
+                    s->got_quit_command = 1;
                     break;
             }
-            *data_size = 0;
+            if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
+                *got_frame_ptr = 0;
+                break;
+            }
         } else {
             /* process audio command */
-            int ret;
             int residual_size = 0;
             int channel = s->cur_chan;
             int32_t coffset;
+
+            /* get Rice code for residual decoding */
             if (cmd != FN_ZERO) {
                 residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
                 /* this is a hack as version 0 differed in defintion of get_sr_golomb_shorten */
@@ -475,6 +529,7 @@ static int shorten_decode_frame(AVCodecContext *avctx,
                     residual_size--;
             }
 
+            /* calculate sample offset using means from previous blocks */
             if (s->nmean == 0)
                 coffset = s->offset[channel][0];
             else {
@@ -485,35 +540,17 @@ static int shorten_decode_frame(AVCodecContext *avctx,
                 if (s->version >= 2)
                     coffset >>= FFMIN(1, s->bitshift);
             }
-            switch (cmd) {
-                case FN_ZERO:
-                    for (i=0; i<s->blocksize; i++)
-                        s->decoded[channel][i] = 0;
-                    break;
-                case FN_DIFF0:
-                    for (i=0; i<s->blocksize; i++)
-                        s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + coffset;
-                    break;
-                case FN_DIFF1:
-                    for (i=0; i<s->blocksize; i++)
-                        s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + s->decoded[channel][i - 1];
-                    break;
-                case FN_DIFF2:
-                    for (i=0; i<s->blocksize; i++)
-                        s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + 2*s->decoded[channel][i-1]
-                                                                                              -   s->decoded[channel][i-2];
-                    break;
-                case FN_DIFF3:
-                    for (i=0; i<s->blocksize; i++)
-                        s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + 3*s->decoded[channel][i-1]
-                                                                                              - 3*s->decoded[channel][i-2]
-                                                                                              +   s->decoded[channel][i-3];
-                    break;
-                case FN_QLPC:
-                    if ((ret = decode_subframe_lpc(s, channel, residual_size, coffset)) < 0)
-                        return ret;
-                    break;
+
+            /* decode samples for this channel */
+            if (cmd == FN_ZERO) {
+                for (i=0; i<s->blocksize; i++)
+                    s->decoded[channel][i] = 0;
+            } else {
+                if ((ret = decode_subframe_lpc(s, cmd, channel, residual_size, coffset)) < 0)
+                    return ret;
             }
+
+            /* update means with info from the current block */
             if (s->nmean > 0) {
                 int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
                 for (i=0; i<s->blocksize; i++)
@@ -527,23 +564,37 @@ static int shorten_decode_frame(AVCodecContext *avctx,
                 else
                     s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
             }
+
+            /* copy wrap samples for use with next block */
             for (i=-s->nwrap; i<0; i++)
                 s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
 
+            /* shift samples to add in unused zero bits which were removed
+               during encoding */
             fix_bitshift(s, s->decoded[channel]);
 
+            /* if this is the last channel in the block, output the samples */
             s->cur_chan++;
             if (s->cur_chan == s->channels) {
-                samples = interleave_buffer(samples, s->channels, s->blocksize, s->decoded);
-                s->cur_chan = 0;
-                *data_size = (int8_t *)samples - (int8_t *)data;
-            } else {
-                *data_size = 0;
+                /* get output buffer */
+                s->frame.nb_samples = s->blocksize;
+                if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
+                    av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+                    return ret;
+                }
+                /* interleave output */
+                interleave_buffer((int16_t *)s->frame.data[0], s->channels,
+                                  s->blocksize, s->decoded);
+
+                *got_frame_ptr   = 1;
+                *(AVFrame *)data = s->frame;
             }
         }
     }
+    if (s->cur_chan < s->channels)
+        *got_frame_ptr = 0;
 
-    //    s->last_blocksize = s->blocksize;
+finish_frame:
     s->bitindex = get_bits_count(&s->gb) - 8*((get_bits_count(&s->gb))/8);
     i= (get_bits_count(&s->gb))/8;
     if (i > buf_size) {
@@ -566,12 +617,13 @@ static av_cold int shorten_decode_close(AVCodecContext *avctx)
     int i;
 
     for (i = 0; i < s->channels; i++) {
-        s->decoded[i] -= s->nwrap;
-        av_freep(&s->decoded[i]);
+        s->decoded[i] = NULL;
+        av_freep(&s->decoded_base[i]);
         av_freep(&s->offset[i]);
     }
     av_freep(&s->bitstream);
     av_freep(&s->coeffs);
+
     return 0;
 }
 
@@ -583,5 +635,6 @@ AVCodec ff_shorten_decoder = {
     .init           = shorten_decode_init,
     .close          = shorten_decode_close,
     .decode         = shorten_decode_frame,
-    .long_name= NULL_IF_CONFIG_SMALL("Shorten"),
+    .capabilities   = CODEC_CAP_DELAY | CODEC_CAP_DR1,
+    .long_name      = NULL_IF_CONFIG_SMALL("Shorten"),
 };