#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 "crc.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
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
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
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
GetBitContext gbc; ///< bitstream reader
AVRandomState dith_state; ///< for dither generation
AVCodecContext *avctx; ///< parent context
+ uint8_t *input_buffer; ///< temp buffer to prevent overread
} AC3DecodeContext;
/**
/*
* Initialize tables at runtime.
*/
-static void ac3_tables_init(void)
+static av_cold void ac3_tables_init(void)
{
int i;
/**
* 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;
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;
}
GetBitContext *gbc = &s->gbc;
int err, i;
- err = ff_ac3_parse_header(gbc->buffer, &hdr);
+ err = ff_ac3_parse_header(gbc, &hdr);
if(err)
return err;
s->fbw_channels = s->channels - s->lfe_on;
s->lfe_ch = s->fbw_channels + 1;
s->frame_size = hdr.frame_size;
+ s->center_mix_level = hdr.center_mix_level;
+ s->surround_mix_level = hdr.surround_mix_level;
/* set default output to all source channels */
s->out_channels = s->channels;
if(s->lfe_on)
s->output_mode |= AC3_OUTPUT_LFEON;
- /* set default mix levels */
- s->center_mix_level = 3; // -4.5dB
- s->surround_mix_level = 4; // -6.0dB
-
- /* 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)
- s->center_mix_level = center_levels[get_bits(gbc, 2)];
- if(s->channel_mode & 4)
- s->surround_mix_level = surround_levels[get_bits(gbc, 2)];
- }
- skip_bits1(gbc); // skip lfeon
-
/* read the rest of the bsi. read twice for dual mono mode. */
i = !(s->channel_mode);
do {
* 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]) {
/**
* 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;
for(i=0; i<256; i++) {
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];
+ v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
+ v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
}
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.
*/
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;
}
}
- 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 */
/**
* 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);
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;
/**
* 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;
}
.init = ac3_decode_init,
.close = ac3_decode_end,
.decode = ac3_decode_frame,
+ .long_name = "ATSC A/52 / AC-3",
};