#include "avcodec.h"
#include "put_bits.h"
-#include "dsputil.h"
+#include "internal.h"
#include "lpc.h"
#include "mathops.h"
+#include "alac_data.h"
#define DEFAULT_FRAME_SIZE 4096
-#define DEFAULT_SAMPLE_SIZE 16
-#define MAX_CHANNELS 8
#define ALAC_EXTRADATA_SIZE 36
#define ALAC_FRAME_HEADER_SIZE 55
#define ALAC_FRAME_FOOTER_SIZE 3
} AlacLPCContext;
typedef struct AlacEncodeContext {
+ int frame_size; /**< current frame size */
+ int verbatim; /**< current frame verbatim mode flag */
int compression_level;
int min_prediction_order;
int max_prediction_order;
int max_coded_frame_size;
int write_sample_size;
- int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
+ int extra_bits;
+ int32_t sample_buf[2][DEFAULT_FRAME_SIZE];
int32_t predictor_buf[DEFAULT_FRAME_SIZE];
int interlacing_shift;
int interlacing_leftweight;
PutBitContext pbctx;
RiceContext rc;
- AlacLPCContext lpc[MAX_CHANNELS];
+ AlacLPCContext lpc[2];
LPCContext lpc_ctx;
AVCodecContext *avctx;
} AlacEncodeContext;
-static void init_sample_buffers(AlacEncodeContext *s,
- const int16_t *input_samples)
+static void init_sample_buffers(AlacEncodeContext *s, int channels,
+ uint8_t const *samples[2])
{
int ch, i;
-
- for (ch = 0; ch < s->avctx->channels; ch++) {
- const int16_t *sptr = input_samples + ch;
- for (i = 0; i < s->avctx->frame_size; i++) {
- s->sample_buf[ch][i] = *sptr;
- sptr += s->avctx->channels;
- }
- }
+ int shift = av_get_bytes_per_sample(s->avctx->sample_fmt) * 8 -
+ s->avctx->bits_per_raw_sample;
+
+#define COPY_SAMPLES(type) do { \
+ for (ch = 0; ch < channels; ch++) { \
+ int32_t *bptr = s->sample_buf[ch]; \
+ const type *sptr = (const type *)samples[ch]; \
+ for (i = 0; i < s->frame_size; i++) \
+ bptr[i] = sptr[i] >> shift; \
+ } \
+ } while (0)
+
+ if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S32P)
+ COPY_SAMPLES(int32_t);
+ else
+ COPY_SAMPLES(int16_t);
}
static void encode_scalar(AlacEncodeContext *s, int x,
}
}
-static void write_frame_header(AlacEncodeContext *s, int is_verbatim)
+static void write_element_header(AlacEncodeContext *s,
+ enum AlacRawDataBlockType element,
+ int instance)
{
- put_bits(&s->pbctx, 3, s->avctx->channels-1); // No. of channels -1
- put_bits(&s->pbctx, 16, 0); // Seems to be zero
- put_bits(&s->pbctx, 1, 1); // Sample count is in the header
- put_bits(&s->pbctx, 2, 0); // FIXME: Wasted bytes field
- put_bits(&s->pbctx, 1, is_verbatim); // Audio block is verbatim
- put_bits32(&s->pbctx, s->avctx->frame_size); // No. of samples in the frame
+ int encode_fs = 0;
+
+ if (s->frame_size < DEFAULT_FRAME_SIZE)
+ encode_fs = 1;
+
+ put_bits(&s->pbctx, 3, element); // element type
+ put_bits(&s->pbctx, 4, instance); // element instance
+ put_bits(&s->pbctx, 12, 0); // unused header bits
+ put_bits(&s->pbctx, 1, encode_fs); // Sample count is in the header
+ put_bits(&s->pbctx, 2, s->extra_bits >> 3); // Extra bytes (for 24-bit)
+ put_bits(&s->pbctx, 1, s->verbatim); // Audio block is verbatim
+ if (encode_fs)
+ put_bits32(&s->pbctx, s->frame_size); // No. of samples in the frame
}
static void calc_predictor_params(AlacEncodeContext *s, int ch)
s->lpc[ch].lpc_coeff[5] = -25;
} else {
opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, s->sample_buf[ch],
- s->avctx->frame_size,
+ s->frame_size,
s->min_prediction_order,
s->max_prediction_order,
ALAC_MAX_LPC_PRECISION, coefs, shift,
static void alac_stereo_decorrelation(AlacEncodeContext *s)
{
int32_t *left = s->sample_buf[0], *right = s->sample_buf[1];
- int i, mode, n = s->avctx->frame_size;
+ int i, mode, n = s->frame_size;
int32_t tmp;
mode = estimate_stereo_mode(left, right, n);
if (lpc.lpc_order == 31) {
s->predictor_buf[0] = s->sample_buf[ch][0];
- for (i = 1; i < s->avctx->frame_size; i++) {
+ for (i = 1; i < s->frame_size; i++) {
s->predictor_buf[i] = s->sample_buf[ch][i ] -
s->sample_buf[ch][i - 1];
}
residual[i] = samples[i] - samples[i-1];
// perform lpc on remaining samples
- for (i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {
+ for (i = lpc.lpc_order + 1; i < s->frame_size; i++) {
int sum = 1 << (lpc.lpc_quant - 1), res_val, j;
for (j = 0; j < lpc.lpc_order; j++) {
int sign_modifier = 0, i, k;
int32_t *samples = s->predictor_buf;
- for (i = 0; i < s->avctx->frame_size;) {
+ for (i = 0; i < s->frame_size;) {
int x;
k = av_log2((history >> 9) + 3);
if (x > 0xFFFF)
history = 0xFFFF;
- if (history < 128 && i < s->avctx->frame_size) {
+ if (history < 128 && i < s->frame_size) {
unsigned int block_size = 0;
k = 7 - av_log2(history) + ((history + 16) >> 6);
- while (*samples == 0 && i < s->avctx->frame_size) {
+ while (*samples == 0 && i < s->frame_size) {
samples++;
i++;
block_size++;
}
}
-static void write_compressed_frame(AlacEncodeContext *s)
+static void write_element(AlacEncodeContext *s,
+ enum AlacRawDataBlockType element, int instance,
+ const uint8_t *samples0, const uint8_t *samples1)
{
- int i, j;
+ uint8_t const *samples[2] = { samples0, samples1 };
+ int i, j, channels;
int prediction_type = 0;
+ PutBitContext *pb = &s->pbctx;
- if (s->avctx->channels == 2)
- alac_stereo_decorrelation(s);
- put_bits(&s->pbctx, 8, s->interlacing_shift);
- put_bits(&s->pbctx, 8, s->interlacing_leftweight);
+ channels = element == TYPE_CPE ? 2 : 1;
+
+ if (s->verbatim) {
+ write_element_header(s, element, instance);
+ /* samples are channel-interleaved in verbatim mode */
+ if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S32P) {
+ int shift = 32 - s->avctx->bits_per_raw_sample;
+ int32_t const *samples_s32[2] = { (const int32_t *)samples0,
+ (const int32_t *)samples1 };
+ for (i = 0; i < s->frame_size; i++)
+ for (j = 0; j < channels; j++)
+ put_sbits(pb, s->avctx->bits_per_raw_sample,
+ samples_s32[j][i] >> shift);
+ } else {
+ int16_t const *samples_s16[2] = { (const int16_t *)samples0,
+ (const int16_t *)samples1 };
+ for (i = 0; i < s->frame_size; i++)
+ for (j = 0; j < channels; j++)
+ put_sbits(pb, s->avctx->bits_per_raw_sample,
+ samples_s16[j][i]);
+ }
+ } else {
+ s->write_sample_size = s->avctx->bits_per_raw_sample - s->extra_bits +
+ channels - 1;
+
+ init_sample_buffers(s, channels, samples);
+ write_element_header(s, element, instance);
+
+ if (channels == 2)
+ alac_stereo_decorrelation(s);
+ else
+ s->interlacing_shift = s->interlacing_leftweight = 0;
+ put_bits(pb, 8, s->interlacing_shift);
+ put_bits(pb, 8, s->interlacing_leftweight);
+
+ for (i = 0; i < channels; i++) {
+ calc_predictor_params(s, i);
+
+ put_bits(pb, 4, prediction_type);
+ put_bits(pb, 4, s->lpc[i].lpc_quant);
+
+ put_bits(pb, 3, s->rc.rice_modifier);
+ put_bits(pb, 5, s->lpc[i].lpc_order);
+ // predictor coeff. table
+ for (j = 0; j < s->lpc[i].lpc_order; j++)
+ put_sbits(pb, 16, s->lpc[i].lpc_coeff[j]);
+ }
- for (i = 0; i < s->avctx->channels; i++) {
- calc_predictor_params(s, i);
+ // write extra bits if needed
+ if (s->extra_bits) {
+ uint32_t mask = (1 << s->extra_bits) - 1;
+ for (i = 0; i < s->frame_size; i++) {
+ for (j = 0; j < channels; j++) {
+ put_bits(pb, s->extra_bits, s->sample_buf[j][i] & mask);
+ s->sample_buf[j][i] >>= s->extra_bits;
+ }
+ }
+ }
- put_bits(&s->pbctx, 4, prediction_type);
- put_bits(&s->pbctx, 4, s->lpc[i].lpc_quant);
+ // apply lpc and entropy coding to audio samples
+ for (i = 0; i < channels; i++) {
+ alac_linear_predictor(s, i);
- put_bits(&s->pbctx, 3, s->rc.rice_modifier);
- put_bits(&s->pbctx, 5, s->lpc[i].lpc_order);
- // predictor coeff. table
- for (j = 0; j < s->lpc[i].lpc_order; j++)
- put_sbits(&s->pbctx, 16, s->lpc[i].lpc_coeff[j]);
+ // TODO: determine when this will actually help. for now it's not used.
+ if (prediction_type == 15) {
+ // 2nd pass 1st order filter
+ for (j = s->frame_size - 1; j > 0; j--)
+ s->predictor_buf[j] -= s->predictor_buf[j - 1];
+ }
+ alac_entropy_coder(s);
+ }
}
+}
- // apply lpc and entropy coding to audio samples
-
- for (i = 0; i < s->avctx->channels; i++) {
- alac_linear_predictor(s, i);
-
- // TODO: determine when this will actually help. for now it's not used.
- if (prediction_type == 15) {
- // 2nd pass 1st order filter
- for (j = s->avctx->frame_size - 1; j > 0; j--)
- s->predictor_buf[j] -= s->predictor_buf[j - 1];
+static int write_frame(AlacEncodeContext *s, AVPacket *avpkt,
+ uint8_t * const *samples)
+{
+ PutBitContext *pb = &s->pbctx;
+ const enum AlacRawDataBlockType *ch_elements = ff_alac_channel_elements[s->avctx->channels - 1];
+ const uint8_t *ch_map = ff_alac_channel_layout_offsets[s->avctx->channels - 1];
+ int ch, element, sce, cpe;
+
+ init_put_bits(pb, avpkt->data, avpkt->size);
+
+ ch = element = sce = cpe = 0;
+ while (ch < s->avctx->channels) {
+ if (ch_elements[element] == TYPE_CPE) {
+ write_element(s, TYPE_CPE, cpe, samples[ch_map[ch]],
+ samples[ch_map[ch + 1]]);
+ cpe++;
+ ch += 2;
+ } else {
+ write_element(s, TYPE_SCE, sce, samples[ch_map[ch]], NULL);
+ sce++;
+ ch++;
}
-
- alac_entropy_coder(s);
+ element++;
}
+
+ put_bits(pb, 3, TYPE_END);
+ flush_put_bits(pb);
+
+ return put_bits_count(pb) >> 3;
}
static av_always_inline int get_max_frame_size(int frame_size, int ch, int bps)
{
- return FFALIGN(55 + bps * ch * frame_size + 3, 8) / 8;
+ int header_bits = 23 + 32 * (frame_size < DEFAULT_FRAME_SIZE);
+ return FFALIGN(header_bits + bps * ch * frame_size + 3, 8) / 8;
}
static av_cold int alac_encode_close(AVCodecContext *avctx)
int ret;
uint8_t *alac_extradata;
- avctx->frame_size = DEFAULT_FRAME_SIZE;
+ avctx->frame_size = s->frame_size = DEFAULT_FRAME_SIZE;
- if (avctx->sample_fmt != AV_SAMPLE_FMT_S16) {
- av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n");
- return -1;
- }
-
- /* TODO: Correctly implement multi-channel ALAC.
- It is similar to multi-channel AAC, in that it has a series of
- single-channel (SCE), channel-pair (CPE), and LFE elements. */
- if (avctx->channels > 2) {
- av_log(avctx, AV_LOG_ERROR, "only mono or stereo input is currently supported\n");
- return AVERROR_PATCHWELCOME;
+ if (avctx->sample_fmt == AV_SAMPLE_FMT_S32P) {
+ if (avctx->bits_per_raw_sample != 24)
+ av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n");
+ avctx->bits_per_raw_sample = 24;
+ } else {
+ avctx->bits_per_raw_sample = 16;
+ s->extra_bits = 0;
}
// Set default compression level
s->max_coded_frame_size = get_max_frame_size(avctx->frame_size,
avctx->channels,
- DEFAULT_SAMPLE_SIZE);
-
- // FIXME: consider wasted_bytes
- s->write_sample_size = DEFAULT_SAMPLE_SIZE + avctx->channels - 1;
+ avctx->bits_per_raw_sample);
avctx->extradata = av_mallocz(ALAC_EXTRADATA_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
if (!avctx->extradata) {
AV_WB32(alac_extradata, ALAC_EXTRADATA_SIZE);
AV_WB32(alac_extradata+4, MKBETAG('a','l','a','c'));
AV_WB32(alac_extradata+12, avctx->frame_size);
- AV_WB8 (alac_extradata+17, DEFAULT_SAMPLE_SIZE);
+ AV_WB8 (alac_extradata+17, avctx->bits_per_raw_sample);
AV_WB8 (alac_extradata+21, avctx->channels);
AV_WB32(alac_extradata+24, s->max_coded_frame_size);
AV_WB32(alac_extradata+28,
- avctx->sample_rate * avctx->channels * DEFAULT_SAMPLE_SIZE); // average bitrate
+ avctx->sample_rate * avctx->channels * avctx->bits_per_raw_sample); // average bitrate
AV_WB32(alac_extradata+32, avctx->sample_rate);
// Set relevant extradata fields
goto error;
}
- avctx->coded_frame = avcodec_alloc_frame();
+ avctx->coded_frame = av_frame_alloc();
if (!avctx->coded_frame) {
ret = AVERROR(ENOMEM);
goto error;
return ret;
}
-static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
- int buf_size, void *data)
+static int alac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
+ const AVFrame *frame, int *got_packet_ptr)
{
AlacEncodeContext *s = avctx->priv_data;
- PutBitContext *pb = &s->pbctx;
- int i, out_bytes, verbatim_flag = 0;
- int max_frame_size;
+ int out_bytes, max_frame_size, ret;
- if (avctx->frame_size < DEFAULT_FRAME_SIZE)
- max_frame_size = get_max_frame_size(avctx->frame_size, avctx->channels,
- DEFAULT_SAMPLE_SIZE);
+ s->frame_size = frame->nb_samples;
+
+ if (frame->nb_samples < DEFAULT_FRAME_SIZE)
+ max_frame_size = get_max_frame_size(s->frame_size, avctx->channels,
+ avctx->bits_per_raw_sample);
else
max_frame_size = s->max_coded_frame_size;
- if (buf_size < 2 * max_frame_size) {
- av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n");
- return -1;
+ if ((ret = ff_alloc_packet(avpkt, 2 * max_frame_size))) {
+ av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
+ return ret;
}
-verbatim:
- init_put_bits(pb, frame, buf_size);
-
- if (s->compression_level == 0 || verbatim_flag) {
- // Verbatim mode
- const int16_t *samples = data;
- write_frame_header(s, 1);
- for (i = 0; i < avctx->frame_size * avctx->channels; i++) {
- put_sbits(pb, 16, *samples++);
- }
+ /* use verbatim mode for compression_level 0 */
+ if (s->compression_level) {
+ s->verbatim = 0;
+ s->extra_bits = avctx->bits_per_raw_sample - 16;
} else {
- init_sample_buffers(s, data);
- write_frame_header(s, 0);
- write_compressed_frame(s);
+ s->verbatim = 1;
+ s->extra_bits = 0;
}
- put_bits(pb, 3, 7);
- flush_put_bits(pb);
- out_bytes = put_bits_count(pb) >> 3;
+ out_bytes = write_frame(s, avpkt, frame->extended_data);
if (out_bytes > max_frame_size) {
/* frame too large. use verbatim mode */
- if (verbatim_flag || s->compression_level == 0) {
- /* still too large. must be an error. */
- av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
- return -1;
- }
- verbatim_flag = 1;
- goto verbatim;
+ s->verbatim = 1;
+ s->extra_bits = 0;
+ out_bytes = write_frame(s, avpkt, frame->extended_data);
}
- return out_bytes;
+ avpkt->size = out_bytes;
+ *got_packet_ptr = 1;
+ return 0;
}
AVCodec ff_alac_encoder = {
.name = "alac",
+ .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
.type = AVMEDIA_TYPE_AUDIO,
- .id = CODEC_ID_ALAC,
+ .id = AV_CODEC_ID_ALAC,
.priv_data_size = sizeof(AlacEncodeContext),
.init = alac_encode_init,
- .encode = alac_encode_frame,
+ .encode2 = alac_encode_frame,
.close = alac_encode_close,
.capabilities = CODEC_CAP_SMALL_LAST_FRAME,
- .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
+ .channel_layouts = ff_alac_channel_layouts,
+ .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S32P,
+ AV_SAMPLE_FMT_S16P,
AV_SAMPLE_FMT_NONE },
- .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
};