#include "libavutil/md5.h"
#include "libavutil/opt.h"
#include "avcodec.h"
-#include "dsputil.h"
+#include "bswapdsp.h"
#include "get_bits.h"
#include "golomb.h"
#include "internal.h"
#define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
#define MAX_LPC_PRECISION 15
#define MAX_LPC_SHIFT 15
-#define MAX_RICE_PARAM 14
+
+enum CodingMode {
+ CODING_MODE_RICE = 4,
+ CODING_MODE_RICE2 = 5,
+};
typedef struct CompressionOptions {
int compression_level;
} CompressionOptions;
typedef struct RiceContext {
+ enum CodingMode coding_mode;
int porder;
int params[MAX_PARTITIONS];
+ uint32_t udata[FLAC_MAX_BLOCKSIZE];
} RiceContext;
typedef struct FlacSubframe {
int channels;
int samplerate;
int sr_code[2];
+ int bps_code;
int max_blocksize;
int min_framesize;
int max_framesize;
struct AVMD5 *md5ctx;
uint8_t *md5_buffer;
unsigned int md5_buffer_size;
- DSPContext dsp;
+ BswapDSPContext bdsp;
FLACDSPContext flac_dsp;
+
+ int flushed;
+ int64_t next_pts;
} FlacEncodeContext;
put_bits(&pb, 24, s->max_framesize);
put_bits(&pb, 20, s->samplerate);
put_bits(&pb, 3, s->channels-1);
- put_bits(&pb, 5, 15); /* bits per sample - 1 */
+ put_bits(&pb, 5, s->avctx->bits_per_raw_sample - 1);
/* write 36-bit sample count in 2 put_bits() calls */
put_bits(&pb, 24, (s->sample_count & 0xFFFFFF000LL) >> 12);
put_bits(&pb, 12, s->sample_count & 0x000000FFFLL);
s->avctx = avctx;
- if (avctx->sample_fmt != AV_SAMPLE_FMT_S16)
- return -1;
+ switch (avctx->sample_fmt) {
+ case AV_SAMPLE_FMT_S16:
+ avctx->bits_per_raw_sample = 16;
+ s->bps_code = 4;
+ break;
+ case AV_SAMPLE_FMT_S32:
+ 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;
+ s->bps_code = 6;
+ break;
+ }
if (channels < 1 || channels > FLAC_MAX_CHANNELS)
return -1;
/* set maximum encoded frame size in verbatim mode */
s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size,
- s->channels, 16);
+ s->channels,
+ s->avctx->bits_per_raw_sample);
/* initialize MD5 context */
s->md5ctx = av_md5_alloc();
s->frame_count = 0;
s->min_framesize = s->max_framesize;
-#if FF_API_OLD_ENCODE_AUDIO
- avctx->coded_frame = avcodec_alloc_frame();
- if (!avctx->coded_frame)
- return AVERROR(ENOMEM);
-#endif
-
ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size,
s->options.max_prediction_order, FF_LPC_TYPE_LEVINSON);
- ff_dsputil_init(&s->dsp, avctx);
- ff_flacdsp_init(&s->flac_dsp, avctx->sample_fmt, 16);
+ ff_bswapdsp_init(&s->bdsp);
+ ff_flacdsp_init(&s->flac_dsp, avctx->sample_fmt,
+ avctx->bits_per_raw_sample);
dprint_compression_options(s);
}
for (ch = 0; ch < s->channels; ch++) {
- frame->subframes[ch].wasted = 0;
- frame->subframes[ch].obits = 16;
+ FlacSubframe *sub = &frame->subframes[ch];
+
+ sub->wasted = 0;
+ sub->obits = s->avctx->bits_per_raw_sample;
+
+ if (sub->obits > 16)
+ sub->rc.coding_mode = CODING_MODE_RICE2;
+ else
+ sub->rc.coding_mode = CODING_MODE_RICE;
}
frame->verbatim_only = 0;
/**
* Copy channel-interleaved input samples into separate subframes.
*/
-static void copy_samples(FlacEncodeContext *s, const int16_t *samples)
+static void copy_samples(FlacEncodeContext *s, const void *samples)
{
int i, j, ch;
FlacFrame *frame;
-
- frame = &s->frame;
- for (i = 0, j = 0; i < frame->blocksize; i++)
- for (ch = 0; ch < s->channels; ch++, j++)
- frame->subframes[ch].samples[i] = samples[j];
+ int shift = av_get_bytes_per_sample(s->avctx->sample_fmt) * 8 -
+ s->avctx->bits_per_raw_sample;
+
+#define COPY_SAMPLES(bits) do { \
+ const int ## bits ## _t *samples0 = samples; \
+ frame = &s->frame; \
+ for (i = 0, j = 0; i < frame->blocksize; i++) \
+ for (ch = 0; ch < s->channels; ch++, j++) \
+ frame->subframes[ch].samples[i] = samples0[j] >> shift; \
+} while (0)
+
+ if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S16)
+ COPY_SAMPLES(16);
+ else
+ COPY_SAMPLES(32);
}
part_end = psize;
for (p = 0; p < 1 << porder; p++) {
int k = sub->rc.params[p];
- count += 4;
+ count += sub->rc.coding_mode;
count += rice_count_exact(&sub->residual[i], part_end - i, k);
i = part_end;
part_end = FFMIN(s->frame.blocksize, part_end + psize);
/**
* Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0.
*/
-static int find_optimal_param(uint64_t sum, int n)
+static int find_optimal_param(uint64_t sum, int n, int max_param)
{
int k;
uint64_t sum2;
return 0;
sum2 = sum - (n >> 1);
k = av_log2(av_clipl_int32(sum2 / n));
- return FFMIN(k, MAX_RICE_PARAM);
+ return FFMIN(k, max_param);
}
uint64_t *sums, int n, int pred_order)
{
int i;
- int k, cnt, part;
+ int k, cnt, part, max_param;
uint64_t all_bits;
+ max_param = (1 << rc->coding_mode) - 2;
+
part = (1 << porder);
all_bits = 4 * part;
cnt = (n >> porder) - pred_order;
for (i = 0; i < part; i++) {
- k = find_optimal_param(sums[i], cnt);
+ k = find_optimal_param(sums[i], cnt, max_param);
rc->params[i] = k;
all_bits += rice_encode_count(sums[i], cnt, k);
cnt = n >> porder;
uint64_t bits[MAX_PARTITION_ORDER+1];
int opt_porder;
RiceContext tmp_rc;
- uint32_t *udata;
- uint64_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
+ uint64_t sums[MAX_PARTITION_ORDER + 1][MAX_PARTITIONS] = { { 0 } };
assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
assert(pmin <= pmax);
- udata = av_malloc(n * sizeof(uint32_t));
+ tmp_rc.coding_mode = rc->coding_mode;
+
for (i = 0; i < n; i++)
- udata[i] = (2*data[i]) ^ (data[i]>>31);
+ rc->udata[i] = (2 * data[i]) ^ (data[i] >> 31);
- calc_sums(pmin, pmax, udata, n, pred_order, sums);
+ calc_sums(pmin, pmax, rc->udata, n, pred_order, sums);
opt_porder = pmin;
bits[pmin] = UINT32_MAX;
}
}
- av_freep(&udata);
return bits[opt_porder];
}
int pmax = get_max_p_order(s->options.max_partition_order,
s->frame.blocksize, pred_order);
- uint64_t bits = 8 + pred_order * sub->obits + 2 + 4;
+ uint64_t bits = 8 + pred_order * sub->obits + 2 + sub->rc.coding_mode;
if (sub->type == FLAC_SUBFRAME_LPC)
bits += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
bits += calc_rice_params(&sub->rc, pmin, pmax, sub->residual,
omethod == ORDER_METHOD_8LEVEL) {
int levels = 1 << omethod;
uint64_t bits[1 << ORDER_METHOD_8LEVEL];
- int order;
+ int order = -1;
int opt_index = levels-1;
opt_order = max_order-1;
bits[opt_index] = UINT32_MAX;
for (i = levels-1; i >= 0; i--) {
+ int last_order = order;
order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
- if (order < 0)
- order = 0;
+ order = av_clip(order, min_order - 1, max_order - 1);
+ if (order == last_order)
+ continue;
s->flac_dsp.lpc_encode(res, smp, n, order+1, coefs[order],
shift[order]);
bits[i] = find_subframe_rice_params(s, sub, order+1);
sub->wasted = v;
sub->obits -= v;
+
+ /* for 24-bit, check if removing wasted bits makes the range better
+ suited for using RICE instead of RICE2 for entropy coding */
+ if (sub->obits <= 17)
+ sub->rc.coding_mode = CODING_MODE_RICE;
}
}
}
-static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
+static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n,
+ int max_rice_param)
{
int i, best;
int32_t lt, rt;
}
/* estimate bit counts */
for (i = 0; i < 4; i++) {
- k = find_optimal_param(2 * sum[i], n);
+ k = find_optimal_param(2 * sum[i], n, max_rice_param);
sum[i] = rice_encode_count( 2 * sum[i], n, k);
}
return;
}
- if (s->options.ch_mode < 0)
- frame->ch_mode = estimate_stereo_mode(left, right, n);
- else
+ if (s->options.ch_mode < 0) {
+ int max_rice_param = (1 << frame->subframes[0].rc.coding_mode) - 2;
+ frame->ch_mode = estimate_stereo_mode(left, right, n, max_rice_param);
+ } else
frame->ch_mode = s->options.ch_mode;
/* perform decorrelation and adjust bits-per-sample */
else
put_bits(&s->pb, 4, frame->ch_mode + FLAC_MAX_CHANNELS - 1);
- put_bits(&s->pb, 3, 4); /* bits-per-sample code */
+ put_bits(&s->pb, 3, s->bps_code);
put_bits(&s->pb, 1, 0);
write_utf8(&s->pb, s->frame_count);
}
/* rice-encoded block */
- put_bits(&s->pb, 2, 0);
+ put_bits(&s->pb, 2, sub->rc.coding_mode - 4);
/* partition order */
porder = sub->rc.porder;
part_end = &sub->residual[psize];
for (p = 0; p < 1 << porder; p++) {
int k = sub->rc.params[p];
- put_bits(&s->pb, 4, k);
+ put_bits(&s->pb, sub->rc.coding_mode, k);
while (res < part_end)
set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0);
part_end = FFMIN(frame_end, part_end + psize);
}
-static int update_md5_sum(FlacEncodeContext *s, const int16_t *samples)
+static int update_md5_sum(FlacEncodeContext *s, const void *samples)
{
const uint8_t *buf;
- int buf_size = s->frame.blocksize * s->channels * 2;
+ int buf_size = s->frame.blocksize * s->channels *
+ ((s->avctx->bits_per_raw_sample + 7) / 8);
- if (HAVE_BIGENDIAN) {
+ if (s->avctx->bits_per_raw_sample > 16 || HAVE_BIGENDIAN) {
av_fast_malloc(&s->md5_buffer, &s->md5_buffer_size, buf_size);
if (!s->md5_buffer)
return AVERROR(ENOMEM);
}
- buf = (const uint8_t *)samples;
+ if (s->avctx->bits_per_raw_sample <= 16) {
+ buf = (const uint8_t *)samples;
#if HAVE_BIGENDIAN
- s->dsp.bswap16_buf((uint16_t *)s->md5_buffer,
- (const uint16_t *)samples, buf_size / 2);
- buf = s->md5_buffer;
+ s->bdsp.bswap16_buf((uint16_t *) s->md5_buffer,
+ (const uint16_t *) samples, buf_size / 2);
+ buf = s->md5_buffer;
#endif
+ } else {
+ int i;
+ const int32_t *samples0 = samples;
+ uint8_t *tmp = s->md5_buffer;
+
+ for (i = 0; i < s->frame.blocksize * s->channels; i++) {
+ int32_t v = samples0[i] >> 8;
+ *tmp++ = (v ) & 0xFF;
+ *tmp++ = (v >> 8) & 0xFF;
+ *tmp++ = (v >> 16) & 0xFF;
+ }
+ buf = s->md5_buffer;
+ }
av_md5_update(s->md5ctx, buf, buf_size);
return 0;
const AVFrame *frame, int *got_packet_ptr)
{
FlacEncodeContext *s;
- const int16_t *samples;
int frame_bytes, out_bytes, ret;
s = avctx->priv_data;
s->max_framesize = s->max_encoded_framesize;
av_md5_final(s->md5ctx, s->md5sum);
write_streaminfo(s, avctx->extradata);
+
+ if (avctx->side_data_only_packets && !s->flushed) {
+ uint8_t *side_data = av_packet_new_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA,
+ avctx->extradata_size);
+ if (!side_data)
+ return AVERROR(ENOMEM);
+ memcpy(side_data, avctx->extradata, avctx->extradata_size);
+
+ avpkt->pts = s->next_pts;
+
+ *got_packet_ptr = 1;
+ s->flushed = 1;
+ }
+
return 0;
}
- samples = (const int16_t *)frame->data[0];
/* change max_framesize for small final frame */
if (frame->nb_samples < s->frame.blocksize) {
s->max_framesize = ff_flac_get_max_frame_size(frame->nb_samples,
- s->channels, 16);
+ s->channels,
+ avctx->bits_per_raw_sample);
}
init_frame(s, frame->nb_samples);
- copy_samples(s, samples);
+ copy_samples(s, frame->data[0]);
channel_decorrelation(s);
frame_bytes = encode_frame(s);
- /* fallback to verbatim mode if the compressed frame is larger than it
+ /* Fall back on verbatim mode if the compressed frame is larger than it
would be if encoded uncompressed. */
if (frame_bytes < 0 || frame_bytes > s->max_framesize) {
s->frame.verbatim_only = 1;
s->frame_count++;
s->sample_count += frame->nb_samples;
- if ((ret = update_md5_sum(s, samples)) < 0) {
+ if ((ret = update_md5_sum(s, frame->data[0])) < 0) {
av_log(avctx, AV_LOG_ERROR, "Error updating MD5 checksum\n");
return ret;
}
avpkt->pts = frame->pts;
avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);
avpkt->size = out_bytes;
+
+ s->next_pts = avpkt->pts + avpkt->duration;
+
*got_packet_ptr = 1;
return 0;
}
}
av_freep(&avctx->extradata);
avctx->extradata_size = 0;
-#if FF_API_OLD_ENCODE_AUDIO
- av_freep(&avctx->coded_frame);
-#endif
return 0;
}
{ "fixed", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_FIXED }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
{ "levinson", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_LEVINSON }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
{ "cholesky", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_CHOLESKY }, INT_MIN, INT_MAX, FLAGS, "lpc_type" },
-{ "lpc_passes", "Number of passes to use for Cholesky factorization during LPC analysis", offsetof(FlacEncodeContext, options.lpc_passes), AV_OPT_TYPE_INT, {.i64 = -1 }, INT_MIN, INT_MAX, FLAGS },
+{ "lpc_passes", "Number of passes to use for Cholesky factorization during LPC analysis", offsetof(FlacEncodeContext, options.lpc_passes), AV_OPT_TYPE_INT, {.i64 = 1 }, 1, INT_MAX, FLAGS },
{ "min_partition_order", NULL, offsetof(FlacEncodeContext, options.min_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
{ "max_partition_order", NULL, offsetof(FlacEncodeContext, options.max_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS },
{ "prediction_order_method", "Search method for selecting prediction order", offsetof(FlacEncodeContext, options.prediction_order_method), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, ORDER_METHOD_LOG, FLAGS, "predm" },
AVCodec ff_flac_encoder = {
.name = "flac",
+ .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
.type = AVMEDIA_TYPE_AUDIO,
.id = AV_CODEC_ID_FLAC,
.priv_data_size = sizeof(FlacEncodeContext),
.close = flac_encode_close,
.capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY,
.sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
+ AV_SAMPLE_FMT_S32,
AV_SAMPLE_FMT_NONE },
- .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
.priv_class = &flac_encoder_class,
};
projects / ffmpeg / blobdiff