#include "libavutil/md5.h"
#include "avcodec.h"
#include "get_bits.h"
-#include "dsputil.h"
#include "golomb.h"
#include "lpc.h"
#include "flac.h"
int bs_code[2];
uint8_t crc8;
int ch_mode;
+ int verbatim_only;
} FlacFrame;
typedef struct FlacEncodeContext {
FlacFrame frame;
CompressionOptions options;
AVCodecContext *avctx;
- DSPContext dsp;
+ LPCContext lpc_ctx;
struct AVMD5 *md5ctx;
} FlacEncodeContext;
int freq = avctx->sample_rate;
int channels = avctx->channels;
FlacEncodeContext *s = avctx->priv_data;
- int i, level;
+ int i, level, ret;
uint8_t *streaminfo;
s->avctx = avctx;
- dsputil_init(&s->dsp, avctx);
-
- if (avctx->sample_fmt != SAMPLE_FMT_S16)
+ if (avctx->sample_fmt != AV_SAMPLE_FMT_S16)
return -1;
if (channels < 1 || channels > FLAC_MAX_CHANNELS)
s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
/* set compression option overrides from AVCodecContext */
-#if LIBAVCODEC_VERSION_MAJOR < 53
+#if FF_API_USE_LPC
/* for compatibility with deprecated AVCodecContext.use_lpc */
if (avctx->use_lpc == 0) {
s->options.lpc_type = AV_LPC_TYPE_FIXED;
if (!avctx->coded_frame)
return AVERROR(ENOMEM);
+ ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size,
+ s->options.max_prediction_order, AV_LPC_TYPE_LEVINSON);
+
dprint_compression_options(s);
- return 0;
+ return ret;
}
for (ch = 0; ch < s->channels; ch++)
frame->subframes[ch].obits = 16;
+
+ frame->verbatim_only = 0;
}
}
+static int rice_count_exact(int32_t *res, int n, int k)
+{
+ int i;
+ int count = 0;
+
+ for (i = 0; i < n; i++) {
+ int32_t v = -2 * res[i] - 1;
+ v ^= v >> 31;
+ count += (v >> k) + 1 + k;
+ }
+ return count;
+}
+
+
+static int subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub,
+ int pred_order)
+{
+ int p, porder, psize;
+ int i, part_end;
+ int count = 0;
+
+ /* subframe header */
+ count += 8;
+
+ /* subframe */
+ if (sub->type == FLAC_SUBFRAME_CONSTANT) {
+ count += sub->obits;
+ } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
+ count += s->frame.blocksize * sub->obits;
+ } else {
+ /* warm-up samples */
+ count += pred_order * sub->obits;
+
+ /* LPC coefficients */
+ if (sub->type == FLAC_SUBFRAME_LPC)
+ count += 4 + 5 + pred_order * s->options.lpc_coeff_precision;
+
+ /* rice-encoded block */
+ count += 2;
+
+ /* partition order */
+ porder = sub->rc.porder;
+ psize = s->frame.blocksize >> porder;
+ count += 4;
+
+ /* residual */
+ i = pred_order;
+ part_end = psize;
+ for (p = 0; p < 1 << porder; p++) {
+ int k = sub->rc.params[p];
+ count += 4;
+ count += rice_count_exact(&sub->residual[i], part_end - i, k);
+ i = part_end;
+ part_end = FFMIN(s->frame.blocksize, part_end + psize);
+ }
+ }
+
+ return count;
+}
+
+
#define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
/**
}
-static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
- int32_t *data, int n, int pred_order,
- int bps)
+static uint32_t find_subframe_rice_params(FlacEncodeContext *s,
+ FlacSubframe *sub, int pred_order)
{
- uint32_t bits;
- pmin = get_max_p_order(pmin, n, pred_order);
- pmax = get_max_p_order(pmax, n, pred_order);
- bits = pred_order * bps + 6;
- bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
+ int pmin = get_max_p_order(s->options.min_partition_order,
+ s->frame.blocksize, pred_order);
+ int pmax = get_max_p_order(s->options.max_partition_order,
+ s->frame.blocksize, pred_order);
+
+ uint32_t bits = 8 + pred_order * sub->obits + 2 + 4;
+ 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,
+ s->frame.blocksize, pred_order);
return bits;
}
-static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
- int32_t *data, int n, int pred_order,
- int bps, int precision)
-{
- uint32_t bits;
- pmin = get_max_p_order(pmin, n, pred_order);
- pmax = get_max_p_order(pmax, n, pred_order);
- bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
- bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
- return bits;
-}
-
-
-static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
-{
- assert(n > 0);
- memcpy(res, smp, n * sizeof(int32_t));
-}
-
-
static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
int order)
{
}
-static int encode_residual(FlacEncodeContext *ctx, int ch)
+static int encode_residual_ch(FlacEncodeContext *s, int ch)
{
int i, n;
- int min_order, max_order, opt_order, precision, omethod;
- int min_porder, max_porder;
+ int min_order, max_order, opt_order, omethod;
FlacFrame *frame;
FlacSubframe *sub;
int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
int shift[MAX_LPC_ORDER];
int32_t *res, *smp;
- frame = &ctx->frame;
+ frame = &s->frame;
sub = &frame->subframes[ch];
res = sub->residual;
smp = sub->samples;
if (i == n) {
sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
res[0] = smp[0];
- return sub->obits;
+ return subframe_count_exact(s, sub, 0);
}
/* VERBATIM */
- if (n < 5) {
+ if (frame->verbatim_only || n < 5) {
sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
- encode_residual_verbatim(res, smp, n);
- return sub->obits * n;
+ memcpy(res, smp, n * sizeof(int32_t));
+ return subframe_count_exact(s, sub, 0);
}
- min_order = ctx->options.min_prediction_order;
- max_order = ctx->options.max_prediction_order;
- min_porder = ctx->options.min_partition_order;
- max_porder = ctx->options.max_partition_order;
- precision = ctx->options.lpc_coeff_precision;
- omethod = ctx->options.prediction_order_method;
+ min_order = s->options.min_prediction_order;
+ max_order = s->options.max_prediction_order;
+ omethod = s->options.prediction_order_method;
/* FIXED */
- if (ctx->options.lpc_type == AV_LPC_TYPE_NONE ||
- ctx->options.lpc_type == AV_LPC_TYPE_FIXED || n <= max_order) {
+ sub->type = FLAC_SUBFRAME_FIXED;
+ if (s->options.lpc_type == AV_LPC_TYPE_NONE ||
+ s->options.lpc_type == AV_LPC_TYPE_FIXED || n <= max_order) {
uint32_t bits[MAX_FIXED_ORDER+1];
if (max_order > MAX_FIXED_ORDER)
max_order = MAX_FIXED_ORDER;
bits[0] = UINT32_MAX;
for (i = min_order; i <= max_order; i++) {
encode_residual_fixed(res, smp, n, i);
- bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res,
- n, i, sub->obits);
+ bits[i] = find_subframe_rice_params(s, sub, i);
if (bits[i] < bits[opt_order])
opt_order = i;
}
sub->order = opt_order;
- sub->type = FLAC_SUBFRAME_FIXED;
sub->type_code = sub->type | sub->order;
if (sub->order != max_order) {
encode_residual_fixed(res, smp, n, sub->order);
- return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n,
- sub->order, sub->obits);
+ find_subframe_rice_params(s, sub, sub->order);
}
- return bits[sub->order];
+ return subframe_count_exact(s, sub, sub->order);
}
/* LPC */
- opt_order = ff_lpc_calc_coefs(&ctx->dsp, smp, n, min_order, max_order,
- precision, coefs, shift, ctx->options.lpc_type,
- ctx->options.lpc_passes, omethod,
+ sub->type = FLAC_SUBFRAME_LPC;
+ opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order,
+ s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type,
+ s->options.lpc_passes, omethod,
MAX_LPC_SHIFT, 0);
if (omethod == ORDER_METHOD_2LEVEL ||
if (order < 0)
order = 0;
encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
- bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
- res, n, order+1, sub->obits, precision);
+ bits[i] = find_subframe_rice_params(s, sub, order+1);
if (bits[i] < bits[opt_index]) {
opt_index = i;
opt_order = order;
bits[0] = UINT32_MAX;
for (i = min_order-1; i < max_order; i++) {
encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
- bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
- res, n, i+1, sub->obits, precision);
+ bits[i] = find_subframe_rice_params(s, sub, i+1);
if (bits[i] < bits[opt_order])
opt_order = i;
}
if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX)
continue;
encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
- bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
- res, n, i+1, sub->obits,
- precision);
+ bits[i] = find_subframe_rice_params(s, sub, i+1);
if (bits[i] < bits[opt_order])
opt_order = i;
}
}
sub->order = opt_order;
- sub->type = FLAC_SUBFRAME_LPC;
sub->type_code = sub->type | (sub->order-1);
sub->shift = shift[sub->order-1];
for (i = 0; i < sub->order; i++)
encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
- return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n,
- sub->order, sub->obits, precision);
+ find_subframe_rice_params(s, sub, sub->order);
+
+ return subframe_count_exact(s, sub, sub->order);
}
-static int encode_residual_v(FlacEncodeContext *ctx, int ch)
+static int count_frame_header(FlacEncodeContext *s)
{
- int i, n;
- FlacFrame *frame;
- FlacSubframe *sub;
- int32_t *res, *smp;
+ uint8_t tmp;
+ int count;
+
+ /*
+ <14> Sync code
+ <1> Reserved
+ <1> Blocking strategy
+ <4> Block size in inter-channel samples
+ <4> Sample rate
+ <4> Channel assignment
+ <3> Sample size in bits
+ <1> Reserved
+ */
+ count = 32;
+
+ /* coded frame number */
+ PUT_UTF8(s->frame_count, tmp, count += 8;)
+
+ /* explicit block size */
+ if (s->frame.bs_code[0] == 6)
+ count += 8;
+ else if (s->frame.bs_code[0] == 7)
+ count += 16;
+
+ /* explicit sample rate */
+ count += ((s->sr_code[0] == 12) + (s->sr_code[0] > 12)) * 8;
+
+ /* frame header CRC-8 */
+ count += 8;
+
+ return count;
+}
- frame = &ctx->frame;
- sub = &frame->subframes[ch];
- res = sub->residual;
- smp = sub->samples;
- n = frame->blocksize;
- /* CONSTANT */
- for (i = 1; i < n; i++)
- if (smp[i] != smp[0])
- break;
- if (i == n) {
- sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
- res[0] = smp[0];
- return sub->obits;
- }
+static int encode_frame(FlacEncodeContext *s)
+{
+ int ch, count;
- /* VERBATIM */
- sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
- encode_residual_verbatim(res, smp, n);
- return sub->obits * n;
+ count = count_frame_header(s);
+
+ for (ch = 0; ch < s->channels; ch++)
+ count += encode_residual_ch(s, ch);
+
+ count += (8 - (count & 7)) & 7; // byte alignment
+ count += 16; // CRC-16
+
+ return count >> 3;
}
/**
* Perform stereo channel decorrelation.
*/
-static void channel_decorrelation(FlacEncodeContext *ctx)
+static void channel_decorrelation(FlacEncodeContext *s)
{
FlacFrame *frame;
int32_t *left, *right;
int i, n;
- frame = &ctx->frame;
+ frame = &s->frame;
n = frame->blocksize;
left = frame->subframes[0].samples;
right = frame->subframes[1].samples;
- if (ctx->channels != 2) {
+ if (s->channels != 2) {
frame->ch_mode = FLAC_CHMODE_INDEPENDENT;
return;
}
}
-static void output_frame_header(FlacEncodeContext *s)
+static void write_frame_header(FlacEncodeContext *s)
{
FlacFrame *frame;
int crc;
}
-static void output_subframe_constant(FlacEncodeContext *s, int ch)
+static void write_subframes(FlacEncodeContext *s)
{
- FlacSubframe *sub;
- int32_t res;
-
- sub = &s->frame.subframes[ch];
- res = sub->residual[0];
- put_sbits(&s->pb, sub->obits, res);
-}
-
-
-static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
-{
- int i;
- FlacFrame *frame;
- FlacSubframe *sub;
- int32_t res;
-
- frame = &s->frame;
- sub = &frame->subframes[ch];
-
- for (i = 0; i < frame->blocksize; i++) {
- res = sub->residual[i];
- put_sbits(&s->pb, sub->obits, res);
- }
-}
-
-
-static void output_residual(FlacEncodeContext *ctx, int ch)
-{
- int i, j, p, n, parts;
- int k, porder, psize, res_cnt;
- FlacFrame *frame;
- FlacSubframe *sub;
- int32_t *res;
-
- frame = &ctx->frame;
- sub = &frame->subframes[ch];
- res = sub->residual;
- n = frame->blocksize;
-
- /* rice-encoded block */
- put_bits(&ctx->pb, 2, 0);
-
- /* partition order */
- porder = sub->rc.porder;
- psize = n >> porder;
- parts = (1 << porder);
- put_bits(&ctx->pb, 4, porder);
- res_cnt = psize - sub->order;
-
- /* residual */
- j = sub->order;
- for (p = 0; p < parts; p++) {
- k = sub->rc.params[p];
- put_bits(&ctx->pb, 4, k);
- if (p == 1)
- res_cnt = psize;
- for (i = 0; i < res_cnt && j < n; i++, j++)
- set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0);
- }
-}
-
-
-static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
-{
- int i;
- FlacFrame *frame;
- FlacSubframe *sub;
-
- frame = &ctx->frame;
- sub = &frame->subframes[ch];
-
- /* warm-up samples */
- for (i = 0; i < sub->order; i++)
- put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
-
- /* residual */
- output_residual(ctx, ch);
-}
-
-
-static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
-{
- int i, cbits;
- FlacFrame *frame;
- FlacSubframe *sub;
-
- frame = &ctx->frame;
- sub = &frame->subframes[ch];
-
- /* warm-up samples */
- for (i = 0; i < sub->order; i++)
- put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
-
- /* LPC coefficients */
- cbits = ctx->options.lpc_coeff_precision;
- put_bits( &ctx->pb, 4, cbits-1);
- put_sbits(&ctx->pb, 5, sub->shift);
- for (i = 0; i < sub->order; i++)
- put_sbits(&ctx->pb, cbits, sub->coefs[i]);
-
- /* residual */
- output_residual(ctx, ch);
-}
-
-
-static void output_subframes(FlacEncodeContext *s)
-{
- FlacFrame *frame;
- FlacSubframe *sub;
int ch;
- frame = &s->frame;
-
for (ch = 0; ch < s->channels; ch++) {
- sub = &frame->subframes[ch];
+ FlacSubframe *sub = &s->frame.subframes[ch];
+ int i, p, porder, psize;
+ int32_t *part_end;
+ int32_t *res = sub->residual;
+ int32_t *frame_end = &sub->residual[s->frame.blocksize];
/* subframe header */
put_bits(&s->pb, 1, 0);
put_bits(&s->pb, 1, 0); /* no wasted bits */
/* subframe */
- if(sub->type == FLAC_SUBFRAME_CONSTANT)
- output_subframe_constant(s, ch);
- else if(sub->type == FLAC_SUBFRAME_VERBATIM)
- output_subframe_verbatim(s, ch);
- else if(sub->type == FLAC_SUBFRAME_FIXED)
- output_subframe_fixed(s, ch);
- else if(sub->type == FLAC_SUBFRAME_LPC)
- output_subframe_lpc(s, ch);
+ if (sub->type == FLAC_SUBFRAME_CONSTANT) {
+ put_sbits(&s->pb, sub->obits, res[0]);
+ } else if (sub->type == FLAC_SUBFRAME_VERBATIM) {
+ while (res < frame_end)
+ put_sbits(&s->pb, sub->obits, *res++);
+ } else {
+ /* warm-up samples */
+ for (i = 0; i < sub->order; i++)
+ put_sbits(&s->pb, sub->obits, *res++);
+
+ /* LPC coefficients */
+ if (sub->type == FLAC_SUBFRAME_LPC) {
+ int cbits = s->options.lpc_coeff_precision;
+ put_bits( &s->pb, 4, cbits-1);
+ put_sbits(&s->pb, 5, sub->shift);
+ for (i = 0; i < sub->order; i++)
+ put_sbits(&s->pb, cbits, sub->coefs[i]);
+ }
+
+ /* rice-encoded block */
+ put_bits(&s->pb, 2, 0);
+
+ /* partition order */
+ porder = sub->rc.porder;
+ psize = s->frame.blocksize >> porder;
+ put_bits(&s->pb, 4, porder);
+
+ /* residual */
+ part_end = &sub->residual[psize];
+ for (p = 0; p < 1 << porder; p++) {
+ int k = sub->rc.params[p];
+ put_bits(&s->pb, 4, 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 void output_frame_footer(FlacEncodeContext *s)
+static void write_frame_footer(FlacEncodeContext *s)
{
int crc;
flush_put_bits(&s->pb);
}
+static int write_frame(FlacEncodeContext *s, uint8_t *frame, int buf_size)
+{
+ init_put_bits(&s->pb, frame, buf_size);
+ write_frame_header(s);
+ write_subframes(s);
+ write_frame_footer(s);
+ return put_bits_count(&s->pb) >> 3;
+}
+
+
static void update_md5_sum(FlacEncodeContext *s, const int16_t *samples)
{
#if HAVE_BIGENDIAN
static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
int buf_size, void *data)
{
- int ch;
FlacEncodeContext *s;
const int16_t *samples = data;
- int out_bytes;
- int reencoded=0;
+ int frame_bytes, out_bytes;
s = avctx->priv_data;
- if (buf_size < s->max_framesize * 2) {
- av_log(avctx, AV_LOG_ERROR, "output buffer too small\n");
- return 0;
- }
-
/* when the last block is reached, update the header in extradata */
if (!data) {
s->max_framesize = s->max_encoded_framesize;
return 0;
}
+ /* change max_framesize for small final frame */
+ if (avctx->frame_size < s->frame.blocksize) {
+ s->max_framesize = ff_flac_get_max_frame_size(avctx->frame_size,
+ s->channels, 16);
+ }
+
init_frame(s);
copy_samples(s, samples);
channel_decorrelation(s);
- for (ch = 0; ch < s->channels; ch++)
- encode_residual(s, ch);
+ frame_bytes = encode_frame(s);
-write_frame:
- init_put_bits(&s->pb, frame, buf_size);
- output_frame_header(s);
- output_subframes(s);
- output_frame_footer(s);
- out_bytes = put_bits_count(&s->pb) >> 3;
-
- if (out_bytes > s->max_framesize) {
- if (reencoded) {
- /* still too large. must be an error. */
- av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
- return -1;
- }
+ /* fallback to verbatim mode if the compressed frame is larger than it
+ would be if encoded uncompressed. */
+ if (frame_bytes > s->max_framesize) {
+ s->frame.verbatim_only = 1;
+ frame_bytes = encode_frame(s);
+ }
- /* frame too large. use verbatim mode */
- for (ch = 0; ch < s->channels; ch++)
- encode_residual_v(s, ch);
- reencoded = 1;
- goto write_frame;
+ if (buf_size < frame_bytes) {
+ av_log(avctx, AV_LOG_ERROR, "output buffer too small\n");
+ return 0;
}
+ out_bytes = write_frame(s, frame, buf_size);
s->frame_count++;
avctx->coded_frame->pts = s->sample_count;
if (avctx->priv_data) {
FlacEncodeContext *s = avctx->priv_data;
av_freep(&s->md5ctx);
+ ff_lpc_end(&s->lpc_ctx);
}
av_freep(&avctx->extradata);
avctx->extradata_size = 0;
}
-AVCodec flac_encoder = {
+AVCodec ff_flac_encoder = {
"flac",
AVMEDIA_TYPE_AUDIO,
CODEC_ID_FLAC,
flac_encode_close,
NULL,
.capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY,
- .sample_fmts = (const enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
+ .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
.long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),
};