#define FLAC_CHMODE_RIGHT_SIDE 9
#define FLAC_CHMODE_MID_SIDE 10
+#define ORDER_METHOD_EST 0
+#define ORDER_METHOD_2LEVEL 1
+#define ORDER_METHOD_4LEVEL 2
+#define ORDER_METHOD_8LEVEL 3
+#define ORDER_METHOD_SEARCH 4
+
#define FLAC_STREAMINFO_SIZE 34
+#define MIN_LPC_ORDER 1
+#define MAX_LPC_ORDER 32
+#define MAX_FIXED_ORDER 4
+#define MAX_PARTITION_ORDER 8
+#define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
+#define MAX_LPC_PRECISION 15
+#define MAX_LPC_SHIFT 15
+#define MAX_RICE_PARAM 14
+
+typedef struct CompressionOptions {
+ int compression_level;
+ int block_time_ms;
+ int use_lpc;
+ int lpc_coeff_precision;
+ int min_prediction_order;
+ int max_prediction_order;
+ int prediction_order_method;
+ int min_partition_order;
+ int max_partition_order;
+} CompressionOptions;
+
+typedef struct RiceContext {
+ int porder;
+ int params[MAX_PARTITIONS];
+} RiceContext;
+
typedef struct FlacSubframe {
int type;
int type_code;
int obits;
int order;
+ int32_t coefs[MAX_LPC_ORDER];
+ int shift;
+ RiceContext rc;
int32_t samples[FLAC_MAX_BLOCKSIZE];
int32_t residual[FLAC_MAX_BLOCKSIZE];
} FlacSubframe;
int max_framesize;
uint32_t frame_count;
FlacFrame frame;
+ CompressionOptions options;
+ AVCodecContext *avctx;
} FlacEncodeContext;
static const int flac_samplerates[16] = {
256, 512, 1024, 2048, 4096, 8192, 16384, 32768
};
-static const int flac_blocksizes_ordered[14] = {
- 0, 192, 256, 512, 576, 1024, 1152, 2048, 2304, 4096, 4608, 8192, 16384, 32768
-};
-
/**
* Writes streaminfo metadata block to byte array
*/
/* MD5 signature = 0 */
}
-#define BLOCK_TIME_MS 105
-
/**
* Sets blocksize based on samplerate
* Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
*/
-static int select_blocksize(int samplerate)
+static int select_blocksize(int samplerate, int block_time_ms)
{
int i;
int target;
int blocksize;
assert(samplerate > 0);
- blocksize = 0;
- target = (samplerate * BLOCK_TIME_MS) / 1000;
- for(i=13; i>=0; i--) {
- if(target >= flac_blocksizes_ordered[i]) {
- blocksize = flac_blocksizes_ordered[i];
- break;
+ blocksize = flac_blocksizes[1];
+ target = (samplerate * block_time_ms) / 1000;
+ for(i=0; i<16; i++) {
+ if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) {
+ blocksize = flac_blocksizes[i];
}
}
- if(blocksize == 0) {
- blocksize = flac_blocksizes_ordered[1];
- }
return blocksize;
}
int freq = avctx->sample_rate;
int channels = avctx->channels;
FlacEncodeContext *s = avctx->priv_data;
- int i;
+ int i, level;
uint8_t *streaminfo;
- if(s == NULL) {
- return -1;
- }
+ s->avctx = avctx;
if(avctx->sample_fmt != SAMPLE_FMT_S16) {
return -1;
s->samplerate = freq;
}
- s->blocksize = select_blocksize(s->samplerate);
- avctx->frame_size = s->blocksize;
+ /* set compression option defaults based on avctx->compression_level */
+ if(avctx->compression_level < 0) {
+ s->options.compression_level = 5;
+ } else {
+ s->options.compression_level = avctx->compression_level;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
+
+ level= s->options.compression_level;
+ if(level > 5) {
+ av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
+ s->options.compression_level);
+ return -1;
+ }
+
+ s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105})[level];
+ s->options.use_lpc = ((int[]){ 0, 0, 0, 1, 1, 1})[level];
+ s->options.min_prediction_order= ((int[]){ 2, 0, 0, 1, 1, 1})[level];
+ s->options.max_prediction_order= ((int[]){ 3, 4, 4, 6, 8, 8})[level];
+ s->options.prediction_order_method = ORDER_METHOD_EST;
+ s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0})[level];
+ s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8})[level];
+
+ /* set compression option overrides from AVCodecContext */
+ if(avctx->use_lpc >= 0) {
+ s->options.use_lpc = !!avctx->use_lpc;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " use lpc: %s\n",
+ s->options.use_lpc? "yes" : "no");
+
+ if(avctx->min_prediction_order >= 0) {
+ if(s->options.use_lpc) {
+ if(avctx->min_prediction_order < MIN_LPC_ORDER ||
+ avctx->min_prediction_order > MAX_LPC_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
+ avctx->min_prediction_order);
+ return -1;
+ }
+ } else {
+ if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
+ avctx->min_prediction_order);
+ return -1;
+ }
+ }
+ s->options.min_prediction_order = avctx->min_prediction_order;
+ }
+ if(avctx->max_prediction_order >= 0) {
+ if(s->options.use_lpc) {
+ if(avctx->max_prediction_order < MIN_LPC_ORDER ||
+ avctx->max_prediction_order > MAX_LPC_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
+ avctx->max_prediction_order);
+ return -1;
+ }
+ } else {
+ if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
+ avctx->max_prediction_order);
+ return -1;
+ }
+ }
+ s->options.max_prediction_order = avctx->max_prediction_order;
+ }
+ if(s->options.max_prediction_order < s->options.min_prediction_order) {
+ av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
+ s->options.min_prediction_order, s->options.max_prediction_order);
+ return -1;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
+ s->options.min_prediction_order, s->options.max_prediction_order);
+
+ if(avctx->prediction_order_method >= 0) {
+ if(avctx->prediction_order_method > ORDER_METHOD_SEARCH) {
+ av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
+ avctx->prediction_order_method);
+ return -1;
+ }
+ s->options.prediction_order_method = avctx->prediction_order_method;
+ }
+ switch(avctx->prediction_order_method) {
+ case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "estimate"); break;
+ case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "2-level"); break;
+ case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "4-level"); break;
+ case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "8-level"); break;
+ case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "full search"); break;
+ }
+
+ if(avctx->min_partition_order >= 0) {
+ if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
+ avctx->min_partition_order);
+ return -1;
+ }
+ s->options.min_partition_order = avctx->min_partition_order;
+ }
+ if(avctx->max_partition_order >= 0) {
+ if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
+ avctx->max_partition_order);
+ return -1;
+ }
+ s->options.max_partition_order = avctx->max_partition_order;
+ }
+ if(s->options.max_partition_order < s->options.min_partition_order) {
+ av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
+ s->options.min_partition_order, s->options.max_partition_order);
+ return -1;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
+ s->options.min_partition_order, s->options.max_partition_order);
+
+ if(avctx->frame_size > 0) {
+ if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
+ avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
+ av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
+ avctx->frame_size);
+ return -1;
+ }
+ s->blocksize = avctx->frame_size;
+ } else {
+ s->blocksize = select_blocksize(s->samplerate, s->options.block_time_ms);
+ avctx->frame_size = s->blocksize;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->blocksize);
+
+ /* set LPC precision */
+ if(avctx->lpc_coeff_precision > 0) {
+ if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
+ av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
+ avctx->lpc_coeff_precision);
+ return -1;
+ }
+ s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
+ } else {
+ /* select LPC precision based on block size */
+ if( s->blocksize <= 192) s->options.lpc_coeff_precision = 7;
+ else if(s->blocksize <= 384) s->options.lpc_coeff_precision = 8;
+ else if(s->blocksize <= 576) s->options.lpc_coeff_precision = 9;
+ else if(s->blocksize <= 1152) s->options.lpc_coeff_precision = 10;
+ else if(s->blocksize <= 2304) s->options.lpc_coeff_precision = 11;
+ else if(s->blocksize <= 4608) s->options.lpc_coeff_precision = 12;
+ else if(s->blocksize <= 8192) s->options.lpc_coeff_precision = 13;
+ else if(s->blocksize <= 16384) s->options.lpc_coeff_precision = 14;
+ else s->options.lpc_coeff_precision = 15;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
+ s->options.lpc_coeff_precision);
- s->max_framesize = 14 + (s->blocksize * s->channels * 2);
+ /* set maximum encoded frame size in verbatim mode */
+ if(s->channels == 2) {
+ s->max_framesize = 14 + ((s->blocksize * 33 + 7) >> 3);
+ } else {
+ s->max_framesize = 14 + (s->blocksize * s->channels * 2);
+ }
streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
write_streaminfo(s, streaminfo);
return 0;
}
-static int init_frame(FlacEncodeContext *s)
+static void init_frame(FlacEncodeContext *s)
{
int i, ch;
FlacFrame *frame;
for(ch=0; ch<s->channels; ch++) {
frame->subframes[ch].obits = 16;
}
- if(s->channels == 2) {
- frame->ch_mode = FLAC_CHMODE_LEFT_RIGHT;
- } else {
- frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
- }
-
- return 0;
}
/**
}
}
-static void encode_residual_verbatim(FlacEncodeContext *s, int ch)
+
+#define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
+
+static int find_optimal_param(uint32_t sum, int n)
{
- FlacFrame *frame;
- FlacSubframe *sub;
- int32_t *res;
- int32_t *smp;
- int n;
+ int k, k_opt;
+ uint32_t nbits[MAX_RICE_PARAM+1];
+
+ k_opt = 0;
+ nbits[0] = UINT32_MAX;
+ for(k=0; k<=MAX_RICE_PARAM; k++) {
+ nbits[k] = rice_encode_count(sum, n, k);
+ if(nbits[k] < nbits[k_opt]) {
+ k_opt = k;
+ }
+ }
+ return k_opt;
+}
- frame = &s->frame;
- sub = &frame->subframes[ch];
- res = sub->residual;
- smp = sub->samples;
- n = frame->blocksize;
+static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
+ uint32_t *sums, int n, int pred_order)
+{
+ int i;
+ int k, cnt, part;
+ uint32_t all_bits;
+
+ part = (1 << porder);
+ all_bits = 0;
+
+ cnt = (n >> porder) - pred_order;
+ for(i=0; i<part; i++) {
+ if(i == 1) cnt = (n >> porder);
+ k = find_optimal_param(sums[i], cnt);
+ rc->params[i] = k;
+ all_bits += rice_encode_count(sums[i], cnt, k);
+ }
+ all_bits += (4 * part);
+
+ rc->porder = porder;
+
+ return all_bits;
+}
+
+static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
+ uint32_t sums[][MAX_PARTITIONS])
+{
+ int i, j;
+ int parts;
+ uint32_t *res, *res_end;
+
+ /* sums for highest level */
+ parts = (1 << pmax);
+ res = &data[pred_order];
+ res_end = &data[n >> pmax];
+ for(i=0; i<parts; i++) {
+ sums[pmax][i] = 0;
+ while(res < res_end){
+ sums[pmax][i] += *(res++);
+ }
+ res_end+= n >> pmax;
+ }
+ /* sums for lower levels */
+ for(i=pmax-1; i>=pmin; i--) {
+ parts = (1 << i);
+ for(j=0; j<parts; j++) {
+ sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
+ }
+ }
+}
+
+static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
+ int32_t *data, int n, int pred_order)
+{
+ int i;
+ uint32_t bits[MAX_PARTITION_ORDER+1];
+ int opt_porder;
+ RiceContext tmp_rc;
+ uint32_t *udata;
+ uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
+
+ assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
+ assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
+ assert(pmin <= pmax);
+
+ udata = av_malloc(n * sizeof(uint32_t));
+ for(i=0; i<n; i++) {
+ udata[i] = (2*data[i]) ^ (data[i]>>31);
+ }
+
+ calc_sums(pmin, pmax, udata, n, pred_order, sums);
+
+ opt_porder = pmin;
+ bits[pmin] = UINT32_MAX;
+ for(i=pmin; i<=pmax; i++) {
+ bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
+ if(bits[i] <= bits[opt_porder]) {
+ opt_porder = i;
+ *rc= tmp_rc;
+ }
+ }
+
+ av_freep(&udata);
+ return bits[opt_porder];
+}
+
+static int get_max_p_order(int max_porder, int n, int order)
+{
+ int porder = FFMIN(max_porder, av_log2(n^(n-1)));
+ if(order > 0)
+ porder = FFMIN(porder, av_log2(n/order));
+ return porder;
+}
+
+static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
+ int32_t *data, int n, int pred_order,
+ int bps)
+{
+ 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);
+ 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;
+}
+
+/**
+ * Apply Welch window function to audio block
+ */
+static void apply_welch_window(const int32_t *data, int len, double *w_data)
+{
+ int i, n2;
+ double w;
+ double c;
+
+ n2 = (len >> 1);
+ c = 2.0 / (len - 1.0);
+ for(i=0; i<n2; i++) {
+ w = c - i - 1.0;
+ w = 1.0 - (w * w);
+ w_data[i] = data[i] * w;
+ w_data[len-1-i] = data[len-1-i] * w;
+ }
+}
+
+/**
+ * Calculates autocorrelation data from audio samples
+ * A Welch window function is applied before calculation.
+ */
+static void compute_autocorr(const int32_t *data, int len, int lag,
+ double *autoc)
+{
+ int i, lag_ptr;
+ double tmp[len + lag];
+ double *data1= tmp + lag;
+
+ apply_welch_window(data, len, data1);
+
+ for(i=0; i<lag; i++){
+ autoc[i] = 1.0;
+ data1[i-lag]= 0.0;
+ }
+
+ for(i=0; i<len; i++){
+ for(lag_ptr= i-lag; lag_ptr<=i; lag_ptr++){
+ autoc[i-lag_ptr] += data1[i] * data1[lag_ptr];
+ }
+ }
+}
+
+/**
+ * Levinson-Durbin recursion.
+ * Produces LPC coefficients from autocorrelation data.
+ */
+static void compute_lpc_coefs(const double *autoc, int max_order,
+ double lpc[][MAX_LPC_ORDER], double *ref)
+{
+ int i, j, i2;
+ double r, err, tmp;
+ double lpc_tmp[MAX_LPC_ORDER];
+
+ for(i=0; i<max_order; i++) lpc_tmp[i] = 0;
+ err = autoc[0];
+
+ for(i=0; i<max_order; i++) {
+ r = -autoc[i+1];
+ for(j=0; j<i; j++) {
+ r -= lpc_tmp[j] * autoc[i-j];
+ }
+ r /= err;
+ ref[i] = fabs(r);
+
+ err *= 1.0 - (r * r);
+
+ i2 = (i >> 1);
+ lpc_tmp[i] = r;
+ for(j=0; j<i2; j++) {
+ tmp = lpc_tmp[j];
+ lpc_tmp[j] += r * lpc_tmp[i-1-j];
+ lpc_tmp[i-1-j] += r * tmp;
+ }
+ if(i & 1) {
+ lpc_tmp[j] += lpc_tmp[j] * r;
+ }
+
+ for(j=0; j<=i; j++) {
+ lpc[i][j] = -lpc_tmp[j];
+ }
+ }
+}
+
+/**
+ * Quantize LPC coefficients
+ */
+static void quantize_lpc_coefs(double *lpc_in, int order, int precision,
+ int32_t *lpc_out, int *shift)
+{
+ int i;
+ double cmax;
+ int32_t qmax;
+ int sh;
+
+ /* define maximum levels */
+ qmax = (1 << (precision - 1)) - 1;
+
+ /* find maximum coefficient value */
+ cmax = 0.0;
+ for(i=0; i<order; i++) {
+ cmax= FFMAX(cmax, fabs(lpc_in[i]));
+ }
+
+ /* if maximum value quantizes to zero, return all zeros */
+ if(cmax * (1 << MAX_LPC_SHIFT) < 1.0) {
+ *shift = 0;
+ memset(lpc_out, 0, sizeof(int32_t) * order);
+ return;
+ }
+
+ /* calculate level shift which scales max coeff to available bits */
+ sh = MAX_LPC_SHIFT;
+ while((cmax * (1 << sh) > qmax) && (sh > 0)) {
+ sh--;
+ }
+
+ /* since negative shift values are unsupported in decoder, scale down
+ coefficients instead */
+ if(sh == 0 && cmax > qmax) {
+ double scale = ((double)qmax) / cmax;
+ for(i=0; i<order; i++) {
+ lpc_in[i] *= scale;
+ }
+ }
+
+ /* output quantized coefficients and level shift */
+ for(i=0; i<order; i++) {
+ lpc_out[i] = (int32_t)(lpc_in[i] * (1 << sh));
+ }
+ *shift = sh;
+}
+
+static int estimate_best_order(double *ref, int max_order)
+{
+ int i, est;
+
+ est = 1;
+ for(i=max_order-1; i>=0; i--) {
+ if(ref[i] > 0.10) {
+ est = i+1;
+ break;
+ }
+ }
+ return est;
+}
- sub->order = 0;
- sub->type = FLAC_SUBFRAME_VERBATIM;
- sub->type_code = sub->type;
+/**
+ * Calculate LPC coefficients for multiple orders
+ */
+static int lpc_calc_coefs(const int32_t *samples, int blocksize, int max_order,
+ int precision, int32_t coefs[][MAX_LPC_ORDER],
+ int *shift)
+{
+ double autoc[MAX_LPC_ORDER+1];
+ double ref[MAX_LPC_ORDER];
+ double lpc[MAX_LPC_ORDER][MAX_LPC_ORDER];
+ int i;
+ int opt_order;
+
+ assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER);
+
+ compute_autocorr(samples, blocksize, max_order+1, autoc);
+
+ compute_lpc_coefs(autoc, max_order, lpc, ref);
+
+ opt_order = estimate_best_order(ref, max_order);
+
+ i = opt_order-1;
+ quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
+ return opt_order;
+}
+
+
+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, int32_t *smp, int n, int order)
+static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
+ int order)
{
int i;
+
+ for(i=0; i<order; i++) {
+ res[i] = smp[i];
+ }
+
+ if(order==0){
+ for(i=order; i<n; i++)
+ res[i]= smp[i];
+ }else if(order==1){
+ for(i=order; i<n; i++)
+ res[i]= smp[i] - smp[i-1];
+ }else if(order==2){
+ for(i=order; i<n; i++)
+ res[i]= smp[i] - 2*smp[i-1] + smp[i-2];
+ }else if(order==3){
+ for(i=order; i<n; i++)
+ res[i]= smp[i] - 3*smp[i-1] + 3*smp[i-2] - smp[i-3];
+ }else{
+ for(i=order; i<n; i++)
+ res[i]= smp[i] - 4*smp[i-1] + 6*smp[i-2] - 4*smp[i-3] + smp[i-4];
+ }
+}
+
+static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
+ int order, const int32_t *coefs, int shift)
+{
+ int i, j;
int32_t pred;
for(i=0; i<order; i++) {
}
for(i=order; i<n; i++) {
pred = 0;
- switch(order) {
- case 0: pred = 0;
- break;
- case 1: pred = smp[i-1];
- break;
- case 2: pred = 2*smp[i-1] - smp[i-2];
- break;
- case 3: pred = 3*smp[i-1] - 3*smp[i-2] + smp[i-3];
- break;
- case 4: pred = 4*smp[i-1] - 6*smp[i-2] + 4*smp[i-3] - smp[i-4];
- break;
+ for(j=0; j<order; j++) {
+ pred += coefs[j] * smp[i-j-1];
}
- res[i] = smp[i] - pred;
+ res[i] = smp[i] - (pred >> shift);
}
}
-static void encode_residual(FlacEncodeContext *s, int ch)
+static int encode_residual(FlacEncodeContext *ctx, int ch)
{
+ int i, n;
+ int min_order, max_order, opt_order, precision;
+ int min_porder, max_porder;
FlacFrame *frame;
FlacSubframe *sub;
- int32_t *res;
- int32_t *smp;
- int n;
+ int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
+ int shift[MAX_LPC_ORDER];
+ int32_t *res, *smp;
- frame = &s->frame;
+ frame = &ctx->frame;
sub = &frame->subframes[ch];
res = sub->residual;
smp = sub->samples;
n = frame->blocksize;
- sub->order = 2;
- sub->type = FLAC_SUBFRAME_FIXED;
- sub->type_code = sub->type | sub->order;
- encode_residual_fixed(res, smp, n, sub->order);
+ /* 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;
+ }
+
+ /* VERBATIM */
+ if(n < 5) {
+ sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
+ encode_residual_verbatim(res, smp, n);
+ return sub->obits * n;
+ }
+
+ 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;
+
+ /* FIXED */
+ if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
+ uint32_t bits[MAX_FIXED_ORDER+1];
+ if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
+ opt_order = 0;
+ 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);
+ 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);
+ }
+ return bits[sub->order];
+ }
+
+ /* LPC */
+ sub->order = lpc_calc_coefs(smp, n, max_order, precision, coefs, shift);
+ 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++) {
+ sub->coefs[i] = coefs[sub->order-1][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);
}
-static void
-put_sbits(PutBitContext *pb, int bits, int32_t val)
+static int encode_residual_v(FlacEncodeContext *ctx, int ch)
{
- uint32_t uval;
+ int i, n;
+ FlacFrame *frame;
+ FlacSubframe *sub;
+ int32_t *res, *smp;
- assert(bits >= 0 && bits <= 31);
- uval = (val < 0) ? (1UL << bits) + val : val;
- put_bits(pb, bits, uval);
+ 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;
+ }
+
+ /* VERBATIM */
+ sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
+ encode_residual_verbatim(res, smp, n);
+ return sub->obits * n;
+}
+
+static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
+{
+ int i, best;
+ int32_t lt, rt;
+ uint64_t sum[4];
+ uint64_t score[4];
+ int k;
+
+ /* calculate sum of 2nd order residual for each channel */
+ sum[0] = sum[1] = sum[2] = sum[3] = 0;
+ for(i=2; i<n; i++) {
+ lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
+ rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
+ sum[2] += ABS((lt + rt) >> 1);
+ sum[3] += ABS(lt - rt);
+ sum[0] += ABS(lt);
+ sum[1] += ABS(rt);
+ }
+ /* estimate bit counts */
+ for(i=0; i<4; i++) {
+ k = find_optimal_param(2*sum[i], n);
+ sum[i] = rice_encode_count(2*sum[i], n, k);
+ }
+
+ /* calculate score for each mode */
+ score[0] = sum[0] + sum[1];
+ score[1] = sum[0] + sum[3];
+ score[2] = sum[1] + sum[3];
+ score[3] = sum[2] + sum[3];
+
+ /* return mode with lowest score */
+ best = 0;
+ for(i=1; i<4; i++) {
+ if(score[i] < score[best]) {
+ best = i;
+ }
+ }
+ if(best == 0) {
+ return FLAC_CHMODE_LEFT_RIGHT;
+ } else if(best == 1) {
+ return FLAC_CHMODE_LEFT_SIDE;
+ } else if(best == 2) {
+ return FLAC_CHMODE_RIGHT_SIDE;
+ } else {
+ return FLAC_CHMODE_MID_SIDE;
+ }
+}
+
+/**
+ * Perform stereo channel decorrelation
+ */
+static void channel_decorrelation(FlacEncodeContext *ctx)
+{
+ FlacFrame *frame;
+ int32_t *left, *right;
+ int i, n;
+
+ frame = &ctx->frame;
+ n = frame->blocksize;
+ left = frame->subframes[0].samples;
+ right = frame->subframes[1].samples;
+
+ if(ctx->channels != 2) {
+ frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
+ return;
+ }
+
+ frame->ch_mode = estimate_stereo_mode(left, right, n);
+
+ /* perform decorrelation and adjust bits-per-sample */
+ if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) {
+ return;
+ }
+ if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
+ int32_t tmp;
+ for(i=0; i<n; i++) {
+ tmp = left[i];
+ left[i] = (tmp + right[i]) >> 1;
+ right[i] = tmp - right[i];
+ }
+ frame->subframes[1].obits++;
+ } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
+ for(i=0; i<n; i++) {
+ right[i] = left[i] - right[i];
+ }
+ frame->subframes[1].obits++;
+ } else {
+ for(i=0; i<n; i++) {
+ left[i] -= right[i];
+ }
+ frame->subframes[0].obits++;
+ }
}
-static void
-write_utf8(PutBitContext *pb, uint32_t val)
+static void put_sbits(PutBitContext *pb, int bits, int32_t val)
{
- int i, bytes, mask, shift;
+ assert(bits >= 0 && bits <= 31);
- bytes = 1;
- if(val >= 0x80) bytes++;
- if(val >= 0x800) bytes++;
- if(val >= 0x10000) bytes++;
- if(val >= 0x200000) bytes++;
- if(val >= 0x4000000) bytes++;
+ put_bits(pb, bits, val & ((1<<bits)-1));
+}
+
+static void write_utf8(PutBitContext *pb, uint32_t val)
+{
+ int bytes, shift;
- if(bytes == 1) {
+ if(val < 0x80){
put_bits(pb, 8, val);
return;
}
+ bytes= (av_log2(val)+4) / 5;
shift = (bytes - 1) * 6;
- mask = 0x80 + ((1 << 7) - (1 << (8 - bytes)));
- put_bits(pb, 8, mask | (val >> shift));
- for(i=0; i<bytes-1; i++) {
+ put_bits(pb, 8, (256 - (256>>bytes)) | (val >> shift));
+ while(shift >= 6){
shift -= 6;
put_bits(pb, 8, 0x80 | ((val >> shift) & 0x3F));
}
}
-static void
-output_frame_header(FlacEncodeContext *s)
+static void output_frame_header(FlacEncodeContext *s)
{
FlacFrame *frame;
int crc;
put_bits(&s->pb, 3, 4); /* bits-per-sample code */
put_bits(&s->pb, 1, 0);
write_utf8(&s->pb, s->frame_count);
- if(frame->bs_code[1] > 0) {
- if(frame->bs_code[1] < 256) {
- put_bits(&s->pb, 8, frame->bs_code[1]);
- } else {
- put_bits(&s->pb, 16, frame->bs_code[1]);
- }
+ if(frame->bs_code[0] == 6) {
+ put_bits(&s->pb, 8, frame->bs_code[1]);
+ } else if(frame->bs_code[0] == 7) {
+ put_bits(&s->pb, 16, frame->bs_code[1]);
}
- if(s->sr_code[1] > 0) {
- if(s->sr_code[1] < 256) {
- put_bits(&s->pb, 8, s->sr_code[1]);
- } else {
- put_bits(&s->pb, 16, s->sr_code[1]);
- }
+ if(s->sr_code[0] == 12) {
+ put_bits(&s->pb, 8, s->sr_code[1]);
+ } else if(s->sr_code[0] > 12) {
+ put_bits(&s->pb, 16, s->sr_code[1]);
}
flush_put_bits(&s->pb);
crc = av_crc(av_crc07, 0, s->pb.buf, put_bits_count(&s->pb)>>3);
put_bits(&s->pb, 8, crc);
}
+static void output_subframe_constant(FlacEncodeContext *s, int ch)
+{
+ 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;
}
}
-static void
-output_residual(FlacEncodeContext *ctx, int ch)
+static void output_residual(FlacEncodeContext *ctx, int ch)
{
- int i, j, p;
+ 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 = 0;
- psize = frame->blocksize;
- //porder = sub->rc.porder;
- //psize = frame->blocksize >> porder;
+ 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<(1 << porder); p++) {
- //k = sub->rc.params[p];
- k = 9;
+ 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<frame->blocksize; i++, j++) {
- set_sr_golomb_flac(&ctx->pb, sub->residual[j], k, INT32_MAX, 0);
+ 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)
+static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
{
int i;
FlacFrame *frame;
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;
put_bits(&s->pb, 1, 0); /* no wasted bits */
/* subframe */
- if(sub->type == FLAC_SUBFRAME_VERBATIM) {
+ if(sub->type == FLAC_SUBFRAME_CONSTANT) {
+ output_subframe_constant(s, ch);
+ } else if(sub->type == FLAC_SUBFRAME_VERBATIM) {
output_subframe_verbatim(s, ch);
- } else {
+ } else if(sub->type == FLAC_SUBFRAME_FIXED) {
output_subframe_fixed(s, ch);
+ } else if(sub->type == FLAC_SUBFRAME_LPC) {
+ output_subframe_lpc(s, ch);
}
}
}
s = avctx->priv_data;
s->blocksize = avctx->frame_size;
- if(init_frame(s)) {
- return 0;
- }
+ init_frame(s);
copy_samples(s, samples);
+ channel_decorrelation(s);
+
for(ch=0; ch<s->channels; ch++) {
encode_residual(s, ch);
}
if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
/* frame too large. use verbatim mode */
for(ch=0; ch<s->channels; ch++) {
- encode_residual_verbatim(s, ch);
+ encode_residual_v(s, ch);
}
init_put_bits(&s->pb, frame, buf_size);
output_frame_header(s);
static int flac_encode_close(AVCodecContext *avctx)
{
+ av_freep(&avctx->extradata);
+ avctx->extradata_size = 0;
av_freep(&avctx->coded_frame);
return 0;
}
projects / ffmpeg / blobdiff