*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*
*/
#undef V_DEBUG
+//#define V_DEBUG
+//#define AV_DEBUG(...) av_log(NULL, AV_LOG_INFO, __VA_ARGS__)
#include <math.h>
/**
* reads 0-32 bits when using the ALT_BITSTREAM_READER_LE bitstream reader
*/
-unsigned int get_bits_long_le(GetBitContext *s, int n){
+static unsigned int get_bits_long_le(GetBitContext *s, int n){
if(n<=17) return get_bits(s, n);
else{
int ret= get_bits(s, 16);
#define ilog(i) av_log2(2*(i))
+#define BARK(x) \
+ (13.1f*atan(0.00074f*(x))+2.24f*atan(1.85e-8f*(x)*(x))+1e-4f*(x))
+
static unsigned int nth_root(unsigned int x, unsigned int n) { // x^(1/n)
unsigned int ret=0, i, j;
av_freep(&vc->residues);
av_freep(&vc->modes);
- ff_mdct_end(&vc->mdct0);
- ff_mdct_end(&vc->mdct1);
+ ff_mdct_end(&vc->mdct[0]);
+ ff_mdct_end(&vc->mdct[1]);
for(i=0;i<vc->codebook_count;++i) {
av_free(vc->codebooks[i].codevectors);
av_freep(&vc->codebooks);
for(i=0;i<vc->floor_count;++i) {
- av_free(vc->floors[i].x_list);
- av_free(vc->floors[i].x_list_order);
- av_free(vc->floors[i].low_neighbour);
- av_free(vc->floors[i].high_neighbour);
+ if(vc->floors[i].floor_type==0) {
+ av_free(vc->floors[i].data.t0.map[0]);
+ av_free(vc->floors[i].data.t0.map[1]);
+ av_free(vc->floors[i].data.t0.book_list);
+ av_free(vc->floors[i].data.t0.lsp);
+ }
+ else {
+ av_free(vc->floors[i].data.t1.x_list);
+ av_free(vc->floors[i].data.t1.x_list_order);
+ av_free(vc->floors[i].data.t1.low_neighbour);
+ av_free(vc->floors[i].data.t1.high_neighbour);
+ }
}
av_freep(&vc->floors);
av_free(vc->mappings[i].mux);
}
av_freep(&vc->mappings);
+
+ if(vc->exp_bias){
+ av_freep(&vc->win[0]);
+ av_freep(&vc->win[1]);
+ }
}
// Parse setup header -------------------------------------------------
return 0;
}
-// Process floors part - only floor type 1 is supported
+// Process floors part
+static uint_fast8_t vorbis_floor0_decode(vorbis_context *vc,
+ vorbis_floor_data *vfu, float *vec);
+static void create_map( vorbis_context * vc, uint_fast8_t floor_number );
+static uint_fast8_t vorbis_floor1_decode(vorbis_context *vc,
+ vorbis_floor_data *vfu, float *vec);
static int vorbis_parse_setup_hdr_floors(vorbis_context *vc) {
GetBitContext *gb=&vc->gb;
uint_fast16_t i,j,k;
uint_fast8_t rangebits;
uint_fast16_t floor1_values=2;
- floor_setup->partitions=get_bits(gb, 5);
+ floor_setup->decode=vorbis_floor1_decode;
- AV_DEBUG(" %d.floor: %d partitions \n", i, floor_setup->partitions);
+ floor_setup->data.t1.partitions=get_bits(gb, 5);
- for(j=0;j<floor_setup->partitions;++j) {
- floor_setup->partition_class[j]=get_bits(gb, 4);
- if (floor_setup->partition_class[j]>maximum_class) maximum_class=floor_setup->partition_class[j];
+ AV_DEBUG(" %d.floor: %d partitions \n", i, floor_setup->data.t1.partitions);
- AV_DEBUG(" %d. floor %d partition class %d \n", i, j, floor_setup->partition_class[j]);
+ for(j=0;j<floor_setup->data.t1.partitions;++j) {
+ floor_setup->data.t1.partition_class[j]=get_bits(gb, 4);
+ if (floor_setup->data.t1.partition_class[j]>maximum_class) maximum_class=floor_setup->data.t1.partition_class[j];
+
+ AV_DEBUG(" %d. floor %d partition class %d \n", i, j, floor_setup->data.t1.partition_class[j]);
}
AV_DEBUG(" maximum class %d \n", maximum_class);
- floor_setup->maximum_class=maximum_class;
+ floor_setup->data.t1.maximum_class=maximum_class;
for(j=0;j<=maximum_class;++j) {
- floor_setup->class_dimensions[j]=get_bits(gb, 3)+1;
- floor_setup->class_subclasses[j]=get_bits(gb, 2);
+ floor_setup->data.t1.class_dimensions[j]=get_bits(gb, 3)+1;
+ floor_setup->data.t1.class_subclasses[j]=get_bits(gb, 2);
- AV_DEBUG(" %d floor %d class dim: %d subclasses %d \n", i, j, floor_setup->class_dimensions[j], floor_setup->class_subclasses[j]);
+ AV_DEBUG(" %d floor %d class dim: %d subclasses %d \n", i, j, floor_setup->data.t1.class_dimensions[j], floor_setup->data.t1.class_subclasses[j]);
- if (floor_setup->class_subclasses[j]) {
- floor_setup->class_masterbook[j]=get_bits(gb, 8);
+ if (floor_setup->data.t1.class_subclasses[j]) {
+ floor_setup->data.t1.class_masterbook[j]=get_bits(gb, 8);
- AV_DEBUG(" masterbook: %d \n", floor_setup->class_masterbook[j]);
+ AV_DEBUG(" masterbook: %d \n", floor_setup->data.t1.class_masterbook[j]);
}
- for(k=0;k<(1<<floor_setup->class_subclasses[j]);++k) {
- floor_setup->subclass_books[j][k]=get_bits(gb, 8)-1;
+ for(k=0;k<(1<<floor_setup->data.t1.class_subclasses[j]);++k) {
+ floor_setup->data.t1.subclass_books[j][k]=(int16_t)get_bits(gb, 8)-1;
- AV_DEBUG(" book %d. : %d \n", k, floor_setup->subclass_books[j][k]);
+ AV_DEBUG(" book %d. : %d \n", k, floor_setup->data.t1.subclass_books[j][k]);
}
}
- floor_setup->multiplier=get_bits(gb, 2)+1;
- floor_setup->x_list_dim=2;
+ floor_setup->data.t1.multiplier=get_bits(gb, 2)+1;
+ floor_setup->data.t1.x_list_dim=2;
- for(j=0;j<floor_setup->partitions;++j) {
- floor_setup->x_list_dim+=floor_setup->class_dimensions[floor_setup->partition_class[j]];
+ for(j=0;j<floor_setup->data.t1.partitions;++j) {
+ floor_setup->data.t1.x_list_dim+=floor_setup->data.t1.class_dimensions[floor_setup->data.t1.partition_class[j]];
}
- floor_setup->x_list=(uint_fast16_t *)av_mallocz(floor_setup->x_list_dim * sizeof(uint_fast16_t));
- floor_setup->x_list_order=(uint_fast16_t *)av_mallocz(floor_setup->x_list_dim * sizeof(uint_fast16_t));
- floor_setup->low_neighbour=(uint_fast16_t *)av_mallocz(floor_setup->x_list_dim * sizeof(uint_fast16_t));
- floor_setup->high_neighbour=(uint_fast16_t *)av_mallocz(floor_setup->x_list_dim * sizeof(uint_fast16_t));
+ floor_setup->data.t1.x_list=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
+ floor_setup->data.t1.x_list_order=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
+ floor_setup->data.t1.low_neighbour=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
+ floor_setup->data.t1.high_neighbour=(uint_fast16_t *)av_mallocz(floor_setup->data.t1.x_list_dim * sizeof(uint_fast16_t));
rangebits=get_bits(gb, 4);
- floor_setup->x_list[0] = 0;
- floor_setup->x_list[1] = (1<<rangebits);
+ floor_setup->data.t1.x_list[0] = 0;
+ floor_setup->data.t1.x_list[1] = (1<<rangebits);
- for(j=0;j<floor_setup->partitions;++j) {
- for(k=0;k<floor_setup->class_dimensions[floor_setup->partition_class[j]];++k,++floor1_values) {
- floor_setup->x_list[floor1_values]=get_bits(gb, rangebits);
+ for(j=0;j<floor_setup->data.t1.partitions;++j) {
+ for(k=0;k<floor_setup->data.t1.class_dimensions[floor_setup->data.t1.partition_class[j]];++k,++floor1_values) {
+ floor_setup->data.t1.x_list[floor1_values]=get_bits(gb, rangebits);
- AV_DEBUG(" %d. floor1 Y coord. %d \n", floor1_values, floor_setup->x_list[floor1_values]);
+ AV_DEBUG(" %d. floor1 Y coord. %d \n", floor1_values, floor_setup->data.t1.x_list[floor1_values]);
}
}
// Precalculate order of x coordinates - needed for decode
- for(k=0;k<floor_setup->x_list_dim;++k) {
- floor_setup->x_list_order[k]=k;
+ for(k=0;k<floor_setup->data.t1.x_list_dim;++k) {
+ floor_setup->data.t1.x_list_order[k]=k;
}
- for(k=0;k<floor_setup->x_list_dim-1;++k) { // FIXME optimize sorting ?
- for(j=k+1;j<floor_setup->x_list_dim;++j) {
- if(floor_setup->x_list[floor_setup->x_list_order[k]]>floor_setup->x_list[floor_setup->x_list_order[j]]) {
- uint_fast16_t tmp=floor_setup->x_list_order[k];
- floor_setup->x_list_order[k]=floor_setup->x_list_order[j];
- floor_setup->x_list_order[j]=tmp;
+ for(k=0;k<floor_setup->data.t1.x_list_dim-1;++k) { // FIXME optimize sorting ?
+ for(j=k+1;j<floor_setup->data.t1.x_list_dim;++j) {
+ if(floor_setup->data.t1.x_list[floor_setup->data.t1.x_list_order[k]]>floor_setup->data.t1.x_list[floor_setup->data.t1.x_list_order[j]]) {
+ uint_fast16_t tmp=floor_setup->data.t1.x_list_order[k];
+ floor_setup->data.t1.x_list_order[k]=floor_setup->data.t1.x_list_order[j];
+ floor_setup->data.t1.x_list_order[j]=tmp;
}
}
}
// Precalculate low and high neighbours
- for(k=2;k<floor_setup->x_list_dim;++k) {
- floor_setup->low_neighbour[k]=0;
- floor_setup->high_neighbour[k]=1; // correct according to SPEC requirements
+ for(k=2;k<floor_setup->data.t1.x_list_dim;++k) {
+ floor_setup->data.t1.low_neighbour[k]=0;
+ floor_setup->data.t1.high_neighbour[k]=1; // correct according to SPEC requirements
for (j=0;j<k;++j) {
- if ((floor_setup->x_list[j]<floor_setup->x_list[k]) &&
- (floor_setup->x_list[j]>floor_setup->x_list[floor_setup->low_neighbour[k]])) {
- floor_setup->low_neighbour[k]=j;
+ if ((floor_setup->data.t1.x_list[j]<floor_setup->data.t1.x_list[k]) &&
+ (floor_setup->data.t1.x_list[j]>floor_setup->data.t1.x_list[floor_setup->data.t1.low_neighbour[k]])) {
+ floor_setup->data.t1.low_neighbour[k]=j;
}
- if ((floor_setup->x_list[j]>floor_setup->x_list[k]) &&
- (floor_setup->x_list[j]<floor_setup->x_list[floor_setup->high_neighbour[k]])) {
- floor_setup->high_neighbour[k]=j;
+ if ((floor_setup->data.t1.x_list[j]>floor_setup->data.t1.x_list[k]) &&
+ (floor_setup->data.t1.x_list[j]<floor_setup->data.t1.x_list[floor_setup->data.t1.high_neighbour[k]])) {
+ floor_setup->data.t1.high_neighbour[k]=j;
}
}
}
}
+ else if(floor_setup->floor_type==0) {
+ uint_fast8_t max_codebook_dim=0;
+
+ floor_setup->decode=vorbis_floor0_decode;
+
+ floor_setup->data.t0.order=get_bits(gb, 8);
+ floor_setup->data.t0.rate=get_bits(gb, 16);
+ floor_setup->data.t0.bark_map_size=get_bits(gb, 16);
+ floor_setup->data.t0.amplitude_bits=get_bits(gb, 6);
+ /* zero would result in a div by zero later *
+ * 2^0 - 1 == 0 */
+ if (floor_setup->data.t0.amplitude_bits == 0) {
+ av_log(vc->avccontext, AV_LOG_ERROR,
+ "Floor 0 amplitude bits is 0.\n");
+ return 1;
+ }
+ floor_setup->data.t0.amplitude_offset=get_bits(gb, 8);
+ floor_setup->data.t0.num_books=get_bits(gb, 4)+1;
+
+ /* allocate mem for booklist */
+ floor_setup->data.t0.book_list=
+ av_malloc(floor_setup->data.t0.num_books);
+ if(!floor_setup->data.t0.book_list) { return 1; }
+ /* read book indexes */
+ {
+ int idx;
+ uint_fast8_t book_idx;
+ for (idx=0;idx<floor_setup->data.t0.num_books;++idx) {
+ book_idx=get_bits(gb, 8);
+ floor_setup->data.t0.book_list[idx]=book_idx;
+ if (vc->codebooks[book_idx].dimensions > max_codebook_dim)
+ max_codebook_dim=vc->codebooks[book_idx].dimensions;
+
+ if (floor_setup->data.t0.book_list[idx]>vc->codebook_count)
+ return 1;
+ }
+ }
+
+ create_map( vc, i );
+
+ /* allocate mem for lsp coefficients */
+ {
+ /* codebook dim is for padding if codebook dim doesn't *
+ * divide order+1 then we need to read more data */
+ floor_setup->data.t0.lsp=
+ av_malloc((floor_setup->data.t0.order+1 + max_codebook_dim)
+ * sizeof(float));
+ if(!floor_setup->data.t0.lsp) { return 1; }
+ }
+
+#ifdef V_DEBUG /* debug output parsed headers */
+ AV_DEBUG("floor0 order: %u\n", floor_setup->data.t0.order);
+ AV_DEBUG("floor0 rate: %u\n", floor_setup->data.t0.rate);
+ AV_DEBUG("floor0 bark map size: %u\n",
+ floor_setup->data.t0.bark_map_size);
+ AV_DEBUG("floor0 amplitude bits: %u\n",
+ floor_setup->data.t0.amplitude_bits);
+ AV_DEBUG("floor0 amplitude offset: %u\n",
+ floor_setup->data.t0.amplitude_offset);
+ AV_DEBUG("floor0 number of books: %u\n",
+ floor_setup->data.t0.num_books);
+ AV_DEBUG("floor0 book list pointer: %p\n",
+ floor_setup->data.t0.book_list);
+ {
+ int idx;
+ for (idx=0;idx<floor_setup->data.t0.num_books;++idx) {
+ AV_DEBUG( " Book %d: %u\n",
+ idx+1,
+ floor_setup->data.t0.book_list[idx] );
+ }
+ }
+#endif
+ }
else {
- av_log(vc->avccontext, AV_LOG_ERROR, "Only floor type 1 supported. \n");
+ av_log(vc->avccontext, AV_LOG_ERROR, "Invalid floor type!\n");
return 1;
}
}
// Process modes part
+static void create_map( vorbis_context * vc, uint_fast8_t floor_number )
+{
+ vorbis_floor * floors=vc->floors;
+ vorbis_floor0 * vf;
+ int idx;
+ int_fast8_t blockflag;
+ int_fast32_t * map;
+ int_fast32_t n; //TODO: could theoretically be smaller?
+
+ for (blockflag=0;blockflag<2;++blockflag)
+ {
+ n=vc->blocksize[blockflag]/2;
+ floors[floor_number].data.t0.map[blockflag]=
+ av_malloc((n+1) * sizeof(int_fast32_t)); // n+sentinel
+
+ map=floors[floor_number].data.t0.map[blockflag];
+ vf=&floors[floor_number].data.t0;
+
+ for (idx=0; idx<n;++idx) {
+ map[idx]=floor( BARK((vf->rate*idx)/(2.0f*n)) *
+ ((vf->bark_map_size)/
+ BARK(vf->rate/2.0f )) );
+ if (vf->bark_map_size-1 < map[idx]) {
+ map[idx]=vf->bark_map_size-1;
+ }
+ }
+ map[n]=-1;
+ vf->map_size[blockflag]=n;
+ }
+
+# ifdef V_DEBUG
+ for(idx=0;idx<=n;++idx) {
+ AV_DEBUG("floor0 map: map at pos %d is %d\n",
+ idx, map[idx]);
+ }
+# endif
+}
+
static int vorbis_parse_setup_hdr_modes(vorbis_context *vc) {
GetBitContext *gb=&vc->gb;
uint_fast8_t i;
vc->bitrate_minimum=get_bits_long_le(gb, 32);
bl0=get_bits(gb, 4);
bl1=get_bits(gb, 4);
- vc->blocksize_0=(1<<bl0);
- vc->blocksize_1=(1<<bl1);
- if (bl0>13 || bl0<6 || bl1>13 || bl1<6) {
+ vc->blocksize[0]=(1<<bl0);
+ vc->blocksize[1]=(1<<bl1);
+ if (bl0>13 || bl0<6 || bl1>13 || bl1<6 || bl1<bl0) {
av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis id header packet corrupt (illegal blocksize). \n");
return 3;
}
- vc->swin=vwin[bl0-6];
- vc->lwin=vwin[bl1-6];
+ // output format int16
+ if (vc->blocksize[1]/2 * vc->audio_channels * 2 >
+ AVCODEC_MAX_AUDIO_FRAME_SIZE) {
+ av_log(vc->avccontext, AV_LOG_ERROR, "Vorbis channel count makes "
+ "output packets too large.\n");
+ return 4;
+ }
+ vc->win[0]=vwin[bl0-6];
+ vc->win[1]=vwin[bl1-6];
+
+ if(vc->exp_bias){
+ int i, j;
+ for(j=0; j<2; j++){
+ float *win = av_malloc(vc->blocksize[j]/2 * sizeof(float));
+ for(i=0; i<vc->blocksize[j]/2; i++)
+ win[i] = vc->win[j][i] * (1<<15);
+ vc->win[j] = win;
+ }
+ }
if ((get_bits1(gb)) == 0) {
av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis id header packet corrupt (framing flag not set). \n");
return 2;
}
- vc->channel_residues=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->channel_floors=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->saved=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->ret=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->buf=(float *)av_malloc(vc->blocksize_1 * sizeof(float));
- vc->buf_tmp=(float *)av_malloc(vc->blocksize_1 * sizeof(float));
+ vc->channel_residues=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->channel_floors=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->saved=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->ret=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->buf=(float *)av_malloc(vc->blocksize[1] * sizeof(float));
+ vc->buf_tmp=(float *)av_malloc(vc->blocksize[1] * sizeof(float));
vc->saved_start=0;
- ff_mdct_init(&vc->mdct0, bl0, 1);
- ff_mdct_init(&vc->mdct1, bl1, 1);
+ ff_mdct_init(&vc->mdct[0], bl0, 1);
+ ff_mdct_init(&vc->mdct[1], bl1, 1);
AV_DEBUG(" vorbis version %d \n audio_channels %d \n audio_samplerate %d \n bitrate_max %d \n bitrate_nom %d \n bitrate_min %d \n blk_0 %d blk_1 %d \n ",
- vc->version, vc->audio_channels, vc->audio_samplerate, vc->bitrate_maximum, vc->bitrate_nominal, vc->bitrate_minimum, vc->blocksize_0, vc->blocksize_1);
+ vc->version, vc->audio_channels, vc->audio_samplerate, vc->bitrate_maximum, vc->bitrate_nominal, vc->bitrate_minimum, vc->blocksize[0], vc->blocksize[1]);
/*
- BLK=vc->blocksize_0;
+ BLK=vc->blocksize[0];
for(i=0;i<BLK/2;++i) {
- vc->swin[i]=sin(0.5*3.14159265358*(sin(((float)i+0.5)/(float)BLK*3.14159265358))*(sin(((float)i+0.5)/(float)BLK*3.14159265358)));
+ vc->win[0][i]=sin(0.5*3.14159265358*(sin(((float)i+0.5)/(float)BLK*3.14159265358))*(sin(((float)i+0.5)/(float)BLK*3.14159265358)));
}
*/
int i, j, hdr_type;
vc->avccontext = avccontext;
+ dsputil_init(&vc->dsp, avccontext);
+
+ if(vc->dsp.float_to_int16 == ff_float_to_int16_c) {
+ vc->add_bias = 385;
+ vc->exp_bias = 0;
+ } else {
+ vc->add_bias = 0;
+ vc->exp_bias = 15<<23;
+ }
if (!headers_len) {
av_log(avccontext, AV_LOG_ERROR, "Extradata corrupt.\n");
// Decode audiopackets -------------------------------------------------
-// Read and decode floor (type 1 only)
+// Read and decode floor
+
+static uint_fast8_t vorbis_floor0_decode(vorbis_context *vc,
+ vorbis_floor_data *vfu, float *vec) {
+ vorbis_floor0 * vf=&vfu->t0;
+ float * lsp=vf->lsp;
+ uint_fast32_t amplitude;
+ uint_fast32_t book_idx;
+ uint_fast8_t blockflag=vc->modes[vc->mode_number].blockflag;
+
+ amplitude=get_bits(&vc->gb, vf->amplitude_bits);
+ if (amplitude>0) {
+ float last = 0;
+ uint_fast16_t lsp_len = 0;
+ uint_fast16_t idx;
+ vorbis_codebook codebook;
+
+ book_idx=get_bits(&vc->gb, ilog(vf->num_books));
+ if ( book_idx >= vf->num_books ) {
+ av_log( vc->avccontext, AV_LOG_ERROR,
+ "floor0 dec: booknumber too high!\n" );
+ //FIXME: look above
+ }
+ AV_DEBUG( "floor0 dec: booknumber: %u\n", book_idx );
+ codebook=vc->codebooks[vf->book_list[book_idx]];
+
+ while (lsp_len<vf->order) {
+ int vec_off;
+
+ AV_DEBUG( "floor0 dec: book dimension: %d\n", codebook.dimensions );
+ AV_DEBUG( "floor0 dec: maximum depth: %d\n", codebook.maxdepth );
+ /* read temp vector */
+ vec_off=get_vlc2(&vc->gb,
+ codebook.vlc.table,
+ codebook.nb_bits,
+ codebook.maxdepth ) *
+ codebook.dimensions;
+ AV_DEBUG( "floor0 dec: vector offset: %d\n", vec_off );
+ /* copy each vector component and add last to it */
+ for (idx=0; idx<codebook.dimensions; ++idx) {
+ lsp[lsp_len+idx]=codebook.codevectors[vec_off+idx]+last;
+ }
+ last=lsp[lsp_len+idx-1]; /* set last to last vector component */
+
+ lsp_len += codebook.dimensions;
+ }
+#ifdef V_DEBUG
+ /* DEBUG: output lsp coeffs */
+ {
+ int idx;
+ for ( idx = 0; idx < lsp_len; ++idx )
+ AV_DEBUG("floor0 dec: coeff at %d is %f\n", idx, lsp[idx] );
+ }
+#endif
+
+ /* synthesize floor output vector */
+ {
+ int i;
+ int order=vf->order;
+ float wstep=M_PI/vf->bark_map_size;
+
+ for(i=0;i<order;i++) { lsp[i]=2.0f*cos(lsp[i]); }
+
+ AV_DEBUG("floor0 synth: map_size=%d; m=%d; wstep=%f\n",
+ vf->map_size, order, wstep);
+
+ i=0;
+ while(i<vf->map_size[blockflag]) {
+ int j, iter_cond=vf->map[blockflag][i];
+ float p=0.5f;
+ float q=0.5f;
+ float two_cos_w=2.0f*cos(wstep*iter_cond); // needed all times
+
+ /* similar part for the q and p products */
+ for(j=0;j<order;j+=2) {
+ q *= lsp[j] -two_cos_w;
+ p *= lsp[j+1]-two_cos_w;
+ }
+ if(j==order) { // even order
+ p *= p*(2.0f-two_cos_w);
+ q *= q*(2.0f+two_cos_w);
+ }
+ else { // odd order
+ q *= two_cos_w-lsp[j]; // one more time for q
+
+ /* final step and square */
+ p *= p*(4.f-two_cos_w*two_cos_w);
+ q *= q;
+ }
+
+ /* calculate linear floor value */
+ {
+ q=exp( (
+ ( (amplitude*vf->amplitude_offset)/
+ (((1<<vf->amplitude_bits)-1) * sqrt(p+q)) )
+ - vf->amplitude_offset ) * .11512925f
+ );
+ }
+
+ /* fill vector */
+ do { vec[i]=q; ++i; }while(vf->map[blockflag][i]==iter_cond);
+ }
+ }
+ }
+ else {
+ /* this channel is unused */
+ return 1;
+ }
+
+ AV_DEBUG(" Floor0 decoded\n");
-static uint_fast8_t vorbis_floor1_decode(vorbis_context *vc, vorbis_floor *vf, float *vec) {
+ return 0;
+}
+static uint_fast8_t vorbis_floor1_decode(vorbis_context *vc, vorbis_floor_data *vfu, float *vec) {
+ vorbis_floor1 * vf=&vfu->t1;
GetBitContext *gb=&vc->gb;
uint_fast16_t range_v[4]={ 256, 128, 86, 64 };
uint_fast16_t range=range_v[vf->multiplier-1];
adx=vf->x_list[high_neigh_offs]-vf->x_list[low_neigh_offs];
ady= ABS(dy);
err=ady*(vf->x_list[i]-vf->x_list[low_neigh_offs]);
- off=err/adx;
+ off=(int16_t)err/(int16_t)adx;
if (dy<0) {
predicted=floor1_Y_final[low_neigh_offs]-off;
} else {
dy=hy-ly;
adx=hx-lx;
ady= (dy<0) ? -dy:dy;//ABS(dy);
- base=dy/adx;
+ base=(int16_t)dy/(int16_t)adx;
AV_DEBUG(" dy %d adx %d base %d = %d \n", dy, adx, base, dy/adx);
voffset=vr->begin;
for(partition_count=0;partition_count<ptns_to_read;) { // SPEC error
if (!pass) {
+ uint_fast32_t inverse_class = inverse[vr->classifications];
for(j_times_ptns_to_read=0, j=0;j<ch_used;++j) {
if (!do_not_decode[j]) {
uint_fast32_t temp=get_vlc2(gb, vc->codebooks[vr->classbook].vlc.table,
AV_DEBUG("Classword: %d \n", temp);
- assert(vr->classifications > 1 && vr->classifications<256 && temp<=65536); //needed for inverse[]
+ assert(vr->classifications > 1 && temp<=65536); //needed for inverse[]
for(i=0;i<c_p_c;++i) {
uint_fast32_t temp2;
- temp2=(((uint_fast64_t)temp) * inverse[vr->classifications])>>32;
- classifs[j_times_ptns_to_read+partition_count+c_p_c-1-i]=temp-temp2*vr->classifications;
+ temp2=(((uint_fast64_t)temp) * inverse_class)>>32;
+ if (partition_count+c_p_c-1-i < ptns_to_read) {
+ classifs[j_times_ptns_to_read+partition_count+c_p_c-1-i]=temp-temp2*vr->classifications;
+ }
temp=temp2;
}
}
if (vqbook>=0) {
uint_fast16_t coffs;
- uint_fast16_t step=vr->partition_size/vc->codebooks[vqbook].dimensions;
+ unsigned dim= vc->codebooks[vqbook].dimensions; // not uint_fast8_t: 64bit is slower here on amd64
+ uint_fast16_t step= dim==1 ? vr->partition_size
+ : FASTDIV(vr->partition_size, dim);
vorbis_codebook codebook= vc->codebooks[vqbook];
if (vr->type==0) {
voffs=voffset+j*vlen;
for(k=0;k<step;++k) {
- coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * codebook.dimensions;
- for(l=0;l<codebook.dimensions;++l) {
+ coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
+ for(l=0;l<dim;++l) {
vec[voffs+k+l*step]+=codebook.codevectors[coffs+l]; // FPMATH
}
}
else if (vr->type==1) {
voffs=voffset+j*vlen;
for(k=0;k<step;++k) {
- coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * codebook.dimensions;
- for(l=0;l<codebook.dimensions;++l, ++voffs) {
+ coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
+ for(l=0;l<dim;++l, ++voffs) {
vec[voffs]+=codebook.codevectors[coffs+l]; // FPMATH
AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d \n", pass, voffs, vec[voffs], codebook.codevectors[coffs+l], coffs);
}
}
}
- else if (vr->type==2 && ch==2 && (voffset&1)==0 && (codebook.dimensions&1)==0) { // most frequent case optimized
+ else if (vr->type==2 && ch==2 && (voffset&1)==0 && (dim&1)==0) { // most frequent case optimized
voffs=voffset>>1;
+ if(dim==2) {
+ for(k=0;k<step;++k) {
+ coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * 2;
+ vec[voffs+k ]+=codebook.codevectors[coffs ]; // FPMATH
+ vec[voffs+k+vlen]+=codebook.codevectors[coffs+1]; // FPMATH
+ }
+ } else
for(k=0;k<step;++k) {
- coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * codebook.dimensions;
- for(l=0;l<codebook.dimensions;l+=2, voffs++) {
+ coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
+ for(l=0;l<dim;l+=2, voffs++) {
vec[voffs ]+=codebook.codevectors[coffs+l ]; // FPMATH
vec[voffs+vlen]+=codebook.codevectors[coffs+l+1]; // FPMATH
voffs=voffset;
for(k=0;k<step;++k) {
- coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * codebook.dimensions;
- for(l=0;l<codebook.dimensions;++l, ++voffs) {
+ coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
+ for(l=0;l<dim;++l, ++voffs) {
vec[voffs/ch+(voffs%ch)*vlen]+=codebook.codevectors[coffs+l]; // FPMATH FIXME use if and counter instead of / and %
AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], codebook.codevectors[coffs+l], coffs, l);
return 0;
}
+void vorbis_inverse_coupling(float *mag, float *ang, int blocksize)
+{
+ int i;
+ for(i=0; i<blocksize; i++)
+ {
+ if (mag[i]>0.0) {
+ if (ang[i]>0.0) {
+ ang[i]=mag[i]-ang[i];
+ } else {
+ float temp=ang[i];
+ ang[i]=mag[i];
+ mag[i]+=temp;
+ }
+ } else {
+ if (ang[i]>0.0) {
+ ang[i]+=mag[i];
+ } else {
+ float temp=ang[i];
+ ang[i]=mag[i];
+ mag[i]-=temp;
+ }
+ }
+ }
+}
+
// Decode the audio packet using the functions above
-#define BIAS 385
static int vorbis_parse_audio_packet(vorbis_context *vc) {
GetBitContext *gb=&vc->gb;
uint_fast8_t res_num=0;
int_fast16_t retlen=0;
uint_fast16_t saved_start=0;
+ float fadd_bias = vc->add_bias;
if (get_bits1(gb)) {
av_log(vc->avccontext, AV_LOG_ERROR, "Not a Vorbis I audio packet.\n");
} else {
mode_number=get_bits(gb, ilog(vc->mode_count-1));
}
+ vc->mode_number=mode_number;
mapping=&vc->mappings[vc->modes[mode_number].mapping];
AV_DEBUG(" Mode number: %d , mapping: %d , blocktype %d \n", mode_number, vc->modes[mode_number].mapping, vc->modes[mode_number].blockflag);
next_window=get_bits1(gb);
}
- blocksize=vc->modes[mode_number].blockflag ? vc->blocksize_1 : vc->blocksize_0;
+ blocksize=vc->blocksize[vc->modes[mode_number].blockflag];
memset(ch_res_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?
memset(ch_floor_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?
-// Decode floor(1)
+// Decode floor
for(i=0;i<vc->audio_channels;++i) {
vorbis_floor *floor;
floor=&vc->floors[mapping->submap_floor[0]];
}
- no_residue[i]=vorbis_floor1_decode(vc, floor, ch_floor_ptr);
+ no_residue[i]=floor->decode(vc, &floor->data, ch_floor_ptr);
ch_floor_ptr+=blocksize/2;
}
mag=vc->channel_residues+res_chan[mapping->magnitude[i]]*blocksize/2;
ang=vc->channel_residues+res_chan[mapping->angle[i]]*blocksize/2;
- for(j=0;j<blocksize/2;++j) {
- float temp;
- if (mag[j]>0.0) {
- if (ang[j]>0.0) {
- ang[j]=mag[j]-ang[j];
- } else {
- temp=ang[j];
- ang[j]=mag[j];
- mag[j]+=temp;
- }
- } else {
- if (ang[j]>0.0) {
- ang[j]+=mag[j];
- } else {
- temp=ang[j];
- ang[j]=mag[j];
- mag[j]-=temp;
- }
- }
- }
+ vc->dsp.vorbis_inverse_coupling(mag, ang, blocksize/2);
}
// Dotproduct
for(j=0, ch_floor_ptr=vc->channel_floors;j<vc->audio_channels;++j,ch_floor_ptr+=blocksize/2) {
ch_res_ptr=vc->channel_residues+res_chan[j]*blocksize/2;
-
- for(i=0;i<blocksize/2;++i) {
- ch_floor_ptr[i]*=ch_res_ptr[i]; //FPMATH
- }
+ vc->dsp.vector_fmul(ch_floor_ptr, ch_res_ptr, blocksize/2);
}
// MDCT, overlap/add, save data for next overlapping FPMATH
for(j=0;j<vc->audio_channels;++j) {
uint_fast8_t step=vc->audio_channels;
uint_fast16_t k;
- float *saved=vc->saved+j*vc->blocksize_1/2;
+ float *saved=vc->saved+j*vc->blocksize[1]/2;
float *ret=vc->ret;
- const float *lwin=vc->lwin;
- const float *swin=vc->swin;
+ const float *lwin=vc->win[1];
+ const float *swin=vc->win[0];
float *buf=vc->buf;
float *buf_tmp=vc->buf_tmp;
saved_start=vc->saved_start;
- ff_imdct_calc(vc->modes[mode_number].blockflag ? &vc->mdct1 : &vc->mdct0, buf, ch_floor_ptr, buf_tmp);
+ vc->mdct[0].fft.imdct_calc(&vc->mdct[vc->modes[mode_number].blockflag], buf, ch_floor_ptr, buf_tmp);
+ //FIXME process channels together, to allow faster simd vector_fmul_add_add?
if (vc->modes[mode_number].blockflag) {
// -- overlap/add
if (previous_window) {
- for(k=j, i=0;i<vc->blocksize_1/2;++i, k+=step) {
- ret[k]=saved[i]+buf[i]*lwin[i]+BIAS;
- }
- retlen=vc->blocksize_1/2;
+ vc->dsp.vector_fmul_add_add(ret+j, buf, lwin, saved, vc->add_bias, vc->blocksize[1]/2, step);
+ retlen=vc->blocksize[1]/2;
} else {
- buf += (vc->blocksize_1-vc->blocksize_0)/4;
- for(k=j, i=0;i<vc->blocksize_0/2;++i, k+=step) {
- ret[k]=saved[i]+buf[i]*swin[i]+BIAS;
- }
- buf += vc->blocksize_0/2;
- for(i=0;i<(vc->blocksize_1-vc->blocksize_0)/4;++i, k+=step) {
- ret[k]=buf[i]+BIAS;
+ int len = (vc->blocksize[1]-vc->blocksize[0])/4;
+ buf += len;
+ vc->dsp.vector_fmul_add_add(ret+j, buf, swin, saved, vc->add_bias, vc->blocksize[0]/2, step);
+ k = vc->blocksize[0]/2*step + j;
+ buf += vc->blocksize[0]/2;
+ if(vc->exp_bias){
+ for(i=0; i<len; i++, k+=step)
+ ((uint32_t*)ret)[k] = ((uint32_t*)buf)[i] + vc->exp_bias; // ret[k]=buf[i]*(1<<bias)
+ } else {
+ for(i=0; i<len; i++, k+=step)
+ ret[k] = buf[i] + fadd_bias;
}
buf=vc->buf;
- retlen=vc->blocksize_0/2+(vc->blocksize_1-vc->blocksize_0)/4;
+ retlen=vc->blocksize[0]/2+len;
}
// -- save
if (next_window) {
- buf += vc->blocksize_1/2;
- lwin += vc->blocksize_1/2-1;
- for(i=0;i<vc->blocksize_1/2;++i) {
- saved[i]=buf[i]*lwin[-i];
- }
+ buf += vc->blocksize[1]/2;
+ vc->dsp.vector_fmul_reverse(saved, buf, lwin, vc->blocksize[1]/2);
saved_start=0;
} else {
- saved_start=(vc->blocksize_1-vc->blocksize_0)/4;
- buf += vc->blocksize_1/2;
- for(i=0;i<saved_start;++i) {
- saved[i]=buf[i];
- }
- swin += vc->blocksize_0/2-1;
- for(i=0;i<vc->blocksize_0/2;++i) {
- saved[saved_start+i]=buf[saved_start+i]*swin[-i];
- }
+ saved_start=(vc->blocksize[1]-vc->blocksize[0])/4;
+ buf += vc->blocksize[1]/2;
+ for(i=0; i<saved_start; i++)
+ ((uint32_t*)saved)[i] = ((uint32_t*)buf)[i] + vc->exp_bias;
+ vc->dsp.vector_fmul_reverse(saved+saved_start, buf+saved_start, swin, vc->blocksize[0]/2);
}
} else {
// --overlap/add
- for(k=j, i=0;i<saved_start;++i, k+=step) {
- ret[k]=saved[i]+BIAS;
- }
- for(i=0;i<vc->blocksize_0/2;++i, k+=step) {
- ret[k]=saved[saved_start+i]+buf[i]*swin[i]+BIAS;
+ if(vc->add_bias) {
+ for(k=j, i=0;i<saved_start;++i, k+=step)
+ ret[k] = saved[i] + fadd_bias;
+ } else {
+ for(k=j, i=0;i<saved_start;++i, k+=step)
+ ret[k] = saved[i];
}
- retlen=saved_start+vc->blocksize_0/2;
+ vc->dsp.vector_fmul_add_add(ret+k, buf, swin, saved+saved_start, vc->add_bias, vc->blocksize[0]/2, step);
+ retlen=saved_start+vc->blocksize[0]/2;
// -- save
- buf += vc->blocksize_0/2;
- swin += vc->blocksize_0/2-1;
- for(i=0;i<vc->blocksize_0/2;++i) {
- saved[i]=buf[i]*swin[-i];
- }
+ buf += vc->blocksize[0]/2;
+ vc->dsp.vector_fmul_reverse(saved, buf, swin, vc->blocksize[0]/2);
saved_start=0;
}
}
AV_DEBUG("parsed %d bytes %d bits, returned %d samples (*ch*bits) \n", get_bits_count(gb)/8, get_bits_count(gb)%8, len);
- for(i=0;i<len;++i) {
- int_fast32_t tmp= ((int32_t*)vc->ret)[i];
- if(tmp & 0xf0000){
-// tmp= (0x43c0ffff - tmp)>>31; //ask gcc devs why this is slower
- if(tmp > 0x43c0ffff) tmp= 0xFFFF;
- else tmp= 0;
- }
- ((int16_t*)data)[i]=tmp - 0x8000;
- }
-
+ vc->dsp.float_to_int16(data, vc->ret, len);
*data_size=len*2;
return buf_size ;
projects / ffmpeg / blobdiff