/* l3loop->c */ #include "types.h" #include "tables.h" #include "l3loop.h" #include "layer3.h" #include "huffman.h" #include "bitstream.h" #include "l3bitstream.h" #include "reservoir.h" #define e 2.71828182845 #define CBLIMIT 21 #define SFB_LMAX 22 #define en_tot_krit 10 #define en_dif_krit 100 #define en_scfsi_band_krit 10 #define xm_scfsi_band_krit 10 static void calc_scfsi(shine_psy_xmin_t *l3_xmin, int ch, int gr, shine_global_config *config); static int part2_length(int gr, int ch, shine_global_config *config); static int bin_search_StepSize(int desired_rate, int ix[GRANULE_SIZE], gr_info * cod_info, shine_global_config *config); static int count_bit(int ix[GRANULE_SIZE], unsigned int start, unsigned int end, unsigned int table ); static int bigv_bitcount(int ix[GRANULE_SIZE], gr_info *gi); static int new_choose_table( int ix[GRANULE_SIZE], unsigned int begin, unsigned int end ); static void bigv_tab_select( int ix[GRANULE_SIZE], gr_info *cod_info ); static void subdivide(gr_info *cod_info, shine_global_config *config ); static int count1_bitcount( int ix[ GRANULE_SIZE ], gr_info *cod_info ); static void calc_runlen( int ix[GRANULE_SIZE], gr_info *cod_info ); static void calc_xmin( gr_info *cod_info, shine_psy_xmin_t *l3_xmin, int gr, int ch ); static int quantize(int ix[GRANULE_SIZE], int stepsize, shine_global_config *config); int sqrt_int(int r) { float x; float rr = r; float y = rr*0.5; *(unsigned int*)&x = (0xbe6f0000 - *(uint32_t*)&rr) >> 1; x = (1.5f*x) - (x*x)*(x*y); if(r>101123) x = (1.5f*x) - (x*x)*(x*y); int is = (int)(x*rr + 0.5f); return is + ((r - is*is)>>31); } #define SQRT_MAGIC_F 0x5f3759df float f_sqrt(const float x) { const float xhalf = 0.5f*x; //float step; union // get bits for floating value { float x; int i; } u; u.x = x; u.i = SQRT_MAGIC_F - (u.i >> 1); // gives initial guess y0 return x*u.x*(1.5f - xhalf*u.x*u.x);// Newton step, repeating increases accuracy } /* * shine_inner_loop: * ---------- * The code selects the best quantizerStepSize for a particular set * of scalefacs. */ int shine_inner_loop(int ix[GRANULE_SIZE], int max_bits, gr_info *cod_info, int gr, int ch, shine_global_config *config ) { int bits, c1bits, bvbits; if(max_bits<0) cod_info->quantizerStepSize--; do { while(quantize(ix,++cod_info->quantizerStepSize,config) > 8192); /* within table range? */ calc_runlen(ix,cod_info); /* rzero,count1,big_values*/ bits = c1bits = count1_bitcount(ix,cod_info); /* count1_table selection*/ subdivide(cod_info, config); /* bigvalues sfb division */ bigv_tab_select(ix,cod_info); /* codebook selection*/ bits += bvbits = bigv_bitcount( ix, cod_info ); /* bit count */ } while(bits>max_bits); return bits; } /* * shine_outer_loop: * ----------- * Function: The outer iteration loop controls the masking conditions * of all scalefactorbands. It computes the best scalefac and * global gain. This module calls the inner iteration loop. */ int shine_outer_loop( int max_bits, shine_psy_xmin_t *l3_xmin, /* the allowed distortion of the scalefactor */ int ix[GRANULE_SIZE], /* vector of quantized values ix(0..575) */ int gr, int ch, shine_global_config *config) { int bits, huff_bits; shine_side_info_t *side_info = &config->side_info; gr_info *cod_info = &side_info->gr[gr].ch[ch].tt; cod_info->quantizerStepSize = bin_search_StepSize(max_bits,ix,cod_info, config); cod_info->part2_length = part2_length(gr,ch,config); huff_bits = max_bits - cod_info->part2_length; bits = shine_inner_loop(ix, huff_bits, cod_info, gr, ch, config ); cod_info->part2_3_length = cod_info->part2_length + bits; return cod_info->part2_3_length; } /* * shine_iteration_loop: * ------------------ */ void shine_iteration_loop(shine_global_config *config) { shine_psy_xmin_t l3_xmin; gr_info *cod_info; int max_bits; int ch, gr, i; int *ix; for(ch=config->wave.channels; ch--; ) { for(gr=0; grmpeg.granules_per_frame; gr++) { /* setup pointers */ ix = config->l3_enc[ch][gr]; config->l3loop->xr = config->mdct_freq[ch][gr]; /* Precalculate the square, abs, and maximum, * for use later on. */ for (i=GRANULE_SIZE, config->l3loop->xrmax=0; i--;) { config->l3loop->xrsq[i] = asm_mulsr(config->l3loop->xr[i],config->l3loop->xr[i]); config->l3loop->xrabs[i] = abs(config->l3loop->xr[i]); if(config->l3loop->xrabs[i]>config->l3loop->xrmax) config->l3loop->xrmax=config->l3loop->xrabs[i]; } cod_info = (gr_info *) &(config->side_info.gr[gr].ch[ch]); cod_info->sfb_lmax = SFB_LMAX - 1; /* gr_deco */ calc_xmin(cod_info, &l3_xmin, gr, ch ); if ( config->mpeg.version == MPEG_I ) calc_scfsi(&l3_xmin,ch,gr,config); /* calculation of number of available bit( per granule ) */ max_bits = shine_max_reservoir_bits(&config->pe[ch][gr],config); /* reset of iteration variables */ memset(config->scalefactor.l[gr][ch],0,sizeof(config->scalefactor.l[gr][ch])); memset(config->scalefactor.s[gr][ch],0,sizeof(config->scalefactor.s[gr][ch])); for ( i=4; i--; ) cod_info->slen[i] = 0; cod_info->part2_3_length = 0; cod_info->big_values = 0; cod_info->count1 = 0; cod_info->scalefac_compress = 0; cod_info->table_select[0] = 0; cod_info->table_select[1] = 0; cod_info->table_select[2] = 0; cod_info->region0_count = 0; cod_info->region1_count = 0; cod_info->part2_length = 0; cod_info->preflag = 0; cod_info->scalefac_scale = 0; cod_info->count1table_select= 0; /* all spectral values zero ? */ if(config->l3loop->xrmax) cod_info->part2_3_length = shine_outer_loop(max_bits,&l3_xmin,ix, gr,ch,config); shine_ResvAdjust(cod_info, config ); cod_info->global_gain = cod_info->quantizerStepSize+210; } /* for gr */ } /* for ch */ shine_ResvFrameEnd(config); } /* * calc_scfsi: * ----------- * calculation of the scalefactor select information ( scfsi ). */ void calc_scfsi( shine_psy_xmin_t *l3_xmin, int ch, int gr, shine_global_config *config ) { shine_side_info_t *l3_side = &config->side_info; /* This is the scfsi_band table from 2.4.2.7 of the IS */ static const int scfsi_band_long[5] = { 0, 6, 11, 16, 21 }; int scfsi_band; unsigned scfsi_set; int sfb, start, end, i; int condition = 0; int temp; const int *scalefac_band_long = &shine_scale_fact_band_index[config->mpeg.samplerate_index][0]; // note. it goes quite a bit faster if you uncomment the next bit and exit // early from scfsi, but you then loose the advantage of common scale factors. /*for(scfsi_band=0;scfsi_band<4;scfsi_band++) l3_side->scfsi[ch][scfsi_band] = 0; return;*/ config->l3loop->xrmaxl[gr] = config->l3loop->xrmax; scfsi_set = 0; /* the total energy of the granule */ for ( temp = 0, i =GRANULE_SIZE; i--; ) temp += config->l3loop->xrsq[i]>>10; /* a bit of scaling to avoid overflow, (not very good) */ if ( temp ) config->l3loop->en_tot[gr] = log((float)temp * 4.768371584e-7) / LN2; /* 1024 / 0x7fffffff */ else config->l3loop->en_tot[gr] = 0; /* the energy of each scalefactor band, en */ /* the allowed distortion of each scalefactor band, xm */ for(sfb=21; sfb--; ) { start = scalefac_band_long[ sfb ]; end = scalefac_band_long[ sfb+1 ]; for ( temp = 0, i = start; i < end; i++ ) temp += config->l3loop->xrsq[i]>>10; if ( temp ) config->l3loop->en[gr][sfb] = log((float)temp * 4.768371584e-7) / LN2; /* 1024 / 0x7fffffff */ else config->l3loop->en[gr][sfb] = 0; if ( l3_xmin->l[gr][ch][sfb]) config->l3loop->xm[gr][sfb] = log( l3_xmin->l[gr][ch][sfb] ) / LN2; else config->l3loop->xm[gr][sfb] = 0; } if(gr==1) { int gr2, tp; for(gr2=2; gr2--; ) { /* The spectral values are not all zero */ if(config->l3loop->xrmaxl[gr2]) condition++; condition++; } if(abs(config->l3loop->en_tot[0]-config->l3loop->en_tot[1]) < en_tot_krit) condition++; for(tp=0,sfb=21; sfb--; ) tp += abs(config->l3loop->en[0][sfb]-config->l3loop->en[1][sfb]); if (tp < en_dif_krit) condition++; if(condition==6) { for(scfsi_band=0;scfsi_band<4;scfsi_band++) { int sum0 = 0, sum1 = 0; l3_side->scfsi[ch][scfsi_band] = 0; start = scfsi_band_long[scfsi_band]; end = scfsi_band_long[scfsi_band+1]; for ( sfb = start; sfb < end; sfb++ ) { sum0 += abs( config->l3loop->en[0][sfb] - config->l3loop->en[1][sfb] ); sum1 += abs( config->l3loop->xm[0][sfb] - config->l3loop->xm[1][sfb] ); } if(sum0scfsi[ch][scfsi_band] = 1; scfsi_set |= (1 << scfsi_band); } else l3_side->scfsi[ch][scfsi_band] = 0; } /* for scfsi_band */ } /* if condition == 6 */ else for(scfsi_band=0;scfsi_band<4;scfsi_band++) l3_side->scfsi[ch][scfsi_band] = 0; } /* if gr == 1 */ } /* * part2_length: * ------------- * calculates the number of bits needed to encode the scalefacs in the * main data block. */ int part2_length(int gr, int ch, shine_global_config *config) { int slen1, slen2, bits; gr_info *gi = &config->side_info.gr[gr].ch[ch].tt; bits = 0; { slen1 = shine_slen1_tab[ gi->scalefac_compress ]; slen2 = shine_slen2_tab[ gi->scalefac_compress ]; if ( !gr || !(config->side_info.scfsi[ch][0]) ) bits += (6 * slen1); if ( !gr || !(config->side_info.scfsi[ch][1]) ) bits += (5 * slen1); if ( !gr || !(config->side_info.scfsi[ch][2]) ) bits += (5 * slen2); if ( !gr || !(config->side_info.scfsi[ch][3]) ) bits += (5 * slen2); } return bits; } /* * calc_xmin: * ---------- * Calculate the allowed distortion for each scalefactor band, * as determined by the psychoacoustic model. * xmin(sb) = ratio(sb) * en(sb) / bw(sb) */ void calc_xmin(gr_info *cod_info, shine_psy_xmin_t *l3_xmin, int gr, int ch ) { int sfb; for ( sfb = cod_info->sfb_lmax; sfb--; ) { /*note. xmin will always be zero with no psychoacoustic model start = scalefac_band_long[ sfb ]; end = scalefac_band_long[ sfb+1 ]; bw = end - start; for ( en = 0, l = start; l < end; l++ ) en += config->l3loop->xrsq[l]; l3_xmin->l[gr][ch][sfb] = ratio->l[gr][ch][sfb] * en / bw;*/ l3_xmin->l[gr][ch][sfb] = 0; } } /* * shine_loop_initialise: * ------------------- * Calculates the look up tables used by the iteration loop. */ void shine_loop_initialise(shine_global_config *config) { int i; /* quantize: stepsize conversion, fourth root of 2 table. * The table is inverted (negative power) from the equation given * in the spec because it is quicker to do x*y than x/y. * The 0.5 is for rounding. */ for(i=128; i--;) { config->l3loop->steptab[i] = pow(2.0,(double)(127-i)/4); if((config->l3loop->steptab[i]*2)>0x7fffffff) /* MAXINT = 2**31 = 2**(124/4) */ config->l3loop->steptabi[i]=0x7fffffff; else /* The table is multiplied by 2 to give an extra bit of accuracy. * In quantize, the long multiply does not shift it's result left one * bit to compensate. */ config->l3loop->steptabi[i] = (int)((config->l3loop->steptab[i]*2) + 0.5); } /* quantize: vector conversion, three quarter power table. * The 0.5 is for rounding, the .0946 comes from the spec. */ for(i=10000; i--;) config->l3loop->int2idx[i] = (int)(sqrt(sqrt((double)i)*(double)i) - 0.0946 + 0.5); } /* * quantize: * --------- * Function: Quantization of the vector xr ( -> ix). * Returns maximum value of ix. */ int quantize(int ix[GRANULE_SIZE], int stepsize, shine_global_config *config ) { int i, max, ln; int scalei; float scale, dbl; scalei = config->l3loop->steptabi[stepsize+127]; /* 2**(-stepsize/4) */ /* a quick check to see if ixmax will be less than 8192 */ /* this speeds up the early calls to bin_search_StepSize */ if((asm_mulr(config->l3loop->xrmax,scalei)) > 165140) /* 8192**(4/3) */ max = 16384; /* no point in continuing, stepsize not big enough */ else for(i=0, max=0;il3loop->xr[i]),scalei); if(ln<10000) /* ln < 10000 catches most values */ ix[i] = config->l3loop->int2idx[ln]; /* quick look up method */ else { /* outside table range so have to do it using floats */ scale = config->l3loop->steptab[stepsize+127]; /* 2**(-stepsize/4) */ dbl = ((float)config->l3loop->xrabs[i]) * scale * 4.656612875e-10; /* 0x7fffffff */ //ix[i] = sqrt_int((int)(f_sqrt(dbl)*dbl)); /* dbl**(3/4) */ ix[i] = (int)sqrt(sqrt(dbl)*dbl); /* dbl**(3/4) */ } /* calculate ixmax while we're here */ /* note. ix cannot be negative */ if(max < ix[i]) max = ix[i]; } return max; } /* * ix_max: * ------- * Function: Calculate the maximum of ix from 0 to 575 */ static inline int ix_max( int ix[GRANULE_SIZE], unsigned int begin, unsigned int end ) { int i; int max = 0; for(i=begin;i 1; i -= 2 ) if ( !ix[i-1] && !ix[i-2] ) rzero++; else break; cod_info->count1 = 0 ; for ( ; i > 3; i -= 4 ) if ( ix[i-1] <= 1 && ix[i-2] <= 1 && ix[i-3] <= 1 && ix[i-4] <= 1 ) cod_info->count1++; else break; cod_info->big_values = i>>1; } /* * count1_bitcount: * ---------------- * Determines the number of bits to encode the quadruples. */ int count1_bitcount(int ix[GRANULE_SIZE], gr_info *cod_info) { int p, i, k; int v, w, x, y, signbits; int sum0 = 0, sum1 = 0; for(i=cod_info->big_values<<1, k=0; kcount1; i+=4, k++) { v = ix[i]; w = ix[i+1]; x = ix[i+2]; y = ix[i+3]; p = v + (w<<1) + (x<<2) + (y<<3); signbits = 0; if(v!=0) signbits++; if(w!=0) signbits++; if(x!=0) signbits++; if(y!=0) signbits++; sum0 += signbits; sum1 += signbits; sum0 += shine_huffman_table[32].hlen[p]; sum1 += shine_huffman_table[33].hlen[p]; } if(sum0count1table_select = 0; return sum0; } else { cod_info->count1table_select = 1; return sum1; } } /* * subdivide: * ---------- * presumable subdivides the bigvalue region which will use separate Huffman tables. */ void subdivide(gr_info *cod_info, shine_global_config *config) { static const struct { unsigned region0_count; unsigned region1_count; } subdv_table[ 23 ] = { {0, 0}, /* 0 bands */ {0, 0}, /* 1 bands */ {0, 0}, /* 2 bands */ {0, 0}, /* 3 bands */ {0, 0}, /* 4 bands */ {0, 1}, /* 5 bands */ {1, 1}, /* 6 bands */ {1, 1}, /* 7 bands */ {1, 2}, /* 8 bands */ {2, 2}, /* 9 bands */ {2, 3}, /* 10 bands */ {2, 3}, /* 11 bands */ {3, 4}, /* 12 bands */ {3, 4}, /* 13 bands */ {3, 4}, /* 14 bands */ {4, 5}, /* 15 bands */ {4, 5}, /* 16 bands */ {4, 6}, /* 17 bands */ {5, 6}, /* 18 bands */ {5, 6}, /* 19 bands */ {5, 7}, /* 20 bands */ {6, 7}, /* 21 bands */ {6, 7}, /* 22 bands */ }; if (!cod_info->big_values) { /* no big_values region */ cod_info->region0_count = 0; cod_info->region1_count = 0; } else { const int *scalefac_band_long = &shine_scale_fact_band_index[config->mpeg.samplerate_index][0]; int bigvalues_region, scfb_anz, thiscount; bigvalues_region = 2 * cod_info->big_values; /* Calculate scfb_anz */ scfb_anz = 0; while ( scalefac_band_long[scfb_anz] < bigvalues_region ) scfb_anz++; for (thiscount = subdv_table[scfb_anz].region0_count; thiscount; thiscount--) { if (scalefac_band_long[thiscount + 1] <= bigvalues_region) break; } cod_info->region0_count = thiscount; cod_info->address1 = scalefac_band_long[thiscount + 1]; scalefac_band_long += cod_info->region0_count + 1; for (thiscount = subdv_table[scfb_anz].region1_count; thiscount; thiscount--) { if (scalefac_band_long[thiscount + 1] <= bigvalues_region) break; } cod_info->region1_count = thiscount; cod_info->address2 = scalefac_band_long[thiscount + 1]; cod_info->address3 = bigvalues_region; } } /* * bigv_tab_select: * ---------------- * Function: Select huffman code tables for bigvalues regions */ void bigv_tab_select( int ix[GRANULE_SIZE], gr_info *cod_info ) { cod_info->table_select[0] = 0; cod_info->table_select[1] = 0; cod_info->table_select[2] = 0; { if ( cod_info->address1 > 0 ) cod_info->table_select[0] = new_choose_table( ix, 0, cod_info->address1 ); if ( cod_info->address2 > cod_info->address1 ) cod_info->table_select[1] = new_choose_table( ix, cod_info->address1, cod_info->address2 ); if ( cod_info->big_values<<1 > cod_info->address2 ) cod_info->table_select[2] = new_choose_table( ix, cod_info->address2, cod_info->big_values<<1 ); } } /* * new_choose_table: * ----------------- * Choose the Huffman table that will encode ix[begin..end] with * the fewest bits. * Note: This code contains knowledge about the sizes and characteristics * of the Huffman tables as defined in the IS (Table B.7), and will not work * with any arbitrary tables. */ int new_choose_table( int ix[GRANULE_SIZE], unsigned int begin, unsigned int end ) { int i, max; int choice[2]; int sum[2]; max = ix_max(ix,begin,end); if(!max) return 0; choice[0] = 0; choice[1] = 0; if(max<15) { /* try tables with no linbits */ for ( i =14; i--; ) if ( shine_huffman_table[i].xlen > max ) { choice[0] = i; break; } sum[0] = count_bit( ix, begin, end, choice[0] ); switch (choice[0]) { case 2: sum[1] = count_bit( ix, begin, end, 3 ); if ( sum[1] <= sum[0] ) choice[0] = 3; break; case 5: sum[1] = count_bit( ix, begin, end, 6 ); if ( sum[1] <= sum[0] ) choice[0] = 6; break; case 7: sum[1] = count_bit( ix, begin, end, 8 ); if ( sum[1] <= sum[0] ) { choice[0] = 8; sum[0] = sum[1]; } sum[1] = count_bit( ix, begin, end, 9 ); if ( sum[1] <= sum[0] ) choice[0] = 9; break; case 10: sum[1] = count_bit( ix, begin, end, 11 ); if ( sum[1] <= sum[0] ) { choice[0] = 11; sum[0] = sum[1]; } sum[1] = count_bit( ix, begin, end, 12 ); if ( sum[1] <= sum[0] ) choice[0] = 12; break; case 13: sum[1] = count_bit( ix, begin, end, 15 ); if ( sum[1] <= sum[0] ) choice[0] = 15; break; } } else { /* try tables with linbits */ max -= 15; for(i=15;i<24;i++) if(shine_huffman_table[i].linmax>=max) { choice[0] = i; break; } for(i=24;i<32;i++) if(shine_huffman_table[i].linmax>=max) { choice[1] = i; break; } sum[0] = count_bit(ix,begin,end,choice[0]); sum[1] = count_bit(ix,begin,end,choice[1]); if (sum[1]table_select[0])) /* region0 */ bits += count_bit(ix, 0, gi->address1, table ); if( (table=gi->table_select[1])) /* region1 */ bits += count_bit(ix, gi->address1, gi->address2, table ); if( (table=gi->table_select[2])) /* region2 */ bits += count_bit(ix, gi->address2, gi->address3, table ); return bits; } /* * count_bit: * ---------- * Function: Count the number of bits necessary to code the subregion. */ int count_bit(int ix[GRANULE_SIZE], unsigned int start, unsigned int end, unsigned int table ) { unsigned linbits, ylen; int i, sum; int x,y; const struct huffcodetab *h; if(!table) return 0; h = &(shine_huffman_table[table]); sum = 0; ylen = h->ylen; linbits = h->linbits; if(table>15) { /* ESC-table is used */ for(i=start;i14) { x = 15; sum += linbits; } if(y>14) { y = 15; sum += linbits; } sum += h->hlen[(x*ylen)+y]; if(x) sum++; if(y) sum++; } } else { /* No ESC-words */ for(i=start;ihlen[(x*ylen)+y]; if(x!=0) sum++; if(y!=0) sum++; } } return sum; } /* * bin_search_StepSize: * -------------------- * Succesive approximation approach to obtaining a initial quantizer * step size. * The following optional code written by Seymour Shlien * will speed up the shine_outer_loop code which is called * by iteration_loop. When BIN_SEARCH is defined, the * shine_outer_loop function precedes the call to the function shine_inner_loop * with a call to bin_search gain defined below, which * returns a good starting quantizerStepSize. */ int bin_search_StepSize(int desired_rate, int ix[GRANULE_SIZE], gr_info * cod_info, shine_global_config *config) { int bit, next, count; next = -120; count = 120; do { int half = count / 2; if (quantize(ix, next + half, config) > 8192) bit = 100000; /* fail */ else { calc_runlen(ix, cod_info); /* rzero,count1,big_values */ bit = count1_bitcount(ix, cod_info); /* count1_table selection */ subdivide(cod_info, config); /* bigvalues sfb division */ bigv_tab_select(ix, cod_info); /* codebook selection */ bit += bigv_bitcount(ix, cod_info); /* bit count */ } if (bit < desired_rate) count = half; else { next += half; count -= half; } } while (count > 1); return next; }