Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | Download | RSS feed

  1. /*
  2.         leyer3.c: the layer 3 decoder
  3.  
  4.         copyright 1995-2006 by the mpg123 project - free software under the terms of the LGPL 2.1
  5.         see COPYING and AUTHORS files in distribution or http://mpg123.de
  6.         initially written by Michael Hipp
  7.  
  8.         Optimize-TODO: put short bands into the band-field without the stride of 3 reals
  9.         Length-optimze: unify long and short band code where it is possible
  10.  
  11.         The int-vs-pointer situation has to be cleaned up.
  12. */
  13.  
  14. //#include <stdlib.h>
  15. //#include "config.h"
  16. #include "mpg123.h"
  17. #include "huffman.h"
  18.  
  19. //#include "common.h"
  20. //#include "debug.h"
  21.  
  22. #include "getbits.h"
  23.  
  24. static real ispow[8207];
  25. static real aa_ca[8],aa_cs[8];
  26. static real COS1[12][6];
  27. static real win[4][36];
  28. static real win1[4][36];
  29. static real gainpow2[256+118+4];
  30. #ifdef USE_3DNOW
  31. real COS9[9];
  32. static real COS6_1,COS6_2;
  33. real tfcos36[9];
  34. #else
  35. static real COS9[9];
  36. static real COS6_1,COS6_2;
  37. static real tfcos36[9];
  38. #endif
  39. static real tfcos12[3];
  40. #define NEW_DCT9
  41. #ifdef NEW_DCT9
  42. static real cos9[3],cos18[3];
  43. #endif
  44.  
  45. struct bandInfoStruct {
  46.   int longIdx[23];
  47.   int longDiff[22];
  48.   int shortIdx[14];
  49.   int shortDiff[13];
  50. };
  51.  
  52. int longLimit[9][23];
  53. int shortLimit[9][14];
  54.  
  55. struct bandInfoStruct bandInfo[9] = {
  56.  
  57. /* MPEG 1.0 */
  58.  { {0,4,8,12,16,20,24,30,36,44,52,62,74, 90,110,134,162,196,238,288,342,418,576},
  59.    {4,4,4,4,4,4,6,6,8, 8,10,12,16,20,24,28,34,42,50,54, 76,158},
  60.    {0,4*3,8*3,12*3,16*3,22*3,30*3,40*3,52*3,66*3, 84*3,106*3,136*3,192*3},
  61.    {4,4,4,4,6,8,10,12,14,18,22,30,56} } ,
  62.  
  63.  { {0,4,8,12,16,20,24,30,36,42,50,60,72, 88,106,128,156,190,230,276,330,384,576},
  64.    {4,4,4,4,4,4,6,6,6, 8,10,12,16,18,22,28,34,40,46,54, 54,192},
  65.    {0,4*3,8*3,12*3,16*3,22*3,28*3,38*3,50*3,64*3, 80*3,100*3,126*3,192*3},
  66.    {4,4,4,4,6,6,10,12,14,16,20,26,66} } ,
  67.  
  68.  { {0,4,8,12,16,20,24,30,36,44,54,66,82,102,126,156,194,240,296,364,448,550,576} ,
  69.    {4,4,4,4,4,4,6,6,8,10,12,16,20,24,30,38,46,56,68,84,102, 26} ,
  70.    {0,4*3,8*3,12*3,16*3,22*3,30*3,42*3,58*3,78*3,104*3,138*3,180*3,192*3} ,
  71.    {4,4,4,4,6,8,12,16,20,26,34,42,12} }  ,
  72.  
  73. /* MPEG 2.0 */
  74.  { {0,6,12,18,24,30,36,44,54,66,80,96,116,140,168,200,238,284,336,396,464,522,576},
  75.    {6,6,6,6,6,6,8,10,12,14,16,20,24,28,32,38,46,52,60,68,58,54 } ,
  76.    {0,4*3,8*3,12*3,18*3,24*3,32*3,42*3,56*3,74*3,100*3,132*3,174*3,192*3} ,
  77.    {4,4,4,6,6,8,10,14,18,26,32,42,18 } } ,
  78.  
  79. /* mhipp trunk has 330 -> 332 without further explanation ... */
  80.  { {0,6,12,18,24,30,36,44,54,66,80,96,114,136,162,194,232,278,330,394,464,540,576},
  81.    {6,6,6,6,6,6,8,10,12,14,16,18,22,26,32,38,46,52,64,70,76,36 } ,
  82.    {0,4*3,8*3,12*3,18*3,26*3,36*3,48*3,62*3,80*3,104*3,136*3,180*3,192*3} ,
  83.    {4,4,4,6,8,10,12,14,18,24,32,44,12 } } ,
  84.  
  85.  { {0,6,12,18,24,30,36,44,54,66,80,96,116,140,168,200,238,284,336,396,464,522,576},
  86.    {6,6,6,6,6,6,8,10,12,14,16,20,24,28,32,38,46,52,60,68,58,54 },
  87.    {0,4*3,8*3,12*3,18*3,26*3,36*3,48*3,62*3,80*3,104*3,134*3,174*3,192*3},
  88.    {4,4,4,6,8,10,12,14,18,24,30,40,18 } } ,
  89. /* MPEG 2.5 */
  90.  { {0,6,12,18,24,30,36,44,54,66,80,96,116,140,168,200,238,284,336,396,464,522,576} ,
  91.    {6,6,6,6,6,6,8,10,12,14,16,20,24,28,32,38,46,52,60,68,58,54},
  92.    {0,12,24,36,54,78,108,144,186,240,312,402,522,576},
  93.    {4,4,4,6,8,10,12,14,18,24,30,40,18} },
  94.  { {0,6,12,18,24,30,36,44,54,66,80,96,116,140,168,200,238,284,336,396,464,522,576} ,
  95.    {6,6,6,6,6,6,8,10,12,14,16,20,24,28,32,38,46,52,60,68,58,54},
  96.    {0,12,24,36,54,78,108,144,186,240,312,402,522,576},
  97.    {4,4,4,6,8,10,12,14,18,24,30,40,18} },
  98.  { {0,12,24,36,48,60,72,88,108,132,160,192,232,280,336,400,476,566,568,570,572,574,576},
  99.    {12,12,12,12,12,12,16,20,24,28,32,40,48,56,64,76,90,2,2,2,2,2},
  100.    {0, 24, 48, 72,108,156,216,288,372,480,486,492,498,576},
  101.    {8,8,8,12,16,20,24,28,36,2,2,2,26} } ,
  102. };
  103.  
  104. static int mapbuf0[9][152];
  105. static int mapbuf1[9][156];
  106. static int mapbuf2[9][44];
  107. static int *map[9][3];
  108. static int *mapend[9][3];
  109.  
  110. static unsigned int n_slen2[512]; /* MPEG 2.0 slen for 'normal' mode */
  111. static unsigned int i_slen2[256]; /* MPEG 2.0 slen for intensity stereo */
  112.  
  113. static real tan1_1[16],tan2_1[16],tan1_2[16],tan2_2[16];
  114. static real pow1_1[2][16],pow2_1[2][16],pow1_2[2][16],pow2_2[2][16];
  115.  
  116. #ifdef GAPLESS
  117. /* still a dirty hack, places in bytes (zero-based)... */
  118. static unsigned long position; /* position in raw decoder bytestream */
  119. static unsigned long begin; /* first byte to play == number to skip */
  120. static unsigned long end; /* last byte to play */
  121. static unsigned long ignore; /* forcedly ignore stuff in between */
  122. static int bytified;
  123.  
  124. /* input in bytes already */
  125. void layer3_gapless_init(unsigned long b, unsigned long e)
  126. {
  127.         bytified = 0;
  128.         position = 0;
  129.         ignore = 0;
  130.         begin = b;
  131.         end = e;
  132.         debug2("layer3_gapless_init: from %lu to %lu samples", begin, end);
  133. }
  134.  
  135. void layer3_gapless_set_position(unsigned long frames, struct frame* fr, struct audio_info_struct *ai)
  136. {
  137.         position = samples_to_bytes(frames*spf(fr), fr, ai);
  138.         debug1("set; position now %lu", position);
  139. }
  140.  
  141. void layer3_gapless_bytify(struct frame *fr, struct audio_info_struct *ai)
  142. {
  143.         if(!bytified)
  144.         {
  145.                 begin = samples_to_bytes(begin, fr, ai);
  146.                 end = samples_to_bytes(end, fr, ai);
  147.                 bytified = 1;
  148.                 debug2("bytified: begin=%lu; end=%5lu", begin, end);
  149.         }
  150. }
  151.  
  152. /* I need initialized fr here! */
  153. void layer3_gapless_set_ignore(unsigned long frames, struct frame *fr, struct audio_info_struct *ai)
  154. {
  155.         ignore = samples_to_bytes(frames*spf(fr), fr, ai);
  156. }
  157.  
  158. /*
  159.         take the (partially or fully) filled and remove stuff for gapless mode if needed
  160.         pcm_point may then be smaller than before...
  161. */
  162. void layer3_gapless_buffercheck()
  163. {
  164.         /* pcm_point bytes added since last position... */
  165.         unsigned long new_pos = position + pcm_point;
  166.         if(begin && (position < begin))
  167.         {
  168.                 debug4("new_pos %lu (old: %lu), begin %lu, pcm_point %i", new_pos, position, begin, pcm_point);
  169.                 if(new_pos < begin)
  170.                 {
  171.                         if(ignore > pcm_point) ignore -= pcm_point;
  172.                         else ignore = 0;
  173.                         pcm_point = 0; /* full of padding/delay */
  174.                 }
  175.                 else
  176.                 {
  177.                         unsigned long ignored = begin-position;
  178.                         /* we need to shift the memory to the left... */
  179.                         debug3("old pcm_point: %i, begin %lu; good bytes: %i", pcm_point, begin, (int)(new_pos-begin));
  180.                         if(ignore > ignored) ignore -= ignored;
  181.                         else ignore = 0;
  182.                         pcm_point -= ignored;
  183.                         debug3("shifting %i bytes from %p to %p", pcm_point, pcm_sample+(int)(begin-position), pcm_sample);
  184.                         memmove(pcm_sample, pcm_sample+(int)(begin-position), pcm_point);
  185.                 }
  186.         }
  187.         /* I don't cover the case with both end and begin in chunk! */
  188.         else if(end && (new_pos > end))
  189.         {
  190.                 ignore = 0;
  191.                 /* either end in current chunk or chunk totally out */
  192.                 debug2("ending at position %lu / point %i", new_pos, pcm_point);
  193.                 if(position < end)      pcm_point -= new_pos-end;
  194.                 else pcm_point = 0;
  195.                 debug1("set pcm_point to %i", pcm_point);
  196.         }
  197.         else if(ignore)
  198.         {
  199.                 if(pcm_point < ignore)
  200.                 {
  201.                         ignore -= pcm_point;
  202.                         debug2("ignored %i bytes; pcm_point = 0; %lu bytes left", pcm_point, ignore);
  203.                         pcm_point = 0;
  204.                 }
  205.                 else
  206.                 {
  207.                         /* we need to shift the memory to the left... */
  208.                         debug3("old pcm_point: %i, to ignore: %lu; good bytes: %i", pcm_point, ignore, pcm_point-(int)ignore);
  209.                         pcm_point -= ignore;
  210.                         debug3("shifting %i bytes from %p to %p", pcm_point, pcm_sample+ignore, pcm_sample);
  211.                         memmove(pcm_sample, pcm_sample+ignore, pcm_point);
  212.                         ignore = 0;
  213.                 }
  214.         }
  215.         position = new_pos;
  216. }
  217. #endif
  218.  
  219. /*
  220.  * init tables for layer-3
  221.  */
  222.  
  223. real hybridIn [2][SBLIMIT][SSLIMIT];
  224. real hybridOut[2][SSLIMIT][SBLIMIT];
  225. static real block[2][2][SBLIMIT*SSLIMIT] = { { { 0, } } };
  226. static int blc[2]={0,0};
  227.  
  228. void reset_mpg()
  229. {
  230.   memset(block,0,sizeof(block));
  231.   blc[0]=0;
  232.   blc[1]=0;
  233.   init_dct();
  234. };
  235.  
  236.  
  237. #pragma warning(disable:4244)
  238. void init_layer3(int down_sample_sblimit)
  239. {
  240.   int i,j,k,l;
  241.  
  242.   for(i=-256;i<118+4;i++)
  243. #ifdef USE_MMX
  244.     if(!param.down_sample)
  245.       gainpow2[i+256] = 16384.0 * pow((double)2.0,-0.25 * (double) (i+210) );
  246.     else
  247. #endif
  248.     gainpow2[i+256] = DOUBLE_TO_REAL(pow_test((double)2.0,-0.25 * (double) (i+210)));
  249.  
  250.   for(i=0;i<8207;i++)
  251.     ispow[i] = DOUBLE_TO_REAL(pow_test((double)i,(double)4.0/3.0));
  252.  
  253.   for (i=0;i<8;i++) {
  254.     static double Ci[8]={-0.6,-0.535,-0.33,-0.185,-0.095,-0.041,-0.0142,-0.0037};
  255.     double sq=sqrt(1.0+Ci[i]*Ci[i]);
  256.     aa_cs[i] = DOUBLE_TO_REAL(1.0/sq);
  257.     aa_ca[i] = DOUBLE_TO_REAL(Ci[i]/sq);
  258.   }
  259.  
  260.   for(i=0;i<18;i++) {
  261.     win[0][i]    = win[1][i]    = DOUBLE_TO_REAL(0.5 * sin( M_PI / 72.0 * (double) (2*(i+0) +1) ) / cos ( M_PI * (double) (2*(i+0) +19) / 72.0 ));
  262.     win[0][i+18] = win[3][i+18] = DOUBLE_TO_REAL(0.5 * sin( M_PI / 72.0 * (double) (2*(i+18)+1) ) / cos ( M_PI * (double) (2*(i+18)+19) / 72.0 ));
  263.   }
  264.   for(i=0;i<6;i++) {
  265.     win[1][i+18] = DOUBLE_TO_REAL(0.5 / cos ( M_PI * (double) (2*(i+18)+19) / 72.0 ));
  266.     win[3][i+12] = DOUBLE_TO_REAL(0.5 / cos ( M_PI * (double) (2*(i+12)+19) / 72.0 ));
  267.     win[1][i+24] = DOUBLE_TO_REAL(0.5 * sin( M_PI / 24.0 * (double) (2*i+13) ) / cos ( M_PI * (double) (2*(i+24)+19) / 72.0 ));
  268.     win[1][i+30] = win[3][i] = DOUBLE_TO_REAL(0.0);
  269.     win[3][i+6 ] = DOUBLE_TO_REAL(0.5 * sin( M_PI / 24.0 * (double) (2*i+1) ) / cos ( M_PI * (double) (2*(i+6 )+19) / 72.0 ));
  270.   }
  271.  
  272.   for(i=0;i<9;i++)
  273.     COS9[i] = DOUBLE_TO_REAL(cos( M_PI / 18.0 * (double) i));
  274.  
  275.   for(i=0;i<9;i++)
  276.     tfcos36[i] = DOUBLE_TO_REAL(0.5 / cos ( M_PI * (double) (i*2+1) / 36.0 ));
  277.   for(i=0;i<3;i++)
  278.     tfcos12[i] = DOUBLE_TO_REAL(0.5 / cos ( M_PI * (double) (i*2+1) / 12.0 ));
  279.  
  280.   COS6_1 = DOUBLE_TO_REAL(cos( M_PI / 6.0 * (double) 1));
  281.   COS6_2 = DOUBLE_TO_REAL(cos( M_PI / 6.0 * (double) 2));
  282.  
  283. #ifdef NEW_DCT9
  284.   cos9[0]  = DOUBLE_TO_REAL(cos(1.0*M_PI/9.0));
  285.   cos9[1]  = DOUBLE_TO_REAL(cos(5.0*M_PI/9.0));
  286.   cos9[2]  = DOUBLE_TO_REAL(cos(7.0*M_PI/9.0));
  287.   cos18[0] = DOUBLE_TO_REAL(cos(1.0*M_PI/18.0));
  288.   cos18[1] = DOUBLE_TO_REAL(cos(11.0*M_PI/18.0));
  289.   cos18[2] = DOUBLE_TO_REAL(cos(13.0*M_PI/18.0));
  290. #endif
  291.  
  292.   for(i=0;i<12;i++) {
  293.     win[2][i]  = DOUBLE_TO_REAL(0.5 * sin( M_PI / 24.0 * (double) (2*i+1) ) / cos ( M_PI * (double) (2*i+7) / 24.0 ));
  294.     for(j=0;j<6;j++)
  295.       COS1[i][j] = DOUBLE_TO_REAL(cos( M_PI / 24.0 * (double) ((2*i+7)*(2*j+1)) ));
  296.   }
  297.  
  298.   for(j=0;j<4;j++) {
  299.     static int len[4] = { 36,36,12,36 };
  300.     for(i=0;i<len[j];i+=2)
  301.       win1[j][i] = + win[j][i];
  302.     for(i=1;i<len[j];i+=2)
  303.       win1[j][i] = - win[j][i];
  304.   }
  305.  
  306.   for(i=0;i<16;i++) {
  307.     double t = tan( (double) i * M_PI / 12.0 );
  308.     tan1_1[i] = DOUBLE_TO_REAL(t / (1.0+t));
  309.     tan2_1[i] = DOUBLE_TO_REAL(1.0 / (1.0 + t));
  310.     tan1_2[i] = DOUBLE_TO_REAL(M_SQRT2 * t / (1.0+t));
  311.     tan2_2[i] = DOUBLE_TO_REAL(M_SQRT2 / (1.0 + t));
  312.  
  313.     for(j=0;j<2;j++) {
  314.       double base = pow_test(2.0,-0.25*(j+1.0));
  315.       double p1=1.0,p2=1.0;
  316.       if(i > 0) {
  317.         if( i & 1 )
  318.           p1 = pow_test(base,(i+1.0)*0.5);
  319.         else
  320.           p2 = pow_test(base,i*0.5);
  321.       }
  322.       pow1_1[j][i] = DOUBLE_TO_REAL(p1);
  323.       pow2_1[j][i] = DOUBLE_TO_REAL(p2);
  324.       pow1_2[j][i] = DOUBLE_TO_REAL(M_SQRT2 * p1);
  325.       pow2_2[j][i] = DOUBLE_TO_REAL(M_SQRT2 * p2);
  326.     }
  327.   }
  328.  
  329.   for(j=0;j<9;j++) {
  330.    struct bandInfoStruct *bi = &bandInfo[j];
  331.    int *mp;
  332.    int cb,lwin;
  333.    int *bdf;
  334.  
  335.    mp = map[j][0] = mapbuf0[j];
  336.    bdf = bi->longDiff;
  337.    for(i=0,cb = 0; cb < 8 ; cb++,i+=*bdf++) {
  338.      *mp++ = (*bdf) >> 1;
  339.      *mp++ = i;
  340.      *mp++ = 3;
  341.      *mp++ = cb;
  342.    }
  343.    bdf = bi->shortDiff+3;
  344.    for(cb=3;cb<13;cb++) {
  345.      int l = (*bdf++) >> 1;
  346.      for(lwin=0;lwin<3;lwin++) {
  347.        *mp++ = l;
  348.        *mp++ = i + lwin;
  349.        *mp++ = lwin;
  350.        *mp++ = cb;
  351.      }
  352.      i += 6*l;
  353.    }
  354.    mapend[j][0] = mp;
  355.  
  356.    mp = map[j][1] = mapbuf1[j];
  357.    bdf = bi->shortDiff+0;
  358.    for(i=0,cb=0;cb<13;cb++) {
  359.      int l = (*bdf++) >> 1;
  360.      for(lwin=0;lwin<3;lwin++) {
  361.        *mp++ = l;
  362.        *mp++ = i + lwin;
  363.        *mp++ = lwin;
  364.        *mp++ = cb;
  365.      }
  366.      i += 6*l;
  367.    }
  368.    mapend[j][1] = mp;
  369.  
  370.    mp = map[j][2] = mapbuf2[j];
  371.    bdf = bi->longDiff;
  372.    for(cb = 0; cb < 22 ; cb++) {
  373.      *mp++ = (*bdf++) >> 1;
  374.      *mp++ = cb;
  375.    }
  376.    mapend[j][2] = mp;
  377.  
  378.   }
  379.  
  380.   for(j=0;j<9;j++) {
  381.     for(i=0;i<23;i++) {
  382.       longLimit[j][i] = (bandInfo[j].longIdx[i] - 1 + 8) / 18 + 1;
  383.       if(longLimit[j][i] > (down_sample_sblimit) )
  384.         longLimit[j][i] = down_sample_sblimit;
  385.     }
  386.     for(i=0;i<14;i++) {
  387.       shortLimit[j][i] = (bandInfo[j].shortIdx[i] - 1) / 18 + 1;
  388.       if(shortLimit[j][i] > (down_sample_sblimit) )
  389.         shortLimit[j][i] = down_sample_sblimit;
  390.     }
  391.   }
  392.  
  393.   for(i=0;i<5;i++) {
  394.     for(j=0;j<6;j++) {
  395.       for(k=0;k<6;k++) {
  396.         int n = k + j * 6 + i * 36;
  397.         i_slen2[n] = i|(j<<3)|(k<<6)|(3<<12);
  398.       }
  399.     }
  400.   }
  401.   for(i=0;i<4;i++) {
  402.     for(j=0;j<4;j++) {
  403.       for(k=0;k<4;k++) {
  404.         int n = k + j * 4 + i * 16;
  405.         i_slen2[n+180] = i|(j<<3)|(k<<6)|(4<<12);
  406.       }
  407.     }
  408.   }
  409.   for(i=0;i<4;i++) {
  410.     for(j=0;j<3;j++) {
  411.       int n = j + i * 3;
  412.       i_slen2[n+244] = i|(j<<3) | (5<<12);
  413.       n_slen2[n+500] = i|(j<<3) | (2<<12) | (1<<15);
  414.     }
  415.   }
  416.  
  417.   for(i=0;i<5;i++) {
  418.     for(j=0;j<5;j++) {
  419.       for(k=0;k<4;k++) {
  420.         for(l=0;l<4;l++) {
  421.           int n = l + k * 4 + j * 16 + i * 80;
  422.           n_slen2[n] = i|(j<<3)|(k<<6)|(l<<9)|(0<<12);
  423.         }
  424.       }
  425.     }
  426.   }
  427.   for(i=0;i<5;i++) {
  428.     for(j=0;j<5;j++) {
  429.       for(k=0;k<4;k++) {
  430.         int n = k + j * 4 + i * 20;
  431.         n_slen2[n+400] = i|(j<<3)|(k<<6)|(1<<12);
  432.       }
  433.     }
  434.   }
  435. }
  436.  
  437. /*
  438.  * read additional side information (for MPEG 1 and MPEG 2)
  439.  */
  440. static int III_get_side_info(struct III_sideinfo *si,int stereo,
  441.  int ms_stereo,long sfreq,int single,int lsf)
  442. {
  443.    int ch, gr;
  444.    int powdiff = (single == 3) ? 4 : 0;
  445.  
  446.    static const int tabs[2][5] = { { 2,9,5,3,4 } , { 1,8,1,2,9 } };
  447.    const int *tab = tabs[lsf];
  448.    
  449.    si->main_data_begin = getbits(tab[1]);
  450.    if (stereo == 1)
  451.      si->private_bits = getbits_fast(tab[2]);
  452.    else
  453.      si->private_bits = getbits_fast(tab[3]);
  454.  
  455.    if(!lsf) {
  456.      for (ch=0; ch<stereo; ch++) {
  457.          si->ch[ch].gr[0].scfsi = -1;
  458.          si->ch[ch].gr[1].scfsi = getbits_fast(4);
  459.      }
  460.    }
  461.  
  462.    for (gr=0; gr<tab[0]; gr++) {
  463.      for (ch=0; ch<stereo; ch++) {
  464.        register struct gr_info_s *gr_info = &(si->ch[ch].gr[gr]);
  465.  
  466.        gr_info->part2_3_length = getbits(12);
  467.        gr_info->big_values = getbits(9);
  468.        if(gr_info->big_values > 288) {
  469.           gr_info->big_values = 288;
  470.        }
  471.        gr_info->pow2gain = gainpow2+256 - getbits_fast(8) + powdiff;
  472.        if(ms_stereo)
  473.          gr_info->pow2gain += 2;
  474.        gr_info->scalefac_compress = getbits(tab[4]);
  475.  
  476.        if(get1bit()) { /* window switch flag  */
  477.          int i;
  478.          gr_info->block_type       = getbits_fast(2);
  479.          gr_info->mixed_block_flag = get1bit();
  480.          gr_info->table_select[0]  = getbits_fast(5);
  481.          gr_info->table_select[1]  = getbits_fast(5);
  482.          /*
  483.           * table_select[2] not needed, because there is no region2,
  484.           * but to satisfy some verifications tools we set it either.
  485.           */
  486.          gr_info->table_select[2] = 0;
  487.          for(i=0;i<3;i++)
  488.            gr_info->full_gain[i] = gr_info->pow2gain + (getbits_fast(3)<<3);
  489.  
  490.          if(gr_info->block_type == 0) {
  491.            /* exit(1); */
  492.            return 1;
  493.          }
  494.      
  495.          /* region_count/start parameters are implicit in this case. */      
  496.          if(!lsf || gr_info->block_type == 2)
  497.            gr_info->region1start = 36>>1;
  498.          else {
  499. /* check this again for 2.5 and sfreq=8 */
  500.            if(sfreq == 8)
  501.              gr_info->region1start = 108>>1;
  502.            else
  503.              gr_info->region1start = 54>>1;
  504.          }
  505.          gr_info->region2start = 576>>1;
  506.        }
  507.        else {
  508.          int i,r0c,r1c;
  509.          for (i=0; i<3; i++)
  510.            gr_info->table_select[i] = getbits_fast(5);
  511.          r0c = getbits_fast(4);
  512.          r1c = getbits_fast(3);
  513.          gr_info->region1start = bandInfo[sfreq].longIdx[r0c+1] >> 1 ;
  514.          gr_info->region2start = bandInfo[sfreq].longIdx[r0c+1+r1c+1] >> 1;
  515.          gr_info->block_type = 0;
  516.          gr_info->mixed_block_flag = 0;
  517.        }
  518.        if(!lsf)
  519.          gr_info->preflag = get1bit();
  520.        gr_info->scalefac_scale = get1bit();
  521.        gr_info->count1table_select = get1bit();
  522.      }
  523.    }
  524.    return 0;
  525. }
  526.  
  527. /*
  528.  * read scalefactors
  529.  */
  530. static int III_get_scale_factors_1(int *scf,struct gr_info_s *gr_info,int ch,int gr)
  531. {
  532.    static const unsigned char slen[2][16] = {
  533.      {0, 0, 0, 0, 3, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4},
  534.      {0, 1, 2, 3, 0, 1, 2, 3, 1, 2, 3, 1, 2, 3, 2, 3}
  535.    };
  536.    int numbits;
  537.    int num0 = slen[0][gr_info->scalefac_compress];
  538.    int num1 = slen[1][gr_info->scalefac_compress];
  539.  
  540.     if (gr_info->block_type == 2) {
  541.       int i=18;
  542.       numbits = (num0 + num1) * 18;
  543.  
  544.       if (gr_info->mixed_block_flag) {
  545.          for (i=8;i;i--)
  546.            *scf++ = getbits_fast(num0);
  547.          i = 9;
  548.          numbits -= num0; /* num0 * 17 + num1 * 18 */
  549.       }
  550.  
  551.       for (;i;i--)
  552.         *scf++ = getbits_fast(num0);
  553.       for (i = 18; i; i--)
  554.         *scf++ = getbits_fast(num1);
  555.       *scf++ = 0; *scf++ = 0; *scf++ = 0; /* short[13][0..2] = 0 */
  556.     }
  557.     else {
  558.       int i;
  559.       int scfsi = gr_info->scfsi;
  560.  
  561.       if(scfsi < 0) { /* scfsi < 0 => granule == 0 */
  562.          for(i=11;i;i--)
  563.            *scf++ = getbits_fast(num0);
  564.          for(i=10;i;i--)
  565.            *scf++ = getbits_fast(num1);
  566.          numbits = (num0 + num1) * 10 + num0;
  567.          *scf++ = 0;
  568.       }
  569.       else {
  570.         numbits = 0;
  571.         if(!(scfsi & 0x8)) {
  572.           for (i=0;i<6;i++)
  573.             *scf++ = getbits_fast(num0);
  574.           numbits += num0 * 6;
  575.         }
  576.         else {
  577.           scf += 6;
  578.         }
  579.  
  580.         if(!(scfsi & 0x4)) {
  581.           for (i=0;i<5;i++)
  582.             *scf++ = getbits_fast(num0);
  583.           numbits += num0 * 5;
  584.         }
  585.         else {
  586.           scf += 5;
  587.         }
  588.  
  589.         if(!(scfsi & 0x2)) {
  590.           for(i=0;i<5;i++)
  591.             *scf++ = getbits_fast(num1);
  592.           numbits += num1 * 5;
  593.         }
  594.         else {
  595.           scf += 5;
  596.         }
  597.  
  598.         if(!(scfsi & 0x1)) {
  599.           for (i=0;i<5;i++)
  600.             *scf++ = getbits_fast(num1);
  601.           numbits += num1 * 5;
  602.         }
  603.         else {
  604.            scf += 5;
  605.         }
  606.         *scf++ = 0;  /* no l[21] in original sources */
  607.       }
  608.     }
  609.     return numbits;
  610. }
  611.  
  612. static int III_get_scale_factors_2(int *scf,struct gr_info_s *gr_info,int i_stereo)
  613. {
  614.   unsigned char *pnt;
  615.   int i,j,n=0,numbits=0;
  616.   unsigned int slen;
  617.  
  618.   static const unsigned char stab[3][6][4] = {
  619.    { { 6, 5, 5,5 } , { 6, 5, 7,3 } , { 11,10,0,0} ,
  620.      { 7, 7, 7,0 } , { 6, 6, 6,3 } , {  8, 8,5,0} } ,
  621.    { { 9, 9, 9,9 } , { 9, 9,12,6 } , { 18,18,0,0} ,
  622.      {12,12,12,0 } , {12, 9, 9,6 } , { 15,12,9,0} } ,
  623.    { { 6, 9, 9,9 } , { 6, 9,12,6 } , { 15,18,0,0} ,
  624.      { 6,15,12,0 } , { 6,12, 9,6 } , {  6,18,9,0} } };
  625.  
  626.   if(i_stereo) /* i_stereo AND second channel -> do_layer3() checks this */
  627.     slen = i_slen2[gr_info->scalefac_compress>>1];
  628.   else
  629.     slen = n_slen2[gr_info->scalefac_compress];
  630.  
  631.   gr_info->preflag = (slen>>15) & 0x1;
  632.  
  633.   n = 0;  
  634.   if( gr_info->block_type == 2 ) {
  635.     n++;
  636.     if(gr_info->mixed_block_flag)
  637.       n++;
  638.   }
  639.  
  640.   pnt = stab[n][(slen>>12)&0x7];
  641.  
  642.   for(i=0;i<4;i++) {
  643.     int num = slen & 0x7;
  644.     slen >>= 3;
  645.     if(num) {
  646.       for(j=0;j<(int)(pnt[i]);j++)
  647.         *scf++ = getbits_fast(num);
  648.       numbits += pnt[i] * num;
  649.     }
  650.     else {
  651.       for(j=0;j<(int)(pnt[i]);j++)
  652.         *scf++ = 0;
  653.     }
  654.   }
  655.  
  656.   n = (n << 1) + 1;
  657.   for(i=0;i<n;i++)
  658.     *scf++ = 0;
  659.  
  660.   return numbits;
  661. }
  662.  
  663. static int pretab1[22] = {0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,2,2,3,3,3,2,0};
  664. static int pretab2[22] = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
  665.  
  666. /*
  667.  * Dequantize samples (includes huffman decoding)
  668.  */
  669. /* 24 is enough because tab13 has max. a 19 bit huffvector */
  670. #define BITSHIFT ((sizeof(long)-1)*8)
  671. #define REFRESH_MASK \
  672.   while(num < BITSHIFT) { \
  673.     mask |= ((unsigned long)getbyte())<<(BITSHIFT-num); \
  674.     num += 8; \
  675.     part2remain -= 8; }
  676.  
  677. static int III_dequantize_sample(real xr[SBLIMIT][SSLIMIT],int *scf,
  678.    struct gr_info_s *gr_info,int sfreq,int part2bits)
  679. {
  680.   int shift = 1 + gr_info->scalefac_scale;
  681.   real *xrpnt = (real *) xr;
  682.   int l[3],l3;
  683.   int part2remain = gr_info->part2_3_length - part2bits;
  684.   int *me;
  685.  
  686.   /* mhipp tree has this split up a bit... */
  687.   int num=getbitoffset();
  688.   long mask = (long) getbits(num)<<(BITSHIFT+8-num);
  689.   part2remain -= num;
  690.  
  691.   {
  692.     int bv       = gr_info->big_values;
  693.     int region1  = gr_info->region1start;
  694.     int region2  = gr_info->region2start;
  695. if(region1 > region2)
  696. {
  697.         return 1;
  698. }
  699.     l3 = ((576>>1)-bv)>>1;  
  700. /*
  701.  * we may lose the 'odd' bit here !!
  702.  * check this later again
  703.  */
  704.     if(bv <= region1) {
  705.       l[0] = bv; l[1] = 0; l[2] = 0;
  706.     }
  707.     else {
  708.       l[0] = region1;
  709.       if(bv <= region2) {
  710.         l[1] = bv - l[0];  l[2] = 0;
  711.       }
  712.       else {
  713.         l[1] = region2 - l[0]; l[2] = bv - region2;
  714.       }
  715.     }
  716.   }
  717.  
  718.   if(gr_info->block_type == 2) {
  719.     /*
  720.      * decoding with short or mixed mode BandIndex table
  721.      */
  722.     int i,max[4];
  723.     int step=0,lwin=3,cb=0;
  724.     register real v = 0.0;
  725.     register int *m,mc;
  726.  
  727.     if(gr_info->mixed_block_flag) {
  728.       max[3] = -1;
  729.       max[0] = max[1] = max[2] = 2;
  730.       m = map[sfreq][0];
  731.       me = mapend[sfreq][0];
  732.     }
  733.     else {
  734.       max[0] = max[1] = max[2] = max[3] = -1;
  735.       /* max[3] not really needed in this case */
  736.       m = map[sfreq][1];
  737.       me = mapend[sfreq][1];
  738.     }
  739.  
  740.     mc = 0;
  741.     for(i=0;i<2;i++) {
  742.       int lp = l[i];
  743.       struct newhuff *h = ht+gr_info->table_select[i];
  744.       for(;lp;lp--,mc--) {
  745.         register int x,y;
  746.         if( (!mc) ) {
  747.           mc    = *m++;
  748.           xrpnt = ((real *) xr) + (*m++);
  749.           lwin  = *m++;
  750.           cb    = *m++;
  751.           if(lwin == 3) {
  752.             v = gr_info->pow2gain[(*scf++) << shift];
  753.             step = 1;
  754.           }
  755.           else {
  756.             v = gr_info->full_gain[lwin][(*scf++) << shift];
  757.             step = 3;
  758.           }
  759.         }
  760.         {
  761.           register short *val = h->table;
  762.           REFRESH_MASK;
  763.           while((y=*val++)<0) {
  764.             if (mask < 0)
  765.               val -= y;
  766.             num--;
  767.             mask <<= 1;
  768.           }
  769.           x = y >> 4;
  770.           y &= 0xf;
  771.         }
  772.         if(x == 15 && h->linbits) {
  773.           max[lwin] = cb;
  774.           REFRESH_MASK;
  775.           x += ((unsigned long) mask) >> (BITSHIFT+8-h->linbits);
  776.           num -= h->linbits+1;
  777.           mask <<= h->linbits;
  778.           if(mask < 0)
  779.             *xrpnt = REAL_MUL(-ispow[x], v);
  780.           else
  781.             *xrpnt = REAL_MUL(ispow[x], v);
  782.           mask <<= 1;
  783.         }
  784.         else if(x) {
  785.           max[lwin] = cb;
  786.           if(mask < 0)
  787.             *xrpnt = REAL_MUL(-ispow[x], v);
  788.           else
  789.             *xrpnt = REAL_MUL(ispow[x], v);
  790.           num--;
  791.           mask <<= 1;
  792.         }
  793.         else
  794.           *xrpnt = DOUBLE_TO_REAL(0.0);
  795.         xrpnt += step;
  796.         if(y == 15 && h->linbits) {
  797.           max[lwin] = cb;
  798.           REFRESH_MASK;
  799.           y += ((unsigned long) mask) >> (BITSHIFT+8-h->linbits);
  800.           num -= h->linbits+1;
  801.           mask <<= h->linbits;
  802.           if(mask < 0)
  803.             *xrpnt = REAL_MUL(-ispow[y], v);
  804.           else
  805.             *xrpnt = REAL_MUL(ispow[y], v);
  806.           mask <<= 1;
  807.         }
  808.         else if(y) {
  809.           max[lwin] = cb;
  810.           if(mask < 0)
  811.             *xrpnt = REAL_MUL(-ispow[y], v);
  812.           else
  813.             *xrpnt = REAL_MUL(ispow[y], v);
  814.           num--;
  815.           mask <<= 1;
  816.         }
  817.         else
  818.           *xrpnt = DOUBLE_TO_REAL(0.0);
  819.         xrpnt += step;
  820.       }
  821.     }
  822.  
  823.     for(;l3 && (part2remain+num > 0);l3--) {
  824.       /* not mixing code and declarations to keep C89 happy */
  825.       struct newhuff* h;
  826.       register short* val;
  827.                         register short a;
  828.       /* This is only a humble hack to prevent a special segfault. */
  829.       /* More insight into the real workings is still needed. */
  830.       /* especially why there are (valid?) files that make xrpnt exceed the array with 4 bytes without segfaulting, more seems to be really bad, though. */
  831.       #ifdef DEBUG
  832.       if(!(xrpnt < &xr[SBLIMIT][0]))
  833.       {
  834.         if(param.verbose) debug2("attempted soft xrpnt overflow (%p !< %p) ?", (void*) xrpnt, (void*) &xr[SBLIMIT][0]);
  835.       }
  836.       #endif
  837.       if(!(xrpnt < &xr[SBLIMIT][0]+5))
  838.       {
  839.         return 2;
  840.       }
  841.       h = htc+gr_info->count1table_select;
  842.       val = h->table;
  843.  
  844.       REFRESH_MASK;
  845.       while((a=*val++)<0) {
  846.         if (mask < 0)
  847.           val -= a;
  848.         num--;
  849.         mask <<= 1;
  850.       }
  851.       if(part2remain+num <= 0) {
  852.         num -= part2remain+num;
  853.         break;
  854.       }
  855.  
  856.       for(i=0;i<4;i++) {
  857.         if(!(i & 1)) {
  858.           if(!mc) {
  859.             mc = *m++;
  860.             xrpnt = ((real *) xr) + (*m++);
  861.             lwin = *m++;
  862.             cb = *m++;
  863.             if(lwin == 3) {
  864.               v = gr_info->pow2gain[(*scf++) << shift];
  865.               step = 1;
  866.             }
  867.             else {
  868.               v = gr_info->full_gain[lwin][(*scf++) << shift];
  869.               step = 3;
  870.             }
  871.           }
  872.           mc--;
  873.         }
  874.         if( (a & (0x8>>i)) ) {
  875.           max[lwin] = cb;
  876.           if(part2remain+num <= 0) {
  877.             break;
  878.           }
  879.           if(mask < 0)
  880.             *xrpnt = -v;
  881.           else
  882.             *xrpnt = v;
  883.           num--;
  884.           mask <<= 1;
  885.         }
  886.         else
  887.           *xrpnt = DOUBLE_TO_REAL(0.0);
  888.         xrpnt += step;
  889.       }
  890.     }
  891.  
  892.     if(lwin < 3) { /* short band? */
  893.       while(1) {
  894.         for(;mc > 0;mc--) {
  895.           *xrpnt = DOUBLE_TO_REAL(0.0); xrpnt += 3; /* short band -> step=3 */
  896.           *xrpnt = DOUBLE_TO_REAL(0.0); xrpnt += 3;
  897.         }
  898.         if(m >= me)
  899.           break;
  900.         mc    = *m++;
  901.         xrpnt = ((real *) xr) + *m++;
  902.         if(*m++ == 0)
  903.           break; /* optimize: field will be set to zero at the end of the function */
  904.         m++; /* cb */
  905.       }
  906.     }
  907.  
  908.     gr_info->maxband[0] = max[0]+1;
  909.     gr_info->maxband[1] = max[1]+1;
  910.     gr_info->maxband[2] = max[2]+1;
  911.     gr_info->maxbandl = max[3]+1;
  912.  
  913.     {
  914.       int rmax = max[0] > max[1] ? max[0] : max[1];
  915.       rmax = (rmax > max[2] ? rmax : max[2]) + 1;
  916.       gr_info->maxb = rmax ? shortLimit[sfreq][rmax] : longLimit[sfreq][max[3]+1];
  917.     }
  918.  
  919.   }
  920.   else {
  921.     /*
  922.      * decoding with 'long' BandIndex table (block_type != 2)
  923.      */
  924.     int *pretab = gr_info->preflag ? pretab1 : pretab2;
  925.     int i,max = -1;
  926.     int cb = 0;
  927.     int *m = map[sfreq][2];
  928.     register real v = 0.0;
  929.     int mc = 0;
  930.  
  931.     /*
  932.      * long hash table values
  933.      */
  934.     for(i=0;i<3;i++) {
  935.       int lp = l[i];
  936.       struct newhuff *h = ht+gr_info->table_select[i];
  937.  
  938.       for(;lp;lp--,mc--) {
  939.         int x,y;
  940.         if(!mc) {
  941.           mc = *m++;
  942.           cb = *m++;
  943.           if(cb == 21)
  944.             v = 0.0;
  945.           else
  946.             v = gr_info->pow2gain[((*scf++) + (*pretab++)) << shift];
  947.  
  948.         }
  949.         {
  950.           register short *val = h->table;
  951.           REFRESH_MASK;
  952.           while((y=*val++)<0) {
  953.             if (mask < 0)
  954.               val -= y;
  955.             num--;
  956.             mask <<= 1;
  957.           }
  958.           x = y >> 4;
  959.           y &= 0xf;
  960.         }
  961.  
  962.         if (x == 15 && h->linbits) {
  963.           max = cb;
  964.           REFRESH_MASK;
  965.           x += ((unsigned long) mask) >> (BITSHIFT+8-h->linbits);
  966.           num -= h->linbits+1;
  967.           mask <<= h->linbits;
  968.           if(mask < 0)
  969.             *xrpnt++ = REAL_MUL(-ispow[x], v);
  970.           else
  971.             *xrpnt++ = REAL_MUL(ispow[x], v);
  972.           mask <<= 1;
  973.         }
  974.         else if(x) {
  975.           max = cb;
  976.           if(mask < 0)
  977.             *xrpnt++ = REAL_MUL(-ispow[x], v);
  978.           else
  979.             *xrpnt++ = REAL_MUL(ispow[x], v);
  980.           num--;
  981.           mask <<= 1;
  982.         }
  983.         else
  984.           *xrpnt++ = DOUBLE_TO_REAL(0.0);
  985.  
  986.         if (y == 15 && h->linbits) {
  987.           max = cb;
  988.           REFRESH_MASK;
  989.           y += ((unsigned long) mask) >> (BITSHIFT+8-h->linbits);
  990.           num -= h->linbits+1;
  991.           mask <<= h->linbits;
  992.           if(mask < 0)
  993.             *xrpnt++ = REAL_MUL(-ispow[y], v);
  994.           else
  995.             *xrpnt++ = REAL_MUL(ispow[y], v);
  996.           mask <<= 1;
  997.         }
  998.         else if(y) {
  999.           max = cb;
  1000.           if(mask < 0)
  1001.             *xrpnt++ = REAL_MUL(-ispow[y], v);
  1002.           else
  1003.             *xrpnt++ = REAL_MUL(ispow[y], v);
  1004.           num--;
  1005.           mask <<= 1;
  1006.         }
  1007.         else
  1008.           *xrpnt++ = DOUBLE_TO_REAL(0.0);
  1009.       }
  1010.     }
  1011.  
  1012.     /*
  1013.      * short (count1table) values
  1014.      */
  1015.     for(;l3 && (part2remain+num > 0);l3--) {
  1016.       struct newhuff *h = htc+gr_info->count1table_select;
  1017.       register short *val = h->table,a;
  1018.  
  1019.       REFRESH_MASK;
  1020.       while((a=*val++)<0) {
  1021.         if (mask < 0)
  1022.           val -= a;
  1023.         num--;
  1024.         mask <<= 1;
  1025.       }
  1026.       if(part2remain+num <= 0) {
  1027.         num -= part2remain+num;
  1028.         break;
  1029.       }
  1030.  
  1031.       for(i=0;i<4;i++) {
  1032.         if(!(i & 1)) {
  1033.           if(!mc) {
  1034.             mc = *m++;
  1035.             cb = *m++;
  1036.             if(cb == 21)
  1037.               v = 0.0;
  1038.             else
  1039.               v = gr_info->pow2gain[((*scf++) + (*pretab++)) << shift];
  1040.           }
  1041.           mc--;
  1042.         }
  1043.         if ( (a & (0x8>>i)) ) {
  1044.           max = cb;
  1045.           if(part2remain+num <= 0) {
  1046.             break;
  1047.           }
  1048.           if(mask < 0)
  1049.             *xrpnt++ = -v;
  1050.           else
  1051.             *xrpnt++ = v;
  1052.           num--;
  1053.           mask <<= 1;
  1054.         }
  1055.         else
  1056.           *xrpnt++ = DOUBLE_TO_REAL(0.0);
  1057.       }
  1058.     }
  1059.  
  1060.     gr_info->maxbandl = max+1;
  1061.     gr_info->maxb = longLimit[sfreq][gr_info->maxbandl];
  1062.   }
  1063.  
  1064.   part2remain += num;
  1065.   backbits(num);
  1066.   num = 0;
  1067.  
  1068.   while(xrpnt < &xr[SBLIMIT][0])
  1069.     *xrpnt++ = DOUBLE_TO_REAL(0.0);
  1070.  
  1071.   while( part2remain > 16 ) {
  1072.     getbits(16); /* Dismiss stuffing Bits */
  1073.     part2remain -= 16;
  1074.   }
  1075.   if(part2remain > 0)
  1076.     getbits(part2remain);
  1077.   else if(part2remain < 0) {
  1078.     return 1; /* -> error */
  1079.   }
  1080.   return 0;
  1081. }
  1082.  
  1083. /*
  1084.  * III_stereo: calculate real channel values for Joint-I-Stereo-mode
  1085.  */
  1086. static void III_i_stereo(real xr_buf[2][SBLIMIT][SSLIMIT],int *scalefac,
  1087.    struct gr_info_s *gr_info,int sfreq,int ms_stereo,int lsf)
  1088. {
  1089.       real (*xr)[SBLIMIT*SSLIMIT] = (real (*)[SBLIMIT*SSLIMIT] ) xr_buf;
  1090.       struct bandInfoStruct *bi = &bandInfo[sfreq];
  1091.  
  1092.       const real *tab1,*tab2;
  1093.  
  1094. #if 1
  1095.       int tab;
  1096. /* TODO: optimize as static */
  1097.       static const real *tabs[3][2][2] = {
  1098.          { { tan1_1,tan2_1 }     , { tan1_2,tan2_2 } },
  1099.          { { pow1_1[0],pow2_1[0] } , { pow1_2[0],pow2_2[0] } } ,
  1100.          { { pow1_1[1],pow2_1[1] } , { pow1_2[1],pow2_2[1] } }
  1101.       };
  1102.  
  1103.       tab = lsf + (gr_info->scalefac_compress & lsf);
  1104.       tab1 = tabs[tab][ms_stereo][0];
  1105.       tab2 = tabs[tab][ms_stereo][1];
  1106. #else
  1107.       if(lsf) {
  1108.         int p = gr_info->scalefac_compress & 0x1;
  1109.         if(ms_stereo) {
  1110.           tab1 = pow1_2[p]; tab2 = pow2_2[p];
  1111.         }
  1112.         else {
  1113.           tab1 = pow1_1[p]; tab2 = pow2_1[p];
  1114.         }
  1115.       }
  1116.       else {
  1117.         if(ms_stereo) {
  1118.           tab1 = tan1_2; tab2 = tan2_2;
  1119.         }
  1120.         else {
  1121.           tab1 = tan1_1; tab2 = tan2_1;
  1122.         }
  1123.       }
  1124. #endif
  1125.  
  1126.       if (gr_info->block_type == 2) {
  1127.          int lwin,do_l = 0;
  1128.          if( gr_info->mixed_block_flag )
  1129.            do_l = 1;
  1130.  
  1131.          for (lwin=0;lwin<3;lwin++) { /* process each window */
  1132.              /* get first band with zero values */
  1133.            int is_p,sb,idx,sfb = gr_info->maxband[lwin];  /* sfb is minimal 3 for mixed mode */
  1134.            if(sfb > 3)
  1135.              do_l = 0;
  1136.  
  1137.            for(;sfb<12;sfb++) {
  1138.              is_p = scalefac[sfb*3+lwin-gr_info->mixed_block_flag]; /* scale: 0-15 */
  1139.              if(is_p != 7) {
  1140.                real t1,t2;
  1141.                sb  = bi->shortDiff[sfb];
  1142.                idx = bi->shortIdx[sfb] + lwin;
  1143.                t1  = tab1[is_p]; t2 = tab2[is_p];
  1144.                for (; sb > 0; sb--,idx+=3) {
  1145.                  real v = xr[0][idx];
  1146.                  xr[0][idx] = REAL_MUL(v, t1);
  1147.                  xr[1][idx] = REAL_MUL(v, t2);
  1148.                }
  1149.              }
  1150.            }
  1151.  
  1152. #if 1
  1153. /* in the original: copy 10 to 11 , here: copy 11 to 12
  1154. maybe still wrong??? (copy 12 to 13?) */
  1155.            is_p = scalefac[11*3+lwin-gr_info->mixed_block_flag]; /* scale: 0-15 */
  1156.            sb   = bi->shortDiff[12];
  1157.            idx  = bi->shortIdx[12] + lwin;
  1158. #else
  1159.            is_p = scalefac[10*3+lwin-gr_info->mixed_block_flag]; /* scale: 0-15 */
  1160.            sb   = bi->shortDiff[11];
  1161.            idx  = bi->shortIdx[11] + lwin;
  1162. #endif
  1163.            if(is_p != 7) {
  1164.              real t1,t2;
  1165.              t1 = tab1[is_p]; t2 = tab2[is_p];
  1166.              for ( ; sb > 0; sb--,idx+=3 ) {  
  1167.                real v = xr[0][idx];
  1168.                xr[0][idx] = REAL_MUL(v, t1);
  1169.                xr[1][idx] = REAL_MUL(v, t2);
  1170.              }
  1171.            }
  1172.          } /* end for(lwin; .. ; . ) */
  1173.  
  1174. /* also check l-part, if ALL bands in the three windows are 'empty'
  1175.  * and mode = mixed_mode
  1176.  */
  1177.          if (do_l) {
  1178.            int sfb = gr_info->maxbandl;
  1179.            int idx;
  1180.            if(sfb > 21) return; /* similarity fix related to CVE-2006-1655 */
  1181.            idx = bi->longIdx[sfb];
  1182.            for ( ; sfb<8; sfb++ ) {
  1183.              int sb = bi->longDiff[sfb];
  1184.              int is_p = scalefac[sfb]; /* scale: 0-15 */
  1185.              if(is_p != 7) {
  1186.                real t1,t2;
  1187.                t1 = tab1[is_p]; t2 = tab2[is_p];
  1188.                for ( ; sb > 0; sb--,idx++) {
  1189.                  real v = xr[0][idx];
  1190.                  xr[0][idx] = REAL_MUL(v, t1);
  1191.                  xr[1][idx] = REAL_MUL(v, t2);
  1192.                }
  1193.              }
  1194.              else
  1195.                idx += sb;
  1196.            }
  1197.          }    
  1198.       }
  1199.       else { /* ((gr_info->block_type != 2)) */
  1200.         int sfb = gr_info->maxbandl;
  1201.         int is_p,idx;
  1202.         if(sfb > 21) return; /* tightened fix for CVE-2006-1655 */
  1203.         idx = bi->longIdx[sfb];
  1204.         for ( ; sfb<21; sfb++) {
  1205.           int sb = bi->longDiff[sfb];
  1206.           is_p = scalefac[sfb]; /* scale: 0-15 */
  1207.           if(is_p != 7) {
  1208.             real t1,t2;
  1209.             t1 = tab1[is_p]; t2 = tab2[is_p];
  1210.             for ( ; sb > 0; sb--,idx++) {
  1211.                real v = xr[0][idx];
  1212.                xr[0][idx] = REAL_MUL(v, t1);
  1213.                xr[1][idx] = REAL_MUL(v, t2);
  1214.             }
  1215.           }
  1216.           else
  1217.             idx += sb;
  1218.         }
  1219.  
  1220.         is_p = scalefac[20];
  1221.         if(is_p != 7) {  /* copy l-band 20 to l-band 21 */
  1222.           int sb;
  1223.           real t1 = tab1[is_p],t2 = tab2[is_p];
  1224.  
  1225.           for ( sb = bi->longDiff[21]; sb > 0; sb--,idx++ ) {
  1226.             real v = xr[0][idx];
  1227.             xr[0][idx] = REAL_MUL(v, t1);
  1228.             xr[1][idx] = REAL_MUL(v, t2);
  1229.           }
  1230.         }
  1231.       } /* ... */
  1232. }
  1233.  
  1234. static void III_antialias(real xr[SBLIMIT][SSLIMIT],struct gr_info_s *gr_info) {
  1235.    int sblim;
  1236.  
  1237.    if(gr_info->block_type == 2) {
  1238.       if(!gr_info->mixed_block_flag)
  1239.         return;
  1240.       sblim = 1;
  1241.    }
  1242.    else {
  1243.      sblim = gr_info->maxb-1;
  1244.    }
  1245.  
  1246.    /* 31 alias-reduction operations between each pair of sub-bands */
  1247.    /* with 8 butterflies between each pair                         */
  1248.  
  1249.    {
  1250.      int sb;
  1251.      real *xr1=(real *) xr[1];
  1252.  
  1253.      for(sb=sblim;sb;sb--,xr1+=10) {
  1254.        int ss;
  1255.        real *cs=aa_cs,*ca=aa_ca;
  1256.        real *xr2 = xr1;
  1257.  
  1258.        for(ss=7;ss>=0;ss--)
  1259.        {       /* upper and lower butterfly inputs */
  1260.          register real bu = *--xr2,bd = *xr1;
  1261.         *xr2   = REAL_MUL(bu, *cs) - REAL_MUL(bd, *ca);
  1262.         *xr1++ = REAL_MUL(bd, *cs++) + REAL_MUL(bu, *ca++);
  1263.        }
  1264.      }
  1265.   }
  1266. }
  1267.  
  1268. /*
  1269. // This is an optimized DCT from Jeff Tsay's maplay 1.2+ package.
  1270. // Saved one multiplication by doing the 'twiddle factor' stuff
  1271. // together with the window mul. (MH)
  1272. //
  1273. // This uses Byeong Gi Lee's Fast Cosine Transform algorithm, but the
  1274. // 9 point IDCT needs to be reduced further. Unfortunately, I don't
  1275. // know how to do that, because 9 is not an even number. - Jeff.
  1276. //
  1277. //////////////////////////////////////////////////////////////////
  1278. //
  1279. // 9 Point Inverse Discrete Cosine Transform
  1280. //
  1281. // This piece of code is Copyright 1997 Mikko Tommila and is freely usable
  1282. // by anybody. The algorithm itself is of course in the public domain.
  1283. //
  1284. // Again derived heuristically from the 9-point WFTA.
  1285. //
  1286. // The algorithm is optimized (?) for speed, not for small rounding errors or
  1287. // good readability.
  1288. //
  1289. // 36 additions, 11 multiplications
  1290. //
  1291. // Again this is very likely sub-optimal.
  1292. //
  1293. // The code is optimized to use a minimum number of temporary variables,
  1294. // so it should compile quite well even on 8-register Intel x86 processors.
  1295. // This makes the code quite obfuscated and very difficult to understand.
  1296. //
  1297. // References:
  1298. // [1] S. Winograd: "On Computing the Discrete Fourier Transform",
  1299. //     Mathematics of Computation, Volume 32, Number 141, January 1978,
  1300. //     Pages 175-199
  1301. */
  1302.  
  1303. /*------------------------------------------------------------------*/
  1304. /*                                                                  */
  1305. /*    Function: Calculation of the inverse MDCT                     */
  1306. /*                                                                  */
  1307. /*------------------------------------------------------------------*/
  1308. #ifdef USE_3DNOW
  1309. void dct36(real *inbuf,real *o1,real *o2,real *wintab,real *tsbuf)
  1310. #else
  1311. static void dct36(real *inbuf,real *o1,real *o2,real *wintab,real *tsbuf)
  1312. #endif
  1313. {
  1314. #ifdef NEW_DCT9
  1315.   real tmp[18];
  1316. #endif
  1317.  
  1318.   {
  1319.     register real *in = inbuf;
  1320.  
  1321.     in[17]+=in[16]; in[16]+=in[15]; in[15]+=in[14];
  1322.     in[14]+=in[13]; in[13]+=in[12]; in[12]+=in[11];
  1323.     in[11]+=in[10]; in[10]+=in[9];  in[9] +=in[8];
  1324.     in[8] +=in[7];  in[7] +=in[6];  in[6] +=in[5];
  1325.     in[5] +=in[4];  in[4] +=in[3];  in[3] +=in[2];
  1326.     in[2] +=in[1];  in[1] +=in[0];
  1327.  
  1328.     in[17]+=in[15]; in[15]+=in[13]; in[13]+=in[11]; in[11]+=in[9];
  1329.     in[9] +=in[7];  in[7] +=in[5];  in[5] +=in[3];  in[3] +=in[1];
  1330.  
  1331.  
  1332. #ifdef NEW_DCT9
  1333. #if 1
  1334.     {
  1335.      real t3;
  1336.      {
  1337.       real t0, t1, t2;
  1338.  
  1339.       t0 = REAL_MUL(COS6_2, (in[8] + in[16] - in[4]));
  1340.       t1 = REAL_MUL(COS6_2, in[12]);
  1341.  
  1342.       t3 = in[0];
  1343.       t2 = t3 - t1 - t1;
  1344.       tmp[1] = tmp[7] = t2 - t0;
  1345.       tmp[4]          = t2 + t0 + t0;
  1346.       t3 += t1;
  1347.  
  1348.       t2 = REAL_MUL(COS6_1, (in[10] + in[14] - in[2]));
  1349.       tmp[1] -= t2;
  1350.       tmp[7] += t2;
  1351.      }
  1352.      {
  1353.       real t0, t1, t2;
  1354.  
  1355.       t0 = REAL_MUL(cos9[0], (in[4] + in[8] ));
  1356.       t1 = REAL_MUL(cos9[1], (in[8] - in[16]));
  1357.       t2 = REAL_MUL(cos9[2], (in[4] + in[16]));
  1358.  
  1359.       tmp[2] = tmp[6] = t3 - t0      - t2;
  1360.       tmp[0] = tmp[8] = t3 + t0 + t1;
  1361.       tmp[3] = tmp[5] = t3      - t1 + t2;
  1362.      }
  1363.     }
  1364.     {
  1365.       real t1, t2, t3;
  1366.  
  1367.       t1 = REAL_MUL(cos18[0], (in[2]  + in[10]));
  1368.       t2 = REAL_MUL(cos18[1], (in[10] - in[14]));
  1369.       t3 = REAL_MUL(COS6_1,    in[6]);
  1370.  
  1371.       {
  1372.         real t0 = t1 + t2 + t3;
  1373.         tmp[0] += t0;
  1374.         tmp[8] -= t0;
  1375.       }
  1376.  
  1377.       t2 -= t3;
  1378.       t1 -= t3;
  1379.  
  1380.       t3 = REAL_MUL(cos18[2], (in[2] + in[14]));
  1381.  
  1382.       t1 += t3;
  1383.       tmp[3] += t1;
  1384.       tmp[5] -= t1;
  1385.  
  1386.       t2 -= t3;
  1387.       tmp[2] += t2;
  1388.       tmp[6] -= t2;
  1389.     }
  1390.  
  1391. #else
  1392.     {
  1393.       real t0, t1, t2, t3, t4, t5, t6, t7;
  1394.  
  1395.       t1 = REAL_MUL(COS6_2, in[12]);
  1396.       t2 = REAL_MUL(COS6_2, (in[8] + in[16] - in[4]));
  1397.  
  1398.       t3 = in[0] + t1;
  1399.       t4 = in[0] - t1 - t1;
  1400.       t5     = t4 - t2;
  1401.       tmp[4] = t4 + t2 + t2;
  1402.  
  1403.       t0 = REAL_MUL(cos9[0], (in[4] + in[8]));
  1404.       t1 = REAL_MUL(cos9[1], (in[8] - in[16]));
  1405.  
  1406.       t2 = REAL_MUL(cos9[2], (in[4] + in[16]));
  1407.  
  1408.       t6 = t3 - t0 - t2;
  1409.       t0 += t3 + t1;
  1410.       t3 += t2 - t1;
  1411.  
  1412.       t2 = REAL_MUL(cos18[0], (in[2]  + in[10]));
  1413.       t4 = REAL_MUL(cos18[1], (in[10] - in[14]));
  1414.       t7 = REAL_MUL(COS6_1, in[6]);
  1415.  
  1416.       t1 = t2 + t4 + t7;
  1417.       tmp[0] = t0 + t1;
  1418.       tmp[8] = t0 - t1;
  1419.       t1 = REAL_MUL(cos18[2], (in[2] + in[14]));
  1420.       t2 += t1 - t7;
  1421.  
  1422.       tmp[3] = t3 + t2;
  1423.       t0 = REAL_MUL(COS6_1, (in[10] + in[14] - in[2]));
  1424.       tmp[5] = t3 - t2;
  1425.  
  1426.       t4 -= t1 + t7;
  1427.  
  1428.       tmp[1] = t5 - t0;
  1429.       tmp[7] = t5 + t0;
  1430.       tmp[2] = t6 + t4;
  1431.       tmp[6] = t6 - t4;
  1432.     }
  1433. #endif
  1434.  
  1435.     {
  1436.       real t0, t1, t2, t3, t4, t5, t6, t7;
  1437.  
  1438.       t1 = REAL_MUL(COS6_2, in[13]);
  1439.       t2 = REAL_MUL(COS6_2, (in[9] + in[17] - in[5]));
  1440.  
  1441.       t3 = in[1] + t1;
  1442.       t4 = in[1] - t1 - t1;
  1443.       t5 = t4 - t2;
  1444.  
  1445.       t0 = REAL_MUL(cos9[0], (in[5] + in[9]));
  1446.       t1 = REAL_MUL(cos9[1], (in[9] - in[17]));
  1447.  
  1448.       tmp[13] = REAL_MUL((t4 + t2 + t2), tfcos36[17-13]);
  1449.       t2 = REAL_MUL(cos9[2], (in[5] + in[17]));
  1450.  
  1451.       t6 = t3 - t0 - t2;
  1452.       t0 += t3 + t1;
  1453.       t3 += t2 - t1;
  1454.  
  1455.       t2 = REAL_MUL(cos18[0], (in[3]  + in[11]));
  1456.       t4 = REAL_MUL(cos18[1], (in[11] - in[15]));
  1457.       t7 = REAL_MUL(COS6_1, in[7]);
  1458.  
  1459.       t1 = t2 + t4 + t7;
  1460.       tmp[17] = REAL_MUL((t0 + t1), tfcos36[17-17]);
  1461.       tmp[9]  = REAL_MUL((t0 - t1), tfcos36[17-9]);
  1462.       t1 = REAL_MUL(cos18[2], (in[3] + in[15]));
  1463.       t2 += t1 - t7;
  1464.  
  1465.       tmp[14] = REAL_MUL((t3 + t2), tfcos36[17-14]);
  1466.       t0 = REAL_MUL(COS6_1, (in[11] + in[15] - in[3]));
  1467.       tmp[12] = REAL_MUL((t3 - t2), tfcos36[17-12]);
  1468.  
  1469.       t4 -= t1 + t7;
  1470.  
  1471.       tmp[16] = REAL_MUL((t5 - t0), tfcos36[17-16]);
  1472.       tmp[10] = REAL_MUL((t5 + t0), tfcos36[17-10]);
  1473.       tmp[15] = REAL_MUL((t6 + t4), tfcos36[17-15]);
  1474.       tmp[11] = REAL_MUL((t6 - t4), tfcos36[17-11]);
  1475.    }
  1476.  
  1477. #define MACRO(v) { \
  1478.     real tmpval; \
  1479.     tmpval = tmp[(v)] + tmp[17-(v)]; \
  1480.     out2[9+(v)] = REAL_MUL(tmpval, w[27+(v)]); \
  1481.     out2[8-(v)] = REAL_MUL(tmpval, w[26-(v)]); \
  1482.     tmpval = tmp[(v)] - tmp[17-(v)]; \
  1483.     ts[SBLIMIT*(8-(v))] = out1[8-(v)] + REAL_MUL(tmpval, w[8-(v)]); \
  1484.     ts[SBLIMIT*(9+(v))] = out1[9+(v)] + REAL_MUL(tmpval, w[9+(v)]); }
  1485.  
  1486. {
  1487.    register real *out2 = o2;
  1488.    register real *w = wintab;
  1489.    register real *out1 = o1;
  1490.    register real *ts = tsbuf;
  1491.  
  1492.    MACRO(0);
  1493.    MACRO(1);
  1494.    MACRO(2);
  1495.    MACRO(3);
  1496.    MACRO(4);
  1497.    MACRO(5);
  1498.    MACRO(6);
  1499.    MACRO(7);
  1500.    MACRO(8);
  1501. }
  1502.  
  1503. #else
  1504.  
  1505.   {
  1506.  
  1507. #define MACRO0(v) { \
  1508.     real tmp; \
  1509.     out2[9+(v)] = REAL_MUL((tmp = sum0 + sum1), w[27+(v)]); \
  1510.     out2[8-(v)] = REAL_MUL(tmp, w[26-(v)]);   } \
  1511.     sum0 -= sum1; \
  1512.     ts[SBLIMIT*(8-(v))] = out1[8-(v)] + REAL_MUL(sum0, w[8-(v)]); \
  1513.     ts[SBLIMIT*(9+(v))] = out1[9+(v)] + REAL_MUL(sum0, w[9+(v)]);
  1514. #define MACRO1(v) { \
  1515.         real sum0,sum1; \
  1516.     sum0 = tmp1a + tmp2a; \
  1517.         sum1 = REAL_MUL((tmp1b + tmp2b), tfcos36[(v)]); \
  1518.         MACRO0(v); }
  1519. #define MACRO2(v) { \
  1520.     real sum0,sum1; \
  1521.     sum0 = tmp2a - tmp1a; \
  1522.     sum1 = REAL_MUL((tmp2b - tmp1b), tfcos36[(v)]); \
  1523.         MACRO0(v); }
  1524.  
  1525.     register const real *c = COS9;
  1526.     register real *out2 = o2;
  1527.         register real *w = wintab;
  1528.         register real *out1 = o1;
  1529.         register real *ts = tsbuf;
  1530.  
  1531.     real ta33,ta66,tb33,tb66;
  1532.  
  1533.     ta33 = REAL_MUL(in[2*3+0], c[3]);
  1534.     ta66 = REAL_MUL(in[2*6+0], c[6]);
  1535.     tb33 = REAL_MUL(in[2*3+1], c[3]);
  1536.     tb66 = REAL_MUL(in[2*6+1], c[6]);
  1537.  
  1538.     {
  1539.       real tmp1a,tmp2a,tmp1b,tmp2b;
  1540.       tmp1a = REAL_MUL(in[2*1+0], c[1]) + ta33 + REAL_MUL(in[2*5+0], c[5]) + REAL_MUL(in[2*7+0], c[7]);
  1541.       tmp1b = REAL_MUL(in[2*1+1], c[1]) + tb33 + REAL_MUL(in[2*5+1], c[5]) + REAL_MUL(in[2*7+1], c[7]);
  1542.       tmp2a = REAL_MUL(in[2*2+0], c[2]) + REAL_MUL(in[2*4+0], c[4]) + ta66 + REAL_MUL(in[2*8+0], c[8]);
  1543.       tmp2b = REAL_MUL(in[2*2+1], c[2]) + REAL_MUL(in[2*4+1], c[4]) + tb66 + REAL_MUL(in[2*8+1], c[8]);
  1544.  
  1545.       MACRO1(0);
  1546.       MACRO2(8);
  1547.     }
  1548.  
  1549.     {
  1550.       real tmp1a,tmp2a,tmp1b,tmp2b;
  1551.       tmp1a = REAL_MUL(( in[2*1+0] - in[2*5+0] - in[2*7+0] ), c[3]);
  1552.       tmp1b = REAL_MUL(( in[2*1+1] - in[2*5+1] - in[2*7+1] ), c[3]);
  1553.       tmp2a = REAL_MUL(( in[2*2+0] - in[2*4+0] - in[2*8+0] ), c[6]) - in[2*6+0] + in[2*0+0];
  1554.       tmp2b = REAL_MUL(( in[2*2+1] - in[2*4+1] - in[2*8+1] ), c[6]) - in[2*6+1] + in[2*0+1];
  1555.  
  1556.       MACRO1(1);
  1557.       MACRO2(7);
  1558.     }
  1559.  
  1560.     {
  1561.       real tmp1a,tmp2a,tmp1b,tmp2b;
  1562.       tmp1a =   REAL_MUL(in[2*1+0], c[5]) - ta33 - REAL_MUL(in[2*5+0], c[7]) + REAL_MUL(in[2*7+0], c[1]);
  1563.       tmp1b =   REAL_MUL(in[2*1+1], c[5]) - tb33 - REAL_MUL(in[2*5+1], c[7]) + REAL_MUL(in[2*7+1], c[1]);
  1564.       tmp2a = - REAL_MUL(in[2*2+0], c[8]) - REAL_MUL(in[2*4+0], c[2]) + ta66 + REAL_MUL(in[2*8+0], c[4]);
  1565.       tmp2b = - REAL_MUL(in[2*2+1], c[8]) - REAL_MUL(in[2*4+1], c[2]) + tb66 + REAL_MUL(in[2*8+1], c[4]);
  1566.  
  1567.       MACRO1(2);
  1568.       MACRO2(6);
  1569.     }
  1570.  
  1571.     {
  1572.       real tmp1a,tmp2a,tmp1b,tmp2b;
  1573.       tmp1a =   REAL_MUL(in[2*1+0], c[7]) - ta33 + REAL_MUL(in[2*5+0], c[1]) - REAL_MUL(in[2*7+0], c[5]);
  1574.       tmp1b =   REAL_MUL(in[2*1+1], c[7]) - tb33 + REAL_MUL(in[2*5+1], c[1]) - REAL_MUL(in[2*7+1], c[5]);
  1575.       tmp2a = - REAL_MUL(in[2*2+0], c[4]) + REAL_MUL(in[2*4+0], c[8]) + ta66 - REAL_MUL(in[2*8+0], c[2]);
  1576.       tmp2b = - REAL_MUL(in[2*2+1], c[4]) + REAL_MUL(in[2*4+1], c[8]) + tb66 - REAL_MUL(in[2*8+1], c[2]);
  1577.  
  1578.       MACRO1(3);
  1579.       MACRO2(5);
  1580.     }
  1581.  
  1582.         {
  1583.                 real sum0,sum1;
  1584.         sum0 =  in[2*0+0] - in[2*2+0] + in[2*4+0] - in[2*6+0] + in[2*8+0];
  1585.         sum1 = REAL_MUL((in[2*0+1] - in[2*2+1] + in[2*4+1] - in[2*6+1] + in[2*8+1] ), tfcos36[4]);
  1586.                 MACRO0(4);
  1587.         }
  1588.   }
  1589. #endif
  1590.  
  1591.   }
  1592. }
  1593.  
  1594. /*
  1595.  * new DCT12
  1596.  */
  1597. static void dct12(real *in,real *rawout1,real *rawout2,register real *wi,register real *ts)
  1598. {
  1599. #define DCT12_PART1 \
  1600.              in5 = in[5*3];  \
  1601.      in5 += (in4 = in[4*3]); \
  1602.      in4 += (in3 = in[3*3]); \
  1603.      in3 += (in2 = in[2*3]); \
  1604.      in2 += (in1 = in[1*3]); \
  1605.      in1 += (in0 = in[0*3]); \
  1606.                              \
  1607.      in5 += in3; in3 += in1; \
  1608.                              \
  1609.      in2 = REAL_MUL(in2, COS6_1); \
  1610.      in3 = REAL_MUL(in3, COS6_1); \
  1611.  
  1612. #define DCT12_PART2 \
  1613.      in0 += REAL_MUL(in4, COS6_2); \
  1614.                           \
  1615.      in4 = in0 + in2;     \
  1616.      in0 -= in2;          \
  1617.                           \
  1618.      in1 += REAL_MUL(in5, COS6_2); \
  1619.                           \
  1620.      in5 = REAL_MUL((in1 + in3), tfcos12[0]); \
  1621.      in1 = REAL_MUL((in1 - in3), tfcos12[2]); \
  1622.                          \
  1623.      in3 = in4 + in5;    \
  1624.      in4 -= in5;         \
  1625.                          \
  1626.      in2 = in0 + in1;    \
  1627.      in0 -= in1;
  1628.  
  1629.  
  1630.    {
  1631.      real in0,in1,in2,in3,in4,in5;
  1632.      register real *out1 = rawout1;
  1633.      ts[SBLIMIT*0] = out1[0]; ts[SBLIMIT*1] = out1[1]; ts[SBLIMIT*2] = out1[2];
  1634.      ts[SBLIMIT*3] = out1[3]; ts[SBLIMIT*4] = out1[4]; ts[SBLIMIT*5] = out1[5];
  1635.  
  1636.      DCT12_PART1
  1637.  
  1638.      {
  1639.        real tmp0,tmp1 = (in0 - in4);
  1640.        {
  1641.          real tmp2 = REAL_MUL((in1 - in5), tfcos12[1]);
  1642.          tmp0 = tmp1 + tmp2;
  1643.          tmp1 -= tmp2;
  1644.        }
  1645.        ts[(17-1)*SBLIMIT] = out1[17-1] + REAL_MUL(tmp0, wi[11-1]);
  1646.        ts[(12+1)*SBLIMIT] = out1[12+1] + REAL_MUL(tmp0, wi[6+1]);
  1647.        ts[(6 +1)*SBLIMIT] = out1[6 +1] + REAL_MUL(tmp1, wi[1]);
  1648.        ts[(11-1)*SBLIMIT] = out1[11-1] + REAL_MUL(tmp1, wi[5-1]);
  1649.      }
  1650.  
  1651.      DCT12_PART2
  1652.  
  1653.      ts[(17-0)*SBLIMIT] = out1[17-0] + REAL_MUL(in2, wi[11-0]);
  1654.      ts[(12+0)*SBLIMIT] = out1[12+0] + REAL_MUL(in2, wi[6+0]);
  1655.      ts[(12+2)*SBLIMIT] = out1[12+2] + REAL_MUL(in3, wi[6+2]);
  1656.      ts[(17-2)*SBLIMIT] = out1[17-2] + REAL_MUL(in3, wi[11-2]);
  1657.  
  1658.      ts[(6 +0)*SBLIMIT]  = out1[6+0] + REAL_MUL(in0, wi[0]);
  1659.      ts[(11-0)*SBLIMIT] = out1[11-0] + REAL_MUL(in0, wi[5-0]);
  1660.      ts[(6 +2)*SBLIMIT]  = out1[6+2] + REAL_MUL(in4, wi[2]);
  1661.      ts[(11-2)*SBLIMIT] = out1[11-2] + REAL_MUL(in4, wi[5-2]);
  1662.   }
  1663.  
  1664.   in++;
  1665.  
  1666.   {
  1667.      real in0,in1,in2,in3,in4,in5;
  1668.      register real *out2 = rawout2;
  1669.  
  1670.      DCT12_PART1
  1671.  
  1672.      {
  1673.        real tmp0,tmp1 = (in0 - in4);
  1674.        {
  1675.          real tmp2 = REAL_MUL((in1 - in5), tfcos12[1]);
  1676.          tmp0 = tmp1 + tmp2;
  1677.          tmp1 -= tmp2;
  1678.        }
  1679.        out2[5-1] = REAL_MUL(tmp0, wi[11-1]);
  1680.        out2[0+1] = REAL_MUL(tmp0, wi[6+1]);
  1681.        ts[(12+1)*SBLIMIT] += REAL_MUL(tmp1, wi[1]);
  1682.        ts[(17-1)*SBLIMIT] += REAL_MUL(tmp1, wi[5-1]);
  1683.      }
  1684.  
  1685.      DCT12_PART2
  1686.  
  1687.      out2[5-0] = REAL_MUL(in2, wi[11-0]);
  1688.      out2[0+0] = REAL_MUL(in2, wi[6+0]);
  1689.      out2[0+2] = REAL_MUL(in3, wi[6+2]);
  1690.      out2[5-2] = REAL_MUL(in3, wi[11-2]);
  1691.  
  1692.      ts[(12+0)*SBLIMIT] += REAL_MUL(in0, wi[0]);
  1693.      ts[(17-0)*SBLIMIT] += REAL_MUL(in0, wi[5-0]);
  1694.      ts[(12+2)*SBLIMIT] += REAL_MUL(in4, wi[2]);
  1695.      ts[(17-2)*SBLIMIT] += REAL_MUL(in4, wi[5-2]);
  1696.   }
  1697.  
  1698.   in++;
  1699.  
  1700.   {
  1701.      real in0,in1,in2,in3,in4,in5;
  1702.      register real *out2 = rawout2;
  1703.      out2[12]=out2[13]=out2[14]=out2[15]=out2[16]=out2[17]=0.0;
  1704.  
  1705.      DCT12_PART1
  1706.  
  1707.      {
  1708.        real tmp0,tmp1 = (in0 - in4);
  1709.        {
  1710.          real tmp2 = REAL_MUL((in1 - in5), tfcos12[1]);
  1711.          tmp0 = tmp1 + tmp2;
  1712.          tmp1 -= tmp2;
  1713.        }
  1714.        out2[11-1] = REAL_MUL(tmp0, wi[11-1]);
  1715.        out2[6 +1] = REAL_MUL(tmp0, wi[6+1]);
  1716.        out2[0+1] += REAL_MUL(tmp1, wi[1]);
  1717.        out2[5-1] += REAL_MUL(tmp1, wi[5-1]);
  1718.      }
  1719.  
  1720.      DCT12_PART2
  1721.  
  1722.      out2[11-0] = REAL_MUL(in2, wi[11-0]);
  1723.      out2[6 +0] = REAL_MUL(in2, wi[6+0]);
  1724.      out2[6 +2] = REAL_MUL(in3, wi[6+2]);
  1725.      out2[11-2] = REAL_MUL(in3, wi[11-2]);
  1726.  
  1727.      out2[0+0] += REAL_MUL(in0, wi[0]);
  1728.      out2[5-0] += REAL_MUL(in0, wi[5-0]);
  1729.      out2[0+2] += REAL_MUL(in4, wi[2]);
  1730.      out2[5-2] += REAL_MUL(in4, wi[5-2]);
  1731.   }
  1732. }
  1733.  
  1734. /*
  1735.  * III_hybrid
  1736.  */
  1737. #ifdef USE_3DNOW
  1738. static void III_hybrid(real fsIn[SBLIMIT][SSLIMIT],real tsOut[SSLIMIT][SBLIMIT],int ch,struct gr_info_s *gr_info,struct frame *fr)
  1739. #else
  1740. static void III_hybrid(real fsIn[SBLIMIT][SSLIMIT],real tsOut[SSLIMIT][SBLIMIT],
  1741.    int ch,struct gr_info_s *gr_info)
  1742. #endif
  1743. {
  1744.     real *tspnt = (real *) tsOut;
  1745.    real *rawout1,*rawout2;
  1746.    int bt,sb = 0;
  1747.  
  1748.    {
  1749.      int b = blc[ch];
  1750.      rawout1=block[b][ch];
  1751.      b=-b+1;
  1752.      rawout2=block[b][ch];
  1753.      blc[ch] = b;
  1754.    }
  1755.  
  1756.    if(gr_info->mixed_block_flag) {
  1757.      sb = 2;
  1758. #ifdef USE_3DNOW
  1759.      (fr->dct36)(fsIn[0],rawout1,rawout2,win[0],tspnt);
  1760.      (fr->dct36)(fsIn[1],rawout1+18,rawout2+18,win1[0],tspnt+1);
  1761. #else
  1762.      dct36(fsIn[0],rawout1,rawout2,win[0],tspnt);
  1763.      dct36(fsIn[1],rawout1+18,rawout2+18,win1[0],tspnt+1);
  1764. #endif
  1765.      rawout1 += 36; rawout2 += 36; tspnt += 2;
  1766.    }
  1767.  
  1768.    bt = gr_info->block_type;
  1769.    if(bt == 2) {
  1770.      for (; sb<gr_info->maxb; sb+=2,tspnt+=2,rawout1+=36,rawout2+=36) {
  1771.        dct12(fsIn[sb]  ,rawout1   ,rawout2   ,win[2] ,tspnt);
  1772.        dct12(fsIn[sb+1],rawout1+18,rawout2+18,win1[2],tspnt+1);
  1773.      }
  1774.    }
  1775.    else {
  1776.      for (; sb<gr_info->maxb; sb+=2,tspnt+=2,rawout1+=36,rawout2+=36) {
  1777. #ifdef USE_3DNOW
  1778.        (fr->dct36)(fsIn[sb],rawout1,rawout2,win[bt],tspnt);
  1779.        (fr->dct36)(fsIn[sb+1],rawout1+18,rawout2+18,win1[bt],tspnt+1);
  1780. #else
  1781.        dct36(fsIn[sb],rawout1,rawout2,win[bt],tspnt);
  1782.        dct36(fsIn[sb+1],rawout1+18,rawout2+18,win1[bt],tspnt+1);
  1783. #endif
  1784.      }
  1785.    }
  1786.  
  1787.    for(;sb<SBLIMIT;sb++,tspnt++) {
  1788.      int i;
  1789.      for(i=0;i<SSLIMIT;i++) {
  1790.        tspnt[i*SBLIMIT] = *rawout1++;
  1791.        *rawout2++ = DOUBLE_TO_REAL(0.0);
  1792.      }
  1793.    }
  1794. }
  1795.  
  1796. int do_layer3(struct frame *fr,byte *pcm_sample,int *pcm_point)
  1797. {
  1798.   int gr, ch, ss,clip=0;
  1799.   int scalefacs[2][39]; /* max 39 for short[13][3] mode, mixed: 38, long: 22 */
  1800.   struct III_sideinfo sideinfo;
  1801.   int stereo = fr->stereo;
  1802.   int single = fr->single;
  1803.   int ms_stereo,i_stereo;
  1804.   int sfreq = fr->sampling_frequency;
  1805.   int stereo1,granules;
  1806.  
  1807.   if(stereo == 1) { /* stream is mono */
  1808.     stereo1 = 1;
  1809.     single = 0;
  1810.   }
  1811.   else if(single >= 0) /* stream is stereo, but force to mono */
  1812.     stereo1 = 1;
  1813.   else
  1814.     stereo1 = 2;
  1815.  
  1816.   if(fr->mode == MPG_MD_JOINT_STEREO) {
  1817.     ms_stereo = (fr->mode_ext & 0x2)>>1;
  1818.     i_stereo  = fr->mode_ext & 0x1;
  1819.   }
  1820.   else
  1821.     ms_stereo = i_stereo = 0;
  1822.  
  1823.   if(fr->lsf) {
  1824.     granules = 1;
  1825. #if 0
  1826.     III_get_side_info_2(&sideinfo,stereo,ms_stereo,sfreq,single);
  1827. #endif
  1828.   }
  1829.   else {
  1830.     granules = 2;
  1831.   }
  1832.   /* quick hack to keep the music playing */
  1833.   /* after having seen this nasty test file... */
  1834.   if(III_get_side_info(&sideinfo,stereo,ms_stereo,sfreq,single,fr->lsf))
  1835.   {
  1836.     return clip;
  1837.   }
  1838.  
  1839.   set_pointer(sideinfo.main_data_begin);
  1840.  
  1841.   for (gr=0;gr<granules;gr++)
  1842.    {
  1843.  
  1844.     {
  1845.       struct gr_info_s *gr_info = &(sideinfo.ch[0].gr[gr]);
  1846.       long part2bits;
  1847.       if(fr->lsf)
  1848.         part2bits = III_get_scale_factors_2(scalefacs[0],gr_info,0);
  1849.       else
  1850.         part2bits = III_get_scale_factors_1(scalefacs[0],gr_info,0,gr);
  1851.  
  1852.       if(III_dequantize_sample(hybridIn[0], scalefacs[0],gr_info,sfreq,part2bits))
  1853.         return clip;
  1854.     }
  1855.  
  1856.     if(stereo == 2) {
  1857.       struct gr_info_s *gr_info = &(sideinfo.ch[1].gr[gr]);
  1858.       long part2bits;
  1859.       if(fr->lsf)
  1860.         part2bits = III_get_scale_factors_2(scalefacs[1],gr_info,i_stereo);
  1861.       else
  1862.         part2bits = III_get_scale_factors_1(scalefacs[1],gr_info,1,gr);
  1863.  
  1864.       if(III_dequantize_sample(hybridIn[1],scalefacs[1],gr_info,sfreq,part2bits))
  1865.           return clip;
  1866.  
  1867.       if(ms_stereo) {
  1868.         int i;
  1869.         int maxb = sideinfo.ch[0].gr[gr].maxb;
  1870.         if(sideinfo.ch[1].gr[gr].maxb > maxb)
  1871.             maxb = sideinfo.ch[1].gr[gr].maxb;
  1872.         for(i=0;i<SSLIMIT*maxb;i++) {
  1873.           real tmp0 = ((real *)hybridIn[0])[i];
  1874.           real tmp1 = ((real *)hybridIn[1])[i];
  1875.           ((real *)hybridIn[0])[i] = tmp0 + tmp1;
  1876.           ((real *)hybridIn[1])[i] = tmp0 - tmp1;
  1877.         }
  1878.       }
  1879.  
  1880.       if(i_stereo)
  1881.         III_i_stereo(hybridIn,scalefacs[1],gr_info,sfreq,ms_stereo,fr->lsf);
  1882.  
  1883.       if(ms_stereo || i_stereo || (single == 3) ) {
  1884.         if(gr_info->maxb > sideinfo.ch[0].gr[gr].maxb)
  1885.           sideinfo.ch[0].gr[gr].maxb = gr_info->maxb;
  1886.         else
  1887.           gr_info->maxb = sideinfo.ch[0].gr[gr].maxb;
  1888.       }
  1889.  
  1890.       switch(single) {
  1891.         case 3:
  1892.           {
  1893.             register int i;
  1894.             register real *in0 = (real *) hybridIn[0],*in1 = (real *) hybridIn[1];
  1895.             for(i=0;i<SSLIMIT*gr_info->maxb;i++,in0++)
  1896.               *in0 = (*in0 + *in1++); /* *0.5 done by pow-scale */
  1897.           }
  1898.           break;
  1899.         case 1:
  1900.           {
  1901.             register int i;
  1902.             register real *in0 = (real *) hybridIn[0],*in1 = (real *) hybridIn[1];
  1903.             for(i=0;i<SSLIMIT*gr_info->maxb;i++)
  1904.               *in0++ = *in1++;
  1905.           }
  1906.           break;
  1907.       }
  1908.     }
  1909.  
  1910.     for(ch=0;ch<stereo1;ch++) {
  1911.       struct gr_info_s *gr_info = &(sideinfo.ch[ch].gr[gr]);
  1912.       III_antialias(hybridIn[ch],gr_info);
  1913. #ifdef USE_3DNOW
  1914.       III_hybrid(hybridIn[ch], hybridOut[ch], ch,gr_info,fr);
  1915. #else
  1916.       III_hybrid(hybridIn[ch], hybridOut[ch], ch,gr_info);
  1917. #endif
  1918.     }
  1919.  
  1920. #ifdef I486_OPT
  1921.     if (fr->synth != synth_1to1 || single >= 0) {
  1922. #endif
  1923.     for(ss=0;ss<SSLIMIT;ss++) {
  1924.       if(single >= 0) {
  1925.         clip += (fr->synth_mono)(hybridOut[0][ss],pcm_sample,pcm_point);
  1926.       }
  1927.       else {
  1928.         int p1=*pcm_point;
  1929.         clip += (fr->synth)(hybridOut[0][ss],0,pcm_sample,&p1);
  1930.         clip += (fr->synth)(hybridOut[1][ss],1,pcm_sample,pcm_point);
  1931.       }
  1932.  
  1933. #ifdef VARMODESUPPORT
  1934.       if (playlimit < 128) {
  1935.         pcm_point -= playlimit >> 1;
  1936.         playlimit = 0;
  1937.       }
  1938.       else
  1939.         playlimit -= 128;
  1940. #endif
  1941.     }
  1942. #ifdef I486_OPT
  1943.     } else {
  1944.       /* Only stereo, 16 bits benefit from the 486 optimization. */
  1945.       ss=0;
  1946.       while (ss < SSLIMIT) {
  1947.         int n;
  1948.         n=(0x40000 - *pcm_point) / (2*2*32);
  1949.         if (n > (SSLIMIT-ss)) n=SSLIMIT-ss;
  1950.        
  1951.         synth_1to1_486(hybridOut[0][ss],0,pcm_sample+*pcm_point,n);
  1952.         synth_1to1_486(hybridOut[1][ss],1,pcm_sample+*pcm_point,n);
  1953.         ss+=n;
  1954.         *pcm_point+=(2*2*32)*n;
  1955.       }
  1956.     }
  1957. #endif
  1958.   }
  1959.  
  1960.   return clip;
  1961. }
  1962.