libzpaqo.cpp

/* libzpaqo.cpp

Part of LIBZPAQ Version 2.02
See accompanying libzpaq.txt for documentation.

Written by Matt Mahoney, Nov. 13, 2010

The LIBZPAQ software is placed in the public domain. It may be used
without restriction. LIBZPAQ is provided "as is" with no warranty.

This file contains optimized versions of
models[], ZPAQL::run(), Predictor::predict() and Predictor::update()
for fast.cfg, mid.cfg, and max.cfg.

It was generated using zpaq 2.02 as follows:

  zpaq cfast x nul:
  zpaq amid x nul:
  zpaq amax x nul:
  zpaq kox x
  rename zpaqopt.cpp libzpaqo.cpp

followed by inserting this comment and removing the declaration
for zpaqdir at the end.
*/


#include "libzpaq.h"
namespace libzpaq {

const char models[]={

  // Model 1
  26,0,1,2,0,0,2,3,16,8,19,0,0,96,4,28,
  59,10,59,112,25,10,59,10,59,112,56,0,

  // Model 2
  69,0,3,3,0,0,8,3,5,8,13,0,8,17,1,8,
  18,2,8,18,3,8,19,4,4,22,24,7,16,0,7,24,
  -1,0,17,104,74,4,95,1,59,112,10,25,59,112,10,25,
  59,112,10,25,59,112,10,25,59,112,10,25,59,10,59,112,
  25,69,-49,8,112,56,0,

  // Model 3
  -60,0,5,9,0,0,22,1,-96,3,5,8,13,1,8,16,
  2,8,18,3,8,19,4,8,19,5,8,20,6,4,22,24,
  3,17,8,19,9,3,13,3,13,3,13,3,14,7,16,0,
  15,24,-1,7,8,0,16,10,-1,6,0,15,16,24,0,9,
  8,17,32,-1,6,8,17,18,16,-1,9,16,19,32,-1,6,
  0,19,20,16,0,0,17,104,74,4,95,2,59,112,10,25,
  59,112,10,25,59,112,10,25,59,112,10,25,59,112,10,25,
  59,10,59,112,10,25,59,112,10,25,69,-73,32,-17,64,47,
  14,-25,91,47,10,25,60,26,48,-122,-105,20,112,63,9,70,
  -33,0,39,3,25,112,26,52,25,25,74,10,4,59,112,25,
  10,4,59,112,25,10,4,59,112,25,65,-113,-44,72,4,59,
  112,8,-113,-40,8,68,-81,60,60,25,69,-49,9,112,25,25,
  25,25,25,112,56,0,

  0,0};

int Predictor::predict() {
  switch(z.select) {
    case 1: {
      // 2 components

      // 0 ICM 16
      if (c8==1 || (c8&0xf0)==16)
        comp[0].c=find(comp[0].ht, 16+2, z.H(0)+16*c8);
      comp[0].cxt=comp[0].ht[comp[0].c+(hmap4&15)];
      p[0]=stretch(comp[0].cm(comp[0].cxt)>>8);

      // 1 ISSE 19 0
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[1].c=find(comp[1].ht, 21, z.H(1)+16*c8);
        comp[1].cxt=comp[1].ht[comp[1].c+(hmap4&15)];
        int *wt=(int*)&comp[1].cm[comp[1].cxt*2];
        p[1]=clamp2k((wt[0]*p[0]+wt[1]*64)>>16);
      }
      return squash(p[1]);
    }
    case 2: {
      // 8 components

      // 0 ICM 5
      if (c8==1 || (c8&0xf0)==16)
        comp[0].c=find(comp[0].ht, 5+2, z.H(0)+16*c8);
      comp[0].cxt=comp[0].ht[comp[0].c+(hmap4&15)];
      p[0]=stretch(comp[0].cm(comp[0].cxt)>>8);

      // 1 ISSE 13 0
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[1].c=find(comp[1].ht, 15, z.H(1)+16*c8);
        comp[1].cxt=comp[1].ht[comp[1].c+(hmap4&15)];
        int *wt=(int*)&comp[1].cm[comp[1].cxt*2];
        p[1]=clamp2k((wt[0]*p[0]+wt[1]*64)>>16);
      }

      // 2 ISSE 17 1
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[2].c=find(comp[2].ht, 19, z.H(2)+16*c8);
        comp[2].cxt=comp[2].ht[comp[2].c+(hmap4&15)];
        int *wt=(int*)&comp[2].cm[comp[2].cxt*2];
        p[2]=clamp2k((wt[0]*p[1]+wt[1]*64)>>16);
      }

      // 3 ISSE 18 2
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[3].c=find(comp[3].ht, 20, z.H(3)+16*c8);
        comp[3].cxt=comp[3].ht[comp[3].c+(hmap4&15)];
        int *wt=(int*)&comp[3].cm[comp[3].cxt*2];
        p[3]=clamp2k((wt[0]*p[2]+wt[1]*64)>>16);
      }

      // 4 ISSE 18 3
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[4].c=find(comp[4].ht, 20, z.H(4)+16*c8);
        comp[4].cxt=comp[4].ht[comp[4].c+(hmap4&15)];
        int *wt=(int*)&comp[4].cm[comp[4].cxt*2];
        p[4]=clamp2k((wt[0]*p[3]+wt[1]*64)>>16);
      }

      // 5 ISSE 19 4
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[5].c=find(comp[5].ht, 21, z.H(5)+16*c8);
        comp[5].cxt=comp[5].ht[comp[5].c+(hmap4&15)];
        int *wt=(int*)&comp[5].cm[comp[5].cxt*2];
        p[5]=clamp2k((wt[0]*p[4]+wt[1]*64)>>16);
      }

      // 6 MATCH 22 24
      if (comp[6].a==0) p[6]=0;
      else {
        comp[6].c=comp[6].ht((comp[6].limit>>3)
           -comp[6].b)>>(7-(comp[6].limit&7))&1;
        p[6]=stretch(comp[6].cxt*(comp[6].c*-2+1)&32767);
      }

      // 7 MIX 16 0 7 24 255
      {
        comp[7].cxt=z.H(7)+(c8&255);
        comp[7].cxt=(comp[7].cxt&(comp[7].c-1))*7;
        int* wt=(int*)&comp[7].cm[comp[7].cxt];
        p[7]=(wt[0]>>8)*p[0];
        p[7]+=(wt[1]>>8)*p[1];
        p[7]+=(wt[2]>>8)*p[2];
        p[7]+=(wt[3]>>8)*p[3];
        p[7]+=(wt[4]>>8)*p[4];
        p[7]+=(wt[5]>>8)*p[5];
        p[7]+=(wt[6]>>8)*p[6];
        p[7]=clamp2k(p[7]>>8);
      }
      return squash(p[7]);
    }
    case 3: {
      // 22 components

      // 0 CONST 160

      // 1 ICM 5
      if (c8==1 || (c8&0xf0)==16)
        comp[1].c=find(comp[1].ht, 5+2, z.H(1)+16*c8);
      comp[1].cxt=comp[1].ht[comp[1].c+(hmap4&15)];
      p[1]=stretch(comp[1].cm(comp[1].cxt)>>8);

      // 2 ISSE 13 1
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[2].c=find(comp[2].ht, 15, z.H(2)+16*c8);
        comp[2].cxt=comp[2].ht[comp[2].c+(hmap4&15)];
        int *wt=(int*)&comp[2].cm[comp[2].cxt*2];
        p[2]=clamp2k((wt[0]*p[1]+wt[1]*64)>>16);
      }

      // 3 ISSE 16 2
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[3].c=find(comp[3].ht, 18, z.H(3)+16*c8);
        comp[3].cxt=comp[3].ht[comp[3].c+(hmap4&15)];
        int *wt=(int*)&comp[3].cm[comp[3].cxt*2];
        p[3]=clamp2k((wt[0]*p[2]+wt[1]*64)>>16);
      }

      // 4 ISSE 18 3
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[4].c=find(comp[4].ht, 20, z.H(4)+16*c8);
        comp[4].cxt=comp[4].ht[comp[4].c+(hmap4&15)];
        int *wt=(int*)&comp[4].cm[comp[4].cxt*2];
        p[4]=clamp2k((wt[0]*p[3]+wt[1]*64)>>16);
      }

      // 5 ISSE 19 4
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[5].c=find(comp[5].ht, 21, z.H(5)+16*c8);
        comp[5].cxt=comp[5].ht[comp[5].c+(hmap4&15)];
        int *wt=(int*)&comp[5].cm[comp[5].cxt*2];
        p[5]=clamp2k((wt[0]*p[4]+wt[1]*64)>>16);
      }

      // 6 ISSE 19 5
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[6].c=find(comp[6].ht, 21, z.H(6)+16*c8);
        comp[6].cxt=comp[6].ht[comp[6].c+(hmap4&15)];
        int *wt=(int*)&comp[6].cm[comp[6].cxt*2];
        p[6]=clamp2k((wt[0]*p[5]+wt[1]*64)>>16);
      }

      // 7 ISSE 20 6
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[7].c=find(comp[7].ht, 22, z.H(7)+16*c8);
        comp[7].cxt=comp[7].ht[comp[7].c+(hmap4&15)];
        int *wt=(int*)&comp[7].cm[comp[7].cxt*2];
        p[7]=clamp2k((wt[0]*p[6]+wt[1]*64)>>16);
      }

      // 8 MATCH 22 24
      if (comp[8].a==0) p[8]=0;
      else {
        comp[8].c=comp[8].ht((comp[8].limit>>3)
           -comp[8].b)>>(7-(comp[8].limit&7))&1;
        p[8]=stretch(comp[8].cxt*(comp[8].c*-2+1)&32767);
      }

      // 9 ICM 17
      if (c8==1 || (c8&0xf0)==16)
        comp[9].c=find(comp[9].ht, 17+2, z.H(9)+16*c8);
      comp[9].cxt=comp[9].ht[comp[9].c+(hmap4&15)];
      p[9]=stretch(comp[9].cm(comp[9].cxt)>>8);

      // 10 ISSE 19 9
      {
        if (c8==1 || (c8&0xf0)==16)
          comp[10].c=find(comp[10].ht, 21, z.H(10)+16*c8);
        comp[10].cxt=comp[10].ht[comp[10].c+(hmap4&15)];
        int *wt=(int*)&comp[10].cm[comp[10].cxt*2];
        p[10]=clamp2k((wt[0]*p[9]+wt[1]*64)>>16);
      }

      // 11 ICM 13
      if (c8==1 || (c8&0xf0)==16)
        comp[11].c=find(comp[11].ht, 13+2, z.H(11)+16*c8);
      comp[11].cxt=comp[11].ht[comp[11].c+(hmap4&15)];
      p[11]=stretch(comp[11].cm(comp[11].cxt)>>8);

      // 12 ICM 13
      if (c8==1 || (c8&0xf0)==16)
        comp[12].c=find(comp[12].ht, 13+2, z.H(12)+16*c8);
      comp[12].cxt=comp[12].ht[comp[12].c+(hmap4&15)];
      p[12]=stretch(comp[12].cm(comp[12].cxt)>>8);

      // 13 ICM 13
      if (c8==1 || (c8&0xf0)==16)
        comp[13].c=find(comp[13].ht, 13+2, z.H(13)+16*c8);
      comp[13].cxt=comp[13].ht[comp[13].c+(hmap4&15)];
      p[13]=stretch(comp[13].cm(comp[13].cxt)>>8);

      // 14 ICM 14
      if (c8==1 || (c8&0xf0)==16)
        comp[14].c=find(comp[14].ht, 14+2, z.H(14)+16*c8);
      comp[14].cxt=comp[14].ht[comp[14].c+(hmap4&15)];
      p[14]=stretch(comp[14].cm(comp[14].cxt)>>8);

      // 15 MIX 16 0 15 24 255
      {
        comp[15].cxt=z.H(15)+(c8&255);
        comp[15].cxt=(comp[15].cxt&(comp[15].c-1))*15;
        int* wt=(int*)&comp[15].cm[comp[15].cxt];
        p[15]=(wt[0]>>8)*p[0];
        p[15]+=(wt[1]>>8)*p[1];
        p[15]+=(wt[2]>>8)*p[2];
        p[15]+=(wt[3]>>8)*p[3];
        p[15]+=(wt[4]>>8)*p[4];
        p[15]+=(wt[5]>>8)*p[5];
        p[15]+=(wt[6]>>8)*p[6];
        p[15]+=(wt[7]>>8)*p[7];
        p[15]+=(wt[8]>>8)*p[8];
        p[15]+=(wt[9]>>8)*p[9];
        p[15]+=(wt[10]>>8)*p[10];
        p[15]+=(wt[11]>>8)*p[11];
        p[15]+=(wt[12]>>8)*p[12];
        p[15]+=(wt[13]>>8)*p[13];
        p[15]+=(wt[14]>>8)*p[14];
        p[15]=clamp2k(p[15]>>8);
      }

      // 16 MIX 8 0 16 10 255
      {
        comp[16].cxt=z.H(16)+(c8&255);
        comp[16].cxt=(comp[16].cxt&(comp[16].c-1))*16;
        int* wt=(int*)&comp[16].cm[comp[16].cxt];
        p[16]=(wt[0]>>8)*p[0];
        p[16]+=(wt[1]>>8)*p[1];
        p[16]+=(wt[2]>>8)*p[2];
        p[16]+=(wt[3]>>8)*p[3];
        p[16]+=(wt[4]>>8)*p[4];
        p[16]+=(wt[5]>>8)*p[5];
        p[16]+=(wt[6]>>8)*p[6];
        p[16]+=(wt[7]>>8)*p[7];
        p[16]+=(wt[8]>>8)*p[8];
        p[16]+=(wt[9]>>8)*p[9];
        p[16]+=(wt[10]>>8)*p[10];
        p[16]+=(wt[11]>>8)*p[11];
        p[16]+=(wt[12]>>8)*p[12];
        p[16]+=(wt[13]>>8)*p[13];
        p[16]+=(wt[14]>>8)*p[14];
        p[16]+=(wt[15]>>8)*p[15];
        p[16]=clamp2k(p[16]>>8);
      }

      // 17 MIX2 0 15 16 24 0
      {
        comp[17].cxt=((z.H(17)+(c8&0))&(comp[17].c-1));
        int w=comp[17].a16[comp[17].cxt];
        p[17]=(w*p[15]+(65536-w)*p[16])>>16;
      }

      // 18 SSE 8 17 32 255
      {
        comp[18].cxt=(z.H(18)+c8)*32;
        int pq=p[17]+992;
        if (pq<0) pq=0;
        if (pq>1983) pq=1983;
        int wt=pq&63;
        pq>>=6;
        comp[18].cxt+=pq;
        p[18]=stretch(((comp[18].cm(comp[18].cxt)>>10)*(64-wt)
           +(comp[18].cm(comp[18].cxt+1)>>10)*wt)>>13);
        comp[18].cxt+=wt>>5;
      }

      // 19 MIX2 8 17 18 16 255
      {
        comp[19].cxt=((z.H(19)+(c8&255))&(comp[19].c-1));
        int w=comp[19].a16[comp[19].cxt];
        p[19]=(w*p[17]+(65536-w)*p[18])>>16;
      }

      // 20 SSE 16 19 32 255
      {
        comp[20].cxt=(z.H(20)+c8)*32;
        int pq=p[19]+992;
        if (pq<0) pq=0;
        if (pq>1983) pq=1983;
        int wt=pq&63;
        pq>>=6;
        comp[20].cxt+=pq;
        p[20]=stretch(((comp[20].cm(comp[20].cxt)>>10)*(64-wt)
           +(comp[20].cm(comp[20].cxt+1)>>10)*wt)>>13);
        comp[20].cxt+=wt>>5;
      }

      // 21 MIX2 0 19 20 16 0
      {
        comp[21].cxt=((z.H(21)+(c8&0))&(comp[21].c-1));
        int w=comp[21].a16[comp[21].cxt];
        p[21]=(w*p[19]+(65536-w)*p[20])>>16;
      }
      return squash(p[21]);
    }
    default: return predict0();
  }
}

void Predictor::update(int y) {
  switch(z.select) {
    case 1: {
      // 2 components

      // 0 ICM 16
      {
        comp[0].ht[comp[0].c+(hmap4&15)]=
            st.next(comp[0].ht[comp[0].c+(hmap4&15)], y);
        U32& pn=comp[0].cm(comp[0].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 1 ISSE 19 0
      {
        int err=y*32767-squash(p[1]);
        int *wt=(int*)&comp[1].cm[comp[1].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[0]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[1].ht[comp[1].c+(hmap4&15)]=st.next(comp[1].cxt, y);
      }
      break;
    }
    case 2: {
      // 8 components

      // 0 ICM 5
      {
        comp[0].ht[comp[0].c+(hmap4&15)]=
            st.next(comp[0].ht[comp[0].c+(hmap4&15)], y);
        U32& pn=comp[0].cm(comp[0].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 1 ISSE 13 0
      {
        int err=y*32767-squash(p[1]);
        int *wt=(int*)&comp[1].cm[comp[1].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[0]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[1].ht[comp[1].c+(hmap4&15)]=st.next(comp[1].cxt, y);
      }

      // 2 ISSE 17 1
      {
        int err=y*32767-squash(p[2]);
        int *wt=(int*)&comp[2].cm[comp[2].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[1]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[2].ht[comp[2].c+(hmap4&15)]=st.next(comp[2].cxt, y);
      }

      // 3 ISSE 18 2
      {
        int err=y*32767-squash(p[3]);
        int *wt=(int*)&comp[3].cm[comp[3].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[2]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[3].ht[comp[3].c+(hmap4&15)]=st.next(comp[3].cxt, y);
      }

      // 4 ISSE 18 3
      {
        int err=y*32767-squash(p[4]);
        int *wt=(int*)&comp[4].cm[comp[4].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[3]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[4].ht[comp[4].c+(hmap4&15)]=st.next(comp[4].cxt, y);
      }

      // 5 ISSE 19 4
      {
        int err=y*32767-squash(p[5]);
        int *wt=(int*)&comp[5].cm[comp[5].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[4]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[5].ht[comp[5].c+(hmap4&15)]=st.next(comp[5].cxt, y);
      }

      // 6 MATCH 22 24
      {
        if (comp[6].c!=y) comp[6].a=0;
        comp[6].ht(comp[6].limit>>3)+=comp[6].ht(comp[6].limit>>3)+y;
        if ((++comp[6].limit&7)==0) {
          int pos=comp[6].limit>>3;
          if (comp[6].a==0) {
            comp[6].b=pos-comp[6].cm(z.H(6));
            if (comp[6].b&(comp[6].ht.size()-1))
              while (comp[6].a<255 && comp[6].ht(pos-comp[6].a-1)
                     ==comp[6].ht(pos-comp[6].a-comp[6].b-1))
                ++comp[6].a;
          }
          else comp[6].a+=comp[6].a<255;
          comp[6].cm(z.H(6))=pos;
          if (comp[6].a>0) comp[6].cxt=2048/comp[6].a;
        }
      }

      // 7 MIX 16 0 7 24 255
      {
        int err=(y*32767-squash(p[7]))*24>>4;
        int* wt=(int*)&comp[7].cm[comp[7].cxt];
          wt[0]=clamp512k(wt[0]+((err*p[0]+(1<<12))>>13));
          wt[1]=clamp512k(wt[1]+((err*p[1]+(1<<12))>>13));
          wt[2]=clamp512k(wt[2]+((err*p[2]+(1<<12))>>13));
          wt[3]=clamp512k(wt[3]+((err*p[3]+(1<<12))>>13));
          wt[4]=clamp512k(wt[4]+((err*p[4]+(1<<12))>>13));
          wt[5]=clamp512k(wt[5]+((err*p[5]+(1<<12))>>13));
          wt[6]=clamp512k(wt[6]+((err*p[6]+(1<<12))>>13));
      }
      break;
    }
    case 3: {
      // 22 components

      // 0 CONST 160

      // 1 ICM 5
      {
        comp[1].ht[comp[1].c+(hmap4&15)]=
            st.next(comp[1].ht[comp[1].c+(hmap4&15)], y);
        U32& pn=comp[1].cm(comp[1].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 2 ISSE 13 1
      {
        int err=y*32767-squash(p[2]);
        int *wt=(int*)&comp[2].cm[comp[2].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[1]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[2].ht[comp[2].c+(hmap4&15)]=st.next(comp[2].cxt, y);
      }

      // 3 ISSE 16 2
      {
        int err=y*32767-squash(p[3]);
        int *wt=(int*)&comp[3].cm[comp[3].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[2]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[3].ht[comp[3].c+(hmap4&15)]=st.next(comp[3].cxt, y);
      }

      // 4 ISSE 18 3
      {
        int err=y*32767-squash(p[4]);
        int *wt=(int*)&comp[4].cm[comp[4].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[3]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[4].ht[comp[4].c+(hmap4&15)]=st.next(comp[4].cxt, y);
      }

      // 5 ISSE 19 4
      {
        int err=y*32767-squash(p[5]);
        int *wt=(int*)&comp[5].cm[comp[5].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[4]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[5].ht[comp[5].c+(hmap4&15)]=st.next(comp[5].cxt, y);
      }

      // 6 ISSE 19 5
      {
        int err=y*32767-squash(p[6]);
        int *wt=(int*)&comp[6].cm[comp[6].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[5]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[6].ht[comp[6].c+(hmap4&15)]=st.next(comp[6].cxt, y);
      }

      // 7 ISSE 20 6
      {
        int err=y*32767-squash(p[7]);
        int *wt=(int*)&comp[7].cm[comp[7].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[6]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[7].ht[comp[7].c+(hmap4&15)]=st.next(comp[7].cxt, y);
      }

      // 8 MATCH 22 24
      {
        if (comp[8].c!=y) comp[8].a=0;
        comp[8].ht(comp[8].limit>>3)+=comp[8].ht(comp[8].limit>>3)+y;
        if ((++comp[8].limit&7)==0) {
          int pos=comp[8].limit>>3;
          if (comp[8].a==0) {
            comp[8].b=pos-comp[8].cm(z.H(8));
            if (comp[8].b&(comp[8].ht.size()-1))
              while (comp[8].a<255 && comp[8].ht(pos-comp[8].a-1)
                     ==comp[8].ht(pos-comp[8].a-comp[8].b-1))
                ++comp[8].a;
          }
          else comp[8].a+=comp[8].a<255;
          comp[8].cm(z.H(8))=pos;
          if (comp[8].a>0) comp[8].cxt=2048/comp[8].a;
        }
      }

      // 9 ICM 17
      {
        comp[9].ht[comp[9].c+(hmap4&15)]=
            st.next(comp[9].ht[comp[9].c+(hmap4&15)], y);
        U32& pn=comp[9].cm(comp[9].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 10 ISSE 19 9
      {
        int err=y*32767-squash(p[10]);
        int *wt=(int*)&comp[10].cm[comp[10].cxt*2];
        wt[0]=clamp512k(wt[0]+((err*p[9]+(1<<12))>>13));
        wt[1]=clamp512k(wt[1]+((err+16)>>5));
        comp[10].ht[comp[10].c+(hmap4&15)]=st.next(comp[10].cxt, y);
      }

      // 11 ICM 13
      {
        comp[11].ht[comp[11].c+(hmap4&15)]=
            st.next(comp[11].ht[comp[11].c+(hmap4&15)], y);
        U32& pn=comp[11].cm(comp[11].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 12 ICM 13
      {
        comp[12].ht[comp[12].c+(hmap4&15)]=
            st.next(comp[12].ht[comp[12].c+(hmap4&15)], y);
        U32& pn=comp[12].cm(comp[12].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 13 ICM 13
      {
        comp[13].ht[comp[13].c+(hmap4&15)]=
            st.next(comp[13].ht[comp[13].c+(hmap4&15)], y);
        U32& pn=comp[13].cm(comp[13].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 14 ICM 14
      {
        comp[14].ht[comp[14].c+(hmap4&15)]=
            st.next(comp[14].ht[comp[14].c+(hmap4&15)], y);
        U32& pn=comp[14].cm(comp[14].cxt);
        pn+=int(y*32767-(pn>>8))>>2;
      }

      // 15 MIX 16 0 15 24 255
      {
        int err=(y*32767-squash(p[15]))*24>>4;
        int* wt=(int*)&comp[15].cm[comp[15].cxt];
          wt[0]=clamp512k(wt[0]+((err*p[0]+(1<<12))>>13));
          wt[1]=clamp512k(wt[1]+((err*p[1]+(1<<12))>>13));
          wt[2]=clamp512k(wt[2]+((err*p[2]+(1<<12))>>13));
          wt[3]=clamp512k(wt[3]+((err*p[3]+(1<<12))>>13));
          wt[4]=clamp512k(wt[4]+((err*p[4]+(1<<12))>>13));
          wt[5]=clamp512k(wt[5]+((err*p[5]+(1<<12))>>13));
          wt[6]=clamp512k(wt[6]+((err*p[6]+(1<<12))>>13));
          wt[7]=clamp512k(wt[7]+((err*p[7]+(1<<12))>>13));
          wt[8]=clamp512k(wt[8]+((err*p[8]+(1<<12))>>13));
          wt[9]=clamp512k(wt[9]+((err*p[9]+(1<<12))>>13));
          wt[10]=clamp512k(wt[10]+((err*p[10]+(1<<12))>>13));
          wt[11]=clamp512k(wt[11]+((err*p[11]+(1<<12))>>13));
          wt[12]=clamp512k(wt[12]+((err*p[12]+(1<<12))>>13));
          wt[13]=clamp512k(wt[13]+((err*p[13]+(1<<12))>>13));
          wt[14]=clamp512k(wt[14]+((err*p[14]+(1<<12))>>13));
      }

      // 16 MIX 8 0 16 10 255
      {
        int err=(y*32767-squash(p[16]))*10>>4;
        int* wt=(int*)&comp[16].cm[comp[16].cxt];
          wt[0]=clamp512k(wt[0]+((err*p[0]+(1<<12))>>13));
          wt[1]=clamp512k(wt[1]+((err*p[1]+(1<<12))>>13));
          wt[2]=clamp512k(wt[2]+((err*p[2]+(1<<12))>>13));
          wt[3]=clamp512k(wt[3]+((err*p[3]+(1<<12))>>13));
          wt[4]=clamp512k(wt[4]+((err*p[4]+(1<<12))>>13));
          wt[5]=clamp512k(wt[5]+((err*p[5]+(1<<12))>>13));
          wt[6]=clamp512k(wt[6]+((err*p[6]+(1<<12))>>13));
          wt[7]=clamp512k(wt[7]+((err*p[7]+(1<<12))>>13));
          wt[8]=clamp512k(wt[8]+((err*p[8]+(1<<12))>>13));
          wt[9]=clamp512k(wt[9]+((err*p[9]+(1<<12))>>13));
          wt[10]=clamp512k(wt[10]+((err*p[10]+(1<<12))>>13));
          wt[11]=clamp512k(wt[11]+((err*p[11]+(1<<12))>>13));
          wt[12]=clamp512k(wt[12]+((err*p[12]+(1<<12))>>13));
          wt[13]=clamp512k(wt[13]+((err*p[13]+(1<<12))>>13));
          wt[14]=clamp512k(wt[14]+((err*p[14]+(1<<12))>>13));
          wt[15]=clamp512k(wt[15]+((err*p[15]+(1<<12))>>13));
      }

      // 17 MIX2 0 15 16 24 0
      {
        int err=(y*32767-squash(p[17]))*24>>5;
        int w=comp[17].a16[comp[17].cxt];
        w+=(err*(p[15]-p[16])+(1<<12))>>13;
        if (w<0) w=0;
        if (w>65535) w=65535;
        comp[17].a16[comp[17].cxt]=w;
      }

      // 18 SSE 8 17 32 255
      train(comp[18], y);

      // 19 MIX2 8 17 18 16 255
      {
        int err=(y*32767-squash(p[19]))*16>>5;
        int w=comp[19].a16[comp[19].cxt];
        w+=(err*(p[17]-p[18])+(1<<12))>>13;
        if (w<0) w=0;
        if (w>65535) w=65535;
        comp[19].a16[comp[19].cxt]=w;
      }

      // 20 SSE 16 19 32 255
      train(comp[20], y);

      // 21 MIX2 0 19 20 16 0
      {
        int err=(y*32767-squash(p[21]))*16>>5;
        int w=comp[21].a16[comp[21].cxt];
        w+=(err*(p[19]-p[20])+(1<<12))>>13;
        if (w<0) w=0;
        if (w>65535) w=65535;
        comp[21].a16[comp[21].cxt]=w;
      }
      break;
    }
    default: return update0(y);
  }
  c8+=c8+y;
  if (c8>=256) {
    z.run(c8-256);
    hmap4=1;
    c8=1;
  }
  else if (c8>=16 && c8<32)
    hmap4=(hmap4&0xf)<<5|y<<4|1;
  else
    hmap4=(hmap4&0x1f0)|(((hmap4&0xf)*2+y)&0xf);
}

void ZPAQL::run(U32 input) {
  switch(select) {
    case 1: {
      a = input;
      m(b) = a;
      a = 0;
      d = 0;
      a = (a+m(b)+512)*773;
      --b;
      a = (a+m(b)+512)*773;
      h(d) = a;
      ++d;
      --b;
      a = (a+m(b)+512)*773;
      --b;
      a = (a+m(b)+512)*773;
      h(d) = a;
      return;
      break;
    }
    case 2: {
      a = input;
      ++c;
      m(c) = a;
      b = c;
      a = 0;
      d = 1;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      --b;
      a = (a+m(b)+512)*773;
      h(d) = a;
      ++d;
      a = m(c);
      a <<= (8&31);
      h(d) = a;
      return;
      break;
    }
    case 3: {
      a = input;
      ++c;
      m(c) = a;
      b = c;
      a = 0;
      d = 2;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      --b;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = (a+m(b)+512)*773;
      h(d) = a;
      --b;
      ++d;
      a = m(c);
      a &= ~ 32;
      f = (a > U32(64));
      if (!f) goto L300057;
      f = (a < U32(91));
      if (!f) goto L300057;
      ++d;
      h(d) = (h(d)+a+512)*773;
      --d;
      swap(h(d));
      a += h(d);
      a *= 20;
      h(d) = a;
      goto L300066;
L300057:
      a = h(d);
      f = (a == U32(0));
      if (f) goto L300065;
      ++d;
      h(d) = a;
      --d;
L300065:
      h(d) = 0;
L300066:
      ++d;
      ++d;
      b = c;
      --b;
      a = 0;
      a = (a+m(b)+512)*773;
      h(d) = a;
      ++d;
      --b;
      a = 0;
      a = (a+m(b)+512)*773;
      h(d) = a;
      ++d;
      --b;
      a = 0;
      a = (a+m(b)+512)*773;
      h(d) = a;
      ++d;
      a = b;
      a -= 212;
      b = a;
      a = 0;
      a = (a+m(b)+512)*773;
      h(d) = a;
      swap(b);
      a -= 216;
      swap(b);
      a = m(b);
      a &= 60;
      h(d) = (h(d)+a+512)*773;
      ++d;
      a = m(c);
      a <<= (9&31);
      h(d) = a;
      ++d;
      ++d;
      ++d;
      ++d;
      ++d;
      h(d) = a;
      return;
      break;
    }
    default: run0(input);
  }
}
}