/*---------------------------------------------------*/ int BZ_API(BZ2_bzDecompress) ( bz_stream *strm ) { DState* s; s = strm->state; while (True) { if (s->state == BZ_X_IDLE) return BZ_SEQUENCE_ERROR; if (s->state == BZ_X_OUTPUT) { #ifdef NSIS_COMPRESS_BZIP2_SMALLMODE unRLE_obuf_to_output_SMALL ( s ); #else unRLE_obuf_to_output_FAST ( s ); #endif if (s->nblock_used == s->save_nblock+1 && s->state_out_len == 0) { s->state = BZ_X_BLKHDR_1; } else { return BZ_OK; } } if (s->state >= BZ_X_BLKHDR_1) { Int32 r = BZ2_decompress ( s ); if (r == BZ_STREAM_END) { return r; } if (s->state != BZ_X_OUTPUT) return r; } } AssertH ( 0, 6001 ); return 0; /*NOTREACHED*/ }
static void fallbackQSort3(UInt32 *fmap, UInt32 *eclass, Int32 loSt, Int32 hiSt) { Int32 unLo, unHi, ltLo, gtHi, n, m; Int32 sp, lo, hi; UInt32 med, r, r3; Int32 stackLo[FALLBACK_QSORT_STACK_SIZE]; Int32 stackHi[FALLBACK_QSORT_STACK_SIZE]; r = 0; sp = 0; fpush(loSt, hiSt); while ( sp > 0 ) { AssertH(sp < FALLBACK_QSORT_STACK_SIZE - 1, 1004); fpop(lo, hi); if ( hi - lo < FALLBACK_QSORT_SMALL_THRESH ) { fallbackSimpleSort(fmap, eclass, lo, hi); continue; } /* Random partitioning. Median of 3 sometimes fails to * avoid bad cases. Median of 9 seems to help but * looks rather expensive. This too seems to work but * is cheaper. Guidance for the magic constants * 7621 and 32768 is taken from Sedgewick's algorithms * book, chapter 35. */ r = ( (r * 7621) + 1 ) % 32768; r3 = r % 3; if ( r3 == 0 ) { med = eclass[fmap[lo]]; } else { if ( r3 == 1 ) { med = eclass[fmap[(lo + hi) >> 1]]; } else { med = eclass[fmap[hi]]; } }
void fallbackSort ( UInt32* fmap, UInt32* eclass, Int32 nblock, Int32 verb ) { Int32 ftab[257]; Int32 ftabCopy[256]; Int32 H, i, j, k, l, r, cc, cc1; Int32 nNotDone; Int32 nBhtab; UChar* eclass8 = (UChar*)eclass; if (verb >= 4) VPrintf0 ( " bucket sorting ...\n" ); for (i = 0; i < 257; i++) ftab[i] = 0; for (i = 0; i < nblock; i++) ftab[eclass8[i]]++; for (i = 0; i < 256; i++) ftabCopy[i] = ftab[i]; for (i = 1; i < 257; i++) ftab[i] += ftab[i-1]; for (i = 0; i < nblock; i++) { j = eclass8[i] + ftab [i]; } AssertH ( j < 256, 1005 ); }
ABC_NAMESPACE_IMPL_START /*---------------------------------------------------*/ #define WEIGHTOF(zz0) ((zz0) & 0xffffff00) #define DEPTHOF(zz1) ((zz1) & 0x000000ff) #define MYMAX(zz2,zz3) ((zz2) > (zz3) ? (zz2) : (zz3)) #define ADDWEIGHTS(zw1,zw2) \ (WEIGHTOF(zw1)+WEIGHTOF(zw2)) | \ (1 + MYMAX(DEPTHOF(zw1),DEPTHOF(zw2))) #define UPHEAP(z) \ { \ Int32 zz, tmp; \ zz = z; tmp = heap[zz]; \ while (weight[tmp] < weight[heap[zz >> 1]]) { \ heap[zz] = heap[zz >> 1]; \ zz >>= 1; \ } \ heap[zz] = tmp; \ } #define DOWNHEAP(z) \ { \ Int32 zz, yy, tmp; \ zz = z; tmp = heap[zz]; \ while (True) { \ yy = zz << 1; \ if (yy > nHeap) break; \ if (yy < nHeap && \ weight[heap[yy+1]] < weight[heap[yy]]) \ yy++; \ if (weight[tmp] < weight[heap[yy]]) break; \ heap[zz] = heap[yy]; \ zz = yy; \ } \ heap[zz] = tmp; \ } /*---------------------------------------------------*/ void BZ2_hbMakeCodeLengths ( UChar *len, Int32 *freq, Int32 alphaSize, Int32 maxLen ) { /*-- Nodes and heap entries run from 1. Entry 0 for both the heap and nodes is a sentinel. --*/ Int32 nNodes, nHeap, n1, n2, i, j, k; Bool tooLong; Int32 heap [ BZ_MAX_ALPHA_SIZE + 2 ]; Int32 weight [ BZ_MAX_ALPHA_SIZE * 2 ]; Int32 parent [ BZ_MAX_ALPHA_SIZE * 2 ]; for (i = 0; i < alphaSize; i++) weight[i+1] = (freq[i] == 0 ? 1 : freq[i]) << 8; while (True) { nNodes = alphaSize; nHeap = 0; heap[0] = 0; weight[0] = 0; parent[0] = -2; for (i = 1; i <= alphaSize; i++) { parent[i] = -1; nHeap++; heap[nHeap] = i; UPHEAP(nHeap); } AssertH( nHeap < (BZ_MAX_ALPHA_SIZE+2), 2001 ); while (nHeap > 1) { n1 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); n2 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); nNodes++; parent[n1] = parent[n2] = nNodes; weight[nNodes] = ADDWEIGHTS(weight[n1], weight[n2]); parent[nNodes] = -1; nHeap++; heap[nHeap] = nNodes; UPHEAP(nHeap); } AssertH( nNodes < (BZ_MAX_ALPHA_SIZE * 2), 2002 ); tooLong = False; for (i = 1; i <= alphaSize; i++) { j = 0; k = i; while (parent[k] >= 0) { k = parent[k]; j++; } len[i-1] = j; if (j > maxLen) tooLong = True; } if (! tooLong) break; /* 17 Oct 04: keep-going condition for the following loop used to be 'i < alphaSize', which missed the last element, theoretically leading to the possibility of the compressor looping. However, this count-scaling step is only needed if one of the generated Huffman code words is longer than maxLen, which up to and including version 1.0.2 was 20 bits, which is extremely unlikely. In version 1.0.3 maxLen was changed to 17 bits, which has minimal effect on compression ratio, but does mean this scaling step is used from time to time, enough to verify that it works. This means that bzip2-1.0.3 and later will only produce Huffman codes with a maximum length of 17 bits. However, in order to preserve backwards compatibility with bitstreams produced by versions pre-1.0.3, the decompressor must still handle lengths of up to 20. */ for (i = 1; i <= alphaSize; i++) { j = weight[i] >> 8; j = 1 + (j / 2); weight[i] = j << 8; } } }
static void fallbackQSort3 ( UInt32* fmap, UInt32* eclass, Int32 loSt, Int32 hiSt ) { Int32 unLo, unHi, ltLo, gtHi, n, m; Int32 sp, lo, hi; UInt32 med, r, r3; Int32 stackLo[FALLBACK_QSORT_STACK_SIZE]; Int32 stackHi[FALLBACK_QSORT_STACK_SIZE]; r = 0; sp = 0; fpush ( loSt, hiSt ); while (sp > 0) { AssertH ( sp < FALLBACK_QSORT_STACK_SIZE, 1004 ); fpop ( lo, hi ); if (hi - lo < FALLBACK_QSORT_SMALL_THRESH) { fallbackSimpleSort ( fmap, eclass, lo, hi ); continue; } /* Random partitioning. Median of 3 sometimes fails to avoid bad cases. Median of 9 seems to help but looks rather expensive. This too seems to work but is cheaper. Guidance for the magic constants 7621 and 32768 is taken from Sedgewick's algorithms book, chapter 35. */ r = ((r * 7621) + 1) % 32768; r3 = r % 3; if (r3 == 0) med = eclass[fmap[lo]]; else if (r3 == 1) med = eclass[fmap[(lo+hi)>>1]]; else med = eclass[fmap[hi]]; unLo = ltLo = lo; unHi = gtHi = hi; while (1) { while (1) { if (unLo > unHi) break; n = (Int32)eclass[fmap[unLo]] - (Int32)med; if (n == 0) { fswap(fmap[unLo], fmap[ltLo]); ltLo++; unLo++; continue; }; if (n > 0) break; unLo++; } while (1) { if (unLo > unHi) break; n = (Int32)eclass[fmap[unHi]] - (Int32)med; if (n == 0) { fswap(fmap[unHi], fmap[gtHi]); gtHi--; unHi--; continue; }; if (n < 0) break; unHi--; } if (unLo > unHi) break; fswap(fmap[unLo], fmap[unHi]); unLo++; unHi--; } AssertD ( unHi == unLo-1, "fallbackQSort3(2)" ); if (gtHi < ltLo) continue; n = fmin(ltLo-lo, unLo-ltLo); fvswap(lo, unLo-n, n); m = fmin(hi-gtHi, gtHi-unHi); fvswap(unLo, hi-m+1, m); n = lo + unLo - ltLo - 1; m = hi - (gtHi - unHi) + 1; if (n - lo > hi - m) { fpush ( lo, n ); fpush ( m, hi ); } else { fpush ( m, hi ); fpush ( lo, n ); } }
/*---------------------------------------------------*/ void BZ2_hbMakeCodeLengths ( UChar *len, Int32 *freq, Int32 alphaSize, Int32 maxLen ) { /*-- Nodes and heap entries run from 1. Entry 0 for both the heap and nodes is a sentinel. --*/ Int32 nNodes, nHeap, n1, n2, i, j, k; Bool tooLong; Int32 heap [ BZ_MAX_ALPHA_SIZE + 2 ]; Int32 weight [ BZ_MAX_ALPHA_SIZE * 2 ]; Int32 parent [ BZ_MAX_ALPHA_SIZE * 2 ]; for (i = 0; i < alphaSize; i++) weight[i+1] = (freq[i] == 0 ? 1 : freq[i]) << 8; while (True) { nNodes = alphaSize; nHeap = 0; heap[0] = 0; weight[0] = 0; parent[0] = -2; for (i = 1; i <= alphaSize; i++) { parent[i] = -1; nHeap++; heap[nHeap] = i; UPHEAP(nHeap); } AssertH( nHeap < (BZ_MAX_ALPHA_SIZE+2), 2001 ); while (nHeap > 1) { n1 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); n2 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); nNodes++; parent[n1] = parent[n2] = nNodes; weight[nNodes] = ADDWEIGHTS(weight[n1], weight[n2]); parent[nNodes] = -1; nHeap++; heap[nHeap] = nNodes; UPHEAP(nHeap); } AssertH( nNodes < (BZ_MAX_ALPHA_SIZE * 2), 2002 ); tooLong = False; for (i = 1; i <= alphaSize; i++) { j = 0; k = i; while (parent[k] >= 0) { k = parent[k]; j++; } len[i-1] = j; if (j > maxLen) tooLong = True; } if (! tooLong) break; /* 17 Oct 04: keep-going condition for the following loop used to be 'i < alphaSize', which missed the last element, theoretically leading to the possibility of the compressor looping. However, this count-scaling step is only needed if one of the generated Huffman code words is longer than maxLen, which up to and including version 1.0.2 was 20 bits, which is extremely unlikely. In version 1.0.3 maxLen was changed to 17 bits, which has minimal effect on compression ratio, but does mean this scaling step is used from time to time, enough to verify that it works. This means that bzip2-1.0.3 and later will only produce Huffman codes with a maximum length of 17 bits. However, in order to preserve backwards compatibility with bitstreams produced by versions pre-1.0.3, the decompressor must still handle lengths of up to 20. */ for (i = 1; i <= alphaSize; i++) { j = weight[i] >> 8; j = 1 + (j / 2); weight[i] = j << 8; } } }
/*---------------------------------------------------*/ void BZ2_hbMakeCodeLengths ( UChar *len, Int32 *freq, Int32 alphaSize, Int32 maxLen ) { /*-- Nodes and heap entries run from 1. Entry 0 for both the heap and nodes is a sentinel. --*/ Int32 nNodes, nHeap, n1, n2, i, j, k; Bool tooLong; static Int32 heap [ BZ_MAX_ALPHA_SIZE + 2 ]; static Int32 weight [ BZ_MAX_ALPHA_SIZE * 2 ]; static Int32 parent [ BZ_MAX_ALPHA_SIZE * 2 ]; for (i = 0; i < alphaSize; i++) weight[i+1] = (freq[i] == 0 ? 1 : freq[i]) << 8; while (True) { nNodes = alphaSize; nHeap = 0; heap[0] = 0; weight[0] = 0; parent[0] = -2; for (i = 1; i <= alphaSize; i++) { parent[i] = -1; nHeap++; heap[nHeap] = i; UPHEAP(nHeap); } AssertH( nHeap < (BZ_MAX_ALPHA_SIZE+2), 2001 ); while (nHeap > 1) { n1 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); n2 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); nNodes++; parent[n1] = parent[n2] = nNodes; weight[nNodes] = ADDWEIGHTS(weight[n1], weight[n2]); parent[nNodes] = -1; nHeap++; heap[nHeap] = nNodes; UPHEAP(nHeap); } AssertH( nNodes < (BZ_MAX_ALPHA_SIZE * 2), 2002 ); tooLong = False; for (i = 1; i <= alphaSize; i++) { j = 0; k = i; while (parent[k] >= 0) { k = parent[k]; j++; } len[i-1] = j; if (j > maxLen) tooLong = True; } if (! tooLong) break; for (i = 1; i < alphaSize; i++) { j = weight[i] >> 8; j = 1 + (j / 2); weight[i] = j << 8; } } }
static void sendMTFValues ( EState* s ) { Int32 v, t, i, j, gs, ge, totc, bt, bc, iter; Int32 nSelectors, alphaSize, minLen, maxLen, selCtr; Int32 nGroups, nBytes; /*-- UChar len [BZ_N_GROUPS][BZ_MAX_ALPHA_SIZE]; is a global since the decoder also needs it. Int32 code[BZ_N_GROUPS][BZ_MAX_ALPHA_SIZE]; Int32 rfreq[BZ_N_GROUPS][BZ_MAX_ALPHA_SIZE]; are also globals only used in this proc. Made global to keep stack frame size small. --*/ UInt16 cost[BZ_N_GROUPS]; Int32 fave[BZ_N_GROUPS]; UInt16* mtfv = s->mtfv; alphaSize = s->nInUse+2; for (t = 0; t < BZ_N_GROUPS; t++) for (v = 0; v < alphaSize; v++) s->len[t][v] = BZ_GREATER_ICOST; /*--- Decide how many coding tables to use ---*/ AssertH ( s->nMTF > 0, 3001 ); if (s->nMTF < 200) nGroups = 2; else if (s->nMTF < 600) nGroups = 3; else if (s->nMTF < 1200) nGroups = 4; else if (s->nMTF < 2400) nGroups = 5; else nGroups = 6; /*--- Generate an initial set of coding tables ---*/ { Int32 nPart, remF, tFreq, aFreq; nPart = nGroups; remF = s->nMTF; gs = 0; while (nPart > 0) { tFreq = remF / nPart; ge = gs-1; aFreq = 0; while (aFreq < tFreq && ge < alphaSize-1) { ge++; aFreq += s->mtfFreq[ge]; } if (ge > gs && nPart != nGroups && nPart != 1 && ((nGroups-nPart) % 2 == 1)) { aFreq -= s->mtfFreq[ge]; ge--; } for (v = 0; v < alphaSize; v++) if (v >= gs && v <= ge) s->len[nPart-1][v] = BZ_LESSER_ICOST; else s->len[nPart-1][v] = BZ_GREATER_ICOST; nPart--; gs = ge+1; remF -= aFreq; } } /*--- Iterate up to BZ_N_ITERS times to improve the tables. ---*/ for (iter = 0; iter < BZ_N_ITERS; iter++) { for (t = 0; t < nGroups; t++) fave[t] = 0; for (t = 0; t < nGroups; t++) for (v = 0; v < alphaSize; v++) s->rfreq[t][v] = 0; /*--- Set up an auxiliary length table which is used to fast-track the common case (nGroups == 6). ---*/ if (nGroups == 6) { for (v = 0; v < alphaSize; v++) { s->len_pack[v][0] = (s->len[1][v] << 16) | s->len[0][v]; s->len_pack[v][1] = (s->len[3][v] << 16) | s->len[2][v]; s->len_pack[v][2] = (s->len[5][v] << 16) | s->len[4][v]; } } nSelectors = 0; totc = 0; gs = 0; while (True) { /*--- Set group start & end marks. --*/ if (gs >= s->nMTF) break; ge = gs + BZ_G_SIZE - 1; if (ge >= s->nMTF) ge = s->nMTF-1; /*-- Calculate the cost of this group as coded by each of the coding tables. --*/ for (t = 0; t < nGroups; t++) cost[t] = 0; if (nGroups == 6 && 50 == ge-gs+1) { /*--- fast track the common case ---*/ register UInt32 cost01, cost23, cost45; register UInt16 icv; cost01 = cost23 = cost45 = 0; # define BZ_ITER(nn) \ icv = mtfv[gs+(nn)]; \ cost01 += s->len_pack[icv][0]; \ cost23 += s->len_pack[icv][1]; \ cost45 += s->len_pack[icv][2]; \ BZ_ITER(0); BZ_ITER(1); BZ_ITER(2); BZ_ITER(3); BZ_ITER(4); BZ_ITER(5); BZ_ITER(6); BZ_ITER(7); BZ_ITER(8); BZ_ITER(9); BZ_ITER(10); BZ_ITER(11); BZ_ITER(12); BZ_ITER(13); BZ_ITER(14); BZ_ITER(15); BZ_ITER(16); BZ_ITER(17); BZ_ITER(18); BZ_ITER(19); BZ_ITER(20); BZ_ITER(21); BZ_ITER(22); BZ_ITER(23); BZ_ITER(24); BZ_ITER(25); BZ_ITER(26); BZ_ITER(27); BZ_ITER(28); BZ_ITER(29); BZ_ITER(30); BZ_ITER(31); BZ_ITER(32); BZ_ITER(33); BZ_ITER(34); BZ_ITER(35); BZ_ITER(36); BZ_ITER(37); BZ_ITER(38); BZ_ITER(39); BZ_ITER(40); BZ_ITER(41); BZ_ITER(42); BZ_ITER(43); BZ_ITER(44); BZ_ITER(45); BZ_ITER(46); BZ_ITER(47); BZ_ITER(48); BZ_ITER(49); # undef BZ_ITER cost[0] = cost01 & 0xffff; cost[1] = cost01 >> 16; cost[2] = cost23 & 0xffff; cost[3] = cost23 >> 16; cost[4] = cost45 & 0xffff; cost[5] = cost45 >> 16; } else { /*--- slow version which correctly handles all situations ---*/ for (i = gs; i <= ge; i++) { UInt16 icv = mtfv[i]; for (t = 0; t < nGroups; t++) cost[t] += s->len[t][icv]; } } /*-- Find the coding table which is best for this group, and record its identity in the selector table. --*/ bc = 999999999; bt = -1; for (t = 0; t < nGroups; t++) if (cost[t] < bc) { bc = cost[t]; bt = t; }; totc += bc; fave[bt]++; s->selector[nSelectors] = bt; nSelectors++; /*-- Increment the symbol frequencies for the selected table. --*/ if (nGroups == 6 && 50 == ge-gs+1) { /*--- fast track the common case ---*/ # define BZ_ITUR(nn) s->rfreq[bt][ mtfv[gs+(nn)] ]++ BZ_ITUR(0); BZ_ITUR(1); BZ_ITUR(2); BZ_ITUR(3); BZ_ITUR(4); BZ_ITUR(5); BZ_ITUR(6); BZ_ITUR(7); BZ_ITUR(8); BZ_ITUR(9); BZ_ITUR(10); BZ_ITUR(11); BZ_ITUR(12); BZ_ITUR(13); BZ_ITUR(14); BZ_ITUR(15); BZ_ITUR(16); BZ_ITUR(17); BZ_ITUR(18); BZ_ITUR(19); BZ_ITUR(20); BZ_ITUR(21); BZ_ITUR(22); BZ_ITUR(23); BZ_ITUR(24); BZ_ITUR(25); BZ_ITUR(26); BZ_ITUR(27); BZ_ITUR(28); BZ_ITUR(29); BZ_ITUR(30); BZ_ITUR(31); BZ_ITUR(32); BZ_ITUR(33); BZ_ITUR(34); BZ_ITUR(35); BZ_ITUR(36); BZ_ITUR(37); BZ_ITUR(38); BZ_ITUR(39); BZ_ITUR(40); BZ_ITUR(41); BZ_ITUR(42); BZ_ITUR(43); BZ_ITUR(44); BZ_ITUR(45); BZ_ITUR(46); BZ_ITUR(47); BZ_ITUR(48); BZ_ITUR(49); # undef BZ_ITUR } else { /*--- slow version which correctly handles all situations ---*/ for (i = gs; i <= ge; i++) s->rfreq[bt][ mtfv[i] ]++; } gs = ge+1; } /*-- Recompute the tables based on the accumulated frequencies. --*/ for (t = 0; t < nGroups; t++) BZ2_hbMakeCodeLengths ( &(s->len[t][0]), &(s->rfreq[t][0]), alphaSize, 20 ); }
/*---------------------------------------------------*/ void BZ2_hbMakeCodeLengths ( UChar *len, Int32 *freq, Int32 alphaSize, Int32 maxLen ) { /*-- Nodes and heap entries run from 1. Entry 0 for both the heap and nodes is a sentinel. --*/ Int32 nNodes, nHeap, n1, n2, i, j, k; Bool tooLong; Int32 heap [ BZ_MAX_ALPHA_SIZE + 2 ]; Int32 weight [ BZ_MAX_ALPHA_SIZE * 2 ]; Int32 parent [ BZ_MAX_ALPHA_SIZE * 2 ]; for (i = 0; i < alphaSize; i++) weight[i+1] = (freq[i] == 0 ? 1 : freq[i]) << 8; while (True) { nNodes = alphaSize; nHeap = 0; heap[0] = 0; weight[0] = 0; parent[0] = -2; for (i = 1; i <= alphaSize; i++) { parent[i] = -1; nHeap++; heap[nHeap] = i; UPHEAP(nHeap); } AssertH( nHeap < (BZ_MAX_ALPHA_SIZE+2), 2001 ); while (nHeap > 1) { n1 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); n2 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1); nNodes++; parent[n1] = parent[n2] = nNodes; weight[nNodes] = ADDWEIGHTS(weight[n1], weight[n2]); parent[nNodes] = -1; nHeap++; heap[nHeap] = nNodes; UPHEAP(nHeap); } AssertH( nNodes < (BZ_MAX_ALPHA_SIZE * 2), 2002 ); tooLong = False; for (i = 1; i <= alphaSize; i++) { j = 0; k = i; while (parent[k] >= 0) { k = parent[k]; j++; } len[i-1] = j; if (j > maxLen) tooLong = True; } if (! tooLong) break; /* bzip2-1.0.3 and later will only produce Huffman codes with a maximum length of 17 bits. However, in order to preserve backwards compatibility with bitstreams produced by previous versions, the decompressor must still handle lengths of up to 20. */ for (i = 1; i <= alphaSize; i++) { j = weight[i] >> 8; j = 1 + (j / 2); weight[i] = j << 8; } } }