bzip2: Support compression for sandbox

Add the missing code to allow bzip2 compression to be used. This is useful
for sandbox tests. These files are taken from the bzip2 1.0.6 release.
The license text is copied to the top of each file as is done with other
bzip2 files in U-Boot. The only other change is to squash a compiler warning
with nBytes.

Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>

3 files changed, 1849 insertions, 0 deletions

diff --git a/lib/bzip2/Makefile b/lib/bzip2/Makefile
index f0b81ad2c2..585d776ba8 100644
--- a/lib/bzip2/Makefile
+++ b/lib/bzip2/Makefile
@@ -4,3 +4,4 @@
 
 obj-y += bzlib.o bzlib_crctable.o bzlib_decompress.o \
 	bzlib_randtable.o bzlib_huffman.o
+obj-$(CONFIG_SANDBOX) += bzlib_compress.o bzlib_blocksort.o
diff --git a/lib/bzip2/bzlib_blocksort.c b/lib/bzip2/bzlib_blocksort.c
new file mode 100644
index 0000000000..2785521502
--- /dev/null
+++ b/lib/bzip2/bzlib_blocksort.c
@@ -0,0 +1,1134 @@
+
+/*-------------------------------------------------------------*/
+/*--- Block sorting machinery                               ---*/
+/*---                                           blocksort.c ---*/
+/*-------------------------------------------------------------*/
+
+/*--
+  This file is a part of bzip2 and/or libbzip2, a program and
+  library for lossless, block-sorting data compression.
+
+  Copyright (C) 1996-2002 Julian R Seward.  All rights reserved.
+
+  Redistribution and use in source and binary forms, with or without
+  modification, are permitted provided that the following conditions
+  are met:
+
+  1. Redistributions of source code must retain the above copyright
+     notice, this list of conditions and the following disclaimer.
+
+  2. The origin of this software must not be misrepresented; you must
+     not claim that you wrote the original software.  If you use this
+     software in a product, an acknowledgment in the product
+     documentation would be appreciated but is not required.
+
+  3. Altered source versions must be plainly marked as such, and must
+     not be misrepresented as being the original software.
+
+  4. The name of the author may not be used to endorse or promote
+     products derived from this software without specific prior written
+     permission.
+
+  THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
+  OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+  ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+  DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+  DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+  GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+  WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+  Julian Seward, Cambridge, UK.
+  jseward@acm.org
+  bzip2/libbzip2 version 1.0.6 of 6 September 2010
+  Copyright (C) 1996-2010 Julian Seward <jseward@bzip.org>
+
+  This program is based on (at least) the work of:
+     Mike Burrows
+     David Wheeler
+     Peter Fenwick
+     Alistair Moffat
+     Radford Neal
+     Ian H. Witten
+     Robert Sedgewick
+     Jon L. Bentley
+
+  For more information on these sources, see the manual.
+--*/
+
+#include "bzlib_private.h"
+
+/*---------------------------------------------*/
+/*--- Fallback O(N log(N)^2) sorting        ---*/
+/*--- algorithm, for repetitive blocks      ---*/
+/*---------------------------------------------*/
+
+/*---------------------------------------------*/
+static 
+__inline__
+void fallbackSimpleSort ( UInt32* fmap, 
+                          UInt32* eclass, 
+                          Int32   lo, 
+                          Int32   hi )
+{
+   Int32 i, j, tmp;
+   UInt32 ec_tmp;
+
+   if (lo == hi) return;
+
+   if (hi - lo > 3) {
+      for ( i = hi-4; i >= lo; i-- ) {
+         tmp = fmap[i];
+         ec_tmp = eclass[tmp];
+         for ( j = i+4; j <= hi && ec_tmp > eclass[fmap[j]]; j += 4 )
+            fmap[j-4] = fmap[j];
+         fmap[j-4] = tmp;
+      }
+   }
+
+   for ( i = hi-1; i >= lo; i-- ) {
+      tmp = fmap[i];
+      ec_tmp = eclass[tmp];
+      for ( j = i+1; j <= hi && ec_tmp > eclass[fmap[j]]; j++ )
+         fmap[j-1] = fmap[j];
+      fmap[j-1] = tmp;
+   }
+}
+
+
+/*---------------------------------------------*/
+#define fswap(zz1, zz2) \
+   { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }
+
+#define fvswap(zzp1, zzp2, zzn)       \
+{                                     \
+   Int32 yyp1 = (zzp1);               \
+   Int32 yyp2 = (zzp2);               \
+   Int32 yyn  = (zzn);                \
+   while (yyn > 0) {                  \
+      fswap(fmap[yyp1], fmap[yyp2]);  \
+      yyp1++; yyp2++; yyn--;          \
+   }                                  \
+}
+
+
+#define fmin(a,b) ((a) < (b)) ? (a) : (b)
+
+#define fpush(lz,hz) { stackLo[sp] = lz; \
+                       stackHi[sp] = hz; \
+                       sp++; }
+
+#define fpop(lz,hz) { sp--;              \
+                      lz = stackLo[sp];  \
+                      hz = stackHi[sp]; }
+
+#define FALLBACK_QSORT_SMALL_THRESH 10
+#define FALLBACK_QSORT_STACK_SIZE   100
+
+
+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]];
+
+      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 );
+      }
+   }
+}
+
+#undef fmin
+#undef fpush
+#undef fpop
+#undef fswap
+#undef fvswap
+#undef FALLBACK_QSORT_SMALL_THRESH
+#undef FALLBACK_QSORT_STACK_SIZE
+
+
+/*---------------------------------------------*/
+/* Pre:
+      nblock > 0
+      eclass exists for [0 .. nblock-1]
+      ((UChar*)eclass) [0 .. nblock-1] holds block
+      ptr exists for [0 .. nblock-1]
+
+   Post:
+      ((UChar*)eclass) [0 .. nblock-1] holds block
+      All other areas of eclass destroyed
+      fmap [0 .. nblock-1] holds sorted order
+      bhtab [ 0 .. 2+(nblock/32) ] destroyed
+*/
+
+#define       SET_BH(zz)  bhtab[(zz) >> 5] |= (1 << ((zz) & 31))
+#define     CLEAR_BH(zz)  bhtab[(zz) >> 5] &= ~(1 << ((zz) & 31))
+#define     ISSET_BH(zz)  (bhtab[(zz) >> 5] & (1 << ((zz) & 31)))
+#define      WORD_BH(zz)  bhtab[(zz) >> 5]
+#define UNALIGNED_BH(zz)  ((zz) & 0x01f)
+
+static
+void fallbackSort ( UInt32* fmap, 
+                    UInt32* eclass, 
+                    UInt32* bhtab,
+                    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;
+
+   /*--
+      Initial 1-char radix sort to generate
+      initial fmap and initial BH bits.
+   --*/
+   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];
+      k = ftab[j] - 1;
+      ftab[j] = k;
+      fmap[k] = i;
+   }
+
+   nBhtab = 2 + (nblock / 32);
+   for (i = 0; i < nBhtab; i++) bhtab[i] = 0;
+   for (i = 0; i < 256; i++) SET_BH(ftab[i]);
+
+   /*--
+      Inductively refine the buckets.  Kind-of an
+      "exponential radix sort" (!), inspired by the
+      Manber-Myers suffix array construction algorithm.
+   --*/
+
+   /*-- set sentinel bits for block-end detection --*/
+   for (i = 0; i < 32; i++) { 
+      SET_BH(nblock + 2*i);
+      CLEAR_BH(nblock + 2*i + 1);
+   }
+
+   /*-- the log(N) loop --*/
+   H = 1;
+   while (1) {
+
+      if (verb >= 4) 
+         VPrintf1 ( "        depth %6d has ", H );
+
+      j = 0;
+      for (i = 0; i < nblock; i++) {
+         if (ISSET_BH(i)) j = i;
+         k = fmap[i] - H; if (k < 0) k += nblock;
+         eclass[k] = j;
+      }
+
+      nNotDone = 0;
+      r = -1;
+      while (1) {
+
+	 /*-- find the next non-singleton bucket --*/
+         k = r + 1;
+         while (ISSET_BH(k) && UNALIGNED_BH(k)) k++;
+         if (ISSET_BH(k)) {
+            while (WORD_BH(k) == 0xffffffff) k += 32;
+            while (ISSET_BH(k)) k++;
+         }
+         l = k - 1;
+         if (l >= nblock) break;
+         while (!ISSET_BH(k) && UNALIGNED_BH(k)) k++;
+         if (!ISSET_BH(k)) {
+            while (WORD_BH(k) == 0x00000000) k += 32;
+            while (!ISSET_BH(k)) k++;
+         }
+         r = k - 1;
+         if (r >= nblock) break;
+
+         /*-- now [l, r] bracket current bucket --*/
+         if (r > l) {
+            nNotDone += (r - l + 1);
+            fallbackQSort3 ( fmap, eclass, l, r );
+
+            /*-- scan bucket and generate header bits-- */
+            cc = -1;
+            for (i = l; i <= r; i++) {
+               cc1 = eclass[fmap[i]];
+               if (cc != cc1) { SET_BH(i); cc = cc1; };
+            }
+         }
+      }
+
+      if (verb >= 4) 
+         VPrintf1 ( "%6d unresolved strings\n", nNotDone );
+
+      H *= 2;
+      if (H > nblock || nNotDone == 0) break;
+   }
+
+   /*-- 
+      Reconstruct the original block in
+      eclass8 [0 .. nblock-1], since the
+      previous phase destroyed it.
+   --*/
+   if (verb >= 4)
+      VPrintf0 ( "        reconstructing block ...\n" );
+   j = 0;
+   for (i = 0; i < nblock; i++) {
+      while (ftabCopy[j] == 0) j++;
+      ftabCopy[j]--;
+      eclass8[fmap[i]] = (UChar)j;
+   }
+   AssertH ( j < 256, 1005 );
+}
+
+#undef       SET_BH
+#undef     CLEAR_BH
+#undef     ISSET_BH
+#undef      WORD_BH
+#undef UNALIGNED_BH
+
+
+/*---------------------------------------------*/
+/*--- The main, O(N^2 log(N)) sorting       ---*/
+/*--- algorithm.  Faster for "normal"       ---*/
+/*--- non-repetitive blocks.                ---*/
+/*---------------------------------------------*/
+
+/*---------------------------------------------*/
+static
+__inline__
+Bool mainGtU ( UInt32  i1, 
+               UInt32  i2,
+               UChar*  block, 
+               UInt16* quadrant,
+               UInt32  nblock,
+               Int32*  budget )
+{
+   Int32  k;
+   UChar  c1, c2;
+   UInt16 s1, s2;
+
+   AssertD ( i1 != i2, "mainGtU" );
+   /* 1 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 2 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 3 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 4 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 5 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 6 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 7 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 8 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 9 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 10 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 11 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 12 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+
+   k = nblock + 8;
+
+   do {
+      /* 1 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 2 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 3 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 4 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 5 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 6 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 7 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 8 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+
+      if (i1 >= nblock) i1 -= nblock;
+      if (i2 >= nblock) i2 -= nblock;
+
+      k -= 8;
+      (*budget)--;
+   }
+      while (k >= 0);
+
+   return False;
+}
+
+
+/*---------------------------------------------*/
+/*--
+   Knuth's increments seem to work better
+   than Incerpi-Sedgewick here.  Possibly
+   because the number of elems to sort is
+   usually small, typically <= 20.
+--*/
+static
+Int32 incs[14] = { 1, 4, 13, 40, 121, 364, 1093, 3280,
+                   9841, 29524, 88573, 265720,
+                   797161, 2391484 };
+
+static
+void mainSimpleSort ( UInt32* ptr,
+                      UChar*  block,
+                      UInt16* quadrant,
+                      Int32   nblock,
+                      Int32   lo, 
+                      Int32   hi, 
+                      Int32   d,
+                      Int32*  budget )
+{
+   Int32 i, j, h, bigN, hp;
+   UInt32 v;
+
+   bigN = hi - lo + 1;
+   if (bigN < 2) return;
+
+   hp = 0;
+   while (incs[hp] < bigN) hp++;
+   hp--;
+
+   for (; hp >= 0; hp--) {
+      h = incs[hp];
+
+      i = lo + h;
+      while (True) {
+
+         /*-- copy 1 --*/
+         if (i > hi) break;
+         v = ptr[i];
+         j = i;
+         while ( mainGtU ( 
+                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
+                 ) ) {
+            ptr[j] = ptr[j-h];
+            j = j - h;
+            if (j <= (lo + h - 1)) break;
+         }
+         ptr[j] = v;
+         i++;
+
+         /*-- copy 2 --*/
+         if (i > hi) break;
+         v = ptr[i];
+         j = i;
+         while ( mainGtU ( 
+                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
+                 ) ) {
+            ptr[j] = ptr[j-h];
+            j = j - h;
+            if (j <= (lo + h - 1)) break;
+         }
+         ptr[j] = v;
+         i++;
+
+         /*-- copy 3 --*/
+         if (i > hi) break;
+         v = ptr[i];
+         j = i;
+         while ( mainGtU ( 
+                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
+                 ) ) {
+            ptr[j] = ptr[j-h];
+            j = j - h;
+            if (j <= (lo + h - 1)) break;
+         }
+         ptr[j] = v;
+         i++;
+
+         if (*budget < 0) return;
+      }
+   }
+}
+
+
+/*---------------------------------------------*/
+/*--
+   The following is an implementation of
+   an elegant 3-way quicksort for strings,
+   described in a paper "Fast Algorithms for
+   Sorting and Searching Strings", by Robert
+   Sedgewick and Jon L. Bentley.
+--*/
+
+#define mswap(zz1, zz2) \
+   { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }
+
+#define mvswap(zzp1, zzp2, zzn)       \
+{                                     \
+   Int32 yyp1 = (zzp1);               \
+   Int32 yyp2 = (zzp2);               \
+   Int32 yyn  = (zzn);                \
+   while (yyn > 0) {                  \
+      mswap(ptr[yyp1], ptr[yyp2]);    \
+      yyp1++; yyp2++; yyn--;          \
+   }                                  \
+}
+
+static 
+__inline__
+UChar mmed3 ( UChar a, UChar b, UChar c )
+{
+   UChar t;
+   if (a > b) { t = a; a = b; b = t; };
+   if (b > c) { 
+      b = c;
+      if (a > b) b = a;
+   }
+   return b;
+}
+
+#define mmin(a,b) ((a) < (b)) ? (a) : (b)
+
+#define mpush(lz,hz,dz) { stackLo[sp] = lz; \
+                          stackHi[sp] = hz; \
+                          stackD [sp] = dz; \
+                          sp++; }
+
+#define mpop(lz,hz,dz) { sp--;             \
+                         lz = stackLo[sp]; \
+                         hz = stackHi[sp]; \
+                         dz = stackD [sp]; }
+
+
+#define mnextsize(az) (nextHi[az]-nextLo[az])
+
+#define mnextswap(az,bz)                                        \
+   { Int32 tz;                                                  \
+     tz = nextLo[az]; nextLo[az] = nextLo[bz]; nextLo[bz] = tz; \
+     tz = nextHi[az]; nextHi[az] = nextHi[bz]; nextHi[bz] = tz; \
+     tz = nextD [az]; nextD [az] = nextD [bz]; nextD [bz] = tz; }
+
+
+#define MAIN_QSORT_SMALL_THRESH 20
+#define MAIN_QSORT_DEPTH_THRESH (BZ_N_RADIX + BZ_N_QSORT)
+#define MAIN_QSORT_STACK_SIZE 100
+
+static
+void mainQSort3 ( UInt32* ptr,
+                  UChar*  block,
+                  UInt16* quadrant,
+                  Int32   nblock,
+                  Int32   loSt, 
+                  Int32   hiSt, 
+                  Int32   dSt,
+                  Int32*  budget )
+{
+   Int32 unLo, unHi, ltLo, gtHi, n, m, med;
+   Int32 sp, lo, hi, d;
+
+   Int32 stackLo[MAIN_QSORT_STACK_SIZE];
+   Int32 stackHi[MAIN_QSORT_STACK_SIZE];
+   Int32 stackD [MAIN_QSORT_STACK_SIZE];
+
+   Int32 nextLo[3];
+   Int32 nextHi[3];
+   Int32 nextD [3];
+
+   sp = 0;
+   mpush ( loSt, hiSt, dSt );
+
+   while (sp > 0) {
+
+      AssertH ( sp < MAIN_QSORT_STACK_SIZE - 2, 1001 );
+
+      mpop ( lo, hi, d );
+      if (hi - lo < MAIN_QSORT_SMALL_THRESH || 
+          d > MAIN_QSORT_DEPTH_THRESH) {
+         mainSimpleSort ( ptr, block, quadrant, nblock, lo, hi, d, budget );
+         if (*budget < 0) return;
+         continue;
+      }
+
+      med = (Int32) 
+            mmed3 ( block[ptr[ lo         ]+d],
+                    block[ptr[ hi         ]+d],
+                    block[ptr[ (lo+hi)>>1 ]+d] );
+
+      unLo = ltLo = lo;
+      unHi = gtHi = hi;
+
+      while (True) {
+         while (True) {
+            if (unLo > unHi) break;
+            n = ((Int32)block[ptr[unLo]+d]) - med;
+            if (n == 0) { 
+               mswap(ptr[unLo], ptr[ltLo]); 
+               ltLo++; unLo++; continue; 
+            };
+            if (n >  0) break;
+            unLo++;
+         }
+         while (True) {
+            if (unLo > unHi) break;
+            n = ((Int32)block[ptr[unHi]+d]) - med;
+            if (n == 0) { 
+               mswap(ptr[unHi], ptr[gtHi]); 
+               gtHi--; unHi--; continue; 
+            };
+            if (n <  0) break;
+            unHi--;
+         }
+         if (unLo > unHi) break;
+         mswap(ptr[unLo], ptr[unHi]); unLo++; unHi--;
+      }
+
+      AssertD ( unHi == unLo-1, "mainQSort3(2)" );
+
+      if (gtHi < ltLo) {
+         mpush(lo, hi, d+1 );
+         continue;
+      }
+
+      n = mmin(ltLo-lo, unLo-ltLo); mvswap(lo, unLo-n, n);
+      m = mmin(hi-gtHi, gtHi-unHi); mvswap(unLo, hi-m+1, m);
+
+      n = lo + unLo - ltLo - 1;
+      m = hi - (gtHi - unHi) + 1;
+
+      nextLo[0] = lo;  nextHi[0] = n;   nextD[0] = d;
+      nextLo[1] = m;   nextHi[1] = hi;  nextD[1] = d;
+      nextLo[2] = n+1; nextHi[2] = m-1; nextD[2] = d+1;
+
+      if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);
+      if (mnextsize(1) < mnextsize(2)) mnextswap(1,2);
+      if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);
+
+      AssertD (mnextsize(0) >= mnextsize(1), "mainQSort3(8)" );
+      AssertD (mnextsize(1) >= mnextsize(2), "mainQSort3(9)" );
+
+      mpush (nextLo[0], nextHi[0], nextD[0]);
+      mpush (nextLo[1], nextHi[1], nextD[1]);
+      mpush (nextLo[2], nextHi[2], nextD[2]);
+   }
+}
+
+#undef mswap
+#undef mvswap
+#undef mpush
+#undef mpop
+#undef mmin
+#undef mnextsize
+#undef mnextswap
+#undef MAIN_QSORT_SMALL_THRESH
+#undef MAIN_QSORT_DEPTH_THRESH
+#undef MAIN_QSORT_STACK_SIZE
+
+
+/*---------------------------------------------*/
+/* Pre:
+      nblock > N_OVERSHOOT
+      block32 exists for [0 .. nblock-1 +N_OVERSHOOT]
+      ((UChar*)block32) [0 .. nblock-1] holds block
+      ptr exists for [0 .. nblock-1]
+
+   Post:
+      ((UChar*)block32) [0 .. nblock-1] holds block
+      All other areas of block32 destroyed
+      ftab [0 .. 65536 ] destroyed
+      ptr [0 .. nblock-1] holds sorted order
+      if (*budget < 0), sorting was abandoned
+*/
+
+#define BIGFREQ(b) (ftab[((b)+1) << 8] - ftab[(b) << 8])
+#define SETMASK (1 << 21)
+#define CLEARMASK (~(SETMASK))
+
+static
+void mainSort ( UInt32* ptr, 
+                UChar*  block,
+                UInt16* quadrant, 
+                UInt32* ftab,
+                Int32   nblock,
+                Int32   verb,
+                Int32*  budget )
+{
+   Int32  i, j, k, ss, sb;
+   Int32  runningOrder[256];
+   Bool   bigDone[256];
+   Int32  copyStart[256];
+   Int32  copyEnd  [256];
+   UChar  c1;
+   Int32  numQSorted;
+   UInt16 s;
+   if (verb >= 4) VPrintf0 ( "        main sort initialise ...\n" );
+
+   /*-- set up the 2-byte frequency table --*/
+   for (i = 65536; i >= 0; i--) ftab[i] = 0;
+
+   j = block[0] << 8;
+   i = nblock-1;
+   for (; i >= 3; i -= 4) {
+      quadrant[i] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i]) << 8);
+      ftab[j]++;
+      quadrant[i-1] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i-1]) << 8);
+      ftab[j]++;
+      quadrant[i-2] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i-2]) << 8);
+      ftab[j]++;
+      quadrant[i-3] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i-3]) << 8);
+      ftab[j]++;
+   }
+   for (; i >= 0; i--) {
+      quadrant[i] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i]) << 8);
+      ftab[j]++;
+   }
+
+   /*-- (emphasises close relationship of block & quadrant) --*/
+   for (i = 0; i < BZ_N_OVERSHOOT; i++) {
+      block   [nblock+i] = block[i];
+      quadrant[nblock+i] = 0;
+   }
+
+   if (verb >= 4) VPrintf0 ( "        bucket sorting ...\n" );
+
+   /*-- Complete the initial radix sort --*/
+   for (i = 1; i <= 65536; i++) ftab[i] += ftab[i-1];
+
+   s = block[0] << 8;
+   i = nblock-1;
+   for (; i >= 3; i -= 4) {
+      s = (s >> 8) | (block[i] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i;
+      s = (s >> 8) | (block[i-1] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i-1;
+      s = (s >> 8) | (block[i-2] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i-2;
+      s = (s >> 8) | (block[i-3] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i-3;
+   }
+   for (; i >= 0; i--) {
+      s = (s >> 8) | (block[i] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i;
+   }
+
+   /*--
+      Now ftab contains the first loc of every small bucket.
+      Calculate the running order, from smallest to largest
+      big bucket.
+   --*/
+   for (i = 0; i <= 255; i++) {
+      bigDone     [i] = False;
+      runningOrder[i] = i;
+   }
+
+   {
+      Int32 vv;
+      Int32 h = 1;
+      do h = 3 * h + 1; while (h <= 256);
+      do {
+         h = h / 3;
+         for (i = h; i <= 255; i++) {
+            vv = runningOrder[i];
+            j = i;
+            while ( BIGFREQ(runningOrder[j-h]) > BIGFREQ(vv) ) {
+               runningOrder[j] = runningOrder[j-h];
+               j = j - h;
+               if (j <= (h - 1)) goto zero;
+            }
+            zero:
+            runningOrder[j] = vv;
+         }
+      } while (h != 1);
+   }
+
+   /*--
+      The main sorting loop.
+   --*/
+
+   numQSorted = 0;
+
+   for (i = 0; i <= 255; i++) {
+
+      /*--
+         Process big buckets, starting with the least full.
+         Basically this is a 3-step process in which we call
+         mainQSort3 to sort the small buckets [ss, j], but
+         also make a big effort to avoid the calls if we can.
+      --*/
+      ss = runningOrder[i];
+
+      /*--
+         Step 1:
+         Complete the big bucket [ss] by quicksorting
+         any unsorted small buckets [ss, j], for j != ss.  
+         Hopefully previous pointer-scanning phases have already
+         completed many of the small buckets [ss, j], so
+         we don't have to sort them at all.
+      --*/
+      for (j = 0; j <= 255; j++) {
+         if (j != ss) {
+            sb = (ss << 8) + j;
+            if ( ! (ftab[sb] & SETMASK) ) {
+               Int32 lo = ftab[sb]   & CLEARMASK;
+               Int32 hi = (ftab[sb+1] & CLEARMASK) - 1;
+               if (hi > lo) {
+                  if (verb >= 4)
+                     VPrintf4 ( "        qsort [0x%x, 0x%x]   "
+                                "done %d   this %d\n",
+                                ss, j, numQSorted, hi - lo + 1 );
+                  mainQSort3 ( 
+                     ptr, block, quadrant, nblock, 
+                     lo, hi, BZ_N_RADIX, budget 
+                  );   
+                  numQSorted += (hi - lo + 1);
+                  if (*budget < 0) return;
+               }
+            }
+            ftab[sb] |= SETMASK;
+         }
+      }
+
+      AssertH ( !bigDone[ss], 1006 );
+
+      /*--
+         Step 2:
+         Now scan this big bucket [ss] so as to synthesise the
+         sorted order for small buckets [t, ss] for all t,
+         including, magically, the bucket [ss,ss] too.
+         This will avoid doing Real Work in subsequent Step 1's.
+      --*/
+      {
+         for (j = 0; j <= 255; j++) {
+            copyStart[j] =  ftab[(j << 8) + ss]     & CLEARMASK;
+            copyEnd  [j] = (ftab[(j << 8) + ss + 1] & CLEARMASK) - 1;
+         }
+         for (j = ftab[ss << 8] & CLEARMASK; j < copyStart[ss]; j++) {
+            k = ptr[j]-1; if (k < 0) k += nblock;
+            c1 = block[k];
+            if (!bigDone[c1])
+               ptr[ copyStart[c1]++ ] = k;
+         }
+         for (j = (ftab[(ss+1) << 8] & CLEARMASK) - 1; j > copyEnd[ss]; j--) {
+            k = ptr[j]-1; if (k < 0) k += nblock;
+            c1 = block[k];
+            if (!bigDone[c1]) 
+               ptr[ copyEnd[c1]-- ] = k;
+         }
+      }
+
+      AssertH ( (copyStart[ss]-1 == copyEnd[ss])
+                || 
+                /* Extremely rare case missing in bzip2-1.0.0 and 1.0.1.
+                   Necessity for this case is demonstrated by compressing 
+                   a sequence of approximately 48.5 million of character 
+                   251; 1.0.0/1.0.1 will then die here. */
+                (copyStart[ss] == 0 && copyEnd[ss] == nblock-1),
+                1007 )
+
+      for (j = 0; j <= 255; j++) ftab[(j << 8) + ss] |= SETMASK;
+
+      /*--
+         Step 3:
+         The [ss] big bucket is now done.  Record this fact,
+         and update the quadrant descriptors.  Remember to
+         update quadrants in the overshoot area too, if
+         necessary.  The "if (i < 255)" test merely skips
+         this updating for the last bucket processed, since
+         updating for the last bucket is pointless.
+
+         The quadrant array provides a way to incrementally
+         cache sort orderings, as they appear, so as to 
+         make subsequent comparisons in fullGtU() complete
+         faster.  For repetitive blocks this makes a big
+         difference (but not big enough to be able to avoid
+         the fallback sorting mechanism, exponential radix sort).
+
+         The precise meaning is: at all times:
+
+            for 0 <= i < nblock and 0 <= j <= nblock
+
+            if block[i] != block[j], 
+
+               then the relative values of quadrant[i] and 
+                    quadrant[j] are meaningless.
+
+               else {
+                  if quadrant[i] < quadrant[j]
+                     then the string starting at i lexicographically
+                     precedes the string starting at j
+
+                  else if quadrant[i] > quadrant[j]
+                     then the string starting at j lexicographically
+                     precedes the string starting at i
+
+                  else
+                     the relative ordering of the strings starting
+                     at i and j has not yet been determined.
+               }
+      --*/
+      bigDone[ss] = True;
+
+      if (i < 255) {
+         Int32 bbStart  = ftab[ss << 8] & CLEARMASK;
+         Int32 bbSize   = (ftab[(ss+1) << 8] & CLEARMASK) - bbStart;
+         Int32 shifts   = 0;
+
+         while ((bbSize >> shifts) > 65534) shifts++;
+
+         for (j = bbSize-1; j >= 0; j--) {
+            Int32 a2update     = ptr[bbStart + j];
+            UInt16 qVal        = (UInt16)(j >> shifts);
+            quadrant[a2update] = qVal;
+            if (a2update < BZ_N_OVERSHOOT)
+               quadrant[a2update + nblock] = qVal;
+         }
+         AssertH ( ((bbSize-1) >> shifts) <= 65535, 1002 );
+      }
+
+   }
+
+   if (verb >= 4)
+      VPrintf3 ( "        %d pointers, %d sorted, %d scanned\n",
+                 nblock, numQSorted, nblock - numQSorted );
+}
+
+#undef BIGFREQ
+#undef SETMASK
+#undef CLEARMASK
+
+
+/*---------------------------------------------*/
+/* Pre:
+      nblock > 0
+      arr2 exists for [0 .. nblock-1 +N_OVERSHOOT]
+      ((UChar*)arr2)  [0 .. nblock-1] holds block
+      arr1 exists for [0 .. nblock-1]
+
+   Post:
+      ((UChar*)arr2) [0 .. nblock-1] holds block
+      All other areas of block destroyed
+      ftab [ 0 .. 65536 ] destroyed
+      arr1 [0 .. nblock-1] holds sorted order
+*/
+void BZ2_blockSort ( EState* s )
+{
+   UInt32* ptr    = s->ptr; 
+   UChar*  block  = s->block;
+   UInt32* ftab   = s->ftab;
+   Int32   nblock = s->nblock;
+   Int32   verb   = s->verbosity;
+   Int32   wfact  = s->workFactor;
+   UInt16* quadrant;
+   Int32   budget;
+   Int32   budgetInit;
+   Int32   i;
+
+   if (nblock < 10000) {
+      fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
+   } else {
+      /* Calculate the location for quadrant, remembering to get
+         the alignment right.  Assumes that &(block[0]) is at least
+         2-byte aligned -- this should be ok since block is really
+         the first section of arr2.
+      */
+      i = nblock+BZ_N_OVERSHOOT;
+      if (i & 1) i++;
+      quadrant = (UInt16*)(&(block[i]));
+
+      /* (wfact-1) / 3 puts the default-factor-30
+         transition point at very roughly the same place as 
+         with v0.1 and v0.9.0.  
+         Not that it particularly matters any more, since the
+         resulting compressed stream is now the same regardless
+         of whether or not we use the main sort or fallback sort.
+      */
+      if (wfact < 1  ) wfact = 1;
+      if (wfact > 100) wfact = 100;
+      budgetInit = nblock * ((wfact-1) / 3);
+      budget = budgetInit;
+
+      mainSort ( ptr, block, quadrant, ftab, nblock, verb, &budget );
+      if (verb >= 3) 
+         VPrintf3 ( "      %d work, %d block, ratio %5.2f\n",
+                    budgetInit - budget,
+                    nblock, 
+                    (float)(budgetInit - budget) /
+                    (float)(nblock==0 ? 1 : nblock) ); 
+      if (budget < 0) {
+         if (verb >= 2) 
+            VPrintf0 ( "    too repetitive; using fallback"
+                       " sorting algorithm\n" );
+         fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
+      }
+   }
+
+   s->origPtr = -1;
+   for (i = 0; i < s->nblock; i++)
+      if (ptr[i] == 0)
+         { s->origPtr = i; break; };
+
+   AssertH( s->origPtr != -1, 1003 );
+}
+
+
+/*-------------------------------------------------------------*/
+/*--- end                                       blocksort.c ---*/
+/*-------------------------------------------------------------*/
diff --git a/lib/bzip2/bzlib_compress.c b/lib/bzip2/bzlib_compress.c
new file mode 100644
index 0000000000..c8da1c72e9
--- /dev/null
+++ b/lib/bzip2/bzlib_compress.c
@@ -0,0 +1,714 @@
+
+/*-------------------------------------------------------------*/
+/*--- Compression machinery (not incl block sorting)        ---*/
+/*---                                            compress.c ---*/
+/*-------------------------------------------------------------*/
+
+/*--
+  This file is a part of bzip2 and/or libbzip2, a program and
+  library for lossless, block-sorting data compression.
+
+  Copyright (C) 1996-2002 Julian R Seward.  All rights reserved.
+
+  Redistribution and use in source and binary forms, with or without
+  modification, are permitted provided that the following conditions
+  are met:
+
+  1. Redistributions of source code must retain the above copyright
+     notice, this list of conditions and the following disclaimer.
+
+  2. The origin of this software must not be misrepresented; you must
+     not claim that you wrote the original software.  If you use this
+     software in a product, an acknowledgment in the product
+     documentation would be appreciated but is not required.
+
+  3. Altered source versions must be plainly marked as such, and must
+     not be misrepresented as being the original software.
+
+  4. The name of the author may not be used to endorse or promote
+     products derived from this software without specific prior written
+     permission.
+
+  THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
+  OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+  ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+  DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+  DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+  GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+  WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+  Julian Seward, Cambridge, UK.
+  jseward@acm.org
+  bzip2/libbzip2 version 1.0.6 of 6 September 2010
+  Copyright (C) 1996-2010 Julian Seward <jseward@bzip.org>
+
+  This program is based on (at least) the work of:
+     Mike Burrows
+     David Wheeler
+     Peter Fenwick
+     Alistair Moffat
+     Radford Neal
+     Ian H. Witten
+     Robert Sedgewick
+     Jon L. Bentley
+
+  For more information on these sources, see the manual.
+--*/
+
+/* CHANGES
+    0.9.0    -- original version.
+    0.9.0a/b -- no changes in this file.
+    0.9.0c   -- changed setting of nGroups in sendMTFValues() 
+                so as to do a bit better on small files
+*/
+
+#include "bzlib_private.h"
+
+
+/*---------------------------------------------------*/
+/*--- Bit stream I/O                              ---*/
+/*---------------------------------------------------*/
+
+/*---------------------------------------------------*/
+void BZ2_bsInitWrite ( EState* s )
+{
+   s->bsLive = 0;
+   s->bsBuff = 0;
+}
+
+
+/*---------------------------------------------------*/
+static
+void bsFinishWrite ( EState* s )
+{
+   while (s->bsLive > 0) {
+      s->zbits[s->numZ] = (UChar)(s->bsBuff >> 24);
+      s->numZ++;
+      s->bsBuff <<= 8;
+      s->bsLive -= 8;
+   }
+}
+
+
+/*---------------------------------------------------*/
+#define bsNEEDW(nz)                           \
+{                                             \
+   while (s->bsLive >= 8) {                   \
+      s->zbits[s->numZ]                       \
+         = (UChar)(s->bsBuff >> 24);          \
+      s->numZ++;                              \
+      s->bsBuff <<= 8;                        \
+      s->bsLive -= 8;                         \
+   }                                          \
+}
+
+
+/*---------------------------------------------------*/
+static
+__inline__
+void bsW ( EState* s, Int32 n, UInt32 v )
+{
+   bsNEEDW ( n );
+   s->bsBuff |= (v << (32 - s->bsLive - n));
+   s->bsLive += n;
+}
+
+
+/*---------------------------------------------------*/
+static
+void bsPutUInt32 ( EState* s, UInt32 u )
+{
+   bsW ( s, 8, (u >> 24) & 0xffL );
+   bsW ( s, 8, (u >> 16) & 0xffL );
+   bsW ( s, 8, (u >>  8) & 0xffL );
+   bsW ( s, 8,  u        & 0xffL );
+}
+
+
+/*---------------------------------------------------*/
+static
+void bsPutUChar ( EState* s, UChar c )
+{
+   bsW( s, 8, (UInt32)c );
+}
+
+
+/*---------------------------------------------------*/
+/*--- The back end proper                         ---*/
+/*---------------------------------------------------*/
+
+/*---------------------------------------------------*/
+static
+void makeMaps_e ( EState* s )
+{
+   Int32 i;
+   s->nInUse = 0;
+   for (i = 0; i < 256; i++)
+      if (s->inUse[i]) {
+         s->unseqToSeq[i] = s->nInUse;
+         s->nInUse++;
+      }
+}
+
+
+/*---------------------------------------------------*/
+static
+void generateMTFValues ( EState* s )
+{
+   UChar   yy[256];
+   Int32   i, j;
+   Int32   zPend;
+   Int32   wr;
+   Int32   EOB;
+
+   /* 
+      After sorting (eg, here),
+         s->arr1 [ 0 .. s->nblock-1 ] holds sorted order,
+         and
+         ((UChar*)s->arr2) [ 0 .. s->nblock-1 ] 
+         holds the original block data.
+
+      The first thing to do is generate the MTF values,
+      and put them in
+         ((UInt16*)s->arr1) [ 0 .. s->nblock-1 ].
+      Because there are strictly fewer or equal MTF values
+      than block values, ptr values in this area are overwritten
+      with MTF values only when they are no longer needed.
+
+      The final compressed bitstream is generated into the
+      area starting at
+         (UChar*) (&((UChar*)s->arr2)[s->nblock])
+
+      These storage aliases are set up in bzCompressInit(),
+      except for the last one, which is arranged in 
+      compressBlock().
+   */
+   UInt32* ptr   = s->ptr;
+   UChar* block  = s->block;
+   UInt16* mtfv  = s->mtfv;
+
+   makeMaps_e ( s );
+   EOB = s->nInUse+1;
+
+   for (i = 0; i <= EOB; i++) s->mtfFreq[i] = 0;
+
+   wr = 0;
+   zPend = 0;
+   for (i = 0; i < s->nInUse; i++) yy[i] = (UChar) i;
+
+   for (i = 0; i < s->nblock; i++) {
+      UChar ll_i;
+      AssertD ( wr <= i, "generateMTFValues(1)" );
+      j = ptr[i]-1; if (j < 0) j += s->nblock;
+      ll_i = s->unseqToSeq[block[j]];
+      AssertD ( ll_i < s->nInUse, "generateMTFValues(2a)" );
+
+      if (yy[0] == ll_i) { 
+         zPend++;
+      } else {
+
+         if (zPend > 0) {
+            zPend--;
+            while (True) {
+               if (zPend & 1) {
+                  mtfv[wr] = BZ_RUNB; wr++; 
+                  s->mtfFreq[BZ_RUNB]++; 
+               } else {
+                  mtfv[wr] = BZ_RUNA; wr++; 
+                  s->mtfFreq[BZ_RUNA]++; 
+               }
+               if (zPend < 2) break;
+               zPend = (zPend - 2) / 2;
+            };
+            zPend = 0;
+         }
+         {
+            register UChar  rtmp;
+            register UChar* ryy_j;
+            register UChar  rll_i;
+            rtmp  = yy[1];
+            yy[1] = yy[0];
+            ryy_j = &(yy[1]);
+            rll_i = ll_i;
+            while ( rll_i != rtmp ) {
+               register UChar rtmp2;
+               ryy_j++;
+               rtmp2  = rtmp;
+               rtmp   = *ryy_j;
+               *ryy_j = rtmp2;
+            };
+            yy[0] = rtmp;
+            j = ryy_j - &(yy[0]);
+            mtfv[wr] = j+1; wr++; s->mtfFreq[j+1]++;
+         }
+
+      }
+   }
+
+   if (zPend > 0) {
+      zPend--;
+      while (True) {
+         if (zPend & 1) {
+            mtfv[wr] = BZ_RUNB; wr++; 
+            s->mtfFreq[BZ_RUNB]++; 
+         } else {
+            mtfv[wr] = BZ_RUNA; wr++; 
+            s->mtfFreq[BZ_RUNA]++; 
+         }
+         if (zPend < 2) break;
+         zPend = (zPend - 2) / 2;
+      };
+      zPend = 0;
+   }
+
+   mtfv[wr] = EOB; wr++; s->mtfFreq[EOB]++;
+
+   s->nMTF = wr;
+}
+
+
+/*---------------------------------------------------*/
+#define BZ_LESSER_ICOST  0
+#define BZ_GREATER_ICOST 15
+
+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;
+
+   if (s->verbosity >= 3)
+      VPrintf3( "      %d in block, %d after MTF & 1-2 coding, "
+                "%d+2 syms in use\n", 
+                s->nblock, s->nMTF, s->nInUse );
+
+   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--;
+         }
+
+         if (s->verbosity >= 3)
+            VPrintf5( "      initial group %d, [%d .. %d], "
+                      "has %d syms (%4.1f%%)\n",
+                      nPart, gs, ge, aFreq, 
+                      (100.0 * (float)aFreq) / (float)(s->nMTF) );
+ 
+         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;
+      }
+      if (s->verbosity >= 3) {
+         VPrintf2 ( "      pass %d: size is %d, grp uses are ", 
+                   iter+1, totc/8 );
+         for (t = 0; t < nGroups; t++)
+            VPrintf1 ( "%d ", fave[t] );
+         VPrintf0 ( "\n" );
+      }
+
+      /*--
+        Recompute the tables based on the accumulated frequencies.
+      --*/
+      /* maxLen was changed from 20 to 17 in bzip2-1.0.3.  See 
+         comment in huffman.c for details. */
+      for (t = 0; t < nGroups; t++)
+         BZ2_hbMakeCodeLengths ( &(s->len[t][0]), &(s->rfreq[t][0]), 
+                                 alphaSize, 17 /*20*/ );
+   }
+
+
+   AssertH( nGroups < 8, 3002 );
+   AssertH( nSelectors < 32768 &&
+            nSelectors <= (2 + (900000 / BZ_G_SIZE)),
+            3003 );
+
+
+   /*--- Compute MTF values for the selectors. ---*/
+   {
+      UChar pos[BZ_N_GROUPS], ll_i, tmp2, tmp;
+      for (i = 0; i < nGroups; i++) pos[i] = i;
+      for (i = 0; i < nSelectors; i++) {
+         ll_i = s->selector[i];
+         j = 0;
+         tmp = pos[j];
+         while ( ll_i != tmp ) {
+            j++;
+            tmp2 = tmp;
+            tmp = pos[j];
+            pos[j] = tmp2;
+         };
+         pos[0] = tmp;
+         s->selectorMtf[i] = j;
+      }
+   };
+
+   /*--- Assign actual codes for the tables. --*/
+   for (t = 0; t < nGroups; t++) {
+      minLen = 32;
+      maxLen = 0;
+      for (i = 0; i < alphaSize; i++) {
+         if (s->len[t][i] > maxLen) maxLen = s->len[t][i];
+         if (s->len[t][i] < minLen) minLen = s->len[t][i];
+      }
+      AssertH ( !(maxLen > 17 /*20*/ ), 3004 );
+      AssertH ( !(minLen < 1),  3005 );
+      BZ2_hbAssignCodes ( &(s->code[t][0]), &(s->len[t][0]), 
+                          minLen, maxLen, alphaSize );
+   }
+
+   /*--- Transmit the mapping table. ---*/
+   { 
+      Bool inUse16[16];
+      for (i = 0; i < 16; i++) {
+          inUse16[i] = False;
+          for (j = 0; j < 16; j++)
+             if (s->inUse[i * 16 + j]) inUse16[i] = True;
+      }
+     
+      nBytes = s->numZ;
+      for (i = 0; i < 16; i++)
+         if (inUse16[i]) bsW(s,1,1); else bsW(s,1,0);
+
+      for (i = 0; i < 16; i++)
+         if (inUse16[i])
+            for (j = 0; j < 16; j++) {
+               if (s->inUse[i * 16 + j]) bsW(s,1,1); else bsW(s,1,0);
+            }
+
+      if (s->verbosity >= 3) 
+         VPrintf1( "      bytes: mapping %d, ", s->numZ-nBytes );
+   }
+
+   /*--- Now the selectors. ---*/
+   nBytes = s->numZ;
+   bsW ( s, 3, nGroups );
+   bsW ( s, 15, nSelectors );
+   for (i = 0; i < nSelectors; i++) { 
+      for (j = 0; j < s->selectorMtf[i]; j++) bsW(s,1,1);
+      bsW(s,1,0);
+   }
+   if (s->verbosity >= 3)
+      VPrintf1( "selectors %d, ", s->numZ-nBytes );
+
+   /*--- Now the coding tables. ---*/
+   nBytes = s->numZ;
+
+   for (t = 0; t < nGroups; t++) {
+      Int32 curr = s->len[t][0];
+      bsW ( s, 5, curr );
+      for (i = 0; i < alphaSize; i++) {
+         while (curr < s->len[t][i]) { bsW(s,2,2); curr++; /* 10 */ };
+         while (curr > s->len[t][i]) { bsW(s,2,3); curr--; /* 11 */ };
+         bsW ( s, 1, 0 );
+      }
+   }
+
+   if (s->verbosity >= 3)
+      VPrintf1 ( "code lengths %d, ", s->numZ-nBytes );
+
+   /*--- And finally, the block data proper ---*/
+   nBytes = s->numZ;
+   selCtr = 0;
+   gs = 0;
+   while (True) {
+      if (gs >= s->nMTF) break;
+      ge = gs + BZ_G_SIZE - 1; 
+      if (ge >= s->nMTF) ge = s->nMTF-1;
+      AssertH ( s->selector[selCtr] < nGroups, 3006 );
+
+      if (nGroups == 6 && 50 == ge-gs+1) {
+            /*--- fast track the common case ---*/
+            UInt16 mtfv_i;
+            UChar* s_len_sel_selCtr 
+               = &(s->len[s->selector[selCtr]][0]);
+            Int32* s_code_sel_selCtr
+               = &(s->code[s->selector[selCtr]][0]);
+
+#           define BZ_ITAH(nn)                      \
+               mtfv_i = mtfv[gs+(nn)];              \
+               bsW ( s,                             \
+                     s_len_sel_selCtr[mtfv_i],      \
+                     s_code_sel_selCtr[mtfv_i] )
+
+            BZ_ITAH(0);  BZ_ITAH(1);  BZ_ITAH(2);  BZ_ITAH(3);  BZ_ITAH(4);
+            BZ_ITAH(5);  BZ_ITAH(6);  BZ_ITAH(7);  BZ_ITAH(8);  BZ_ITAH(9);
+            BZ_ITAH(10); BZ_ITAH(11); BZ_ITAH(12); BZ_ITAH(13); BZ_ITAH(14);
+            BZ_ITAH(15); BZ_ITAH(16); BZ_ITAH(17); BZ_ITAH(18); BZ_ITAH(19);
+            BZ_ITAH(20); BZ_ITAH(21); BZ_ITAH(22); BZ_ITAH(23); BZ_ITAH(24);
+            BZ_ITAH(25); BZ_ITAH(26); BZ_ITAH(27); BZ_ITAH(28); BZ_ITAH(29);
+            BZ_ITAH(30); BZ_ITAH(31); BZ_ITAH(32); BZ_ITAH(33); BZ_ITAH(34);
+            BZ_ITAH(35); BZ_ITAH(36); BZ_ITAH(37); BZ_ITAH(38); BZ_ITAH(39);
+            BZ_ITAH(40); BZ_ITAH(41); BZ_ITAH(42); BZ_ITAH(43); BZ_ITAH(44);
+            BZ_ITAH(45); BZ_ITAH(46); BZ_ITAH(47); BZ_ITAH(48); BZ_ITAH(49);
+
+#           undef BZ_ITAH
+
+      } else {
+	 /*--- slow version which correctly handles all situations ---*/
+         for (i = gs; i <= ge; i++) {
+            bsW ( s, 
+                  s->len  [s->selector[selCtr]] [mtfv[i]],
+                  s->code [s->selector[selCtr]] [mtfv[i]] );
+         }
+      }
+
+
+      gs = ge+1;
+      selCtr++;
+   }
+   AssertH( selCtr == nSelectors, 3007 );
+
+   if (s->verbosity >= 3)
+      VPrintf1( "codes %d\n", s->numZ-nBytes );
+   else /* squash compiler 'used but not set' warning */
+      nBytes = nBytes;
+}
+
+
+/*---------------------------------------------------*/
+void BZ2_compressBlock ( EState* s, Bool is_last_block )
+{
+   if (s->nblock > 0) {
+
+      BZ_FINALISE_CRC ( s->blockCRC );
+      s->combinedCRC = (s->combinedCRC << 1) | (s->combinedCRC >> 31);
+      s->combinedCRC ^= s->blockCRC;
+      if (s->blockNo > 1) s->numZ = 0;
+
+      if (s->verbosity >= 2)
+         VPrintf4( "    block %d: crc = 0x%08x, "
+                   "combined CRC = 0x%08x, size = %d\n",
+                   s->blockNo, s->blockCRC, s->combinedCRC, s->nblock );
+
+      BZ2_blockSort ( s );
+   }
+
+   s->zbits = (UChar*) (&((UChar*)s->arr2)[s->nblock]);
+
+   /*-- If this is the first block, create the stream header. --*/
+   if (s->blockNo == 1) {
+      BZ2_bsInitWrite ( s );
+      bsPutUChar ( s, BZ_HDR_B );
+      bsPutUChar ( s, BZ_HDR_Z );
+      bsPutUChar ( s, BZ_HDR_h );
+      bsPutUChar ( s, (UChar)(BZ_HDR_0 + s->blockSize100k) );
+   }
+
+   if (s->nblock > 0) {
+
+      bsPutUChar ( s, 0x31 ); bsPutUChar ( s, 0x41 );
+      bsPutUChar ( s, 0x59 ); bsPutUChar ( s, 0x26 );
+      bsPutUChar ( s, 0x53 ); bsPutUChar ( s, 0x59 );
+
+      /*-- Now the block's CRC, so it is in a known place. --*/
+      bsPutUInt32 ( s, s->blockCRC );
+
+      /*-- 
+         Now a single bit indicating (non-)randomisation. 
+         As of version 0.9.5, we use a better sorting algorithm
+         which makes randomisation unnecessary.  So always set
+         the randomised bit to 'no'.  Of course, the decoder
+         still needs to be able to handle randomised blocks
+         so as to maintain backwards compatibility with
+         older versions of bzip2.
+      --*/
+      bsW(s,1,0);
+
+      bsW ( s, 24, s->origPtr );
+      generateMTFValues ( s );
+      sendMTFValues ( s );
+   }
+
+
+   /*-- If this is the last block, add the stream trailer. --*/
+   if (is_last_block) {
+
+      bsPutUChar ( s, 0x17 ); bsPutUChar ( s, 0x72 );
+      bsPutUChar ( s, 0x45 ); bsPutUChar ( s, 0x38 );
+      bsPutUChar ( s, 0x50 ); bsPutUChar ( s, 0x90 );
+      bsPutUInt32 ( s, s->combinedCRC );
+      if (s->verbosity >= 2)
+         VPrintf1( "    final combined CRC = 0x%08x\n   ", s->combinedCRC );
+      bsFinishWrite ( s );
+   }
+}
+
+
+/*-------------------------------------------------------------*/
+/*--- end                                        compress.c ---*/
+/*-------------------------------------------------------------*/