blob: 1e677878e7bd9c68b026f8c72b0de9f01e15459c [file] [log] [blame]
/* { dg-skip-if "AArch64 does not support these bounds." { aarch64*-*-* } { "--param stack-clash-protection-*" } } */
/*-------------------------------------------------------------*/
/*--- Block sorting machinery ---*/
/*--- blocksort.c ---*/
/*-------------------------------------------------------------*/
/* ------------------------------------------------------------------
This file is part of bzip2/libbzip2, a program and library for
lossless, block-sorting data compression.
bzip2/libbzip2 version 1.0.6 of 6 September 2010
Copyright (C) 1996-2010 Julian Seward <jseward@bzip.org>
Please read the WARNING, DISCLAIMER and PATENTS sections in the
README file.
This program is released under the terms of the license contained
in the file LICENSE.
------------------------------------------------------------------ */
typedef char Char;
typedef unsigned char Bool;
typedef unsigned char UChar;
#if __SIZEOF_INT__ == 2
typedef long Int32;
typedef unsigned long UInt32;
#else
typedef int Int32;
typedef unsigned int UInt32;
#endif
typedef short Int16;
typedef unsigned short UInt16;
#define True ((Bool)1)
#define False ((Bool)0)
#define BZ_M_IDLE 1
#define BZ_M_RUNNING 2
#define BZ_M_FLUSHING 3
#define BZ_M_FINISHING 4
#define BZ_S_OUTPUT 1
#define BZ_S_INPUT 2
#define BZ_N_RADIX 2
#define BZ_N_QSORT 12
#define BZ_N_SHELL 18
#define BZ_N_OVERSHOOT (BZ_N_RADIX + BZ_N_QSORT + BZ_N_SHELL + 2)
/*---------------------------------------------*/
/*--- Fallback O(N log(N)^2) sorting ---*/
/*--- algorithm, for repetitive blocks ---*/
/*---------------------------------------------*/
/*---------------------------------------------*/
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
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) {
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--;
}
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)
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.
--*/
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) {
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; };
}
}
}
H *= 2;
if (H > nblock || nNotDone == 0) break;
}
/*--
Reconstruct the original block in
eclass8 [0 .. nblock-1], since the
previous phase destroyed it.
--*/
j = 0;
for (i = 0; i < nblock; i++) {
while (ftabCopy[j] == 0) j++;
ftabCopy[j]--;
eclass8[fmap[i]] = (UChar)j;
}
}
#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. ---*/
/*---------------------------------------------*/
/*---------------------------------------------*/
Bool mainGtU ( UInt32 i1,
UInt32 i2,
UChar* block,
UInt16* quadrant,
UInt32 nblock,
Int32* budget )
{
Int32 k;
UChar c1, c2;
UInt16 s1, s2;
/* 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.
--*/
Int32 incs[14] = { 1, 4, 13, 40, 121, 364, 1093, 3280,
9841, 29524, 88573, 265720,
797161, 2391484 };
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--; \
} \
}
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
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) {
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--;
}
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);
mpush (nextLo[0], nextHi[0], nextD[0]);
mpush (nextLo[1], nextHi[1], nextD[1]);
mpush (nextLo[2], nextHi[2], nextD[2]);
}
}