[Courgette Experimental] Replace QSufSort with libdivsufsort

courgette/third_party/divsufsort/trsort.cc

+/*
+ * trsort.cc for libdivsufsort
+ * Copyright (c) 2003-2008 Yuta Mori All Rights Reserved.
+ *
+ * For the terms under which this work may be distributed, please see
+ * the adjoining file "LICENSE".
+ *
+ * Changelog:
+ * 2016-06-02 - Initial commit and adaption to use PagedArray.
+ *                --Samuel Huang <huangs@chromium.org>
+ */
+
+#include "courgette/third_party/divsufsort/divsufsort_private.h"
+
+#define TR_INSERTIONSORT_THRESHOLD (8)
+#define TR_STACKSIZE (64)
+
+#define STACK_PUSH5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(ssize < STACK_SIZE);\
+    stack[ssize].a = (_a), stack[ssize].b = (_b),\
+    stack[ssize].c = (_c), stack[ssize].d = (_d), stack[ssize++].e = (_e);\
+  } while(0)
+#define STACK_POP5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(0 <= ssize);\
+    if(ssize == 0) { return; }\
+    (_a) = stack[--ssize].a, (_b) = stack[ssize].b,\
+    (_c) = stack[ssize].c, (_d) = stack[ssize].d, (_e) = stack[ssize].e;\
+  } while(0)
+
+
+namespace divsuf {
+
+namespace {
+
+/*- Private Functions -*/
+
+const saint_t lg_table[256]= {
+ -1,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,
+  5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7
+};
+
+inline
+saint_t
+tr_ilg(saidx_t n) {
+  return (n & 0xffff0000) ?
+          ((n & 0xff000000) ?
+            24 + lg_table[(n >> 24) & 0xff] :
+            16 + lg_table[(n >> 16) & 0xff]) :
+          ((n & 0x0000ff00) ?
+             8 + lg_table[(n >>  8) & 0xff] :
+             0 + lg_table[(n >>  0) & 0xff]);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Simple insertionsort for small size groups. */
+void
+tr_insertionsort(const_saidx_it ISAd, saidx_it first, saidx_it last) {
+  saidx_it a, b;
+  saidx_t t, r;
+
+  for(a = first + 1; a < last; ++a) {
+    for(t = *a, b = a - 1; 0 > (r = ISAd[t] - ISAd[*b]);) {
+      do { *(b + 1) = *b; } while((first <= --b) && (*b < 0));
+      if(b < first) { break; }
+    }
+    if(r == 0) { *b = ~*b; }
+    *(b + 1) = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+inline
+void
+tr_fixdown(const_saidx_it ISAd, saidx_it SA, saidx_t i, saidx_t size) {
+  saidx_t j, k;
+  saidx_t v;
+  saidx_t c, d, e;
+
+  for(v = SA[i], c = ISAd[v]; (j = 2 * i + 1) < size; SA[i] = SA[k], i = k) {
+    d = ISAd[SA[k = j++]];
+    if(d < (e = ISAd[SA[j]])) { k = j; d = e; }
+    if(d <= c) { break; }
+  }
+  SA[i] = v;
+}
+
+/* Simple top-down heapsort. */
+void
+tr_heapsort(const_saidx_it ISAd, saidx_it SA, saidx_t size) {
+  saidx_t i, m;
+  saidx_t t;
+
+  m = size;
+  if((size % 2) == 0) {
+    m--;
+    if(ISAd[SA[m / 2]] < ISAd[SA[m]]) { SWAP(SA[m], SA[m / 2]); }
+  }
+
+  for(i = m / 2 - 1; 0 <= i; --i) { tr_fixdown(ISAd, SA, i, m); }
+  if((size % 2) == 0) { SWAP(SA[0], SA[m]); tr_fixdown(ISAd, SA, 0, m); }
+  for(i = m - 1; 0 < i; --i) {
+    t = SA[0], SA[0] = SA[i];
+    tr_fixdown(ISAd, SA, 0, i);
+    SA[i] = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Returns the median of three elements. */
+inline
+saidx_it
+tr_median3(const_saidx_it ISAd, saidx_it v1, saidx_it v2, saidx_it v3) {
+  saidx_it t;
+  if(ISAd[*v1] > ISAd[*v2]) { SWAP(v1, v2); }
+  if(ISAd[*v2] > ISAd[*v3]) {
+    if(ISAd[*v1] > ISAd[*v3]) { return v1; }
+    else { return v3; }
+  }
+  return v2;
+}
+
+/* Returns the median of five elements. */
+inline
+saidx_it
+tr_median5(const_saidx_it ISAd,
+           saidx_it v1, saidx_it v2, saidx_it v3, saidx_it v4, saidx_it v5) {
+  saidx_it t;
+  if(ISAd[*v2] > ISAd[*v3]) { SWAP(v2, v3); }
+  if(ISAd[*v4] > ISAd[*v5]) { SWAP(v4, v5); }
+  if(ISAd[*v2] > ISAd[*v4]) { SWAP(v2, v4); SWAP(v3, v5); }
+  if(ISAd[*v1] > ISAd[*v3]) { SWAP(v1, v3); }
+  if(ISAd[*v1] > ISAd[*v4]) { SWAP(v1, v4); SWAP(v3, v5); }
+  if(ISAd[*v3] > ISAd[*v4]) { return v4; }
+  return v3;
+}
+
+/* Returns the pivot element. */
+inline
+saidx_it
+tr_pivot(const_saidx_it ISAd, saidx_it first, saidx_it last) {
+  saidx_it middle;
+  saidx_t t;
+
+  t = last - first;
+  middle = first + t / 2;
+
+  if(t <= 512) {
+    if(t <= 32) {
+      return tr_median3(ISAd, first, middle, last - 1);
+    } else {
+      t >>= 2;
+      return tr_median5(ISAd, first, first + t, middle, last - 1 - t, last - 1);
+    }
+  }
+  t >>= 3;
+  first  = tr_median3(ISAd, first, first + t, first + (t << 1));
+  middle = tr_median3(ISAd, middle - t, middle, middle + t);
+  last   = tr_median3(ISAd, last - 1 - (t << 1), last - 1 - t, last - 1);
+  return tr_median3(ISAd, first, middle, last);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+typedef struct _trbudget_t trbudget_t;
+struct _trbudget_t {
+  saidx_t chance;
+  saidx_t remain;
+  saidx_t incval;
+  saidx_t count;
+};
+
+inline
+void
+trbudget_init(trbudget_t *budget, saidx_t chance, saidx_t incval) {
+  budget->chance = chance;
+  budget->remain = budget->incval = incval;
+}
+
+inline
+saint_t
+trbudget_check(trbudget_t *budget, saidx_t size) {
+  if(size <= budget->remain) { budget->remain -= size; return 1; }
+  if(budget->chance == 0) { budget->count += size; return 0; }
+  budget->remain += budget->incval - size;
+  budget->chance -= 1;
+  return 1;
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+inline
+void
+tr_partition(const_saidx_it ISAd,
+             saidx_it first, saidx_it middle, saidx_it last,
+             saidx_it* pa, saidx_it* pb, saidx_t v) {
+  saidx_it a, b, c, d, e, f;
+  saidx_t t, s;
+  saidx_t x = 0;
+
+  for(b = middle - 1; (++b < last) && ((x = ISAd[*b]) == v);) { }
+  if(((a = b) < last) && (x < v)) {
+    for(; (++b < last) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+  }
+  for(c = last; (b < --c) && ((x = ISAd[*c]) == v);) { }
+  if((b < (d = c)) && (x > v)) {
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+  for(; b < c;) {
+    SWAP(*b, *c);
+    for(; (++b < c) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+
+  if(a <= d) {
+    c = b - 1;
+    if((s = a - first) > (t = b - a)) { s = t; }
+    for(e = first, f = b - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    if((s = d - c) > (t = last - d - 1)) { s = t; }
+    for(e = b, f = last - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    first += (b - a), last -= (d - c);
+  }
+  *pa = first, *pb = last;
+}
+
+void
+tr_copy(saidx_it ISA, const_saidx_it SA,
+        saidx_it first, saidx_it a, saidx_it b, saidx_it last,
+        saidx_t depth) {
+  /* sort suffixes of middle partition
+     by using sorted order of suffixes of left and right partition. */
+  saidx_it c, d, e;
+  saidx_t s, v;
+
+  v = b - SA - 1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      ISA[s] = d - SA;
+    }
+  }
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      ISA[s] = d - SA;
+    }
+  }
+}
+
+void
+tr_partialcopy(saidx_it ISA, const_saidx_it SA,
+               saidx_it first, saidx_it a, saidx_it b, saidx_it last,
+               saidx_t depth) {
+  saidx_it c, d, e;
+  saidx_t s, v;
+  saidx_t rank, lastrank, newrank = -1;
+
+  v = b - SA - 1;
+  lastrank = -1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+
+  lastrank = -1;
+  for(e = d; first <= e; --e) {
+    rank = ISA[*e];
+    if(lastrank != rank) { lastrank = rank; newrank = e - SA; }
+    if(newrank != rank) { ISA[*e] = newrank; }
+  }
+
+  lastrank = -1;
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+}
+
+void
+tr_introsort(saidx_it ISA, const_saidx_it ISAd,
+             saidx_it SA, saidx_it first, saidx_it last,
+             trbudget_t *budget) {
+#define STACK_SIZE TR_STACKSIZE
+  struct { const_saidx_it a; saidx_it b, c; saint_t d, e; }stack[STACK_SIZE];
+  saidx_it a, b, c;
+  saidx_t t;
+  saidx_t v, x = 0;
+  saidx_t incr = ISAd - ISA;
+  saint_t limit, next;
+  saint_t ssize, trlink = -1;
+
+  for(ssize = 0, limit = tr_ilg(last - first);;) {
+
+    if(limit < 0) {
+      if(limit == -1) {
+        /* tandem repeat partition */
+        tr_partition(ISAd - incr, first, first, last, &a, &b, last - SA - 1);
+
+        /* update ranks */
+        if(a < last) {
+          for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+        }
+        if(b < last) {
+          for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; }
+        }
+
+        /* push */
+        if(1 < (b - a)) {
+          STACK_PUSH5(NULL, a, b, 0, 0);
+          STACK_PUSH5(ISAd - incr, first, last, -2, trlink);
+          trlink = ssize - 2;
+        }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, tr_ilg(last - b), trlink);
+            last = a, limit = tr_ilg(a - first);
+          } else if(1 < (last - b)) {
+            first = b, limit = tr_ilg(last - b);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, tr_ilg(a - first), trlink);
+            first = b, limit = tr_ilg(last - b);
+          } else if(1 < (a - first)) {
+            last = a, limit = tr_ilg(a - first);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      } else if(limit == -2) {
+        /* tandem repeat copy */
+        a = stack[--ssize].b, b = stack[ssize].c;
+        if(stack[ssize].d == 0) {
+          tr_copy(ISA, SA, first, a, b, last, ISAd - ISA);
+        } else {
+          if(0 <= trlink) { stack[trlink].d = -1; }
+          tr_partialcopy(ISA, SA, first, a, b, last, ISAd - ISA);
+        }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      } else {
+        /* sorted partition */
+        if(0 <= *first) {
+          a = first;
+          do { ISA[*a] = a - SA; } while((++a < last) && (0 <= *a));
+          first = a;
+        }
+        if(first < last) {
+          a = first; do { *a = ~*a; } while(*++a < 0);
+          next = (ISA[*a] != ISAd[*a]) ? tr_ilg(a - first + 1) : -1;
+          if(++a < last) { for(b = first, v = a - SA - 1; b < a; ++b) { ISA[*b] = v; } }
+
+          /* push */
+          if(trbudget_check(budget, a - first)) {
+            if((a - first) <= (last - a)) {
+              STACK_PUSH5(ISAd, a, last, -3, trlink);
+              ISAd += incr, last = a, limit = next;
+            } else {
+              if(1 < (last - a)) {
+                STACK_PUSH5(ISAd + incr, first, a, next, trlink);
+                first = a, limit = -3;
+              } else {
+                ISAd += incr, last = a, limit = next;
+              }
+            }
+          } else {
+            if(0 <= trlink) { stack[trlink].d = -1; }
+            if(1 < (last - a)) {
+              first = a, limit = -3;
+            } else {
+              STACK_POP5(ISAd, first, last, limit, trlink);
+            }
+          }
+        } else {
+          STACK_POP5(ISAd, first, last, limit, trlink);
+        }
+      }
+      continue;
+    }
+
+    if((last - first) <= TR_INSERTIONSORT_THRESHOLD) {
+      tr_insertionsort(ISAd, first, last);
+      limit = -3;
+      continue;
+    }
+
+    if(limit-- == 0) {
+      tr_heapsort(ISAd, first, last - first);
+      for(a = last - 1; first < a; a = b) {
+        for(x = ISAd[*a], b = a - 1; (first <= b) && (ISAd[*b] == x); --b) { *b = ~*b; }
+      }
+      limit = -3;
+      continue;
+    }
+
+    /* choose pivot */
+    a = tr_pivot(ISAd, first, last);
+    SWAP(*first, *a);
+    v = ISAd[*first];
+
+    /* partition */
+    tr_partition(ISAd, first, first + 1, last, &a, &b, v);
+    if((last - first) != (b - a)) {
+      next = (ISA[*a] != v) ? tr_ilg(b - a) : -1;
+
+      /* update ranks */
+      for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+      if(b < last) { for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; } }
+
+      /* push */
+      if((1 < (b - a)) && (trbudget_check(budget, b - a))) {
+        if((a - first) <= (last - b)) {
+          if((last - b) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              last = a;
+            } else if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((a - first) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        } else {
+          if((a - first) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              first = b;
+            } else if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((last - b) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        }
+      } else {
+        if((1 < (b - a)) && (0 <= trlink)) { stack[trlink].d = -1; }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            last = a;
+          } else if(1 < (last - b)) {
+            first = b;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            first = b;
+          } else if(1 < (a - first)) {
+            last = a;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      }
+    } else {
+      if(trbudget_check(budget, last - first)) {
+        limit = tr_ilg(last - first), ISAd += incr;
+      } else {
+        if(0 <= trlink) { stack[trlink].d = -1; }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      }
+    }
+  }
+#undef STACK_SIZE
+}
+
+}  // namespace
+
+/*---------------------------------------------------------------------------*/
+
+/*- Function -*/
+
+/* Tandem repeat sort */
+void
+trsort(saidx_it ISA, saidx_it SA, saidx_t n, saidx_t depth) {
+  saidx_it ISAd;
+  saidx_it first, last;
+  trbudget_t budget;
+  saidx_t t, skip, unsorted;
+
+  trbudget_init(&budget, tr_ilg(n) * 2 / 3, n);
+/*  trbudget_init(&budget, tr_ilg(n) * 3 / 4, n); */
+  for(ISAd = ISA + depth; -n < *SA; ISAd += ISAd - ISA) {
+    first = SA;
+    skip = 0;
+    unsorted = 0;
+    do {
+      if((t = *first) < 0) { first -= t; skip += t; }
+      else {
+        if(skip != 0) { *(first + skip) = skip; skip = 0; }
+        last = SA + ISA[t] + 1;
+        if(1 < (last - first)) {
+          budget.count = 0;
+          tr_introsort(ISA, ISAd, SA, first, last, &budget);
+          if(budget.count != 0) { unsorted += budget.count; }
+          else { skip = first - last; }
+        } else if((last - first) == 1) {
+          skip = -1;
+        }
+        first = last;
+      }
+    } while(first < (SA + n));
+    if(skip != 0) { *(first + skip) = skip; }
+    if(unsorted == 0) { break; }
+  }
+}
+
+}  // namespace divsuf