/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */

#include "prlong.h"

static PRInt64 ll_zero = PR_INT64(0x0000000000000000);
static PRInt64 ll_maxint = PR_INT64(0x7fffffffffffffff);
static PRInt64 ll_minint = PR_INT64(0x8000000000000000);
static PRUint64 ll_maxuint = PR_UINT64(0xffffffffffffffff);

PR_IMPLEMENT(PRInt64) LL_Zero(void) { return ll_zero; }
PR_IMPLEMENT(PRInt64) LL_MaxInt(void) { return ll_maxint; }
PR_IMPLEMENT(PRInt64) LL_MinInt(void) { return ll_minint; }
PR_IMPLEMENT(PRUint64) LL_MaxUint(void) { return ll_maxuint; }

#ifndef HAVE_LONG_LONG
/*
** Divide 64-bit a by 32-bit b, which must be normalized so its high bit is 1.
*/
static void norm_udivmod32(PRUint32* qp, PRUint32* rp, PRUint64 a, PRUint32 b) {
  PRUint32 d1, d0, q1, q0;
  PRUint32 r1, r0, m;

  d1 = _hi16(b);
  d0 = _lo16(b);
  r1 = a.hi % d1;
  q1 = a.hi / d1;
  m = q1 * d0;
  r1 = (r1 << 16) | _hi16(a.lo);
  if (r1 < m) {
    q1--, r1 += b;
    if (r1 >= b /* i.e., we didn't get a carry when adding to r1 */
        && r1 < m) {
      q1--, r1 += b;
    }
  }
  r1 -= m;
  r0 = r1 % d1;
  q0 = r1 / d1;
  m = q0 * d0;
  r0 = (r0 << 16) | _lo16(a.lo);
  if (r0 < m) {
    q0--, r0 += b;
    if (r0 >= b && r0 < m) {
      q0--, r0 += b;
    }
  }
  *qp = (q1 << 16) | q0;
  *rp = r0 - m;
}

static PRUint32 CountLeadingZeros(PRUint32 a) {
  PRUint32 t;
  PRUint32 r = 32;

  if ((t = a >> 16) != 0) {
    r -= 16, a = t;
  }
  if ((t = a >> 8) != 0) {
    r -= 8, a = t;
  }
  if ((t = a >> 4) != 0) {
    r -= 4, a = t;
  }
  if ((t = a >> 2) != 0) {
    r -= 2, a = t;
  }
  if ((t = a >> 1) != 0) {
    r -= 1, a = t;
  }
  if (a & 1) {
    r--;
  }
  return r;
}

PR_IMPLEMENT(void)
ll_udivmod(PRUint64* qp, PRUint64* rp, PRUint64 a, PRUint64 b) {
  PRUint32 n0, n1, n2;
  PRUint32 q0, q1;
  PRUint32 rsh, lsh;

  n0 = a.lo;
  n1 = a.hi;

  if (b.hi == 0) {
    if (b.lo > n1) {
      /* (0 q0) = (n1 n0) / (0 D0) */

      lsh = CountLeadingZeros(b.lo);

      if (lsh) {
        /*
         * Normalize, i.e. make the most significant bit of the
         * denominator be set.
         */
        b.lo = b.lo << lsh;
        n1 = (n1 << lsh) | (n0 >> (32 - lsh));
        n0 = n0 << lsh;
      }

      a.lo = n0, a.hi = n1;
      norm_udivmod32(&q0, &n0, a, b.lo);
      q1 = 0;

      /* remainder is in n0 >> lsh */
    } else {
      /* (q1 q0) = (n1 n0) / (0 d0) */

      if (b.lo == 0) {   /* user wants to divide by zero! */
        b.lo = 1 / b.lo; /* so go ahead and crash */
      }

      lsh = CountLeadingZeros(b.lo);

      if (lsh == 0) {
        /*
         * From (n1 >= b.lo)
         *   && (the most significant bit of b.lo is set),
         * conclude that
         *  (the most significant bit of n1 is set)
         *   && (the leading quotient digit q1 = 1).
         *
         * This special case is necessary, not an optimization
         * (Shifts counts of 32 are undefined).
         */
        n1 -= b.lo;
        q1 = 1;
      } else {
        /*
         * Normalize.
         */
        rsh = 32 - lsh;

        b.lo = b.lo << lsh;
        n2 = n1 >> rsh;
        n1 = (n1 << lsh) | (n0 >> rsh);
        n0 = n0 << lsh;

        a.lo = n1, a.hi = n2;
        norm_udivmod32(&q1, &n1, a, b.lo);
      }

      /* n1 != b.lo... */

      a.lo = n0, a.hi = n1;
      norm_udivmod32(&q0, &n0, a, b.lo);

      /* remainder in n0 >> lsh */
    }

    if (rp) {
      rp->lo = n0 >> lsh;
      rp->hi = 0;
    }
  } else {
    if (b.hi > n1) {
      /* (0 0) = (n1 n0) / (D1 d0) */

      q0 = 0;
      q1 = 0;

      /* remainder in (n1 n0) */
      if (rp) {
        rp->lo = n0;
        rp->hi = n1;
      }
    } else {
      /* (0 q0) = (n1 n0) / (d1 d0) */

      lsh = CountLeadingZeros(b.hi);
      if (lsh == 0) {
        /*
         * From (n1 >= b.hi)
         *   && (the most significant bit of b.hi is set),
         * conclude that
         *      (the most significant bit of n1 is set)
         *   && (the quotient digit q0 = 0 or 1).
         *
         * This special case is necessary, not an optimization.
         */

        /*
         * The condition on the next line takes advantage of that
         * n1 >= b.hi (true due to control flow).
         */
        if (n1 > b.hi || n0 >= b.lo) {
          q0 = 1;
          a.lo = n0, a.hi = n1;
          LL_SUB(a, a, b);
        } else {
          q0 = 0;
        }
        q1 = 0;

        if (rp) {
          rp->lo = n0;
          rp->hi = n1;
        }
      } else {
        PRInt64 m;

        /*
         * Normalize.
         */
        rsh = 32 - lsh;

        b.hi = (b.hi << lsh) | (b.lo >> rsh);
        b.lo = b.lo << lsh;
        n2 = n1 >> rsh;
        n1 = (n1 << lsh) | (n0 >> rsh);
        n0 = n0 << lsh;

        a.lo = n1, a.hi = n2;
        norm_udivmod32(&q0, &n1, a, b.hi);
        LL_MUL32(m, q0, b.lo);

        if ((m.hi > n1) || ((m.hi == n1) && (m.lo > n0))) {
          q0--;
          LL_SUB(m, m, b);
        }

        q1 = 0;

        /* Remainder is ((n1 n0) - (m1 m0)) >> lsh */
        if (rp) {
          a.lo = n0, a.hi = n1;
          LL_SUB(a, a, m);
          rp->lo = (a.hi << rsh) | (a.lo >> lsh);
          rp->hi = a.hi >> lsh;
        }
      }
    }
  }

  if (qp) {
    qp->lo = q0;
    qp->hi = q1;
  }
}
#endif /* !HAVE_LONG_LONG */