1  /* mpn_sub_err2_n -- sub_n with two error terms
       2  
       3     Contributed by David Harvey.
       4  
       5     THE FUNCTION IN THIS FILE IS INTERNAL WITH A MUTABLE INTERFACE.  IT IS ONLY
       6     SAFE TO REACH IT THROUGH DOCUMENTED INTERFACES.  IN FACT, IT IS ALMOST
       7     GUARANTEED THAT IT'LL CHANGE OR DISAPPEAR IN A FUTURE GNU MP RELEASE.
       8  
       9  Copyright 2011 Free Software Foundation, Inc.
      10  
      11  This file is part of the GNU MP Library.
      12  
      13  The GNU MP Library is free software; you can redistribute it and/or modify
      14  it under the terms of either:
      15  
      16    * the GNU Lesser General Public License as published by the Free
      17      Software Foundation; either version 3 of the License, or (at your
      18      option) any later version.
      19  
      20  or
      21  
      22    * the GNU General Public License as published by the Free Software
      23      Foundation; either version 2 of the License, or (at your option) any
      24      later version.
      25  
      26  or both in parallel, as here.
      27  
      28  The GNU MP Library is distributed in the hope that it will be useful, but
      29  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
      30  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
      31  for more details.
      32  
      33  You should have received copies of the GNU General Public License and the
      34  GNU Lesser General Public License along with the GNU MP Library.  If not,
      35  see https://www.gnu.org/licenses/.  */
      36  
      37  #include "gmp-impl.h"
      38  
      39  /*
      40    Computes:
      41  
      42    (1) {rp,n} := {up,n} - {vp,n} (just like mpn_sub_n) with incoming borrow cy,
      43    return value is borrow out.
      44  
      45    (2) Let c[i+1] = borrow from i-th limb subtraction (c[0] = cy).
      46    Computes c[1]*yp1[n-1] + ... + c[n]*yp1[0],
      47             c[1]*yp2[n-1] + ... + c[n]*yp2[0],
      48    stores two-limb results at {ep,2} and {ep+2,2} respectively.
      49  
      50    Requires n >= 1.
      51  
      52    None of the outputs may overlap each other or any of the inputs, except
      53    that {rp,n} may be equal to {up,n} or {vp,n}.
      54  */
      55  mp_limb_t
      56  mpn_sub_err2_n (mp_ptr rp, mp_srcptr up, mp_srcptr vp,
      57                  mp_ptr ep, mp_srcptr yp1, mp_srcptr yp2,
      58                  mp_size_t n, mp_limb_t cy)
      59  {
      60    mp_limb_t el1, eh1, el2, eh2, ul, vl, yl1, yl2, zl1, zl2, rl, sl, cy1, cy2;
      61  
      62    ASSERT (n >= 1);
      63    ASSERT (MPN_SAME_OR_SEPARATE_P (rp, up, n));
      64    ASSERT (MPN_SAME_OR_SEPARATE_P (rp, vp, n));
      65    ASSERT (! MPN_OVERLAP_P (rp, n, yp1, n));
      66    ASSERT (! MPN_OVERLAP_P (rp, n, yp2, n));
      67    ASSERT (! MPN_OVERLAP_P (ep, 4, up, n));
      68    ASSERT (! MPN_OVERLAP_P (ep, 4, vp, n));
      69    ASSERT (! MPN_OVERLAP_P (ep, 4, yp1, n));
      70    ASSERT (! MPN_OVERLAP_P (ep, 4, yp2, n));
      71    ASSERT (! MPN_OVERLAP_P (ep, 4, rp, n));
      72  
      73    yp1 += n - 1;
      74    yp2 += n - 1;
      75    el1 = eh1 = 0;
      76    el2 = eh2 = 0;
      77  
      78    do
      79      {
      80        yl1 = *yp1--;
      81        yl2 = *yp2--;
      82        ul = *up++;
      83        vl = *vp++;
      84  
      85        /* ordinary sub_n */
      86        SUBC_LIMB (cy1, sl, ul, vl);
      87        SUBC_LIMB (cy2, rl, sl, cy);
      88        cy = cy1 | cy2;
      89        *rp++ = rl;
      90  
      91        /* update (eh1:el1) */
      92        zl1 = (-cy) & yl1;
      93        el1 += zl1;
      94        eh1 += el1 < zl1;
      95  
      96        /* update (eh2:el2) */
      97        zl2 = (-cy) & yl2;
      98        el2 += zl2;
      99        eh2 += el2 < zl2;
     100      }
     101    while (--n);
     102  
     103  #if GMP_NAIL_BITS != 0
     104    eh1 = (eh1 << GMP_NAIL_BITS) + (el1 >> GMP_NUMB_BITS);
     105    el1 &= GMP_NUMB_MASK;
     106    eh2 = (eh2 << GMP_NAIL_BITS) + (el2 >> GMP_NUMB_BITS);
     107    el2 &= GMP_NUMB_MASK;
     108  #endif
     109  
     110    ep[0] = el1;
     111    ep[1] = eh1;
     112    ep[2] = el2;
     113    ep[3] = eh2;
     114  
     115    return cy;
     116  }