1  /* mpn_sec_div_qr, mpn_sec_div_r -- Compute Q = floor(U / V), U = U mod V.
       2     Side-channel silent under the assumption that the used instructions are
       3     side-channel silent.
       4  
       5     Contributed to the GNU project by Torbjörn Granlund.
       6  
       7  Copyright 2011-2015 Free Software Foundation, Inc.
       8  
       9  This file is part of the GNU MP Library.
      10  
      11  The GNU MP Library is free software; you can redistribute it and/or modify
      12  it under the terms of either:
      13  
      14    * the GNU Lesser General Public License as published by the Free
      15      Software Foundation; either version 3 of the License, or (at your
      16      option) any later version.
      17  
      18  or
      19  
      20    * the GNU General Public License as published by the Free Software
      21      Foundation; either version 2 of the License, or (at your option) any
      22      later version.
      23  
      24  or both in parallel, as here.
      25  
      26  The GNU MP Library is distributed in the hope that it will be useful, but
      27  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
      28  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
      29  for more details.
      30  
      31  You should have received copies of the GNU General Public License and the
      32  GNU Lesser General Public License along with the GNU MP Library.  If not,
      33  see https://www.gnu.org/licenses/.  */
      34  
      35  #include "gmp-impl.h"
      36  #include "longlong.h"
      37  
      38  #if OPERATION_sec_div_qr
      39  #define FNAME mpn_sec_div_qr
      40  #define FNAME_itch mpn_sec_div_qr_itch
      41  #define Q(q) q,
      42  #define RETTYPE mp_limb_t
      43  #endif
      44  #if OPERATION_sec_div_r
      45  #define FNAME mpn_sec_div_r
      46  #define FNAME_itch mpn_sec_div_r_itch
      47  #define Q(q)
      48  #define RETTYPE void
      49  #endif
      50  
      51  mp_size_t
      52  FNAME_itch (mp_size_t nn, mp_size_t dn)
      53  {
      54  #if OPERATION_sec_div_qr
      55  /* Needs (nn + dn + 1) + mpn_sec_pi1_div_qr's needs of (2nn' - dn + 1) for a
      56     total of 3nn + 4 limbs at tp.  Note that mpn_sec_pi1_div_qr's nn is one
      57     greater than ours, therefore +4 and not just +2.  */
      58    return 3 * nn + 4;
      59  #endif
      60  #if OPERATION_sec_div_r
      61  /* Needs (nn + dn + 1) + mpn_sec_pi1_div_r's needs of (dn + 1) for a total of
      62     nn + 2dn + 2 limbs at tp.  */
      63    return nn + 2 * dn + 2;
      64  #endif
      65  }
      66  
      67  RETTYPE
      68  FNAME (Q(mp_ptr qp)
      69         mp_ptr np, mp_size_t nn,
      70         mp_srcptr dp, mp_size_t dn,
      71         mp_ptr tp)
      72  {
      73    mp_limb_t d1, d0;
      74    unsigned int cnt;
      75    mp_limb_t inv32;
      76  
      77    ASSERT (dn >= 1);
      78    ASSERT (nn >= dn);
      79    ASSERT (dp[dn - 1] != 0);
      80  
      81    d1 = dp[dn - 1];
      82    count_leading_zeros (cnt, d1);
      83  
      84    if (cnt != 0)
      85      {
      86        mp_limb_t qh, cy;
      87        mp_ptr np2, dp2;
      88        dp2 = tp;					/* dn limbs */
      89        mpn_lshift (dp2, dp, dn, cnt);
      90  
      91        np2 = tp + dn;				/* (nn + 1) limbs */
      92        cy = mpn_lshift (np2, np, nn, cnt);
      93        np2[nn++] = cy;
      94  
      95        d0 = dp2[dn - 1];
      96        d0 += (~d0 != 0);
      97        invert_limb (inv32, d0);
      98  
      99        /* We add nn + dn to tp here, not nn + 1 + dn, as expected.  This is
     100  	 since nn here will have been incremented.  */
     101  #if OPERATION_sec_div_qr
     102        qh = mpn_sec_pi1_div_qr (np2 + dn, np2, nn, dp2, dn, inv32, tp + nn + dn);
     103        ASSERT (qh == 0);		/* FIXME: this indicates inefficiency! */
     104        MPN_COPY (qp, np2 + dn, nn - dn - 1);
     105        qh = np2[nn - 1];
     106  #else
     107        mpn_sec_pi1_div_r (np2, nn, dp2, dn, inv32, tp + nn + dn);
     108  #endif
     109  
     110        mpn_rshift (np, np2, dn, cnt);
     111  
     112  #if OPERATION_sec_div_qr
     113        return qh;
     114  #endif
     115      }
     116    else
     117      {
     118        /* FIXME: Consider copying np => np2 here, adding a 0-limb at the top.
     119  	 That would simplify the underlying pi1 function, since then it could
     120  	 assume nn > dn.  */
     121        d0 = dp[dn - 1];
     122        d0 += (~d0 != 0);
     123        invert_limb (inv32, d0);
     124  
     125  #if OPERATION_sec_div_qr
     126        return mpn_sec_pi1_div_qr (qp, np, nn, dp, dn, inv32, tp);
     127  #else
     128        mpn_sec_pi1_div_r (np, nn, dp, dn, inv32, tp);
     129  #endif
     130      }
     131  }