Home | History | Annotate | Line # | Download | only in generic
dcpi1_bdiv_q.c revision 1.1.1.1.8.1
      1 /* mpn_dcpi1_bdiv_q -- divide-and-conquer Hensel division with precomputed
      2    inverse, returning quotient.
      3 
      4    Contributed to the GNU project by Niels Mller and Torbjorn Granlund.
      5 
      6    THE FUNCTIONS IN THIS FILE ARE INTERNAL WITH MUTABLE INTERFACES.  IT IS ONLY
      7    SAFE TO REACH THEM THROUGH DOCUMENTED INTERFACES.  IN FACT, IT IS ALMOST
      8    GUARANTEED THAT THEY WILL CHANGE OR DISAPPEAR IN A FUTURE GMP RELEASE.
      9 
     10 Copyright 2006, 2007, 2009, 2010, 2011 Free Software Foundation, Inc.
     11 
     12 This file is part of the GNU MP Library.
     13 
     14 The GNU MP Library is free software; you can redistribute it and/or modify
     15 it under the terms of the GNU Lesser General Public License as published by
     16 the Free Software Foundation; either version 3 of the License, or (at your
     17 option) any later version.
     18 
     19 The GNU MP Library is distributed in the hope that it will be useful, but
     20 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
     21 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
     22 License for more details.
     23 
     24 You should have received a copy of the GNU Lesser General Public License
     25 along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.  */
     26 
     27 #include "gmp.h"
     28 #include "gmp-impl.h"
     29 
     30 
     31 mp_size_t
     32 mpn_dcpi1_bdiv_q_n_itch (mp_size_t n)
     33 {
     34   /* NOTE: Depends on mullo_n interface */
     35   return n;
     36 }
     37 
     38 /* Computes Q = N / D mod B^n, destroys N.
     39 
     40    N = {np,n}
     41    D = {dp,n}
     42 */
     43 
     44 void
     45 mpn_dcpi1_bdiv_q_n (mp_ptr qp,
     46 		    mp_ptr np, mp_srcptr dp, mp_size_t n,
     47 		    mp_limb_t dinv, mp_ptr tp)
     48 {
     49   while (ABOVE_THRESHOLD (n, DC_BDIV_Q_THRESHOLD))
     50     {
     51       mp_size_t lo, hi;
     52       mp_limb_t cy;
     53 
     54       lo = n >> 1;			/* floor(n/2) */
     55       hi = n - lo;			/* ceil(n/2) */
     56 
     57       cy = mpn_dcpi1_bdiv_qr_n (qp, np, dp, lo, dinv, tp);
     58 
     59       mpn_mullo_n (tp, qp, dp + hi, lo);
     60       mpn_sub_n (np + hi, np + hi, tp, lo);
     61 
     62       if (lo < hi)
     63 	{
     64 	  cy += mpn_submul_1 (np + lo, qp, lo, dp[lo]);
     65 	  np[n - 1] -= cy;
     66 	}
     67       qp += lo;
     68       np += lo;
     69       n -= lo;
     70     }
     71   mpn_sbpi1_bdiv_q (qp, np, n, dp, n, dinv);
     72 }
     73 
     74 /* Computes Q = N / D mod B^nn, destroys N.
     75 
     76    N = {np,nn}
     77    D = {dp,dn}
     78 */
     79 
     80 void
     81 mpn_dcpi1_bdiv_q (mp_ptr qp,
     82 		  mp_ptr np, mp_size_t nn,
     83 		  mp_srcptr dp, mp_size_t dn,
     84 		  mp_limb_t dinv)
     85 {
     86   mp_size_t qn;
     87   mp_limb_t cy;
     88   mp_ptr tp;
     89   TMP_DECL;
     90 
     91   TMP_MARK;
     92 
     93   ASSERT (dn >= 2);
     94   ASSERT (nn - dn >= 0);
     95   ASSERT (dp[0] & 1);
     96 
     97   tp = TMP_SALLOC_LIMBS (dn);
     98 
     99   qn = nn;
    100 
    101   if (qn > dn)
    102     {
    103       /* Reduce qn mod dn in a super-efficient manner.  */
    104       do
    105 	qn -= dn;
    106       while (qn > dn);
    107 
    108       /* Perform the typically smaller block first.  */
    109       if (BELOW_THRESHOLD (qn, DC_BDIV_QR_THRESHOLD))
    110 	cy = mpn_sbpi1_bdiv_qr (qp, np, 2 * qn, dp, qn, dinv);
    111       else
    112 	cy = mpn_dcpi1_bdiv_qr_n (qp, np, dp, qn, dinv, tp);
    113 
    114       if (qn != dn)
    115 	{
    116 	  if (qn > dn - qn)
    117 	    mpn_mul (tp, qp, qn, dp + qn, dn - qn);
    118 	  else
    119 	    mpn_mul (tp, dp + qn, dn - qn, qp, qn);
    120 	  mpn_incr_u (tp + qn, cy);
    121 
    122 	  mpn_sub (np + qn, np + qn, nn - qn, tp, dn);
    123 	  cy = 0;
    124 	}
    125 
    126       np += qn;
    127       qp += qn;
    128 
    129       qn = nn - qn;
    130       while (qn > dn)
    131 	{
    132 	  mpn_sub_1 (np + dn, np + dn, qn - dn, cy);
    133 	  cy = mpn_dcpi1_bdiv_qr_n (qp, np, dp, dn, dinv, tp);
    134 	  qp += dn;
    135 	  np += dn;
    136 	  qn -= dn;
    137 	}
    138       mpn_dcpi1_bdiv_q_n (qp, np, dp, dn, dinv, tp);
    139     }
    140   else
    141     {
    142       if (BELOW_THRESHOLD (qn, DC_BDIV_Q_THRESHOLD))
    143 	mpn_sbpi1_bdiv_q (qp, np, qn, dp, qn, dinv);
    144       else
    145 	mpn_dcpi1_bdiv_q_n (qp, np, dp, qn, dinv, tp);
    146     }
    147 
    148   TMP_FREE;
    149 }
    150