Home | History | Annotate | Line # | Download | only in generic
toom2_sqr.c revision 1.1.1.3 
      1 /* mpn_toom2_sqr -- Square {ap,an}.
      2 
      3    Contributed to the GNU project by Torbjorn Granlund.
      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 WILL CHANGE OR DISAPPEAR IN A FUTURE GNU MP RELEASE.
      8 
      9 Copyright 2006-2010, 2012, 2014 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 
     38 #include "gmp.h"
     39 #include "gmp-impl.h"
     40 
     41 /* Evaluate in: -1, 0, +inf
     42 
     43   <-s--><--n-->
     44    ____ ______
     45   |_a1_|___a0_|
     46 
     47   v0  =  a0     ^2  #   A(0)^2
     48   vm1 = (a0- a1)^2  #  A(-1)^2
     49   vinf=      a1 ^2  # A(inf)^2
     50 */
     51 
     52 #if TUNE_PROGRAM_BUILD || WANT_FAT_BINARY
     53 #define MAYBE_sqr_toom2   1
     54 #else
     55 #define MAYBE_sqr_toom2							\
     56   (SQR_TOOM3_THRESHOLD >= 2 * SQR_TOOM2_THRESHOLD)
     57 #endif
     58 
     59 #define TOOM2_SQR_REC(p, a, n, ws)					\
     60   do {									\
     61     if (! MAYBE_sqr_toom2						\
     62 	|| BELOW_THRESHOLD (n, SQR_TOOM2_THRESHOLD))			\
     63       mpn_sqr_basecase (p, a, n);					\
     64     else								\
     65       mpn_toom2_sqr (p, a, n, ws);					\
     66   } while (0)
     67 
     68 void
     69 mpn_toom2_sqr (mp_ptr pp,
     70 	       mp_srcptr ap, mp_size_t an,
     71 	       mp_ptr scratch)
     72 {
     73   const int __gmpn_cpuvec_initialized = 1;
     74   mp_size_t n, s;
     75   mp_limb_t cy, cy2;
     76   mp_ptr asm1;
     77 
     78 #define a0  ap
     79 #define a1  (ap + n)
     80 
     81   s = an >> 1;
     82   n = an - s;
     83 
     84   ASSERT (0 < s && s <= n && s >= n - 1);
     85 
     86   asm1 = pp;
     87 
     88   /* Compute asm1.  */
     89   if (s == n)
     90     {
     91       if (mpn_cmp (a0, a1, n) < 0)
     92 	{
     93 	  mpn_sub_n (asm1, a1, a0, n);
     94 	}
     95       else
     96 	{
     97 	  mpn_sub_n (asm1, a0, a1, n);
     98 	}
     99     }
    100   else /* n - s == 1 */
    101     {
    102       if (a0[s] == 0 && mpn_cmp (a0, a1, s) < 0)
    103 	{
    104 	  mpn_sub_n (asm1, a1, a0, s);
    105 	  asm1[s] = 0;
    106 	}
    107       else
    108 	{
    109 	  asm1[s] = a0[s] - mpn_sub_n (asm1, a0, a1, s);
    110 	}
    111     }
    112 
    113 #define v0	pp				/* 2n */
    114 #define vinf	(pp + 2 * n)			/* s+s */
    115 #define vm1	scratch				/* 2n */
    116 #define scratch_out	scratch + 2 * n
    117 
    118   /* vm1, 2n limbs */
    119   TOOM2_SQR_REC (vm1, asm1, n, scratch_out);
    120 
    121   /* vinf, s+s limbs */
    122   TOOM2_SQR_REC (vinf, a1, s, scratch_out);
    123 
    124   /* v0, 2n limbs */
    125   TOOM2_SQR_REC (v0, ap, n, scratch_out);
    126 
    127   /* H(v0) + L(vinf) */
    128   cy = mpn_add_n (pp + 2 * n, v0 + n, vinf, n);
    129 
    130   /* L(v0) + H(v0) */
    131   cy2 = cy + mpn_add_n (pp + n, pp + 2 * n, v0, n);
    132 
    133   /* L(vinf) + H(vinf) */
    134   cy += mpn_add (pp + 2 * n, pp + 2 * n, n, vinf + n, s + s - n);
    135 
    136   cy -= mpn_sub_n (pp + n, pp + n, vm1, 2 * n);
    137 
    138   ASSERT (cy + 1  <= 3);
    139   ASSERT (cy2 <= 2);
    140 
    141   MPN_INCR_U (pp + 2 * n, s + s, cy2);
    142   if (LIKELY (cy <= 2))
    143     MPN_INCR_U (pp + 3 * n, s + s - n, cy);
    144   else
    145     MPN_DECR_U (pp + 3 * n, s + s - n, 1);
    146 }
    147