source/libs/gmp/gmp-src/mpn/cray/add_n.c

   1 /* Cray PVP mpn_add_n -- add two limb vectors and store their sum in a third
   2    limb vector.
   3
   4 Copyright 1996, 2000, 2001 Free Software Foundation, Inc.
   5
   6 This file is part of the GNU MP Library.
   7
   8 The GNU MP Library is free software; you can redistribute it and/or modify
   9 it under the terms of either:
  10
  11   * the GNU Lesser General Public License as published by the Free
  12     Software Foundation; either version 3 of the License, or (at your
  13     option) any later version.
  14
  15 or
  16
  17   * the GNU General Public License as published by the Free Software
  18     Foundation; either version 2 of the License, or (at your option) any
  19     later version.
  20
  21 or both in parallel, as here.
  22
  23 The GNU MP Library is distributed in the hope that it will be useful, but
  24 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  25 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  26 for more details.
  27
  28 You should have received copies of the GNU General Public License and the
  29 GNU Lesser General Public License along with the GNU MP Library.  If not,
  30 see https://www.gnu.org/licenses/.  */
  31
  32 /* This code runs at 4 cycles/limb.  It may be possible to bring it down
  33    to 3 cycles/limb.  */
  34
  35 #include "gmp.h"
  36 #include "gmp-impl.h"
  37
  38 mp_limb_t
  39 mpn_add_n (mp_ptr rp, mp_srcptr up, mp_srcptr vp, mp_size_t n)
  40 {
  41   mp_limb_t cy[n];
  42   mp_limb_t a, b, r, s0, c0, c1;
  43   mp_size_t i;
  44   int more_carries;
  45
  46   /* Main add loop.  Generate a raw output sum in rp[] and a carry vector
  47      in cy[].  */
  48 #pragma _CRI ivdep
  49   for (i = 0; i < n; i++)
  50     {
  51       a = up[i];
  52       b = vp[i];
  53       s0 = a + b;
  54       rp[i] = s0;
  55       c0 = ((a & b) | ((a | b) & ~s0)) >> 63;
  56       cy[i] = c0;
  57     }
  58   /* Carry add loop.  Add the carry vector cy[] to the raw sum rp[] and
  59      store the new sum back to rp[0].  If this generates further carry, set
  60      more_carries.  */
  61   more_carries = 0;
  62 #pragma _CRI ivdep
  63   for (i = 1; i < n; i++)
  64     {
  65       r = rp[i];
  66       c0 = cy[i - 1];
  67       s0 = r + c0;
  68       rp[i] = s0;
  69       c0 = (r & ~s0) >> 63;
  70       more_carries += c0;
  71     }
  72   /* If that second loop generated carry, handle that in scalar loop.  */
  73   if (more_carries)
  74     {
  75       mp_limb_t cyrec = 0;
  76       /* Look for places where rp[k] is zero and cy[k-1] is non-zero.
  77          These are where we got a recurrency carry.  */
  78       for (i = 1; i < n; i++)
  79         {
  80           r = rp[i];
  81           c0 = (r == 0 && cy[i - 1] != 0);
  82           s0 = r + cyrec;
  83           rp[i] = s0;
  84           c1 = (r & ~s0) >> 63;
  85           cyrec = c0 | c1;
  86         }
  87       return cyrec | cy[n - 1];
  88     }
  89
  90   return cy[n - 1];
  91 }