Annotation of OpenXM_contrib/gmp/mpn/cray/add_n.c, Revision 1.1.1.2
1.1.1.2 ! ohara 1: /* Cray PVP mpn_add_n -- add two limb vectors and store their sum in a third
! 2: limb vector.
1.1 maekawa 3:
1.1.1.2 ! ohara 4: Copyright 1996, 2000, 2001 Free Software Foundation, Inc.
1.1 maekawa 5:
1.1.1.2 ! ohara 6: This file is part of the GNU MP Library.
1.1 maekawa 7:
1.1.1.2 ! ohara 8: The GNU MP Library is free software; you can redistribute it and/or modify
! 9: it under the terms of the GNU Lesser General Public License as published by
! 10: the Free Software Foundation; either version 2.1 of the License, or (at your
! 11: option) any later version.
! 12:
! 13: The GNU MP Library is distributed in the hope that it will be useful, but
! 14: WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
! 15: or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
! 16: License for more details.
! 17:
! 18: You should have received a copy of the GNU Lesser General Public License
! 19: along with the GNU MP Library; see the file COPYING.LIB. If not, write to
! 20: the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
! 21: MA 02111-1307, USA. */
! 22:
! 23: /* This code runs at 4 cycles/limb. It may be possible to bring it down
! 24: to 3 cycles/limb. */
1.1 maekawa 25:
26: #include "gmp.h"
27: #include "gmp-impl.h"
28:
29: mp_limb_t
1.1.1.2 ! ohara 30: mpn_add_n (mp_ptr rp, mp_srcptr up, mp_srcptr vp, mp_size_t n)
1.1 maekawa 31: {
1.1.1.2 ! ohara 32: mp_limb_t cy[n];
! 33: mp_limb_t a, b, r, s0, c0, c1;
1.1 maekawa 34: mp_size_t i;
1.1.1.2 ! ohara 35: int more_carries;
1.1 maekawa 36:
1.1.1.2 ! ohara 37: /* Main add loop. Generate a raw output sum in rp[] and a carry vector
! 38: in cy[]. */
! 39: #pragma _CRI ivdep
! 40: for (i = 0; i < n; i++)
1.1 maekawa 41: {
1.1.1.2 ! ohara 42: a = up[i];
! 43: b = vp[i];
! 44: s0 = a + b;
! 45: rp[i] = s0;
! 46: c0 = ((a & b) | ((a | b) & ~s0)) >> 63;
! 47: cy[i] = c0;
1.1 maekawa 48: }
1.1.1.2 ! ohara 49: /* Carry add loop. Add the carry vector cy[] to the raw sum rp[] and
! 50: store the new sum back to rp[0]. If this generates further carry, set
! 51: more_carries. */
! 52: more_carries = 0;
! 53: #pragma _CRI ivdep
! 54: for (i = 1; i < n; i++)
1.1 maekawa 55: {
1.1.1.2 ! ohara 56: r = rp[i];
! 57: c0 = cy[i - 1];
! 58: s0 = r + c0;
! 59: rp[i] = s0;
! 60: c0 = (r & ~s0) >> 63;
! 61: more_carries += c0;
1.1 maekawa 62: }
1.1.1.2 ! ohara 63: /* If that second loop generated carry, handle that in scalar loop. */
1.1 maekawa 64: if (more_carries)
65: {
1.1.1.2 ! ohara 66: mp_limb_t cyrec = 0;
! 67: /* Look for places where rp[k] is zero and cy[k-1] is non-zero.
! 68: These are where we got a recurrency carry. */
! 69: for (i = 1; i < n; i++)
1.1 maekawa 70: {
1.1.1.2 ! ohara 71: r = rp[i];
! 72: c0 = (r == 0 && cy[i - 1] != 0);
! 73: s0 = r + cyrec;
! 74: rp[i] = s0;
! 75: c1 = (r & ~s0) >> 63;
! 76: cyrec = c0 | c1;
1.1 maekawa 77: }
1.1.1.2 ! ohara 78: return cyrec | cy[n - 1];
1.1 maekawa 79: }
80:
1.1.1.2 ! ohara 81: return cy[n - 1];
1.1 maekawa 82: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>