Annotation of OpenXM_contrib/gmp/mpn/cray/ieee/mul_basecase.c, Revision 1.1
1.1 ! ohara 1: /* Cray PVP/IEEE mpn_mul_basecase.
! 2:
! 3: Copyright 2000, 2001 Free Software Foundation, Inc.
! 4:
! 5: This file is part of the GNU MP Library.
! 6:
! 7: The GNU MP Library is free software; you can redistribute it and/or modify
! 8: it under the terms of the GNU Lesser General Public License as published by
! 9: the Free Software Foundation; either version 2.1 of the License, or (at your
! 10: option) any later version.
! 11:
! 12: The GNU MP Library is distributed in the hope that it will be useful, but
! 13: WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
! 14: or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
! 15: License for more details.
! 16:
! 17: You should have received a copy of the GNU Lesser General Public License
! 18: along with the GNU MP Library; see the file COPYING.LIB. If not, write to
! 19: the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
! 20: MA 02111-1307, USA. */
! 21:
! 22: /* The most critical loop of this code runs at about 5 cycles/limb on a T90.
! 23: That is not perfect, mainly due to vector register shortage. */
! 24:
! 25: #include <intrinsics.h>
! 26: #include "gmp.h"
! 27: #include "gmp-impl.h"
! 28:
! 29: void
! 30: mpn_mul_basecase (mp_ptr rp,
! 31: mp_srcptr up, mp_size_t un,
! 32: mp_srcptr vp, mp_size_t vn)
! 33: {
! 34: mp_limb_t cy[un + vn];
! 35: mp_limb_t vl;
! 36: mp_limb_t a, b, r, s0, s1, c0, c1;
! 37: mp_size_t i, j;
! 38: int more_carries;
! 39:
! 40: for (i = 0; i < un + vn; i++)
! 41: {
! 42: rp[i] = 0;
! 43: cy[i] = 0;
! 44: }
! 45:
! 46: #pragma _CRI novector
! 47: for (j = 0; j < vn; j++)
! 48: {
! 49: vl = vp[j];
! 50:
! 51: a = up[0] * vl;
! 52: r = rp[j];
! 53: s0 = a + r;
! 54: rp[j] = s0;
! 55: c0 = ((a & r) | ((a | r) & ~s0)) >> 63;
! 56: cy[j] += c0;
! 57:
! 58: #pragma _CRI ivdep
! 59: for (i = 1; i < un; i++)
! 60: {
! 61: a = up[i] * vl;
! 62: b = _int_mult_upper (up[i - 1], vl);
! 63: s0 = a + b;
! 64: c0 = ((a & b) | ((a | b) & ~s0)) >> 63;
! 65: r = rp[j + i];
! 66: s1 = s0 + r;
! 67: rp[j + i] = s1;
! 68: c1 = ((s0 & r) | ((s0 | r) & ~s1)) >> 63;
! 69: cy[j + i] += c0 + c1;
! 70: }
! 71: rp[j + un] = _int_mult_upper (up[un - 1], vl);
! 72: }
! 73:
! 74: more_carries = 0;
! 75: #pragma _CRI ivdep
! 76: for (i = 1; i < un + vn; i++)
! 77: {
! 78: r = rp[i];
! 79: c0 = cy[i - 1];
! 80: s0 = r + c0;
! 81: rp[i] = s0;
! 82: c0 = (r & ~s0) >> 63;
! 83: more_carries += c0;
! 84: }
! 85: /* If that second loop generated carry, handle that in scalar loop. */
! 86: if (more_carries)
! 87: {
! 88: mp_limb_t cyrec = 0;
! 89: for (i = 1; i < un + vn; i++)
! 90: {
! 91: r = rp[i];
! 92: c0 = (r < cy[i - 1]);
! 93: s0 = r + cyrec;
! 94: rp[i] = s0;
! 95: c1 = (r & ~s0) >> 63;
! 96: cyrec = c0 | c1;
! 97: }
! 98: }
! 99: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>