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