Annotation of OpenXM_contrib/gmp/mpn/pa32/hppa2_0/sqr_diagonal.asm, Revision 1.1.1.1
1.1 ohara 1: dnl HP-PA 32-bit mpn_sqr_diagonal optimized for the PA8x00.
2:
3: dnl Copyright 2001, 2002 Free Software Foundation, Inc.
4:
5: dnl This file is part of the GNU MP Library.
6:
7: dnl The GNU MP Library is free software; you can redistribute it and/or modify
8: dnl it under the terms of the GNU Lesser General Public License as published
9: dnl by the Free Software Foundation; either version 2.1 of the License, or (at
10: dnl your option) any later version.
11:
12: dnl The GNU MP Library is distributed in the hope that it will be useful, but
13: dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
14: dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
15: dnl License for more details.
16:
17: dnl You should have received a copy of the GNU Lesser General Public License
18: dnl along with the GNU MP Library; see the file COPYING.LIB. If not, write to
19: dnl the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
20: dnl MA 02111-1307, USA.
21:
22: include(`../config.m4')
23:
24: C This code runs at 6 cycles/limb on the PA7100 and 2 cycles/limb on PA8x00.
25: C The 2-way unrolling is actually not helping the PA7100.
26:
27: C INPUT PARAMETERS
28: define(`rp',`%r26')
29: define(`up',`%r25')
30: define(`n',`%r24')
31:
32: ASM_START()
33: PROLOGUE(mpn_sqr_diagonal)
34:
35: fldws,ma 4(up),%fr4r
36: addib,= -1,n,L(end1)
37: ldo 4(rp),rp
38:
39: fldws,ma 4(up),%fr6r
40: addib,= -1,n,L(end2)
41: xmpyu %fr4r,%fr4r,%fr5
42:
43: fldws,ma 4(up),%fr4r
44: addib,= -1,n,L(end3)
45: xmpyu %fr6r,%fr6r,%fr7
46:
47:
48: .label L(loop)
49: fldws,ma 4(up),%fr6r
50: fstws %fr5r,-4(rp)
51: fstws,ma %fr5l,8(rp)
52: addib,= -1,n,L(exite)
53: xmpyu %fr4r,%fr4r,%fr5
54: fldws,ma 4(up),%fr4r
55: fstws %fr7r,-4(rp)
56: fstws,ma %fr7l,8(rp)
57: addib,<> -1,n,L(loop)
58: xmpyu %fr6r,%fr6r,%fr7
59:
60: .label L(exito)
61: fstws %fr5r,-4(rp)
62: fstws %fr5l,0(rp)
63: xmpyu %fr4r,%fr4r,%fr5
64: fstws %fr7r,4(rp)
65: fstws %fr7l,8(rp)
66: fstws,mb %fr5r,12(rp)
67: bv 0(%r2)
68: fstws %fr5l,4(rp)
69:
70: .label L(exite)
71: fstws %fr7r,-4(rp)
72: fstws %fr7l,0(rp)
73: xmpyu %fr6r,%fr6r,%fr7
74: fstws %fr5r,4(rp)
75: fstws %fr5l,8(rp)
76: fstws,mb %fr7r,12(rp)
77: bv 0(%r2)
78: fstws %fr7l,4(rp)
79:
80: .label L(end1)
81: xmpyu %fr4r,%fr4r,%fr5
82: fstws %fr5r,-4(rp)
83: bv 0(%r2)
84: fstws,ma %fr5l,8(rp)
85:
86: .label L(end2)
87: xmpyu %fr6r,%fr6r,%fr7
88: fstws %fr5r,-4(rp)
89: fstws %fr5l,0(rp)
90: fstws %fr7r,4(rp)
91: bv 0(%r2)
92: fstws %fr7l,8(rp)
93:
94: .label L(end3)
95: fstws %fr5r,-4(rp)
96: fstws %fr5l,0(rp)
97: xmpyu %fr4r,%fr4r,%fr5
98: fstws %fr7r,4(rp)
99: fstws %fr7l,8(rp)
100: fstws,mb %fr5r,12(rp)
101: bv 0(%r2)
102: fstws %fr5l,4(rp)
103: EPILOGUE(mpn_sqr_diagonal)
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>