![[BACK]](/icons/cvsweb/back.gif){kind=icon}
Return to mul_1.asm CVS log ![[TXT]](/icons/cvsweb/text.gif){kind=icon}
![[DIR]](/icons/cvsweb/dir.gif){kind=icon}
Up to [local] / OpenXM_contrib / gmp / mpn / x86|
Return to mul_1.asm CVS log | Up to [local] / OpenXM_contrib / gmp / mpn / x86 |
1.1 maekawa 1: dnl x86 mpn_mul_1 (for 386, 486, and Pentium Pro) -- Multiply a limb vector
2: dnl with a limb and store the result in a second limb vector.
3:
1.1.1.2 ! ohara 4: dnl Copyright 1992, 1994, 1997, 1998, 1999, 2000, 2001, 2002 Free Software
1.1 maekawa 5: dnl Foundation, Inc.
6: dnl
7: dnl This file is part of the GNU MP Library.
8: dnl
9: dnl The GNU MP Library is free software; you can redistribute it and/or
10: dnl modify it under the terms of the GNU Lesser General Public License as
11: dnl published by the Free Software Foundation; either version 2.1 of the
12: dnl License, or (at your option) any later version.
13: dnl
14: dnl The GNU MP Library is distributed in the hope that it will be useful,
15: dnl but WITHOUT ANY WARRANTY; without even the implied warranty of
16: dnl MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17: dnl Lesser General Public License for more details.
18: dnl
19: dnl You should have received a copy of the GNU Lesser General Public
20: dnl License along with the GNU MP Library; see the file COPYING.LIB. If
21: dnl not, write to the Free Software Foundation, Inc., 59 Temple Place -
22: dnl Suite 330, Boston, MA 02111-1307, USA.
23:
24: include(`../config.m4')
1.1.1.2 ! ohara 25:
! 26:
! 27: C cycles/limb
! 28: C P5: 12.5
! 29: C P6: 5.5
! 30: C K6: 10.5
! 31: C K7: 4.5
! 32: C P4: 19
1.1 maekawa 33:
34:
35: C mp_limb_t mpn_mul_1 (mp_ptr dst, mp_srcptr src, mp_size_t size,
36: C mp_limb_t multiplier);
37:
38: defframe(PARAM_MULTIPLIER,16)
39: defframe(PARAM_SIZE, 12)
40: defframe(PARAM_SRC, 8)
41: defframe(PARAM_DST, 4)
42:
43: TEXT
44: ALIGN(8)
45: PROLOGUE(mpn_mul_1)
46: deflit(`FRAME',0)
47:
48: pushl %edi
49: pushl %esi
50: pushl %ebx
51: pushl %ebp
52: deflit(`FRAME',16)
53:
54: movl PARAM_DST,%edi
55: movl PARAM_SRC,%esi
56: movl PARAM_SIZE,%ecx
57:
58: xorl %ebx,%ebx
59: andl $3,%ecx
60: jz L(end0)
61:
62: L(oop0):
63: movl (%esi),%eax
64: mull PARAM_MULTIPLIER
65: leal 4(%esi),%esi
66: addl %ebx,%eax
67: movl $0,%ebx
68: adcl %ebx,%edx
69: movl %eax,(%edi)
70: movl %edx,%ebx C propagate carry into cylimb
71:
72: leal 4(%edi),%edi
73: decl %ecx
74: jnz L(oop0)
75:
76: L(end0):
77: movl PARAM_SIZE,%ecx
78: shrl $2,%ecx
79: jz L(end)
80:
81:
82: ALIGN(8)
83: L(oop): movl (%esi),%eax
84: mull PARAM_MULTIPLIER
85: addl %eax,%ebx
86: movl $0,%ebp
87: adcl %edx,%ebp
88:
89: movl 4(%esi),%eax
90: mull PARAM_MULTIPLIER
91: movl %ebx,(%edi)
92: addl %eax,%ebp C new lo + cylimb
93: movl $0,%ebx
94: adcl %edx,%ebx
95:
96: movl 8(%esi),%eax
97: mull PARAM_MULTIPLIER
98: movl %ebp,4(%edi)
99: addl %eax,%ebx C new lo + cylimb
100: movl $0,%ebp
101: adcl %edx,%ebp
102:
103: movl 12(%esi),%eax
104: mull PARAM_MULTIPLIER
105: movl %ebx,8(%edi)
106: addl %eax,%ebp C new lo + cylimb
107: movl $0,%ebx
108: adcl %edx,%ebx
109:
110: movl %ebp,12(%edi)
111:
112: leal 16(%esi),%esi
113: leal 16(%edi),%edi
114: decl %ecx
115: jnz L(oop)
116:
117: L(end): movl %ebx,%eax
118:
119: popl %ebp
120: popl %ebx
121: popl %esi
122: popl %edi
123: ret
124:
125: EPILOGUE()