![[BACK]](/icons/cvsweb/back.gif){kind=icon}
Return to mod_34lsub1.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 mod_34lsub1.asm CVS log | Up to [local] / OpenXM_contrib / gmp / mpn / x86 |
1.1 ohara 1: dnl Generic x86 mpn_mod_32lsub1 -- mpn remainder modulo 2^24-1.
2:
3: dnl Copyright 2000, 2001, 2002 Free Software Foundation, Inc.
4: dnl
5: dnl This file is part of the GNU MP Library.
6: dnl
7: dnl The GNU MP Library is free software; you can redistribute it and/or
8: dnl modify it under the terms of the GNU Lesser General Public License as
9: dnl published by the Free Software Foundation; either version 2.1 of the
10: dnl License, or (at your option) any later version.
11: dnl
12: dnl The GNU MP Library is distributed in the hope that it will be useful,
13: dnl but WITHOUT ANY WARRANTY; without even the implied warranty of
14: dnl MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15: dnl Lesser General Public License for more details.
16: dnl
17: dnl You should have received a copy of the GNU Lesser General Public
18: dnl License along with the GNU MP Library; see the file COPYING.LIB. If
19: dnl not, write to the Free Software Foundation, Inc., 59 Temple Place -
20: dnl Suite 330, Boston, MA 02111-1307, USA.
21:
22: include(`../config.m4')
23:
24:
25: C cycles/limb
26: C P5: 3.0
27: C P6: 3.66
28: C K6: 3.0
29: C K7: 1.3
30: C P4: 9
31:
32:
33: C mp_limb_t mpn_mod_34lsub1 (mp_srcptr src, mp_size_t size)
34: C
35:
36: defframe(PARAM_SIZE, 8)
37: defframe(PARAM_SRC, 4)
38:
39: dnl re-use parameter space
40: define(SAVE_EBX, `PARAM_SRC')
41:
42: TEXT
43: ALIGN(16)
44: PROLOGUE(mpn_mod_34lsub1)
45: deflit(`FRAME',0)
46:
47: movl PARAM_SIZE, %ecx
48: movl PARAM_SRC, %edx
49:
50: subl $2, %ecx
51: ja L(three_or_more)
52:
53: movl (%edx), %eax
54: jb L(one)
55:
56: movl 4(%edx), %ecx
57: movl %eax, %edx
58: shrl $24, %eax C src[0] low
59:
60: andl $0xFFFFFF, %edx C src[0] high
61: addl %edx, %eax
62: movl %ecx, %edx
63:
64: andl $0xFFFF, %ecx
65: shrl $16, %edx C src[1] high
66: addl %edx, %eax
67:
68: shll $8, %ecx C src[1] low
69: addl %ecx, %eax
70:
71: L(one):
72: ret
73:
74:
75: L(three_or_more):
76: C eax
77: C ebx
78: C ecx size-2
79: C edx src
80: C esi
81: C edi
82: C ebp
83:
84: movl %ebx, SAVE_EBX C and arrange 16-byte loop alignment
85: xorl %ebx, %ebx
86:
87: pushl %esi FRAME_pushl()
88: xorl %esi, %esi
89:
90: pushl %edi FRAME_pushl()
91: xorl %eax, %eax C and clear carry flag
92:
93:
94: C offset 0x40 here
95: L(top):
96: C eax acc 0mod3
97: C ebx acc 1mod3
98: C ecx counter, limbs
99: C edx src
100: C esi acc 2mod3
101: C edi
102: C ebp
103:
104: leal 12(%edx), %edx
105: leal -2(%ecx), %ecx
106:
107: adcl -12(%edx), %eax
108: adcl -8(%edx), %ebx
109: adcl -4(%edx), %esi
110:
111: decl %ecx
112: jg L(top)
113:
114:
115: C ecx is -2, -1 or 0 representing 0, 1 or 2 more limbs, respectively
116:
117: movl $0xFFFFFFFF, %edi
118: incl %ecx
119: js L(combine)
120:
121: adcl (%edx), %eax
122: movl $0xFFFFFF00, %edi
123: decl %ecx
124: js L(combine)
125:
126: adcl 4(%edx), %ebx
127: movl $0xFFFF0000, %edi
128:
129:
130: L(combine):
131: C eax acc 0mod3
132: C ebx acc 1mod3
133: C ecx
134: C edx
135: C esi acc 2mod3
136: C edi mask
137: C ebp
138:
139: sbbl %ecx, %ecx C carry
140: movl %eax, %edx C 0mod3
141:
142: shrl $24, %eax C 0mod3 high
143: andl %edi, %ecx C carry masked
144:
145: subl %ecx, %eax C apply carry
146: movl %ebx, %edi C 1mod3
147:
148: shrl $16, %ebx C 1mod3 high
149: andl $0x00FFFFFF, %edx C 0mod3 low
150:
151: addl %edx, %eax C apply 0mod3 low
152: andl $0xFFFF, %edi
153:
154: shll $8, %edi C 1mod3 low
155: addl %ebx, %eax C apply 1mod3 high
156:
157: addl %edi, %eax C apply 1mod3 low
158: movl %esi, %edx C 2mod3
159:
160: shrl $8, %esi C 2mod3 high
161: andl $0xFF, %edx C 2mod3 low
162:
163: shll $16, %edx C 2mod3 low
164: addl %esi, %eax C apply 2mod3 high
165:
166: addl %edx, %eax C apply 2mod3 low
167: popl %edi FRAME_popl()
168:
169: movl SAVE_EBX, %ebx
170: popl %esi FRAME_popl()
171:
172: ret
173:
174: EPILOGUE()