1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
|
dnl Intel Pentium mpn_rshift -- mpn right shift.
dnl
dnl cycles/limb
dnl P5,P54: 6.0
dnl P55: 5.375
dnl Copyright (C) 1992, 1994, 1995, 1996, 1999, 2000 Free Software
dnl Foundation, Inc.
dnl
dnl This file is part of the GNU MP Library.
dnl
dnl The GNU MP Library is free software; you can redistribute it and/or
dnl modify it under the terms of the GNU Lesser General Public License as
dnl published by the Free Software Foundation; either version 2.1 of the
dnl License, or (at your option) any later version.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful,
dnl but WITHOUT ANY WARRANTY; without even the implied warranty of
dnl MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
dnl Lesser General Public License for more details.
dnl
dnl You should have received a copy of the GNU Lesser General Public
dnl License along with the GNU MP Library; see the file COPYING.LIB. If
dnl not, write to the Free Software Foundation, Inc., 59 Temple Place -
dnl Suite 330, Boston, MA 02111-1307, USA.
include(`../config.m4')
C mp_limb_t mpn_rshift (mp_ptr dst, mp_srcptr src, mp_size_t size,
C unsigned shift);
C
C The main shift-by-N loop should run at 5.375 c/l and that's what P55 does,
C but P5 and P54 run only at 6.0 c/l, which is 4 cycles lost somewhere.
defframe(PARAM_SHIFT,16)
defframe(PARAM_SIZE, 12)
defframe(PARAM_SRC, 8)
defframe(PARAM_DST, 4)
.text
ALIGN(8)
PROLOGUE(mpn_rshift)
pushl %edi
pushl %esi
pushl %ebx
pushl %ebp
deflit(`FRAME',16)
movl PARAM_DST,%edi
movl PARAM_SRC,%esi
movl PARAM_SIZE,%ebp
movl PARAM_SHIFT,%ecx
C We can use faster code for shift-by-1 under certain conditions.
cmp $1,%ecx
jne L(normal)
leal 4(%edi),%eax
cmpl %esi,%eax
jnc L(special) C jump if res_ptr + 1 >= s_ptr
leal (%edi,%ebp,4),%eax
cmpl %eax,%esi
jnc L(special) C jump if s_ptr >= res_ptr + size
L(normal):
movl (%esi),%edx
addl $4,%esi
xorl %eax,%eax
shrdl( %cl, %edx, %eax) C compute carry limb
pushl %eax C push carry limb onto stack
decl %ebp
pushl %ebp
shrl $3,%ebp
jz L(end)
movl (%edi),%eax C fetch destination cache line
ALIGN(4)
L(oop): movl 28(%edi),%eax C fetch destination cache line
movl %edx,%ebx
movl (%esi),%eax
movl 4(%esi),%edx
shrdl( %cl, %eax, %ebx)
shrdl( %cl, %edx, %eax)
movl %ebx,(%edi)
movl %eax,4(%edi)
movl 8(%esi),%ebx
movl 12(%esi),%eax
shrdl( %cl, %ebx, %edx)
shrdl( %cl, %eax, %ebx)
movl %edx,8(%edi)
movl %ebx,12(%edi)
movl 16(%esi),%edx
movl 20(%esi),%ebx
shrdl( %cl, %edx, %eax)
shrdl( %cl, %ebx, %edx)
movl %eax,16(%edi)
movl %edx,20(%edi)
movl 24(%esi),%eax
movl 28(%esi),%edx
shrdl( %cl, %eax, %ebx)
shrdl( %cl, %edx, %eax)
movl %ebx,24(%edi)
movl %eax,28(%edi)
addl $32,%esi
addl $32,%edi
decl %ebp
jnz L(oop)
L(end): popl %ebp
andl $7,%ebp
jz L(end2)
L(oop2):
movl (%esi),%eax
shrdl( %cl,%eax,%edx) C compute result limb
movl %edx,(%edi)
movl %eax,%edx
addl $4,%esi
addl $4,%edi
decl %ebp
jnz L(oop2)
L(end2):
shrl %cl,%edx C compute most significant limb
movl %edx,(%edi) C store it
popl %eax C pop carry limb
popl %ebp
popl %ebx
popl %esi
popl %edi
ret
C We loop from least significant end of the arrays, which is only
C permissable if the source and destination don't overlap, since the
C function is documented to work for overlapping source and destination.
L(special):
leal -4(%edi,%ebp,4),%edi
leal -4(%esi,%ebp,4),%esi
movl (%esi),%edx
subl $4,%esi
decl %ebp
pushl %ebp
shrl $3,%ebp
shrl %edx
incl %ebp
decl %ebp
jz L(Lend)
movl (%edi),%eax C fetch destination cache line
ALIGN(4)
L(Loop):
movl -28(%edi),%eax C fetch destination cache line
movl %edx,%ebx
movl (%esi),%eax
movl -4(%esi),%edx
rcrl %eax
movl %ebx,(%edi)
rcrl %edx
movl %eax,-4(%edi)
movl -8(%esi),%ebx
movl -12(%esi),%eax
rcrl %ebx
movl %edx,-8(%edi)
rcrl %eax
movl %ebx,-12(%edi)
movl -16(%esi),%edx
movl -20(%esi),%ebx
rcrl %edx
movl %eax,-16(%edi)
rcrl %ebx
movl %edx,-20(%edi)
movl -24(%esi),%eax
movl -28(%esi),%edx
rcrl %eax
movl %ebx,-24(%edi)
rcrl %edx
movl %eax,-28(%edi)
leal -32(%esi),%esi C use leal not to clobber carry
leal -32(%edi),%edi
decl %ebp
jnz L(Loop)
L(Lend):
popl %ebp
sbbl %eax,%eax C save carry in %eax
andl $7,%ebp
jz L(Lend2)
addl %eax,%eax C restore carry from eax
L(Loop2):
movl %edx,%ebx
movl (%esi),%edx
rcrl %edx
movl %ebx,(%edi)
leal -4(%esi),%esi C use leal not to clobber carry
leal -4(%edi),%edi
decl %ebp
jnz L(Loop2)
jmp L(L1)
L(Lend2):
addl %eax,%eax C restore carry from eax
L(L1): movl %edx,(%edi) C store last limb
movl $0,%eax
rcrl %eax
popl %ebp
popl %ebx
popl %esi
popl %edi
ret
EPILOGUE()
|
