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
|
dnl AMD K7 mpn_copyi -- copy limb vector, incrementing.
dnl Copyright 1999, 2000, 2002, 2003 Free Software Foundation, Inc.
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 modify
dnl it under the terms of either:
dnl
dnl * the GNU Lesser General Public License as published by the Free
dnl Software Foundation; either version 3 of the License, or (at your
dnl option) any later version.
dnl
dnl or
dnl
dnl * the GNU General Public License as published by the Free Software
dnl Foundation; either version 2 of the License, or (at your option) any
dnl later version.
dnl
dnl or both in parallel, as here.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful, but
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
dnl for more details.
dnl
dnl You should have received copies of the GNU General Public License and the
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
dnl see https://www.gnu.org/licenses/.
include(`../config.m4')
C alignment dst/src, A=0mod8 N=4mod8
C A/A A/N N/A N/N
C K7 0.75 1.0 1.0 0.75
C void mpn_copyi (mp_ptr dst, mp_srcptr src, mp_size_t size);
C
C Copy src,size to dst,size.
C
C This code at 0.75 or 1.0 c/l is always faster than a plain rep movsl at
C 1.33 c/l.
C
C The K7 can do a 64-bit load and 64-bit store in one cycle (optimization
C guile 22007 appendix B), so 0.5 c/l should be possible, however nothing
C under 0.7 c/l is known. Apparently only two 32-bit stores can be done in
C one cycle, so perhaps some scheduling is needed to ensure it's a
C load+store in each cycle, not store+store.
C
C If both source and destination are unaligned then one limb is processed at
C the start to make them aligned and so get 0.75 c/l, whereas if they'd been
C used unaligned it would be 1.5 c/l.
defframe(PARAM_SIZE,12)
defframe(PARAM_SRC, 8)
defframe(PARAM_DST, 4)
dnl parameter space reused
define(SAVE_EBX,`PARAM_SIZE')
dnl minimum 5 since the unrolled code can't handle less than 5
deflit(UNROLL_THRESHOLD, 5)
TEXT
ALIGN(32)
PROLOGUE(mpn_copyi)
deflit(`FRAME',0)
movl PARAM_SIZE, %ecx
movl %ebx, SAVE_EBX
movl PARAM_SRC, %eax
movl PARAM_DST, %edx
cmpl $UNROLL_THRESHOLD, %ecx
jae L(unroll)
orl %ecx, %ecx
jz L(simple_done)
L(simple):
C eax src, incrementing
C ebx scratch
C ecx counter
C edx dst, incrementing
C
C this loop is 2 cycles/limb
movl (%eax), %ebx
movl %ebx, (%edx)
decl %ecx
leal 4(%eax), %eax
leal 4(%edx), %edx
jnz L(simple)
L(simple_done):
movl SAVE_EBX, %ebx
ret
L(unroll):
movl %eax, %ebx
leal -12(%eax,%ecx,4), %eax C src end - 12
subl $3, %ecx C size-3
andl %edx, %ebx
leal (%edx,%ecx,4), %edx C dst end - 12
negl %ecx
testl $4, %ebx C testl to pad code closer to 16 bytes for L(top)
jz L(aligned)
C both src and dst unaligned, process one limb to align them
movl (%eax,%ecx,4), %ebx
movl %ebx, (%edx,%ecx,4)
incl %ecx
L(aligned):
ALIGN(16)
L(top):
C eax src end - 12
C ebx
C ecx counter, negative, limbs
C edx dst end - 12
movq (%eax,%ecx,4), %mm0
movq 8(%eax,%ecx,4), %mm1
addl $4, %ecx
movq %mm0, -16(%edx,%ecx,4)
movq %mm1, -16+8(%edx,%ecx,4)
ja L(top) C jump no carry and not zero
C now %ecx is 0 to 3 representing respectively 3 to 0 limbs remaining
testb $2, %cl
jnz L(finish_not_two)
movq (%eax,%ecx,4), %mm0
movq %mm0, (%edx,%ecx,4)
L(finish_not_two):
testb $1, %cl
jnz L(done)
movl 8(%eax), %ebx
movl %ebx, 8(%edx)
L(done):
movl SAVE_EBX, %ebx
emms
ret
EPILOGUE()
|
