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
|
dnl AMD64 mpn_copyi optimised for CPUs with fast SSE.
dnl Copyright 2003, 2005, 2007, 2011, 2012 Free Software Foundation, Inc.
dnl This file is part of the GNU MP Library.
dnl The GNU MP Library is free software; you can redistribute it and/or modify
dnl it under the terms of the GNU Lesser General Public License as published
dnl by the Free Software Foundation; either version 3 of the License, or (at
dnl your option) any later version.
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 Lesser General Public
dnl License for more details.
dnl You should have received a copy of the GNU Lesser General Public License
dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/.
include(`../config.m4')
C cycles/limb good for cpu?
C AMD K8,K9
C AMD K10 0.85 1.64 Y/N
C AMD bd1 1.4 1.4 Y
C AMD bobcat
C Intel P4 2.3 2.3 Y
C Intel core2 1.0 1.0
C Intel NHM 0.5 0.67 Y
C Intel SBR 0.5 0.75 Y
C Intel atom
C VIA nano 1.16 5.16 Y/N
C We try to do as many 16-byte operations as possible. The top-most and
C bottom-most writes might need 8-byte operations. We can always write using
C aligned 16-byte operations, we read with both aligned and unaligned 16-byte
C operations.
C Instead of having separate loops for reading aligned and unaligned, we read
C using MOVDQU. This seems to work great except for core2; there performance
C doubles when reading using MOVDQA (for aligned source). It is unclear how to
C best handle the unaligned case there.
C INPUT PARAMETERS
define(`rp', `%rdi')
define(`up', `%rsi')
define(`n', `%rdx')
ABI_SUPPORT(DOS64)
ABI_SUPPORT(STD64)
dnl define(`movdqu', lddqu)
ASM_START()
TEXT
ALIGN(64)
PROLOGUE(mpn_copyi)
DOS64_ENTRY(3)
cmp $3, n
jc L(bc)
test $8, R8(rp) C is rp 16-byte aligned?
jz L(ali) C jump if rp aligned
movsq C copy single limb
dec n
sub $16, n
jc L(sma)
ALIGN(16)
L(top): movdqu (up), %xmm0
movdqu 16(up), %xmm1
movdqu 32(up), %xmm2
movdqu 48(up), %xmm3
movdqu 64(up), %xmm4
movdqu 80(up), %xmm5
movdqu 96(up), %xmm6
movdqu 112(up), %xmm7
lea 128(up), up
movdqa %xmm0, (rp)
movdqa %xmm1, 16(rp)
movdqa %xmm2, 32(rp)
movdqa %xmm3, 48(rp)
movdqa %xmm4, 64(rp)
movdqa %xmm5, 80(rp)
movdqa %xmm6, 96(rp)
movdqa %xmm7, 112(rp)
lea 128(rp), rp
L(ali): sub $16, n
jnc L(top)
L(sma): test $8, R8(n)
jz 1f
movdqu (up), %xmm0
movdqu 16(up), %xmm1
movdqu 32(up), %xmm2
movdqu 48(up), %xmm3
lea 64(up), up
movdqa %xmm0, (rp)
movdqa %xmm1, 16(rp)
movdqa %xmm2, 32(rp)
movdqa %xmm3, 48(rp)
lea 64(rp), rp
1:
test $4, R8(n)
jz 1f
movdqu (up), %xmm0
movdqu 16(up), %xmm1
lea 32(up), up
movdqa %xmm0, (rp)
movdqa %xmm1, 16(rp)
lea 32(rp), rp
1:
test $2, R8(n)
jz 1f
movdqu (up), %xmm0
lea 16(up), up
movdqa %xmm0, (rp)
lea 16(rp), rp
ALIGN(16)
1:
L(end): bt $0, n
jnc 1f
mov (up), %r8
mov %r8, (rp)
1:
DOS64_EXIT()
ret
C Basecase code. Needed for good small operands speed, not for
C correctness as the above code is currently written.
L(bc): sub $2, n
jc L(end)
ALIGN(16)
1: mov (up), %rax
mov 8(up), %rcx
lea 16(up), up
mov %rax, (rp)
mov %rcx, 8(rp)
lea 16(rp), rp
sub $2, n
jnc 1b
bt $0, n
jnc L(ret)
mov (up), %rax
mov %rax, (rp)
L(ret): DOS64_EXIT()
ret
EPILOGUE()
|