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
|
dnl AMD K6-2 mpn_copyd -- copy limb vector, decrementing.
dnl Copyright 2001, 2002 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 K6-2: 1.0 cycles/limb
C void mpn_copyd (mp_ptr dst, mp_srcptr src, mp_size_t size);
C
C The loop here is no faster than a rep movsl at 1.0 c/l, but it avoids a 30
C cycle startup time, which amounts for instance to a 2x speedup at 15
C limbs.
C
C If dst is 4mod8 the loop would be 1.17 c/l, but that's avoided by
C processing one limb separately to make it aligned. This and a final odd
C limb are handled in a branch-free fashion, ending up re-copying if the
C special case isn't needed.
C
C Alternatives:
C
C There used to be a big unrolled version of this, running at 0.56 c/l if
C the destination was aligned, but that seemed rather excessive for the
C relative importance of copyd.
C
C If the destination alignment is ignored and just left to run at 1.17 c/l
C some code size and a fixed few cycles can be saved. Considering how few
C uses copyd finds perhaps that should be favoured. The current code has
C the attraction of being no slower than a basic rep movsl though.
defframe(PARAM_SIZE,12)
defframe(PARAM_SRC, 8)
defframe(PARAM_DST, 4)
dnl re-using parameter space
define(SAVE_EBX,`PARAM_SIZE')
TEXT
ALIGN(16)
PROLOGUE(mpn_copyd)
deflit(`FRAME',0)
movl PARAM_SIZE, %ecx
movl %ebx, SAVE_EBX
movl PARAM_SRC, %eax
movl PARAM_DST, %edx
subl $1, %ecx C better code alignment than decl
jb L(zero)
jz L(one_more)
leal 4(%edx,%ecx,4), %ebx
Zdisp( movd, 0,(%eax,%ecx,4), %mm0) C high limb
Zdisp( movd, %mm0, 0,(%edx,%ecx,4)) C Zdisp for good code alignment
cmpl $1, %ecx
je L(one_more)
shrl $2, %ebx
andl $1, %ebx C 1 if dst[size-2] unaligned
subl %ebx, %ecx
nop C code alignment
L(top):
C eax src
C ebx
C ecx counter
C edx dst
movq -4(%eax,%ecx,4), %mm0
subl $2, %ecx
movq %mm0, 4(%edx,%ecx,4)
ja L(top)
L(one_more):
movd (%eax), %mm0
movd %mm0, (%edx)
movl SAVE_EBX, %ebx
emms_or_femms
L(zero):
ret
EPILOGUE()
|
