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
|
dnl Intel Pentium-4 mpn_rsh1add_n -- mpn (x+y)/2
dnl Copyright 2001-2004 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 cycles/limb (approx)
C dst!=src1,2 dst==src1 dst==src2
C P4: 4.5 6.5 6.5
C mp_limb_t mpn_rsh1add_n (mp_ptr wp, mp_srcptr xp, mp_srcptr yp,
C mp_size_t size);
C
C The slightly strange combination of indexing and pointer incrementing
C that's used seems to work best. Not sure why, but for instance leal
C incrementing on %esi is a 1 or 2 cycle slowdown.
C
C The dependent chain is paddq combining the carry and next (shifted) part,
C plus psrlq to move the new carry down. That, and just 4 mmx instructions
C in total, makes 4 c/l the target speed, which is almost achieved for
C separate src/dst but when src==dst the write combining anomalies slow it
C down.
defframe(PARAM_SIZE, 16)
defframe(PARAM_YP, 12)
defframe(PARAM_XP, 8)
defframe(PARAM_WP, 4)
dnl re-use parameter space
define(SAVE_EBX,`PARAM_XP')
define(SAVE_ESI,`PARAM_YP')
TEXT
ALIGN(8)
PROLOGUE(mpn_rsh1add_n)
deflit(`FRAME',0)
movl PARAM_XP, %edx
movl %ebx, SAVE_EBX
movl PARAM_YP, %ebx
movl %esi, SAVE_ESI
movl PARAM_WP, %esi
movd (%edx), %mm0 C xp[0]
movd (%ebx), %mm1 C yp[0]
movl PARAM_SIZE, %ecx
movl (%edx), %eax C xp[0]
addl (%ebx), %eax C xp[0]+yp[0]
paddq %mm1, %mm0 C xp[0]+yp[0]
leal (%esi,%ecx,4), %esi C wp end
negl %ecx C -size
psrlq $1, %mm0 C (xp[0]+yp[0])/2
and $1, %eax C return value, rsh1 bit of xp[0]+yp[0]
addl $1, %ecx C -(size-1)
jz L(done)
L(top):
C eax return value
C ebx yp end
C ecx counter, limbs, -(size-1) to -1 inclusive
C edx xp end
C esi wp end
C mm0 carry (32 bits)
movd 4(%edx), %mm1 C xp[i+1]
movd 4(%ebx), %mm2 C yp[i+1]
leal 4(%edx), %edx
leal 4(%ebx), %ebx
paddq %mm2, %mm1 C xp[i+1]+yp[i+1]
psllq $31, %mm1 C low bit at 31, further 32 above
paddq %mm1, %mm0 C 31 and carry from prev add
movd %mm0, -4(%esi,%ecx,4) C low ready to store dst[i]
psrlq $32, %mm0 C high becomes new carry
addl $1, %ecx
jnz L(top)
L(done):
movd %mm0, -4(%esi) C dst[size-1]
movl SAVE_EBX, %ebx
movl SAVE_ESI, %esi
emms
ret
EPILOGUE()
|