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
|
dnl Alpha mpn_com -- mpn one's complement.
dnl Copyright 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 cycles/limb
C EV4: 4.75
C EV5: 2.0
C EV6: 1.5
C mp_limb_t mpn_com (mp_ptr dst, mp_srcptr src, mp_size_t size);
C
C For ev5 the main loop is 7 cycles plus 1 taken branch bubble, for a total
C 2.0 c/l. In general, a pattern like this unrolled to N limbs per loop
C will be 1.5+2/N c/l.
C
C 2 cycles of loop control are unavoidable, for pointer updates and the
C taken branch bubble, but also since ldq cannot issue two cycles after stq
C (and with a run of stqs that means neither of two cycles at the end of the
C loop.
C
C The fbeq is forced into the second cycle of the loop using unops, since
C the first time through it must wait for the cvtqt result. Once that
C result is ready (a 1 cycle stall) then both the branch and following loads
C can issue together.
C
C The main loop handles an odd count of limbs, being two limbs loaded before
C each size test, plus one pipelined around from the previous iteration (or
C setup in the entry sequence).
C
C An even number of limbs is handled by an explicit dst[0]=~src[0] in the
C entry sequence, and an increment of the pointers. For an odd size there's
C no increment and the first store in the loop (r24) is a repeat of dst[0].
C
C Note that the load for r24 after the possible pointer increment is done
C before the explicit store to dst[0], in case src==dst.
ASM_START()
FLOAT64(L(dat), 2.0)
ALIGN(16)
PROLOGUE(mpn_com,gp)
C r16 dst
C r17 src
C r18 size
lda r30, -16(r30) C temporary stack space
lda r7, -3(r18) C size - 3
ldq r20, 0(r17) C src[0]
srl r7, 1, r6 C (size-3)/2
stq r6, 8(r30) C (size-3)/2
and r7, 1, r5 C 1 if size even
LEA( r8, L(dat))
s8addq r5, r17, r17 C skip src[0] if even
ornot r31, r20, r20 C ~src[0]
unop
ldt f0, 8(r30) C (size-3)/2
ldq r24, 0(r17) C src[0 or 1]
stq r20, 0(r16) C dst[0]
s8addq r5, r16, r19 C skip dst[0] if even
ldt f1, 0(r8) C data 2.0
lda r30, 16(r30) C restore stack
unop
cvtqt f0, f0 C (size-3)/2 as float
ornot r31, r24, r24
blt r7, L(done_1) C if size<=2
unop
unop
C 16-byte alignment here
L(top):
C r17 src, incrementing
C r19 dst, incrementing
C r24 dst[i] result, ready to store
C f0 (size-3)/2, decrementing
C f1 2.0
ldq r20, 8(r17) C src[i+1]
ldq r21, 16(r17) C src[i+2]
unop
unop
fbeq f0, L(done_2)
unop
ldq r22, 24(r17) C src[i+3]
ldq r23, 32(r17) C src[i+4]
stq r24, 0(r19) C dst[i]
ornot r31, r20, r20
subt f0, f1, f0 C count -= 2
unop
stq r20, 8(r19) C dst[i+1]
ornot r31, r21, r21
unop
unop
stq r21, 16(r19) C dst[i+2]
ornot r31, r22, r22
stq r22, 24(r19) C dst[i+3]
ornot r31, r23, r24
lda r17, 32(r17) C src += 4
lda r19, 32(r19) C dst += 4
unop
fbge f0, L(top)
L(done_1):
C r19 &dst[size-1]
C r24 result for dst[size-1]
stq r24, 0(r19) C dst[size-1]
ret r31, (r26), 1
L(done_2):
C r19 &dst[size-3]
C r20 src[size-2]
C r21 src[size-1]
C r24 result for dst[size-3]
stq r24, 0(r19) C dst[size-3]
ornot r31, r20, r20
stq r20, 8(r19) C dst[size-2]
ornot r31, r21, r21
stq r21, 16(r19) C dst[size-1]
ret r31, (r26), 1
EPILOGUE()
ASM_END()
|
