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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
|
dnl AMD64 mpn_com optimised for CPUs with fast SSE copying and SSSE3.
dnl Copyright 2012, 2013 Free Software Foundation, Inc.
dnl Contributed to the GNU project by Torbjorn Granlund.
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 cycles/limb cycles/limb good
C aligned unaligned best seen for cpu?
C AMD K8,K9 2.0 illop 1.0/1.0 N
C AMD K10 0.85 illop Y/N
C AMD bd1 1.39 ? 1.45 Y/N
C AMD bobcat 1.97 ? 8.17 1.5/1.5 N
C Intel P4 2.26 illop Y/N
C Intel core2 0.52 0.82 opt/0.74 Y
C Intel NHM 0.52 0.65 opt/opt Y
C Intel SBR 0.51 0.55 opt/0.51 Y
C Intel atom 1.16 1.70 opt/opt Y
C VIA nano 1.09 1.10 opt/opt Y
C We use only 16-byte operations, except for unaligned top-most and bottom-most
C limbs. We use the SSSE3 palignr instruction when rp - up = 8 (mod 16). That
C instruction is better adapted to mpn_copyd's needs, we need to contort the
C code to use it here.
C
C For operands of < COM_SSE_THRESHOLD limbs, we use a plain 64-bit loop, taken
C from the x86_64 default code.
C INPUT PARAMETERS
define(`rp', `%rdi')
define(`up', `%rsi')
define(`n', `%rdx')
C There are three instructions for loading an aligned 128-bit quantity. We use
C movaps, since it has the shortest coding.
define(`movdqa', ``movaps'')
ifdef(`COM_SSE_THRESHOLD',`',`define(`COM_SSE_THRESHOLD', 7)')
ASM_START()
TEXT
ALIGN(64)
PROLOGUE(mpn_com)
FUNC_ENTRY(3)
cmp $COM_SSE_THRESHOLD, n
jbe L(bc)
pcmpeqb %xmm7, %xmm7 C set to 111...111
test $8, R8(rp) C is rp 16-byte aligned?
jz L(rp_aligned) C jump if rp aligned
mov (up), %r8
lea 8(up), up
not %r8
mov %r8, (rp)
lea 8(rp), rp
dec n
L(rp_aligned):
test $8, R8(up)
jnz L(uent)
ifelse(eval(COM_SSE_THRESHOLD >= 8),1,
` sub $8, n',
` jmp L(am)')
ALIGN(16)
L(atop):movdqa 0(up), %xmm0
movdqa 16(up), %xmm1
movdqa 32(up), %xmm2
movdqa 48(up), %xmm3
lea 64(up), up
pxor %xmm7, %xmm0
pxor %xmm7, %xmm1
pxor %xmm7, %xmm2
pxor %xmm7, %xmm3
movdqa %xmm0, (rp)
movdqa %xmm1, 16(rp)
movdqa %xmm2, 32(rp)
movdqa %xmm3, 48(rp)
lea 64(rp), rp
L(am): sub $8, n
jnc L(atop)
test $4, R8(n)
jz 1f
movdqa (up), %xmm0
movdqa 16(up), %xmm1
lea 32(up), up
pxor %xmm7, %xmm0
pxor %xmm7, %xmm1
movdqa %xmm0, (rp)
movdqa %xmm1, 16(rp)
lea 32(rp), rp
1: test $2, R8(n)
jz 1f
movdqa (up), %xmm0
lea 16(up), up
pxor %xmm7, %xmm0
movdqa %xmm0, (rp)
lea 16(rp), rp
1: test $1, R8(n)
jz 1f
mov (up), %r8
not %r8
mov %r8, (rp)
1: FUNC_EXIT()
ret
L(uent):
C Code handling up - rp = 8 (mod 16)
C FIXME: The code below only handles overlap if it is close to complete, or
C quite separate: up-rp < 5 or up-up > 15 limbs
lea -40(up), %rax C 40 = 5 * GMP_LIMB_BYTES
sub rp, %rax
cmp $80, %rax C 80 = (15-5) * GMP_LIMB_BYTES
jbe L(bc) C deflect to plain loop
sub $16, n
jc L(uend)
movdqa 120(up), %xmm3
sub $16, n
jmp L(um)
ALIGN(16)
L(utop):movdqa 120(up), %xmm3
pxor %xmm7, %xmm0
movdqa %xmm0, -128(rp)
sub $16, n
L(um): movdqa 104(up), %xmm2
palignr($8, %xmm2, %xmm3)
movdqa 88(up), %xmm1
pxor %xmm7, %xmm3
movdqa %xmm3, 112(rp)
palignr($8, %xmm1, %xmm2)
movdqa 72(up), %xmm0
pxor %xmm7, %xmm2
movdqa %xmm2, 96(rp)
palignr($8, %xmm0, %xmm1)
movdqa 56(up), %xmm3
pxor %xmm7, %xmm1
movdqa %xmm1, 80(rp)
palignr($8, %xmm3, %xmm0)
movdqa 40(up), %xmm2
pxor %xmm7, %xmm0
movdqa %xmm0, 64(rp)
palignr($8, %xmm2, %xmm3)
movdqa 24(up), %xmm1
pxor %xmm7, %xmm3
movdqa %xmm3, 48(rp)
palignr($8, %xmm1, %xmm2)
movdqa 8(up), %xmm0
pxor %xmm7, %xmm2
movdqa %xmm2, 32(rp)
palignr($8, %xmm0, %xmm1)
movdqa -8(up), %xmm3
pxor %xmm7, %xmm1
movdqa %xmm1, 16(rp)
palignr($8, %xmm3, %xmm0)
lea 128(up), up
lea 128(rp), rp
jnc L(utop)
pxor %xmm7, %xmm0
movdqa %xmm0, -128(rp)
L(uend):test $8, R8(n)
jz 1f
movdqa 56(up), %xmm3
movdqa 40(up), %xmm2
palignr($8, %xmm2, %xmm3)
movdqa 24(up), %xmm1
pxor %xmm7, %xmm3
movdqa %xmm3, 48(rp)
palignr($8, %xmm1, %xmm2)
movdqa 8(up), %xmm0
pxor %xmm7, %xmm2
movdqa %xmm2, 32(rp)
palignr($8, %xmm0, %xmm1)
movdqa -8(up), %xmm3
pxor %xmm7, %xmm1
movdqa %xmm1, 16(rp)
palignr($8, %xmm3, %xmm0)
lea 64(up), up
pxor %xmm7, %xmm0
movdqa %xmm0, (rp)
lea 64(rp), rp
1: test $4, R8(n)
jz 1f
movdqa 24(up), %xmm1
movdqa 8(up), %xmm0
palignr($8, %xmm0, %xmm1)
movdqa -8(up), %xmm3
pxor %xmm7, %xmm1
movdqa %xmm1, 16(rp)
palignr($8, %xmm3, %xmm0)
lea 32(up), up
pxor %xmm7, %xmm0
movdqa %xmm0, (rp)
lea 32(rp), rp
1: test $2, R8(n)
jz 1f
movdqa 8(up), %xmm0
movdqa -8(up), %xmm3
palignr($8, %xmm3, %xmm0)
lea 16(up), up
pxor %xmm7, %xmm0
movdqa %xmm0, (rp)
lea 16(rp), rp
1: test $1, R8(n)
jz 1f
mov (up), %r8
not %r8
mov %r8, (rp)
1: FUNC_EXIT()
ret
C Basecase code. Needed for good small operands speed, not for
C correctness as the above code is currently written.
L(bc): lea -8(rp), rp
sub $4, R32(n)
jc L(end)
ifelse(eval(1 || COM_SSE_THRESHOLD >= 8),1,
` ALIGN(16)')
L(top): mov (up), %r8
mov 8(up), %r9
lea 32(rp), rp
mov 16(up), %r10
mov 24(up), %r11
lea 32(up), up
not %r8
not %r9
not %r10
not %r11
mov %r8, -24(rp)
mov %r9, -16(rp)
ifelse(eval(1 || COM_SSE_THRESHOLD >= 8),1,
` sub $4, R32(n)')
mov %r10, -8(rp)
mov %r11, (rp)
ifelse(eval(1 || COM_SSE_THRESHOLD >= 8),1,
` jnc L(top)')
L(end): test $1, R8(n)
jz 1f
mov (up), %r8
not %r8
mov %r8, 8(rp)
lea 8(rp), rp
lea 8(up), up
1: test $2, R8(n)
jz 1f
mov (up), %r8
mov 8(up), %r9
not %r8
not %r9
mov %r8, 8(rp)
mov %r9, 16(rp)
1: FUNC_EXIT()
ret
EPILOGUE()
|
