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
|
dnl Intel Pentium-4 mpn_divexact_1 -- mpn by limb exact division.
dnl Rearranged from mpn/x86/pentium4/sse2/dive_1.asm by Marco Bodrato.
dnl Copyright 2001, 2002, 2007, 2011 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 P4: 19.0 cycles/limb
C Pairs of movd's are used to avoid unaligned loads. Despite the loads not
C being on the dependent chain and there being plenty of cycles available,
C using an unaligned movq on every second iteration measured about 23 c/l.
C
defframe(PARAM_SHIFT, 24)
defframe(PARAM_INVERSE,20)
defframe(PARAM_DIVISOR,16)
defframe(PARAM_SIZE, 12)
defframe(PARAM_SRC, 8)
defframe(PARAM_DST, 4)
TEXT
C mp_limb_t
C mpn_pi1_bdiv_q_1 (mp_ptr dst, mp_srcptr src, mp_size_t size, mp_limb_t divisor,
C mp_limb_t inverse, int shift)
ALIGN(32)
PROLOGUE(mpn_pi1_bdiv_q_1)
deflit(`FRAME',0)
movl PARAM_SIZE, %edx
movl PARAM_SRC, %eax
movl PARAM_DIVISOR, %ecx
movd %ecx, %mm6
movl PARAM_SHIFT, %ecx
movd %ecx, %mm7 C shift
C
movl PARAM_INVERSE, %ecx
movd %ecx, %mm5 C inv
movl PARAM_DST, %ecx
pxor %mm1, %mm1 C initial carry limb
pxor %mm0, %mm0 C initial carry bit
subl $1, %edx
jz L(done)
pcmpeqd %mm4, %mm4
psrlq $32, %mm4 C 0x00000000FFFFFFFF
C The dependent chain here is as follows.
C
C latency
C psubq s = (src-cbit) - climb 2
C pmuludq q = s*inverse 8
C pmuludq prod = q*divisor 8
C psrlq climb = high(prod) 2
C --
C 20
C
C Yet the loop measures 19.0 c/l, so obviously there's something gained
C there over a straight reading of the chip documentation.
L(top):
C eax src, incrementing
C ebx
C ecx dst, incrementing
C edx counter, size-1 iterations
C
C mm0 carry bit
C mm1 carry limb
C mm4 0x00000000FFFFFFFF
C mm5 inverse
C mm6 divisor
C mm7 shift
movd (%eax), %mm2
movd 4(%eax), %mm3
addl $4, %eax
punpckldq %mm3, %mm2
psrlq %mm7, %mm2
pand %mm4, %mm2 C src
psubq %mm0, %mm2 C src - cbit
psubq %mm1, %mm2 C src - cbit - climb
movq %mm2, %mm0
psrlq $63, %mm0 C new cbit
pmuludq %mm5, %mm2 C s*inverse
movd %mm2, (%ecx) C q
addl $4, %ecx
movq %mm6, %mm1
pmuludq %mm2, %mm1 C q*divisor
psrlq $32, %mm1 C new climb
L(entry):
subl $1, %edx
jnz L(top)
L(done):
movd (%eax), %mm2
psrlq %mm7, %mm2 C src
psubq %mm0, %mm2 C src - cbit
psubq %mm1, %mm2 C src - cbit - climb
pmuludq %mm5, %mm2 C s*inverse
movd %mm2, (%ecx) C q
emms
ret
EPILOGUE()
ALIGN(16)
C mp_limb_t mpn_bdiv_q_1 (mp_ptr dst, mp_srcptr src, mp_size_t size,
C mp_limb_t divisor);
C
PROLOGUE(mpn_bdiv_q_1)
deflit(`FRAME',0)
movl PARAM_SIZE, %edx
movl PARAM_DIVISOR, %ecx
C eax src
C ebx
C ecx divisor
C edx size-1
movl %ecx, %eax
bsfl %ecx, %ecx C trailing twos
shrl %cl, %eax C d = divisor without twos
movd %eax, %mm6
movd %ecx, %mm7 C shift
shrl %eax C d/2
andl $127, %eax C d/2, 7 bits
ifdef(`PIC',`
LEA( binvert_limb_table, %ecx)
movzbl (%eax,%ecx), %eax C inv 8 bits
',`
movzbl binvert_limb_table(%eax), %eax C inv 8 bits
')
C
movd %eax, %mm5 C inv
movd %eax, %mm0 C inv
pmuludq %mm5, %mm5 C inv*inv
C
pmuludq %mm6, %mm5 C inv*inv*d
paddd %mm0, %mm0 C 2*inv
C
psubd %mm5, %mm0 C inv = 2*inv - inv*inv*d
pxor %mm5, %mm5
paddd %mm0, %mm5
pmuludq %mm0, %mm0 C inv*inv
pcmpeqd %mm4, %mm4
psrlq $32, %mm4 C 0x00000000FFFFFFFF
C
pmuludq %mm6, %mm0 C inv*inv*d
paddd %mm5, %mm5 C 2*inv
movl PARAM_SRC, %eax
movl PARAM_DST, %ecx
pxor %mm1, %mm1 C initial carry limb
C
psubd %mm0, %mm5 C inv = 2*inv - inv*inv*d
ASSERT(e,` C expect d*inv == 1 mod 2^GMP_LIMB_BITS
pushl %eax FRAME_pushl()
movq %mm6, %mm0
pmuludq %mm5, %mm0
movd %mm0, %eax
cmpl $1, %eax
popl %eax FRAME_popl()')
pxor %mm0, %mm0 C initial carry bit
jmp L(entry)
EPILOGUE()
ASM_END()
|