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
|
/*
* Copyright (c) 2023 The FreeBSD Foundation
*
* This software was developed by Robert Clausecker <fuz@FreeBSD.org>
* under sponsorship from the FreeBSD Foundation.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ''AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE
*/
#include <machine/asm.h>
#include "amd64_archlevel.h"
#define ALIGN_TEXT .p2align 4, 0x90
.weak stpncpy
.set stpncpy, __stpncpy
ARCHFUNCS(__stpncpy)
ARCHFUNC(__stpncpy, scalar)
ARCHFUNC(__stpncpy, baseline)
ENDARCHFUNCS(__stpncpy)
ARCHENTRY(__stpncpy, scalar)
push %rbp # establish stack frame
mov %rsp, %rbp
push %rdx
push %rdi
push %rsi
push %rax # dummy push for alignment
mov %rsi, %rdi
xor %esi, %esi
call CNAME(__memchr) # memchr(src, '\0', len)
pop %rcx # dummy pop
pop %rsi
mov -16(%rbp), %rdi
test %rax, %rax # NUL found?
jz .Lfullcopy
mov %rax, %rdx
sub %rsi, %rdx # copy until the NUL byte
add %rdx, -16(%rbp) # advance destination by string length
sub %rdx, -8(%rbp) # and shorten buffer size by string length
call CNAME(memcpy)
pop %rdi
pop %rdx
xor %esi, %esi
pop %rbp
jmp CNAME(memset) # clear remaining buffer
.Lfullcopy:
mov -8(%rbp), %rdx
call CNAME(memcpy) # copy whole string
add -8(%rbp), %rax # point to dest[n]
leave
ret
ARCHEND(__stpncpy, scalar)
/*
* this mask allows us to generate masks of 16-n 0xff bytes
* followed by n 0x00 bytes by loading from .Lmask+n.
*/
.section .rodata
.Lmask: .quad 0xffffffffffffffff
.quad 0xffffffffffffffff
.quad 0x0000000000000000
.quad 0x0000000000000000
/* stpncpy(char *restrict rdi, const char *rsi, size_t rdx) */
ARCHENTRY(__stpncpy, baseline)
#define bounce (-3*16-8) /* location of on-stack bounce buffer */
test %rdx, %rdx # no bytes to copy?
jz .L0
mov %esi, %ecx
and $~0xf, %rsi # align source to 16 bytes
movdqa (%rsi), %xmm0 # load head
and $0xf, %ecx # offset from alignment
mov $-1, %r9d
lea -32(%rcx), %rax # set up overflow-proof comparison rdx+rcx<=32
shl %cl, %r9d # mask of bytes belonging to the string
sub %rcx, %rdi # adjust RDI to correspond to RSI
pxor %xmm1, %xmm1
movdqa %xmm0, bounce(%rsp) # stash copy of head on the stack
pcmpeqb %xmm1, %xmm0
pmovmskb %xmm0, %r8d
lea (%rdx, %rcx, 1), %r10 # buffer length from alignment boundary
add %rdx, %rax # less than 2 chunks (32 bytes) to play with?
jnc .Lrunt # if yes, use special runt processing
movdqu %xmm1, -16(%rdi, %r10, 1) # clear final bytes of destination
and %r9d, %r8d # end of string within head?
jnz .Lheadnul
movdqu (%rsi, %rcx, 1), %xmm2 # load head from source buffer
movdqu %xmm2, (%rdi, %rcx, 1) # an deposit
add $16, %rsi
add $16, %rdi
sub $32, %r10
/* main loop unrolled twice */
ALIGN_TEXT
0: movdqa (%rsi), %xmm0
pxor %xmm1, %xmm1
pcmpeqb %xmm0, %xmm1 # NUL byte encountered?
pmovmskb %xmm1, %r8d
test %r8d, %r8d
jnz 3f
movdqu %xmm0, (%rdi)
cmp $16, %r10 # more than a full chunk left?
jbe 1f
movdqa 16(%rsi), %xmm0
add $32, %rdi # advance pointers to next chunk
add $32, %rsi
pxor %xmm1, %xmm1
pcmpeqb %xmm0, %xmm1 # NUL byte encountered?
pmovmskb %xmm1, %r8d
test %r8d, %r8d
jnz 2f
movdqu %xmm0, -16(%rdi)
sub $32, %r10 # more than another full chunk left?
ja 0b
sub $16, %rdi # undo second advancement
sub $16, %rsi
add $16, %r10d # restore number of remaining bytes
/* 1--16 bytes left but string has not ended yet */
1: pxor %xmm1, %xmm1
pcmpeqb 16(%rsi), %xmm1 # NUL byte in source tail?
pmovmskb %xmm1, %r8d
bts %r10d, %r8d # treat end of buffer as NUL
tzcnt %r8d, %r8d # where is the NUL byte?
movdqu (%rsi, %r8, 1), %xmm0 # load source tail before NUL
lea 16(%rdi, %r8, 1), %rax # point return value to NUL byte
# or end of buffer
movdqu %xmm0, (%rdi, %r8, 1) # store tail into the buffer
ret
2: sub $16, %rdi # undo second advancement
sub $16, %rsi
sub $16, %r10
/* string has ended and buffer has not */
3: tzcnt %r8d, %r8d # where did the string end?
lea .Lmask+16(%rip), %rcx
lea (%rdi, %r8, 1), %rax # where the NUL byte will be
neg %r8
movdqu (%rcx, %r8, 1), %xmm1 # mask with FF where the string is,
# 00 where it is not
pand %xmm1, %xmm0 # mask out bytes after the string
movdqu %xmm0, (%rdi) # store masked current chunk
pxor %xmm1, %xmm1
sub $16, %r10 # another full chunk left?
jbe 1f
/* clear remaining destination buffer (tail has been cleared earlier) */
ALIGN_TEXT
0: movdqu %xmm1, 16(%rdi)
cmp $16, %r10
jbe 1f
movdqu %xmm1, 32(%rdi)
add $32, %rdi
sub $32, %r10
ja 0b
1: ret
/* at least two chunks to play with and NUL while processing head */
.Lheadnul:
movdqu bounce(%rsp, %rcx, 1), %xmm0 # load start of source from stack
tzcnt %r8d, %r8d # find location of NUL byte
movdqu %xmm0, (%rdi, %rcx, 1) # deposit head in the destination
movdqu %xmm1, (%rdi, %r8, 1) # clear out following bytes
movdqu %xmm1, 16(%rdi) # clear out second chunk
lea (%rdi, %r8, 1), %rax # make RAX point to the NUL byte
add $32, %rdi # advance past first two chunks
sub $32+16, %r10 # advance past first three chunks
jbe 1f # did we pass the end of the buffer?
/* clear remaining destination buffer (tail has been cleared earlier) */
ALIGN_TEXT
0: movdqu %xmm1, (%rdi) # clear out buffer chunk
cmp $16, %r10
jbe 1f
movdqu %xmm1, 16(%rdi)
add $32, %rdi
sub $32, %r10
ja 0b
1: ret
/* 1--32 bytes to copy, bounce through the stack */
.Lrunt: movdqa %xmm1, bounce+16(%rsp) # clear out rest of on-stack copy
bts %r10d, %r8d # treat end of buffer as end of string
and %r9w, %r8w # end of string within first buffer?
jnz 0f # if yes, do not inspect second buffer
movdqa 16(%rsi), %xmm0 # load second chunk of input
movdqa %xmm0, bounce+16(%rsp) # stash copy on stack
pcmpeqb %xmm1, %xmm0 # NUL in second chunk?
pmovmskb %xmm0, %r9d
shl $16, %r9d
or %r9d, %r8d # merge found NUL bytes into NUL mask
/* end of string after one buffer */
0: tzcnt %r8d, %r8d # location of last char in string
movdqu %xmm1, bounce(%rsp, %r8, 1) # clear bytes behind string
lea bounce(%rsp, %rcx, 1), %rsi # start of string copy on stack
lea (%rdi, %r8, 1), %rax # return pointer to NUL byte
cmp $16, %edx # at least 16 bytes to transfer?
jae .L1631
mov (%rsi), %r8 # load string head
cmp $8, %edx # at least 8 bytes to transfer?
jae .L0815
cmp $4, %edx # at least 4 bytes to transfer?
jae .L0407
movzwl -2(%rsi, %rdx, 1), %esi # load last two bytes of string
mov %r8b, (%rdi, %rcx, 1) # store first byte
cmp $2, %edx # at least 2 bytes to transfer?
jb .L1
mov %si, -2(%rdi, %r10, 1) # store last two bytes of string
.L1: ret
.L1631: movdqu (%rsi), %xmm0 # load first 16 bytes of string
movdqu -16(%rsi, %rdx, 1), %xmm1 # load last 16 bytes of string
movdqu %xmm0, (%rdi, %rcx, 1)
movdqu %xmm1, -16(%rdi, %r10, 1)
ret
.L0815: mov -8(%rsi, %rdx, 1), %rdx # load last 8 bytes of string
mov %r8, (%rdi, %rcx, 1)
mov %rdx, -8(%rdi, %r10, 1)
ret
.L0407: mov -4(%rsi, %rdx, 1), %edx # load last four bytes of string
mov %r8d, (%rdi, %rcx, 1)
mov %edx, -4(%rdi, %r10, 1)
ret
/* length 0 buffer: just return dest */
.L0: mov %rdi, %rax
ret
ARCHEND(__stpncpy, baseline)
.section .note.GNU-stack,"",%progbits
|