about summary refs log tree commit diff
path: root/sysdeps/x86_64/multiarch/memchr-evex-base.S
blob: 6ebc9a66e812a6444106c3d173c093c4e3d0b917 (plain) (blame)
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
303
304
/* Placeholder function, not used by any processor at the moment.
   Copyright (C) 2022 Free Software Foundation, Inc.
   This file is part of the GNU C Library.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; if not, see
   <https://www.gnu.org/licenses/>.  */

/* UNUSED. Exists purely as reference implementation.  */

#include <isa-level.h>

#if ISA_SHOULD_BUILD (4)

# include <sysdep.h>

# ifdef USE_AS_WMEMCHR
#  define CHAR_SIZE	4
#  define VPBROADCAST   vpbroadcastd
#  define VPCMPEQ	vpcmpeqd
#  define VPCMPNE	vpcmpneqd
#  define VPMINU	vpminud
#  define VPTESTNM	vptestnmd
# else
#  define CHAR_SIZE	1
#  define VPBROADCAST   vpbroadcastb
#  define VPCMPEQ	vpcmpeqb
#  define VPCMPNE	vpcmpneqb
#  define VPMINU	vpminub
#  define VPTESTNM	vptestnmb
# endif

# define PAGE_SIZE	4096
# define CHAR_PER_VEC	(VEC_SIZE / CHAR_SIZE)

	.section SECTION(.text), "ax", @progbits
/* Aligning entry point to 64 byte, provides better performance for
   one vector length string.  */
ENTRY_P2ALIGN (MEMCHR, 6)
# ifndef USE_AS_RAWMEMCHR
	/* Check for zero length.  */
	test	%RDX_LP, %RDX_LP
	jz	L(zero)

#  ifdef __ILP32__
	/* Clear the upper 32 bits.  */
	movl	%edx, %edx
#  endif
# endif

	/* Broadcast CHAR to VMM(1).  */
	VPBROADCAST %esi, %VMM(1)
	movl	%edi, %eax
	andl	$(PAGE_SIZE - 1), %eax
	cmpl	$(PAGE_SIZE - VEC_SIZE), %eax
	ja	L(page_cross)

	/* Compare [w]char for null, mask bit will be set for match.  */
	VPCMPEQ	(%rdi), %VMM(1), %k0

	KMOV	%k0, %VRCX
# ifndef USE_AS_RAWMEMCHR
	mov	%rdx, %rsi
	/* Need to use bsfq here as upper 32 bit of rsi may zero out
	   for 'bsf %ecx, %esi', if %ecx is 0.  */
	bsfq	%rcx, %rsi
	cmp	$CHAR_PER_VEC, %rsi
	ja	L(align_more)
#  ifdef USE_AS_WMEMCHR
	leaq	(%rdi, %rsi, CHAR_SIZE), %rdi
#  else
	addq	%rsi, %rdi
#  endif
	xor	%eax, %eax
	cmp	%rsi, %rdx
	cmova	%rdi, %rax
# else
	bsf     %VRCX, %VRAX
	jz	L(align_more)
	add	%rdi, %rax
# endif
	ret

	.p2align 5,,5
L(page_cross):
	movl	%eax, %ecx
	andl	$(VEC_SIZE - 1), %ecx
# ifdef USE_AS_WMEMCHR
	shrl	$2, %ecx
# endif
	xorq	%rdi, %rax
	VPCMPEQ (PAGE_SIZE - VEC_SIZE)(%rax), %VMM(1), %k0
	KMOV    %k0, %VRSI
	shr	%cl, %VRSI
# ifndef USE_AS_RAWMEMCHR
	jnz	L(page_cross_end)
	movl	$CHAR_PER_VEC, %eax
	sub	%ecx, %eax
	cmp	%rax, %rdx
	ja	L(align_more)
# else
	jz	L(align_more)
# endif

L(page_cross_end):
# ifndef USE_AS_RAWMEMCHR
	bsf	%VRSI, %VRCX
	jz	L(zero)
	leaq	(%rdi, %rcx, CHAR_SIZE), %rdi
	xor	%eax, %eax
	cmp	%rcx, %rdx
	cmova	%rdi, %rax
# else
	bsf	%VRSI, %VRAX
	add	%rdi, %rax
# endif
	ret

# ifndef USE_AS_RAWMEMCHR
L(zero):
	xorl	%eax, %eax
	ret
# endif

L(ret_vec_x2):
	subq	$-VEC_SIZE, %rdi
L(ret_vec_x1):
	bsf     %VRAX, %VRAX
# ifndef USE_AS_RAWMEMCHR
	cmp	%rax, %rdx
	jbe	L(zero)
# endif
# ifdef USE_AS_WMEMCHR
	leaq	(%rdi, %rax, CHAR_SIZE), %rax
# else
	add	%rdi, %rax
# endif
	ret

	.p2align 5,,5
L(align_more):
# ifndef USE_AS_RAWMEMCHR
	mov	%rdi, %rax
# endif
	subq	$-VEC_SIZE, %rdi
	/* Align rdi to VEC_SIZE.  */
	andq	$-VEC_SIZE, %rdi

# ifndef USE_AS_RAWMEMCHR
	subq	%rdi, %rax
#  ifdef USE_AS_WMEMCHR
	sar	$2, %rax
#  endif
	addq	%rax, %rdx
# endif

	/* Loop unroll 4 times for 4 vector loop.  */
	VPCMPEQ	(%rdi), %VMM(1), %k0

	KMOV	%k0, %VRAX
	test	%VRAX, %VRAX
	jnz	L(ret_vec_x1)

# ifndef USE_AS_RAWMEMCHR
	subq	$CHAR_PER_VEC, %rdx
	jbe	L(zero)
# endif

	VPCMPEQ	VEC_SIZE(%rdi), %VMM(1), %k0

	KMOV	%k0, %VRAX
	test	%VRAX, %VRAX
	jnz	L(ret_vec_x2)

# ifndef USE_AS_RAWMEMCHR
	subq	$CHAR_PER_VEC, %rdx
	jbe	L(zero)
# endif

	VPCMPEQ	(VEC_SIZE * 2)(%rdi), %VMM(1), %k0

	KMOV	%k0, %VRAX
	test	%VRAX, %VRAX
	jnz	L(ret_vec_x3)

# ifndef USE_AS_RAWMEMCHR
	subq	$CHAR_PER_VEC, %rdx
	jbe	L(zero)
# endif

	VPCMPEQ	(VEC_SIZE * 3)(%rdi), %VMM(1), %k0

	KMOV	%k0, %VRAX
	test	%VRAX, %VRAX
	jnz	L(ret_vec_x4)

# ifndef USE_AS_RAWMEMCHR
	subq	$CHAR_PER_VEC, %rdx
	jbe	L(zero)
	/* Save pointer to find alignment adjustment.  */
	movq	%rdi, %rax
# endif
	/* Align address to VEC_SIZE * 4 for loop.  */
	andq	$-(VEC_SIZE * 4), %rdi

	/* Add alignment difference to rdx.  */
# ifndef USE_AS_RAWMEMCHR
	subq	%rdi, %rax
#  ifdef USE_AS_WMEMCHR
	shr	$2, %VRAX
#  endif
	addq	%rax, %rdx
# endif

	/* 4 vector loop.  */
	.p2align 5,,11
L(loop):

	VPCMPNE	(VEC_SIZE * 4)(%rdi), %VMM(1), %k1
	vpxorq  (VEC_SIZE * 5)(%rdi), %VMM(1), %VMM(2)
	vpxorq  (VEC_SIZE * 6)(%rdi), %VMM(1), %VMM(3)
	VPCMPEQ (VEC_SIZE * 7)(%rdi), %VMM(1), %k3
	VPMINU  %VMM(2), %VMM(3), %VMM(3){%k1}{z}
	VPTESTNM %VMM(3), %VMM(3), %k2

	subq	$-(VEC_SIZE * 4), %rdi
	KORTEST	%k2, %k3
# ifdef USE_AS_RAWMEMCHR
	jz	L(loop)
# else
	jnz	L(loopend)
	subq	$(CHAR_PER_VEC * 4), %rdx
	ja	L(loop)
L(zero_2):
	xor	%eax, %eax
	ret
# endif

L(loopend):
	VPCMPEQ	(%rdi), %VMM(1), %k1
	KMOV	%k1, %VRAX
	test	%VRAX, %VRAX
	jnz	L(ret_vec_x1)

# ifndef USE_AS_RAWMEMCHR
	subq	$CHAR_PER_VEC, %rdx
	jbe	L(zero_2)
# endif

	VPCMPEQ	VEC_SIZE(%rdi), %VMM(1), %k1
	KMOV	%k1, %VRAX
	test	%VRAX, %VRAX
	jnz	L(ret_vec_x2)

# ifndef USE_AS_RAWMEMCHR
	subq	$CHAR_PER_VEC, %rdx
	jbe	L(zero_2)
# endif

	VPCMPEQ	(VEC_SIZE * 2)(%rdi), %VMM(1), %k1
	KMOV	%k1, %VRAX
	test	%VRAX, %VRAX
	jnz	L(ret_vec_x3)

# ifndef USE_AS_RAWMEMCHR
	subq	$CHAR_PER_VEC, %rdx
	jbe	L(zero_2)
# endif

	/* At this point null [w]char must be in the fourth vector so no
	   need to check.  */
	KMOV	%k3, %VRAX

L(ret_vec_x4):
	bsf	%VRAX, %VRAX
# ifndef USE_AS_RAWMEMCHR
	cmp	%rax, %rdx
	jbe	L(zero)
# endif
	leaq	(VEC_SIZE * 3)(%rdi, %rax, CHAR_SIZE), %rax
	ret

	.p2align 5,,5
L(ret_vec_x3):
	bsf	%VRAX, %VRAX
# ifndef USE_AS_RAWMEMCHR
	cmp	%rax, %rdx
	jbe	L(zero)
# endif
	leaq	(VEC_SIZE * 2)(%rdi, %rax, CHAR_SIZE), %rax
	ret

END (MEMCHR)
#endif