about summary refs log tree commit diff
path: root/sysdeps/x86_64/multiarch/strrchr-sse2.S
blob: 9bc4fc3dfea3b6244a78220140f55e3d9b05cbec (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
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
/* strrchr optimized with SSE2.
   Copyright (C) 2017-2023 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/>.  */

#include <isa-level.h>

/* ISA level >= 2 because there are no {wcs|str}rchr-sse4
   implementations.  */
#if ISA_SHOULD_BUILD (2)

# include <sysdep.h>

# ifndef STRRCHR
#  define STRRCHR __strrchr_sse2
# endif

# ifdef USE_AS_WCSRCHR
#  define PCMPEQ	pcmpeqd
#  define CHAR_SIZE	4
#  define PMINU	pminud
# else
#  define PCMPEQ	pcmpeqb
#  define CHAR_SIZE	1
#  define PMINU	pminub
# endif

# define PAGE_SIZE	4096
# define VEC_SIZE	16

	.text
ENTRY(STRRCHR)
	movd	%esi, %xmm0
	movq	%rdi, %rax
	andl	$(PAGE_SIZE - 1), %eax
# ifndef USE_AS_WCSRCHR
	punpcklbw %xmm0, %xmm0
	punpcklwd %xmm0, %xmm0
# endif
	pshufd	$0, %xmm0, %xmm0
	cmpl	$(PAGE_SIZE - VEC_SIZE), %eax
	ja	L(cross_page)

L(cross_page_continue):
	movups	(%rdi), %xmm1
	pxor	%xmm2, %xmm2
	PCMPEQ	%xmm1, %xmm2
	pmovmskb %xmm2, %ecx
	testl	%ecx, %ecx
	jz	L(aligned_more)

	PCMPEQ	%xmm0, %xmm1
	pmovmskb %xmm1, %eax
	leal	-1(%rcx), %edx
	xorl	%edx, %ecx
	andl	%ecx, %eax
	jz	L(ret0)
	bsrl	%eax, %eax
	addq	%rdi, %rax
	/* We are off by 3 for wcsrchr if search CHAR is non-zero. If
	   search CHAR is zero we are correct. Either way `andq
	   -CHAR_SIZE, %rax` gets the correct result.  */
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
L(ret0):
	ret

	/* Returns for first vec x1/x2 have hard coded backward search
	   path for earlier matches.  */
	.p2align 4
L(first_vec_x0_test):
	PCMPEQ	%xmm0, %xmm1
	pmovmskb %xmm1, %eax
	testl	%eax, %eax
	jz	L(ret0)
	bsrl	%eax, %eax
	addq	%r8, %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	.p2align 4
L(first_vec_x1):
	PCMPEQ	%xmm0, %xmm2
	pmovmskb %xmm2, %eax
	leal	-1(%rcx), %edx
	xorl	%edx, %ecx
	andl	%ecx, %eax
	jz	L(first_vec_x0_test)
	bsrl	%eax, %eax
	leaq	(VEC_SIZE)(%rdi, %rax), %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	.p2align 4
L(first_vec_x1_test):
	PCMPEQ	%xmm0, %xmm2
	pmovmskb %xmm2, %eax
	testl	%eax, %eax
	jz	L(first_vec_x0_test)
	bsrl	%eax, %eax
	leaq	(VEC_SIZE)(%rdi, %rax), %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	.p2align 4
L(first_vec_x2):
	PCMPEQ	%xmm0, %xmm3
	pmovmskb %xmm3, %eax
	leal	-1(%rcx), %edx
	xorl	%edx, %ecx
	andl	%ecx, %eax
	jz	L(first_vec_x1_test)
	bsrl	%eax, %eax
	leaq	(VEC_SIZE * 2)(%rdi, %rax), %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	.p2align 4
L(aligned_more):
	/* Save original pointer if match was in VEC 0.  */
	movq	%rdi, %r8
	andq	$-VEC_SIZE, %rdi

	movaps	VEC_SIZE(%rdi), %xmm2
	pxor	%xmm3, %xmm3
	PCMPEQ	%xmm2, %xmm3
	pmovmskb %xmm3, %ecx
	testl	%ecx, %ecx
	jnz	L(first_vec_x1)

	movaps	(VEC_SIZE * 2)(%rdi), %xmm3
	pxor	%xmm4, %xmm4
	PCMPEQ	%xmm3, %xmm4
	pmovmskb %xmm4, %ecx
	testl	%ecx, %ecx
	jnz	L(first_vec_x2)

	addq	$VEC_SIZE, %rdi
	/* Save pointer again before realigning.  */
	movq	%rdi, %rsi
	andq	$-(VEC_SIZE * 2), %rdi
	.p2align 4
L(first_loop):
	/* Do 2x VEC at a time.  */
	movaps	(VEC_SIZE * 2)(%rdi), %xmm4
	movaps	(VEC_SIZE * 3)(%rdi), %xmm5
	/* Since SSE2 no pminud so wcsrchr needs seperate logic for
	   detecting zero. Note if this is found to be a bottleneck it
	   may be worth adding an SSE4.1 wcsrchr implementation.  */
# ifdef USE_AS_WCSRCHR
	movaps	%xmm5, %xmm6
	pxor	%xmm8, %xmm8

	PCMPEQ	%xmm8, %xmm5
	PCMPEQ	%xmm4, %xmm8
	por	%xmm5, %xmm8
# else
	movaps	%xmm5, %xmm6
	PMINU	%xmm4, %xmm5
# endif

	movaps	%xmm4, %xmm9
	PCMPEQ	%xmm0, %xmm4
	PCMPEQ	%xmm0, %xmm6
	movaps	%xmm6, %xmm7
	por	%xmm4, %xmm6
# ifndef USE_AS_WCSRCHR
	pxor	%xmm8, %xmm8
	PCMPEQ	%xmm5, %xmm8
# endif
	pmovmskb %xmm8, %ecx
	pmovmskb %xmm6, %eax

	addq	$(VEC_SIZE * 2), %rdi
	/* Use `addl` 1) so we can undo it with `subl` and 2) it can
	   macro-fuse with `jz`.  */
	addl	%ecx, %eax
	jz	L(first_loop)

	/* Check if there is zero match.  */
	testl	%ecx, %ecx
	jz	L(second_loop_match)

	/* Check if there was a match in last iteration.  */
	subl	%ecx, %eax
	jnz	L(new_match)

L(first_loop_old_match):
	PCMPEQ	%xmm0, %xmm2
	PCMPEQ	%xmm0, %xmm3
	pmovmskb %xmm2, %ecx
	pmovmskb %xmm3, %eax
	addl	%eax, %ecx
	jz	L(first_vec_x0_test)
	/* NB: We could move this shift to before the branch and save a
	   bit of code size / performance on the fall through. The
	   branch leads to the null case which generally seems hotter
	   than char in first 3x VEC.  */
	sall	$16, %eax
	orl	%ecx, %eax

	bsrl	%eax, %eax
	addq	%rsi, %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	.p2align 4
L(new_match):
	pxor	%xmm6, %xmm6
	PCMPEQ	%xmm9, %xmm6
	pmovmskb %xmm6, %eax
	sall	$16, %ecx
	orl	%eax, %ecx

	/* We can't reuse either of the old comparisons as since we mask
	   of zeros after first zero (instead of using the full
	   comparison) we can't gurantee no interference between match
	   after end of string and valid match.  */
	pmovmskb %xmm4, %eax
	pmovmskb %xmm7, %edx
	sall	$16, %edx
	orl	%edx, %eax

	leal	-1(%ecx), %edx
	xorl	%edx, %ecx
	andl	%ecx, %eax
	jz	L(first_loop_old_match)
	bsrl	%eax, %eax
	addq	%rdi, %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	/* Save minimum state for getting most recent match. We can
	   throw out all previous work.  */
	.p2align 4
L(second_loop_match):
	movq	%rdi, %rsi
	movaps	%xmm4, %xmm2
	movaps	%xmm7, %xmm3

	.p2align 4
L(second_loop):
	movaps	(VEC_SIZE * 2)(%rdi), %xmm4
	movaps	(VEC_SIZE * 3)(%rdi), %xmm5
	/* Since SSE2 no pminud so wcsrchr needs seperate logic for
	   detecting zero. Note if this is found to be a bottleneck it
	   may be worth adding an SSE4.1 wcsrchr implementation.  */
# ifdef USE_AS_WCSRCHR
	movaps	%xmm5, %xmm6
	pxor	%xmm8, %xmm8

	PCMPEQ	%xmm8, %xmm5
	PCMPEQ	%xmm4, %xmm8
	por	%xmm5, %xmm8
# else
	movaps	%xmm5, %xmm6
	PMINU	%xmm4, %xmm5
# endif

	movaps	%xmm4, %xmm9
	PCMPEQ	%xmm0, %xmm4
	PCMPEQ	%xmm0, %xmm6
	movaps	%xmm6, %xmm7
	por	%xmm4, %xmm6
# ifndef USE_AS_WCSRCHR
	pxor	%xmm8, %xmm8
	PCMPEQ	%xmm5, %xmm8
# endif

	pmovmskb %xmm8, %ecx
	pmovmskb %xmm6, %eax

	addq	$(VEC_SIZE * 2), %rdi
	/* Either null term or new occurence of CHAR.  */
	addl	%ecx, %eax
	jz	L(second_loop)

	/* No null term so much be new occurence of CHAR.  */
	testl	%ecx, %ecx
	jz	L(second_loop_match)


	subl	%ecx, %eax
	jnz	L(second_loop_new_match)

L(second_loop_old_match):
	pmovmskb %xmm2, %ecx
	pmovmskb %xmm3, %eax
	sall	$16, %eax
	orl	%ecx, %eax
	bsrl	%eax, %eax
	addq	%rsi, %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	.p2align 4
L(second_loop_new_match):
	pxor	%xmm6, %xmm6
	PCMPEQ	%xmm9, %xmm6
	pmovmskb %xmm6, %eax
	sall	$16, %ecx
	orl	%eax, %ecx

	/* We can't reuse either of the old comparisons as since we mask
	   of zeros after first zero (instead of using the full
	   comparison) we can't gurantee no interference between match
	   after end of string and valid match.  */
	pmovmskb %xmm4, %eax
	pmovmskb %xmm7, %edx
	sall	$16, %edx
	orl	%edx, %eax

	leal	-1(%ecx), %edx
	xorl	%edx, %ecx
	andl	%ecx, %eax
	jz	L(second_loop_old_match)
	bsrl	%eax, %eax
	addq	%rdi, %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
	ret

	.p2align 4,, 4
L(cross_page):
	movq	%rdi, %rsi
	andq	$-VEC_SIZE, %rsi
	movaps	(%rsi), %xmm1
	pxor	%xmm2, %xmm2
	PCMPEQ	%xmm1, %xmm2
	pmovmskb %xmm2, %edx
	movl	%edi, %ecx
	andl	$(VEC_SIZE - 1), %ecx
	sarl	%cl, %edx
	jz	L(cross_page_continue)
	PCMPEQ	%xmm0, %xmm1
	pmovmskb %xmm1, %eax
	sarl	%cl, %eax
	leal	-1(%rdx), %ecx
	xorl	%edx, %ecx
	andl	%ecx, %eax
	jz	L(ret1)
	bsrl	%eax, %eax
	addq	%rdi, %rax
# ifdef USE_AS_WCSRCHR
	andq	$-CHAR_SIZE, %rax
# endif
L(ret1):
	ret
END(STRRCHR)
#endif