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
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
|
From 3ee56bbc56faa7b85a6513340db4a4fdd6ce709d Mon Sep 17 00:00:00 2001
From: dengjianbo <dengjianbo@loongson.cn>
Date: Mon, 28 Aug 2023 10:08:36 +0800
Subject: [PATCH 15/29] LoongArch: Add ifunc support for memchr{aligned, lsx,
lasx}
According to glibc memchr microbenchmark, this implementation could reduce
the runtime as following:
Name Percent of runtime reduced
memchr-lasx 37%-83%
memchr-lsx 30%-66%
memchr-aligned 0%-15%
Signed-off-by: Peng Fan <fanpeng@loongson.cn>
Signed-off-by: ticat_fp <fanpeng@loongson.cn>
---
sysdeps/loongarch/lp64/multiarch/Makefile | 3 +
.../lp64/multiarch/ifunc-impl-list.c | 7 ++
.../loongarch/lp64/multiarch/ifunc-memchr.h | 40 ++++++
.../loongarch/lp64/multiarch/memchr-aligned.S | 95 ++++++++++++++
.../loongarch/lp64/multiarch/memchr-lasx.S | 117 ++++++++++++++++++
sysdeps/loongarch/lp64/multiarch/memchr-lsx.S | 102 +++++++++++++++
sysdeps/loongarch/lp64/multiarch/memchr.c | 37 ++++++
7 files changed, 401 insertions(+)
create mode 100644 sysdeps/loongarch/lp64/multiarch/ifunc-memchr.h
create mode 100644 sysdeps/loongarch/lp64/multiarch/memchr-aligned.S
create mode 100644 sysdeps/loongarch/lp64/multiarch/memchr-lasx.S
create mode 100644 sysdeps/loongarch/lp64/multiarch/memchr-lsx.S
create mode 100644 sysdeps/loongarch/lp64/multiarch/memchr.c
diff --git a/sysdeps/loongarch/lp64/multiarch/Makefile b/sysdeps/loongarch/lp64/multiarch/Makefile
index 64416b02..2f4802cf 100644
--- a/sysdeps/loongarch/lp64/multiarch/Makefile
+++ b/sysdeps/loongarch/lp64/multiarch/Makefile
@@ -24,5 +24,8 @@ sysdep_routines += \
rawmemchr-aligned \
rawmemchr-lsx \
rawmemchr-lasx \
+ memchr-aligned \
+ memchr-lsx \
+ memchr-lasx \
# sysdep_routines
endif
diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
index 3db9af14..a567b9cf 100644
--- a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
@@ -102,5 +102,12 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, rawmemchr, 1, __rawmemchr_aligned)
)
+ IFUNC_IMPL (i, name, memchr,
+#if !defined __loongarch_soft_float
+ IFUNC_IMPL_ADD (array, i, memchr, SUPPORT_LASX, __memchr_lasx)
+ IFUNC_IMPL_ADD (array, i, memchr, SUPPORT_LSX, __memchr_lsx)
+#endif
+ IFUNC_IMPL_ADD (array, i, memchr, 1, __memchr_aligned)
+ )
return i;
}
diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-memchr.h b/sysdeps/loongarch/lp64/multiarch/ifunc-memchr.h
new file mode 100644
index 00000000..9060ccd5
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-memchr.h
@@ -0,0 +1,40 @@
+/* Common definition for memchr ifunc selections.
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 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 <ldsodefs.h>
+#include <ifunc-init.h>
+
+#if !defined __loongarch_soft_float
+extern __typeof (REDIRECT_NAME) OPTIMIZE (lasx) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (lsx) attribute_hidden;
+#endif
+extern __typeof (REDIRECT_NAME) OPTIMIZE (aligned) attribute_hidden;
+
+static inline void *
+IFUNC_SELECTOR (void)
+{
+#if !defined __loongarch_soft_float
+ if (SUPPORT_LASX)
+ return OPTIMIZE (lasx);
+ else if (SUPPORT_LSX)
+ return OPTIMIZE (lsx);
+ else
+#endif
+ return OPTIMIZE (aligned);
+}
diff --git a/sysdeps/loongarch/lp64/multiarch/memchr-aligned.S b/sysdeps/loongarch/lp64/multiarch/memchr-aligned.S
new file mode 100644
index 00000000..81d0d004
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memchr-aligned.S
@@ -0,0 +1,95 @@
+/* Optimized memchr implementation using basic LoongArch instructions.
+ Copyright (C) 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 <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc)
+# define MEMCHR_NAME __memchr_aligned
+#else
+# define MEMCHR_NAME memchr
+#endif
+
+LEAF(MEMCHR_NAME, 6)
+ beqz a2, L(out)
+ andi t1, a0, 0x7
+ add.d a5, a0, a2
+ bstrins.d a0, zero, 2, 0
+
+ ld.d t0, a0, 0
+ bstrins.d a1, a1, 15, 8
+ lu12i.w a3, 0x01010
+ slli.d t2, t1, 03
+
+ bstrins.d a1, a1, 31, 16
+ ori a3, a3, 0x101
+ li.d t7, -1
+ li.d t8, 8
+
+ bstrins.d a1, a1, 63, 32
+ bstrins.d a3, a3, 63, 32
+ sll.d t2, t7, t2
+ xor t0, t0, a1
+
+
+ addi.d a6, a5, -1
+ slli.d a4, a3, 7
+ sub.d t1, t8, t1
+ orn t0, t0, t2
+
+ sub.d t2, t0, a3
+ andn t3, a4, t0
+ bstrins.d a6, zero, 2, 0
+ and t0, t2, t3
+
+ bgeu t1, a2, L(end)
+L(loop):
+ bnez t0, L(found)
+ ld.d t1, a0, 8
+ xor t0, t1, a1
+
+ addi.d a0, a0, 8
+ sub.d t2, t0, a3
+ andn t3, a4, t0
+ and t0, t2, t3
+
+
+ bne a0, a6, L(loop)
+L(end):
+ sub.d t1, a5, a6
+ ctz.d t0, t0
+ srli.d t0, t0, 3
+
+ sltu t1, t0, t1
+ add.d a0, a0, t0
+ maskeqz a0, a0, t1
+ jr ra
+
+L(found):
+ ctz.d t0, t0
+ srli.d t0, t0, 3
+ add.d a0, a0, t0
+ jr ra
+
+L(out):
+ move a0, zero
+ jr ra
+END(MEMCHR_NAME)
+
+libc_hidden_builtin_def (MEMCHR_NAME)
diff --git a/sysdeps/loongarch/lp64/multiarch/memchr-lasx.S b/sysdeps/loongarch/lp64/multiarch/memchr-lasx.S
new file mode 100644
index 00000000..a26cdf48
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memchr-lasx.S
@@ -0,0 +1,117 @@
+/* Optimized memchr implementation using LoongArch LASX instructions.
+ Copyright (C) 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 <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc) && !defined __loongarch_soft_float
+
+# define MEMCHR __memchr_lasx
+
+LEAF(MEMCHR, 6)
+ beqz a2, L(ret0)
+ add.d a3, a0, a2
+ andi t0, a0, 0x3f
+ bstrins.d a0, zero, 5, 0
+
+ xvld xr0, a0, 0
+ xvld xr1, a0, 32
+ li.d t1, -1
+ li.d t2, 64
+
+ xvreplgr2vr.b xr2, a1
+ sll.d t3, t1, t0
+ sub.d t2, t2, t0
+ xvseq.b xr0, xr0, xr2
+
+ xvseq.b xr1, xr1, xr2
+ xvmsknz.b xr0, xr0
+ xvmsknz.b xr1, xr1
+ xvpickve.w xr3, xr0, 4
+
+
+ xvpickve.w xr4, xr1, 4
+ vilvl.h vr0, vr3, vr0
+ vilvl.h vr1, vr4, vr1
+ vilvl.w vr0, vr1, vr0
+
+ movfr2gr.d t0, fa0
+ and t0, t0, t3
+ bgeu t2, a2, L(end)
+ bnez t0, L(found)
+
+ addi.d a4, a3, -1
+ bstrins.d a4, zero, 5, 0
+L(loop):
+ xvld xr0, a0, 64
+ xvld xr1, a0, 96
+
+ addi.d a0, a0, 64
+ xvseq.b xr0, xr0, xr2
+ xvseq.b xr1, xr1, xr2
+ beq a0, a4, L(out)
+
+
+ xvmax.bu xr3, xr0, xr1
+ xvseteqz.v fcc0, xr3
+ bcnez fcc0, L(loop)
+ xvmsknz.b xr0, xr0
+
+ xvmsknz.b xr1, xr1
+ xvpickve.w xr3, xr0, 4
+ xvpickve.w xr4, xr1, 4
+ vilvl.h vr0, vr3, vr0
+
+ vilvl.h vr1, vr4, vr1
+ vilvl.w vr0, vr1, vr0
+ movfr2gr.d t0, fa0
+L(found):
+ ctz.d t1, t0
+
+ add.d a0, a0, t1
+ jr ra
+L(ret0):
+ move a0, zero
+ jr ra
+
+
+L(out):
+ xvmsknz.b xr0, xr0
+ xvmsknz.b xr1, xr1
+ xvpickve.w xr3, xr0, 4
+ xvpickve.w xr4, xr1, 4
+
+ vilvl.h vr0, vr3, vr0
+ vilvl.h vr1, vr4, vr1
+ vilvl.w vr0, vr1, vr0
+ movfr2gr.d t0, fa0
+
+L(end):
+ sub.d t2, zero, a3
+ srl.d t1, t1, t2
+ and t0, t0, t1
+ ctz.d t1, t0
+
+ add.d a0, a0, t1
+ maskeqz a0, a0, t0
+ jr ra
+END(MEMCHR)
+
+libc_hidden_builtin_def (MEMCHR)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/memchr-lsx.S b/sysdeps/loongarch/lp64/multiarch/memchr-lsx.S
new file mode 100644
index 00000000..a73ecd25
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memchr-lsx.S
@@ -0,0 +1,102 @@
+/* Optimized memchr implementation using LoongArch LSX instructions.
+ Copyright (C) 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 <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc) && !defined __loongarch_soft_float
+
+# define MEMCHR __memchr_lsx
+
+LEAF(MEMCHR, 6)
+ beqz a2, L(ret0)
+ add.d a3, a0, a2
+ andi t0, a0, 0x1f
+ bstrins.d a0, zero, 4, 0
+
+ vld vr0, a0, 0
+ vld vr1, a0, 16
+ li.d t1, -1
+ li.d t2, 32
+
+ vreplgr2vr.b vr2, a1
+ sll.d t3, t1, t0
+ sub.d t2, t2, t0
+ vseq.b vr0, vr0, vr2
+
+ vseq.b vr1, vr1, vr2
+ vmsknz.b vr0, vr0
+ vmsknz.b vr1, vr1
+ vilvl.h vr0, vr1, vr0
+
+
+ movfr2gr.s t0, fa0
+ and t0, t0, t3
+ bgeu t2, a2, L(end)
+ bnez t0, L(found)
+
+ addi.d a4, a3, -1
+ bstrins.d a4, zero, 4, 0
+L(loop):
+ vld vr0, a0, 32
+ vld vr1, a0, 48
+
+ addi.d a0, a0, 32
+ vseq.b vr0, vr0, vr2
+ vseq.b vr1, vr1, vr2
+ beq a0, a4, L(out)
+
+ vmax.bu vr3, vr0, vr1
+ vseteqz.v fcc0, vr3
+ bcnez fcc0, L(loop)
+ vmsknz.b vr0, vr0
+
+
+ vmsknz.b vr1, vr1
+ vilvl.h vr0, vr1, vr0
+ movfr2gr.s t0, fa0
+L(found):
+ ctz.w t0, t0
+
+ add.d a0, a0, t0
+ jr ra
+L(ret0):
+ move a0, zero
+ jr ra
+
+L(out):
+ vmsknz.b vr0, vr0
+ vmsknz.b vr1, vr1
+ vilvl.h vr0, vr1, vr0
+ movfr2gr.s t0, fa0
+
+L(end):
+ sub.d t2, zero, a3
+ srl.w t1, t1, t2
+ and t0, t0, t1
+ ctz.w t1, t0
+
+
+ add.d a0, a0, t1
+ maskeqz a0, a0, t0
+ jr ra
+END(MEMCHR)
+
+libc_hidden_builtin_def (MEMCHR)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/memchr.c b/sysdeps/loongarch/lp64/multiarch/memchr.c
new file mode 100644
index 00000000..059479c0
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memchr.c
@@ -0,0 +1,37 @@
+/* Multiple versions of memchr.
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 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/>. */
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+# define memchr __redirect_memchr
+# include <string.h>
+# undef memchr
+
+# define SYMBOL_NAME memchr
+# include "ifunc-memchr.h"
+
+libc_ifunc_redirected (__redirect_memchr, memchr,
+ IFUNC_SELECTOR ());
+
+# ifdef SHARED
+__hidden_ver1 (memchr, __GI_memchr, __redirect_memchr)
+ __attribute__ ((visibility ("hidden"))) __attribute_copy__ (memchr);
+# endif
+
+#endif
--
2.33.0
|
