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
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
|
/*
Optimized memcpy for x86-64.
Copyright (C) 2007-2015 Free Software Foundation, Inc.
Contributed by Evandro Menezes <evandro.menezes@amd.com>, 2007.
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
<http://www.gnu.org/licenses/>.
*/
#include <sysdep.h>
#include "asm-syntax.h"
/* Stack slots in the red-zone. */
#ifdef USE_AS_MEMPCPY
# define RETVAL (0)
#else
# define RETVAL (-8)
#endif
#define SAVE0 (RETVAL - 8)
#define SAVE1 (SAVE0 - 8)
#define SAVE2 (SAVE1 - 8)
#define SAVE3 (SAVE2 - 8)
.text
#if defined SHARED && IS_IN (libc)
ENTRY_CHK (__memcpy_chk_sse2)
cmpq %rdx, %rcx
jb HIDDEN_JUMPTARGET (__chk_fail)
END_CHK (__memcpy_chk_sse2)
#endif
ENTRY(__memcpy_sse2) /* (void *, const void*, size_t) */
/* Handle tiny blocks. */
L(1try): /* up to 32B */
cmpq $32, %rdx
#ifndef USE_AS_MEMPCPY
movq %rdi, %rax /* save return value */
#endif
jae L(1after)
L(1): /* 1-byte once */
testb $1, %dl
jz L(1a)
movzbl (%rsi), %ecx
movb %cl, (%rdi)
incq %rsi
incq %rdi
.p2align 4,, 4
L(1a): /* 2-byte once */
testb $2, %dl
jz L(1b)
movzwl (%rsi), %ecx
movw %cx, (%rdi)
addq $2, %rsi
addq $2, %rdi
.p2align 4,, 4
L(1b): /* 4-byte once */
testb $4, %dl
jz L(1c)
movl (%rsi), %ecx
movl %ecx, (%rdi)
addq $4, %rsi
addq $4, %rdi
.p2align 4,, 4
L(1c): /* 8-byte once */
testb $8, %dl
jz L(1d)
movq (%rsi), %rcx
movq %rcx, (%rdi)
addq $8, %rsi
addq $8, %rdi
.p2align 4,, 4
L(1d): /* 16-byte loop */
andl $0xf0, %edx
jz L(exit)
.p2align 4
L(1loop):
movq (%rsi), %rcx
movq 8(%rsi), %r8
movq %rcx, (%rdi)
movq %r8, 8(%rdi)
subl $16, %edx
leaq 16(%rsi), %rsi
leaq 16(%rdi), %rdi
jnz L(1loop)
.p2align 4,, 4
L(exit): /* exit */
#ifdef USE_AS_MEMPCPY
movq %rdi, %rax /* return value */
#else
rep
#endif
retq
.p2align 4
L(1after):
#ifndef USE_AS_MEMPCPY
movq %rax, RETVAL(%rsp) /* save return value */
#endif
/* Align to the natural word size. */
L(aligntry):
movl %esi, %ecx /* align by source */
andl $7, %ecx
jz L(alignafter) /* already aligned */
L(align): /* align */
leaq -8(%rcx, %rdx), %rdx /* calculate remaining bytes */
subl $8, %ecx
.p2align 4
L(alignloop): /* 1-byte alignment loop */
movzbl (%rsi), %eax
movb %al, (%rdi)
incl %ecx
leaq 1(%rsi), %rsi
leaq 1(%rdi), %rdi
jnz L(alignloop)
.p2align 4
L(alignafter):
/* Handle mid-sized blocks. */
L(32try): /* up to 1KB */
cmpq $1024, %rdx
ja L(32after)
L(32): /* 32-byte loop */
movl %edx, %ecx
shrl $5, %ecx
jz L(32skip)
.p2align 4
L(32loop):
decl %ecx
movq (%rsi), %rax
movq 8(%rsi), %r8
movq 16(%rsi), %r9
movq 24(%rsi), %r10
movq %rax, (%rdi)
movq %r8, 8(%rdi)
movq %r9, 16(%rdi)
movq %r10, 24(%rdi)
leaq 32(%rsi), %rsi
leaq 32(%rdi), %rdi
jz L(32skip) /* help out smaller blocks */
decl %ecx
movq (%rsi), %rax
movq 8(%rsi), %r8
movq 16(%rsi), %r9
movq 24(%rsi), %r10
movq %rax, (%rdi)
movq %r8, 8(%rdi)
movq %r9, 16(%rdi)
movq %r10, 24(%rdi)
leaq 32(%rsi), %rsi
leaq 32(%rdi), %rdi
jnz L(32loop)
.p2align 4
L(32skip):
andl $31, %edx /* check for left overs */
#ifdef USE_AS_MEMPCPY
jnz L(1)
movq %rdi, %rax
#else
movq RETVAL(%rsp), %rax
jnz L(1)
rep
#endif
retq /* exit */
.p2align 4
L(32after):
/*
In order to minimize code-size in RTLD, algorithms specific for
larger blocks are excluded when building for RTLD.
*/
/* Handle blocks smaller than 1/2 L1. */
L(fasttry): /* first 1/2 L1 */
#if IS_IN (libc) /* only up to this algorithm outside of libc.so */
mov __x86_data_cache_size_half(%rip), %R11_LP
cmpq %rdx, %r11 /* calculate the smaller of */
cmovaq %rdx, %r11 /* remaining bytes and 1/2 L1 */
#endif
L(fast): /* good ol' MOVS */
#if IS_IN (libc)
movq %r11, %rcx
andq $-8, %r11
#else
movq %rdx, %rcx
#endif
shrq $3, %rcx
jz L(fastskip)
rep
movsq
.p2align 4,, 4
L(fastskip):
#if IS_IN (libc)
subq %r11, %rdx /* check for more */
testq $-8, %rdx
jnz L(fastafter)
#endif
andl $7, %edx /* check for left overs */
#ifdef USE_AS_MEMPCPY
jnz L(1)
movq %rdi, %rax
#else
movq RETVAL(%rsp), %rax
jnz L(1)
rep
#endif
retq /* exit */
#if IS_IN (libc) /* none of the algorithms below for RTLD */
.p2align 4
L(fastafter):
/* Handle large blocks smaller than 1/2 L2. */
L(pretry): /* first 1/2 L2 */
mov __x86_shared_cache_size_half (%rip), %R8_LP
cmpq %rdx, %r8 /* calculate the lesser of */
cmovaq %rdx, %r8 /* remaining bytes and 1/2 L2 */
L(pre): /* 64-byte with prefetching */
movq %r8, %rcx
andq $-64, %r8
shrq $6, %rcx
jz L(preskip)
movq %r14, SAVE0(%rsp)
cfi_rel_offset (%r14, SAVE0)
movq %r13, SAVE1(%rsp)
cfi_rel_offset (%r13, SAVE1)
movq %r12, SAVE2(%rsp)
cfi_rel_offset (%r12, SAVE2)
movq %rbx, SAVE3(%rsp)
cfi_rel_offset (%rbx, SAVE3)
cmpl $0, __x86_prefetchw(%rip)
jz L(preloop) /* check if PREFETCHW OK */
.p2align 4
/* ... when PREFETCHW is available (less cache-probe traffic in MP systems). */
L(prewloop): /* cache-line in state M */
decq %rcx
movq (%rsi), %rax
movq 8 (%rsi), %rbx
movq 16 (%rsi), %r9
movq 24 (%rsi), %r10
movq 32 (%rsi), %r11
movq 40 (%rsi), %r12
movq 48 (%rsi), %r13
movq 56 (%rsi), %r14
prefetcht0 0 + 896 (%rsi)
prefetcht0 64 + 896 (%rsi)
movq %rax, (%rdi)
movq %rbx, 8(%rdi)
movq %r9, 16(%rdi)
movq %r10, 24(%rdi)
movq %r11, 32(%rdi)
movq %r12, 40(%rdi)
movq %r13, 48(%rdi)
movq %r14, 56(%rdi)
leaq 64(%rsi), %rsi
leaq 64(%rdi), %rdi
jz L(prebail)
decq %rcx
movq (%rsi), %rax
movq 8(%rsi), %rbx
movq 16(%rsi), %r9
movq 24(%rsi), %r10
movq 32(%rsi), %r11
movq 40(%rsi), %r12
movq 48(%rsi), %r13
movq 56(%rsi), %r14
movq %rax, (%rdi)
movq %rbx, 8(%rdi)
movq %r9, 16(%rdi)
movq %r10, 24(%rdi)
movq %r11, 32(%rdi)
movq %r12, 40(%rdi)
movq %r13, 48(%rdi)
movq %r14, 56(%rdi)
prefetchw 896 - 64(%rdi)
prefetchw 896 - 0(%rdi)
leaq 64(%rsi), %rsi
leaq 64(%rdi), %rdi
jnz L(prewloop)
jmp L(prebail)
.p2align 4
/* ... when PREFETCHW is not available. */
L(preloop): /* cache-line in state E */
decq %rcx
movq (%rsi), %rax
movq 8(%rsi), %rbx
movq 16(%rsi), %r9
movq 24(%rsi), %r10
movq 32(%rsi), %r11
movq 40(%rsi), %r12
movq 48(%rsi), %r13
movq 56(%rsi), %r14
prefetcht0 896 + 0(%rsi)
prefetcht0 896 + 64(%rsi)
movq %rax, (%rdi)
movq %rbx, 8(%rdi)
movq %r9, 16(%rdi)
movq %r10, 24(%rdi)
movq %r11, 32(%rdi)
movq %r12, 40(%rdi)
movq %r13, 48(%rdi)
movq %r14, 56(%rdi)
leaq 64 (%rsi), %rsi
leaq 64 (%rdi), %rdi
jz L(prebail)
decq %rcx
movq (%rsi), %rax
movq 8(%rsi), %rbx
movq 16(%rsi), %r9
movq 24(%rsi), %r10
movq 32(%rsi), %r11
movq 40(%rsi), %r12
movq 48(%rsi), %r13
movq 56(%rsi), %r14
prefetcht0 896 - 64(%rdi)
prefetcht0 896 - 0(%rdi)
movq %rax, (%rdi)
movq %rbx, 8(%rdi)
movq %r9, 16(%rdi)
movq %r10, 24(%rdi)
movq %r11, 32(%rdi)
movq %r12, 40(%rdi)
movq %r13, 48(%rdi)
movq %r14, 56(%rdi)
leaq 64(%rsi), %rsi
leaq 64(%rdi), %rdi
jnz L(preloop)
L(prebail):
movq SAVE3(%rsp), %rbx
cfi_restore (%rbx)
movq SAVE2(%rsp), %r12
cfi_restore (%r12)
movq SAVE1(%rsp), %r13
cfi_restore (%r13)
movq SAVE0(%rsp), %r14
cfi_restore (%r14)
/* .p2align 4 */
L(preskip):
subq %r8, %rdx /* check for more */
testq $-64, %rdx
jnz L(preafter)
andl $63, %edx /* check for left overs */
#ifdef USE_AS_MEMPCPY
jnz L(1)
movq %rdi, %rax
#else
movq RETVAL(%rsp), %rax
jnz L(1)
rep
#endif
retq /* exit */
.p2align 4
L(preafter):
/* Handle huge blocks. */
L(NTtry):
L(NT): /* non-temporal 128-byte */
movq %rdx, %rcx
shrq $7, %rcx
jz L(NTskip)
movq %r14, SAVE0(%rsp)
cfi_rel_offset (%r14, SAVE0)
movq %r13, SAVE1(%rsp)
cfi_rel_offset (%r13, SAVE1)
movq %r12, SAVE2(%rsp)
cfi_rel_offset (%r12, SAVE2)
.p2align 4
L(NTloop):
prefetchnta 768(%rsi)
prefetchnta 832(%rsi)
decq %rcx
movq (%rsi), %rax
movq 8(%rsi), %r8
movq 16(%rsi), %r9
movq 24(%rsi), %r10
movq 32(%rsi), %r11
movq 40(%rsi), %r12
movq 48(%rsi), %r13
movq 56(%rsi), %r14
movntiq %rax, (%rdi)
movntiq %r8, 8(%rdi)
movntiq %r9, 16(%rdi)
movntiq %r10, 24(%rdi)
movntiq %r11, 32(%rdi)
movntiq %r12, 40(%rdi)
movntiq %r13, 48(%rdi)
movntiq %r14, 56(%rdi)
movq 64(%rsi), %rax
movq 72(%rsi), %r8
movq 80(%rsi), %r9
movq 88(%rsi), %r10
movq 96(%rsi), %r11
movq 104(%rsi), %r12
movq 112(%rsi), %r13
movq 120(%rsi), %r14
movntiq %rax, 64(%rdi)
movntiq %r8, 72(%rdi)
movntiq %r9, 80(%rdi)
movntiq %r10, 88(%rdi)
movntiq %r11, 96(%rdi)
movntiq %r12, 104(%rdi)
movntiq %r13, 112(%rdi)
movntiq %r14, 120(%rdi)
leaq 128(%rsi), %rsi
leaq 128(%rdi), %rdi
jnz L(NTloop)
sfence /* serialize memory stores */
movq SAVE2(%rsp), %r12
cfi_restore (%r12)
movq SAVE1(%rsp), %r13
cfi_restore (%r13)
movq SAVE0(%rsp), %r14
cfi_restore (%r14)
L(NTskip):
andl $127, %edx /* check for left overs */
#ifdef USE_AS_MEMPCPY
jnz L(1)
movq %rdi, %rax
#else
movq RETVAL(%rsp), %rax
jnz L(1)
rep
#endif
retq /* exit */
#endif /* IS_IN (libc) */
END(__memcpy_sse2)
|
