about summary refs log tree commit diff
path: root/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S
blob: 1bb12e81b09d36787b2715fabd5fe319cfc7ca22 (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
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
/* memcpy optimized with AVX512 for KNL hardware.
   Copyright (C) 2016 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
   <http://www.gnu.org/licenses/>.  */

#include <sysdep.h>

#if defined HAVE_AVX512_ASM_SUPPORT && IS_IN (libc) \
    && (defined SHARED \
	|| defined USE_AS_MEMMOVE \
	|| !defined USE_MULTIARCH)

#include "asm-syntax.h"
#ifndef MEMCPY
# define MEMCPY		__memcpy_avx512_no_vzeroupper
# define MEMCPY_CHK	__memcpy_chk_avx512_no_vzeroupper
#endif

	.section .text,"ax",@progbits
#if !defined USE_AS_BCOPY
ENTRY (MEMCPY_CHK)
	cmpq	%rdx, %rcx
	jb	HIDDEN_JUMPTARGET (__chk_fail)
END (MEMCPY_CHK)
#endif

ENTRY (MEMCPY)
	mov	%rdi, %rax
#ifdef USE_AS_MEMPCPY
	add	%rdx, %rax
#endif
	lea	(%rsi, %rdx), %rcx
	lea	(%rdi, %rdx), %r9
	cmp	$512, %rdx
	ja	L(512bytesormore)

L(check):
	cmp	$16, %rdx
	jbe	L(less_16bytes)
	cmp	$256, %rdx
	jb	L(less_256bytes)
	vmovups	(%rsi), %zmm0
	vmovups 0x40(%rsi), %zmm1
	vmovups 0x80(%rsi), %zmm2
	vmovups 0xC0(%rsi), %zmm3
	vmovups	-0x100(%rcx), %zmm4
	vmovups -0xC0(%rcx), %zmm5
	vmovups -0x80(%rcx), %zmm6
	vmovups -0x40(%rcx), %zmm7
	vmovups %zmm0, (%rdi)
	vmovups %zmm1, 0x40(%rdi)
	vmovups %zmm2, 0x80(%rdi)
	vmovups %zmm3, 0xC0(%rdi)
	vmovups	%zmm4, -0x100(%r9)
	vmovups %zmm5, -0xC0(%r9)
	vmovups %zmm6, -0x80(%r9)
	vmovups %zmm7, -0x40(%r9)
	ret

L(less_256bytes):
	cmp	$128, %dl
	jb	L(less_128bytes)
	vmovups	(%rsi), %zmm0
	vmovups 0x40(%rsi), %zmm1
	vmovups -0x80(%rcx), %zmm2
	vmovups -0x40(%rcx), %zmm3
	vmovups	%zmm0, (%rdi)
	vmovups %zmm1, 0x40(%rdi)
	vmovups %zmm2, -0x80(%r9)
	vmovups %zmm3, -0x40(%r9)
	ret

L(less_128bytes):
	cmp	$64, %dl
	jb	L(less_64bytes)
	vmovdqu (%rsi), %ymm0
	vmovdqu 0x20(%rsi), %ymm1
	vmovdqu -0x40(%rcx), %ymm2
	vmovdqu -0x20(%rcx), %ymm3
	vmovdqu %ymm0, (%rdi)
	vmovdqu %ymm1, 0x20(%rdi)
	vmovdqu %ymm2, -0x40(%r9)
	vmovdqu %ymm3, -0x20(%r9)
	ret

L(less_64bytes):
	cmp	$32, %dl
	jb	L(less_32bytes)
	vmovdqu	(%rsi), %ymm0
	vmovdqu -0x20(%rcx), %ymm1
	vmovdqu	%ymm0, (%rdi)
	vmovdqu	%ymm1, -0x20(%r9)
	ret

L(less_32bytes):
	vmovdqu (%rsi), %xmm0
	vmovdqu -0x10(%rcx), %xmm1
	vmovdqu %xmm0, (%rdi)
	vmovdqu %xmm1, -0x10(%r9)
	ret

L(less_16bytes):
	cmp	$8, %dl
	jb	L(less_8bytes)
	movq	(%rsi), %rsi
	movq	-0x8(%rcx), %rcx
	movq	%rsi, (%rdi)
	movq	%rcx, -0x8(%r9)
	ret

L(less_8bytes):
	cmp	$4, %dl
	jb	L(less_4bytes)
	mov	(%rsi), %esi
	mov	-0x4(%rcx), %ecx
	mov	%esi, (%rdi)
	mov	%ecx, -0x4(%r9)
	ret

L(less_4bytes):
	cmp	$2, %dl
	jb	L(less_2bytes)
	mov	(%rsi), %si
	mov	-0x2(%rcx), %cx
	mov	%si, (%rdi)
	mov	%cx, -0x2(%r9)
	ret

L(less_2bytes):
	cmp	$1, %dl
	jb	L(less_1bytes)
	mov	(%rsi), %cl
	mov	%cl, (%rdi)
L(less_1bytes):
	ret

L(512bytesormore):
#ifdef SHARED_CACHE_SIZE_HALF
	mov	$SHARED_CACHE_SIZE_HALF, %r8
#else
	mov	__x86_shared_cache_size_half(%rip), %r8
#endif
	cmp	%r8, %rdx
	jae	L(preloop_large)
	cmp	$1024, %rdx
	ja	L(1024bytesormore)
	prefetcht1 (%rsi)
	prefetcht1 0x40(%rsi)
	prefetcht1 0x80(%rsi)
	prefetcht1 0xC0(%rsi)
	prefetcht1 0x100(%rsi)
	prefetcht1 0x140(%rsi)
	prefetcht1 0x180(%rsi)
	prefetcht1 0x1C0(%rsi)
	prefetcht1 -0x200(%rcx)
	prefetcht1 -0x1C0(%rcx)
	prefetcht1 -0x180(%rcx)
	prefetcht1 -0x140(%rcx)
	prefetcht1 -0x100(%rcx)
	prefetcht1 -0xC0(%rcx)
	prefetcht1 -0x80(%rcx)
	prefetcht1 -0x40(%rcx)
	vmovups	(%rsi), %zmm0
	vmovups 0x40(%rsi), %zmm1
	vmovups 0x80(%rsi), %zmm2
	vmovups 0xC0(%rsi), %zmm3
	vmovups	0x100(%rsi), %zmm4
	vmovups 0x140(%rsi), %zmm5
	vmovups 0x180(%rsi), %zmm6
	vmovups 0x1C0(%rsi), %zmm7
	vmovups	-0x200(%rcx), %zmm8
	vmovups -0x1C0(%rcx), %zmm9
	vmovups -0x180(%rcx), %zmm10
	vmovups -0x140(%rcx), %zmm11
	vmovups	-0x100(%rcx), %zmm12
	vmovups -0xC0(%rcx), %zmm13
	vmovups -0x80(%rcx), %zmm14
	vmovups -0x40(%rcx), %zmm15
	vmovups %zmm0, (%rdi)
	vmovups %zmm1, 0x40(%rdi)
	vmovups %zmm2, 0x80(%rdi)
	vmovups %zmm3, 0xC0(%rdi)
	vmovups %zmm4, 0x100(%rdi)
	vmovups %zmm5, 0x140(%rdi)
	vmovups %zmm6, 0x180(%rdi)
	vmovups %zmm7, 0x1C0(%rdi)
	vmovups	%zmm8, -0x200(%r9)
	vmovups %zmm9, -0x1C0(%r9)
	vmovups %zmm10, -0x180(%r9)
	vmovups %zmm11, -0x140(%r9)
	vmovups	%zmm12, -0x100(%r9)
	vmovups %zmm13, -0xC0(%r9)
	vmovups %zmm14, -0x80(%r9)
	vmovups %zmm15, -0x40(%r9)
	ret

L(1024bytesormore):
	cmp	%rsi, %rdi
	ja	L(1024bytesormore_bkw)
	sub	$512, %r9
	vmovups -0x200(%rcx), %zmm8
	vmovups -0x1C0(%rcx), %zmm9
	vmovups -0x180(%rcx), %zmm10
	vmovups -0x140(%rcx), %zmm11
	vmovups	-0x100(%rcx), %zmm12
	vmovups -0xC0(%rcx), %zmm13
	vmovups -0x80(%rcx), %zmm14
	vmovups -0x40(%rcx), %zmm15
	prefetcht1 (%rsi)
	prefetcht1 0x40(%rsi)
	prefetcht1 0x80(%rsi)
	prefetcht1 0xC0(%rsi)
	prefetcht1 0x100(%rsi)
	prefetcht1 0x140(%rsi)
	prefetcht1 0x180(%rsi)
	prefetcht1 0x1C0(%rsi)

/* Loop with unaligned memory access.  */
L(gobble_512bytes_loop):
	vmovups	(%rsi), %zmm0
	vmovups 0x40(%rsi), %zmm1
	vmovups 0x80(%rsi), %zmm2
	vmovups 0xC0(%rsi), %zmm3
	vmovups	0x100(%rsi), %zmm4
	vmovups 0x140(%rsi), %zmm5
	vmovups 0x180(%rsi), %zmm6
	vmovups 0x1C0(%rsi), %zmm7
	add	$512, %rsi
	prefetcht1 (%rsi)
	prefetcht1 0x40(%rsi)
	prefetcht1 0x80(%rsi)
	prefetcht1 0xC0(%rsi)
	prefetcht1 0x100(%rsi)
	prefetcht1 0x140(%rsi)
	prefetcht1 0x180(%rsi)
	prefetcht1 0x1C0(%rsi)
	vmovups	%zmm0, (%rdi)
	vmovups %zmm1, 0x40(%rdi)
	vmovups %zmm2, 0x80(%rdi)
	vmovups %zmm3, 0xC0(%rdi)
	vmovups	%zmm4, 0x100(%rdi)
	vmovups %zmm5, 0x140(%rdi)
	vmovups %zmm6, 0x180(%rdi)
	vmovups %zmm7, 0x1C0(%rdi)
	add	$512, %rdi
	cmp	%r9, %rdi
	jb	L(gobble_512bytes_loop)
	vmovups %zmm8, (%r9)
	vmovups %zmm9, 0x40(%r9)
	vmovups %zmm10, 0x80(%r9)
	vmovups %zmm11, 0xC0(%r9)
	vmovups %zmm12, 0x100(%r9)
	vmovups %zmm13, 0x140(%r9)
	vmovups %zmm14, 0x180(%r9)
	vmovups %zmm15, 0x1C0(%r9)
	ret

L(1024bytesormore_bkw):
	add	$512, %rdi
	vmovups	0x1C0(%rsi), %zmm8
	vmovups 0x180(%rsi), %zmm9
	vmovups 0x140(%rsi), %zmm10
	vmovups 0x100(%rsi), %zmm11
	vmovups	0xC0(%rsi), %zmm12
	vmovups 0x80(%rsi), %zmm13
	vmovups 0x40(%rsi), %zmm14
	vmovups (%rsi), %zmm15
	prefetcht1 -0x40(%rcx)
	prefetcht1 -0x80(%rcx)
	prefetcht1 -0xC0(%rcx)
	prefetcht1 -0x100(%rcx)
	prefetcht1 -0x140(%rcx)
	prefetcht1 -0x180(%rcx)
	prefetcht1 -0x1C0(%rcx)
	prefetcht1 -0x200(%rcx)

/* Backward loop with unaligned memory access.  */
L(gobble_512bytes_loop_bkw):
	vmovups -0x40(%rcx), %zmm0
	vmovups -0x80(%rcx), %zmm1
	vmovups -0xC0(%rcx), %zmm2
	vmovups	-0x100(%rcx), %zmm3
	vmovups -0x140(%rcx), %zmm4
	vmovups -0x180(%rcx), %zmm5
	vmovups -0x1C0(%rcx), %zmm6
	vmovups	-0x200(%rcx), %zmm7
	sub	$512, %rcx
	prefetcht1 -0x40(%rcx)
	prefetcht1 -0x80(%rcx)
	prefetcht1 -0xC0(%rcx)
	prefetcht1 -0x100(%rcx)
	prefetcht1 -0x140(%rcx)
	prefetcht1 -0x180(%rcx)
	prefetcht1 -0x1C0(%rcx)
	prefetcht1 -0x200(%rcx)
	vmovups %zmm0, -0x40(%r9)
	vmovups %zmm1, -0x80(%r9)
	vmovups %zmm2, -0xC0(%r9)
	vmovups	%zmm3, -0x100(%r9)
	vmovups %zmm4, -0x140(%r9)
	vmovups %zmm5, -0x180(%r9)
	vmovups %zmm6, -0x1C0(%r9)
	vmovups	%zmm7, -0x200(%r9)
	sub	$512, %r9
	cmp	%rdi, %r9
	ja	L(gobble_512bytes_loop_bkw)
	vmovups %zmm8, -0x40(%rdi)
	vmovups %zmm9, -0x80(%rdi)
	vmovups %zmm10, -0xC0(%rdi)
	vmovups %zmm11, -0x100(%rdi)
	vmovups %zmm12, -0x140(%rdi)
	vmovups %zmm13, -0x180(%rdi)
	vmovups %zmm14, -0x1C0(%rdi)
	vmovups %zmm15, -0x200(%rdi)
	ret

L(preloop_large):
	cmp	%rsi, %rdi
	ja	L(preloop_large_bkw)
	vmovups	(%rsi), %zmm4
	vmovups	0x40(%rsi), %zmm5

/* Align destination for access with non-temporal stores in the loop.  */
	mov	%rdi, %r8
	and	$-0x80, %rdi
	add	$0x80, %rdi
	sub	%rdi, %r8
	sub	%r8, %rsi
	add	%r8, %rdx
L(gobble_256bytes_nt_loop):
	prefetcht1 0x200(%rsi)
	prefetcht1 0x240(%rsi)
	prefetcht1 0x280(%rsi)
	prefetcht1 0x2C0(%rsi)
	prefetcht1 0x300(%rsi)
	prefetcht1 0x340(%rsi)
	prefetcht1 0x380(%rsi)
	prefetcht1 0x3C0(%rsi)
	vmovdqu64 (%rsi), %zmm0
	vmovdqu64 0x40(%rsi), %zmm1
	vmovdqu64 0x80(%rsi), %zmm2
	vmovdqu64 0xC0(%rsi), %zmm3
	vmovntdq %zmm0, (%rdi)
	vmovntdq %zmm1, 0x40(%rdi)
	vmovntdq %zmm2, 0x80(%rdi)
	vmovntdq %zmm3, 0xC0(%rdi)
	sub	$256, %rdx
	add	$256, %rsi
	add	$256, %rdi
	cmp	$256, %rdx
	ja	L(gobble_256bytes_nt_loop)
	sfence
	vmovups	%zmm4, (%rax)
	vmovups	%zmm5, 0x40(%rax)
	jmp	L(check)

L(preloop_large_bkw):
	vmovups -0x80(%rcx), %zmm4
	vmovups -0x40(%rcx), %zmm5

/* Align end of destination for access with non-temporal stores.  */
	mov	%r9, %r8
	and	$-0x80, %r9
	sub	%r9, %r8
	sub	%r8, %rcx
	sub	%r8, %rdx
	add	%r9, %r8
L(gobble_256bytes_nt_loop_bkw):
	prefetcht1 -0x400(%rcx)
	prefetcht1 -0x3C0(%rcx)
	prefetcht1 -0x380(%rcx)
	prefetcht1 -0x340(%rcx)
	prefetcht1 -0x300(%rcx)
	prefetcht1 -0x2C0(%rcx)
	prefetcht1 -0x280(%rcx)
	prefetcht1 -0x240(%rcx)
	vmovdqu64 -0x100(%rcx), %zmm0
	vmovdqu64 -0xC0(%rcx), %zmm1
	vmovdqu64 -0x80(%rcx), %zmm2
	vmovdqu64 -0x40(%rcx), %zmm3
	vmovntdq %zmm0,	-0x100(%r9)
	vmovntdq %zmm1,	-0xC0(%r9)
	vmovntdq %zmm2,	-0x80(%r9)
	vmovntdq %zmm3,	-0x40(%r9)
	sub	$256, %rdx
	sub	$256, %rcx
	sub	$256, %r9
	cmp	$256, %rdx
	ja	L(gobble_256bytes_nt_loop_bkw)
	sfence
	vmovups	%zmm4, -0x80(%r8)
	vmovups	%zmm5, -0x40(%r8)
	jmp	L(check)
END (MEMCPY)
#endif