aboutsummaryrefslogtreecommitdiffstats
path: root/arch/xtensa/lib/memcopy.S
blob: c48b80acb5f014813c9d509f425e95e4f3b872e0 (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
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
570
571
572
573
574
/*
 * arch/xtensa/lib/hal/memcopy.S -- Core HAL library functions
 * xthal_memcpy and xthal_bcopy
 *
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 *
 * Copyright (C) 2002 - 2012 Tensilica Inc.
 */

#include <variant/core.h>

	.macro	src_b	r, w0, w1
#ifdef __XTENSA_EB__
	src	\r, \w0, \w1
#else
	src	\r, \w1, \w0
#endif
	.endm

	.macro	ssa8	r
#ifdef __XTENSA_EB__
	ssa8b	\r
#else
	ssa8l	\r
#endif
	.endm

/*
 * void *memcpy(void *dst, const void *src, size_t len);
 *
 * This function is intended to do the same thing as the standard
 * library function memcpy() for most cases.
 * However, where the source and/or destination references
 * an instruction RAM or ROM or a data RAM or ROM, that
 * source and/or destination will always be accessed with
 * 32-bit load and store instructions (as required for these
 * types of devices).
 *
 * !!!!!!!  XTFIXME:
 * !!!!!!!  Handling of IRAM/IROM has not yet
 * !!!!!!!  been implemented.
 *
 * The (general case) algorithm is as follows:
 *   If destination is unaligned, align it by conditionally
 *     copying 1 and 2 bytes.
 *   If source is aligned,
 *     do 16 bytes with a loop, and then finish up with
 *     8, 4, 2, and 1 byte copies conditional on the length;
 *   else (if source is unaligned),
 *     do the same, but use SRC to align the source data.
 *   This code tries to use fall-through branches for the common
 *     case of aligned source and destination and multiple
 *     of 4 (or 8) length.
 *
 * Register use:
 *	a0/ return address
 *	a1/ stack pointer
 *	a2/ return value
 *	a3/ src
 *	a4/ length
 *	a5/ dst
 *	a6/ tmp
 *	a7/ tmp
 *	a8/ tmp
 *	a9/ tmp
 *	a10/ tmp
 *	a11/ tmp
 */

	.text

/*
 * Byte by byte copy
 */
	.align	4
	.byte	0		# 1 mod 4 alignment for LOOPNEZ
				# (0 mod 4 alignment for LBEG)
.Lbytecopy:
#if XCHAL_HAVE_LOOPS
	loopnez	a4, .Lbytecopydone
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a4, .Lbytecopydone
	add	a7, a3, a4	# a7 = end address for source
#endif /* !XCHAL_HAVE_LOOPS */
.Lnextbyte:
	l8ui	a6, a3, 0
	addi	a3, a3, 1
	s8i	a6, a5, 0
	addi	a5, a5, 1
#if !XCHAL_HAVE_LOOPS
	bne	a3, a7, .Lnextbyte # continue loop if $a3:src != $a7:src_end
#endif /* !XCHAL_HAVE_LOOPS */
.Lbytecopydone:
	retw

/*
 * Destination is unaligned
 */

	.align	4
.Ldst1mod2:	# dst is only byte aligned
	_bltui	a4, 7, .Lbytecopy	# do short copies byte by byte

	# copy 1 byte
	l8ui	a6, a3,  0
	addi	a3, a3,  1
	addi	a4, a4, -1
	s8i	a6, a5,  0
	addi	a5, a5,  1
	_bbci.l	a5, 1, .Ldstaligned	# if dst is now aligned, then
					# return to main algorithm
.Ldst2mod4:	# dst 16-bit aligned
	# copy 2 bytes
	_bltui	a4, 6, .Lbytecopy	# do short copies byte by byte
	l8ui	a6, a3,  0
	l8ui	a7, a3,  1
	addi	a3, a3,  2
	addi	a4, a4, -2
	s8i	a6, a5,  0
	s8i	a7, a5,  1
	addi	a5, a5,  2
	j	.Ldstaligned	# dst is now aligned, return to main algorithm

	.align	4
	.global	memcpy
	.type   memcpy,@function
memcpy:

	entry	sp, 16		# minimal stack frame
	# a2/ dst, a3/ src, a4/ len
	mov	a5, a2		# copy dst so that a2 is return value
.Lcommon:
	_bbsi.l	a2, 0, .Ldst1mod2	# if dst is 1 mod 2
	_bbsi.l	a2, 1, .Ldst2mod4	# if dst is 2 mod 4
.Ldstaligned:	# return here from .Ldst?mod? once dst is aligned
	srli	a7, a4, 4	# number of loop iterations with 16B
				# per iteration
	movi	a8, 3		# if source is not aligned,
	_bany	a3, a8, .Lsrcunaligned	# then use shifting copy
	/*
	 * Destination and source are word-aligned, use word copy.
	 */
	# copy 16 bytes per iteration for word-aligned dst and word-aligned src
#if XCHAL_HAVE_LOOPS
	loopnez	a7, .Loop1done
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a7, .Loop1done
	slli	a8, a7, 4
	add	a8, a8, a3	# a8 = end of last 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.Loop1:
	l32i	a6, a3,  0
	l32i	a7, a3,  4
	s32i	a6, a5,  0
	l32i	a6, a3,  8
	s32i	a7, a5,  4
	l32i	a7, a3, 12
	s32i	a6, a5,  8
	addi	a3, a3, 16
	s32i	a7, a5, 12
	addi	a5, a5, 16
#if !XCHAL_HAVE_LOOPS
	bne	a3, a8, .Loop1  # continue loop if a3:src != a8:src_end
#endif /* !XCHAL_HAVE_LOOPS */
.Loop1done:
	bbci.l	a4, 3, .L2
	# copy 8 bytes
	l32i	a6, a3,  0
	l32i	a7, a3,  4
	addi	a3, a3,  8
	s32i	a6, a5,  0
	s32i	a7, a5,  4
	addi	a5, a5,  8
.L2:
	bbsi.l	a4, 2, .L3
	bbsi.l	a4, 1, .L4
	bbsi.l	a4, 0, .L5
	retw
.L3:
	# copy 4 bytes
	l32i	a6, a3,  0
	addi	a3, a3,  4
	s32i	a6, a5,  0
	addi	a5, a5,  4
	bbsi.l	a4, 1, .L4
	bbsi.l	a4, 0, .L5
	retw
.L4:
	# copy 2 bytes
	l16ui	a6, a3,  0
	addi	a3, a3,  2
	s16i	a6, a5,  0
	addi	a5, a5,  2
	bbsi.l	a4, 0, .L5
	retw
.L5:
	# copy 1 byte
	l8ui	a6, a3,  0
	s8i	a6, a5,  0
	retw

/*
 * Destination is aligned, Source is unaligned
 */

	.align	4
.Lsrcunaligned:
	_beqz	a4, .Ldone	# avoid loading anything for zero-length copies
	# copy 16 bytes per iteration for word-aligned dst and unaligned src
	ssa8	a3		# set shift amount from byte offset
#define SIM_CHECKS_ALIGNMENT	1	/* set to 1 when running on ISS (simulator) with the
					   lint or ferret client, or 0 to save a few cycles */
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
	and	a11, a3, a8	# save unalignment offset for below
	sub	a3, a3, a11	# align a3
#endif
	l32i	a6, a3, 0	# load first word
#if XCHAL_HAVE_LOOPS
	loopnez	a7, .Loop2done
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a7, .Loop2done
	slli	a10, a7, 4
	add	a10, a10, a3	# a10 = end of last 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.Loop2:
	l32i	a7, a3,  4
	l32i	a8, a3,  8
	src_b	a6, a6, a7
	s32i	a6, a5,  0
	l32i	a9, a3, 12
	src_b	a7, a7, a8
	s32i	a7, a5,  4
	l32i	a6, a3, 16
	src_b	a8, a8, a9
	s32i	a8, a5,  8
	addi	a3, a3, 16
	src_b	a9, a9, a6
	s32i	a9, a5, 12
	addi	a5, a5, 16
#if !XCHAL_HAVE_LOOPS
	bne	a3, a10, .Loop2 # continue loop if a3:src != a10:src_end
#endif /* !XCHAL_HAVE_LOOPS */
.Loop2done:
	bbci.l	a4, 3, .L12
	# copy 8 bytes
	l32i	a7, a3,  4
	l32i	a8, a3,  8
	src_b	a6, a6, a7
	s32i	a6, a5,  0
	addi	a3, a3,  8
	src_b	a7, a7, a8
	s32i	a7, a5,  4
	addi	a5, a5,  8
	mov	a6, a8
.L12:
	bbci.l	a4, 2, .L13
	# copy 4 bytes
	l32i	a7, a3,  4
	addi	a3, a3,  4
	src_b	a6, a6, a7
	s32i	a6, a5,  0
	addi	a5, a5,  4
	mov	a6, a7
.L13:
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
	add	a3, a3, a11	# readjust a3 with correct misalignment
#endif
	bbsi.l	a4, 1, .L14
	bbsi.l	a4, 0, .L15
.Ldone:	retw
.L14:
	# copy 2 bytes
	l8ui	a6, a3,  0
	l8ui	a7, a3,  1
	addi	a3, a3,  2
	s8i	a6, a5,  0
	s8i	a7, a5,  1
	addi	a5, a5,  2
	bbsi.l	a4, 0, .L15
	retw
.L15:
	# copy 1 byte
	l8ui	a6, a3,  0
	s8i	a6, a5,  0
	retw


/*
 * void bcopy(const void *src, void *dest, size_t n);
 */
	.align	4
	.global	bcopy
	.type   bcopy,@function
bcopy:
	entry	sp, 16		# minimal stack frame
	# a2=src, a3=dst, a4=len
	mov	a5, a3
	mov	a3, a2
	mov	a2, a5
	j	.Lmovecommon	# go to common code for memmove+bcopy

/*
 * void *memmove(void *dst, const void *src, size_t len);
 *
 * This function is intended to do the same thing as the standard
 * library function memmove() for most cases.
 * However, where the source and/or destination references
 * an instruction RAM or ROM or a data RAM or ROM, that
 * source and/or destination will always be accessed with
 * 32-bit load and store instructions (as required for these
 * types of devices).
 *
 * !!!!!!!  XTFIXME:
 * !!!!!!!  Handling of IRAM/IROM has not yet
 * !!!!!!!  been implemented.
 *
 * The (general case) algorithm is as follows:
 *   If end of source doesn't overlap destination then use memcpy.
 *   Otherwise do memcpy backwards.
 *
 * Register use:
 *	a0/ return address
 *	a1/ stack pointer
 *	a2/ return value
 *	a3/ src
 *	a4/ length
 *	a5/ dst
 *	a6/ tmp
 *	a7/ tmp
 *	a8/ tmp
 *	a9/ tmp
 *	a10/ tmp
 *	a11/ tmp
 */

/*
 * Byte by byte copy
 */
	.align	4
	.byte	0		# 1 mod 4 alignment for LOOPNEZ
				# (0 mod 4 alignment for LBEG)
.Lbackbytecopy:
#if XCHAL_HAVE_LOOPS
	loopnez	a4, .Lbackbytecopydone
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a4, .Lbackbytecopydone
	sub	a7, a3, a4	# a7 = start address for source
#endif /* !XCHAL_HAVE_LOOPS */
.Lbacknextbyte:
	addi	a3, a3, -1
	l8ui	a6, a3, 0
	addi	a5, a5, -1
	s8i	a6, a5, 0
#if !XCHAL_HAVE_LOOPS
	bne	a3, a7, .Lbacknextbyte # continue loop if
				       # $a3:src != $a7:src_start
#endif /* !XCHAL_HAVE_LOOPS */
.Lbackbytecopydone:
	retw

/*
 * Destination is unaligned
 */

	.align	4
.Lbackdst1mod2:	# dst is only byte aligned
	_bltui	a4, 7, .Lbackbytecopy	# do short copies byte by byte

	# copy 1 byte
	addi	a3, a3, -1
	l8ui	a6, a3,  0
	addi	a5, a5, -1
	s8i	a6, a5,  0
	addi	a4, a4, -1
	_bbci.l	a5, 1, .Lbackdstaligned	# if dst is now aligned, then
					# return to main algorithm
.Lbackdst2mod4:	# dst 16-bit aligned
	# copy 2 bytes
	_bltui	a4, 6, .Lbackbytecopy	# do short copies byte by byte
	addi	a3, a3, -2
	l8ui	a6, a3,  0
	l8ui	a7, a3,  1
	addi	a5, a5, -2
	s8i	a6, a5,  0
	s8i	a7, a5,  1
	addi	a4, a4, -2
	j	.Lbackdstaligned	# dst is now aligned,
					# return to main algorithm

	.align	4
	.global	memmove
	.type   memmove,@function
memmove:

	entry	sp, 16		# minimal stack frame
	# a2/ dst, a3/ src, a4/ len
	mov	a5, a2		# copy dst so that a2 is return value
.Lmovecommon:
	sub	a6, a5, a3
	bgeu	a6, a4, .Lcommon

	add	a5, a5, a4
	add	a3, a3, a4

	_bbsi.l	a5, 0, .Lbackdst1mod2	# if dst is 1 mod 2
	_bbsi.l	a5, 1, .Lbackdst2mod4	# if dst is 2 mod 4
.Lbackdstaligned:	# return here from .Lbackdst?mod? once dst is aligned
	srli	a7, a4, 4	# number of loop iterations with 16B
				# per iteration
	movi	a8, 3		# if source is not aligned,
	_bany	a3, a8, .Lbacksrcunaligned	# then use shifting copy
	/*
	 * Destination and source are word-aligned, use word copy.
	 */
	# copy 16 bytes per iteration for word-aligned dst and word-aligned src
#if XCHAL_HAVE_LOOPS
	loopnez	a7, .backLoop1done
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a7, .backLoop1done
	slli	a8, a7, 4
	sub	a8, a3, a8	# a8 = start of first 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop1:
	addi	a3, a3, -16
	l32i	a7, a3, 12
	l32i	a6, a3,  8
	addi	a5, a5, -16
	s32i	a7, a5, 12
	l32i	a7, a3,  4
	s32i	a6, a5,  8
	l32i	a6, a3,  0
	s32i	a7, a5,  4
	s32i	a6, a5,  0
#if !XCHAL_HAVE_LOOPS
	bne	a3, a8, .backLoop1  # continue loop if a3:src != a8:src_start
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop1done:
	bbci.l	a4, 3, .Lback2
	# copy 8 bytes
	addi	a3, a3, -8
	l32i	a6, a3,  0
	l32i	a7, a3,  4
	addi	a5, a5, -8
	s32i	a6, a5,  0
	s32i	a7, a5,  4
.Lback2:
	bbsi.l	a4, 2, .Lback3
	bbsi.l	a4, 1, .Lback4
	bbsi.l	a4, 0, .Lback5
	retw
.Lback3:
	# copy 4 bytes
	addi	a3, a3, -4
	l32i	a6, a3,  0
	addi	a5, a5, -4
	s32i	a6, a5,  0
	bbsi.l	a4, 1, .Lback4
	bbsi.l	a4, 0, .Lback5
	retw
.Lback4:
	# copy 2 bytes
	addi	a3, a3, -2
	l16ui	a6, a3,  0
	addi	a5, a5, -2
	s16i	a6, a5,  0
	bbsi.l	a4, 0, .Lback5
	retw
.Lback5:
	# copy 1 byte
	addi	a3, a3, -1
	l8ui	a6, a3,  0
	addi	a5, a5, -1
	s8i	a6, a5,  0
	retw

/*
 * Destination is aligned, Source is unaligned
 */

	.align	4
.Lbacksrcunaligned:
	_beqz	a4, .Lbackdone	# avoid loading anything for zero-length copies
	# copy 16 bytes per iteration for word-aligned dst and unaligned src
	ssa8	a3		# set shift amount from byte offset
#define SIM_CHECKS_ALIGNMENT	1	/* set to 1 when running on ISS with
					 * the lint or ferret client, or 0
					 * to save a few cycles */
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
	and	a11, a3, a8	# save unalignment offset for below
	sub	a3, a3, a11	# align a3
#endif
	l32i	a6, a3, 0	# load first word
#if XCHAL_HAVE_LOOPS
	loopnez	a7, .backLoop2done
#else /* !XCHAL_HAVE_LOOPS */
	beqz	a7, .backLoop2done
	slli	a10, a7, 4
	sub	a10, a3, a10	# a10 = start of first 16B source chunk
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop2:
	addi	a3, a3, -16
	l32i	a7, a3, 12
	l32i	a8, a3,  8
	addi	a5, a5, -16
	src_b	a6, a7, a6
	s32i	a6, a5, 12
	l32i	a9, a3,  4
	src_b	a7, a8, a7
	s32i	a7, a5,  8
	l32i	a6, a3,  0
	src_b	a8, a9, a8
	s32i	a8, a5,  4
	src_b	a9, a6, a9
	s32i	a9, a5,  0
#if !XCHAL_HAVE_LOOPS
	bne	a3, a10, .backLoop2 # continue loop if a3:src != a10:src_start
#endif /* !XCHAL_HAVE_LOOPS */
.backLoop2done:
	bbci.l	a4, 3, .Lback12
	# copy 8 bytes
	addi	a3, a3, -8
	l32i	a7, a3,  4
	l32i	a8, a3,  0
	addi	a5, a5, -8
	src_b	a6, a7, a6
	s32i	a6, a5,  4
	src_b	a7, a8, a7
	s32i	a7, a5,  0
	mov	a6, a8
.Lback12:
	bbci.l	a4, 2, .Lback13
	# copy 4 bytes
	addi	a3, a3, -4
	l32i	a7, a3,  0
	addi	a5, a5, -4
	src_b	a6, a7, a6
	s32i	a6, a5,  0
	mov	a6, a7
.Lback13:
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || SIM_CHECKS_ALIGNMENT
	add	a3, a3, a11	# readjust a3 with correct misalignment
#endif
	bbsi.l	a4, 1, .Lback14
	bbsi.l	a4, 0, .Lback15
.Lbackdone:
	retw
.Lback14:
	# copy 2 bytes
	addi	a3, a3, -2
	l8ui	a6, a3,  0
	l8ui	a7, a3,  1
	addi	a5, a5, -2
	s8i	a6, a5,  0
	s8i	a7, a5,  1
	bbsi.l	a4, 0, .Lback15
	retw
.Lback15:
	# copy 1 byte
	addi	a3, a3, -1
	addi	a5, a5, -1
	l8ui	a6, a3,  0
	s8i	a6, a5,  0
	retw


/*
 * Local Variables:
 * mode:fundamental
 * comment-start: "# "
 * comment-start-skip: "# *"
 * End:
 */