about summary refs log tree commit diff
path: root/sysdeps/sparc/sparc64/sparcv9b/memcpy.S
blob: de747680c19ce42bf91f691f1691a80843b920a5 (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
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
/* Copy SIZE bytes from SRC to DEST.
   For UltraSPARC-III.
   Copyright (C) 2001 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by David S. Miller (davem@redhat.com)

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public License as
   published by the Free Software Foundation; either version 2 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
   Library General Public License for more details.

   You should have received a copy of the GNU Library General Public
   License along with the GNU C Library; see the file COPYING.LIB.  If not,
   write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
   Boston, MA 02111-1307, USA.  */

#include <sysdep.h>
#define ASI_BLK_P 0xf0
#define FPRS_FEF  0x04
#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs
#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
#define SMALL_COPY_USES_FPU

#ifndef XCC
#define USE_BPR
#define XCC xcc
#endif

	.text
	.align	32

ENTRY(bcopy)
	sub		%o1, %o0, %o4			/* IEU0		Group		*/
	mov		%o0, %g3			/* IEU1				*/
	cmp		%o4, %o2			/* IEU1		Group		*/
	mov		%o1, %o0			/* IEU0				*/
	bgeu,pt		%XCC, 100f			/* CTI				*/
	 mov		%g3, %o1			/* IEU0		Group		*/
#ifndef USE_BPR
	srl		%o2, 0, %o2			/* IEU1				*/
#endif
	brnz,pn		%o2, 220f			/* CTI		Group		*/
	 add		%o0, %o2, %o0			/* IEU0				*/
	retl
	 nop
END(bcopy)

	/* Special/non-trivial issues of this code:
	 *
	 * 1) %o5 is preserved from VISEntryHalf to VISExitHalf
	 * 2) Only low 32 FPU registers are used so that only the
	 *    lower half of the FPU register set is dirtied by this
	 *    code.  This is especially important in the kernel.
	 * 3) This code never prefetches cachelines past the end
	 *    of the source buffer.
	 *
	 * The cheetah's flexible spine, oversized liver, enlarged heart,
	 * slender muscular body, and claws make it the swiftest hunter
	 * in Africa and the fastest animal on land.  Can reach speeds
	 * of up to 2.4GB per second.
	 */
	.align		32
ENTRY(memcpy)

100: /* %o0=dst, %o1=src, %o2=len */
#ifndef __KERNEL__
	/* Save away original 'dst' for memcpy return value. */
	mov		%o0, %g3			! A0	Group
#endif
	/* Anything to copy at all? */
	cmp		%o2, 0				! A1
	ble,pn		%XCC, 102f			! BR

	/* Extremely small copy? */
218:	 cmp		%o2, 31				! A0	Group
	ble,pn		%XCC, 101f			! BR

	/* Large enough to use unrolled prefetch loops? */
	 cmp		%o2, 0x100			! A1
	bge,a,pt	%XCC, 103f			! BR	Group
	 andcc		%o0, 0x3f, %g2			! A0

	ba,pt		%XCC, 108f			! BR	Group
	 andcc		%o0, 0x7, %g2			! A0

	.align		32
101:
	/* Copy %o2 bytes from src to dst, one byte at a time. */
	ldub		[%o1 + 0x00], %o3		! MS	Group
	add		%o1, 0x1, %o1			! A0
	add		%o0, 0x1, %o0			! A1
	subcc		%o2, 1, %o2			! A0	Group

	bg,pt		%XCC, 101b			! BR
	 stb		%o3, [%o0 + -1]			! MS	Group (1-cycle stall)

102:
#ifdef __KERNEL__
	retl						! BR	Group (0-4 cycle stall)
	 clr		%o0				! A0
#else
	retl						! BR	Group (0-4 cycle stall)
	 mov		%g3, %o0			! A0
#endif

	/* Here len >= (6 * 64) and condition codes reflect execution
	 * of "andcc %o0, 0x7, %g2", done by caller.
	 */
	.align		64
103:
	/* Is 'dst' already aligned on an 64-byte boundary? */
	be,pt		%XCC, 2f			! BR

	/* Compute abs((dst & 0x3f) - 0x40) into %g2.  This is the number
	 * of bytes to copy to make 'dst' 64-byte aligned.  We pre-
	 * subtract this from 'len'.
	 */
	 sub		%g2, 0x40, %g2			! A0	Group
	sub		%g0, %g2, %g2			! A0	Group
	sub		%o2, %g2, %o2			! A0	Group

	/* Copy %g2 bytes from src to dst, one byte at a time. */
1:	ldub		[%o1 + 0x00], %o3		! MS	(Group)
	add		%o1, 0x1, %o1			! A1
	add		%o0, 0x1, %o0			! A0	Group
	subcc		%g2, 0x1, %g2			! A1

	bg,pt		%XCC, 1b			! BR	Group
	 stb		%o3, [%o0 + -1]			! MS	Group

2:	VISEntryHalf					! MS+MS
	and		%o1, 0x7, %g1			! A1
	ba,pt		%XCC, 104f			! BR
	 alignaddr	%o1, %g0, %o1			! MS	      (Break-after)

	.align		64
104:
	prefetch	[%o1 + 0x000], #one_read	! MS	Group1
	prefetch	[%o1 + 0x040], #one_read	! MS	Group2
	andn		%o2, (0x40 - 1), %o4		! A0
	prefetch	[%o1 + 0x080], #one_read	! MS	Group3
	cmp		%o4, 0x140			! A0
	prefetch	[%o1 + 0x0c0], #one_read	! MS	Group4
	ldd		[%o1 + 0x000], %f0		! MS	Group5 (%f0 results at G8)
	bge,a,pt	%XCC, 1f			! BR

	 prefetch	[%o1 + 0x100], #one_read	! MS	Group6
1:	ldd		[%o1 + 0x008], %f2		! AX	       (%f2 results at G9)
	cmp		%o4, 0x180			! A1
	bge,a,pt	%XCC, 1f			! BR
	 prefetch	[%o1 + 0x140], #one_read	! MS	Group7
1:	ldd		[%o1 + 0x010], %f4		! AX	       (%f4 results at G10)
	cmp		%o4, 0x1c0			! A1
	bge,a,pt	%XCC, 1f			! BR

	 prefetch	[%o1 + 0x180], #one_read	! MS	Group8
1:	faligndata	%f0, %f2, %f16			! FGA	Group9 (%f16 at G12)
	ldd		[%o1 + 0x018], %f6		! AX	       (%f6 results at G12)
	faligndata	%f2, %f4, %f18			! FGA	Group10 (%f18 results at G13)
	ldd		[%o1 + 0x020], %f8		! MS	        (%f8 results at G13)
	faligndata	%f4, %f6, %f20			! FGA	Group12	(1-cycle stall,%f20 at G15)
	ldd		[%o1 + 0x028], %f10		! MS		(%f10 results at G15)
	faligndata	%f6, %f8, %f22			! FGA	Group13 (%f22 results at G16)

	ldd		[%o1 + 0x030], %f12		! MS		(%f12 results at G16)
	faligndata	%f8, %f10, %f24			! FGA	Group15 (1-cycle stall,%f24 at G18)
	ldd		[%o1 + 0x038], %f14		! MS		(%f14 results at G18)
	faligndata	%f10, %f12, %f26		! FGA	Group16 (%f26 results at G19)
	ldd		[%o1 + 0x040], %f0		! MS		(%f0 results at G19)

	/* We only use the first loop if len > (7 * 64). */
	subcc		%o4, 0x1c0, %o4			! A0	Group17
	bg,pt		%XCC, 105f			! BR
	 add		%o1, 0x40, %o1			! A1

	add		%o4, 0x140, %o4			! A0	Group18
	ba,pt		%XCC, 106f			! BR
	 srl		%o4, 6, %o3			! A0	Group19
	nop
	nop
	nop
	nop
	nop

	nop
	nop

	/* This loop performs the copy and queues new prefetches.
	 * We drop into the second loop when len <= (5 * 64).  Note
	 * that this (5 * 64) factor has been subtracted from len
	 * already.
	 */
105:
	ldd		[%o1 + 0x008], %f2		! MS	Group2	(%f2 results at G5)
	faligndata	%f12, %f14, %f28		! FGA		(%f28 results at G5)
	ldd		[%o1 + 0x010], %f4		! MS	Group3	(%f4 results at G6)
	faligndata	%f14, %f0, %f30			! FGA	Group4	(1-cycle stall, %f30 at G7)
	stda		%f16, [%o0] ASI_BLK_P		! MS
	ldd		[%o1 + 0x018], %f6		! AX		(%f6 results at G7)

	faligndata	%f0, %f2, %f16			! FGA	Group12 (7-cycle stall)
	ldd		[%o1 + 0x020], %f8		! MS	        (%f8 results at G15)
	faligndata	%f2, %f4, %f18			! FGA	Group13	(%f18 results at G16)
	ldd		[%o1 + 0x028], %f10		! MS		(%f10 results at G16)
	faligndata	%f4, %f6, %f20			! FGA	Group14	(%f20 results at G17)
	ldd		[%o1 + 0x030], %f12		! MS		(%f12 results at G17)
	faligndata	%f6, %f8, %f22			! FGA	Group15	(%f22 results at G18)
	ldd		[%o1 + 0x038], %f14		! MS		(%f14 results at G18)

	faligndata	%f8, %f10, %f24			! FGA	Group16	(%f24 results at G19)
	ldd		[%o1 + 0x040], %f0		! AX		(%f0 results at G19)
	prefetch	[%o1 + 0x180], #one_read	! MS
	faligndata	%f10, %f12, %f26		! FGA	Group17	(%f26 results at G20)
	subcc		%o4, 0x40, %o4			! A0
	add		%o1, 0x40, %o1			! A1
	bg,pt		%XCC, 105b			! BR
	 add		%o0, 0x40, %o0			! A0	Group18

	mov		5, %o3				! A1

	/* This loop performs on the copy, no new prefetches are
	 * queued.  We do things this way so that we do not perform
	 * any spurious prefetches past the end of the src buffer.
	 */
106:
	ldd		[%o1 + 0x008], %f2		! MS
	faligndata	%f12, %f14, %f28		! FGA	Group2
	ldd		[%o1 + 0x010], %f4		! MS
	faligndata	%f14, %f0, %f30			! FGA	Group4	(1-cycle stall)
	stda		%f16, [%o0] ASI_BLK_P		! MS
	ldd		[%o1 + 0x018], %f6		! AX
	faligndata	%f0, %f2, %f16			! FGA	Group12 (7-cycle stall)

	ldd		[%o1 + 0x020], %f8		! MS
	faligndata	%f2, %f4, %f18			! FGA	Group13
	ldd		[%o1 + 0x028], %f10		! MS
	faligndata	%f4, %f6, %f20			! FGA	Group14
	ldd		[%o1 + 0x030], %f12		! MS
	faligndata	%f6, %f8, %f22			! FGA	Group15
	ldd		[%o1 + 0x038], %f14		! MS
	faligndata	%f8, %f10, %f24			! FGA	Group16

	ldd		[%o1 + 0x040], %f0		! AX
	faligndata	%f10, %f12, %f26		! FGA	Group17
	subcc		%o3, 0x01, %o3			! A0
	add		%o1, 0x40, %o1			! A1
	bg,pt		%XCC, 106b			! BR
	 add		%o0, 0x40, %o0			! A0	Group18

	/* Finally we copy the last full 64-byte block. */
	ldd		[%o1 + 0x008], %f2		! MS
	faligndata	%f12, %f14, %f28		! FGA
	ldd		[%o1 + 0x010], %f4		! MS	Group19
	faligndata	%f14, %f0, %f30			! FGA
	stda		%f16, [%o0] ASI_BLK_P		! MS	Group20
	ldd		[%o1 + 0x018], %f6		! AX
	faligndata	%f0, %f2, %f16			! FGA	Group11 (7-cycle stall)
	ldd		[%o1 + 0x020], %f8		! MS
	faligndata	%f2, %f4, %f18			! FGA	Group12
	ldd		[%o1 + 0x028], %f10		! MS
	faligndata	%f4, %f6, %f20			! FGA	Group13
	ldd		[%o1 + 0x030], %f12		! MS
	faligndata	%f6, %f8, %f22			! FGA	Group14
	ldd		[%o1 + 0x038], %f14		! MS
	faligndata	%f8, %f10, %f24			! FGA	Group15
	cmp		%g1, 0				! A0
	be,pt		%XCC, 1f			! BR
	 add		%o0, 0x40, %o0			! A1
	ldd		[%o1 + 0x040], %f0		! MS
1:	faligndata	%f10, %f12, %f26		! FGA	Group16
	faligndata	%f12, %f14, %f28		! FGA	Group17
	faligndata	%f14, %f0, %f30			! FGA	Group18
	stda		%f16, [%o0] ASI_BLK_P		! MS
	add		%o0, 0x40, %o0			! A0
	add		%o1, 0x40, %o1			! A1
	membar		#Sync				! MS	Group26 (7-cycle stall)

	/* Now we copy the (len modulo 64) bytes at the end.
	 * Note how we borrow the %f0 loaded above.
	 *
	 * Also notice how this code is careful not to perform a
	 * load past the end of the src buffer just like similar
	 * code found in 'toosmall' processing.
	 */
	and		%o2, 0x3f, %o2			! A0	Group
	andcc		%o2, 0x38, %g2			! A0	Group
	be,pn		%XCC, 107f			! BR
	 subcc		%g2, 0x8, %g2			! A1
	be,pn		%XCC, 107f			! BR	Group
	 cmp		%g1, 0				! A0

	be,a,pt		%XCC, 1f			! BR	Group
	 ldd		[%o1 + 0x00], %f0		! MS

1:	ldd		[%o1 + 0x08], %f2		! MS	Group
	add		%o1, 0x8, %o1			! A0
	sub		%o2, 0x8, %o2			! A1
	subcc		%g2, 0x8, %g2			! A0	Group
	faligndata	%f0, %f2, %f8			! FGA	Group
	std		%f8, [%o0 + 0x00]		! MS	(XXX does it stall here? XXX)
	be,pn		%XCC, 107f			! BR
	 add		%o0, 0x8, %o0			! A0
	ldd		[%o1 + 0x08], %f0		! MS	Group
	add		%o1, 0x8, %o1			! A0
	sub		%o2, 0x8, %o2			! A1
	subcc		%g2, 0x8, %g2			! A0	Group
	faligndata	%f2, %f0, %f8			! FGA
	std		%f8, [%o0 + 0x00]		! MS	(XXX does it stall here? XXX)
	bne,pn		%XCC, 1b			! BR
	 add		%o0, 0x8, %o0			! A0	Group

	/* If anything is left, we copy it one byte at a time.
	 * Note that %g1 is (src & 0x3) saved above before the
	 * alignaddr was performed.
	 */
107:
	cmp		%o2, 0
	add		%o1, %g1, %o1
	VISExitHalf
	be,pn		%XCC, 102b
	 nop
	ba,a,pt		%XCC, 101b

	/* If we get here, then 32 <= len < (6 * 64) */
108:

#ifdef SMALL_COPY_USES_FPU

	/* Is 'dst' already aligned on an 8-byte boundary? */
	be,pt		%XCC, 2f			! BR	Group

	/* Compute abs((dst & 7) - 8) into %g2.  This is the number
	 * of bytes to copy to make 'dst' 8-byte aligned.  We pre-
	 * subtract this from 'len'.
	 */
	 sub		%g2, 0x8, %g2			! A0
	sub		%g0, %g2, %g2			! A0	Group (reg-dep)
	sub		%o2, %g2, %o2			! A0	Group (reg-dep)

	/* Copy %g2 bytes from src to dst, one byte at a time. */
1:	ldub		[%o1 + 0x00], %o3		! MS	(Group) (%o3 in 3 cycles)
	add		%o1, 0x1, %o1			! A1
	add		%o0, 0x1, %o0			! A0	Group
	subcc		%g2, 0x1, %g2			! A1

	bg,pt		%XCC, 1b			! BR	Group
	 stb		%o3, [%o0 + -1]			! MS	Group

2:	VISEntryHalf					! MS+MS

	/* Compute (len - (len % 8)) into %g2.  This is guarenteed
	 * to be nonzero.
	 */
	andn		%o2, 0x7, %g2			! A0	Group

	/* You may read this and believe that it allows reading
	 * one 8-byte longword past the end of src.  It actually
	 * does not, as %g2 is subtracted as loads are done from
	 * src, so we always stop before running off the end.
	 * Also, we are guarenteed to have at least 0x10 bytes
	 * to move here.
	 */
	sub		%g2, 0x8, %g2			! A0	Group (reg-dep)
	alignaddr	%o1, %g0, %g1			! MS	      (Break-after)
	ldd		[%g1 + 0x00], %f0		! MS	Group (1-cycle stall)
	add		%g1, 0x8, %g1			! A0

1:	ldd		[%g1 + 0x00], %f2		! MS	Group
	add		%g1, 0x8, %g1			! A0
	sub		%o2, 0x8, %o2			! A1
	subcc		%g2, 0x8, %g2			! A0	Group

	faligndata	%f0, %f2, %f8			! FGA	Group (1-cycle stall)
	std		%f8, [%o0 + 0x00]		! MS	Group (2-cycle stall)
	add		%o1, 0x8, %o1			! A0
	be,pn		%XCC, 2f			! BR

	 add		%o0, 0x8, %o0			! A1
	ldd		[%g1 + 0x00], %f0		! MS	Group
	add		%g1, 0x8, %g1			! A0
	sub		%o2, 0x8, %o2			! A1

	subcc		%g2, 0x8, %g2			! A0	Group
	faligndata	%f2, %f0, %f8			! FGA	Group (1-cycle stall)
	std		%f8, [%o0 + 0x00]		! MS	Group (2-cycle stall)
	add		%o1, 0x8, %o1			! A0

	bne,pn		%XCC, 1b			! BR
	 add		%o0, 0x8, %o0			! A1

	/* Nothing left to copy? */
2:	cmp		%o2, 0				! A0	Group
	VISExitHalf					! A0+MS
	be,pn		%XCC, 102b			! BR	Group
	 nop						! A0
	ba,a,pt		%XCC, 101b			! BR	Group

#else /* !(SMALL_COPY_USES_FPU) */

	xor		%o1, %o0, %g2
	andcc		%g2, 0x7, %g0
	bne,pn		%XCC, 101b
	 andcc		%o1, 0x7, %g2

	be,pt		%XCC, 2f
	 sub		%g2, 0x8, %g2
	sub		%g0, %g2, %g2
	sub		%o2, %g2, %o2

1:	ldub		[%o1 + 0x00], %o3
	add		%o1, 0x1, %o1
	add		%o0, 0x1, %o0
	subcc		%g2, 0x1, %g2
	bg,pt		%XCC, 1b
	 stb		%o3, [%o0 + -1]

2:	andn		%o2, 0x7, %g2
	sub		%o2, %g2, %o2

3:	ldx		[%o1 + 0x00], %o3
	add		%o1, 0x8, %o1
	add		%o0, 0x8, %o0
	subcc		%g2, 0x8, %g2
	bg,pt		%XCC, 3b
	 stx		%o3, [%o0 + -8]

	cmp		%o2, 0
	bne,pn		%XCC, 101b
	 nop
	ba,a,pt		%XCC, 102b

#endif /* !(SMALL_COPY_USES_FPU) */
END(memcpy)

#define RMOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3)	\
	ldx		[%src - offset - 0x20], %t0; 		\
	ldx		[%src - offset - 0x18], %t1; 		\
	ldx		[%src - offset - 0x10], %t2; 		\
	ldx		[%src - offset - 0x08], %t3; 		\
	stw		%t0, [%dst - offset - 0x1c]; 		\
	srlx		%t0, 32, %t0;				\
	stw		%t0, [%dst - offset - 0x20]; 		\
	stw		%t1, [%dst - offset - 0x14]; 		\
	srlx		%t1, 32, %t1;				\
	stw		%t1, [%dst - offset - 0x18]; 		\
	stw		%t2, [%dst - offset - 0x0c]; 		\
	srlx		%t2, 32, %t2;				\
	stw		%t2, [%dst - offset - 0x10]; 		\
	stw		%t3, [%dst - offset - 0x04];		\
	srlx		%t3, 32, %t3;				\
	stw		%t3, [%dst - offset - 0x08];

#define RMOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3)	\
	ldx		[%src - offset - 0x20], %t0; 		\
	ldx		[%src - offset - 0x18], %t1; 		\
	ldx		[%src - offset - 0x10], %t2; 		\
	ldx		[%src - offset - 0x08], %t3; 		\
	stx		%t0, [%dst - offset - 0x20]; 		\
	stx		%t1, [%dst - offset - 0x18]; 		\
	stx		%t2, [%dst - offset - 0x10]; 		\
	stx		%t3, [%dst - offset - 0x08];		\
	ldx		[%src - offset - 0x40], %t0; 		\
	ldx		[%src - offset - 0x38], %t1; 		\
	ldx		[%src - offset - 0x30], %t2; 		\
	ldx		[%src - offset - 0x28], %t3; 		\
	stx		%t0, [%dst - offset - 0x40]; 		\
	stx		%t1, [%dst - offset - 0x38]; 		\
	stx		%t2, [%dst - offset - 0x30]; 		\
	stx		%t3, [%dst - offset - 0x28];

#define RMOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3)	\
	ldx		[%src + offset + 0x00], %t0;		\
	ldx		[%src + offset + 0x08], %t1; 		\
	stw		%t0, [%dst + offset + 0x04]; 		\
	srlx		%t0, 32, %t2;				\
	stw		%t2, [%dst + offset + 0x00]; 		\
	stw		%t1, [%dst + offset + 0x0c]; 		\
	srlx		%t1, 32, %t3;				\
	stw		%t3, [%dst + offset + 0x08];

#define RMOVE_LASTALIGNCHUNK(src, dst, offset, t0, t1)		\
	ldx		[%src + offset + 0x00], %t0; 		\
	ldx		[%src + offset + 0x08], %t1; 		\
	stx		%t0, [%dst + offset + 0x00]; 		\
	stx		%t1, [%dst + offset + 0x08];

	.align		32
228:	andcc		%o2, 1, %g0			/* IEU1		Group		*/
	be,pt		%icc, 2f+4			/* CTI				*/
1:	 ldub		[%o1 - 1], %o5			/* LOAD		Group		*/
	sub		%o1, 1, %o1			/* IEU0				*/
	sub		%o0, 1, %o0			/* IEU1				*/
	subcc		%o2, 1, %o2			/* IEU1		Group		*/
	be,pn		%xcc, 229f			/* CTI				*/
	 stb		%o5, [%o0]			/* Store			*/
2:	ldub		[%o1 - 1], %o5			/* LOAD		Group		*/
	sub		%o0, 2, %o0			/* IEU0				*/
	ldub		[%o1 - 2], %g5			/* LOAD		Group		*/
	sub		%o1, 2, %o1			/* IEU0				*/
	subcc		%o2, 2, %o2			/* IEU1		Group		*/
	stb		%o5, [%o0 + 1]			/* Store			*/
	bne,pt		%xcc, 2b			/* CTI				*/
	 stb		%g5, [%o0]			/* Store			*/
229:	retl
	 mov		%g4, %o0

	.align		32
ENTRY(memmove)
	mov		%o0, %g3
#ifndef USE_BPR
	srl		%o2, 0, %o2			/* IEU1		Group		*/
#endif
	brz,pn		%o2, 102b			/* CTI		Group		*/
	 sub		%o0, %o1, %o4			/* IEU0				*/
	cmp		%o4, %o2			/* IEU1		Group		*/
	bgeu,pt		%XCC, 218b			/* CTI				*/
	 mov		%o0, %g4			/* IEU0				*/
	add		%o0, %o2, %o0			/* IEU0		Group		*/
220:	add		%o1, %o2, %o1			/* IEU1				*/
	cmp		%o2, 15				/* IEU1		Group		*/
	bleu,pn		%xcc, 228b			/* CTI				*/
	 andcc		%o0, 7, %g2			/* IEU1		Group		*/
	sub		%o0, %o1, %g5			/* IEU0				*/
	andcc		%g5, 3, %o5			/* IEU1		Group		*/
	bne,pn		%xcc, 232f			/* CTI				*/
	 andcc		%o1, 3, %g0			/* IEU1		Group		*/
	be,a,pt		%xcc, 236f			/* CTI				*/
	 andcc		%o1, 4, %g0			/* IEU1		Group		*/
	andcc		%o1, 1, %g0			/* IEU1		Group		*/
	be,pn		%xcc, 4f			/* CTI				*/
	 andcc		%o1, 2, %g0			/* IEU1		Group		*/
	ldub		[%o1 - 1], %g2			/* Load		Group		*/
	sub		%o1, 1, %o1			/* IEU0				*/
	sub		%o0, 1, %o0			/* IEU1				*/
	sub		%o2, 1, %o2			/* IEU0		Group		*/
	be,pn		%xcc, 5f			/* CTI		Group		*/
	 stb		%g2, [%o0]			/* Store			*/
4:	lduh		[%o1 - 2], %g2			/* Load		Group		*/
	sub		%o1, 2, %o1			/* IEU0				*/
	sub		%o0, 2, %o0			/* IEU1				*/
	sub		%o2, 2, %o2			/* IEU0				*/
	sth		%g2, [%o0]			/* Store	Group + bubble	*/
5:	andcc		%o1, 4, %g0			/* IEU1				*/
236:	be,a,pn		%xcc, 2f			/* CTI				*/
	 andcc		%o2, -128, %g7			/* IEU1		Group		*/
	lduw		[%o1 - 4], %g5			/* Load		Group		*/
	sub		%o1, 4, %o1			/* IEU0				*/
	sub		%o0, 4, %o0			/* IEU1				*/
	sub		%o2, 4, %o2			/* IEU0		Group		*/
	stw		%g5, [%o0]			/* Store			*/
	andcc		%o2, -128, %g7			/* IEU1		Group		*/
2:	be,pn		%xcc, 235f			/* CTI				*/
	 andcc		%o0, 4, %g0			/* IEU1		Group		*/
	be,pn		%xcc, 282f + 4			/* CTI		Group		*/
5:	RMOVE_BIGCHUNK(o1, o0, 0x00, g1, g3, g5, o5)
	RMOVE_BIGCHUNK(o1, o0, 0x20, g1, g3, g5, o5)
	RMOVE_BIGCHUNK(o1, o0, 0x40, g1, g3, g5, o5)
	RMOVE_BIGCHUNK(o1, o0, 0x60, g1, g3, g5, o5)
	subcc		%g7, 128, %g7			/* IEU1		Group		*/
	sub		%o1, 128, %o1			/* IEU0				*/
	bne,pt		%xcc, 5b			/* CTI				*/
	 sub		%o0, 128, %o0			/* IEU0		Group		*/
235:	andcc		%o2, 0x70, %g7			/* IEU1		Group		*/
41:	be,pn		%xcc, 280f			/* CTI				*/
	 andcc		%o2, 8, %g0			/* IEU1		Group		*/
							/* Clk1 8-(			*/
							/* Clk2 8-(			*/
							/* Clk3 8-(			*/
							/* Clk4 8-(			*/
279:	rd		%pc, %o5			/* PDU		Group		*/
	sll		%g7, 1, %g5			/* IEU0		Group		*/
	sub		%o1, %g7, %o1			/* IEU1				*/
	sub		%o5, %g5, %o5			/* IEU0  	Group		*/
	jmpl		%o5 + %lo(280f - 279b), %g0	/* CTI		Group brk forced*/
	 sub		%o0, %g7, %o0			/* IEU0		Group		*/
	RMOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g5, o5)
	RMOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g5, o5)
	RMOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g5, o5)
	RMOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g5, o5)
	RMOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g5, o5)
	RMOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g5, o5)
	RMOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g5, o5)
280:	be,pt		%xcc, 281f			/* CTI				*/
	 andcc		%o2, 4, %g0			/* IEU1				*/
	ldx		[%o1 - 8], %g2			/* Load		Group		*/
	sub		%o0, 8, %o0			/* IEU0				*/
	stw		%g2, [%o0 + 4]			/* Store	Group		*/
	sub		%o1, 8, %o1			/* IEU1				*/
	srlx		%g2, 32, %g2			/* IEU0		Group		*/
	stw		%g2, [%o0]			/* Store			*/
281:	be,pt		%xcc, 1f			/* CTI				*/
	 andcc		%o2, 2, %g0			/* IEU1		Group		*/
	lduw		[%o1 - 4], %g2			/* Load		Group		*/
	sub		%o1, 4, %o1			/* IEU0				*/
	stw		%g2, [%o0 - 4]			/* Store	Group		*/
	sub		%o0, 4, %o0			/* IEU0				*/
1:	be,pt		%xcc, 1f			/* CTI				*/
	 andcc		%o2, 1, %g0			/* IEU1		Group		*/
	lduh		[%o1 - 2], %g2			/* Load		Group		*/
	sub		%o1, 2, %o1			/* IEU0				*/
	sth		%g2, [%o0 - 2]			/* Store	Group		*/
	sub		%o0, 2, %o0			/* IEU0				*/
1:	be,pt		%xcc, 211f			/* CTI				*/
	 nop						/* IEU1				*/
	ldub		[%o1 - 1], %g2			/* Load		Group		*/
	stb		%g2, [%o0 - 1]			/* Store	Group + bubble	*/
211:	retl
	 mov		%g4, %o0

282:	RMOVE_BIGALIGNCHUNK(o1, o0, 0x00, g1, g3, g5, o5)
	RMOVE_BIGALIGNCHUNK(o1, o0, 0x40, g1, g3, g5, o5)
	subcc		%g7, 128, %g7			/* IEU1		Group		*/
	sub		%o1, 128, %o1			/* IEU0				*/
	bne,pt		%xcc, 282b			/* CTI				*/
	 sub		%o0, 128, %o0			/* IEU0		Group		*/
	andcc		%o2, 0x70, %g7			/* IEU1				*/
	be,pn		%xcc, 284f			/* CTI				*/
	 andcc		%o2, 8, %g0			/* IEU1		Group		*/
							/* Clk1 8-(			*/
							/* Clk2 8-(			*/
							/* Clk3 8-(			*/
							/* Clk4 8-(			*/
283:	rd		%pc, %o5			/* PDU		Group		*/
	sub		%o1, %g7, %o1			/* IEU0		Group		*/
	sub		%o5, %g7, %o5			/* IEU1				*/
	jmpl		%o5 + %lo(284f - 283b), %g0	/* CTI		Group brk forced*/
	 sub		%o0, %g7, %o0			/* IEU0		Group		*/
	RMOVE_LASTALIGNCHUNK(o1, o0, 0x60, g2, g3)
	RMOVE_LASTALIGNCHUNK(o1, o0, 0x50, g2, g3)
	RMOVE_LASTALIGNCHUNK(o1, o0, 0x40, g2, g3)
	RMOVE_LASTALIGNCHUNK(o1, o0, 0x30, g2, g3)
	RMOVE_LASTALIGNCHUNK(o1, o0, 0x20, g2, g3)
	RMOVE_LASTALIGNCHUNK(o1, o0, 0x10, g2, g3)
	RMOVE_LASTALIGNCHUNK(o1, o0, 0x00, g2, g3)
284:	be,pt		%xcc, 285f			/* CTI		Group		*/
	 andcc		%o2, 4, %g0			/* IEU1				*/
	ldx		[%o1 - 8], %g2			/* Load		Group		*/
	sub		%o0, 8, %o0			/* IEU0				*/
	sub		%o1, 8, %o1			/* IEU0		Group		*/
	stx		%g2, [%o0]			/* Store			*/
285:	be,pt		%xcc, 1f			/* CTI				*/
	 andcc		%o2, 2, %g0			/* IEU1		Group		*/
	lduw		[%o1 - 4], %g2			/* Load		Group		*/
	sub		%o0, 4, %o0			/* IEU0				*/
	sub		%o1, 4, %o1			/* IEU0		Group		*/
	stw		%g2, [%o0]			/* Store			*/
1:	be,pt		%xcc, 1f			/* CTI				*/
	 andcc		%o2, 1, %g0			/* IEU1		Group		*/
	lduh		[%o1 - 2], %g2			/* Load		Group		*/
	sub		%o0, 2, %o0			/* IEU0				*/
	sub		%o1, 2, %o1			/* IEU0		Group		*/
	sth		%g2, [%o0]			/* Store			*/
1:	be,pt		%xcc, 1f			/* CTI				*/
	 nop						/* IEU0		Group		*/
	ldub		[%o1 - 1], %g2			/* Load		Group		*/
	stb		%g2, [%o0 - 1]			/* Store	Group + bubble	*/
1:	retl
	 mov		%g4, %o0

232:	brz,pt		%g2, 2f				/* CTI		Group		*/
	 sub		%o2, %g2, %o2			/* IEU0		Group		*/
1:	ldub		[%o1 - 1], %g5			/* Load		Group		*/
	sub		%o1, 1, %o1			/* IEU0				*/
	sub		%o0, 1, %o0			/* IEU1				*/
	subcc		%g2, 1, %g2			/* IEU1		Group		*/
	bne,pt		%xcc, 1b			/* CTI				*/
	 stb		%g5, [%o0]			/* Store			*/
2:	andn		%o2, 7, %g5 			/* IEU0		Group		*/
	and		%o2, 7, %o2			/* IEU1				*/
	fmovd		%f0, %f2			/* FPU				*/
	alignaddr	%o1, %g0, %g1			/* GRU		Group		*/
	ldd		[%g1], %f4			/* Load		Group		*/
1:	ldd		[%g1 - 8], %f6			/* Load		Group		*/
	sub		%g1, 8, %g1			/* IEU0		Group		*/
	subcc		%g5, 8, %g5			/* IEU1				*/
	faligndata	%f6, %f4, %f0			/* GRU		Group		*/
	std		%f0, [%o0 - 8]			/* Store			*/
	sub		%o1, 8, %o1			/* IEU0		Group		*/
	be,pn		%xcc, 233f			/* CTI				*/
	 sub		%o0, 8, %o0			/* IEU1				*/
	ldd		[%g1 - 8], %f4			/* Load		Group		*/
	sub		%g1, 8, %g1			/* IEU0				*/
	subcc		%g5, 8, %g5			/* IEU1				*/
	faligndata	%f4, %f6, %f0			/* GRU		Group		*/
	std		%f0, [%o0 - 8]			/* Store			*/
	sub		%o1, 8, %o1			/* IEU0				*/
	bne,pn		%xcc, 1b			/* CTI		Group		*/
	 sub		%o0, 8, %o0			/* IEU0				*/
233:	brz,pn		%o2, 234f			/* CTI		Group		*/
	 nop						/* IEU0				*/
237:	ldub		[%o1 - 1], %g5			/* LOAD				*/
	sub		%o1, 1, %o1			/* IEU0				*/
	sub		%o0, 1, %o0			/* IEU1				*/
	subcc		%o2, 1, %o2			/* IEU1				*/
	bne,pt		%xcc, 237b			/* CTI				*/
	 stb		%g5, [%o0]			/* Store	Group		*/
234:	wr		%g0, FPRS_FEF, %fprs
	retl
	 mov		%g4, %o0
END(memmove)

#ifdef USE_BPR
weak_alias(memcpy, __align_cpy_1)
weak_alias(memcpy, __align_cpy_2)
weak_alias(memcpy, __align_cpy_4)
weak_alias(memcpy, __align_cpy_8)
weak_alias(memcpy, __align_cpy_16)
#endif