summaryrefslogtreecommitdiffstats
path: root/arch/sparc/lib/M7memcpy.S
blob: 2c88175d3abffa84fde9d6893fad5494b7ba6747 (plain)
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
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
/*
 * M7memcpy: Optimized SPARC M7 memcpy
 *
 * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
 */

	.file	"M7memcpy.S"

/*
 * memcpy(s1, s2, len)
 *
 * Copy s2 to s1, always copy n bytes.
 * Note: this C code does not work for overlapped copies.
 *
 * Fast assembler language version of the following C-program for memcpy
 * which represents the `standard' for the C-library.
 *
 *	void *
 *	memcpy(void *s, const void *s0, size_t n)
 *	{
 *		if (n != 0) {
 *		    char *s1 = s;
 *		    const char *s2 = s0;
 *		    do {
 *			*s1++ = *s2++;
 *		    } while (--n != 0);
 *		}
 *		return (s);
 *	}
 *
 *
 * SPARC T7/M7 Flow :
 *
 * if (count < SMALL_MAX) {
 *   if count < SHORTCOPY              (SHORTCOPY=3)
 *	copy bytes; exit with dst addr
 *   if src & dst aligned on word boundary but not long word boundary,
 *     copy with ldw/stw; branch to finish_up
 *   if src & dst aligned on long word boundary
 *     copy with ldx/stx; branch to finish_up
 *   if src & dst not aligned and length <= SHORTCHECK   (SHORTCHECK=14)
 *     copy bytes; exit with dst addr
 *   move enough bytes to get src to word boundary
 *   if dst now on word boundary
 * move_words:
 *     copy words; branch to finish_up
 *   if dst now on half word boundary
 *     load words, shift half words, store words; branch to finish_up
 *   if dst on byte 1
 *     load words, shift 3 bytes, store words; branch to finish_up
 *   if dst on byte 3
 *     load words, shift 1 byte, store words; branch to finish_up
 * finish_up:
 *     copy bytes; exit with dst addr
 * } else {                                         More than SMALL_MAX bytes
 *   move bytes until dst is on long word boundary
 *   if( src is on long word boundary ) {
 *     if (count < MED_MAX) {
 * finish_long:					   src/dst aligned on 8 bytes
 *       copy with ldx/stx in 8-way unrolled loop;
 *       copy final 0-63 bytes; exit with dst addr
 *     } else {				     src/dst aligned; count > MED_MAX
 *       align dst on 64 byte boundary; for main data movement:
 *       prefetch src data to L2 cache; let HW prefetch move data to L1 cache
 *       Use BIS (block initializing store) to avoid copying store cache
 *       lines from memory. But pre-store first element of each cache line
 *       ST_CHUNK lines in advance of the rest of that cache line. That
 *       gives time for replacement cache lines to be written back without
 *       excess STQ and Miss Buffer filling. Repeat until near the end,
 *       then finish up storing before going to finish_long.
 *     }
 *   } else {                                   src/dst not aligned on 8 bytes
 *     if src is word aligned and count < MED_WMAX
 *       move words in 8-way unrolled loop
 *       move final 0-31 bytes; exit with dst addr
 *     if count < MED_UMAX
 *       use alignaddr/faligndata combined with ldd/std in 8-way
 *       unrolled loop to move data.
 *       go to unalign_done
 *     else
 *       setup alignaddr for faligndata instructions
 *       align dst on 64 byte boundary; prefetch src data to L1 cache
 *       loadx8, falign, block-store, prefetch loop
 *	 (only use block-init-store when src/dst on 8 byte boundaries.)
 * unalign_done:
 *       move remaining bytes for unaligned cases. exit with dst addr.
 * }
 *
 */

#include <asm/visasm.h>
#include <asm/asi.h>

#if !defined(EX_LD) && !defined(EX_ST)
#define NON_USER_COPY
#endif

#ifndef EX_LD
#define EX_LD(x)	x
#endif
#ifndef EX_LD_FP
#define EX_LD_FP(x)	x
#endif

#ifndef EX_ST
#define EX_ST(x)	x
#endif
#ifndef EX_ST_FP
#define EX_ST_FP(x)	x
#endif

#ifndef EX_RETVAL
#define EX_RETVAL(x)    x
#endif

#ifndef LOAD
#define LOAD(type,addr,dest)	type [addr], dest
#endif

#ifndef STORE
#define STORE(type,src,addr)	type src, [addr]
#endif

/*
 * ASI_BLK_INIT_QUAD_LDD_P/ASI_BLK_INIT_QUAD_LDD_S marks the cache
 * line as "least recently used" which means if many threads are
 * active, it has a high probability of being pushed out of the cache
 * between the first initializing store and the final stores.
 * Thus, we use ASI_ST_BLKINIT_MRU_P/ASI_ST_BLKINIT_MRU_S which
 * marks the cache line as "most recently used" for all
 * but the last cache line
 */
#ifndef STORE_ASI
#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
#define STORE_ASI	ASI_BLK_INIT_QUAD_LDD_P
#else
#define STORE_ASI	0x80		/* ASI_P */
#endif
#endif

#ifndef STORE_MRU_ASI
#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
#define STORE_MRU_ASI	ASI_ST_BLKINIT_MRU_P
#else
#define STORE_MRU_ASI	0x80		/* ASI_P */
#endif
#endif

#ifndef STORE_INIT
#define STORE_INIT(src,addr)	stxa src, [addr] STORE_ASI
#endif

#ifndef STORE_INIT_MRU
#define STORE_INIT_MRU(src,addr)	stxa src, [addr] STORE_MRU_ASI
#endif

#ifndef FUNC_NAME
#define FUNC_NAME	M7memcpy
#endif

#ifndef PREAMBLE
#define PREAMBLE
#endif

#define	BLOCK_SIZE	64
#define	SHORTCOPY	3
#define	SHORTCHECK	14
#define	SHORT_LONG	64	/* max copy for short longword-aligned case */
				/* must be at least 64 */
#define	SMALL_MAX	128
#define	MED_UMAX	1024	/* max copy for medium un-aligned case */
#define	MED_WMAX	1024	/* max copy for medium word-aligned case */
#define	MED_MAX		1024	/* max copy for medium longword-aligned case */
#define ST_CHUNK	24	/* ST_CHUNK - block of values for BIS Store */
#define ALIGN_PRE	24	/* distance for aligned prefetch loop */

	.register	%g2,#scratch

	.section	".text"
	.global		FUNC_NAME
	.type		FUNC_NAME, #function
	.align		16
FUNC_NAME:
	srlx            %o2, 31, %g2
	cmp             %g2, 0
	tne             %xcc, 5
	PREAMBLE
	mov		%o0, %g1	! save %o0
	brz,pn          %o2, .Lsmallx
	 cmp            %o2, 3
	ble,pn          %icc, .Ltiny_cp
	 cmp            %o2, 19
	ble,pn          %icc, .Lsmall_cp
	 or             %o0, %o1, %g2
	cmp             %o2, SMALL_MAX
	bl,pn           %icc, .Lmedium_cp
	 nop

.Lmedium:
	neg	%o0, %o5
	andcc	%o5, 7, %o5		! bytes till DST 8 byte aligned
	brz,pt	%o5, .Ldst_aligned_on_8

	! %o5 has the bytes to be written in partial store.
	 sub	%o2, %o5, %o2
	sub	%o1, %o0, %o1		! %o1 gets the difference
7:					! dst aligning loop
	add	%o1, %o0, %o4
	EX_LD(LOAD(ldub, %o4, %o4))	! load one byte
	subcc	%o5, 1, %o5
	EX_ST(STORE(stb, %o4, %o0))
	bgu,pt	%xcc, 7b
	 add	%o0, 1, %o0		! advance dst
	add	%o1, %o0, %o1		! restore %o1
.Ldst_aligned_on_8:
	andcc	%o1, 7, %o5
	brnz,pt	%o5, .Lsrc_dst_unaligned_on_8
	 nop

.Lsrc_dst_aligned_on_8:
	! check if we are copying MED_MAX or more bytes
	set MED_MAX, %o3
	cmp %o2, %o3 			! limit to store buffer size
	bgu,pn	%xcc, .Llarge_align8_copy
	 nop

/*
 * Special case for handling when src and dest are both long word aligned
 * and total data to move is less than MED_MAX bytes
 */
.Lmedlong:
	subcc	%o2, 63, %o2		! adjust length to allow cc test
	ble,pn	%xcc, .Lmedl63		! skip big loop if less than 64 bytes
	 nop
.Lmedl64:
	EX_LD(LOAD(ldx, %o1, %o4))	! load
	subcc	%o2, 64, %o2		! decrement length count
	EX_ST(STORE(stx, %o4, %o0))	! and store
	EX_LD(LOAD(ldx, %o1+8, %o3))	! a block of 64 bytes
	EX_ST(STORE(stx, %o3, %o0+8))
	EX_LD(LOAD(ldx, %o1+16, %o4))
	EX_ST(STORE(stx, %o4, %o0+16))
	EX_LD(LOAD(ldx, %o1+24, %o3))
	EX_ST(STORE(stx, %o3, %o0+24))
	EX_LD(LOAD(ldx, %o1+32, %o4))	! load
	EX_ST(STORE(stx, %o4, %o0+32))	! and store
	EX_LD(LOAD(ldx, %o1+40, %o3))	! a block of 64 bytes
	add	%o1, 64, %o1		! increase src ptr by 64
	EX_ST(STORE(stx, %o3, %o0+40))
	EX_LD(LOAD(ldx, %o1-16, %o4))
	add	%o0, 64, %o0		! increase dst ptr by 64
	EX_ST(STORE(stx, %o4, %o0-16))
	EX_LD(LOAD(ldx, %o1-8, %o3))
	bgu,pt	%xcc, .Lmedl64		! repeat if at least 64 bytes left
	 EX_ST(STORE(stx, %o3, %o0-8))
.Lmedl63:
	addcc	%o2, 32, %o2		! adjust remaining count
	ble,pt	%xcc, .Lmedl31		! to skip if 31 or fewer bytes left
	 nop
	EX_LD(LOAD(ldx, %o1, %o4))	! load
	sub	%o2, 32, %o2		! decrement length count
	EX_ST(STORE(stx, %o4, %o0))	! and store
	EX_LD(LOAD(ldx, %o1+8, %o3))	! a block of 32 bytes
	add	%o1, 32, %o1		! increase src ptr by 32
	EX_ST(STORE(stx, %o3, %o0+8))
	EX_LD(LOAD(ldx, %o1-16, %o4))
	add	%o0, 32, %o0		! increase dst ptr by 32
	EX_ST(STORE(stx, %o4, %o0-16))
	EX_LD(LOAD(ldx, %o1-8, %o3))
	EX_ST(STORE(stx, %o3, %o0-8))
.Lmedl31:
	addcc	%o2, 16, %o2		! adjust remaining count
	ble,pt	%xcc, .Lmedl15		! skip if 15 or fewer bytes left
	 nop				!
	EX_LD(LOAD(ldx, %o1, %o4))
	add	%o1, 16, %o1		! increase src ptr by 16
	EX_ST(STORE(stx, %o4, %o0))
	sub	%o2, 16, %o2		! decrease count by 16
	EX_LD(LOAD(ldx, %o1-8, %o3))
	add	%o0, 16, %o0		! increase dst ptr by 16
	EX_ST(STORE(stx, %o3, %o0-8))
.Lmedl15:
	addcc	%o2, 15, %o2		! restore count
	bz,pt	%xcc, .Lsmallx	! exit if finished
	 cmp	%o2, 8
	blt,pt	%xcc, .Lmedw7		! skip if 7 or fewer bytes left
	 tst	%o2
	EX_LD(LOAD(ldx, %o1, %o4))	! load 8 bytes
	add	%o1, 8, %o1		! increase src ptr by 8
	add	%o0, 8, %o0		! increase dst ptr by 8
	subcc	%o2, 8, %o2		! decrease count by 8
	bnz,pn	%xcc, .Lmedw7
	EX_ST(STORE(stx, %o4, %o0-8))	! and store 8 bytes
	retl
	 mov	EX_RETVAL(%g1), %o0	! restore %o0

	.align 16
.Lsrc_dst_unaligned_on_8:
	! DST is 8-byte aligned, src is not
2:
	andcc	%o1, 0x3, %o5		! test word alignment
	bnz,pt	%xcc, .Lunalignsetup	! branch to skip if not word aligned
	 nop

/*
 * Handle all cases where src and dest are aligned on word
 * boundaries. Use unrolled loops for better performance.
 * This option wins over standard large data move when
 * source and destination is in cache for.Lmedium
 * to short data moves.
 */
	set MED_WMAX, %o3
	cmp %o2, %o3 			! limit to store buffer size
	bge,pt	%xcc, .Lunalignrejoin	! otherwise rejoin main loop
	 nop

	subcc	%o2, 31, %o2		! adjust length to allow cc test
					! for end of loop
	ble,pt	%xcc, .Lmedw31		! skip big loop if less than 16
.Lmedw32:
	EX_LD(LOAD(ld, %o1, %o4))	! move a block of 32 bytes
	sllx	%o4, 32, %o5
	EX_LD(LOAD(ld, %o1+4, %o4))
	or	%o4, %o5, %o5
	EX_ST(STORE(stx, %o5, %o0))
	subcc	%o2, 32, %o2		! decrement length count
	EX_LD(LOAD(ld, %o1+8, %o4))
	sllx	%o4, 32, %o5
	EX_LD(LOAD(ld, %o1+12, %o4))
	or	%o4, %o5, %o5
	EX_ST(STORE(stx, %o5, %o0+8))
	add	%o1, 32, %o1		! increase src ptr by 32
	EX_LD(LOAD(ld, %o1-16, %o4))
	sllx	%o4, 32, %o5
	EX_LD(LOAD(ld, %o1-12, %o4))
	or	%o4, %o5, %o5
	EX_ST(STORE(stx, %o5, %o0+16))
	add	%o0, 32, %o0		! increase dst ptr by 32
	EX_LD(LOAD(ld, %o1-8, %o4))
	sllx	%o4, 32, %o5
	EX_LD(LOAD(ld, %o1-4, %o4))
	or	%o4, %o5, %o5
	bgu,pt	%xcc, .Lmedw32		! repeat if at least 32 bytes left
	 EX_ST(STORE(stx, %o5, %o0-8))
.Lmedw31:
	addcc	%o2, 31, %o2		! restore count

	bz,pt	%xcc, .Lsmallx	! exit if finished
	 nop
	cmp	%o2, 16
	blt,pt	%xcc, .Lmedw15
	 nop
	EX_LD(LOAD(ld, %o1, %o4))	! move a block of 16 bytes
	sllx	%o4, 32, %o5
	subcc	%o2, 16, %o2		! decrement length count
	EX_LD(LOAD(ld, %o1+4, %o4))
	or	%o4, %o5, %o5
	EX_ST(STORE(stx, %o5, %o0))
	add	%o1, 16, %o1		! increase src ptr by 16
	EX_LD(LOAD(ld, %o1-8, %o4))
	add	%o0, 16, %o0		! increase dst ptr by 16
	sllx	%o4, 32, %o5
	EX_LD(LOAD(ld, %o1-4, %o4))
	or	%o4, %o5, %o5
	EX_ST(STORE(stx, %o5, %o0-8))
.Lmedw15:
	bz,pt	%xcc, .Lsmallx	! exit if finished
	 cmp	%o2, 8
	blt,pn	%xcc, .Lmedw7		! skip if 7 or fewer bytes left
	 tst	%o2
	EX_LD(LOAD(ld, %o1, %o4))	! load 4 bytes
	subcc	%o2, 8, %o2		! decrease count by 8
	EX_ST(STORE(stw, %o4, %o0))	! and store 4 bytes
	add	%o1, 8, %o1		! increase src ptr by 8
	EX_LD(LOAD(ld, %o1-4, %o3))	! load 4 bytes
	add	%o0, 8, %o0		! increase dst ptr by 8
	EX_ST(STORE(stw, %o3, %o0-4))	! and store 4 bytes
	bz,pt	%xcc, .Lsmallx	! exit if finished
.Lmedw7:				! count is ge 1, less than 8
	cmp	%o2, 4			! check for 4 bytes left
	blt,pn	%xcc, .Lsmallleft3	! skip if 3 or fewer bytes left
	 nop				!
	EX_LD(LOAD(ld, %o1, %o4))	! load 4 bytes
	add	%o1, 4, %o1		! increase src ptr by 4
	add	%o0, 4, %o0		! increase dst ptr by 4
	subcc	%o2, 4, %o2		! decrease count by 4
	bnz	.Lsmallleft3
	 EX_ST(STORE(stw, %o4, %o0-4))! and store 4 bytes
	retl
	 mov	EX_RETVAL(%g1), %o0

	.align 16
.Llarge_align8_copy:			! Src and dst share 8 byte alignment
	! align dst to 64 byte boundary
	andcc	%o0, 0x3f, %o3		! %o3 == 0 means dst is 64 byte aligned
	brz,pn	%o3, .Laligned_to_64
	 andcc	%o0, 8, %o3		! odd long words to move?
	brz,pt	%o3, .Laligned_to_16
	 nop
	EX_LD(LOAD(ldx, %o1, %o4))
	sub	%o2, 8, %o2
	add	%o1, 8, %o1		! increment src ptr
	add	%o0, 8, %o0		! increment dst ptr
	EX_ST(STORE(stx, %o4, %o0-8))
.Laligned_to_16:
	andcc	%o0, 16, %o3		! pair of long words to move?
	brz,pt	%o3, .Laligned_to_32
	 nop
	EX_LD(LOAD(ldx, %o1, %o4))
	sub	%o2, 16, %o2
	EX_ST(STORE(stx, %o4, %o0))
	add	%o1, 16, %o1		! increment src ptr
	EX_LD(LOAD(ldx, %o1-8, %o4))
	add	%o0, 16, %o0		! increment dst ptr
	EX_ST(STORE(stx, %o4, %o0-8))
.Laligned_to_32:
	andcc	%o0, 32, %o3		! four long words to move?
	brz,pt	%o3, .Laligned_to_64
	 nop
	EX_LD(LOAD(ldx, %o1, %o4))
	sub	%o2, 32, %o2
	EX_ST(STORE(stx, %o4, %o0))
	EX_LD(LOAD(ldx, %o1+8, %o4))
	EX_ST(STORE(stx, %o4, %o0+8))
	EX_LD(LOAD(ldx, %o1+16, %o4))
	EX_ST(STORE(stx, %o4, %o0+16))
	add	%o1, 32, %o1		! increment src ptr
	EX_LD(LOAD(ldx, %o1-8, %o4))
	add	%o0, 32, %o0		! increment dst ptr
	EX_ST(STORE(stx, %o4, %o0-8))
.Laligned_to_64:
!
!	Using block init store (BIS) instructions to avoid fetching cache
!	lines from memory. Use ST_CHUNK stores to first element of each cache
!	line (similar to prefetching) to avoid overfilling STQ or miss buffers.
!	Gives existing cache lines time to be moved out of L1/L2/L3 cache.
!	Initial stores using MRU version of BIS to keep cache line in
!	cache until we are ready to store final element of cache line.
!	Then store last element using the LRU version of BIS.
!
	andn	%o2, 0x3f, %o5		! %o5 is multiple of block size
	and	%o2, 0x3f, %o2		! residue bytes in %o2
!
!	We use STORE_MRU_ASI for the first seven stores to each cache line
!	followed by STORE_ASI (mark as LRU) for the last store. That
!	mixed approach reduces the probability that the cache line is removed
!	before we finish setting it, while minimizing the effects on
!	other cached values during a large memcpy
!
!	ST_CHUNK batches up initial BIS operations for several cache lines
!	to allow multiple requests to not be blocked by overflowing the
!	the store miss buffer. Then the matching stores for all those
!	BIS operations are executed.
!

	sub	%o0, 8, %o0		! adjust %o0 for ASI alignment
.Lalign_loop:
	cmp	%o5, ST_CHUNK*64
	blu,pt	%xcc, .Lalign_loop_fin
	 mov	ST_CHUNK,%o3
.Lalign_loop_start:
	prefetch [%o1 + (ALIGN_PRE * BLOCK_SIZE)], 21
	subcc	%o3, 1, %o3
	EX_LD(LOAD(ldx, %o1, %o4))
	add	%o1, 64, %o1
	add	%o0, 8, %o0
	EX_ST(STORE_INIT_MRU(%o4, %o0))
	bgu	%xcc,.Lalign_loop_start
	 add	%o0, 56, %o0

	mov	ST_CHUNK,%o3
	sllx	%o3, 6, %o4		! ST_CHUNK*64
	sub	%o1, %o4, %o1		! reset %o1
	sub	%o0, %o4, %o0		! reset %o0

.Lalign_loop_rest:
	EX_LD(LOAD(ldx, %o1+8, %o4))
	add	%o0, 16, %o0
	EX_ST(STORE_INIT_MRU(%o4, %o0))
	EX_LD(LOAD(ldx, %o1+16, %o4))
	add	%o0, 8, %o0
	EX_ST(STORE_INIT_MRU(%o4, %o0))
	subcc	%o3, 1, %o3
	EX_LD(LOAD(ldx, %o1+24, %o4))
	add	%o0, 8, %o0
	EX_ST(STORE_INIT_MRU(%o4, %o0))
	EX_LD(LOAD(ldx, %o1+32, %o4))
	add	%o0, 8, %o0
	EX_ST(STORE_INIT_MRU(%o4, %o0))
	EX_LD(LOAD(ldx, %o1+40, %o4))
	add	%o0, 8, %o0
	EX_ST(STORE_INIT_MRU(%o4, %o0))
	EX_LD(LOAD(ldx, %o1+48, %o4))
	add	%o1, 64, %o1
	add	%o0, 8, %o0
	EX_ST(STORE_INIT_MRU(%o4, %o0))
	add	%o0, 8, %o0
	EX_LD(LOAD(ldx, %o1-8, %o4))
	sub	%o5, 64, %o5
	bgu	%xcc,.Lalign_loop_rest
	! mark cache line as LRU
	 EX_ST(STORE_INIT(%o4, %o0))

	cmp	%o5, ST_CHUNK*64
	bgu,pt	%xcc, .Lalign_loop_start
	 mov	ST_CHUNK,%o3

	cmp	%o5, 0
	beq	.Lalign_done
	 nop
.Lalign_loop_fin:
	EX_LD(LOAD(ldx, %o1, %o4))
	EX_ST(STORE(stx, %o4, %o0+8))
	EX_LD(LOAD(ldx, %o1+8, %o4))
	EX_ST(STORE(stx, %o4, %o0+8+8))
	EX_LD(LOAD(ldx, %o1+16, %o4))
	EX_ST(STORE(stx, %o4, %o0+8+16))
	subcc	%o5, 64, %o5
	EX_LD(LOAD(ldx, %o1+24, %o4))
	EX_ST(STORE(stx, %o4, %o0+8+24))
	EX_LD(LOAD(ldx, %o1+32, %o4))
	EX_ST(STORE(stx, %o4, %o0+8+32))
	EX_LD(LOAD(ldx, %o1+40, %o4))
	EX_ST(STORE(stx, %o4, %o0+8+40))
	EX_LD(LOAD(ldx, %o1+48, %o4))
	add	%o1, 64, %o1
	EX_ST(STORE(stx, %o4, %o0+8+48))
	add	%o0, 64, %o0
	EX_LD(LOAD(ldx, %o1-8, %o4))
	bgu	%xcc,.Lalign_loop_fin
	 EX_ST(STORE(stx, %o4, %o0))

.Lalign_done:
	add	%o0, 8, %o0		! restore %o0 from ASI alignment
	membar	#StoreStore
	sub	%o2, 63, %o2		! adjust length to allow cc test
	ba	.Lmedl63		! in .Lmedl63
	 nop

	.align 16
	! Dst is on 8 byte boundary; src is not; remaining count > SMALL_MAX
.Lunalignsetup:
.Lunalignrejoin:
	mov	%g1, %o3	! save %g1 as VISEntryHalf clobbers it
#ifdef NON_USER_COPY
	VISEntryHalfFast(.Lmedium_vis_entry_fail_cp)
#else
	VISEntryHalf
#endif
	mov	%o3, %g1	! restore %g1

	set MED_UMAX, %o3
	cmp %o2, %o3 		! check for.Lmedium unaligned limit
	bge,pt	%xcc,.Lunalign_large
	 prefetch [%o1 + (4 * BLOCK_SIZE)], 20
	andn	%o2, 0x3f, %o5		! %o5 is multiple of block size
	and	%o2, 0x3f, %o2		! residue bytes in %o2
	cmp	%o2, 8			! Insure we do not load beyond
	bgt	.Lunalign_adjust	! end of source buffer
	 andn	%o1, 0x7, %o4		! %o4 has long word aligned src address
	add	%o2, 64, %o2		! adjust to leave loop
	sub	%o5, 64, %o5		! early if necessary
.Lunalign_adjust:
	alignaddr %o1, %g0, %g0		! generate %gsr
	add	%o1, %o5, %o1		! advance %o1 to after blocks
	EX_LD_FP(LOAD(ldd, %o4, %f0))
.Lunalign_loop:
	EX_LD_FP(LOAD(ldd, %o4+8, %f2))
	faligndata %f0, %f2, %f16
	EX_LD_FP(LOAD(ldd, %o4+16, %f4))
	subcc	%o5, BLOCK_SIZE, %o5
	EX_ST_FP(STORE(std, %f16, %o0))
	faligndata %f2, %f4, %f18
	EX_LD_FP(LOAD(ldd, %o4+24, %f6))
	EX_ST_FP(STORE(std, %f18, %o0+8))
	faligndata %f4, %f6, %f20
	EX_LD_FP(LOAD(ldd, %o4+32, %f8))
	EX_ST_FP(STORE(std, %f20, %o0+16))
	faligndata %f6, %f8, %f22
	EX_LD_FP(LOAD(ldd, %o4+40, %f10))
	EX_ST_FP(STORE(std, %f22, %o0+24))
	faligndata %f8, %f10, %f24
	EX_LD_FP(LOAD(ldd, %o4+48, %f12))
	EX_ST_FP(STORE(std, %f24, %o0+32))
	faligndata %f10, %f12, %f26
	EX_LD_FP(LOAD(ldd, %o4+56, %f14))
	add	%o4, BLOCK_SIZE, %o4
	EX_ST_FP(STORE(std, %f26, %o0+40))
	faligndata %f12, %f14, %f28
	EX_LD_FP(LOAD(ldd, %o4, %f0))
	EX_ST_FP(STORE(std, %f28, %o0+48))
	faligndata %f14, %f0, %f30
	EX_ST_FP(STORE(std, %f30, %o0+56))
	add	%o0, BLOCK_SIZE, %o0
	bgu,pt	%xcc, .Lunalign_loop
	 prefetch [%o4 + (5 * BLOCK_SIZE)], 20
	ba	.Lunalign_done
	 nop

.Lunalign_large:
	andcc	%o0, 0x3f, %o3		! is dst 64-byte block aligned?
	bz	%xcc, .Lunalignsrc
	 sub	%o3, 64, %o3		! %o3 will be multiple of 8
	neg	%o3			! bytes until dest is 64 byte aligned
	sub	%o2, %o3, %o2		! update cnt with bytes to be moved
	! Move bytes according to source alignment
	andcc	%o1, 0x1, %o5
	bnz	%xcc, .Lunalignbyte	! check for byte alignment
	 nop
	andcc	%o1, 2, %o5		! check for half word alignment
	bnz	%xcc, .Lunalignhalf
	 nop
	! Src is word aligned
.Lunalignword:
	EX_LD_FP(LOAD(ld, %o1, %o4))	! load 4 bytes
	add	%o1, 8, %o1		! increase src ptr by 8
	EX_ST_FP(STORE(stw, %o4, %o0))	! and store 4 bytes
	subcc	%o3, 8, %o3		! decrease count by 8
	EX_LD_FP(LOAD(ld, %o1-4, %o4))	! load 4 bytes
	add	%o0, 8, %o0		! increase dst ptr by 8
	bnz	%xcc, .Lunalignword
	 EX_ST_FP(STORE(stw, %o4, %o0-4))! and store 4 bytes
	ba	.Lunalignsrc
	 nop

	! Src is half-word aligned
.Lunalignhalf:
	EX_LD_FP(LOAD(lduh, %o1, %o4))	! load 2 bytes
	sllx	%o4, 32, %o5		! shift left
	EX_LD_FP(LOAD(lduw, %o1+2, %o4))
	or	%o4, %o5, %o5
	sllx	%o5, 16, %o5
	EX_LD_FP(LOAD(lduh, %o1+6, %o4))
	or	%o4, %o5, %o5
	EX_ST_FP(STORE(stx, %o5, %o0))
	add	%o1, 8, %o1
	subcc	%o3, 8, %o3
	bnz	%xcc, .Lunalignhalf
	 add	%o0, 8, %o0
	ba	.Lunalignsrc
	 nop

	! Src is Byte aligned
.Lunalignbyte:
	sub	%o0, %o1, %o0		! share pointer advance
.Lunalignbyte_loop:
	EX_LD_FP(LOAD(ldub, %o1, %o4))
	sllx	%o4, 56, %o5
	EX_LD_FP(LOAD(lduh, %o1+1, %o4))
	sllx	%o4, 40, %o4
	or	%o4, %o5, %o5
	EX_LD_FP(LOAD(lduh, %o1+3, %o4))
	sllx	%o4, 24, %o4
	or	%o4, %o5, %o5
	EX_LD_FP(LOAD(lduh, %o1+5, %o4))
	sllx	%o4,  8, %o4
	or	%o4, %o5, %o5
	EX_LD_FP(LOAD(ldub, %o1+7, %o4))
	or	%o4, %o5, %o5
	add	%o0, %o1, %o0
	EX_ST_FP(STORE(stx, %o5, %o0))
	sub	%o0, %o1, %o0
	subcc	%o3, 8, %o3
	bnz	%xcc, .Lunalignbyte_loop
	 add	%o1, 8, %o1
	add	%o0,%o1, %o0 		! restore pointer

	! Destination is now block (64 byte aligned)
.Lunalignsrc:
	andn	%o2, 0x3f, %o5		! %o5 is multiple of block size
	and	%o2, 0x3f, %o2		! residue bytes in %o2
	add	%o2, 64, %o2		! Insure we do not load beyond
	sub	%o5, 64, %o5		! end of source buffer

	andn	%o1, 0x7, %o4		! %o4 has long word aligned src address
	alignaddr %o1, %g0, %g0		! generate %gsr
	add	%o1, %o5, %o1		! advance %o1 to after blocks

	EX_LD_FP(LOAD(ldd, %o4, %f14))
	add	%o4, 8, %o4
.Lunalign_sloop:
	EX_LD_FP(LOAD(ldd, %o4, %f16))
	faligndata %f14, %f16, %f0
	EX_LD_FP(LOAD(ldd, %o4+8, %f18))
	faligndata %f16, %f18, %f2
	EX_LD_FP(LOAD(ldd, %o4+16, %f20))
	faligndata %f18, %f20, %f4
	EX_ST_FP(STORE(std, %f0, %o0))
	subcc	%o5, 64, %o5
	EX_LD_FP(LOAD(ldd, %o4+24, %f22))
	faligndata %f20, %f22, %f6
	EX_ST_FP(STORE(std, %f2, %o0+8))
	EX_LD_FP(LOAD(ldd, %o4+32, %f24))
	faligndata %f22, %f24, %f8
	EX_ST_FP(STORE(std, %f4, %o0+16))
	EX_LD_FP(LOAD(ldd, %o4+40, %f26))
	faligndata %f24, %f26, %f10
	EX_ST_FP(STORE(std, %f6, %o0+24))
	EX_LD_FP(LOAD(ldd, %o4+48, %f28))
	faligndata %f26, %f28, %f12
	EX_ST_FP(STORE(std, %f8, %o0+32))
	add	%o4, 64, %o4
	EX_LD_FP(LOAD(ldd, %o4-8, %f30))
	faligndata %f28, %f30, %f14
	EX_ST_FP(STORE(std, %f10, %o0+40))
	EX_ST_FP(STORE(std, %f12, %o0+48))
	add	%o0, 64, %o0
	EX_ST_FP(STORE(std, %f14, %o0-8))
	fsrc2	%f30, %f14
	bgu,pt	%xcc, .Lunalign_sloop
	 prefetch [%o4 + (8 * BLOCK_SIZE)], 20

.Lunalign_done:
	! Handle trailing bytes, 64 to 127
	! Dest long word aligned, Src not long word aligned
	cmp	%o2, 15
	bleu	%xcc, .Lunalign_short

	 andn	%o2, 0x7, %o5		! %o5 is multiple of 8
	and	%o2, 0x7, %o2		! residue bytes in %o2
	add	%o2, 8, %o2
	sub	%o5, 8, %o5		! insure we do not load past end of src
	andn	%o1, 0x7, %o4		! %o4 has long word aligned src address
	add	%o1, %o5, %o1		! advance %o1 to after multiple of 8
	EX_LD_FP(LOAD(ldd, %o4, %f0))	! fetch partial word
.Lunalign_by8:
	EX_LD_FP(LOAD(ldd, %o4+8, %f2))
	add	%o4, 8, %o4
	faligndata %f0, %f2, %f16
	subcc	%o5, 8, %o5
	EX_ST_FP(STORE(std, %f16, %o0))
	fsrc2	%f2, %f0
	bgu,pt	%xcc, .Lunalign_by8
	 add	%o0, 8, %o0

.Lunalign_short:
#ifdef NON_USER_COPY
	VISExitHalfFast
#else
	VISExitHalf
#endif
	ba	.Lsmallrest
	 nop

/*
 * This is a special case of nested memcpy. This can happen when kernel
 * calls unaligned memcpy back to back without saving FP registers. We need
 * traps(context switch) to save/restore FP registers. If the kernel calls
 * memcpy without this trap sequence we will hit FP corruption. Let's use
 * the normal integer load/store method in this case.
 */

#ifdef NON_USER_COPY
.Lmedium_vis_entry_fail_cp:
	or	%o0, %o1, %g2
#endif
.Lmedium_cp:
	LOAD(prefetch, %o1 + 0x40, #n_reads_strong)
	andcc	%g2, 0x7, %g0
	bne,pn	%xcc, .Lmedium_unaligned_cp
	 nop

.Lmedium_noprefetch_cp:
	andncc	%o2, 0x20 - 1, %o5
	be,pn	%xcc, 2f
	 sub	%o2, %o5, %o2
1:	EX_LD(LOAD(ldx, %o1 + 0x00, %o3))
	EX_LD(LOAD(ldx, %o1 + 0x08, %g2))
	EX_LD(LOAD(ldx, %o1 + 0x10, %g7))
	EX_LD(LOAD(ldx, %o1 + 0x18, %o4))
	add	%o1, 0x20, %o1
	subcc	%o5, 0x20, %o5
	EX_ST(STORE(stx, %o3, %o0 + 0x00))
	EX_ST(STORE(stx, %g2, %o0 + 0x08))
	EX_ST(STORE(stx, %g7, %o0 + 0x10))
	EX_ST(STORE(stx, %o4, %o0 + 0x18))
	bne,pt	%xcc, 1b
	 add	%o0, 0x20, %o0
2:	andcc	%o2, 0x18, %o5
	be,pt	%xcc, 3f
	 sub	%o2, %o5, %o2
1:	EX_LD(LOAD(ldx, %o1 + 0x00, %o3))
	add	%o1, 0x08, %o1
	add	%o0, 0x08, %o0
	subcc	%o5, 0x08, %o5
	bne,pt	%xcc, 1b
	 EX_ST(STORE(stx, %o3, %o0 - 0x08))
3:	brz,pt	%o2, .Lexit_cp
	 cmp	%o2, 0x04
	bl,pn	%xcc, .Ltiny_cp
	 nop
	EX_LD(LOAD(lduw, %o1 + 0x00, %o3))
	add	%o1, 0x04, %o1
	add	%o0, 0x04, %o0
	subcc	%o2, 0x04, %o2
	bne,pn	%xcc, .Ltiny_cp
	 EX_ST(STORE(stw, %o3, %o0 - 0x04))
	ba,a,pt	%xcc, .Lexit_cp

.Lmedium_unaligned_cp:
	/* First get dest 8 byte aligned.  */
	sub	%g0, %o0, %o3
	and	%o3, 0x7, %o3
	brz,pt	%o3, 2f
	 sub	%o2, %o3, %o2

1:	EX_LD(LOAD(ldub, %o1 + 0x00, %g2))
	add	%o1, 1, %o1
	subcc	%o3, 1, %o3
	add	%o0, 1, %o0
	bne,pt	%xcc, 1b
	 EX_ST(STORE(stb, %g2, %o0 - 0x01))
2:
	and	%o1, 0x7, %o3
	brz,pn	%o3, .Lmedium_noprefetch_cp
	 sll	%o3, 3, %o3
	mov	64, %g2
	sub	%g2, %o3, %g2
	andn	%o1, 0x7, %o1
	EX_LD(LOAD(ldx, %o1 + 0x00, %o4))
	sllx	%o4, %o3, %o4
	andn	%o2, 0x08 - 1, %o5
	sub	%o2, %o5, %o2

1:	EX_LD(LOAD(ldx, %o1 + 0x08, %g3))
	add	%o1, 0x08, %o1
	subcc	%o5, 0x08, %o5
	srlx	%g3, %g2, %g7
	or	%g7, %o4, %g7
	EX_ST(STORE(stx, %g7, %o0 + 0x00))
	add	%o0, 0x08, %o0
	bne,pt	%xcc, 1b
	 sllx	%g3, %o3, %o4
	srl	%o3, 3, %o3
	add	%o1, %o3, %o1
	brz,pn	%o2, .Lexit_cp
	 nop
	ba,pt	%xcc, .Lsmall_unaligned_cp

.Ltiny_cp:
	EX_LD(LOAD(ldub, %o1 + 0x00, %o3))
	subcc	%o2, 1, %o2
	be,pn	%xcc, .Lexit_cp
	 EX_ST(STORE(stb, %o3, %o0 + 0x00))
	EX_LD(LOAD(ldub, %o1 + 0x01, %o3))
	subcc	%o2, 1, %o2
	be,pn	%xcc, .Lexit_cp
	 EX_ST(STORE(stb, %o3, %o0 + 0x01))
	EX_LD(LOAD(ldub, %o1 + 0x02, %o3))
	ba,pt	%xcc, .Lexit_cp
	 EX_ST(STORE(stb, %o3, %o0 + 0x02))

.Lsmall_cp:
	andcc	%g2, 0x3, %g0
	bne,pn	%xcc, .Lsmall_unaligned_cp
	 andn	%o2, 0x4 - 1, %o5
	sub	%o2, %o5, %o2
1:
	EX_LD(LOAD(lduw, %o1 + 0x00, %o3))
	add	%o1, 0x04, %o1
	subcc	%o5, 0x04, %o5
	add	%o0, 0x04, %o0
	bne,pt	%xcc, 1b
	 EX_ST(STORE(stw, %o3, %o0 - 0x04))
	brz,pt	%o2, .Lexit_cp
	 nop
	ba,a,pt	%xcc, .Ltiny_cp

.Lsmall_unaligned_cp:
1:	EX_LD(LOAD(ldub, %o1 + 0x00, %o3))
	add	%o1, 1, %o1
	add	%o0, 1, %o0
	subcc	%o2, 1, %o2
	bne,pt	%xcc, 1b
	 EX_ST(STORE(stb, %o3, %o0 - 0x01))
	ba,a,pt	%xcc, .Lexit_cp

.Lsmallrest:
	tst	%o2
	bz,pt	%xcc, .Lsmallx
	 cmp	%o2, 4
	blt,pn	%xcc, .Lsmallleft3
	 nop
	sub	%o2, 3, %o2
.Lsmallnotalign4:
	EX_LD(LOAD(ldub, %o1, %o3))! read byte
	subcc	%o2, 4, %o2		! reduce count by 4
	EX_ST(STORE(stb, %o3, %o0))	! write byte
	EX_LD(LOAD(ldub, %o1+1, %o3))! repeat for total of 4 bytes
	add	%o1, 4, %o1		! advance SRC by 4
	EX_ST(STORE(stb, %o3, %o0+1))
	EX_LD(LOAD(ldub, %o1-2, %o3))
	add	%o0, 4, %o0		! advance DST by 4
	EX_ST(STORE(stb, %o3, %o0-2))
	EX_LD(LOAD(ldub, %o1-1, %o3))
	bgu,pt	%xcc, .Lsmallnotalign4	! loop til 3 or fewer bytes remain
	EX_ST(STORE(stb, %o3, %o0-1))
	addcc	%o2, 3, %o2		! restore count
	bz,pt	%xcc, .Lsmallx
.Lsmallleft3:				! 1, 2, or 3 bytes remain
	subcc	%o2, 1, %o2
	EX_LD(LOAD(ldub, %o1, %o3))	! load one byte
	bz,pt	%xcc, .Lsmallx
	 EX_ST(STORE(stb, %o3, %o0))	! store one byte
	EX_LD(LOAD(ldub, %o1+1, %o3))	! load second byte
	subcc	%o2, 1, %o2
	bz,pt	%xcc, .Lsmallx
	 EX_ST(STORE(stb, %o3, %o0+1))! store second byte
	EX_LD(LOAD(ldub, %o1+2, %o3))	! load third byte
	EX_ST(STORE(stb, %o3, %o0+2))	! store third byte
.Lsmallx:
	retl
	 mov	EX_RETVAL(%g1), %o0
.Lsmallfin:
	tst	%o2
	bnz,pn	%xcc, .Lsmallleft3
	 nop
	retl
	 mov	EX_RETVAL(%g1), %o0	! restore %o0
.Lexit_cp:
	retl
	 mov	EX_RETVAL(%g1), %o0
	.size  FUNC_NAME, .-FUNC_NAME