summaryrefslogtreecommitdiff
path: root/libc/string/sh/sh4/memcpy.S
blob: 6a229a06c7cdcb357e2baa75a6648edeac710459 (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
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
/*
 * "memcpy" implementation of SuperH
 *
 * Copyright (C) 1999  Niibe Yutaka
 * Copyright (c) 2002  STMicroelectronics Ltd
 *   Modified from memcpy.S and micro-optimised for SH4
 *   Stuart Menefy (stuart.menefy@st.com)
 *
 * Copyright (c) 2009  STMicroelectronics Ltd
 *   Optimised using prefetching and 64bit data transfer via FPU
 *   Author: Giuseppe Cavallaro <peppe.cavallaro@st.com>
 */

/*
 * void *memcpy(void *dst, const void *src, size_t n);
 *
 * It is assumed that there is no overlap between src and dst.
 * If there is an overlap, then the results are undefined.
 */

#include <sysdep.h>
#include <endian.h>

#if defined (__LITTLE_ENDIAN__) && defined (__SH_FPU_ANY__)
#define	MEMCPY_USES_FPU
/* Use paired single precision load or store mode for 64-bit tranfering.
 * FPSCR.SZ=1,FPSCR.SZ=0 is well defined on both SH4-200 and SH4-300.
 * Currenlty it has been only implemented and tested for little endian mode. */
.macro FPU_SET_PAIRED_PREC
	sts	fpscr, r7
	mov	#0x10, r0	! PR=0 SZ=1
	shll16	r0
	lds	r0, fpscr
.endm
.macro RESTORE_FPSCR
	lds	r7, fpscr
.endm
.macro DALLOC
	! Cache allocate + store on dst-32.
	add	#-32, r1
	movca.l	r0, @r1
	add	#32, r1
.endm

#endif

	!
	!	GHIJ KLMN OPQR -->  ...G HIJK LMNO PQR.
	!

	! Size is 16 or greater, and may have trailing bytes

	.balign	32
.Lcase1:
	! Read a long word and write a long word at once
	! At the start of each iteration, r7 contains last long load
	add	#-1,r5		!  79 EX
	mov	r4,r2		!   5 MT (0 cycles latency)

	mov.l	@(r0,r5),r7	!  21 LS (2 cycles latency)
	add	#-4,r5		!  50 EX

	add	#7,r2		!  79 EX
	!
#ifdef __LITTLE_ENDIAN__
	! 6 cycles, 4 bytes per iteration
3:	mov.l	@(r0,r5),r1	!  21 LS (latency=2)	! NMLK
	mov	r7, r3		!   5 MT (latency=0)	! RQPO

	cmp/hi	r2,r0		!  57 MT
	shll16	r3		! 103 EX

	mov	r1,r6		!   5 MT (latency=0)
	shll8	r3		! 102 EX		! Oxxx

	shlr8	r6		! 106 EX		! xNML
	mov	r1, r7		!   5 MT (latency=0)

	or	r6,r3		!  82 EX		! ONML
	bt/s	3b		! 109 BR

	 mov.l	r3,@-r0		!  30 LS
#else
3:	mov.l	@(r0,r5),r1	!  21 LS (latency=2)	! KLMN
	mov	r7,r3		!   5 MT (latency=0)	! OPQR

	cmp/hi	r2,r0		!  57 MT
	shlr16	r3		! 107 EX

	shlr8	r3		! 106 EX		! xxxO
	mov	r1,r6		!   5 MT (latency=0)

	shll8	r6		! 102 EX		! LMNx
	mov	r1,r7		!   5 MT (latency=0)

	or	r6,r3		!  82 EX		! LMNO
	bt/s	3b		! 109 BR

	 mov.l	r3,@-r0		!  30 LS
#endif
	! Finally, copy a byte at once, if necessary

	add	#4,r5		!  50 EX
	cmp/eq	r4,r0		!  54 MT

	add	#-6,r2		!  50 EX
	bt	9f		! 109 BR

8:	cmp/hi	r2,r0		!  57 MT
	mov.b	@(r0,r5),r1	!  20 LS (latency=2)

	bt/s	8b		! 109 BR

	 mov.b	r1,@-r0		!  29 LS

9:	rts
	 nop


	!
	!	GHIJ KLMN OPQR -->  .GHI JKLM NOPQ R...
	!

	! Size is 16 or greater, and may have trailing bytes

	.balign	32
.Lcase3:
	! Read a long word and write a long word at once
	! At the start of each iteration, r7 contains last long load
	add	#-3,r5		! 79 EX
	mov	r4,r2		!  5 MT (0 cycles latency)

	mov.l	@(r0,r5),r7	! 21 LS (2 cycles latency)
	add	#-4,r5		! 50 EX

	add	#7,r2		!  79 EX
	!
#ifdef __LITTLE_ENDIAN__
	! 6 cycles, 4 bytes per iteration
3:	mov.l	@(r0,r5),r1	!  21 LS (latency=2)	! NMLK
	mov	r7, r3		!   5 MT (latency=0)	! RQPO

	cmp/hi	r2,r0		!  57 MT
	shll8	r3		! 102 EX		! QPOx

	mov	r1,r6		!   5 MT (latency=0)
	shlr16	r6		! 107 EX

	shlr8	r6		! 106 EX		! xxxN
	mov	r1, r7		!   5 MT (latency=0)

	or	r6,r3		!  82 EX		! QPON
	bt/s	3b		! 109 BR

	 mov.l	r3,@-r0		!  30 LS
#else
3:	mov	r7,r3		! OPQR
	shlr8	r3		! xOPQ
	mov.l	@(r0,r5),r7	! KLMN
	mov	r7,r6
	shll16	r6
	shll8	r6		! Nxxx
	or	r6,r3		! NOPQ
	cmp/hi	r2,r0
	bt/s	3b
	 mov.l	r3,@-r0
#endif

	! Finally, copy a byte at once, if necessary

	add	#6,r5		!  50 EX
	cmp/eq	r4,r0		!  54 MT

	add	#-6,r2		!  50 EX
	bt	9f		! 109 BR

8:	cmp/hi	r2,r0		!  57 MT
	mov.b	@(r0,r5),r1	!  20 LS (latency=2)

	bt/s	8b		! 109 BR

	 mov.b	r1,@-r0		!  29 LS

9:	rts
	 nop

ENTRY(memcpy)

	! Calculate the invariants which will be used in the remainder
	! of the code:
	!
	!      r4   -->  [ ...  ] DST             [ ...  ] SRC
	!	         [ ...  ]                 [ ...  ]
	!	           :                        :
	!      r0   -->  [ ...  ]       r0+r5 --> [ ...  ]
	!
	!

	! Short circuit the common case of src, dst and len being 32 bit aligned
	! and test for zero length move

	mov	r6, r0		!   5 MT (0 cycle latency)
	or	r4, r0		!  82 EX

	or	r5, r0		!  82 EX
	tst	r6, r6		!  86 MT

	bt/s	99f		! 111 BR		(zero len)
	 tst	#3, r0		!  87 MT

	mov	r4, r0		!   5 MT (0 cycle latency)
	add	r6, r0		!  49 EX

	bt/s	.Lcase00	! 111 BR		(aligned)
	 sub	r4, r5		!  75 EX

	! Arguments are not nicely long word aligned or zero len.
	! Check for small copies, and if so do a simple byte at a time copy.
	!
	! Deciding on an exact value of 'small' is not easy, as the point at which
	! using the optimised routines become worthwhile varies (these are the
	! cycle counts for differnet sizes using byte-at-a-time vs. optimised):
	!	size	byte-at-time	long	word	byte
	!	16	42		39-40	46-50	50-55
	!	24	58		43-44	54-58	62-67
	!	36	82		49-50	66-70	80-85
	! However the penalty for getting it 'wrong' is much higher for long word
	! aligned data (and this is more common), so use a value of 16.

	mov	#16, r1		!   6 EX
	cmp/gt	r6,r1		!  56 MT

	add	#-1,r5		!  50 EX
	bf/s	6f		! 108 BR		(not small)

	 mov	r5, r3		!   5 MT (latency=0)
	shlr	r6		! 104 EX

	mov.b	@(r0,r5),r1	!  20 LS (latency=2)
	bf/s	4f		! 111 BR

	 add	#-1,r3		!  50 EX
	tst	r6, r6		!  86 MT

	bt/s	98f		! 110 BR
	 mov.b	r1,@-r0		!  29 LS

	! 4 cycles, 2 bytes per iteration
3:	mov.b	@(r0,r5),r1	!  20 LS (latency=2)

4:	mov.b	@(r0,r3),r2	!  20 LS (latency=2)
	dt	r6		!  67 EX

	mov.b	r1,@-r0		!  29 LS
	bf/s	3b		! 111 BR

	 mov.b	r2,@-r0		!  29 LS
98:
	rts
	 nop

99:	rts
	 mov	r4, r0

	! Size is not small, so its worthwhile looking for optimisations.
	! First align destination to a long word boundary.
	!
	! r5 = normal value -1

6:	tst	#3, r0		!  87 MT
        mov	#3, r3		!   6 EX

	bt/s	2f		! 111 BR
	 and	r0,r3		!  78 EX

	! 3 cycles, 1 byte per iteration
1:	dt	r3		!  67 EX
	mov.b	@(r0,r5),r1	!  19 LS (latency=2)

	add	#-1, r6		!  79 EX
	bf/s	1b		! 109 BR

	 mov.b	r1,@-r0		!  28 LS

2:	add	#1, r5		!  79 EX

	! Now select the appropriate bulk transfer code based on relative
	! alignment of src and dst.

	mov	r0, r3		!   5 MT (latency=0)

	mov	r5, r0		!   5 MT (latency=0)
	tst	#1, r0		!  87 MT

	bf/s	1f		! 111 BR
	 mov	#64, r7		!   6 EX

	! bit 0 clear

	cmp/ge	r7, r6		!  55 MT

	bt/s	2f		! 111 BR
	 tst	#2, r0		!  87 MT

	! small
	bt/s	.Lcase0
	 mov	r3, r0

	bra	.Lcase2
	 nop

	! big
2:	bt/s	.Lcase0b
	 mov	r3, r0

	bra	.Lcase2b
	 nop

	! bit 0 set
1:	tst	#2, r0		! 87 MT

	bt/s	.Lcase1
	 mov	r3, r0

	bra	.Lcase3
	 nop


	!
	!	GHIJ KLMN OPQR -->  GHIJ KLMN OPQR
	!

	! src, dst and size are all long word aligned
	! size is non-zero

	.balign	32
.Lcase00:
	mov	#64, r1		!   6 EX
	mov	r5, r3		!   5 MT (latency=0)

	cmp/gt	r6, r1		!  56 MT
	add	#-4, r5		!  50 EX

	bf	.Lcase00b	! 108 BR		(big loop)
	shlr2	r6		! 105 EX

	shlr	r6		! 104 EX
	mov.l	@(r0, r5), r1	!  21 LS (latency=2)

	bf/s	4f		! 111 BR
	 add	#-8, r3		!  50 EX

	tst	r6, r6		!  86 MT
	bt/s	5f		! 110 BR

	 mov.l	r1,@-r0		!  30 LS

	! 4 cycles, 2 long words per iteration
3:	mov.l	@(r0, r5), r1	!  21 LS (latency=2)

4:	mov.l	@(r0, r3), r2	!  21 LS (latency=2)
	dt	r6		!  67 EX

	mov.l	r1, @-r0	!  30 LS
	bf/s	3b		! 109 BR

	 mov.l	r2, @-r0	!  30 LS

5:	rts
	 nop


	! Size is 16 or greater and less than 64, but may have trailing bytes

	.balign	32
.Lcase0:
	add	#-4, r5		!  50 EX
	mov	r4, r7		!   5 MT (latency=0)

	mov.l	@(r0, r5), r1	!  21 LS (latency=2)
	mov	#4, r2		!   6 EX

	add	#11, r7		!  50 EX
	tst	r2, r6		!  86 MT

	mov	r5, r3		!   5 MT (latency=0)
	bt/s	4f		! 111 BR

	 add	#-4, r3		!  50 EX
	mov.l	r1,@-r0		!  30 LS

	! 4 cycles, 2 long words per iteration
3:	mov.l	@(r0, r5), r1	!  21 LS (latency=2)

4:	mov.l	@(r0, r3), r2	!  21 LS (latency=2)
	cmp/hi	r7, r0

	mov.l	r1, @-r0	!  30 LS
	bt/s	3b		! 109 BR

	 mov.l	r2, @-r0	!  30 LS

	! Copy the final 0-3 bytes

	add	#3,r5		!  50 EX

	cmp/eq	r0, r4		!  54 MT
	add	#-10, r7	!  50 EX

	bt	9f		! 110 BR

	! 3 cycles, 1 byte per iteration
1:	mov.b	@(r0,r5),r1	!  19 LS
	cmp/hi	r7,r0		!  57 MT

	bt/s	1b		! 111 BR
	 mov.b	r1,@-r0		!  28 LS

9:	rts
	 nop

	! Size is at least 64 bytes, so will be going round the big loop at least once.
	!
	!   r2 = rounded up r4
	!   r3 = rounded down r0

	.balign	32
.Lcase0b:
	add	#-4, r5		!  50 EX

.Lcase00b:
	mov	r0, r3		!   5 MT (latency=0)
	mov	#(~0x1f), r1	!   6 EX

	and	r1, r3		!  78 EX
	mov	r4, r2		!   5 MT (latency=0)

	cmp/eq	r3, r0		!  54 MT
	add	#0x1f, r2	!  50 EX

	bt/s	1f		! 110 BR
	 and	r1, r2		!  78 EX

	! copy initial words until cache line aligned

	mov.l	@(r0, r5), r1	!  21 LS (latency=2)
	tst	#4, r0		!  87 MT

	mov	r5, r6		!   5 MT (latency=0)
	add	#-4, r6		!  50 EX

	bt/s	4f		! 111 BR
	 add	#8, r3		!  50 EX

	tst	#0x18, r0	!  87 MT

	bt/s	1f		! 109 BR
	 mov.l	r1,@-r0		!  30 LS

	! 4 cycles, 2 long words per iteration
3:	mov.l	@(r0, r5), r1	!  21 LS (latency=2)

4:	mov.l	@(r0, r6), r7	!  21 LS (latency=2)
	cmp/eq	r3, r0		!  54 MT

	mov.l	r1, @-r0	!  30 LS
	bf/s	3b		! 109 BR

	 mov.l	r7, @-r0	!  30 LS

#ifdef MEMCPY_USES_FPU
	! Copy the cache line aligned blocks by using the FPU registers.
	! If src and dst are well aligned adopt 64-bit data transfer.
	! We also need r0 as a temporary (for movca), so 'undo' the invariant:
	!   r5:	 src (was r0+r5)
	!   r1:	 dest (was r0)
1:
	add	r0, r5
	mov	r0, r1

	mov	r1, r3		! MT
	sub	r2, r3		! EX (r3 - r2 -> r3)
	mov	#-5, r0
	shld	r0, r3		! number of the cache lines

	mov	#8, r0
	cmp/ge	r0, r3		! Check if there are many cache lines to copy.
	bf	45f		! Copy cache line aligned blocks without pref.
	mov	r5, r0
	add	#-0x7c, r0
	tst	#7, r0		! src is 8byte aligned
	bf	45f

	! Many cache lines have to be copied and the buffers are well aligned.
	! Aggressive prefetching and FPU in single paired precision.
	mov	r0, r5
	mov	r5, r6
	add	#-0x80, r6	! prefetch head

	! store FPU (in single precision mode, do not check R15 align).
	fmov	fr12, @-r15
	fmov	fr13, @-r15
	fmov	fr14, @-r15
	fmov	fr15, @-r15

	FPU_SET_PAIRED_PREC

	mov	#4, r0
67:
	add	#-0x20, r6
	pref	@r6
	add	#-0x20, r6
	pref	@r6

	fmov	@r5+, dr0
	fmov	@r5+, dr2
	fmov	@r5+, dr4
	fmov	@r5+, dr6
	fmov	@r5+, dr8
	fmov	@r5+, dr10
	fmov	@r5+, dr12
	fmov	@r5+, dr14
	fmov	@r5+, xd0
	fmov	@r5+, xd2
	fmov	@r5+, xd4
	fmov	@r5+, xd6
	fmov	@r5+, xd8
	fmov	@r5+, xd10
	fmov	@r5+, xd12
	fmov	@r5+, xd14

	DALLOC
	fmov	xd14, @-r1
	fmov	xd12, @-r1
	fmov	xd10, @-r1
	fmov	xd8, @-r1
	DALLOC
	fmov	xd6, @-r1
	fmov	xd4, @-r1
	fmov	xd2, @-r1
	fmov	xd0, @-r1
	DALLOC
	fmov	dr14, @-r1
	fmov	dr12, @-r1
	fmov	dr10, @-r1
	fmov	dr8, @-r1
	DALLOC
	fmov	dr6, @-r1
	add	#-0x80, r5
	fmov	dr4, @-r1
	add	#-0x80, r5
	fmov	dr2, @-r1
	add	#-0x20, r6
	fmov	dr0, @-r1
	add	#-4, r3
	pref	@r6
	add	#-0x20, r6
	cmp/ge	r0, r3
	bt/s	67b
	 pref	@r6

	RESTORE_FPSCR

	! Restore FPU callee save registers
	fmov	@r15+, fr15
	fmov	@r15+, fr14
	fmov	@r15+, fr13
	fmov	@r15+, fr12

	! Other cache lines could be copied: so use the FPU in single paired
	! precision without prefetching. No check for alignment is necessary.

	mov	#1, r0
	cmp/ge	r0, r3
	bt/s	3f
	 add	#0x60, r5

	bra	5f
	 nop

	! No prefetch and FPU in single precision.
45:
	add	#-0x1c, r5
	mov	r5, r0
	tst	#7, r0
	bt	3f

2:	fmov.s	@r5+, fr0
	fmov.s	@r5+, fr1
	fmov.s	@r5+, fr2
	fmov.s	@r5+, fr3
	fmov.s	@r5+, fr4
	fmov.s	@r5+, fr5
	fmov.s	@r5+, fr6
	fmov.s	@r5+, fr7

	DALLOC

	fmov.s	fr7, @-r1
	fmov.s	fr6, @-r1
	fmov.s	fr5, @-r1
	fmov.s	fr4, @-r1
	fmov.s	fr3, @-r1
	fmov.s	fr2, @-r1
	fmov.s	fr1, @-r1
	fmov.s	fr0, @-r1

	cmp/eq	r2,r1

	bf/s	2b
	 add	#-0x40, r5

	bra	5f
	 nop

	! No prefetch and FPU in single paired precision.

3:	FPU_SET_PAIRED_PREC

4:	fmov	@r5+, dr0
	fmov	@r5+, dr2
	fmov	@r5+, dr4
	fmov	@r5+, dr6

	DALLOC

	fmov	dr6, @-r1
	fmov	dr4, @-r1
	fmov	dr2, @-r1
	fmov	dr0, @-r1
	cmp/eq	r2,r1

	bf/s	4b
	 add	#-0x40, r5

	RESTORE_FPSCR

5:	mov	r1, r0

	cmp/eq	r4, r0		!  54 MT
	bf/s	1f		! 109 BR
	 sub	r1, r5		!  75 EX

	rts
	 nop
1:
#else
	! Copy the cache line aligned blocks
	!
	! In use: r0, r2, r4, r5
	! Scratch: r1, r3, r6, r7
	!
	! We could do this with the four scratch registers, but if src
	! and dest hit the same cache line, this will thrash, so make
	! use of additional registers.
	!
	! We also need r0 as a temporary (for movca), so 'undo' the invariant:
	!   r5:	 src (was r0+r5)
	!   r1:	 dest (was r0)
	! this can be reversed at the end, so we don't need to save any extra
	! state.
	!
1:	mov.l	r8, @-r15	!  30 LS
	add	r0, r5		!  49 EX

	mov.l	r9, @-r15	!  30 LS
	mov	r0, r1		!   5 MT (latency=0)

	mov.l	r10, @-r15	!  30 LS
	add	#-0x1c, r5	!  50 EX

	mov.l	r11, @-r15	!  30 LS

	! 16 cycles, 32 bytes per iteration
2:	mov.l	@(0x00,r5),r0	! 18 LS (latency=2)
	add	#-0x20, r1	! 50 EX
	mov.l	@(0x04,r5),r3	! 18 LS (latency=2)
	mov.l	@(0x08,r5),r6	! 18 LS (latency=2)
	mov.l	@(0x0c,r5),r7	! 18 LS (latency=2)
	mov.l	@(0x10,r5),r8	! 18 LS (latency=2)
	mov.l	@(0x14,r5),r9	! 18 LS (latency=2)
	mov.l	@(0x18,r5),r10	! 18 LS (latency=2)
	mov.l	@(0x1c,r5),r11	! 18 LS (latency=2)
	movca.l	r0,@r1		! 40 LS (latency=3-7)
	mov.l	r3,@(0x04,r1)	! 33 LS
	mov.l	r6,@(0x08,r1)	! 33 LS
	mov.l	r7,@(0x0c,r1)	! 33 LS

	mov.l	r8,@(0x10,r1)	! 33 LS
	add	#-0x20, r5	! 50 EX

	mov.l	r9,@(0x14,r1)	! 33 LS
	cmp/eq	r2,r1		! 54 MT

	mov.l	r10,@(0x18,r1)	!  33 LS
	bf/s	2b		! 109 BR

	 mov.l	r11,@(0x1c,r1)	!  33 LS

	mov	r1, r0		!   5 MT (latency=0)

	mov.l	@r15+, r11	!  15 LS
	sub	r1, r5		!  75 EX

	mov.l	@r15+, r10	!  15 LS
	cmp/eq	r4, r0		!  54 MT

	bf/s	1f		! 109 BR
	 mov.l	 @r15+, r9	!  15 LS

	rts
1:	 mov.l	@r15+, r8	!  15 LS
#endif
	sub	r4, r1		!  75 EX		(len remaining)

	! number of trailing bytes is non-zero
	!
	! invariants restored (r5 already decremented by 4)
	! also r1=num bytes remaining

	mov	#4, r2		!   6 EX
	mov	r4, r7		!   5 MT (latency=0)

	add	#0x1c, r5	!  50 EX		(back to -4)
	cmp/hs	r2, r1		!  58 MT

	bf/s	5f		! 108 BR
	 add	 #11, r7	!  50 EX

	mov.l	@(r0, r5), r6	!  21 LS (latency=2)
	tst	r2, r1		!  86 MT

	mov	r5, r3		!   5 MT (latency=0)
	bt/s	4f		! 111 BR

	 add	#-4, r3		!  50 EX
	cmp/hs	r2, r1		!  58 MT

	bt/s	5f		! 111 BR
	 mov.l	r6,@-r0		!  30 LS

	! 4 cycles, 2 long words per iteration
3:	mov.l	@(r0, r5), r6	!  21 LS (latency=2)

4:	mov.l	@(r0, r3), r2	!  21 LS (latency=2)
	cmp/hi	r7, r0

	mov.l	r6, @-r0	!  30 LS
	bt/s	3b		! 109 BR

	 mov.l	r2, @-r0	!  30 LS

	! Copy the final 0-3 bytes

5:	cmp/eq	r0, r4		!  54 MT
	add	#-10, r7	!  50 EX

	bt	9f		! 110 BR
	add	#3,r5		!  50 EX

	! 3 cycles, 1 byte per iteration
1:	mov.b	@(r0,r5),r1	!  19 LS
	cmp/hi	r7,r0		!  57 MT

	bt/s	1b		! 111 BR
	 mov.b	r1,@-r0		!  28 LS

9:	rts
	 nop

	!
	!	GHIJ KLMN OPQR -->  ..GH IJKL MNOP QR..
	!

	.balign	32
.Lcase2:
	! Size is 16 or greater and less then 64, but may have trailing bytes

2:	mov	r5, r6		!   5 MT (latency=0)
	add	#-2,r5		!  50 EX

	mov	r4,r2		!   5 MT (latency=0)
	add	#-4,r6		!  50 EX

	add	#7,r2		!  50 EX
3:	mov.w	@(r0,r5),r1	!  20 LS (latency=2)

	mov.w	@(r0,r6),r3	!  20 LS (latency=2)
	cmp/hi	r2,r0		!  57 MT

	mov.w	r1,@-r0		!  29 LS
	bt/s	3b		! 111 BR

	 mov.w	r3,@-r0		!  29 LS

	bra	10f
	 nop


	.balign	32
.Lcase2b:
	! Size is at least 64 bytes, so will be going round the big loop at least once.
	!
	!   r2 = rounded up r4
	!   r3 = rounded down r0

	mov	r0, r3		!   5 MT (latency=0)
	mov	#(~0x1f), r1	!   6 EX

	and	r1, r3		!  78 EX
	mov	r4, r2		!   5 MT (latency=0)

	cmp/eq	r3, r0		!  54 MT
	add	#0x1f, r2	!  50 EX

	add	#-2, r5		!  50 EX
	bt/s	1f		! 110 BR
	 and	r1, r2		!  78 EX

	! Copy a short word one at a time until we are cache line aligned
	!   Normal values: r0, r2, r3, r4
	!   Unused: r1, r6, r7
	!   Mod: r5 (=r5-2)
	!
	add	#2, r3		!  50 EX

2:	mov.w	@(r0,r5),r1	!  20 LS (latency=2)
	cmp/eq	r3,r0		!  54 MT

	bf/s	2b		! 111 BR

	 mov.w	r1,@-r0		!  29 LS

	! Copy the cache line aligned blocks
	!
	! In use: r0, r2, r4, r5 (=r5-2)
	! Scratch: r1, r3, r6, r7
	!
	! We could do this with the four scratch registers, but if src
	! and dest hit the same cache line, this will thrash, so make
	! use of additional registers.
	!
	! We also need r0 as a temporary (for movca), so 'undo' the invariant:
	!   r5:	 src (was r0+r5)
	!   r1:	 dest (was r0)
	! this can be reversed at the end, so we don't need to save any extra
	! state.
	!
1:	mov.l	r8, @-r15	!  30 LS
	add	r0, r5		!  49 EX

	mov.l	r9, @-r15	!  30 LS
	mov	r0, r1		!   5 MT (latency=0)

	mov.l	r10, @-r15	!  30 LS
	add	#-0x1e, r5	!  50 EX

	mov.l	r11, @-r15	!  30 LS

	mov.l	r12, @-r15	!  30 LS

	! 17 cycles, 32 bytes per iteration
#ifdef __LITTLE_ENDIAN__
2:	mov.w	@r5+, r0	!  14 LS (latency=2)		..JI
	add	#-0x20, r1	!  50 EX

	mov.l	@r5+, r3	!  15 LS (latency=2)		NMLK

	mov.l	@r5+, r6	!  15 LS (latency=2)		RQPO
	shll16	r0		! 103 EX			JI..

	mov.l	@r5+, r7	!  15 LS (latency=2)
	xtrct	r3, r0		!  48 EX			LKJI

	mov.l	@r5+, r8	!  15 LS (latency=2)
	xtrct	r6, r3		!  48 EX			PONM

	mov.l	@r5+, r9	!  15 LS (latency=2)
	xtrct	r7, r6		!  48 EX

	mov.l	@r5+, r10	!  15 LS (latency=2)
	xtrct	r8, r7		!  48 EX

	mov.l	@r5+, r11	!  15 LS (latency=2)
	xtrct	r9, r8		!  48 EX

	mov.w	@r5+, r12	!  15 LS (latency=2)
	xtrct	r10, r9		!  48 EX

	movca.l	r0,@r1		!  40 LS (latency=3-7)
	xtrct	r11, r10	!  48 EX

	mov.l	r3, @(0x04,r1)	!  33 LS
	xtrct	r12, r11	!  48 EX

	mov.l	r6, @(0x08,r1)	!  33 LS

	mov.l	r7, @(0x0c,r1)	!  33 LS

	mov.l	r8, @(0x10,r1)	!  33 LS
	add	#-0x40, r5	!  50 EX

	mov.l	r9, @(0x14,r1)	!  33 LS
	cmp/eq	r2,r1		!  54 MT

	mov.l	r10, @(0x18,r1)	!  33 LS
	bf/s	2b		! 109 BR

	 mov.l	r11, @(0x1c,r1)	!  33 LS
#else
2:	mov.w	@(0x1e,r5), r0	!  17 LS (latency=2)
	add	#-2, r5		!  50 EX

	mov.l	@(0x1c,r5), r3	!  18 LS (latency=2)
	add	#-4, r1		!  50 EX

	mov.l	@(0x18,r5), r6	!  18 LS (latency=2)
	shll16	r0		! 103 EX

	mov.l	@(0x14,r5), r7	!  18 LS (latency=2)
	xtrct	r3, r0		!  48 EX

	mov.l	@(0x10,r5), r8	!  18 LS (latency=2)
	xtrct	r6, r3		!  48 EX

	mov.l	@(0x0c,r5), r9	!  18 LS (latency=2)
	xtrct	r7, r6		!  48 EX

	mov.l	@(0x08,r5), r10	!  18 LS (latency=2)
	xtrct	r8, r7		!  48 EX

	mov.l	@(0x04,r5), r11	!  18 LS (latency=2)
	xtrct	r9, r8		!  48 EX

	mov.l   @(0x00,r5), r12 !  18 LS (latency=2)
	xtrct	r10, r9		!  48 EX

	movca.l	r0,@r1		!  40 LS (latency=3-7)
	add	#-0x1c, r1	!  50 EX

	mov.l	r3, @(0x18,r1)	!  33 LS
	xtrct	r11, r10	!  48 EX

	mov.l	r6, @(0x14,r1)	!  33 LS
	xtrct	r12, r11	!  48 EX

	mov.l	r7, @(0x10,r1)	!  33 LS

	mov.l	r8, @(0x0c,r1)	!  33 LS
	add	#-0x1e, r5	!  50 EX

	mov.l	r9, @(0x08,r1)	!  33 LS
	cmp/eq	r2,r1		!  54 MT

	mov.l	r10, @(0x04,r1)	!  33 LS
	bf/s	2b		! 109 BR

	 mov.l	r11, @(0x00,r1)	!  33 LS
#endif

	mov.l	@r15+, r12
	mov	r1, r0		!   5 MT (latency=0)

	mov.l	@r15+, r11	!  15 LS
	sub	r1, r5		!  75 EX

	mov.l	@r15+, r10	!  15 LS
	cmp/eq	r4, r0		!  54 MT

	bf/s	1f		! 109 BR
	 mov.l	 @r15+, r9	!  15 LS

	rts
1:	 mov.l	@r15+, r8	!  15 LS

	add	#0x1e, r5	!  50 EX

	! Finish off a short word at a time
	! r5 must be invariant - 2
10:	mov	r4,r2		!   5 MT (latency=0)
	add	#1,r2		!  50 EX

	cmp/hi	r2, r0		!  57 MT
	bf/s	1f		! 109 BR

	 add	#2, r2		!  50 EX

3:	mov.w	@(r0,r5),r1	!  20 LS
	cmp/hi	r2,r0		!  57 MT

	bt/s	3b		! 109 BR

	 mov.w	r1,@-r0		!  29 LS
1:

	!
	! Finally, copy the last byte if necessary
	cmp/eq	r4,r0		!  54 MT
	bt/s	9b
	 add	#1,r5
	mov.b	@(r0,r5),r1
	rts
	 mov.b	r1,@-r0

END(memcpy)
libc_hidden_def (memcpy)