sh/lib/memcpy-sh4.S

28 	! Read a long word and write a long word at once
34 	add	#-4,r5		!  50 EX
74 	! Finally, copy a byte at once, if necessary
76 	add	#4,r5		!  50 EX
79 	add	#-6,r2		!  50 EX
101 	! Read a long word and write a long word at once
107 	add	#-4,r5		! 50 EX
142 	! Finally, copy a byte at once, if necessary
144 	add	#6,r5		!  50 EX
147 	add	#-6,r2		!  50 EX
193 	! Check for small copies, and if so do a simple byte at a time copy.
197 	! cycle counts for differnet sizes using byte-at-a-time vs. optimised):
199 	!	16	42		39-40	46-50	50-55
201 	!	36	82		49-50	66-70	80-85
203 	! aligned data (and this is more common), so use a value of 16.
207 	add	#-1,r5		!  50 EX
216 	 add	#-1,r3		!  50 EX
240 	! First align destination to a long word boundary.
316 	add	#-4, r5		!  50 EX
325 	 add	#-8, r3		!  50 EX
351 	add	#-4, r5		!  50 EX
357 	add	#11, r7		!  50 EX
363 	 add	#-4, r3		!  50 EX
379 	add	#3,r5		!  50 EX
382 	add	#-10, r7	!  50 EX
403 	add	#-4, r5		!  50 EX
413 	add	#0x1f, r2	!  50 EX
424 	add	#-4, r6		!  50 EX
427 	 add	#8, r3		!  50 EX
454 	! We also need r0 as a temporary (for movca), so 'undo' the invariant:
467 	add	#-0x1c, r5	!  50 EX
473 	add	#-0x20, r1	! 50 EX
487 	add	#-0x20, r5	! 50 EX
520 	add	#0x1c, r5	!  50 EX		(back to -4)
524 	 add	 #11, r7	!  50 EX
532 	 add	#-4, r3		!  50 EX
552 	add	#-10, r7	!  50 EX
555 	add	#3,r5		!  50 EX
576 	add	#-2,r5		!  50 EX
579 	add	#-4,r6		!  50 EX
581 	add	#7,r2		!  50 EX
610 	add	#0x1f, r2	!  50 EX
612 	add	#-2, r5		!  50 EX
616 	! Copy a short word one at a time until we are cache line aligned
621 	add	#2, r3		!  50 EX
639 	! We also need r0 as a temporary (for movca), so 'undo' the invariant:
652 	add	#-0x1e, r5	!  50 EX
661 	add	#-0x20, r1	!  50 EX
697 	add	#-0x40, r5	!  50 EX
708 	add	#-2, r5		!  50 EX
711 	add	#-4, r1		!  50 EX
735 	add	#-0x1c, r1	!  50 EX
746 	add	#-0x1e, r5	!  50 EX
772 	add	#0x1e, r5	!  50 EX
774 	! Finish off a short word at a time
777 	add	#1,r2		!  50 EX
782 	 add	#2, r2		!  50 EX