parisc/lib/lusercopy.S

1 /* SPDX-License-Identifier: GPL-2.0-or-later */
5  *    Copyright (C) 2000-2002 Hewlett-Packard (John Marvin)
6  *    Copyright (C) 2000 Richard Hirst <rhirst with parisc-linux.org>
8  *    Copyright (C) 2003 Randolph Chung <tausq with parisc-linux.org>
40 	addib,<>    -1,%r25,$lclu_loop
48 	ldo         1(%r25),%r25
58  * - sr1 already contains space of source region
59  * - sr2 already contains space of destination region
62  * - number of bytes that could not be copied.
65  * This code is based on a C-implementation of a copy routine written by
69  * conditions. In the optimal case, we copy by loops that copy 32- or 16-bytes
71  * aligning the destination and then using shift-and-write method, or in a few
72  * cases by falling back to a byte-at-a-time copy.
75  * often >10x faster than a simple byte-at-a-time copy, even for strangely
78  * it by 30-40% for aligned copies because of the loop unrolling, but in some
83  * - add cache prefetching
84  * - try not to use the post-increment address modifiers; they may create
123 	/* only do 64-bit copies if we can get aligned. */
127 	/* loop until we are 64-bit aligned */
134 	ldo	-1(len),len
146 	ldo	16(src),src
151 	ldo	16(src),src
165 	ldo	-32(len),len
172 	ldo	-4(len),len
179 	/* loop until we are 32-bit aligned */
186 	ldo	-1(len),len
201 	ldo	16(src),src
215 	ldo	-16(len),len
221 	ldo	1(src),src
224 	ldo	-1(len),len
237 	/* align until dst is 32bit-word-aligned */
241 	ldo	1(src),src
244 	ldo	-1(len),len
260 	 * Copy from a not-aligned src to an aligned dst using shifts.
290 	ldo -1(len),len
316 	ldo -4(len),len
326 	/* calculate new src, dst and len and jump to byte-copy loop */
338 	ldo 1(len),len
345 	ldo 2(len),len