2 * Copyright (c) 2013 The NetBSD Foundation, Inc.
5 * This code is derived from software contributed to The NetBSD Foundation
6 * by Matt Thomas of 3am Software Foundry.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
18 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
19 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
20 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
21 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 * POSSIBILITY OF SUCH DAMAGE.
30 #include <machine/asm.h>
32 RCSID("$NetBSD: memset_naive.S,v 1.1 2013/01/08 20:15:00 matt Exp $")
35 * This isn't quite as simple/short as it could be but the truly trivial
36 * memset was an order of magnitude slower than this.
40 /* LINTSTUB: void *memset(void *, int, size_t) */
41 mov ip, r0 /* need to preserve r0 */
42 cmp r2, #10 /* 10 bytes or less? */
43 ble .Lbyte_by_byte /* yes, bytewise is faster */
44 ands r3, r1, #0xff /* we are dealing with bytes */
45 orrne r3, r3, r3, lsl #8 /* move value into 2nd byte lane */
46 orrne r3, r3, r3, lsl #16 /* move value into all byte lanes */
47 mov r1, r2 /* move count */
48 ands r2, ip, #7 /* are we dword aligned? */
49 beq 1f /* yes we are */
50 rsb r2, r2, #8 /* how many bytes until aligned? */
51 sub r1, r1, r2 /* subtract from count */
52 tst ip, #1 /* halfword aligned? */
53 strneb r3, [ip], #1 /* nope, write a byte */
54 tst ip, #2 /* word aligned? */
55 strneh r3, [ip], #2 /* nope, write a halfword */
56 tst ip, #4 /* dword aligned? */
57 strne r3, [ip], #4 /* nope, write a word */
59 * At this point, we are dword aligned.
61 1: mov r2, r3 /* duplicate fill value */
62 2: subs r1, r1, #16 /* can we write 16 bytes? */
63 stmgeia ip!, {r2,r3} /* yes, write the first 8 of them */
64 stmgeia ip!, {r2,r3} /* yes, write the second 8 of them */
65 bgt 2b /* more left to fill */
66 RETc(eq) /* no, return */
68 * Our count went negative but the bits below 16 haven't changed.
69 * So we are effectively testing modulo 16.
71 tst r1, #8 /* can we write at least 8 bytes? */
72 stmneia ip!, {r2,r3} /* so do it */
73 tst r1, #4 /* can we write at least 4 bytes? */
74 strne r3, [ip], #4 /* so do it */
75 tst r1, #2 /* can we write at least 2 bytes? */
76 strneh r3, [ip], #2 /* so do it */
77 tst r1, #1 /* can we write 1 bytes? */
78 strneb r3, [ip], #1 /* so do it */
82 subs r2, r2, #1 /* can we write a byte? */
83 RETc(lt) /* no, return */
84 strb r3, [ip], #1 /* write a byte */
85 b .Lbyte_by_byte /* do next byte */