2 * arch/xtensa/lib/memset.S
4 * ANSI C standard library function memset
5 * (Well, almost. .fixup code might return zero.)
7 * This file is subject to the terms and conditions of the GNU General
8 * Public License. See the file "COPYING" in the main directory of
9 * this archive for more details.
11 * Copyright (C) 2002 Tensilica Inc.
14 #include <linux/linkage.h>
15 #include <asm/asmmacro.h>
19 * void *memset(void *dst, int c, size_t length)
21 * The algorithm is as follows:
22 * Create a word with c in all byte positions
23 * If the destination is aligned,
24 * do 16B chucks with a loop, and then finish up with
25 * 8B, 4B, 2B, and 1B stores conditional on the length.
26 * If destination is unaligned, align it by conditionally
27 * setting 1B and 2B and then go to aligned case.
28 * This code tries to use fall-through branches for the common
29 * case of an aligned destination (except for the branches to
30 * the alignment labels).
38 # a2/ dst, a3/ c, a4/ length
39 extui a3, a3, 0, 8 # mask to just 8 bits
40 slli a7, a3, 8 # duplicate character in all bytes of word
44 mov a5, a2 # copy dst so that a2 is return value
45 movi a6, 3 # for alignment tests
46 bany a2, a6, .Ldstunaligned # if dst is unaligned
47 .L0: # return here from .Ldstunaligned when dst is aligned
48 srli a7, a4, 4 # number of loop iterations with 16B
54 * Destination is word-aligned.
56 # set 16 bytes per iteration for word-aligned dst
57 .align 4 # 1 mod 4 alignment for LOOPNEZ
58 .byte 0 # (0 mod 4 alignment for LBEG)
61 loopnez a7, .Loop1done
62 #else /* !XCHAL_HAVE_LOOPS */
65 add a6, a6, a5 # a6 = end of last 16B chunk
66 #endif /* !XCHAL_HAVE_LOOPS */
68 EX(10f) s32i a3, a5, 0
69 EX(10f) s32i a3, a5, 4
70 EX(10f) s32i a3, a5, 8
71 EX(10f) s32i a3, a5, 12
75 #endif /* !XCHAL_HAVE_LOOPS */
79 EX(10f) s32i a3, a5, 0
80 EX(10f) s32i a3, a5, 4
85 EX(10f) s32i a3, a5, 0
90 EX(10f) s16i a3, a5, 0
101 * Destination is unaligned
105 bltui a4, 8, .Lbyteset # do short copies byte by byte
106 bbci.l a5, 0, .L20 # branch if dst alignment half-aligned
107 # dst is only byte aligned
109 EX(10f) s8i a3, a5, 0
112 # now retest if dst aligned
113 bbci.l a5, 1, .L0 # if now aligned, return to main algorithm
117 EX(10f) s16i a3, a5, 0
120 j .L0 # dst is now aligned, return to main algorithm
126 .byte 0 # 1 mod 4 alignment for LOOPNEZ
127 # (0 mod 4 alignment for LBEG)
130 loopnez a4, .Lbytesetdone
131 #else /* !XCHAL_HAVE_LOOPS */
132 beqz a4, .Lbytesetdone
133 add a6, a5, a4 # a6 = ending address
134 #endif /* !XCHAL_HAVE_LOOPS */
136 EX(10f) s8i a3, a5, 0
138 #if !XCHAL_HAVE_LOOPS
139 blt a5, a6, .Lbyteloop
140 #endif /* !XCHAL_HAVE_LOOPS */
146 .section .fixup, "ax"
149 /* We return zero if a failure occurred. */